Yang Yang-5G Wireless Systems. Simulation and Evaluation Techniques-Springer (2017)

Wireless Networks
YangYang
JingXu
GuangShi
Cheng-XiangWang
5G Wireless
Systems
Simulation and Evaluation Techniques
Wireless Networks
Series editor
Xuemin Sherman Shen
University of Waterloo, Waterloo, Ontario, Canada
More information about this series at http://www.springer.com/series/14180
Yang Yang Jing Xu Guang Shi
Cheng-Xiang Wang
5G Wireless Systems
Simulation and Evaluation Techniques
Yang Yang Jing Xu
CAS Key Lab of Wireless Sensor Shanghai Research Center for Wireless
Network and Communication Communications
Shanghai Institute of Microsystem Shanghai, China
and Information Technology
Shanghai, China
Guang Shi Cheng-Xiang Wang

China Institute of Communications School of Engineering & Physical Sciences
Beijing, Beijing Heriot-Watt University
China Edinburgh, United Kingdom
ISSN 2366-1186 ISSN 2366-1445 (electronic)

Wireless Networks
ISBN 978-3-319-61868-5 ISBN 978-3-319-61869-2 (eBook)
DOI 10.1007/978-3-319-61869-2
Library of Congress Control Number: 2017947041
Springer International Publishing AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
It is anticipated that the Fifth Generation (5G) mobile communication systems will
start to be commercialized and deployed in 2018 for new mobile services in three
key scenarios, i.e., enhanced Mobile Broadband (eMBB), massive Machine Type
Communications (mMTC), and Ultra-Reliable and Low Latency Communications
(uRLLC). Based on the preliminary research for 5G standardization by telecom
industry leaders since 2012, the International Telecommunication Union (ITU) has
identified and announced the 5G vision and Key Performance Indicators (KPI) on
spectrum efficiency, energy efficiency, peak data rate, traffic density, device con-
nectivity, radio latency and reliability for achieving more comprehensive and better
service provisioning and user experience. At present, the telecom industry is
actively developing a variety of enabling technologies, such as massive MIMO,
Ultra-Dense Network (UDN), and mmWAVE to accelerate the ongoing 5G stan-
dardization and precommercial trial processes. On the other side, new 5G technol-
ogies drive the design and development of corresponding simulation platforms,
evaluation methods, field trails, and application scenarios to compare, screen, and
improve 5G candidate technologies for the ongoing 5G standardization, system
development, and performance enhancement. In view of these 5G technological
trends and testing requirements, this book aims at addressing the technical chal-
lenges and sharing our practical experiences in the simulation and evaluation of a
series of 5G candidate technologies. In particular, it reviews the latest research and
development activities of 5G candidate technologies in the literature, analyzes the
real challenges in testing and evaluating these technologies, proposes different
technical approaches by combining advanced software and hardware capabilities,
and presents the convincing evaluation results in realistic mobile environments and
application scenarios, which are based on our long-term dedicated research and
practical experiences in wireless technology evaluation and testbed development.
This book reviews and provides key testing and evaluation methods for 5G candi-
date technologies from the perspective of technical R&D engineers. In order to
facilitate readers with different backgrounds to better understand important con-
cepts and methods, we include in this book many examples of various simulation
v
vi Preface
and testing cases, which can be used as technical references by 5G researchers,

engineers, and postgraduate students in wireless mobile communications.
This book is organized as follows. Chapter 1 reviews the vision and technical
requirements of 5G mobile systems. Some 5G candidate technologies and the
challenges in technical evaluation and verification methods are then discussed.
Chapter 2 introduces the evolution history of a variety of testing and evaluation
technologies, thus offering a comprehensive and systematic understanding of the
importance and difficulties of testing technologies. Since wireless channel charac-
teristics play a key role in system design and performance evaluation, Chapter 3
introduces state-of-the-art wireless channel measurement, modeling, and simula-
tion methods in detail. Chapter 4 is focused on the development of a large-scale
system-level software simulation platform by using parallel computing technology.
Based on Hardware-in-the Loop (HIL) technology, Chapter 5 develops a hardware
and software co-simulation platform with a real-time channel emulator in the loop,
thus balancing the simulation speed and system flexibility. Further, Chapter 6
presents a truly real-time hardware testing and evaluation platform with a set of
advanced measuring equipments and instruments for small-scale 5G technology
evaluation. Finally, Chapter 7 gives the layout and design of a real 5G testbed with
six macro-cells for field trials and future mobile services.
Last but not least, we would like to acknowledge the generous financial supports
over the last ten years from the National Science and Technology Major Projects
New Generation Mobile Wireless Broadband Communication Networks,
National Natural Science Foundation of China, Ministry of Science and Technol-
ogy of China, Ministry of Industry and Information Technology of China, Chinese
Academy of Sciences, Science and Technology Commission of Shanghai Munic-
ipality, and Shanghai Municipal Commission of Economy and Informatization. In
addition, we are very grateful to the following colleagues for their kind help and
constructive comments on earlier versions of this book, i.e., Xin Yang, Haowen
Wang, Hui Xu, Kai Li, Xiumei Yang, Jian Sun, Jinling Du, Yuanping Zhu,
Chenping, Panrongwei, Guannan Song, Guowei Zhang, and Kaili Wang.
Shanghai, China Yang Yang

Shanghai, China Xu Jing
Beijing, China Shi Guang
Edinburgh, UK Wang Cheng-Xiang
Contents
1 Candidate Technologies and Evaluation Challenges for 5G . . . . . . . 1

1.1 5G Requirements Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 5G Application Types . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 5G Application Scenarios and Requirements
Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 5G Research and Development Process . . . . . . . . . . . . . . . . . . . . 4
1.3 5G Candidate Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 Ultra-Dense Wireless Network . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Large Scale Antenna Technology . . . . . . . . . . . . . . . . . . 7
1.3.3 Millimeter Wave Communications Technology . . . . . . . . 8
1.3.4 Flexible Spectrum Usage . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.5 Waveform and Multiple Access . . . . . . . . . . . . . . . . . . . 11
1.3.6 Device to Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Challenges in 5G Testing and Evaluation . . . . . . . . . . . . . . . . . . . 13
1.4.1 Challenges Posed by New Technologies to Testing
and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.2 Four Elements of Testing and Evaluation
Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Evolution of Testing Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1 The Importance of Testing Technology . . . . . . . . . . . . . . . . . . . . 21
2.2 Testing Technology Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.1 Development of Testing Instruments . . . . . . . . . . . . . . . . 23
2.2.2 Development Trend of Testing Technology . . . . . . . . . . . 33
2.3 Challenges of Testing Technology . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 Challenges of Multi-Function and High-Performance . . . 38
2.3.2 Challenges of Multi-channel . . . . . . . . . . . . . . . . . . . . . . 39
2.3.3 Challenges of High Throughput . . . . . . . . . . . . . . . . . . . 42
vii
viii Contents
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 Channel Measurement and Modeling . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 Requirements for 5G Wireless Channel Model . . . . . . . . . . . . . . . 45
3.2 Channel Modeling Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2.1 Measurement-Based GSCM . . . . . . . . . . . . . . . . . . . . . . 51
3.2.2 Regular-shape GSCM . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.3 CSCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.4 Extended SV Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.5 Ray Tracing based Model . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.6 Comparison of Modeling Methods . . . . . . . . . . . . . . . . . 60
3.3 Channel Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.3.1 Channel Measurement Methods . . . . . . . . . . . . . . . . . . . 62
3.3.2 Channel Measurement Activities . . . . . . . . . . . . . . . . . . 71
3.4 Channel Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.1 Path Parameters Extraction . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.2 Channel Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . 103
3.5 Existing Channel Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.5.1 3GPP SCM/3D/D2D/HF . . . . . . . . . . . . . . . . . . . . . . . . 113
3.5.2 WINNER I/II/+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
3.5.3 ITU IMT-Advanced, IMT-2020 . . . . . . . . . . . . . . . . . . . 116
3.5.4 COST 259/273/2100/IC1004 . . . . . . . . . . . . . . . . . . . . . 117
3.5.5 IEEE 802.11 TGn/TGac . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.5.6 QuaDRiGa, mmMAGIC . . . . . . . . . . . . . . . . . . . . . . . . 119
3.5.7 IEEE 802.15.3c/ IEEE 802.11ad/aj/ay . . . . . . . . . . . . . . . 120
3.5.8 MiWEBA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.5.9 METIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.5.10 5GCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.5.11 Comparison of Existing Models . . . . . . . . . . . . . . . . . . . 123
3.6 Stochastic Channel Generation . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.6.1 Definition of Simulation Scenarios . . . . . . . . . . . . . . . . . 127
3.6.2 Generation of Large Scale Parameters . . . . . . . . . . . . . . . 129
3.6.3 Generation of Small Scale Parameters . . . . . . . . . . . . . . . 135
3.6.4 Generation of Channel Coefficient . . . . . . . . . . . . . . . . . 141
3.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4 Software Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.1 Overview on Software Simulation . . . . . . . . . . . . . . . . . . . . . . . . 157
4.2 5G Software Simulation Requirements . . . . . . . . . . . . . . . . . . . . 161
4.2.1 Simulation Requirements Analysis . . . . . . . . . . . . . . . . . 161
4.2.2 Technological Impact Analysis . . . . . . . . . . . . . . . . . . . . 164
4.3 5G Software Link Level Simulation . . . . . . . . . . . . . . . . . . . . . . 167
4.3.1 Link Technology Overview . . . . . . . . . . . . . . . . . . . . . . 167
Contents ix
4.3.2 Link Simulation Realization . . . . . . . . . . . . . . . . . . . . . . 168

4.3.3 Introduction to Simulation Cases . . . . . . . . . . . . . . . . . . 173
4.4 5G Software System Level Simulation . . . . . . . . . . . . . . . . . . . . 188
4.4.1 Test Evaluation Methods . . . . . . . . . . . . . . . . . . . . . . . . 188
4.4.2 Key Simulation Technologies . . . . . . . . . . . . . . . . . . . . . 190
4.4.3 Introduction to Simulation Cases . . . . . . . . . . . . . . . . . . 204
4.5 Software Visualization of 5G Network Simulation . . . . . . . . . . . . 226
4.5.1 Architecture Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 226
4.5.2 Interface Demonstration . . . . . . . . . . . . . . . . . . . . . . . . . 227
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
5 Evaluation Test of Software and Hardware Co-simulation . . . . . . . . 235
5.1 Overview of Software and Hardware Co-Simulation
Evaluation and Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
5.1.1 Requirements of 5G Software and Hardware
Simulation Evaluation and Test . . . . . . . . . . . . . . . . . . . 236
5.1.2 Forms and Applications of Software
and Hardware Co-Simulation Testing
and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
5.2 Evaluation and Test of Software and Hardware
Link-Level Co-Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.2.1 HIL-Based Link Simulation Composition . . . . . . . . . . . . 240
5.2.2 Realization of Link Simulation Testing
and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
5.2.3 Case Introduction of Software and Hardware
Simulation Evaluation Test . . . . . . . . . . . . . . . . . . . . . . 247
5.3 Evaluation and Test on Hardware and Software
System-level Co-simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
5.3.1 Composition of Hardware and Software
System-level Co-simulation . . . . . . . . . . . . . . . . . . . . . . 277
5.3.2 Key Technologies of Hardware and Software
System-Level Co-simulation . . . . . . . . . . . . . . . . . . . . . 278
5.3.3 Case Study of Simulation . . . . . . . . . . . . . . . . . . . . . . . . 284
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
6 5G Hardware Test Evaluation Platform . . . . . . . . . . . . . . . . . . . . . . 301
6.1 Overview of Typical Hardware Platform Used
for 5G Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
6.2 Test Platform of MIMO Parallel Channel . . . . . . . . . . . . . . . . . . 303
6.2.1 5G Channel Test Requirements . . . . . . . . . . . . . . . . . . . 303
6.2.2 Status and Shortcomings . . . . . . . . . . . . . . . . . . . . . . . . 304
6.2.3 Key Technical Challenges . . . . . . . . . . . . . . . . . . . . . . . 306
6.2.4 Architecture of MIMO Parallel Channel Sounder . . . . . . 307
x Contents
6.2.5 Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

6.2.6 Channel Measurement and Test . . . . . . . . . . . . . . . . . . . 318
6.3 OTA Test Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
6.3.1 5G Requirements for OTA Test . . . . . . . . . . . . . . . . . . . 320
6.3.2 Status and Shortcomings of OTA Test Scheme . . . . . . . . 321
6.3.3 Key Technical Difficulties and Challenges . . . . . . . . . . . 329
6.3.4 OTA Platform Example . . . . . . . . . . . . . . . . . . . . . . . . . 331
6.4 5G Open Source Community . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
6.4.1 Introduction to Requirements . . . . . . . . . . . . . . . . . . . . . 332
6.4.2 Development Status of Communications Platform
Based on General Purpose Processor . . . . . . . . . . . . . . . 335
6.4.3 Introduction to Open 5G Universal Platform . . . . . . . . . . 336
6.4.4 Key Technologies of Universal Platform Based
on General Purpose Processor . . . . . . . . . . . . . . . . . . . . 340
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
7 Field Trial Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
7.1 Requirements and Technical Challenges . . . . . . . . . . . . . . . . . . . 359
7.1.1 Various Application Scenarios . . . . . . . . . . . . . . . . . . . . 359
7.1.2 Technical Challenges in Field Trial . . . . . . . . . . . . . . . . 361
7.1.3 Development Status of 5G Test Bed in Foreign
Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
7.1.4 Development and Evolution of HetNet Convergence . . . . 364
7.2 5G Test Field Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
7.2.1 Network Architecture of HetNets Convergence . . . . . . . . 366
7.2.2 Abundant 5G Typical Scenarios . . . . . . . . . . . . . . . . . . . 366
7.2.3 3GPP Wireless Network Design . . . . . . . . . . . . . . . . . . . 369
7.2.4 WLAN Wireless Network Design . . . . . . . . . . . . . . . . . . 385
7.2.5 Large-Scale Test User Application Scenarios . . . . . . . . . 403
7.3 Typical Case Analysis of Field Trials . . . . . . . . . . . . . . . . . . . . . 403
7.3.1 Case One: Research on the Irregularity of Wireless
Cellular Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
7.3.2 Case Two: Characteristics Analysis on the Radio
Wave Propagation Between Multiple Frequency
Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
7.3.3 Case Three: CoMP Key Technologies Test . . . . . . . . . . . 409
7.3.4 Case Four: Active Antenna System (AAS) Key
Technologies Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
7.3.5 Case Five: Inter-Cell Interference Coordination
(ICIC) Key Technologies Test . . . . . . . . . . . . . . . . . . . . 413
7.3.6 Case Six: Public WLAN Network Coverage
Performance Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
Contents xi
7.4 Wireless Network Data Intelligence Analysis . . . . . . . . . . . . . . . . 419

7.4.1 Background and Necessity . . . . . . . . . . . . . . . . . . . . . . . 419
7.4.2 Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
7.4.3 Application Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 426
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
Abbreviations
5G The Fifth Generation mobile communications system

2D Two-Dimensional
2G 2-Generation wireless telephone technology
3D Three-Dimensional
3G the 3rd Generation mobile communications system
3GPP the 3rd Generation Partnership Project
4G Fourth Generation mobile communications system
5GIC 5G Innovation Center
A2GSM Ad hoc assisted GSM
AAA Authentication, Authorization, Accounting
AAS Active Antenna System
A-BFT Associated Beam Forming training Time
AC Access Controller
ACK ACKnowledgement
A-D Anderson-Darling
AD/DA Analogy Digital/Digital Analogy converter
ADC Analog-to-Digital Converter
AFD Average Fade Duration
AMC Adaptive Modulation and Coding
A-MPDU Aggregated MAC Protocol Data Unit
ANO Automated Network Organization
AOA Angle Of Arrival
AOD Azimuth angle Of Departure
AP Access Point
APDP Averaged PDP
ARM Advanced RISC Machines
ARS Ad hoc Relay network Stations
AS Angle Spread
ASA Azimuth angular Spread at Arrival
ASIC Application-Specific Integrated Circuit
xiii
xiv Abbreviations
AT Announcement Time
AUT UT Austin
AWGN Additive White Gaussian Noise
AWRI Advanced Wireless Research Initiative
AXI Advanced eXtensible Interface
AXIe Advanced eXtensible Interface express
B/S Browser/Server
BBU BaseBand Unit
BDS BeiDou Navigation Satellite System
BER Bit Error Rate
BFS Breadth-First Search algorithm
BI Beacon Intervals
B-ISDN Broadband Integrated Services Digital Network
BLER BLock Error Rate
BRP Beam Refinement Phase
BS Base Station
BSC Base Station Controller
BT Beacon Time
CA Carrier Aggregation
CAPEX CAPital EXpenditure
CAPWAP Control and Provision of Wireless Access Points
CB Coherence Bandwidth
CBP Competition-Based access Period
CC Cooperative Communications
CDD Cyclic Delay Diversity
CDD-OFDM Cyclic Delay Diversity Orthogonal Frequency Division
Multiplexing
CDF Cumulative Distribution Function
CDL Cluster Delay Line
CDM Cyclic Delay Modulation
CDMA Code-Division Multiple Address
CDMA2000 Code Division Multiple Access 2000
CDN Content Delivery Network
CI Close-In the reference short
CIF CI with Frequency
CIR Channel Impulse Response
CMD Correlation Matrix Distance
CoMP Coordinated Multiple Points
COST European Cooperation in Science and Technology
COTS Commercial Off The Shelf
COW Cluster of Workstation/PC
CP Cyclic Prefix
CPCI Compact PCI
CPRI Common Public Radio Interface
Abbreviations xv
CPU Central Processing Unit

CQI Channel Quality Indicator
C-RAN Cloud Radio Access Network
CRC Cyclic Redundancy Check
CRLBs Cramer Rao Lower Bounds
CRS Cell Reference Signal
CS Compressive Sensing
CS/CBF Coordinated Scheduling/BeamForming
CSCM Correlation-based Stochastic Channel Model
CSF Cluster Shadow Fading
CSI Channel State Information
CTD-ML Circular Trap Detection based ML decoder
decoder
CTIA Cellular Telecommunications and Internet Association
CTS Clear To Send
CTTL China Telecommunication Technology Labs
CUDA Compute Unified Device Architecture
CVA Circular Viterbi Algorithm
CW Continuous Wave
D2D Device to Device
DAB Digital Audio Broadcasting
DAC Digital-to-Analog Converter
DC Direct Current
DCT-CDA Discrete Cosine Transform-CDA
DDC Direct Digital Controller
DDP Delay-Doppler Profile
DFT Discrete Fourier Transform
DL DownLink
DMA Direct Memory Access
DMC Diffuse scattering Multipath Components
DoD Department of Defense
DOF Degrees Of Freedom
DS Delay Spread
DSD Doppler power Spectral Density
DSM Distributed Shared Memory multiprocessor
DSP Digital Signal Processors
DTT Data Transmission Time
DUT Device Under Test
DVB Digital Video Broadcasting
DVB-H Digital Video Broadcasting Handheld
EB Elektrobit
ECU Electronic Control Unit
EDA Electronics Design Automation
EKF Extended Kalman Filter
xvi Abbreviations
EM Expectation-Maximization
EOA Elevation angle Of Arrival
EOD Elevation angle Of Departure
EPC Evolved Packet Core
ER Effective Roughness
ESNR Estimated Signal-to-Noise Ratio
ESPRIT Estimation of Signal Parameter via Rotational Invariance
Techniques
EU European Union
EuQoS End-to-end Quality Of Service Support over Heterogeneous
Networks
EUT Equipment Under Test
EVA Extended Vehicular A model
EVM Error Vector Magnitude
FB Flexible Backhauling
FBMC Filter Bank based Multi-Carrier
FCC Federal Communications Commission
FDD Frequency Division Duplex
FDMA Frequency Division Multiple Address
FD-SAGE Frequency Domain SAGE
FER Frame Error Rate
FFR Fractional Frequency Reuse
FFT Fast Fourier Transform
FIFO First Input First Output
F-OFDM Filtered OFDM
FPGA Field Programmable Gate Array
FR Fraunhofer Region
FST Fast Session Transfer
FT France Telecom
GBC General purpose Baseband Computing
GBCM Geometric-Based stochastic Channel Model
GCS Global Coordinate System
GIS Geographic Information System
GO Geometrical Optics
GoF Goodness of Fit
GPIB General Purpose Interface Bus
GPP General Purpose Processor
GPS Global Positioning System
GPU Graphics Processing Unit
GSCM Geometric-based Stochastic Channel Model
GSM Global System for Mobile communication
GSM-R Global System for Mobile Communications-Railway
GTD Geometrical Theory of Diffraction
HARQ Hybrid Automatic Repeat reQuest
Abbreviations xvii
H-CRAN Heterogeneous Cloud Radio Access Network

HD High Definition
HDL Hardware Description Language
HeNB Home eNodeB
HetNet Heterogeneous Network
HF High Frequency
HFB High Frequencies Band
HIL Hardware-In-the-Loop
HSDPA High-Speed Downlink Packet Access
HSPA High-Speed Packet Access
HSR High-Speed Rail
HSS Home Subscriber Server
HTRDP High Technology Research and Development Program
I/O Input/Output
I2I Indoor-to-Indoor
IBM International Business Machines Corporation
IBSS Independent Basic Service Set
iCAR integrated Cellular and Ad hoc Relaying System
ICI Inter-Cell Interference
ICIC Inter-Cell Interference Coordination
IDFT Inverse Discrete Fourier Transform
IDMA Interleave Division Multiple Access
IETF Internet Engineering Task Force
IF Intermediate Frequency
IFDMA Interleaved FDMA
IFFT Inverse Fast Fourier Transform
IID Independent and Identically Distributed
IMC Intel Mobile Communications
IMS IP Multimedia Subsystem
IMT-2020 International Mobile Telecommunications-2020
IMT-A International Mobile Telecommunications Advanced
InH Indoor Hotspot
INNL Intel Nizhny Novgorod Lab
IOI Inter-Operator Interference
IoT Internet of Things
IoV Internet of Vehicles
IP Internet Protocol
IPP Integrated Performance Primitives
IPSec Internet Protocol Security
ISA Industry Standard Architecture
ISD Inter-Site Distance
ISDN Integrated Services Digital Network
ISI-SAGE Initialization and Search Improved SAGE
ISM Industrial Scientific Medical
xviii Abbreviations
IT Information Technology
ITU International Telecommunications Union
JP Joint Processing
KBSM Kronecker product based CBSM
KEST Kalman Enhanced Super resolution Tracking algorithm
KF K-Factor
KPI Key Performance Indicators
K-S Kolmogorov-Smirnov
LAA LTE Assisted Access
LAN Local Area Network
LCR Level Crossing Rate
LCS Local Coordinate System
LDPC Low Density Parity Check
LDSMA Low Density Spreading Multiple Access
LFDMA Localized FDMA
LMDS Local Multipoint Distribution Services
LO Local Oscillator
LOS Light Of Sight
Low Power Low Power Single Carrier PHYsical layer
SCPHY
LS Least Square
LSF Local Scattering Function
LSPs Large-Scale Parameters
LTE Long-Term Evolution
LTE-A LTE-Advanced
LTE-U LTE use Unlicensed band
LUT Look Up Tables
LXI LAN eXtension for Instrumentation
M2 M Machine to Machine
MAC Media Access Control
MAGNET My personal Adaptive Global NET
MAN Metropolitan Area Network
MCD Multipath Component Distance
MCN Multi-hop Cellular Network
MCS Modulation and Coding Scheme
MED Maximum Excess Delay
METIS Mobile and wireless communications Enablers for Twenty-
twenty (2020) Information Society
MIC Many Integrated Cores
MIMO Multiple Input and Multiple Output
MiWEBA Millimetre-Wave Evolution for Backhaul and Access
ML Maximum Likelihood
MLE Maximum Likelihood Estimation
MME Mobility Management Entity
Abbreviations xix
MMSE Minimum Mean Squared Error

MOBYDICK MOBilitY and DIfferentiated serviCes in a future IP networK
MoM Method of Moments
MPCs Multi-Path Components
MPDU Message Protocol Data Unit
MPP Massively Parallel Processor
MSDU MAC Service Data Unit
MT Mobile Terminal
MUI MultiUser Interference
MU-MIMO MultiUser MIMO
MUSIC MUltiple SIgnal Classification
NACK Negative ACKnowledgement
NFV Network Functions Virtualization
NGN Next Generation Network
NI National Instrument
NICT National Institute of Information and Communications
Technology
NIST National Institute of Standards and Technology
NLOS Non-Light Of Sight
NOMA Non-Orthogonal Multiple Access
NSF National Science Foundation
NTB Non-Transparent Bridge
NYU New York University
O2I Outdoor-to-Indoor
O2O Outdoor-to-Outdoor
OAI Open Air Interface
OCXO Oven Controlled Crystal Oscillator
OEW Open-Ended Waveguide
OFDM Orthogonal Frequency Division Multiplexing
OFDMA Orthogonal Frequency Division Multiple Access
OFDMPHY OFDM PHYsical layer
OLAP OnLine Analysis & Process
OLTP OnLine Traffic Process
OPEX OPerating EXpense
OSI Open System Interconnection
OTA Over The Air
OTT Over The Top
P2P Peer to Peer
PAPR Peak-to-Average Power Ratio
PAS Power Angle Spectrum
PBSS Personal Basic Service Set
PCA Principal Component Analysis
PCA-CDA Principal Component Analysis-CDA
PCC Policy Control and Charging
xx Abbreviations
PCFICH Physical Control Format Indicator CHannel

PCI Peripheral Component Interconnect
PCIe Peripheral Component Interface express
PCP PBSS Control Point
PCRF Policy and Charging Rules Function
PDCCH Physical Downlink Control CHannel
PDCP Packet Data Convergence Protocol
PDF Probability Density Function
PDMA Power Domain Multiple Access
PDN Public Data Network
PDP Power Delay Profile
PDSCH Physical Downlink Shared CHannel
PF Particle Filter
PGW Packet data network GateWay
PHICH Physical HARQ Indicator CHannel
PHY Physical Layer
PHY Control PHYsical Control layer
PL Path Loss
PLI Programming Language Interface
PLL Phase-Locked Loop
PMC Production Material Control
PN Pseudo-random Noise
PPS PCP Power Save
PRB Physical Resource Block
P-SAGE Parallel sounding SAGE
PSS Primary Synchronization Signal
PUCCH Physical Uplink Control CHannels
PUSCH Physical Uplink Shared CHannel
PVP Parallel Vector Processor
PXI PCI eXtensions for Instrumentation
PXIe PCI eXtensions for Instrumentation express
PXImc PXI multi-computing
PXISA PXI Systems Alliance
QAM Quadrature Amplitude Modulation
Q-D Quasi-Deterministic
QoS Quality of Service
QPSK Quadrature Phase Shift Keying
RAM Random Access Memory
RB Resource Block
RE Resource Element
RF Radio Frequency
RI Rank Indication
RIMAX Richters Maximum Likelihood Framework For Parameter
Estimation
Abbreviations xxi
Rma Rural Macro

RMSE Root Mean Squared Error
RN Relay Node
RNC Radio Network Controller
ROM Read Only Memory
RRC Radio Resource Control
RRH Remote Radio Head
RRU Remote Radio Unit
RS Reference Signal
RSRP Reference Signal Receiving Power
RSRQ Reference Signal Receiving Quality
RT Ray Tracing
RTS Request To Send
RVs Random Variables
SAE System Architecture Evolution
SAGE Space-Alternating Generalized Expectation-maximization
SBR Shooting and Bouncing Ray
sBS small Base Stations
SC Specular Component
SC-FDE Single-Carrier Frequency Domain Equalization
SC-FDMA Single-Carrier Frequency-Division Multiple Access
SCM Spatial Channel Model
SCMA Sparse Code Multiple Access
SCME Spatial Channel Model Extension
SCPHY Single Carrier PHYsical layer
SCTD Simplified CTD
SD Spectral Divergence
SDMA Space Division Multiple Access
SDN Software Defined Network
SDP Based Semi-Definite Programming Based algorithm
SDR Software Defined Radio
SDRAM Synchronous Dynamic Random Access Memory
SF Shadow Fading
SFC Shadow Fading Correlation
SFE Spatial Fading Emulator
SFG Sweep Frequency Generator
SGi Short Guard interval
SGW Serving GateWay
SI Self Interference
SIMD Single Instruction Multiple Data
SIMO Single Input Multiple Output
SINR Signal to Interference plus Noise Ratio
SIR Signal Interference Ratio
SISO Single Input Single Output
xxii Abbreviations
SLS Sector Level Sweep

Sma Suburban Macro
SMP Symmetric MultiProcessor
SNR Signal-to-Noise Ratio
SON Self-Organizing Network
SOPRANO Self-Organizing Packet Radio Ad hoc Networks with Overlay
SoS Sum of Sinusoids
SP Service Period
SPM Standard Propagation Model
SR Scheduling Request
SS Small Scale
SSPs Small-Scale Parameters
SSS Secondary Synchronization Signal
STA STAtion
STBC Space-Time Block Coding
STRC Shared Trigger Reference Clock
STTC Space-Time Trellis Codes
SV Saleh-Valenzuela
SVB-SAGE Sparse Variational Bayesian SAGE
SWR Standing Wave Ratio
T-CDA Treelet-based Compressive Data Aggregation
TCP/IP Transmission Control Protocol/Internet Protocol
TDD Time Division Duplex
TDL Tapped Delay Line
TDM Time Division Multiplexing
TDMA Time Division Multiple Address
TDMS Technical Data Management Streaming
TDP Time-Delay Profile
TD-SCDMA Time Division-Synchronous Code Division Multiple
TR Technical Report
Tracking Tracking phase
TRP Total Radiated Power
TRS Total Radiated Sensitivity
TTI Transmission Time Interval
TUI Technical University of Ilmenau
UCA Uniformly Cylindrical Array
UCAN Unified Cellular and Ad hoc Network architecture
UC-LMC User-Centric Local Mobile Cloud
UDN Ultra Dense Network
UDP User Datagram Protocol
UFMC Universal Filtered Multi-Carrier
UL UpLink
ULA Uniform Linear Array
UMa Urban Macro
Abbreviations xxiii
UMass University of Massachusetts

UMB Ultra Mobile Broadband
UMi Urban Micro
UMTS Universal Mobile Telecommunications
UMVUE Uniformly Minimum Variance Unbiased Estimation
USB Universal Serial Bus
USRP Universal Software Radio Peripheral
UT User Terminal
UTD Uniform Theory of Diffraction
UWB Ultra Wide Band
V2 V Vehicle to Vehicle
VHDL Very-High-Speed Integrated Circuit Hardware Description
Language
VISA Virtual Instrumentation Software Architecture
VME Versa Module Eurocard
V-MIMO Virtual MIMO
VNA Vector Network Analyzer
VoIP Voice over Internet Protocol
VR Visibility Region
VSA Vector Signal Analyzer
VSG Vector Signal Generator
VT Virginia Tech
VXI VMEbus eXtensions for Instrumentation
WCDMA Wideband Code Division Multiple Access
WEB World Wide Web
WG1 Working Group 1
WiMAX Worldwide Interoperability for Microwave Access
WINNER Wireless World Initiative New Radio
WLAN Wireless Local Area Network
WTP Wireless Termination Point
XPR cross-polarization Power Ratio
ZF Zero Forcing
ZOA Zenith angle Of Arrival
ZOD Zenith angle Of Departure
ZSA Zimuth angular Spread at Arrival
ZSD Zenith angle Spread of Departure
Chapter 1
Candidate Technologies and Evaluation
Challenges for 5G
With several decades of booming development, mobile communications technol-

ogy has penetrated into many related fields of our daily life. Led by various
emerging applications, users have increasing higher requirements for wireless
services, posing almost stringent requirements for the technical indicators of
network. Therefore, the Fifth Generation mobile communications system
(5G) emerges at a historic moment, devotes itself to open curtain of comprehensive
informational era and provide excellent user experience. Recently, researches on
key technologies for 5G are springing up in the industry, and the corresponding
testing, evaluation and verification system need to be established and improved.
This chapter will briefly introduce the features of candidate technologies for 5G,
and then analyze the challenges faced by 5G evaluation system.
1.1 5G Requirements Analysis
Mobile communications industry has been developing at an amazing speed, and at

present it has become one of the important pillar industries in the global economic
development. There is no doubt that mobile communications technology is chang-
ing peoples life and work, and will continue to have an important impact on social
development [1, 2]. Correspondingly, people have ever-increasing dependence and
demand for mobile communications. In recent years, diversified businesses are
emerging in mobile communications, which has led to the rapid development of
storing and processing technology of massive data. Meanwhile, many break-
throughs have been made on the research and development of artificial intelligence
processor and real-time equipment. The emergence of these new technologies has
brought great convenience to peoples life. Meanwhile, it has also put forward a
greater challenge to the modern mobile communications technologies [3].
Therefore, 5G is facing both opportunities and challenges. For users, 5Gs vision
is information comes as you wish, and everything is in touch [4]. We are going to
Springer International Publishing AG 2018 1

Y. Yang et al., 5G Wireless Systems, Wireless Networks,
DOI 10.1007/978-3-319-61869-2_1
2 1 Candidate Technologies and Evaluation Challenges for 5G
Fig. 1.1 Usage scenarios of IMT for 2020 and beyond
personally feel a gluttonous feast of the information age. The purpose of 5G age is
to construct a stable, convenient and economic information ecosystem for human
beings. As shown in Fig. 1.1, various features of the information age will be
included in the development of 5G, and users can enjoy more convenient intelligent
life [5]. With the popularity of wearable devices, the types and number of mobile
terminals will experience explosive growth. Predictably, in the future, demand for
virtual reality and augmented reality experience, demand for cloudization of mas-
sive office data, wireless control of industrial manufacturing or production pro-
cesses, remote medical surgery, automation in a smart grid, transportation safety
and other aspects, not only require 5G network data transmission rate to reach a
very high level, but also require real-time experience with almost zero latency. In
addition, cost reduction and energy saving should also be considered.
1.1.1 5G Application Types
The birth of 5G largely benefited from the large scale growth of the mobile Internet
and the Internet of Things (IoT), and the application of 5G also mainly lies in the
development of these two networks [69]. In recent years, mobile Internet, as the
carrier of main businesses of mobile data communications, has greatly promoted
the development of various fields of information service. Various service providers
made full use of the advantages of their resources and services and developed
1.1 5G Requirements Analysis 3
numerous refreshing applications, reaching an effect of flowers blooming

together in market. By 2020, users will see more convenient functions realized
via mobile Internet, and demand for virtual reality and augmented reality in game
experience will be met. With the further development of mobile Internet, informa-
tion transmission rate will grow by thousands of times.
IoT is an ideal example of the full application of new generation technology. In
all walks of life, human beings are able to manage production and live in a more
precise and dynamic manner via IoT, achieving an intelligent state and improving
resource utilization and productivity level [10]. IoT, as a main tool to realize
intelligent life, work and production, extends the communications from person-to-
person to person-to-thing and thing-to-thing [11]. The application of IoT is extremely
wide, including environmental protection, intelligent transport, public safety, gov-
ernment work, home safety, intelligent firefighting, environmental monitoring, light-
ing control, health care, food traceability, floriculture, water system monitoring,
enemy spy, information collection and many other fields. We can say that IoT will
be the next important productive power promoting society progress [12]. There-
fore, it is of prime importance to foster the development of IoT, and the development
of IoT depends on the development of communications technology. It is conceivable
that when IoT is everywhere in our life, which means the everything connected is
realized, information transmission will be very frequent among thing-to-thing,
person-to-thing and person-to-person, and this change not only brings vitality and
opportunities but also poses a big challenge to mobile communications [13].
1.1.2 5G Application Scenarios and Requirements Indicators
Application scenarios of 5G are related to every aspect in peoples daily life, work,
entertainment and transportation, and wireless communications will show different
characteristics in all kinds of different scenarios. For example, the crowed or dense
mobile devices areas such as residential areas, stadiums and marketplaces, wireless
communications will have the characteristics of high traffic volume density and
high number of connections, while on transport tools such as subway and high-
speed railways, the high mobility feature of wireless communications will be
prominent. At present, the Fourth Generation mobile communications system
(4G) is not able to satisfy the requirements of some special scenarios featuring
high traffic volume density, high number of connections and high mobility [14].
In crowded scenarios such as stadiums which need ultrahigh traffic volume
density and ultrahigh connection density, we need the wireless communications
transmission rate as high as that of the optical fiber so that it can carry the businesses
like photo transmission, video transmission, live broadcast and other services. In
high-speed mobility scenarios, e.g., High-Speed Rail (HSR), the traffic volume
density and connection are relatively lower than those of stadiums. Since HSRs
speed is usually above 200 km/h, it has high requirements for the wireless commu-
nication systems to support high-speed mobility.
Table 1.1 5G performance requirements

Value
Peak data rate >10 Gbps
User experienced data rate >0. 1 Gbps
Connection density Million connections/square kilometers
Service density 10 Gbps/square kilometers
End-to-end latency Millisecond order
Although its very convenient for us to access the Internet now, half of our world
is still beyond the coverage of Internet after several decades when mobile terminals
came into being. With the development and change of Internet, the denotation of
Internet is expanded accordingly and more and more devices are connected with
each other. Cisco System forecasts that by the year of 2019, the whole world will
have 11.5 billion devices to be connected [15], including some hard-to-connect
devices which are under water or beyond the coverage of satellite, so it becomes
more and more important to meet the requirements for wide coverage in future.
It is predicted that in a long period in the future, mobile data traffic will continue
to show explosive growth [4]: from 2010 to 2020, global mobile data traffic growth
will increase by more than 200 times while from 2010 to 2030, more than 20,000
times. Meanwhile, the growth of Chinese mobile data traffic is higher than the
global average level. It is predicted that from 2010 to 2020, it will grow by more
than 300 times, while from 2010 to 2030, more than 40,000 times.
Therefore, based on the requirements above, 5Gs overall goals are: much faster,
more efficient, and more intelligent. Specific 5G performance requirements are
shown in Table 1.1 [16].
To satisfy users experience in multiple dimensions, combination of different
technologies is needed to reach the 5G performance requirements in the above
table. For example, ultra-dense wireless communications technology can make
contributions to improving performance indicators of user experienced data rate,
connection density and service density through increasing the base station deploy-
ment density. Massive antenna technology can effectively improve the spectrum
utilization efficiency through increasing antennas number and it has an important
significance in improving peak data rate, user experienced data rate, connection
density and service density. Millimeter wave communications technology can
increase usable spectrum in a large scale, which is good for improving performance
indicators of peak data rate, user experienced data rate and service density.
1.2 5G Research and Development Process
From 2009, Long Term Evolution (LTE) set off a boom across the globe. At the end
of 2012, European Union (EU) invested 27 million euros and launched the first 5G
research project in the world Mobile and wireless communications Enablers for
1.3 5G Candidate Technologies 5
Twenty-twenty (2020) Information Society (METIS). There were 29 project mem-

bers participating in this project, including main equipment manufacturers, opera-
tors, car manufacturers and academic institutions, and the research contents covered
5G scenarios and requirements, network architecture and various types of new
wireless technology [17, 18]. In 2013, Chinas 863 Plan invested 300 million
RMB to research and development of 5G system, and the research contents
included wireless network architecture, massive antenna, ultra-dense wireless net-
work, soft base station platform, wireless network virtualization, millimeter wave
indoor wireless access, as well as evaluation, test and verification. In order to
promote the development of 5G system, governments worldwide have established
technology communications platforms such as Chinas International Mobile
Telecommunications-2020 (IMT-2020) (5G) advance group, Japans 2020 and
beyond ad hoc, Korea has 5G forum and so on.
At present, the International Telecommunications Union (ITU) has defined a
clear 5G work timetable [16]: ITU completed the formulation of 5G vision and key
technical indicators in 2015. In 2016, it presented 5G technology performance
requirements and evaluation methods. In 2017, it begun to collect 5G international
standards. And by the end of 2020, it will complete technical specifications of 5G.
The 3rd Generation Partnership Project (3GPP), as an international mobile com-
munications standardization organization, has identified 5G research plan and
initiated the research on requirements for 5G technology in 2015, and started 5G
technology specialized seminar in the second half of 2015. It plans to complete the
study stage of 5G technology scheme from 2016 to 2017, and complete the
formulation of 5G technology specification Release 14 from 2018 to 2019.
IEEE 802.11 standardization group, in order to meet the massive business in hot
spot and indoor areas in the future, intends to support 5GHz unlicensed band
transmission in 802.11ax, meanwhile, use orthogonal frequency division multiple
access and interference compensation technology, which can make the peak trans-
mission rate up to 10Gbps, and Wireless Local Area Network (WLAN) perfor-
mance more robust. According to the IEEE 802.11 working groups plan, it is
expected to complete 802.11ax standard specification in 2018.
1.3 5G Candidate Technologies
In order to meet the wireless service growth needs of 1000 times in the next ten
years, the capacity of the wireless networks can be expanded from three directions
in rising the spectrum utilization, enhancing the spatial multiplexing and bandwidth
expansion. For example, it can effectively improve spectrum efficiency, throughput
per unit area and power efficiency through deploying ultra-dense small base station,
and at the same time shorting the distance between wireless access networks and
terminals. It can greatly increase 5G systems available frequency bandwidth
through expanding the use of unlicensed spectrum, high frequency band and
millimeter wave band. And through massive MIMO it can further tap the potentials
of the space, and greatly improve spectrum utilization. The characteristics and
performance of various wireless candidate technologies vary greatly, and we will
briefly introduce the several typical 5G candidate technologies in the following.
1.3.1 Ultra-Dense Wireless Network
How can we effectively increase the capacity of wireless network? A simple and
effective way is to decrease the coverage area of each cell, and to deploy the cellular
network in a dense way to improve the spatial multiplexing of frequency resources
and accordingly enhance the throughput per unit area. Meanwhile, due to the
shortened distance between base station and terminals, transmission loss will be
reduced, thus greatly improving the power efficiency.
Statistics show that in wireless networks, more than 70% wireless data generated
from indoors and hotspots [19]. If we want to significantly improve the network
capacity through utilizing the outdoor macro cells, there will be the two limitations.
On the one hand, due to the scarcity of macro cell site resources, the deployment
density could not be further increased. On the other hand, when the macro cell base
stations realize the indoor coverage, wireless signals will experience severe pene-
tration loss, leading to very poor indoor coverage performance and thus making it
very difficult to meet the business needs of indoor and hotspots. Therefore, a large
number of various types of small access points will be densely deployed, forming
ultra-dense wireless network. These small access points include Home eNodeB
(HeNB), WLAN Access Point (AP), Relay Node (RN), Micro Cell, Pico Cell and
etc. In particular, both Home eNodeB and WLAN access points can be connected to
the operators core network through the cable broadband in a plug-and-play way,
and their deployments are very flexible and convenient.
Since the ultra-dense wireless network access points may be randomly deployed
by the end users according to their own needs, and business needs change more
frequently, the traditional network planning will have to be faced with huge
challenges. According to the deployment scenarios and application requirements,
ultra-dense wireless networks will have multiple features:
Ultra-dense wireless network sites need to have the capability of ad hoc net-
working, such as neighbor cell discovery by itself, physical cell identifier
configuration, access point adaptive activation, carrier adaptive selection and
re-selections.
Because it is very difficult to receive wireless positioning satellite signal of
Global Positioning System (GPS) and BeiDou Navigation Satellite System
(BDS), ultra-dense wireless networks need to realize wireless network synchro-
nization through air interfaces.
Due to access points limited coverage, ultra-dense wireless networks need to
have the capacity to distinguish users moving speed, adaptively connect fast-
moving users and quasi-stationary users respectively with macro coverage and
ultra-dense wireless networks.
In order to enhance the mobile robustness, we can separate the user plane and
control plane, and keep the control plane in the macro cell to reduce the
handover latency and ensure the consistency of user experience.
With relatively small coverage of every access point, high line-of-sight trans-
mission probability, and less path transmission loss, ultra-dense wireless net-
works can make full use of high-frequency band and millimeter wave
transmission, whose spectrum resources are still sufficient.
Since the hotspots and indoor business requirements change frequently and
meanwhile we need to reduce the deployment cost, the ultra-dense wireless
backhaul network can use wireless link, and the spectrum resource of wireless
backhaul and access links can be shared.
In order to better measure and evaluate the cost of ultra-dense wireless networks,
J. G. Andrews et al. [3] defined base station density gain ( > 0). is the gain of
data transmission rate with the increasing density of the network. If we define the
original data transmission rate as R1 , the density of base station is 1BSs/km2.
Through the ultra density networks, the data transmission rate is defined as R2 , the
density of base station is 2BSs/km2. Then the corresponding density gain is:
R2 1
1:1
R1 2
According to this formula, if the network density increases twice and at the same
time, the data transmission rate increases twice, then the density gain is 1.
1.3.2 Large Scale Antenna Technology
In practical scenarios, the point-to-point multiple antenna technology needs multi-

ple antennas to be deployed on terminals with limited physical size, making it hard
to effectively improve the system spectrum utilization. Therefore, downlink Multi-
User MIMO (MU-MIMO) and uplink Virtual MIMO (V-MIMO) are introduced to
overcome the limitations of multiple antennas, which can effectively improve the
system spectrum utilization. On the other hand, multiple antennas technology has
evolved from passive to active, from Two-Dimensional (2D) to Three-Dimensional
(3D), from high order MIMO to the development of large-scale array, and the
spectrum efficiency will be expected to increase by dozens of times or even higher.
Due to the introduction of the active antenna array, the base station side can support
128 cooperative antennas, or even a larger array of antennas. In addition, the
original 2D antenna array expanded to 3D antenna array, the formation of a novel
3DMIMO technology can support multi-user smart beamforming for reducing
user interference. It can be predicted that massive MIMO, combined with
technologies of high frequency band and millimeter wave, will further improve the
performance of wireless signal coverage [21].
Since the proposal of massive MIMO, it got immediate attention from academia
and industry. Operators, equipment manufacturers and research institutions all
showed great interest and made a series of achievements [2124]. From 2010 to
2013, led by Bell Lab, Lund University and Linkoping University in Sweden, Rice
University in the U. S. and many other research institutions, the international
academia has made extensive exploration of massive MIMO channel capacity,
transmission, Channel State Information (CSI) acquisition and testing.
Massive MIMO has certain advantages in technology, for example:
Assuming that channels are independent of each other, since the base station
uses a large number of transmit and receive antennas, the transmitter or receiver
based on filter matching can effectively suppress multiuser interference. There-
fore, the optimal performance can be achieved via linear complexity greatly
improving system spectrum utilization.
The narrow beam formed by massive MIMO can greatly improve the energy
efficiency.
Correspondingly, the research on massive MIMO is also faced with some
challenges:
The pairing freedom has greatly increased between downlink multiuser MIMO
and uplink virtual MIMO user, and it needs to design a realizable and high
performance radio resource scheduler;
In the uplink transmission of large scale antenna system, a large number of users
will be multiplexed. Under the limited condition of bandwidth and orthogonal
code sequence, pilot channel will be polluted, which will affect the performance
of uplink receiver.
1.3.3 Millimeter Wave Communications Technology
According to the latest spectrum requirements researches [20], in 2020 the worlds
incremental spectrum requirements will be 10002000 MHz while the low fre-
quency resource has been largely depleted. Compared with the deployed low
frequency band, the available frequency resources of millimeter wave band
(30300 GHz) are quite abundant, which is about 200 times of the low frequency
band. Therefore, the industry began to explore how to use the millimeter wave band
(30300 GHz) in wireless communications. Millimeter wave band has always been
considered to be not suitable for wireless transmission, due to relatively large path
loss, absorption by atmosphere and rain, relatively poor capacity of diffraction, big
phase noise, high cost of measurement equipment. However, with increasingly
developed semiconductor technology, the cost of equipment has been declining
rapidly. At present, in satellite communications, Local Multipoint Distribution
Services (LMDS), cellular communications backhaul link and wireless transmis-

sion standard 802.11ad, the millimeter wave band has been used.
Because millimeter wave has relatively small wave length, at least more than
10 times of antennas can be deployed in the same region, which is very suitable for
building massive Multiple Input and Multiple Output (MIMO), greatly improving
the channel capacity. However, due to the propagation characteristics of millimeter
wave band, its application in cellular wireless communications has several
challenges:
The transmission loss of millimeter wave is proportional to the square of the
frequency, and it needs to be compensated by the high gain beamforming.
Millimeter wave transmission is very sensitive to the object occlusion, and the
transmission loss is very large.
When line-of-sight transmission path is existed, the transmission quality of
millimeter wave is very high, but the transmission path is longer, the probability
of line-of-sight transmission is lower.
Channel fading coherence time is proportional to the carrier frequency, and the
millimeter wave frequency is very high, which results in that the small scale
fading of channel is very fast in the same moving speed conditions compared
with the low frequency.
The latest channel measurement results of 28 and 73 GHz show that the
millimeter wave transmission can be effectively in non-line-of-sight environments
[25, 26]. Due to the special propagation characteristics, in addition to line-of-sight
models and non-line-of-sight models, a third state is proposed to explicitly model
the possibility of outages. That is to say, the three-state millimeter wave transmis-
sion, including line-of-sight, non-line-of-sight and outage, can objectively reflect
the path loss rule of the millimeter wave [26, 27]. Outage state is a good description
of the characteristics of the object occlusion and penetration of millimeter wave.
In [26], the measurement data analysis shows that the probability of line-of-sight
transmission and non-line-of-sight transmission is closely related to the transmis-
sion distance, and the probability functions for the three states are given as
Eq. (1.2).

Pout d max 0; 1 eaout dbout
PLOS d 1 pout d ealos d 1:2
PNLOS d 1 pout d pLOS d
where aLOS , aout and bout are the parameters decided by the measurement data of
typical network deploying scenarios. When the terminal is close to the base station,
millimeter wave transmission has close-to-zero probability of outage state, and it
mainly relies on line-of-sight transmission. When the terminal is moving away from
the base station, the millimeter wave transmission is in the state of line-of-sight
transmission or non-line-of-sight transmission. When the terminal is far away from
the base station, millimeter wave transmission is in non-line-of-sight transmission
or outage states. Of course, line-of-sight and non-line-of-sight transmissions have

different corresponding path loss models. The model is very similar to the relay
path transmission model, so it can be considered as an extension of relay path
transmission model.
1.3.4 Flexible Spectrum Usage
In the process of the rapid development of mobile communications, one of the

bottlenecks is the shortage of spectrum resources. Yet the requirements for radio
frequency spectrum resources in wireless communications have been expanding.
The competition between different radio technologies and applications is becoming
more and more intense, highlighting the increasing shortage of spectrum resource.
Under this circumstance, on the one hand, we need to expand the spectrum
resources like millimeter wave which is abundant but hasnt yet been used in a
large scale. On the other hand, we need to try to improve the spectrum utilization of
existing resources. The flexible use of spectrum resources is an effective way to
improve the efficiency of spectrum utilization. Flexible spectrum usage includes the
spectrum refarming in operator itself [28], LTE use Unlicensed band (LTE-U) [29],
peer spectrum sharing between operators [30] and other potential ways.
Firstly, the spectrum refarming in operator itself is mainly based on the same
operator with different modes to share frequency resource. For example, with the
extensive deployment of LTE network sites, 2-Generation wireless telephone
technology (2G) and the 3rd Generation mobile communications system
(3G) users will gradually migrate into the LTE network, and the spectrum utiliza-
tion of 2G and 3G networks will have lower efficiency. It can be predicted that
when 5G network deployment is completed in the future, there will be more users
migrating into 5G network. Then 5G network can exchange the information
through 2G, 3G and LTE networks and obtain user distribution and spectrum
usage in 2G, 3G and LTE, and then determine how to use the spectrum resources
strategy of 2G, 3G and LTE network, and actively control interference between
networks.
Secondly, the sharing of LTE-U spectrum resources mainly refers to how the
cellular network use unlicensed band. WLAN plays a very important role in
LTE-U. Sharing of unlicensed frequency band between cellular networks and
WLAN has two possible ways: (1) LTE Assisted Access (LAA) [31], i.e., cellular
network configures the carrier of LTE-U as auxiliary carrier though Carrier Aggre-
gation (CA), cooperating with main carrier of licensed band to provide service for
users. (2) Independent LTE-U communications, which is that cellular network only
uses unlicensed bands for communications. Compared with WLAN, since cellular
network uses technique of turbo codes, Hybrid Automatic Repeat reQuest (HARQ),
adaptive modulation and coding and frequency selective scheduling, it can achieve
greater coverage and higher network throughput in unlicensed band.
Finally, in peer spectrum sharing between operators, operators can jointly

purchase spectrum, and then fairly share spectrum resources, or composite spec-
trum sharing resource pool through holding out part of licensed spectrum which is
owned by a plurality frequency license holders. Multiple operators need to follow a
certain negotiated spectrum sharing strategy, exchange spectrum using information
and form the spectrum use database so that the spectrum can be used in a dynamic
way in space and time dimensions. This sharing way can not only improve the
spectrum utilization, but also ease the operators financial pressure in purchasing
spectrum.
1.3.5 Waveform and Multiple Access
Although 4Gs physical layer uses the core technology based on Orthogonal
Frequency Division Multiplexing (OFDM), but the OFDM waveform itself has
some defects, for example: (1) Slow rolling-off leads to high out-of-band leaking.
Therefore, larger spectrum guard interval is needed. (2) To avoid interference
between carriers, (coarse) synchronization is needed between nodes transmission.
In the hierarchical network, base stations with different coverage should be syn-
chronized. Because the application scenarios of 5G is far more complex than 4G,
and its latency and the number of access requirements are very strict. However,
OFDMs slow rolling-down characteristic and strict synchronization requirement
cannot adapt to real-time business asynchronous fast access and efficient use of
non-continuous spectrum.
Targeting at the disadvantages of OFDM, a lot of alternative waveforms have
been proposed, such as Filter Bank based Multi-Carrier (FBMC), Filtered OFDM
(F-OFDM) and Universal Filtered Multi-Carrier (UFMC) [3], etc. By filtering the
sub-band or sub-carrier, the rate of spectrum rolling-off is increased and the
frequency spectrum leakage is reduced, so that the time frequency synchronization
requirement is reduced, the frequency protection band and the time domain pro-
tection interval is removed. The above new waveform technologies can be well
combined with OFDM and MIMO, and increase the flexibility of 5G air interface
design, so as to match different traffic latency and data rate requirements.
Mobile Internet and IoT are the driving force of the development of 5G, and the
requirements for various applications greatly vary. For example, various types of
real-time traffic transmission in the 5G network have set down requirements for
millisecond order end-to-end latency. Undoubtedly, such a strict traffic latency
requirement will pose very high requirements on the physical layer design (includ-
ing symbol duration, synchronization process, random access and frame structure,
etc.). In addition, the traditional cellular communications establish and complete
synchronization based on connection, and they schedule with the form of the
network resource. That is to say, firstly the end-to-end connection is established,
and then data is transmitted. In IoT applications led by machine communications,
wireless sensor nodes are mainly energy constrained, and transmission data packets
are usually quite small. Hence, the requirements are very high for traffic latency and
energy efficiency. If we still use the traditional cellular communications method
based on the connection, it will bring a lot of unnecessary network expenditure and
lead to too long activation time of wireless sensor node, which is not good for
reducing energy consumption. In order to support fast access to businesses and
simplify the channel accessing process and signaling process, 5G has proposed non
orthogonal multiple access method, which is to use the modulation, spread spec-
trum, power and space dimension for joint mapping so that users may transmit data
without being scheduled, and the spectrum utilization can be effectively improved.
1.3.6 Device to Device
Device to device (D2D) means that two terminal devices directly communicate
without base station and the core network. Its typical applications include cellular
assisted D2D communications and Vehicle to Vehicle (V2V) direct communica-
tions. Cellular assisted D2D communications is a new technology that allows
terminals to multiplex cell resource for direct communications under the control
of a cellular system.
D2D communications can increase the spectrum efficiency of the cellular com-
munications system, reduce the terminal transmitting power, and to a certain extent,
solve the scarcity of spectrum resources in the wireless communications system. In
addition, it can bring additional benefits including: reducing the burden of cellular
network, reducing the mobile terminal battery power consumption, increasing the
transmission rate and improving the robustness of network infrastructure failures,
whats more, supporting the new way of point-to-point data services in a small area.
D2D communications technology is also faced with the following problems:
Since the D2D communications will multiplex cellular resources, the mutual
interference between D2D communications and cellular communications arises.
When D2D communications multiplex uplink resources, in the system it is the
base station that is interfered by the D2D communications while the base station
can control the interference by adjusting the transmitting power of D2D com-
munications and the multiplexed resources. When D2D communications multi-
plex downlink resources, in the system it is any downlink user that is interfered
by the D2D communications. The interference is not controllable, which may
result in the link failure.
The synchronous mode of D2D communications can improve energy efficiency.
When the terminal is in the coverage of base station, the terminal can make the
base station as a time synchronization source, so as to realize the synchronous
D2D communications mode. When the D2D communications terminals are not
in the coverage of the base station, we can select a terminal as the cluster head
for D2D communications to transmit synchronous signals. When there is a
multi-hop in D2D communications, it will lead to multi-hop transmission of
1.4 Challenges in 5G Testing and Evaluation 13
synchronization time, resulting in multiple synchronization sources. Alterna-

tively, when multiple D2D communications groups coexist with each other, it
can lead to multiple synchronization sources. The problem of multiple synchro-
nization sources also needs to be solved in the research of D2D communications
technology.
1.4 Challenges in 5G Testing and Evaluation
1.4.1 Challenges Posed by New Technologies to Testing

and Evaluation
According to the network requirements and technical indicators defined in 5G white

paper [16], 5G network system will achieve huge growth in network capacity and
connected user number. Meanwhile it put forward the new network vision and design
principles based on network maintenance, energy efficiency, user experience and many
other aspects. In order to reach all kinds of technical indicators of the 5G network, the
research on new network architecture and several types of candidate technologies is
going on intensively. Compared with 4G network, 5G is not only more flexible in the
network architecture, but also more complex in signal processing mechanism used in
transmission system, and more multi-dimensional in performance evaluation indica-
tors. There is no doubt that both the proposed new network architecture and the
emergence of various types of transmission technology will pose new challenges to
5G air interface technology standardization, program design and simulation.
For physical layer transmission technology, 5G system will introduce new
waveform and non-orthogonal multiple access at the physical layer to achieve
the required traffic latency in air interface. But in order to overcome the
interference caused by the non-orthogonal transmission technology, we need
to introduce nonlinear receiver, and in the performance evaluation we need to
consider the effect of user pairing and multi-dimensional parameters, which will
greatly increase the difficulty of physical layer technology link simulation and
evaluation. Also in order to simulate and evaluate the system of new wave and
non-orthogonal multiple access, we also need to establish mapping method of
the nonlinear receiver link in system.
In order to further explore the spatial freedom and improve the network through-
put, 5G will introduce massive MIMO technology, which will greatly enhance
the baseband signal dimensions, and make the link simulation evaluation com-
plexity increase by thousand times. And in simulation evaluation system, mas-
sive MIMO and MU-MIMO technology will greatly increase computational
interference complexity. And multi-channel synchronous, isolated, multi-
channel data storage pose severe challenges to the test implementation.
The new channel propagation model will be introduced based on high frequency
band transmission technology, D2D technology and massive MIMO technology.
On the one hand, how to build a more accurate channel propagation model for
the 5G candidate technologies will pose challenges for corresponding testing
and verification. On the other hand, the new channel model based on expanding
the traditional channel model will bring the increased types and dimensions of
parameters, which will lead to increased computation complexity and storage
space for the link simulation and system simulation evaluation of the above
technologies.
In terms of the wireless network technology, the user plane and control plane are
separated, and the wireless network resources are controlled and optimized in a
centralized way, which will lead to a sharp increase in the space of the feasible
solution of control and optimization. Through multi-core Central Processing
Unit (CPU), Graphics Processing Unit (GPU) and Field Programmable Gate
Array (FPGA) computing resources, we can build a heterogeneous computing
platform, and distribute the above computing resources for the baseband signal
processing operations which have higher calculation complexity. We need to
design scheduling algorithm for heterogeneous computing resources, accurately
estimate the consumed time of heterogeneous computing and interface data
transmission, and meanwhile design the synchronized mechanism for computing
tasks to make full use of the heterogeneous computing platform.
1.4.2 Four Elements of Testing and Evaluation

Requirements
All in all, to meet the requirements for realness, comprehensiveness, rapidity and
flexibility is an important challenge for the evaluation of 5G testing and evaluation.
Realness
In the 5G testing and evaluation, the requirement for realness is reflected in the 5G
wireless channel model, verification methods, user experience, and other aspects.
From the point of establishing channel model, describing the channel character-
istics in a numerical way with testing and measurement methods, and providing a
close-to-objective physical channel simulation environment in a real and reproduc-
ible way, are of self-evident importance for 5G communications technology veri-
fication. Compared with 3G/4G system, new application scenarios are added in 5G,
which have the various characteristics of ultra high traffic volume density, ultra
high connection density, ultra high mobility, etc. From the point of the network
topology, various link types in 5G communications system will coexist in a same
region, extending from the traditional macrocell and microcell, to picocell,
femtocell and nomadic base station, and support D2D, Machine to Machine
(M2M), V2V and the fully-connected network. In the complex and diversified
network architecture, traditional channel model does not give sufficient consider-
ation to the consistency of small and medium scale parameter space, which may
lead to the exaggerated simulation performance of in technological evaluation (for
example, MU-MIMO) [32]. In addition, two-way mobility of D2D/V2V will
introduce Doppler model which is different from the traditional model, and mas-
sively intensive scattering exists in both transmitter and the receiver and the
stationary cycle is short, all of which need to be considered in the channel model.
Under the condition of high mobility, the features of fast time-varying, more severe
fading and mixed propagation scenarios make it difficult for the current channel to
be well applied in 5G high speed mobile scenarios. Therefore, wide-range propa-
gation scenarios and diversified network topologies have posed challenges to the
5G channel model. At the technological level, a variety of networks and transmis-
sion technologies emerge in the 5G research, which pose the following challenges
to the channel model: unique transmission characteristics of radio waves in higher
frequency and bandwidth, including non-line-of-sight path loss, high resolution
angle characteristics and outdoor mobile channel; in the greater antenna array
(including antenna unit number and the physical dimensions), the original plane
wave propagation model assumption is no longer applicable, and scattered clusters
non-stationary feature is reflected not only in the time axis, but it s also changing
along the array. In the third chapter, we will describe how to build a real and reliable
5G channel model from the aspects of the characteristics of the channel model,
measurement methods and parameters extraction and data analysis.
In terms of the verification methods, according to the traditional methods, the
system design of wireless link in the mobile communications field is mainly based
on the software simulation methods to verify and evaluate the algorithms. However,
there is a certain gap between the system environment under the pure software
simulation and the real situation. First of all, the software simulation assumes that
the hardware design can perfectly realize the software algorithms, and thus we
cannot introduce the impact of hardware conditions on the communications system.
However, in reality algorithm design is often restricted by the hardware conditions,
and the software algorithms are often greatly reduced when realized in hardware.
For example, in the actual hardware environment, the algorithms with particularly
large amount of computation and extremely high calculation accuracy requirements
are often unable to obtain the optimal performance. And the precisely designed
algorithms are often vulnerable to the outside interference. The algorithms at the
front of transmitter and receiver have to consider power, the impact on other
hardware components and many other factors. If we only use software simulation
for system verification and evaluation, its easy to ignore the above problems,
which will make the designed algorithm design stay in the theoretical stage and
cannot meet the needs of the real situation. Secondly, in the former software
simulation systems, simplified methods are often introduced in network and chan-
nel modeling, simulation business abstraction, design and realization and other
processes for convenience of processing. These simplified methods usually reduce
the simulation computational cost at the cost of authenticity. Thirdly, accurate math
modeling of the complex time-varying nonlinear system is very difficult. No matter
how good the model is, it cannot replace the real scenarios.
In order to more accurately reflect the impact of the actual link process on the
system simulation, we can use a combination of hardware and software, using real
hardware platform, deploying the link to be simulated in the hardware (such as
FPGA, channel simulator) for real-time calculation, and thus improving the calcu-
lation accuracy through hardware simulation; in addition, introducing hardware in
the loop technology in 5G testing, can not only get closer to the real environment,
increase the reliability of the simulation test, but also test the communications
equipment under extreme conditions, which is difficult to achieve in the laboratory
environment. For the corresponding contents, please refer to Chaps. 5 and 6 of this
book. Furthermore, in terms of environmental authenticity, in the process of
research and development of 5G key technologies, in addition to all kinds of
simulation and laboratory evaluation, we also have to verify the technological
feasibility through outfield experiments, and realize the verification of the tested
technology or prototype equipment in real scenario through algorithm realization
! data acquisition ! system optimization ! outfield verification, providing field
basis for standardization. Therefore, in Chap. 8 of this book, we will analyze the
planning for test outfield from the perspective of the diversity of application
scenarios, integration of heterogeneous networks and large scale users.
Comprehensiveness
The demand for testing and evaluating comprehensiveness mainly involves two
aspects: on the one hand, it lies in the comprehensive support for the evaluation of
5G performance indicators; on the other hand, it lies in the comprehensive support
for the diversification of candidate technologies. 5G evaluation indicators system
will include the objective indicators like transmission and network as well as the
user experience indicators. Transmission is mainly reflected in the physical layers
transceiver performance, link layer performance indicators, while the network layer
is reflected in the evaluation indicators for system performance. In evaluating the
performance of 5G system, we need to not only comprehensively measure all kinds
of KPI, but also consider the relationship between each indicator, and guarantee
users consistent experience in pace, time, network, business, and other dimensions.
5G candidate technologies can be divided into two categories of air interface
technology and network technology. The air interface technology includes massive
MIMO, full duplex, novel multiple access and waveform, high frequency range
communications and spectrum sharing and flexible using. Network technologies
include Cloud Radio Access Network (C-RAN), Software Defined Network/Net-
work Functions Virtualization (SDN/NFV), ultra-dense network, multi-RAT and
terminal direct communications, etc. As for 5G software simulation technology,
this book will, from the architecture level, element level and module level, give a
detailed analysis on various 5G candidate technologies, propose and implement the
corresponding resource model and design scheme, and build a comprehensive
performance evaluation system, which will be shown in detail in Chap. 4.
Rapidity
Requirement for rapid test performance is mainly derived from the huge growth in
5G network traffic and a variety of performance. In order to meet the KPI s in 5G
white paper, the computing performance of the simulation system must grow by
more than 1000 times before meeting the requirements of the timely evaluation of
simulation task. With such rapid growth in computational performance, a system-
atic and brand new design and realization are needed in simulation systems
hardware platform, software platform, and simulation application. It can be
predicted that within a period of time, multi-core and Many Integrated Cores
(MIC) parallel computing mode will become the main technical means to enhance
computational efficiency, and cloud computing, super computer will gradually
become the calculation means of simulation system. For the simulation system,
the key problem is how to complete the software system concurrent design and
coding implementation on the new and powerful hardware platform with powerful
computational capabilities.
Flexibility
Affected by flexible network architecture, network resource virtualization manage-

ment, parallel computing needs and other factors, simulation and validation system
needs to have enough flexibility. Architecture design, module design, and interface
design have the characteristics of coupling, modularization, interface expansion,
easy integration and so on, which have posed higher requirements for the network
resources model design, software function module design, network element and the
inter module interface design and other aspects.
In this book, Chap. 4 will introduce, from the perspective of the application of
multi-core simulation technology and parallel processing, elaborate on how to deal
with the challenges in rapidity and flexibility software simulation testing
verification requirements, so as to implement 5G link-level simulation and
system-level simulation. Furthermore, Chaps. 5 and 6 of this book, from the
perspective of software and hardware co-simulation and hardware platform of
software defined radio, will describe how to use these two technologies to enhance
5G key technology verification and evaluation.
From the point of system comprehensiveness, this book presents the 5G test
system covering 5G channel model (Chap. 3), software simulation system
(Chap. 4), software and hardware co-simulation technology (Chap. 5),
hardware platform of software defined radio (Chap. 6), and the actual outfield
testing and verification system (Chap. 8), displaying various technological
means of the whole 5G testing and verification system from all-round angles.
Relationship between chapters and the four elements of evaluation are shown as
Fig. 1.2.
Realness Comprehen- Rapidity Flexibility

siveness
Chapter 3 Chapter 5 Chapter 6 Chapter 7

Chapter 4 Software and
Channel Measurement Hardware Co-
and Modeling software simulation simulation Hardware Platform Field Trial Network
Fig. 1.2 Relationship between chapter setting and the four elements of evaluation
1.5 Summary
Starting from 5G application requirements, this chapter firstly introduces the typical
5G application, deployment scenarios, key technical indicators, as well as the plans
and progress of global 5G research and development. After providing a brief
overview of various 5G candidate technologies and their characteristics and ana-
lyzing the challenges posed by these candidate technologies to 5G testing and
verification, we expounded the challenges brought by the four elements require-
ments for realness, comprehensiveness, rapidity and flexibility in testing and
evaluation, as well as the relationship between chapter setting and these four
elements of evaluation, playing a leading role in outlining the subsequent chapters.
References
1. C. Wang, F. Haider, X. Gao, et al. Cellular Architecture and Key Technologies for 5G Wireless
Communication Networks. IEEE Communications Magazine, 2014, 52(2):122130.
2. X. Chu, D. Lopez-Perez, Y. Yang, and F. Gunnarsson. Heterogeneous Cellular Networks:
Theory, Simulation and Deployment. New York, USA: Cambridge University Press, 2013.
3. J. G. Andrews, S. Buzzi, W. Choi, et al. What Will 5G Be? IEEE Journal on Selected Areas in
Communications, 2014, 32(6):10651082.
4. IMT-2020 (5G) Promotion Group. White Paper on 5G Vision and Requirements, 2014. http://
www.imt-2020.cn/zh/documents/download/1.
5. ITU. (2015). Recomendation ITU-R M.20830: IMT VisionFramework and overall objec-
tives of the future development of IMT for 2020 and beyond. Technical report, ITU-R.
6. J. Yang, Y. Qiao, X. Zhang, et al. Characterizing User Behavior in Mobile Internet. IEEE
Transactions on Emerging Topics in Computing, 2015, 3(1):95106.
7. W. Huang, Z. Chen, W. Dong, et al. Mobile Internet big data platform in China Unicom.
Tsinghua Science and Technology, 2014, 19(1):95101.
8. L. Xu, W. He, S. Li. Internet of Things in Industries: A Survey. IEEE Transactions on
Industrial Informatics, 2014, 10(4):22332243.
9. Z. Sheng, S. Yang, Y. Yu, et al. A survey on the IETF protocol suite for the internet of things:
standards, challenges, and opportunities. IEEE Wireless Communications, 2013, 20(6):9198.
10. C. Perera, A. Zaslavsky, P. Christen, D. Georgakopoulos. Context Aware Computing for The
Internet of Things: A Survey. IEEE Communications Surveys & Tutorials, 2014, 16
(1):414454.
References 19
11. C. Perera, C. H. Liu, S. Jayawardena, C. Min. A Survey on Internet of Things from Industrial
Market Perspective. IEEE Access, 2014, 2:16601679.
12. A. Zanella, N. Bui, A. Castellani, et al. Internet of Things for Smart Cities. IEEE Internet of
Things Journal, 2014 , 1(1):2232.
13. C. Tsai, C. Lai, M. Chiang, L. T. Yang. Data Mining for Internet of Things: A Survey. IEEE
Communications Surveys & Tutorials, 2014, 16(1):7797.
14. IMT-2020(5G)Promotion Group. White Paper on 5G Wireless Technology Architecture,
2015. http://www.imt-2020.cn/zh/documents/download/61.
15. Cisco, Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update,
20142019. White Paper, 2015.
16. IMT-2020(5G)Promotion Group. White Paper on 5G Concept, 2015. http://www.imt-2020.cn/
zh/documents/download/23
17. A. Osseiran, V. Braun, T. Hidekazu, et al. The foundation of the Mobile and Wireless
Communications System for 2020 and beyond Challenges, Enablers and Technology Solu-
tions. IEEE Vehicular Technology Conference, 2013:15.
18. A. Osseiran, F. Boccardi, V. Braun, et al. Scenarios for 5G mobile and wireless communica-
tions: the vision of the METIS project. IEEE Communications Magazine, 2014, 52(5):2635.
19. Qualcomm, The 1000x Mobile Data Challenge, 2013. https://www.qualcomm.com/docu
ments/1000xmobile-data-challenge.
20. T. Wang, B. Huang, J. Pang. Current Situation and Prospect of Spectrum Requirements
Forecasting of the Future IMT System. Telecom science, 2013, 29(4):125130.
21. G. Zhen, D. Ling, M. De, et al. MmWave Massive MIMO based Wireless Backhaul for the 5G
Ultra-Dense Network. IEEE Wireless Communications, 2015, 22(5):1321.
22. E. Larsson, O. Edfors, F. Tufvesson, and T. Marzetta. Massive MIMO for next generation
wireless systems. IEEE Communications Magazine, 2014, 52(2):186195.
23. J. Shen, J. Zhang, K. B. Letaief. Downlink User Capacity of Massive MIMO Under Pilot
Contamination. IEEE Transactions on Wireless Communications, 2015, 14(6):31833193.
24. L. Lu, G. Y. Li, A. L. Swindlehurst, et al. An Overview of Massive MIMO: Benefits and
Challenges. IEEE Journal of Selected Topics in Signal Processing, 2014, 8(5):742758.
25. G. R. MacCartney, T. S. Rappaport. 73 GHz millimeter wave propagation measurements for
outdoor urban mobile and backhaul communications in New York City. Proceedings of IEEE
International Conference on Communications, 2014: 48624867.
26. M. R. Akdeniz, Y. Liu, M. K. Samimi, et al. Millimeter wave channel modeling and cellular
capacity evaluation. IEEE Journal on Selected Areas in Communications, 2014, 32
(6):11641179.
27. A. Ghosh, T. A. Thomas, M. C. Cudak, et al. Millimeter-wave enhanced local area systems: a
high-data-rate approach for future wireless networks. IEEE Journal on Selected Areas in
Communications, 2014, 32(6):11521163.
28. Nokia, Optimizing mobile broadband performance by spectrum refarming. White paper,
2014., http://networks.nokia.com/file/35861/optimizing-mobile-broadband-performance-by-
spectrum-refarming.
29. F. M. Abinader, E. P. L. Almeida, F. S. Chaves, et al. Enabling the coexistence of LTE and
Wi-Fi in unlicensed bands. IEEE Communications Magazine, 2014, 52(11):5461.
30. B. Singh, S. Hailu, K. Koufos, et al. Coordination protocol for inter-operator spectrum sharing
in co-primary 5G small cell networks. IEEE Communications Magazine, 2015, 53(7):3440.
31. H. Zhang, X. Chu; W. Guo, et al. Coexistence of Wi-Fi and heterogeneous small cell networks
sharing unlicensed spectrum. IEEE Communications Magazine, 2015, 53(3):158164.
32. METIS. Deliverable D1. 4: METIS Channel Models, 2015. https://www.metis2020.com/wp-
content/uploads/METIS_D1.4_v3.pdf.
Chapter 2
Evolution of Testing Technology
Testing and evaluation are indispensable links of technology and product inspec-
tion. The final birth of every new technology or product must go through strict
testing and evaluating process. In the 5G era, the testing tools and evaluation
methods are faced with new challenges along with the continuous development of
the new 5G technologies. In the section of testing technology evolution of this
chapter, we will give a brief review of the history of testing technologies, and make
analysis on the development trend of the testing technology ecosystem according to
the existing technologies. In the section of testing technology challenges, based on
the characteristics of wireless communications technology and 5G technology, we
will sort out and analyze the challenges of wireless communications testing tech-
nologies in terms of overall performance, the number of RF channels, and high
throughput, etc.
2.1 The Importance of Testing Technology
Testing means measurement, inspection and test. In different application fields,

concepts of testing are different, but they all play important roles. For example, in
the medical field, testing is an important step and indicator to determine whether
a person is healthy or not; in the teaching area, testing is an important criterion to
review peoples knowledge and skills, etc. Testing in this book means to measure
and test the performance and precision of science, technology and equipment, and
lay more emphasis on the examination and verification of the new wireless com-
munications technology.
In the field of wireless communications, testing technology plays a decisive role
in the entire industry. When it comes to the new technological revolution, Acade-
mician Xuesen Qian said, The key technology of new technology revolution is
information technology. Information technology is composed of three parts, namely
measurement technology, computer technology and communications technology,

DOI 10.1007/978-3-319-61869-2_2
22 2 Evolution of Testing Technology
among which measurement technology is the key and basis. [1] Computer, com-
munications and instruments are three internationally recognized information tech-
nologies and they are the industries that have achieved the fastest development
since the twenty-first century. Technological development of different fields usu-
ally mutually supports and promotes each other. Thanks to the advance of computer
and communications technology, measuring instruments have achieved rapid
development and changes in the past 30 years, which will be described in subse-
quent chapters of this book. As one of the three key technologies of the information
industry, testing and measuring technology has become the foundation and devel-
opment guarantee for the electronic information industry. Measuring instrument is
a countrys strategic equipment, whose development level has become a sign of
national scientific and technological level, comprehensive national strength and
international competitiveness. In communications, radar, navigation, electronic
countermeasure, space technology, measurement and control, aerospace and
many other fields, the electronic measuring instrument is indispensable technical
equipment. In modern manufacturing industry, the advanced instruments are
needed in research and design, production process control, and product testing as
a support to testing technology.
Instruments are the realization carrier of testing technology. Moreover, testing is
an important step for checking the entire products quality, and also an important
basis to help the researchers to improve the product. Any new technology or
product must be fully tested and evaluated before entering the industrial production
and becoming a standard technology or qualified product in the market. Testing
technology needs to run through the entire process, from the pre-design stage to
design stage to production completion stage. Take the production of mobile phones
as an example. Before the development and production, engineers need to test all
kinds of components related to the phone. During the development and production,
they need to take qualitative or quantitative tests of the machining parts, semi-
finished mobile phone and all kinds of mobile phone parameters. And after the
production completed, the comprehensive tests of mobile phones functionality,
performance and reliability are needed, such as mobile phones radio frequency
conformance test, WLAN test, GPS test and so on, so as to judge whether a phone
meets various protocol standards and authority requirements. Different tests are
taken to determine whether mobile phone is up to the standard or not. After
organizing and analyzing the test data, if the testing shows that it does not meet
the standards, then it will go back to optimize the performance and be tested again.
The test data can be used to assist the researchers with performance optimization
and problem solving. The product should not enter the market to become a qualified
mobile phone until the test results finally meet the standards. The quality and
efficiency of products are restricted by the quality and efficiency of testing tech-
nology. Only the accurate and efficient testing process can guarantee more efficient
production of high performance products for the fast-updating terminal market
nowadays. Take the software testing as another example. Engineers run or test
software manually or automatically to find bugs or defects in the software, to help
researchers in software design optimization and provide quality assurance for
2.2 Testing Technology Evolution 23
software. The process of compiling software is the process of testing, debugging

and optimizing software, and the software testing is throughout the entire period of
the software definition and development. Testing technology plays an important
role in all stages of a product or technology, from idea budding to prototype to
standardization. The testing and measurement solutions of each stage are important
guarantees for the successful application of new technologies and new products.
The development of testing technology is mutually enhanced with research and
application of new technologies and new products. It provides an important testing
tool for the development of new equipment while testing instruments are indis-
pensable tools for testing technology. Only when equipped with powerful instru-
ments can we turn a lot of theories into the reality. Now the development of test
instruments is more and more going toward the direction of multi-interface, multi-
function, intelligence, high density, and high speed. Some instrument manufac-
turers even join equipment manufacturers in the standards development and equip-
ment manufacturing, which greatly reflects the importance of testing and
measurement technology [2].
2.2 Testing Technology Evolution
2.2.1 Development of Testing Instruments
Instruments are combinations of various sciences and technologies with many

varieties and wide applications. By the intended purpose and usage, instruments
can be divided into measuring instruments, radio test instruments, automobile
instruments, aircraft instruments, navigation instruments, geological survey instru-
ments, time measuring instruments, teaching instruments, medical instruments and
many other types [3]. The testing instruments in this book are all related to the field
of wireless technology, falling into the category of radio test instruments.
From the mobile phones with simple call function to the multimode smartphones
and netbooks with various functions like GPS, Wi-Fi, Bluetooth, payment function,
and digital television, personal mobile terminals have more and more functions and
applications with more and more complex performances. Accordingly, to ensure
more reliable performances and more powerful integrated functions, the mature
testing equipment and advanced and reliable testing methods are needed. Powerful
testing instruments are indispensable to the stable and reliable running of new
wireless communications equipment [4].
Going through the history of testing instruments, we can take the beginning of
automated testing technology and network-based remote services as critical points.
By far, the development of test instruments can be roughly divided into three eras:
Instrument 1.0 Era, which means the times of traditional stand-alone and pure
manual testing instruments; Instrument 2.0 Era, which is the software radio era
trigged by the technology progress of computer and communications; and
Instrument 3.0 Era, which has introduced interconnected network testing and
evolved from simple testing instruments to the integration of testing equipment and
services. At present, the testing industry is still in Instrument 2.0 Era and the early
stage of Instrument 3.0 Era. Whether changes will take place in the development of
instruments in the future and whether the change will be fast or slow all need our
technical workers continued joint efforts. The development trend of testing tech-
nologies will be described in the subsequent chapters. It should be noted that the
division of different eras is not absolute. Instruments of another era will not
immediately disappear when new technology emerges. Instead, they are more likely
to continue to progress with the technology development or be replaced and stop
production after a certain length of time. The renewal of old and new instruments is
a process of coexistence and substitution.
The development of testing technologies is actually the development process of
the test instruments. This section will lead readers to review the development and
evolution process of the testing instruments. The main realization carrier of testing
technologies is the testing instruments. Its easier for the readers to understand the
development of testing technologies in the corresponding periods with the intro-
duction to the development of testing instruments.
Instrument 1.0 Era
Unlike the high-end testing instrument that the test engineers use now, the initial
test instruments were simple and the measurement results were displayed by the
pointer. They had single functions and low openness, and were known as the
analog instruments, such as the analog voltmeter and ammeter. Afterwards,
with the emergence of electronic tube technology, electronic tube instruments
came into being, such as the early oscilloscope, etc. Subsequently along with the
emergence of transistors and integrated circuits, the instrument industry, combining
with integrated circuit chips, produced the instruments based on integrated circuit
chips, which were known as digital instruments. The basic working principle of
the digital instruments is that the analog signals are converted to digital signals in
the measurement process, and the test results are finally displayed and output in the
digital form. Compared with the analog instruments, digital instruments have more
intuitive and clearer test results, faster response, and relatively higher accuracy, but
the instruments then depended on manual operation. In the 1970s, the testing
instruments began to use the microprocessor. Simple intelligent instruments
began to appear in the field. For the first time, the engineers have tasted the
sweetness of intelligence. The introduction of microprocessors has greatly
improved the performance and automation degree of the instruments, facilitating
automatic range conversion, automatic zero adjustment, trigger level automatic
adjustment, automatic calibration, self-diagnosis and many other functions [5].
Instrument 2.0 Era
As time goes by, technologies in different fields are all developed. Owing to the
development and progress of the computer technology, software, data processing,
data bus, and reprogrammable chips, test instruments began to transform from the
functional fixed discrete instruments to flexible instruments based on the software
design. The instrument based on the software design allows users to redefine the
software function according to their own needs. The development of network
technology also experienced the user defined stage of Web2.0. Similarly, we call
this stage of the instruments as Instrument 2.0 Era. Therefore, the previous
traditional instrument times can be called as Instrument 1.0 Era.
In Instrument 1.0 Era, engineers totally relied on hardware to realize the testing
and measurement. It was expert and manufacturing plant who took charge of design
and manufacturing. Although users were allowed to put forward some opinions and
demands, they could not be realized immediately. Moreover, users were unlikely to
participate in the design and manufacture of products. The hardware itself and its
analysis function were defined by the instrument suppliers, and self-definition
function was not provided to users. Even if the instrument was connected to the
computer, the transmitted information waste test results defined by equipment
manufacturers. Users are unable to obtain the original measure data to do self-
defined analysis. During the development of the testing instruments, with the help
of the progress in computer technology and communications technology, the
instrument industry had seen fast development and changes from the traditional
hardware instruments. The emergence of bus technology made automated testing
possible, and the rapid growth of the bus technology led to the fast development of
the automated testing. The development of automated testing offers a guarantee for
wide applications of wireless communications products. Traditional testing has
complicated procedures, high operations difficulty and strict requirements on the
test engineers, while automatic testing has overcome those disadvantages and
brings many advantages, such as having testing specifications, high testing speed,
avoidance of maloperations, and being able to guarantee the accuracy and authen-
ticity of the testing results to the greatest extent [6]. Among them, the most
representative one is the proposed concepts of the PCI eXtensions for Instrumen-
tation (PXI) bus technology and Software Defined Radio (SDR), which changed the
single direction of the traditional testing technology to the stage when users could
redevelop the instruments according to their own needs, bringing us to Instrument
2.0 Era, namely, the software defined radio era. In this era, engineers had more
control rights to the instruments. After getting access to the original real-time data,
engineers could use software to design their user interface and define measurement
tasks in order to obtain the desired results.
During this period, the instrument technology had experienced great improve-
ment. The major improvement and change compared with the 1.0 Era mainly
stemmed from and were reflected in the development and emergence of bus
technology, instrument module technology, SDR, hybrid systems and many other
representative technologies. The combination of multiple techniques helped to

create a number of high-performance testing systems that increased the flexibility
of testing and brought higher performance and lower cost.
Bus Technology
Since the birth of the first-generation test bus technology, i.e., General Purpose
Interface Bus (GPIB) technology, in the 1970s, testing has gradually developed
from a single manual operation to a large-scale automatic testing system. After-
wards, all measuring instruments with GPIB standard interfaces sprang up, and the
unified standards of interconnected instruments were gradually formed. The stan-
dards allow testing engineers to assemble a variety of automatic measurement
systems with powerful functions by very convenient means. Meanwhile, with the
passing of time and the improvement of test requirements, the bus technology is not
limited to GPIB anymore. Other bus technologies have been put forward and
improved. The representative bus technologies include VMEbus eXtensions for
Instrumentation (VXI), Peripheral Component Interconnect (PCI), Peripheral Com-
ponent Interface expres (PCIe), PXI, PCI eXtensions for Instrumentation expres
(PXIe), LAN eXtension for Instrumentation (LXI) and Advanced eXtensible Inter-
face expres (AXIe), etc.
Through the test bus, the data communications between different units and
modules in the testing instruments are realized. Thanks to the development of bus
technology, testing is no longer limited to the manual operation of a single instru-
ment and the automated testing becomes possible. This progress plays an important
role in promoting the development of automated testing system, so the bus technol-
ogy is an important technology of the 2.0 Era and the main changes in the testing
instruments cannot be separated from the development of the bus technology.
The birth of the GPIB technology enables engineers to directly connect the test
instruments with computers. Almost every device has a GPIB interface, which is
firm, reliable, universal, simple and convenient. The simplest is to connect one
computer with a testing instrument. However, it will be limited if the test needs to
use multiple instruments. In that case, with low data transmission rate, high cost,
limited bandwidth and poor reliability, the synchronization and trigger function of
multiple instruments are unable to be provided, which will affect the test
performance.
VXI has higher bandwidth and lower latency than GPIB. But it is difficult to be
applied to other fields because of its high cost. It is mainly used in military and
aerospace. In addition, VXI is based on the outdated VME bus and the modern
computer does not support this kind of bus structure, which has affected the
development of the bus technology.
Intel put forward the concept of PCI bus for the first time in 1990s, which was
used to connect peripherals and computer backplane. The PCI bus is a parallel bus
that can work in 33 MHz and 32 bit. The maximum theoretical bandwidth is
132 Mbytes/s, and it employs the shared bus topology.
After the PCI bus was used in the instrument field, under the combined effect of
various factors, the PXI technology bus was obtained through the corresponding
expansion of the PCI bus. It is a new bus standard launched by American National
Instrument (NI) Company in 1997. PXI is based on the mature PCI bus technology,
so it has faster bus transmission rate, smaller volume and better cost performance
compared with VXI. In addition, PXI is capable of providing nanosecond timing
and synchronization features as well as robust industrial characteristics. Finally,
thanks to the flexibility of the software and the continuous updating of the modular
hardware, users can upgrade the entire test system at any time with the least
investment. Such an excellent scalability and flexible software architecture make
the system integration based on PXI modular instrumentation platform more
common.
In 2004, PCI Express was launched. It is an expansion on the PCI bus. The
testing system with the Instrument 2.0 technology can enhance the performance of
data stream transmission and data analysis when adopting this kind of bus. Com-
paring with the PCI bus, PCI Express uses the point-to-point bus topology, which
provides a separate data transmission path for each device. All slots of PCI Express
have dedicated bandwidth to connect the PC memory so that they do not have to
share bandwidth like traditional PCI. Data will be transmitted and received in the
packet form through the symmetric channels with the bandwidth of 250 Mbytes/s in
each direction. Multiple channels gathering together can maximally form a 32 times
channel width, significantly improving the data transmission bandwidth, minimiz-
ing the demand on memory and speeding up the transmission of data stream.
PXI Systems Alliance (PXISA) official organization officially launched the
hardware and software standards for PXI Express in the third quarter of 2005. By
using the technology of PCI Express on the backplane, PXI Express was able to
increase the bandwidth by 45 times, from the original 132 MB/s of PXI to current
6 GB/s. At the same time, the compatibility of the software and hardware was
retained with original PXI modules. The performance was enhanced so much that
PXI Express could enter more application fields that were used to be controlled by
dedicated instruments, such as Intermediate Frequency (IF) and Radio Frequency
(RF) digital instruments, communications protocol verification and so on. Thanks
to the software compatibility, PCI Express and PXI can share the same software
architecture. In order to provide hardware compatibility, the latest Compact PCI
Express standard has defined hybrid slots to support modules based on the archi-
tecture of PXI or PCI Express simultaneously [7]. PXI Express also offers a
100 MHz differential system clock, differential signals, and differential trigger
time and synchronization characteristics. Prior to the launch of the technology,
some fields had to only rely on expensive dedicated device to solve the problem.
PCI Express technology solves the problem in these fields with high performance,
for example, high bandwidth IF instruments for communications system testing, the
protocol based on low voltage difference signals, interfaces of high-speed digital
protocols such as the FireWire protocol and the optical fiber channel protocol,
large-scale channel data identification systems for structural and incentive test,
high-speed image recognition and data stream processing, etc.
The AXIe bus technology, an open standard for modular test equipment, has
been proposed in recent years. It inherits the advantages of modular architecture of
Advanced TCA and refers to the existing standards of VXI, PXI, LXI and IVI. It has
larger circuit boards, higher data transmission rate, greater power and better heat
dissipation than VXI or PXI. Moreover, a modular flexible platform with long life
cycle, high performance and strong scalability are provided. Its goal is to create an
ecosystem composed of various components, products and systems, promote the
development of general instruments and semiconductor testing, provide maximum
scalability and meet the needs of various platforms including the general rack
stacking system, modular system, semiconductor AT, etc. [8].
Software Defined Radio Technology
Software Defined Radio, known as the third revolution in the information field, is
the most representative technology of the Instrument 2.0 Era. The software radio is
to take hardware as the basic platform of wireless communications and implement
the maximal functions of wireless communications and personal communications
with software.
In the 1990s, as a new concept and system of wireless communications, software
radio began to receive attention at home and abroad. It gave communications
system good universality and flexibility, and made it easy for system interconnec-
tion and upgrading. The technology turned out to be a major breakthrough in the
field of radio and was primarily applied in military field. The basic idea is to let all
the tactical radios in use be based on the same hardware platform, install different
software to form different types of radio, complete functions of different natures,
and get the software programmable capability [9, 10]. At the beginning of the
twenty-first century, with the efforts of many companies, the application of soft-
ware radio was transformed from military to civilian fields, such as multi-band
multi-mode mobile terminals, multi-band multi-mode base stations, WLAN and
universal gateways, etc. The software defined radio technology, as a new wireless
communications system structure with strong flexibility and openness, has naturally
become the strategic base of global communications [11].
Because of the increase in users demand, the shortage of spectrum resources
and the huge attraction of new business have brought a lot of stress to equipment
manufacturers and operators and promoted the updating of wireless technology
standards. As there are significant differences between various wireless technology
standards and systems, the existing hardware-based wireless communications sys-
tem is difficult to adapt to this situation, which has led to the concept of Software
Defined Radio and the updated technology and equipment. The update of technol-
ogy and equipment usually causes the waste of equipment and investment, while
the software defined radio technology is able to save the inconvenience caused by
technology updating through the self-defined software.
The architecture of software radio is different from that of the traditional
software. In conventional wireless communications systems, the RF part, up/down
frequency conversion, filtering and baseband processing are all in the analog form.
A certain band or type of modulation of communications system corresponds to a
certain specialized hardware structure. The low-frequency part of the digital radio
system adopts the digital circuit, but the RF and IF parts are still inseparable from the
analog circuit [10]. Compared with conventional radio systems, software radio
realizes all of the communications functions by the software. The key idea of
software radio is to construct a standardized, modular universal hardware platform,
to realize various functions by software, and to make the broadband A/D and D/A
convert to IF, near the RF side of the antenna, and strive to carry out the digital
processing from IF. In addition, different from digital radio using digital circuits,
software radio adopts high-speed DSP/CPU. The software radio system must work
in parallel with a number of CPUs, and the digital signal processing data should be
exchanged at high speed, which requires the system bus to have a very high data
transmission rate. DSP devices are used to replace the dedicated digital circuit board
so that the system hardware structure and functions are relatively independent.
Modular design is also employed to give the platform openness, scalability and
compatibility. Based on the relatively universal hardware platform, software radio
realizes different communications functions by loading different software. It can
rapidly change channel access methods or modulation modes and adapt to different
standards by utilizing different software, thus forming the highly-flexible multi-
mode terminals and multi-functional base stations to achieve interconnections.
In practical applications, software radio requires a very high speed of hardware
and software processing. Due to the limitation of the technological level of hard-
ware, the concept of pure software radio has not been widely used in practical
products. The SDR technology based on the concept of software radio has attracted
more and more attention. Software defined radio is a system, which must have the
ability of reprogramming and reconstructing, so that the equipment can be used in
various standards and multiple frequency bands and realize a variety of functions. It
will not only use programmable devices to implement digital baseband signal
processing, but also carry out programming and reconstruction on analog circuits
of radio frequency and intermediate frequency, and have the ability of
reprogramming, reset, providing and changing services, supporting multiple stan-
dards and the ability of intelligent spectrum utilization, etc.
The software defined radio technology is mainly realized based on single-chip
FPGA or DSP. However, with the continuous increase in the volume of collected
data, signal processing capabilities of software radio platform also need to be
further enhanced. In order to improve modern signal processing capacity, the new
technology is also proposed. For example software defined radio technology based
on Compact PCI (CPCI) has attracted many teams and manufacturers to develop the
software radio platform system in line with the CPCI standard [12].
Modular Instrument Technology
Different from simple length measurement, measurement in the wireless commu-

nications field has become more and more systematic. Many brand-new test
methods and instrument concepts emerge one after another. High-speed communi-
cations test instruments featuring modular, software, and integration continue to
spring up, which complement and get closely integrated with traditional methods,
expanding the applications continuously and forming a technical highlight in the
communication-centered application fields.
As mentioned in [7], VXI was the earliest bus which introduced the concept of
modular instrument. It successfully reduced the size of the traditional instrument
system and improved the level of system integration. It mainly used to meet the
needs of high-end application of automated test, and had been successfully applied
in the military aviation test, the manufacturing test and so on. Like the tested
terminals, the test system is developing from the hardware-centric, single-purpose
and limited-functions stage to the software-centric, multi-purposes and unlimited-
functions stage. The modular instrument technology is the epitome of the techno-
logical progress.
Traditional desktop instruments generally have only a single function. They
achieve instrument functions through the man-machine operation step by step.
Modular instruments, on the other hand, highly integrated with computers and
suitable for PCI and PXI platforms, are highly-flexible plug-in computer boards.
The functions of modular instrument are similar to the traditional desktop instru-
ments but better. The modular instrument is an important part of the SDR technol-
ogy. Being defined by software enables it to define the measurement and analysis in
real time, breaking the rule that the traditional instrument manufacturer defines the
fixed functions. The users can realize the required test tasks within a short time
through the functions of software-defined modular instruments. They can apply the
customized data analysis algorithms, create a customized user interface, change the
situation of purely displaying the test results of traditional instruments, and add
more test engineers ideas and testing intentions, thus giving engineers more
initiatives. The difference of modular instruments with the traditional instruments
mainly comes from the progress of the bus technology, with which different
modular instruments share a power supply, a chassis and a controller. The bus
can guarantee the data transmission channel between the modules. Through the bus
control, different functions of the modular instruments can be integrated to reduce
the volume of the instruments and simplify the testing complexity.
With the aid of modular instruments, engineers can choose different kinds of
modular instruments according to their measurement needs and set up a test system.
Due to the adoption of the software defined modular structure, system measurement
can be realized through the corresponding software configuration operation. The
service cycle of modular instruments can be increased through software upgrading.
Different test requirements can be achieved through software programming. The
life of the instruments can be ensured through repeatable and flexible use of
modular instruments, relieving the pressure of increasingly complex equipment
and technology on the test time. Especially, after the introduction of new wireless
standards, the engineers using traditional instruments need to wait for suppliers to
develop a corresponding desktop instrument before testing the standards. However,
with the new software defined radio technology and modular instruments
technology, the engineers are able to test the wireless standards in the standard-
setting process with the universal module instruments and user-defined wireless
protocols and algorithms. Due to the different characteristics from the traditional
instruments, the modular instrument technology has become the representative
technology of the Instrument 2.0 Era.
Hybrid Test System
On account of the advance of the computer technology, the bus technology and their
penetration in instrumentation, the hybrid systems composed of different bus
technologies and instruments are gradually appearing in the testing field. Users of
such kind of hybrid systems can not only enjoy the high speed and flexibility of
modular instruments, but also use existing discrete instruments for some special
measurements.
In a hybrid test system, different components of multiple automated test plat-
forms are integrated in a system, including PXI, PCI, GPIB, VXI, Universal Serial
Bus (USB), Local Area Network (LAN) and LXI, and other different buses. Its
emergence is not a coincidence. Although the bus technology is developing forward
and PXI has been widely recognized and used, other bus technologies cannot
completely disappear. The instruments that include other bus technologies, such
as capture card based on the Industry Standard Architecture (ISA) bus technology
and GPIB control cards, are still trusted and being used by many engineers and
manufacturers, and the bus cannot be completely replaced by PXI. Moreover, from
the test engineers point of view, when designing a test system, there are various
factors to balance. Now the products are becoming more and more complex, the
requirements for the mixed signal test are also getting higher and higher, so it is
necessary to use the advantages of different bus test platforms and build a hybrid
test system to meet the testing demands.
In the examples of the equipment system indicated in Fig. 2.1, the hardware
employs the GPIB, PXI, LXI and many other test buses to form a hybrid system.
What the test engineers should do is not just simply connect several instruments,
i.e. to make the hybrid system that is composed of different products of different
manufacturers and different buses work normally and smoothly, they also need to
consider the software architecture of the hybrid system. Having unified software
architecture can greatly simplify the complexity of system programming and avoid
the problem of compatibility between different instruments.
For example, NI Company offers unified software architecture composed of the
measurement and control service layer and the application development layer. The
architecture provides software for the hybrid systems. If the hardware is the
skeleton of the hybrid system, then the software is the soul to control the whole
system. The measurement and control service layer includes a flexible device driver
and is used to connect the software and hardware and simplify the test code of the
hardware configuration part. The company also proposed Virtual Instrumentation
Software Architecture (VISA) standards to provide the API with high-performance,
Fig. 2.1 A kind of hardware components of the hybrid bus test system [7]
programmability and continuous upgrading to help with the development. VISA,

responsible for the communications with driver software, is independent of the
instrumental bus and will help hardware seamlessly integrate into the software.
Correspondingly, LabVIEW, the graphical programming language developed
mainly by the company, programs in a graphical manner, helping users to effi-
ciently and quickly develop and test applications.
The emergence and use of hybrid systems have brought tangible benefits to the
engineers. For example, some test of engineers may need high throughput and
excellent integration provided by the modular instrument bus like PXI and PCI
Express. At the same time, they may also need discrete instruments based on USB
or LAN (including LXI) for some specific test functions. In addition, if there are
new testing needs, users can use hybrid systems. Engineers can easily upgrade or
add new components in the existing system without redesigning the whole system,
which ensures the continuity and reusability of the test system. The hybrid system is
logically called a representative product of Instrument 2.0 Era, and with the
development of technologies such as the bus and software technologies, it will
continue to evolve.
Instrument 3.0 Era
Different from the evolution time from Instrument 1.0 to 2.0, after the Instrument
2.0, thanks to the rapid development of software technology and network technol-
ogy, the 3.0 Era has arrived.
Instrument 3.0 Era reflects the characteristics of openness, union, and service.
Generally, there are two kinds of definitions for the 3.0 Era. From the standpoint of
test measurement industry, the Instrument 3.0 is defined as follows: in addition to
the function and performance of the equipment itself, the technical service appears
with the instrument. From the technical and academic point of view, the Instrument
3.0 can be defined as follows: in addition to the characteristics of Instrument 2.0,
features like network-based combination, cross-layer test and remote test have
emerged. Cloud testing, software and hardware joint multi-user system simulation
test is a series of applications confirming the characteristics of the instrument 3.0
technology.
2.2.2 Development Trend of Testing Technology
Through the above description on the development, the key technologies and
challenges of testing technology, it is not difficult to find that to cope with the
changing challenges of new standards and technologies, engineers are facing many
difficulties in testing wireless devices and more complex testing work. Although
different testing techniques have developed towards software-based, modular and
intelligent directions, it is still unable to meet the needs of the rapid technical
development. In this section, we summarizes the main development trend of
wireless testing technology based on the testing technology development status
and the development needs of wireless communications,
Coexistence of New and Traditional Test Equipment
The instrument system for test and measurement has been developing towards the
software-centered modular system. Accordingly, users can integrate test into the
design process in a faster and more flexible way, shorten the development time, and
improve the test efficiency. While the software defined instruments depending on
software have become increasingly popular and been widely used, the synthesis of
the instruments is also proposed. However, the traditional instruments are also
advancing with the bus technology. Therefore, the emergence and widespread
adoption of modular instruments and software defined radio technology cannot
immediately replace the traditional test instruments. The renewal of the traditional
test instruments will still be competitive, and the two will continue to coexist.
The Leading Trend of Many-Core/Parallel Technology and FPGA Real

Time Technology
In the testing field, many-core parallel technology and FPGA real time technology
will become important supports for the development of testing technology, and can
lead the development trend of testing technology.
Many-Core Parallel Technology
Over the years, the microprocessor has been developed from single core into
multiple core. Commercial equipment with dual-core or even eight cores are now
everywhere. Owing to the software defined instruments, users can enjoy the huge
performance improvement of automated test applications brought by the multi-core
processor immediately. While multi-core is no longer satisfactory, many-core
comes into being. Moores law states that the number of transistors would be
doubled every year, and processor manufacturers can use these increasing transis-
tors to create more cores. Today, desktops, mobile and ultra-mobile computing
commonly use dual-core and quad core processors, and servers usually use ten or
more cores. More cores are being plugged into smaller, less-power encapsulations.
The number of cores in a server processor is increasing drastically to more than
10, and the processor is developing towards many-core. Super computers enable us
to understand what the future processors will be like. The fastest computers in the
world use millions of cores. Although these multi megawatt super computers are
not suitable for the test stations in the production workshop, integrating more
functions into smaller and smaller space means that many-core processors will be
able to shrink to equal size of other processors due to reducing the volume of power
supply. An example of this tendency is the coprocessors of Intel Xeon Phi class,
which provide up to 61 cores and 244 threads of parallel execution capability.
Market demand is driving the improvement in the performance of graphics cards
and augment of the number of cores. Although the test and measurement applica-
tions basically do not use the graphics function, the new processor with more kernel
can provide higher performance to test applications specifically designed for a
higher number of cores. The migration from single core, multi-core to the many-
core processor has been very common in the field of high-performance computing.
With more advantages of parallel processing being found, many-core processors
will continue to be adopted by more routine applications.
The emergence and development of multi-core/many-core processors, is not only
an inevitable choice of the development of the semiconductor industry, but also a very
reasonable architecture. This architecture is well adapted to the current computing
environment and application mode of Internet. Although there are still large quantities
of problems to be solved, covering from the internal structure, such as network on chip
and cache organization and coherence protocol, to the external environment, including
programming model and key system software, there is no doubt that as the core
component of future computing platform, multi-core/many-core processor is welcom-
ing the best time for development [13]. The development of multi-core/many-core
technology will surely promote the development of the test industry.
Real-Time FPGA Technology
The functions of the FPGA technology will become more and more prominent in
the future. In many cases, when the test system is developed on a multi-level, it is
always a trade-off between customized hardware and commercial general compo-

nents to find the best balance point between the throughput and the test coverage.
FPGA can realize the advantages of customized hardware tools through the flexible
software system. The FPGA technology has many advantages. One key benefit that
FPGA brings to the test application is to liberate the master processor from a large
number of data streams or internal tasks. FPGA is suitable for uninterrupted
filtering, modulation/demodulation, encryption/decryption or other data processing.
Another key benefit is to ensure the inherent parallelism of data. Chip is able to
allow any number of parallel data to process channels. Finally, FPGA can
reconfigure traditional hardware. FPGA is defined by the software first and then
downloaded to a FPGA chip for an actual execution. Only through downloading a
new program to the FPGA chip can the hardware be reconfigured. Although in the
past FPGA was applied in some forms in the testing field, there is no significant
return on investment. It is mainly because the two key technical difficulties. Firstly,
FPGA is a kind of discrete programmable chip. To apply this technology in
practice, users must design peripheral circuits, including digital to analog converter,
analog to digital converter, clock, power and some necessary components for FPGA
programming. Likewise, the Equipment Under Test (EUT) that can communicate
with the peripheral equipment must be provided. Secondly, the hardware descrip-
tion languages of FPGA, such as Verilog or Very-High-Speed Integrated Circuit
Hardware Description Language (VHDL), have not been widely used in the
industry. Only a few digital development engineers are able to use these languages.
In the system of Instrument 2.0, the first problem has been solved by using standard
commercial universal module combined with the FPGA technology. Many manu-
facturers provide FPGA based hardware systems, which have already embedded a
lot of peripheral circuits in the development boards or platforms. In addition, the
current trend shows that the hardware description languages have become more and
more abstract, and been provided to all the engineering disciplines for application.
At the same time, in order to support the design of FPGA, the important work has
been completed in the C to VHDL language conversion and graphical programming
technology, which ensures the development of the FPGA technology and reflects
the importance of the FPGA technology.
As design and testing requirements become higher and higher, the FPGA
technology is introduced into the FPGA based customized instruments. FPGA
with high performance and reconfigurable characteristics provides the support for
the application of FPGA. Especially, in the 5G wireless communications system in
the future, performance requirements will be higher, and FPGA will be needed to
ensure that the response is real time and the inflow and outflow speed of the data are
high in the real-time system simulation and high-speed memory test applications.
As the core of 5G communications test equipment, the digital signal processor
based on FPGA has the following advantages [14]:
1. FPGA has an internal parallel mechanism, which can perform complex mathe-
matical calculations simultaneously without occupying the host processor.
2. According to the ability of reprogram, FPGA can be used for testing applications
of a variety of current or future communications standards.
3. FPGAs powerful functions can reduce the cost and size of the RF test
equipment.
4. Compared with traditional firmware or software-led tests, using FPGA signifi-
cantly shortens the test time, so that large-scale industrial applications become
possible.
The advantages and application fundamentals of FPGA ensure its bright future
in testing and application. The test industry has developed and applied the products
combining with FPGA. For example, R-series data collection and FlexRIO products
family provided by NI have integrated the high-performance FPGA into the readily
available Input/Output (I/O) broad card for users to customize and repeat configu-
rations according to the applications. At the same time, with the easy and intuitive
graphical programming of LabVIEW FPGA and without the need to write the
underlying VHDL code, users can rapidly configure and program FPGA functions
for test automation and control applications [15], simplifying the application
complexity of FPGA technology.
Test Ecosystem Becomes the Trend
The challenges of testing technology have depicted the future of test instrument
times. Through the introduction to the evolution and challenges of test instruments,
we can find that in addition to the above trends of future testing technology, an
upstream and downstream test system will be formed, i.e. a software-centered test
ecosystem.
How to keep up with the rapid development of many communications technol-
ogies and standards, how to improve the testing budget cost performance, how to
flexibly redefine testing requirements and methods, how to effectively use the
multi-core technology, how to use real-time processing technology to improve
the test throughput -- the answers to all these problems will point to the solution
of software centric [15].
FPGA Based IP to the PIN Technology
For decades, the electronic and communications industry has been pursuing the
ideal state of mutual improvement of design and testing. In view of the difference
between design and testing, this goal has not been reached yet. In the design phase,
the latest Electronics Design Automation (EDA) software is applied to the system
level design, while the testing area is slightly independent and lagging. Therefore,
with the latest software-centered electronic communications equipment, a new test
solution is often needed to be found.
Adopting system-level approach, integrating the concepts of design and test,
and expanding software architecture to FPGA is one of the effective means to
balance the development of the two areas and improve the efficiency of
Fig. 2.2 V Graph to realize IP to the Pin [14]
communications testing. The way to integrate design and test is to deploy the
designed IP cores to Device Under Test (DUT) and integrated test platform. This
deployment process is called IP to the Pin [16], because it enables tester-defined
software IP to be close to the hardware I/O pins of integrated test platform as much
as possible. These software IP can include: data acquisition, signal generation,
digital protocol, mathematical operation, RF and real-time signal processing, etc.
Regardless of throughput or power consumption of a single devices original data
processing, FPGA is better than digital signal processor, traditional processor, and
even graphics processor [17].
The specific implementation of IP to the Pin technology can be expressed as the
V graph. Each phase of the design has a corresponding verification or testing
phase. By sharing IP, the design and test team can respectively move along the two
sides of the V graph, from modeling and design of the top layer to the implemen-
tation of the bottom layer, carrying out the corresponding test at each stage
(Fig. 2.2).
Heterogeneous Computing Architecture Supporting Parallel Testing

and Massive Signal Processing
The heterogeneous computing architecture is a system for assigning data processing

and program execution tasks among different computing nodes, so that each node
can handle the most appropriate test and calculation task. This technology can be
used to deal with the increasingly complex testing calculations in future wireless
mobile communications, to store and process massive amounts data of MIMO and
massive MIMO in the RF back end, and to enable the multiple nodes to make
Software environment of
heterogeneous computing Multi-functional integrated bus
connection and triggering
MIMO signal analyzer

online signal processing)
MIMO signal source
FPGA unit (real-time
Channel simulator
behavior control and
Multi-core CPU (test
analysis)
Cloud computing GPU server
(offline data analysis (online/offline
and storage) algorithm analysis)
Fig. 2.3 Heterogeneous computing architecture for complex mobile communications testing
simultaneous spectrum sensing and detection, and collaborative test of the PHY
layer and the Media Access Control (MAC) layer.
Take the MIMO RF test system using the heterogeneous computing architecture
as an example. CPU can be used to control the program execution. FPGA is used for
online demodulation, and the GPU is used for the calculation of multi-antenna test
parameters. Finally all the processing results are stored on a remote server. The
following figure shows a heterogeneous computing architecture that can be applied
to the 5G communications test (Fig. 2.3).
Along with the rapid development of high bandwidth and high data rate of 5G
mobile communications, the combination of the heterogeneous computing archi-
tecture and the multi-core parallel programming technology will be the indispens-
able main technology with which 5G test is able to deal with massive data
processing and improve the parallel testing.
2.3 Challenges of Testing Technology
2.3.1 Challenges of Multi-Function and High-Performance
The demand for wireless communications is increasing, and the requirements of

new technical standards are continuously improving. The development of digital
simulation technology and the increase of the frequency band of RF test are urging
the development of the testing technology. For example, the increasing bandwidth
brings challenges to the testing technology. WLAN standards are generally applied
in the bandwidth which is lower than 100 MHz, so it does not need special methods
of the testing technology. However, the Ultra Wide Band (UWB) technology
requires the bandwidth to be 500 MHz, which brings a great impact on the design
of the testing instruments, because the test methods and techniques with different
bandwidths are very different. In addition, with the development of technologies,
the RF channel and the I/Q baseband channel both challenge the multi-channel
instrument technologies. Moreover, the development of the digital simulation
technology, speed improvement of digital interface and other growing technologies
2.3 Challenges of Testing Technology 39
challenge the testing industry, and meanwhile promote the change of the testing
technology.
The popular WLAN (802.11) protocol also drives the increase of RF channel
number, and supports multiple antennas and multiple communications frequencies.
In addition to the increasing RF channels, the IQ baseband channels of these
transceivers are also increasing. All of these have put forward higher requests to
the test ability of the baseband mixed signals in the test system. The independent
development of the digital and the analog simulation technologies also promotes
the change of testing, which brings the challenge to the test engineers.
Due to the various applications, the current devices usually integrate some parts
of different technologies like the cellular wireless technology, the short distance
wireless technology and the GPS technology. Therefore, it is not rare to have
multiple RF channels on a single chip. The brand-new structure of new testing
instruments should be able to handle the interface test efficiently. At the same time,
the coexistence demand of LTE, Wideband Code Division Multiple Access
(WCDMA)/ High-Speed Packet Access (HSPA), Global System for Mobile com-
munication (GSM) and other networks promotes the development of multi-mode
base stations and multi-mode terminals, and the corresponding test solutions of
multi-mode measurements must be specially designed. Unlike the 2G and 3G tests,
the 4G test not only tests the physical layer, but also comes up to the MAC layer
measurement before getting a more comprehensive understanding of the product
performance. It also presents new challenges for wireless communications testing,
and the wireless testing technology needs to adapt to the needs of the forward
progress. In addition, high-speed data transmission system has developed in the
direction of high frequency. The high frequency testing is also an inevitable
challenge in testing.
2.3.2 Challenges of Multi-channel
In order to promote its service support capacity, 5G will have new breakthrough in
the wireless transmission technology and the network technology. In terms of the
wireless transmission technology, technologies which can further explore the
potential of spectrum efficiency improvement will be introduced, such as the
advanced multiple access technology, multi-antenna technology, the modulation
and coding technology, the new waveform design technology and so on. Among
them the wireless transmission technology based on the large scale MIMO
(or massive MIMO) will be likely to make the spectrum efficiency and power
efficiency to upgrade an order of magnitude on the basis of 4G. This section
introduces the development and application of the MIMO technology, and dis-
cusses the possible challenges of the testing technology brought by the massive
MIMO technology in 5G.
As an effective means to improve the system spectrum efficiency and the
transmission reliability, the multi-antenna technology has been applied to a variety
of wireless communications systems, such as the 3G system, LTE, LTE-Advanced

(LTE-A), WLAN, etc. According to information theory, the more number of
antennas are, the more obvious increase in the spectrum efficiency and reliability
is. Especially when the number of transmitting and receiving antennas is large, the
MIMO channel capacity will have a close-to-linear increase along with the mini-
mum number of transmitting and receiving antennas. Therefore, it provides an
effective way to increase the capacity of the system by adopting a large number
of antennas. Due to the limitation of technical conditions such as the multi-antenna
occupied space, the complexity of the implementation and so on, the present
wireless communications systems are equipped with only a few antennas at the
transceivers. For example, there are at most four antennas in the LTE system and
eight antennas in the LTE-A system. However, due to its huge capacity and
reliability gain, the research on massive MIMO system has attracted great atten-
tions, such as the research works on MU-MIMO where the base station is equipped
with much larger numbers of antennas than that of mobile stations. In 2010,
Marzetta in Bell Labs studied the MU-MIMO technology with an unlimited number
of antennas were configured in each base station for multi cells with Time Division
Duplexing (TDD). He proposed the concept of massive MIMO and found some
different characteristics with those in a single cell with a limited number of
antennas. Thereafter, many researchers studied the configuration of the base station
with limited number of antennas on this basis. In massive MIMO, the number of
antennas configured by a base station is very large (usually tens to hundreds of
antennas, which is 1 to 2 orders of magnitude of the existing systems), serving
several users on the same time-frequency resources simultaneously. With regard to
the antenna configuration, these antennas can be centrally configured on one base
station to form a centralized massive MIMO system, and can also be configured in
multiple nodes to form a distributed massive MIMO system. It is worth mentioning
that Chinas scholars have been at the forefront of the international academia in the
study of the distributed MIMO. The benefits of massive MIMO are mainly reflected
in the following aspects. Firstly, the spatial resolution of massive MIMO is signif-
icantly enhanced compared with the existing MIMO. It can further utilize the space
dimension resources to make a plurality of users in the network communicate with
the base station on the same time-frequency resource, so that the spectrum effi-
ciency can be greatly improved without increasing base station density and band-
width. Secondly, massive MIMO can concentrate the beam in a very narrow range,
greatly reducing the interference. Thirdly, the transmission power can be reduced
significantly, so as to improve the power efficiency. Fourthly, when the number of
antennas is large enough, the simple linear precoding and the linear detector tends
to be optimal, and noise and uncorrelated interference can be ignored.
The research on massive MIMO has been mainly concentrated in the channel
model, capacity and transmission performance analysis, precoding, channel esti-
mation and signal detection in the recent two years. But due to little work of the
experimental model, the channel model has not been widely recognized. In addi-
tion, different from the traditional evaluation of the antenna radiation pattern, 5G
tests also need to complete the evaluation and verification of the end-to-end
2.3 Challenges of Testing Technology 41
receiving performance of the wireless system equipment. All of the above require-
ments need a large number of Over The Air (OTA) tests for support. The fifth
generation mobile communications technology in China defined that the number of
cooperative antenna at the side of the future 5G base station should not be less than
128 [18]. The number of antennas on this massive MIMO technology has risen to
the hundreds, greatly larger than that of traditional techniques. The factors such as
the number of test channels, the synchronization and isolation among multi chan-
nels, the storage of multi-channel data, present severe challenge to the massive
MIMO test implementation.
When validating the characteristics of the MIMO device, or achieving some of
the non-product MIMO systems like RADAR and beam forming, the multi-channel
RF architecture is needed. From the previous 22MIMO system to the present
88MIMO system as well as the future massive MIMO, scalability has become the
key requirement for the next generation of the RF test system. Testing MIMO
wireless devices usually does not require a multi-channel architecture, because it
generally does not require the full spectrum characteristics test. But MIMO itself
needs to add a test channel. In order to the implement parallel test of multi-protocol
wireless devices, engineers need to upgrade the existing RF devices to more
channels at a low cost, and it should be flexible enough to measure multiple
frequency bands. According to the above requirements, it can be seen that the
new generation of RF devices should have a fully parallel architecture of hardware
and software, and have advanced synchronization. In addition, the new RF devices
must be ready to be used, and provide a higher accuracy of synchronization, not
just to realize synchronization by using reference clock (usually 10 MHz) and
trigger. The reason is that although the traditional synchronization methods can
ensure the synchronous acquisition of the signal, it cannot ensure the signal phase
synchronization. The software part is even more important in the whole test
architecture, because it requires a lot of computations to deal with many kinds of
infinite standards. The modern software architectures can implement parallel data
streams, which make one or more processing units to be dedicated to a RF channel.
There are universal parallel processing architectures in the current market, includ-
ing multi-core processors, multi-thread, multi-core and FPGA. There are also some
other promising technologies, such as the Intel Turbo Boost technology used in the
latest generation of the Intel Nehalem. It automatically allows the processors to
overspeed in excess of the basic operating frequency, as long as it runs below the
rated power and the limits of the current and temperature. In order to fully utilize
these processor technologies, engineers need to apply the parallel programming
technology on the algorithm level and the application software level, such as task
parallelism, data parallelism and pipeline operations. The multi-channel test archi-
tecture reduces the total test time, increases the test throughput, and improves the
performance of the instruments. However, the flexibility of the architecture is also
very important. For example, the MIMO configuration is a typical dynamic con-
figuration, and the phase and amplitude of each transmitter can be used to optimize
the performance and direction of signals. Every addition of a MIMO transmitter
will make the software complexity grow in an exponential way.
Similarly, the emerging wireless technologies, for example, the MIMO antenna
systems, have a great impact on the design of the transceiver, which will also affect
the RF devices. The future multi-channel wireless system should be based on the
low-cost architecture, in which the signals and the software bear the characteristics
of parallelism.
Therefore, the application of the massive MIMO technology in 5G has brought
unprecedented challenges to the traditional RF test technology at aspects of both
testing methods and testing devices.
2.3.3 Challenges of High Throughput
The vigorous development of mobile Internet is the main driving force of 5G

mobile communications. Mobile Internet will be the basic business platform of
various future emerging businesses. The existing fixed Internet services will be
increasingly provided to the users in a wireless way. The wide applications of cloud
computing and backstage services will put forward higher requirements on the
transmission quality and the system capacity to 5G mobile communications system.
The main development goal of the 5G mobile communications system is to closely
connect with other wireless mobile communications technologies, to provide the
omnipresent fundamental business capacity to the rapid development of mobile
Internet. According to the present preliminary estimates of the industry, future
wireless mobile networks including 5G, will promote service ability in three
dimensions at the same time: (1) improve the utilization rate of resources by
more than 10 times on the basis of 4G by introducing new wireless transmission
technologies; (2) increase the system throughput rate by about 25 times by intro-
ducing new system structures (such as ultra-dense cells, etc.) and deeper intelli-
gence capabilities; (3) further explore new frequency resources (such as high
frequency band, millimeter wave and visible light, etc.), so that the frequency
resources of future wireless mobile communications can be expanded to around
4 times. Therefore, the amount of the related data from 5G terminals is bound to
increase drastically. It puts forward higher requirements on the 5G test systems:
how to capture, analyze, store and manage the massive data in real time.
As mentioned in the previous section, the applications of massive MIMO in 5G
has brought the need for the multi-channel RF testing. Similarly, massive data has
put forward new requirements on the RF test in both the method and the equipment.
At present, advanced testing instruments have begun to take these requirements into
account. For example, the Keithleys new-generation MIMO test platform makes it
simpler and cheaper to increase new signal standards and MIMO options. The
platform includes the Keithley 2920 RF vector signal generator, the 2820 vector
signal analyzer, the 2895MIMO synchronization unit and the Signal Meister wave-
form generation software. However, when faced with 5G, such devices are still far
from enough. More advanced 5G test equipments are essential. Besides the require-
ments on equipment, the more suitable testing method for massive data computing
References 43
is also one of the main research focus. At present, the heterogeneous computing
architecture is a 5G technology with much application potential. Along with the
rapid development of high bandwidth and high data rate of 5G mobile communi-
cations, the combination of heterogeneous computing architecture and multi-core
parallel programming technology will be the indispensable main technology with
which 5G test is able to deal with massive data processing and improve the parallel
testing.
2.4 Summary
This chapter focuses on the theme of testing technology evolution to explain the
importance of the testing technology in the industry chain, and guides the readers to
review the whole development process of the testing instrument industry. Due to
the development of the computer technology and the micro processing technology,
the traditional testing instruments are continuously developing towards the
bus-based, intelligent, distributed, modular and network directions. The improved
testing instruments are applied to the wireless communications, which facilitate the
test in the communications field. There is usually a system level test in the wireless
field. The development of the testing technology has avoided the stack-based test
methods needed for the system level test, simplified the test volume, and reduced
the human error. Although the test instruments have made great progress, we still
need to consider various requirements from the wireless field especially the future
wireless communications technology field. The constantly increasing requirements
are posing ever higher requirements on the communications technology. The
communications technology is also developing rapidly. All of these are challenging
the testing technology. This chapter analyzes the possible challenges as well as the
future development trends of the testing technology. With the constantly emerging
challenges, the testing technology will surely make great progress accordingly. On
the basis of the existing testing technology, it will grow toward the healthy direction
of the ecological test system.
References
1. X. You, Z. Pan, X. Gao, S. Cao and H. Wu. 5G Mobile Communication Development Trend
and Some Key Technologies. China Science, Information Science, 2014, 44(5):551563.
2. C. Wang, F. Haider, X. Gao, et al. Cellular architecture and key technologies for 5G wireless
communication networks, IEEE Communications Magazine, 2014, 52(2):122130.
3. E. Larsson, O. Edfors, F. Tufvesson, T. Marzetta, Massive MIMO for next generation wireless
systems. IEEE Communications Magazine, 2014, 52(2):186195.
4. J. G. Andrews, S. Buzzi, W. Choi, and et al. What Will 5G Be? IEEE Journal on Selected Areas
in Communications, 2014, 32(6):10651082.
5. IMT-2020 (5G) Promotion Group, 5G Network Technology Architecture, 2015.
6. A. Gohil, H. Modi, S. K. Patel. 5G technology of mobile communication: A survey. Intelligent

Systems and Signal Processing (ISSP), 2013, 288292.
7. X. Ge; H. Cheng, M. Guizani and H. Tao, 5G wireless backhaul networks: challenges and
research advances. IEEE Network, 2014, 28(6):611.
8. A. Osseiran, F. Boccardi, V. Braun and et al. Scenarios for 5G mobile and wireless commu-
nications: the vision of the METIS project. IEEE Communications Magazine, 2014, 52
(5):2635.
9. P. Prinen. A brief overview of 5G research activities. International Conference on 5g for
Ubiquitous Connectivity. 2014:1722.
10. X. Shen. Device-to-device communication in 5G cellular networks. IEEE Network, 2015, 29
(2):23.
11. S. Chen, J. Zhao. The requirements, challenges, and technologies for 5G of terrestrial mobile
telecommunication. IEEE Communications Magazine, 2014, 52(5):3643.
12. N. D. Han, Y. Chung, M. Jo. Green data centers for cloud-assisted mobile ad hoc networks in
5G. IEEE Network, 2015, 29(2):7076.
13. E. Dahlman, G. Mildh, S. Parkvall and et al. 5G wireless access: requirements and realization.
IEEE Communications Magazine, 2014, 52(12):4247.
14. W. Nam, D. Bai, J. Lee and I. Kang. Advanced interference management for 5G cellular
networks. IEEE Communications Magazine, 2014, 52(5):5260.
15. M. Condoluci, M. Dohler, G. Araniti, A. Molinaro and K. Zheng. Toward 5G densenets:
architectural advances for effective machine-type communications over femtocells. IEEE
Communications Magazine, 2015, 53(1):143141.
16. O. Galinina, A. Pyattaev, S. Andreev and et al. 5G Multi-RAT LTE-WiFi Ultra-Dense Small
Cells: Performance Dynamics, Architecture, and Trends. IEEE Journal on Selected Areas in
Communications, 2015, 33(6):12241240.
17. F. Boccardi, R. W. Heath, A. Lozano and et al. Five disruptive technology directions for 5G.
IEEE Communications Magazine, 2014, 52(2):7480.
18. P. Ameigeiras, J. J. Ramos-Munoz, L. Schumacher and et al. Link-level access cloud archi-
tecture design based on SDN for 5G networks. IEEE Network, 2014, 29(2):2431.
Chapter 3
Channel Measurement and Modeling
The wireless channel model is a deterministric or stochastic numerical description

of physical propagation enviroments based on channel measurement or analysis and
simulation on the basis of propagation theories. It provides an effective and simple
means to approximately express the channel characteristics of the wireless trans-
mission. Channel models can be used for link and system-level performance
evaluation, optimization, and comparative studies on wireless communications
technologies, algorithms, products and system. Channel model is of great impor-
tance to all wireless communications (including 5G mobile communication)
research. This chapter focuses on the progress in 5G channel measurement and
channel modeling. Section 3.1 introduces the requirements for 5G wireless channel
model. Section 3.2 introduces and compares five commonly used channel modeling
methods. Section 3.3 introduces the knowledge on channel measurement, channel
measurement system and the currently ongoing 5G wireless channel measurement
activities. Section 3.4 describes several channel parameter extraction algorithms
and statistical analysis of channel parameters. Section 3.5 introduces several
existing channel models. Section 3.6 gives an example to show the generating
process of stochastic channel models. And the last section is the summary of this
chapter.
3.1 Requirements for 5G Wireless Channel Model
The 5G typical scenarios will touch many aspects of life in the future, such as
residence, work, leisure, and transportation, mainly involving dense residential
areas, office, stadiums, indoor shopping mall, open-air festivals, subways, high-
ways and high-speed rails. Compared with 3G/4G system, 5G adds some new
application scenarios, which have diversified features like ultra-high traffic volume
density, ultra-high connection density and ultra-high mobility. A variety of tech-
nologies are used to serve the end users, such as massive MIMO, millimeter wave

DOI 10.1007/978-3-319-61869-2_3
46 3 Channel Measurement and Modeling
(mmWave) communications, ultra dense networks, and D2D, etc. The application
of these new technologies and the requirements for 5G mobile communications
system present new challenges to the wireless channel models. The 5G channel
models should support wide range propagation scenarios, higher frequency and
larger bandwidth, as well as larger antenna array and so on, and keep consistency in
space, time, frequency and antenna, which are mainly showed in the following
aspects [1]:
1. Support wide range propagation scenarios and diverse network topologies
The vision of 5G is that anybody and anything can get access to information and
share data with each other whenever and wherever they are. IMT 2020 5G promo-
tion group of China has interpreted this as Information a finger away, everything in
touch [2]. Current cellular mobile communications networks serve the static and
mobile users with fixed base station(BS), so traditional channel model developed to
serve for this type of applications.While under the 5G architecture, network topol-
ogy should support not only cellular networks, but also the communication of
device to device (D2D), machine to machine (M2M) and vehicle-to-vehicle
(V2V) as well as the whole interconnected network.Accordingly, the 5G channel
models should support mobile to mobile links and networks.
In D2D/V2V communications, the user antenna is generally as low as 1-2.5 m,
which limits its coverage to be smaller than the ordinary cellular system. Besides,
massive dense scatterers exist around both the transmitter and receiver. The sta-
tionary interval of channel is short due to the mobility of both ends. All of those are
the unique characteristics of D2D/V2V transmission. Many channel measurement
activities have been carried out in D2D/V2V, and several preliminary models have
been presented. For example, METIS (an EU 5G R&D projects) proposed a
D2D/V2V model with several scenarios [1], 3GPP proposed a preliminary D2D
channel model [3], and MiWEBA (an EU-Japan joint project) proposed a D2D
models in 60GHz mmWave band [4].
2. Support higher communications frequency and larger bandwidth
5G may operate in a larger frequency range from 350 MHz to 100 GHz. Taking
into account the available bandwidth and communications capacity demand, the
bandwidth of 5G system may be higher than 500 MHz, even 1~2 GHz or higher.
Thus, the 5G channel models should also meets such requirements. In 3GPP-HF
channel model, if the bandwidth of a channel is beyond c/D(wherein c is the speed
of light, D is the aperture size of antenna), then such bandwidth can be refered to as
Big Bandwidth and special processing should be applied [6]. High frequency
(HF) band is brought into keen focus for its large available communications
bandwidth. For example, METIS defines its medium and high priority frequency
bands to develop, including 10 GHz, 28~29 GHz, 32~33 GHz, 43 GHz, 46~50
GHz, 56~76 GHz and 81~86 GHz [5]. HF band (6~100 GHz) has many radio
transmission characteristics that are different from the frequency band below 6 GHz
(sub-6 GHz) [4]:
Firstly, according to the Friis Formula of free space propagation, propagation in
HF band will undergo higher path loss (PL) due to short wavelength. Friis Formula
3.1 Requirements for 5G Wireless Channel Model 47
also points out that the antenna gain is proportional to the square of frequency.
Given a fixed physical aperture size and transmit power, higher receive power will
be obtained when propagation in HF band than in low frequency band [7]. There-
fore, high gain directional antennas or beamforming technology should be adopted
for HF band to ensure communications over several hundred meters distance.
Meanwhile, the antenna also needs to be able to adjust its beam directing to
the UTs.
Secondly, diffraction in HF band weakens due to shorter wavelength, or to say
the propagation has quasi-optical characteristics, which results in lack of rich
scattering in HF band than the sub-6 GHz band. In the line of sight (LOS)
conditions, the received signal is comprised of LOS path and a few low-order
reflection paths, whereas in the non-line of sight (NLOS) conditions, signal prop-
agation may mainly depend on reflection and diffraction. Thus the channel shows
temporal and spatial sparsity. Blockage of human or vehicles will attenuate the
signal vastly. Quasi-optical properties also make it possible to use ray tracing
technique to assist channel modeling.
Despite the fact that the propagation characteristics of mmWave have been
widely investigated, especially in the 60 GHz, many important characteristics,
such as blockage loss, penetration loss, high resolution angle characteristics, fre-
quency dependence and outdoor mobile channel, still need further measurement
and exploration. In the aspect of channel model, several HF channel models
(including 60 GHz) have been proposed [1, 4, 612].
3. Support largeassive antenna array or massive MIMO
The existing channel models [13, 14] assume the superposition of plane waves
bouncing from the scatterers (i.e. scatterers are in the the far field of transmitting
and receiving), and assume that the antenna array is in smaller size. In both ends of
the antenna array, the arrival or departure direction of the radio waves only lead to
little difference in the phase of the signal while the amplitude of the signal remains
constant, i.e. the propagation characteristics of the both ends of the antenna array
are similar. When applying larger antenna technologies, including massive MIMO
and pencil beamforming, the antenna number can reach tens to hundreds, which
will bring about two impacts, as shown in Fig. 3.1.
Firstly, planar wavefront assumption for the conventional MIMO channels is no
longer applicable and should be replaced with spherical wavefront assumption. The
radio waves propagating in the form of plane wave need conform to the far field
assumption, i.e. the distance between the transmitter/receiver between scaterers
should be larger than the Fraunhofer distance [16], Rf 2D2/, D is the aperture size
of antenna array, and is wavelength. D and corresponding Fraunhofer distance of
array will increase with the number of antennas, assuming antenna element spacing
and remain unchanged, i.e. the far field of antenna system expands. The receiver
may be in the near field zone (also known as the Fresnel zone) of transmitter, or the
scatterers will be in the near field zone of transmitter and receiver. In this case, the
wavefront should be modeled as spherical wavefront instead of phane wavefront.
1st cluster
UT 2
BS 2nd cluster
Antenna
Array 3rd cluster
UT 1 UT 3
4th cluster
5th cluster
Fig. 3.1 Non-stationary and near field effect in Massive MIMO system [15]
Secondly, the researchers from the Lund University in Sweden have verified
through measurement that any scattering cluster in the Massive MIMO systems can
not affect all the antennas in the array. That is to say each antenna of the array has
different scattering clusters which can affect its signal propagation. In traditional
MIMO system, the scattering clusters that affect signal propagation only change
with time due to the relative movement among the transmitter, receiver and
scatterers Whereas in the massive MIMO system, scattering clusters also change
along the axis of antenna array.
Additionally, as the large antenna array own high spatial resolution to distin-
guish closely located users in horizontal and perpendicular directions, the channel
model must provide fine three-dimension (3D, includes both azimuth and elevation)
angle information. The angular resolution should be at least one degree or less. The
model should support a varity of antenna array, such as linear, planar, cylinder and
spherical array. Therefore, for massive antenna array, we need to accurately model
the azimuth and elevation angles at departure or arrival direction for each scattered
multipath, as well as the distances or locations of first-bounce/last-bounce scatterer
for spherical wavefront modeling. By far, several channel models have been
proposed according to the aforementioned requirements [17, 18].
4. Support spatial consistency and dual mobility
Spatial consistency (continuity) has two meanings: one is that the channels of
close links are of highly correlation, and the other is that the channel should evolve
smoothly rather than discontinuity or interruption when transmitter/receiver moves
or scenarios switch. The latter is commonly called the ability of dynamic
simulation.5G communications network will contain a variety of link types,
which will coexist in the same area, from the traditional macro cell, micro area,
to picocell, and femtocell, as well as nomadic BS and D2D connection between
UTs in the future. All of these ask 5G channel models to support spatial consis-
tency.At present, most of the widely used channel models [13, 14] consider the
spatial consistency of large scale parameters (LSPs), such as Path Loss, Shadow
3.1 Requirements for 5G Wireless Channel Model 49
Last bounce
al,m
a1,m UT track
bl,m
r Original Rx
ar,l,m,s er,s
Tx position
rr,s
Rx location at the s-th
Fig. 3.2 Scatterer location and AOA update in dynamic simulation [19]
Fading (SF), Delay Spread (DS), Angular Spread (AS) and Ricean K-factor.
However, the spatial consistency of small scale parameters (SSPs) is considered
inadequately. Moreover, The LSPs is generated with snapshot randomly and
assume the scattering environment of different snapshots are statistically indepen-
dent and the scatterers seen from close mobile devices are not related, which leads
to the exaggeration of the simulation performance of some multiple antenna
technologies such as multiple-user MIMO (MU-MIMO) [1]. In addition, most
traditional channel models consider the situation that one end of the link is fixed
and the other end can move arbitrarily. New Doppler model should be introduced to
support bidirectional mobility of D2D/V2V.
Withcoexistence and density increase of links, and with the application of
D2D/V2V, to support spatial consistency becomes especially important for of
wireless channel models. Space consistency model can be built through defining
the scatterers geometric position of the first bounce (the transmitter to scatterer)
and the last bounce (scatterer to the receiver) of each scattering path. As shown in
Fig. 3.2, QuaDriGa [19] channel model library supports dynamic evolution of the
channel caused by the movement of user terminal (UT). The channel coefficients
are calculated based on the location of the last bounce scatterer, the distance and
relative angle of arrival (AOA) changes between it and MT, and the distance
between mobile station and BS.
5. Support high mobility
Typical 5G high mobility scenarios involve mobile to mobile communication
(e.g., V2V), and mobility to infrastructure communication (e.g., subway, high-
speed rail, etc.). 5G should be able to support the mobility speed of above
500 km/h of high-speed rail, the ultra-high user density of 6 people /m2 on subway,
and the millisecond level end-to-end delay on highways. The channels in high
mobility condition have the following characteristics.
Fast time-varying characteristics: Different from the low mobility channel, high
mobility channels are fast time-varyingand show non-stationary characteristic. In
other words, the channel can only be stationary in a shorter period of time. Large
Doppler frequency shift or spectral spread: For either mobile to mobile channels, or
mobile to fixed channels, either the transmitter or the receiver is in high-speed
mobility.
More severe fading: The received signal is superposition of a number of

multipath signals with different amplitudes, phases, frequency shifts and delays.
The phase of each path is related to carrier wavelength, delay and Doppler shift.
The small path delay will bring great phase change with the increase of carrier
frequency. The signal phase caused by Doppler shift changes over time. Therefore,
superposition of those multipath signals with opposite phases may lead to severe
signal fading in some locations.
Mixed propagation scenarios: High-speed rail will encounter one or several
scenarios when it runs, mainly including open area, cutting, viaduct, tunnel and
station. Subway will be faced with two main scenarios: tunnel and platform.
Automobile will meet a variety of scenarios in driving process, including city
blocks, expressway, tunnels, bridges, and so on. These require the channel model
to support smooth evolvtion in switching scenarios.
At present, the existing channel models cannot precisely describe the character-
istics of the wireless channel in high-speed rail scenario. LTE-A system provides a
channel model that can support the two high speed rail scenarios, i.e. open area and
tunnel. However, the non-stationarity of the channel is not considered. WINNER II
model considers the high speed rail at the speed of 350 km/h and the impacts caused
by mobile relay, but it still assumes the channel is stationary. A few Some
measurement have been done in this area, such as the high speed rail/subway
measurement by State Key Lab of Rail Traffic Control & Safety in China, the
vehicle measurement by Lund University in Sweden and the measurement of high-
speed mobile scenarios by WINNER II project. As the high mobility increases the
difficulty and cost of measurement, more and more channel modeling is carried out
by means of ray tracing technology.
3.2 Channel Modeling Method
Channel modeling can be broadly split into deterministic and stochastic channel
modeling or their combination. When detailed environment data (including
man-made objects such as houses, buildings, bridges, roads, etc. as well as natural
objects such as foliage, rocks, ground, etc.) is available to a sufficient degree,
wireless propagation is a deterministic process that allows predicting its character-
istics at every point in space. This is the basis of deterministic channel model and
Ray Tracing is a such method.Stochastic channel modeling is one kind of method to
obtained huge measurement data set that containing the underlying statistical
properties of wireless channel by conducting channel measurements in a large
variety of locations and environments, from which path parameters are extracted
and used to statistically represent the channel parameters. Stochastic channel
models can be roughly divided into several categories as Geometric-based Stochas-
tic Channel Model (GSCM), Correlation-based Stochastic Channel Model
(CSCM), extended Saleh-Valenzuela (SV) stochastic channel model, and Ray
Tracing based Channel Model. GSCM is also known as Geometric-Based
3.2 Channel Modeling Method 51
Stochastic channel Model (GBCM) in some literature. There are two types of
GSCM. One is constructed based on the channel measurement, and the other is
based on the regular geometric shape of scatterers. Their model structure are same,
but the former is based on the measured data to extract the path parameters, and
obtain statistical characteristics through data analysis. The latter places emphasis on
deriving the autocorrelation functions, cross-correlation functions and other statis-
tical characteristics of the spatial, temporal and frequency domain of the channel.
Channel Parameters are usually divided into Path Loss (PL), large-scale param-
eters (LSPs, such as shadowing, delay spread, angular spread, etc.) and small-scale
parameters (such as delay, angle of arrival and departure, etc.), which jointly reflect
the channel fading characteristics. Path loss is usually expressed in one or two
formulas and a set of numerical values of parameters, reflecting the relationships
with transmission environment, distance and frequency and so on. Large-scale
parameters can be regarded as a statistical average in a channel segment (or a
quasi-static channel area), within which the LSP or probability distribution of LSPs
do not change significantly. The length channel segment is dependent on the
propagation environment. It is about dozens of wave-length. Small-scale parameter
is used to describe statistical characteristics of path parameter in a wavelength
range. Path Loss and Shadowing Model.
Friis formula gives the signal propagation model in free space. For the actual
scenarios, a more general path loss model is constructed by introducing a path loss
exponent dependent on environment. For LOS condition, a common path loss
model can be presented as [12]
PLdB 20log10 4f =c 10nlog10 d=1m X 3:1
This is called Close-in Reference short (CI) model [20], where f is the carrier
frequency (Hz), c is the speed of light, d is the distance between transmitter and
receiver (m). The first item at right side of the equation is the path loss in free space
at the distance 1 m. Xs is shadowing. The relationships between path loss and
frequency is same with Friis formula in free space. For NLOS, another model is
usually been used,
PLdB 20log10 d 10log10 f X 3:2
This equation is called ABG model [12] (named by three factors Alpha, Beta and
Gamma). It is the extension of FI (Floating Intercept) model [20] to reflect fre-
quency relevance.
3.2.1 Measurement-Based GSCM
MIMO channel is the superposition of rays with different power, delay and angular
information which can be determined by the geometry among the transmitter (Tx),
the receiver (Rx) and the scatterers. The rays with similar delays and angles are
The n-th N
Ray (n,m) rx,n,m
BS antenna
MS v,
tx,n,m rx,n,m
rx,n
N tx,n,m
BS tx,n MS,LoS,
MS antenna
MS antenna array
BS,LoS,
MS moving
BS antenna array
Fig. 3.3 Clusters and rays of GBSM channel model
grouped into one cluster [21], so that the channel can be descripted by a cluster-ray
structure as shown in Fig. 3.3. The GSCM modeling method separates antennas and
propagation channel. The user can combine arbitary type and layout of antennas
with the propagation channel model to obtain transmission channel model. The
whole 3D MIMO channel model from transmitting antenna array to receiving
antenna array is composed of the sub-channel Hu,s(t;) from transmitting antenna
element s to receiving antenna element u. Hu,s(t;) can be expressed as
" #T " #" #
X N XMn Frx, u, rx, n, m n, m n ,m Ftx, s, tx, n, m
H u, s t;
n1 m1 Frx, u, rx, n, m n
, m n, m Ftx, s, tx, n, m 3:3

ej2 rx, n, m d rx, u = ej2 tx, n, m d tx, s = ej2n, m t n, m
Where N is the number of clusters between antenna element u and s. Mn is the

number of ray in the n-th cluster. Ftx, s, and Ftx, s, are the radiattion patterns of
u for vertical and horizontal polarization respectively, Ftx, s, and Ftx, s, are the
elevation angle and azimuth radiation field patterns of s respectively. More com-
monly, vertical direction is called elevation (offset from horizontal plane) or zenith
(offset from the vertical upwards z-axis) and horizontal direction is called azimuth.
[n, m n
, m n, m n, m] are the complex gains of four polarization pairs of the m-th ray
in cluster n (using ray(n,m) for short), namely elevation-elevation (), elevation-
azimuth (), azimuth-elevation () and azimuth-azimuth (). The AOA (rx, n, m)
of ray (n,m) contains elevation angle of arrival (EOA) rx, n, m, and azimuth angle of
arrival (AoA) rx, n, m. AoD (tx, n, m) contains elevation angle of departure (EOD) tx,
n, m, and azimuth angle of departure (AOD) tx, n, m. rx, n, m and tx, n, m are the unit
vectors of AoA and AoD respectively. dtx, s and drx, u are the location vector of s and
u. is the wavelength of carrier. n, m and n, m are the Doppler shift and propagation
delay of ray n,m. If the wireless channel is dynamic, all the channel parameters are
time-varying, i.e. the function of t.
The clusters and rays in channel is parameterized by the path loss, shadowing
and other parameters in large scale and small scale. In order to keep spatial
consistency, the channel model considers the correlation of different LSPs in the
same station and the correlation of same LSP in the different stations. Cluster
parameters (including the number, arrival rate, power decay rate, angular spread)
and ray parameters within clusters (arrival rate, the average time of arrival, power
decay rate, etc.) are important parameters of the model, the existing GSCM usually
assumes the numbers of the clusters and the number of rays within one cluster and
are fixed, arrival interval of clusters follow exponential distribution, clusters power
decrease exponentially with delay, and cluster angles follow wrapped Gaussian or
Laplace distribution. The rays within one cluster have the same delay and power,
and different angles.
GSCM can adapt to the different scenarios and different types of antennas easily
by using rational parameters. For some new scenarios, like different high-speed rail
scenarios, such as the plains, hills, etc., theres no need to build any new model. One
only need to measure the channel of specific scenario and then update scenario-
related parameters and parameters of clusters and rays in the model. The common
GSCM modeling steps can are divided into three stages as illustrated in Fig. 3.4 and
described below [13].
Stage 1: preparation and measurement. Determine the generic channel model
formula and the parameters to be measured. Draw up a detailed measurement plan,
take into account all aspects of measurement: first, determine the scenarios of
wireless channel; select the measurement environment; determine antenna height,
speed of transmitter and receiver, the placement of channel sounder, measurement
routes and link budget, measurement time and other general requirements. Then
make actual measurements and store measurement data onto a mass memory.
Generic Measurement Channel Measurement

1 Model planning Measurements data
Data post-
Parameter
2 processing
PDFs
/analysis
Parameter
generation MIMO Channel
Channel
3 Matrix
realizations
Simulations
Array generation
info.
Fig. 3.4 GSCM basic modeling steps

Frequency response of the system RF chains and the radiation pattern of antenna
system are required to be measured and used in system calibration, subsequent data
processing and simulation.
Stage II: post-processing of the measured data. Different analysis methods are
applied depending on the required parameters. Typically, high-resolution parameter
estimation algorithm, such as Expectation-Maximization (EM), Space-Alternating
Generalized EM (SAGE), RIMAX or other algorithms, is used to extract path
parameters from the measured data. Output of data post-processing could be, e.g.,
a set of impulse responses, path-loss data, or multidimensional propagation param-
eters, such as gain or polarization gain, delay, AoA, AoD, as well as dense
scattering multipath components (DMC) and so on. The rays with similar param-
eters are grouped into one cluster, so that the parameters are divided into inter-
cluster ones and intra-cluster ones. Make statistical analysis on the post-processed
data to obtain probability distribution functions (PDFs) and the corresponding
numerical values of statistical parameters. The Goodness of Fit (GoF) tools may
be used to select the optimal distribution function among several candidates.
Stage III: the generation of simulation model. At first, clusters and ray param-
eters are generated according to their PDFs and statistical parameters. Then MIMO
transmission matrix is obtained by combining the generated parameters with
antenna information. At last, the time-varying Channel Impulse Response (CIR)
is generated to be used in simulations.
Measurement-based GSCM modeling method has been widely recognized
and used, such as COST 259/273/2100, SCM, SCME, WINNER /I/II/+,
IMT-Advanced, IEEE 802.16m, 3GPP 3D MIMO, 3GPP D2D, QuaDRiGa,
mmMAGIC, 5GCM, METIS and so on. Many new 5G channel model is modified
upon the basis model, e.g. by introducing correlation of cluster number N and
Tx/Rx (u/s) position to model the visibility of scatterers by antennas, by introducing
the correlation between tx,n,m ( rx,n,m) and s(u) position to support the spherical
wavefront propagation, by adding a another Doppler shift to characterize dual
mobility and so on. QuaDRiGa model library supports the realization of dynamic
simulation (time evolution) which depends on determing the locations of first-
bounce and last-bounce scatterers. With the locations of scatters, it is easier to
model spherical wavefront, spatial non-stationarity in Massive MIMO.
3.2.2 Regular-shape GSCM
The scatterers distribute randomly in the actual environment. Whereas, the regular-
shape GSCM distributes scatterers with similar properties onto regular geometric
shapes for simplifying characteristics analysis of the MIMO channel. The charac-
teristics of the propagation channel are described according to the geometric
relationships among the transmitter, receiver and scatterers. Only one-order or
two-order reflections during the propagation process are taken into account, while
high-order reflections are ignored because of higher attenuation. As shown in
BS MS BS MS
(a) One-ring model (b) Two-ring model
BS MS
BS MS
(c) Ellipse model (d) Rectangular model
BS MS
(e) One-sphere model (f) Elliptic-cylinder model
Fig. 3.5 Scatterer regular geometric sharpe-based model
Fig. 3.5, the basic shapes of 2D distribution of scatterers mainly includes: one-ring,
two-ring, ellipse and rectangular [22]; the basic shapes of 3D distribution of
scatterers mainly includes one-sphere, two-sphere, and elliptic-cylinder etc.. The
geometric model in practical application can be a combination of these basic
shapes, and it is also possible to add new geometric shapes.
The regular-shape GSCM is a parametric modeling method with high flexibility.
Different channel characteristics can be described accurately by adopting combi-
nation of several geometry shapes and adjusting the related parameters. For some
5G typical scenarios and channel characteristics, Chengxiang Wangs team at the
Heriot-Watt University has presented a series of channel models based on this
method, mainly including: the modeling of Massive MIMO, M2M as well as High
Speed Rail (HSR) [17, 23, 24]. For massive MIMO channel modeling, S. Wu et al.
built a 2D massive MIMO channel model based on non-stationary twin clusters,
from which space-time correlation function and power spectrum density are inves-
tigated [17]. Spherical wavefront was assumed and its impact on the statistical
properties of the channel model was investigated. Birth-death process was incor-
porated to capture the dynamic properties of clusters on both the array and time
axes. In channel modeling for V2V scenario, Yi Yuan et al. extended the geometric
model represented as two circles and an eclipse in 2D scenario into two spheres and
an ellipse cylinder in 3D scenario, which could be used to simulate the propagation

environment of the V2V channel [23]. In addition, the Chanel non-stationarity and
the influence of vehicle density on the channels were also taken into account.
Ghazal et al. proposed a non-stationary elliptical channel model of the high speed
rail and analyzed its related statistical characteristics [24]. The advantage of this
model was that it could be adapted to different high-speed rail scenarios (except
tunnel) by adjusting several parameters.
3.2.3 CSCM
CSCM describes the MIMO channel by the correlation matrix of antennas. It is

favoured due to ease of use for analysis of MIMO capacity and system perfor-
mance. Assuming that the number of receiving and transmitting antennas are Nr and
Nt respectively, the MIMO channel matrix H is a Nr by Nt matrix, and the full
correlation matrix of the channel RH is a (Nr*Nt) by (Nr*Nt) matrix. Usually it is
very hard to use RH directly to generate channel. The reason is that it is too large and
needs to specify (Nr*Nt)2 values, and each elements in the RH is difficult to
correspond to the physical propagation channel, so that the channel correlation
matrix model with low complexity is expected. The existing CSCM model mainly
includes: independent and identically distributed (IID) channel model with mini-
mum complexity, Kronecker product based CBSM (KBSM), VCR [25], and
Weichselberger channel model [26].
IID model assumes that all Tx antennas and Rx antennas are independent.
Obviously this assumption does not in line with the actual situation. This model
is commonly used to show the performance advantage of MIMO than SISO. KBSM
assumes that Tx and Rx have independent correlation of antennas, namely ignoring
the coupling between Tx and Rx antennas. MIMO wideband access system (IEEE
802.11 n, 802.16 m/ac) using this model. Both VCR and Weichselberger model
consider the coupling between Tx and Rx, but both require the calculation of
eigenvalue decomposition. It is quite cumbersome to use, especially in the appli-
cation of large-scale antenna array. Therefore some more simplified models have
been proposed. For example, J. Hoydis et al., proposed a CBSM for the configura-
tion that the BS is equipped with massive antennas and the user has a single antenna
[27]. When the antennas number in MIMO Massive system increase extremely, the
mutual coupling between antenna elements cannot be ignored. Masouros et al.
propose a new model which considers the effect of antenna impedance, load
impedance, and mutual coupling impedance matrix [28]. For more detailed analysis
of antenna coupling, please refer to [2931] and their references.
IEEE 802.16m document has given two specific calculation methods to convert
GSCM parameters to antenna correlation matrix. One method is to use 20 sub-paths
of each cluster to approximate the Power Angle Spectrum (PAS) with shape of
Laplacian function, and to calculate the antenna correlation matrix by using
AOD/AOA parameters and antenna geometric relations [32].
3.2.4 Extended SV Model
R. Valenzuela and A. M. Saleh of the AT&T Bell Labs proposed the SV model by
analyzing the indoor channel measurement data [33]. The channel is composed of
multiple clusters and rays. The number of clusters and the number of rays within a
cluster obey Poisson distribution, namely the arrival of the cluster and the intra-
cluster rays are two Poisson processes with different arrival rates. The arrival
intervals of clusters or intra-cluster rays obey exponential distribution with different
means. The average power of the cluster or intra-cluster ray decay exponentially
with delay. The instaneous power of the clusters or intra-cluster rays obey Rayleigh
distribution or lognormal distribution. The initial SV model can only describe the
delay information of the channel, and its resolution is higher than 10ns subjected to
the sounding equipments. After that, scholars continued to improve the time
resolution of the model to a nanosecond or less with the help of many measurement
in ultra wide band and mmWave band. Moreover, angle information and polariza-
tion characteristics are also integrated to develop the extended SV model. Herein-
after the IEEE 802.11ad channel model is taken as an example to show the detail of
the extended SV model. The CIR, h, can be expressed as [9]
X
i i
ht; tx ; tx ; rx ; rx Hi Ci t T i ; tx tx ; tx tx ; rx rxi ; rx rxi
i
X
i;k
Ci t; tx ; tx ; rx ; rx i;k t i;k tx tx
k
i;k i;k
tx tx rx rx rx rxi;k
3:4
Where t, tx, tx, rx, rx are time, azimuth angle of departure, elevation angle of
departure, azimuth angle of arrival, elevation angle of arrival, respectively. H(i) is a
2 2 polarization matrix. C(i) is the CIR of the i-th cluster. () denotes Dirac
impulse function. T(i), tx(i), tx(i), rx(i), rx(i) are the delay, arrival and departure
angle of the i-th cluster respectively. (i,k) is the amplitude of the k-th path in the i-th
cluster. (i,k), tx(i,k), tx(i,k), rx(i,k), rx(i,k) are the delay and angle value of the k-th
path in the i-th cluster relative to the central ray within the cluster respectively. The
intra-cluster rays are divided into forward part and backward part, represented by
subscript f and b respectively. The distributions of all of the parameters are as
follows.
h i
p T i jT i1 exp T i T i1 , l > 0

i;k i;k1 i;k i;k1

p f j f
f f exp f f
b b b b b b
,k > 0 3:5

i;k 2 i;k
f =b j0;0 j K f =b exp T i = exp f =b = f =b
2
The arrival of clusters and intra-cluster rays obey the Poisson distribution, where
and f/b respectively are the arrival rate of clusters and arrival rate of intra-cluster
forward/backward rays. and f/b respectively are the power decay rate of the
clusters and the forward/backward rays. The average power of the k-th path in the i-

i;k 2
th cluster is denoted as f =b , which is also the mean of instantaneous rays power
that obey Rayleigh distribution or lognormal distribution. Kf/b denotes the Ricean
K-factor of forward/backward rays. For the angle of clusters and intra-cluster rays,
different models have different assumptions. IEEE 802.15.3c deems the angles of
clusters are uniformly distributed between 0 and 2; the azimuth angles of rays
within a cluster follow Gaussian distribution or Laplacian distribution [8]. IEEE
802.11ad uses ray tracing method to determine the azimuth and zenith angle of
clusters. The azimuth and zenith angles of the intra-cluster rays are Gaussian
distributed with zero mean and variance of 5 degrees [9]. The extended SV
model is also independent of the antenna just as GSCM, so the MIMO channel
model can be obtained by applying specific antenna information.
3.2.5 Ray Tracing based Model
The propagation prediction of electromagnetic wave can be solve by analytical

method based on electromagnetic theory or by corresponding computational elec-
tromagnetics (MOM, FDTD, etc.). However, both methods are hard due to high
complexity under the actual propagation environment. Ray Tracing (RT) method
which based on relatively simple Geometrical Optics (GO), Geometric Theory of
Diffraction (GTD) and Uniform Theory of Diffraction (UTD) is therefore used to
predict the propagation properties of radio wave. RT based channel modeling have
high flexibility which channel measurements cannot achieve, and it can support
modelling for D2D, massive MIMO, arbitrary frequencies, wide frequency band,
and spatial consistency. Its basic principle is to use rays to simulate the electro-
magnetic wave propagation process, such as direct ray, reflection, scattering, and
diffraction. By collecting the geometric dimension and electromagnetic properties
including complex relative permittivity and conductivity of materials, the propa-
gation environment can be constructed into a 2D or 3D digitized map composed of
geometric structures and objects. Searching the propagation paths from the trans-
mitter antenna to receiver antenna, RT method calculates path gain and shadowing
using classical electromagnetic theory, and further obtain other channel parameters
including path delays, angle, and CIR using antenna radiation pattern [34]. Calcu-
lation the field strength of rays and path tracking of rays are two main aspects of
Ray tracing technique.
RT treats the propagation of electromagnetic wave by means of rays propaga-

tion. Based on three well-built propagation theory of optics (GO, GTD, and UTD),
RT can describe the basic propagation behavior of rays, such as line of sight,
reflection, transmission, diffraction and diffuse scattering., and calculate the elec-
tric field intensity contribution of each ray. GO is only suitable for direct ray,
reflection and transmission of electromagnetic waves. It fails for diffraction arising
when electromagnetic waves encountering discontinuous surfaces of objects. Keller
presented the GTD theory by classifying the diffraction as edge diffraction, wedge
diffraction and curved surface diffraction, and gives the formula to calculate
diffraction coefficients under the three conditions. The field strength of diffraction
wave can be obtained through multiplying the field strength of incident wave by the
diffraction coefficient at the diffraction point. While GTD theory has the disconti-
nuity problem of electromagnetic field in the transition region of diffraction,
namely between the bright and dark region of diffraction objects. The Uniform
Theory of Diffraction (UTD) proposed by Pathak and KouyoumJian overcomes the
disadvantage [35].
Ray tracing can be divided into forward and backward algorithms. The forward
algorithms include test ray method and Shooting and Bouncing Ray (SBR). The
backward algorithms include mirror method, minimal light path method, etc. The
most widely used SBR method needs to emit a large number of dense three-
dimensional cone ray tubes into different directions, covering the entire three-
dimensional spherical space with the transmitting source as its center. Each ray
tube is ergodic of all obstacles in scenario, and undergoes reflection, diffraction
propagation until the reception point is contained in the ray tube. Mirror method
bases on geometric optics. The source and the reception antennas are regarded as
points in geometry, and the mirror point on the reflection plane that corresponds to
the source point is found through the plane symmetry method, and then the
propagation path of rays in space is obtained. The complexity of the mirror method
can increase exponentially with the increase of the complexity of the scenario or the
order of mirror images, so it can only be applied to the scenarios with a few
obstacles and simple geometric shapes. It cannot describe and calculate diffraction
and scattering.
Adding one-order or two-order diffuse scattering into RT, namely so called
effective roughness (ER) approach, can model the dense multipath component
(DMC) and obtain PDPs which matched well with measurements [36]. Since RT
does not need high-cost channel measurement devices and hard measurement
compaigns, it have gathered great interest. Several novel 5G channel model were
proposed by combining the RT model and the statistical model based on measure-
ment, e.g., the Q-D model proposed by MiWEBA [4]. In the IEEE 802.11ad/ay and
MiWEBA model, RT modeling method is used to generate the cluster parameters of
60 GHz mmWave channel, and the inter-cluster parameters are obtained by channel
measurements. METIS project first proposed a systematic channel model based on
RT, which considered the reflection, diffraction, diffuse reflection, blockage and
other propagation mechanisms, and claimed that the model could satisfy all of the
requirements of 5G system channel models (see Table 3.1 or Table 4.1 [1]).
Table 3.1 Channel modeling methods adopted by a variety of channel models

GSCM 3GPP SCM/3D/D2D, SCME, WINNER(I, II and +), IMT Advanced, COST
259/273/2100/IC1004, IEEE 802.16m, QuaDRiGa, METIS(GSCM)
CSCM IEEE802.11n/ac, IEEE 802.16m
Extended SV IEEE802.11n/ac, IEEE 802.15.3c, IEEE 802.11ad/ay, MiWEBA
RT IEEE 802.11ad/ay, METIS(Map), METIS(GSCM with Point cloud field
prediction), MiWEBA
3.2.6 Comparison of Modeling Methods
There are pros and cons for GSCM, CSCM, Extended SV and RT modeling
methods. The first three types could separate the antennas from the propagation
channels, thus the transmission channel can be obtained by applying any type of
antennas with the propagation channel. GSCMs and extended SV model use a
cluster-ray structure to describe the propagation channel. Each cluster is composed
of multiple rays, and the rays within one cluster have similar amplitude, phase,
AOA and AoD. The RT channel model is expressed with multiple rays. In CSCM,
each tap has its antenna correlation matrix, which may restrain the angle informa-
tion of rays. Specific antenna correlation matrics are required to be calculated for
different configurations and radiation pattern of the antennas, which make the
model lack of flexibility.
By rational parameterization, GSCM can describe different scenarios accurately
and flexibly. GSCM model does not directly output the antenna correlation matrix.
But if needed, it can be obtained from channel path parameters [32]. The accuracy
of RT based model is closely related to the quality of electronic maps built for the
enviroments. Some notable scatterers (e.g., buildings, desk, etc.) are easy to be
modeled, but small irregular objects (such as trees, shrubs, and street lamp) may be
ignored or cannot be accurately modeled which leads to improper results.
MiWEBA model need channel measurement to obtain the statistical features of
small scale parameters. Extended SV is suitable for channel modeling with high
bandwidth and high delay resolution. It could give intuitive and accurate descrip-
tion of densely arrival of broadband signals in the indoor and outdoor environment.
However, SV model lacks the support for spatial consistency.
The advantage of CSCM model is that it can provide spatial correlation matrix
which make analyzing MIMO theoretical performance easier. However, it has some
disadvantages. It makes over-simplification on the true channel. There is big
difference in channel correlation matrices for LOS and NLOS, which make smooth
transiting from each other impossible. Correlations in time-domain and space-
domain are independent, which make the possible existed temporal-spatial corre-
lation be hard to describe. The characteristics of LSPs, such as DS and AS changing
over time and locations, are difficult to be captured in the model, because these
parameters are hidden in the channel correlation matrix. CSCM is difficult to model
3.3 Channel Measurement 61
the spatial non-stationary phenomenon and the spherical wavefront in near field
propagation.
In simulation, the computational complexity of a channel model is very impor-
tant. When talking about the channel modeling method, people used to consider that
CSCM has lower complexity and higher computational efficiency than GSCM
because of its relatively simple concept. The complexity of WINNER model
(GSCM) and CSCM were revealed and compared in [37]. The number of GSCM
sub-paths was set to 10 or 20, and the tap with a certain Doppler spectrum were
generated by 8-order IIR filters in CSCM. It was found that the complexity
(represented as the number of real calculations at each tap) of them is the same
order of magnitude when the number of MIMO antennas is small. When the number
of MIMO antennas is larger (>16), the complexity of CSCM is surprisingly higher.
RT calculation is too complicated to generate the CIRs in link-level and system-
level simulation in real-time. Off-line calculations of CIR on each UT location are
required to establish a CIRs database for simulation, which will ask for huge storage
resources. In addition, the computational quantity of channel convolution in sim-
ulation is very large and usually several times of that of CIR generation.
In addition to the above several channel modeling methods, recently, some
scholars presented propagation graph theory for channel modeling. It takes the
positions of transmitter, receiver, and scatterers, and propagation coefficient
between them as input, establishes three transmission matrics for expressing the
propagation from transmitter to the scatterers, between scatterers, and from scat-
terers to receiver, and calculates the channel matrix analytically. The location of the
scatterers can be gained by measurement or ray tracing technology and propagation
coefficient can be obtained by theoretical calculation based on GO, GTD, UTD. ER
method can be combined with propagation graph to model the channel and obtain
the PDP conformly alined with the measurement [38]. Compared with the ray
tracing, propagation graph method is less computational, so that it can generate
random channel samples in real-time.
The channel modeling methods used by a variety of standardized organizations
or project teams are shown in Table 3.1. From the table, we can find that GSCM was
widely used in 3G / 4G channel modeling. With the progress of 5G R&D, more and
more groups combine ray tracing and measurement to obtain the channel models,
such as METIS, MiWEBA, 5GCM channel models and son on.
3.3 Channel Measurement
Channel measurement, also known as channel sounding, is a kind of research

activity which aims at extracting the channel parameters for acquiring the knowl-
edge of wireless channel. Channel measurement is a direct and most important
mean to obtain channel information and understand the characteristics of the
channel. The transmitter emits sounding signals to the receiver through the wireless
channel to be investaged. By exploiting the received signals, multiple propagation
paths can be identified. The parameters of every resolvable path are outputted, with
which a variety of important channel statistical information can be obtained, such as
the angular distribution, coherence time, DS, etc. Sometimes it is enough to use the
measurement results for transmission performace analysis. But more often, it is
better to make a general or representative description of a certain type of propaga-
tion scenario. In this case the channel models are needed, and channel measurement
can provide measurement data for channel modeling and verify the effectivity of
the model.
The differences between the channel measurement methods exist in many
aspects, such as the form of the stimulating signals, the number of RF channels,
the types and number of antennas, etc. This section is going to introduce various
measurement methods in a unified framework, and several common channel sound-
ing systems in academia and industry, as well as various channel measurement
activities.
3.3.1 Channel Measurement Methods
Large bandwidth and multiple antennas are the main features of 5G, so in this
subsection we take the broadband MIMO channel as the sounding target. According
to the numbers of channels being measured simultaneously, channel measurement
methods can be divided into Single Input Single Output (SISO) [3943], Multiple
Input Multiple Output (MIMO), and the compromising Single Input Multiple
Output (SIMO) [44]. MIMO method simultaneously measures multiple channels
in parallel. It has the highest measurement speed and can be applied to the high-
speed mobility scenario. However, multiple parallel RF units are required for
MIMO measurement method. Besides the complex structure and high cost, the
inherent requirements of consistency and synchronization increase the difficulty of
system implementation and calibration. In addition, multiple concurrent sounding
signals can interfere with each other. Even when orthogonal sequences are used as
the excitation signals, under multipath channel conditions, the orthogonality
between these sequences with different delay is hard to be guaranteed. The inter-
ference will still be produced, so that a more complex processing algorithm is
needed. SISO only needs one RF chains at the transmitter and the receiver respec-
tively, and adopts the Time Division Multiplexing (TDM) mode to sounding the
subchannel iteratively. There is one electronic switch box at both ends to synchro-
nously switch different antenna element connecting to the RF chain. This method
has relatively simple structure and low cost. The number of antennas at transmitter
and receiver can be flexibly configured. Because there is only one RF chain, the
calibration and consistency between the multiple channels can be achieved much
easier. In addition, since only one probe signal is transmitted, there is no demand for
the orthogonality. However, its measurement time is generally longer, which is not
suitable for the measurement of high mobility scenarios. The switch boxes on
transmitter and receiver act according to a designed sequence. For example, the
Receiver
Transmitter Multi- Multi-
Down-converter
channel channel
Up-converter Low- Storage
parallel parallel or pass or Display
Signal or Multi- band- A/D
Post-
Multi- plexing pass
processing
plexing (switch)
(switch)
Local Local
Clock Reference Clock Reference GPS/OCXO/Rb/Cs
Fig. 3.6 Illustration of a unified modular MIMO channel sounder
possible switching sequence is as follows. The Tx box switches and keeps connec-
tion to one transmitting antenna when the Rx box switches to each of the receiving
antennas and followed by subchannel sounding sequentially. After that the Rx box
switches from the last antenna back to the first antenna when finishing sounding a
group of subchannels, at the same time the Tx box switches to next antenna and
begins sounding next group of subchannels. SIMO is a compromise between SISO
and MIMO methods. It uses switched signal transmission and parallel signal
reception, and obtain a certain balance among system complexity, cost, and mea-
surement speed. Although a few systems [45, 46] have MIMO sounding ability in
terms of the system structure and the principle of operation, they are SIMO in
nature. It is possible to utilize the advantages of both MIMO and SISO, and extend a
MIMO sounder to support larger arrays through a switch system.
A unified modular channel sounder, as shown in Fig. 3.6, is used as a framework
to illustrate various methods, techniques, and devices in the channel measurement.
We mainly focus on four aspects of the measurement system: antenna system
(including type and switching of antenna), RF unit (local oscillator and up/down
converter), sounding signal (baseband part), and transmitter/receiver synchroniza-
tion method.
1. Antenna and antenna array
The transmitting and receiving antenna elements used in sounding have many
types. In order to ensure the radiation and collection of radio signal in all the spaces
and angles, omnidirectional or low gain antennas are adopted, such as half-wave
dipole antenna, patch antenna, diskcone antenna, biconical antenna and open-ended
waveguide (OEW), etc. In order to ensure the sufficient signal strength, directional
antennas are adopted, such as standard gain horn antenna, parabolic antenna, lens
antenna, etc. Directional antennas can only guarantee the signal sounding of certain
angles. In order to achieve coverage of all space angles, multiple directional
antennas splicing in different directions or rotating a directional antenna for scan-
ning is needed. When the polarization characteristics of the channel need to be
detected, the dual polarization antenna element is necessary.
Table 3.2 antenna array type set 1 [13]

Name ODA_5G25 PLA_5G25 UCA_5G25 SPH_5
Manufacturer Elektrobit Elektrobit Elektrobit HUT(Aalto)
Array type Omnidirectional Rectangular Uniform circu- Hemisphere
array array lar array
Polarization Dual polariza- Dual polarization Perpendicularity Dual polarization
tion (+/- 45 ) (+/- 45 ) (H/V)
Center fre- 5.25 5.25 5.25 5
quency [GHz]
Number of 50 (25 dual) 32 (16 dual) 8 32 (16 dual)
antenna
elements
Element type Patch Patch Patch Patch
Picture
The resolution of direction or angle of the multipath channel is closely related to

the structure of the antenna array and the number of the antenna elements. In order
to obtain high angle resolution, large antenna array is usaually adopted in channel
sounding. The antenna array can be formed in two types: virtual antenna array and
real antenna array. The virtual antenna array shifts only one antenna element to the
required locations (directions) through spatial translation or rotation by electrome-
chanical device. CIR is measured independently at each location. For virtual
antenna array, only serial sounding mode can be used. In existing channel mea-
surement systems such as Elektrobit (EB) Propsound and Medav RUSK MIMO
channel sounder, the real antenna arrays are widely used like linear array, area
array, circular array, spherical array, quasi spherical array, etc., as shown in
Table 3.2 and 3.3. In the mmWave sounding system, the antenna element rotation
and translation are often used (e.g., Table 3.4) to form a virtual antenna array.
Recently, the mmWave channel sounding is carried out by using the electronically
controlled antenna array [4]. Such modular antenna array can be used to form the
scanning beam with different directions in the space via phase control. In addition
to the channel sounding, this antenna can quickly and adaptively change the
direction of beam focusing, which can be applied to the actual mmWave commu-
nications system. Real antenna array can be used in SISO, SIMO or MIMO
sounding. When its used in SISO or SIMO sounding, the switching box is required
to switch between different elements or polarization elements. Real antenna array
system can achieve high speed sounding, which is suitable for sounding in multiple
locations and mobile scenarios. Since the mechanical moving speed is much slower
Table 3.3 antenna array type set 2 [13]

Name PULA8 UCA16 PUCPA24 SPUCPA4 24
Manufacturer IRK Dresden TU Ilmenau IRK Dresden IRK Dresden
Array Uniform linear Uniform circular Uniform circular Multi uniform
structure array array array circular array
Polarization Dual (V/H) Perpendicularity Dual (V/H) Dual (V/H)
Center fre- 5.2 5.2 5.2 5.2
quency
[GHz]
Bandwidth 120 120 120 120
[MHz]
Maximum 27 (40) 27 25 24
power [dBm]
Number of 8 16 24 96
elements
Array type Patch Diskcone Patch Patch
Size Unit interval Diameter 10.85 Diameter 19.5 Diameter 19.5 cm,
0.4943 cm cm ring interval
0.4943
element
direction
Picture
Table 3.4 Millimeter wave antennas

Name LB-22-25 NICT [47] MD249-AA OEW 2650
Manufacturer Ainfoinc Flann Satimo
Polarization Linear Linear Linear Linear
Frequency 33~50 GHz 50~75 GHz 50~75 GHz 26.5~40 GHz
[GHz]
Antenna type Pyramid standard gain Conical horn Monopole OEW
horn antenna antenna
Beam width 8.6 /7.5 30 360 /60 64.5 /
(Azimuth/ 112.45 @26.5 GHz
zenith) 53.3 /
71.7 @40 GHz
Picture
than the switching speed of the electric switch, the virtual antenna array is only
suitable for the measurement with static or low speed. In addition, in the measure-
ment system using real antenna array, when a new frequency band is measured, not
only a new antenna array must be developed, but also a full set of complex
calibration need to be done for it. In this aspect, the virtual antenna array only
needs one antenna element, so it is much easier in terms of the cost and the
calibration complexity.
2. RF unit
The important parameters of the RF unit include the number of RF channels, RF
frequency range, and RF signal bandwidth. Parallel sounding mode requires mul-
tiple parallel RF units. It is necessary to consider the consistency and mutual
interference of the RF units. Presently, most models have designated the applicable
frequency range or sounding bandwidth, such as 3GPP SCM is 1~3 GHz/5 MHz,
WINNER II/+ and IMT-A 0.45~6 GHz/100 MHz, 3GPP D2D and 3D 1~4 GHz/
100 MHz. For sub-6 GHz bands, the existing channel models supporting 100 MHz
signal bandwidth seems to be enough, but larger bandwidth requirement will
continue to emerge. For example, the highest bandwidth in IEEE 802.11ac is
160 MHz. Therefore, the channel models have to support higher bandwidth, such
as more than 500 MHz. In the 60 GHz mmWave band, the bandwidth of sounding
system which was used for the development of IEEE 802.11ad is 800 MHz [48]. In
the future mmWave applications, the bandwidth of the signal waveform may
exceed 2 GHz, which requires that the RF unit of the mmWave channel sounder
should be able to support bandwidth larger than 2 GHz.
In the mmWave channel sounding, due to the cable loss of high frequency signal
is great, many sounding systems use an independent external mixer which is close
to the antennas. Lower frequency excitation signal is transmitted to the mixer
through a cable. The lower frequency receiving signal outputted by the
downconverter is transmitted to the signal analysis system through the cable. The
sounding system based on the Vector Network Analyzer (VNA) requires special
considerations: (1) the local oscillator (LO) signals for the mixers at both transmit-
ter and receiver should be from the same local oscillator source, so that the
consistency of reference frequency and phase in each measurement frequency
point can be guaranteed. (2) Sometimes in order to obtain a longer sounding
distance, a long cable is used. Even the low frequency signal still has huge
propagation loss, the signal strength need to be guaranteed by using an electric
power amplifier or fiber optic [1, 49].
3. Sounding signal and detection technology
Different sounding signals have large impacts on the performance of sounding
systems and correspond to different parameter extraction algorithms. How to
choose sounding signals to achieve the optimal performance of the channel sounder
is very worthwhile to explore. The pros and cons of a sounding signal can be
evaluated from several aspects, such as, signal duration, signal bandwidth, time-
bandwidth product, power spectral density, peak-to-average power ratio (PAPR)
and correlation property. Belows are some commonly used channel sounding
signals.
(a) Periodic pulse signal

Pulse signal (high bandwidth) is the most intuitive choice for channel sounding.
Shortening the sounding signal pulse width can improve the path delay resolution.
However, the narrow pulse width of sounding signal causes the high peak power
and arsing serious nonlinear distortion in circuit. One has to reduce the signal power
for lower distortion, as a result the sounding system has limited measurement range.
For digital realized broadband sounding system, according to Shannon-Nyquist
sampling theorem, the receiver should have an ADC with ultra high sampling rate,
which will be highly costy expensive. Tamir et. al, suggest using compressed
sensing technology to reduce the sampling frequency [50]. In the early stage of
channel sounding, periodic pulse was used more extensively. Later, the spread
spectrum sequence and signal with flat envelope in frequency domain have been
more widely used.
(b) Time-domain flat signal
The time-domain flat signal is the signal whose envelope is relatively flat in the
time domain, which is different from the signal whose envelope is flat in frequency
domain. These signals include the sequences with good periodic autocorrelation
properties, such as Pseudo-random Noise (PN) (e.g., M sequence), Chirp
sequences, Zadoff-Chu sequences (also the frequency-domain flat signals), etc..
Systems applying this type of time-domain signal have been demonstrated excellent
performance in delay resolution and system dynamic range, so it has been widely
used, such as the Propsound channel sounder.
Two different processing methods can be used in the receiver when the PN
sequences are sent in transmitter. One method is to use correlator or matched filter.
Related operations are performed at each sampling time, and the CIR is directly
obtained, or the signal is first recorded and then processed. In recent years, some
high time and spatial resolution parameter extraction algorithms, such as RIMAX,
SAGE, etc., have been widely used in channel measurement. This processing
method requires the high-speed ADC to sample and perform analogue-to-digital
conversion, as well as store a large number of data. Another method is sliding
correlation technology. Transmitter and receiver use the same PN sequence, but
with very small chip rate difference. Due to different chip rates (clock frequency),
the PN sequence of transmitter and receiver will produce relative sliding. When the
two sequence chips realign due to propagation delay, maximum correlation peak
will be produced so as to achieve the despreading of the spread spectrum signal.
The advantage of this technique is that it does not require high speed sampling, and
Newhall et al. [43] use this technique to sound mmWave channels. However, since
the single measurement time is determined by the clock frequency difference of
sending and receiving PN sequences and the maximum DS length. If the frequency
difference is small, i.e., the relative sliding of the two PN sequences is slower, a
longer time is needed to complete a measurement. And hence this method is not
suitable for the sounding of high mobility scenario.
(c) Frequency-domain flat signal

The frequency-domain flat signal is the signal whose envelope is relatively flat in
the frequency domain, which is different from the signal whose envelope is flat in
time domain. Early narrowband channel sounding techniques use sinusoidal signal
as the sounding signal. After the signal goes through the propagation channel, the
changes of signal amplitude and phase are the channel frequency response at this
frequency. In the VNA-based broadband sounding system, the whole sounding
frequency band is divided into a number of frequency points. At each frequency
point, sinusoidal wave sounding signal is still used, in order to scan all the
frequencies. This sweep frequency measurement method is generally slow, which
is not suitable for measurement of dynamic scenarios. At different frequency
points, the RF units at the transmitter and receiver must ensure the consistent
frequency and phase. In order to overcome the limitation of the sweep frequency
method, broadband frequency domain measurements commonly use the multi-
carrier (OFDM) signal, which is used to detect the whole frequency band at a
time. In addition, frequency domain excitation symbol X( f ) may be the PN
sequence composed of {+1, 1}. And the frequency response can be obtained
very easily without division. However, it should be noted that the multi-carrier
signals need to be carefully designed to have a lower level of PAPR.
4. Synchronization method
The sounding system has reference clocks both at the transmitter and the
receiver, from which the baseband clock, ADC/DAC sampling driving clock, and
LO (may use a different reference clock), are obtained by means of frequency
doubling, phase lock, etc. The transmitter and receiver need to ensure the consis-
tency of baseband clock, sampling clock, LO frequency and phase. If they are
inconsistent and cannot be accurately estimated and compensated, the difference
parts will be regarded as a part of channel and thus the outputted channel param-
eters will be inaccurate. Therefore, the sounding system should minimize the error
in reference clocks of transmitter and receiver, or it should try to use one reference
clock. This requirement is easy to achieve in the short distance sounding (such as
indoor sounding). For the VNA-based broadband sounding system, if one VNA is
used for both transmitting and receiving, the clock sharing between channels should
be achieved within the VNA. If both signal source and VNA are used for transmit-
ting and receiving, the reference clocks of the two devices should be connected by
cable. Due to the use of single frequency step sweeping, the consistency of the
phases at both ends with different frequencies should be ensured, which requires
that the RF signals of the signal source should be connected to the VNA as a
reference with coupler and long cable. The coupling mode can be electrically
amplified, or it can be realized by means of photoelectricity/electro-optical con-
version and fiber optic extension [52]. In outdoor mobile environment, the shared
reference clock usually uses GPS-tamed Oven Controlled Crystal Oscillator
(OCXO) or the atomic clock with high precision and high stability (rubidium or
cesium clock), or both. Even so, it still cant avoid the deviation of the clock. And
the jitter or phase noise of each clock will affect the sounding performance.
In addition, when SISO or SIMO is used for MIMO sounding, the synchronous
switching of the switch arrays at the receiver and the transmitter must be strictly
controlled. The clock of the switching control circuit is also derived from the same
reference clock.
5. Channel sounding system
The following is a summary of some commonly used channel sounding systems,
as shown in Table 3.5. The early systems that undertook channel sounding tasks for
4G or earlier communications systems only paid attention to frequency bands
below 6 GHz. These systems include Propsound [39] (stop production), RUSK
[40], the channel sounder of Aalto University (Aalto U., formerly known as the
Helsinki University of Technology, HUT) [41] the channel sounder of NTT
DoCoMo [42], etc. Many systems are applied in measurement tasks of various
standard channel models, e.g., Propsound, Medav RUSK and the Aalto University
system are widely used in the development of the channel modeling of COST,
WINNER, IMT-Advanced, 3GPP 3D, etc. In recent years, these systems are also
used in the channel sounding of massive MIMO. With the increase of the demand
for the sounding of more high frequency band, the broadband mmWave sounding
systems [1, 43, 48] and VNA-based measurement systems [1, 49, 51] were devel-
oped for a specific frequency (customization). Of course, the precious sounding
system can also extend frequency to support the mmWave sounding. The
VNA-based measurement system is often used in the sounding of massive
MIMO. From the consideration of reducing the cost in hardware and calibration,
most of the existing systems use a RF unit. But there are also some sounders with
multiple receivers (such as Durham University sounder [45]) and some sounders
with multiple receivers and multiple transmitters such as the Tokyo Institute of
Technology (TIT) sounder [45, 46], etc.
Figure 3.7 shows the MIMO channel sounding which is achieved using SISO
method by sequential switching of the receiver and transmitter antenna (SISO).
Here a simple analysis of the capability of the dynamic channel sounding of the
system will be shown. Lets take that Nt denotes the number of transmitting
antennas, Nr denotes the number of receiving antennas, and max denotes the
channel maximal excess delay. Take the PN sequence as the sounding sequence
and SISO sounding as an example. A periodic sounding sequence is composed of
K chips. The length of the chips is Tp, and the length of the sequence is TaKTp. A
longer sounding sequence Ta can get greater processing gain (see SAGE algorithm
in Sect. 3.4), but it also limits the measurable Doppler shift. The time span To of one
sounding should equal the sum of the length of sounding sequence Ta and the
maximal DS max, i.e., ToTa+max. If we take into account the antenna switching
time Tw, following the assumption in the RUSK, the measurement time and
switching time are equal, i.e., TwTo. The measurement time of a subchannel is
2To. The time spent in a snapshot to iterate all of the sub-channels is Tsnap2ToNt
Nr, i.e., the time in which a full channel sounding is carried out. For time-varying
channel, the sounding period is denoted as Tsnap. If the gap between snapshots is
zero, i.e., TfTsnap, the maximum Doppler shift that can be measured by the system
70
Table 3.5 Channel sounding system parameters

Channel Sounding
sounding RF Frequency Bandwidth Antenna units Antenna array Antenna signal Synchronization
system units range (GHz) type type units type method
Propsound Tx1 <6 GHz, 0.2 Arbitrary (monopole, dual polar- Fixed array, can Arbitrary PN or GPS, OCXO,
RUSK Rx1 scalable 0.24 ized, diskcone, biconical, patch, be customized Limited OFDM Rubidium clock,
Alto U. 0.2 horn antennas), often omnidirec- by sound- cesium clock
tional antennas ing speed
DOCOMO 0.05
Durham U. Tx 1 ~0.06
Rx 8
TIT U. Tx1, 11 GHz, 0.4 Arbitrary, can be customized Arbitrary
Rx1 can be
modular, customized
scalable
Broadband Tx 1 >6G <1 Short distance, omnidirectional or Rotate or trans- A few
high fre- Rx 1~2 directional; late 2D/3D to
quency Long distance, often directional form virtual
sounder antennas array
VNA-based Tx 1 Arbitrary >1 1 Frequency LO extension
sounding Rx 1 scanning [1, 42]
system Transmitted sig-
nal extension
[4447]
3 Channel Measurement and Modeling
Tx switching
Rx switching
2To
2NrTo
Tsnap=2NtNrTo
Tf
Tsnap
Fig. 3.7 Timing sequence in MIMO channel sounding via switching transmitting and receiving
antennas
is fD 1/Tsnap/2 1/(4Nt Nr)/(KTp+max). Thus it is can be seen that, in the system

sounding capabilities, large Doppler shift and more antennas/ large processing gain/
large DS are mutually exclusive. To obtain support of larger DS (larger coverage),
the system has to sacrifice the mobility or the number of channels or the processing
gain. To ensure support of mobility, the system has to sacrifice the sounding range
or number of channels or the processing gain. Such contradictions mainly exist in
the mobile environments in which either the transceivers or the scatterers move. It
is essential to carefully balance various measurement objectives in the measure-
ment process. In addition, the channel
pcoherence
time can been roughly estimated
in terms of Doppler shift, T c 9=16 =f D , and a snapshot should be completed in
that time, i.e., Tsnap<Tc
3.3.2 Channel Measurement Activities
In recent years, with the ongoing progress in 5G R&D, many communication

enterprises, universities, and research institutes all over the world have carried
out extensive channel measurement campaigns. These measurements are divided
into three categories, namely the measurement of Massive MIMO, D2D/V2V/HSR
and mmWave band. The measurement campaigns, parameter settings and results
are briefly reviewed as follows.
3.3.2.1 Massive MIMO measurement
Massive MIMO is a promising technology for 5G communication. The measure-

ment activities related to Massive MIMO are carried out by various institutions
shown in Table 3.6. The main research institutions include the Lund University in
Sweden, Aalborg University in Denmark, Beijing Jiaotong University in China, etc.
In recent years, the reseachers of Lund University had completed a lot of work
on massive MIMO channel measurements and modeling. They used Medav RUSK
with real antenna arrays and VNA with virtual antenna arrays to perform the
measurements. Some new propagation phenomena were observed, which greatly
promote the understanding of Massive MIMO channel and the development of
channel model. Their works are outlined as follows002E
In [52], the BS was placed indoors and configured with a polarized uniform
cylindrical array with 128 ports (4*16 patch antenna elements, each with two
polarized input ports). Only one element in a polarized uniform antenna array
with 32 ports (2*8 patch antenna units, each with two polarized input ports) was
used as the user antenna. The channel correlation between two users was directly
analyzed and found that it would decline rapidly (orthogonality will increase) when
the number of BS antennas increased. When there were 20 antennas at the BS, the
performance of linear pre-coding was close to the performance of optimal dirty
paper coding.
In [53], measurements for controlled outdoor scenario in both LOS and NLOS
conditions were carried out by using a VNA based system with 2.6 GHz carrier
frequency, 50 MHz bandwidth and 1601 frequency points. An approximately
omnidirectional and perpendicularly polarized antennas were used at both side,
while at the receiver side the antenna was moved along a 7.3 m straight sliding trail
with step of half wavelength to form a virtual antenna array with 128 elements. The
path gain, Ricean K-factor, and PAS were analyzed and found obvious changes in
the PAS of the incident wave along the linear array. The specific non-stationary
characteristic and near-field effects of massive MIMO were demonstrated and both
would helped to reduce the correlation between the users. Gao et al. extended the
COTS 2100 channel model to support the newly discovered propagation charac-
teristics [18]. The statistical model of cluster number as well as the visibility region
and visibility gain of clusters at the BS side was obtained from the measured data.
In [54, 55], measurements were carried out outdoor with a larger space using two
systems at in same position, and the results were compared. A uniformly cylindrical
array (UCA) with 128 ports the same as [52] and a virtual uniformly linear array
(ULA) with 128 position the same as [53] were used at the BS, while a perpendic-
ularly polarized omnidirectional antenna was used at the UT. The measurement
results of using the ULA are similar with the results in [53]. The power variation
over the antenna arrays that is critical to the performance of massive MIMO were
observed. For a ULA, the power variation came from different large scale fading
that each element experienced. While for UCA, these large power variation were
resulted from the polarization and directional radiation pattern of each patch
Table 3.6 Summary of massive MIMO measurement activities
Channel Frequency/
Research institutes Scenario Antenna configuration sounder Bandwidth parameters
Lund University in O2I [52] Tx: 1 patch; RUSK 2.6 GHz/50 MHz channel correlation between users
Sweden Rx UCA(128 patch)
Outdoors [53] Tx: 1 omnidirectional VNA 2.6 GHz/50 MHz path gain, K- factor, PAS, eigenvalue, channel
Rx: virtual ULA (128 omnidi- correlation between users
rectional), 7.3 m long
Outdoors Tx: 1 omnidirectional; VNA 2.6 GHz/50 MHz Large-scale fading, angular resolution, singular
[54, 55] Rx1: virtual ULA(128); RUSK value spread [55]
Rx2: 128 UCA
Outdoors [56] Tx: 8 omnidirectional RUSK 2.6 GHz/40 MHz Singular value spread
Rx: UCA(128 elements)
Alcatel-Lucent Outdoors [57] Tx: UCA (112 patch) Customized 2.6 GHz/20 MHz Correlation, inverse condition number
Bell Labs in Rx: 2 omnidirectional (2 m (LTE)
German spaced)
Aalborg Univer- Indoors [58] Tx: 8*2 patch (6 orientations Customized 5.8 GHz/ Correlation matrix, channel singular value,
sity in Denmark and combinations) 100 MHz condition numbers, power variation over array
Rx: 64 omnidirectional
(3 array modes)
BJTU Indoors, Tx: 128 virtual ULA(~12.9 m) VSG 1427~1518 MHz Small scale parameters
outdoors [59] & virtual UCA (biconical); +Mixer, 4400~4500 MHz
Rx 1/2 omnidirectional VSA
Outdoors [60] Tx: 1 biconical; VSG 3.33 GHz/ PADS, PDP, PAS, DS, AS
Rx: 64 virtual ULA(biconical) VSA 100 MHz
73
antenna elements. The ULA could provide better azimuth resolution than the UCA,
however the latter could provide the resolution in both azimuth and zenith dimen-
sion. The analysis results showed that both arrays could approach the capacity that a
independently idendifical channel (i.i.d) provided even for closely located users in
the LOS case. When the number of BS antennas is 10 times of the number of users
(the number of antennas), it could be considered as a massive MIMO array. The
performance gain would be diminishing if the number of BS antennas continued to
increase.
In [56], the UCA in [52] was used as the receiving antenna at the B. Eight users
walked randomly in a circle with radius 5 m at about 0.5 m/s speed and the user
antennas were connected to the transmitter of a RUSK channel sounder, which
constitute a synchronous massive MIMO measurement system to investigate the
spatial separability of adjacent users. As reported in [54, 55], even users located
close to each other in the LOS case can be spatially separated in a massive MIMO
system.
A massive MIMO measurement was also carried out by Alcatel-Lucent Bell Lab
in Germany [57]. Seven antennas perpendicular spaced with half wavelength was
rotated around a circle with radius 1 m and 3.5 degree steps pointing to 16 angular
directions to form a virtual antenna array with 112 elements. A LTE-like system
with a subcarrier interval of 15 kHz but only 400 subcarriers (every 3rd sub-carrier
of total 1200 subcarriers) being valid was used to sound the channel and analyzed
the channel characteristics. It mainly studied the orthogonality of the channel at
different measurement positions with correlation coefficients and inverse condition
number. The analysis results confirmed the theoretical advantages of Massive
MIMO over convention MIMO. Using more antennas could improve the orthogo-
nality of the channels between users, but the performance would not improve
significantly when antenna number exceeds a certain amount.
As a part the METIS project, a massive MIMO measurement was completed by
the Aalborg University in Denmark using self-developed sounding system with
5.8 GHz carrier frequency, 100 MHz bandwidth [58]. With 16 transmitting anten-
nas and 64 receiving antennas, a 6416 channel matrix was obtained. The 64 iden-
tical receiving antenna elements (seperated into 8 groups, and 8 antennas per group)
formed three array shapes with different aperture, i.e. ultra large linear array (6 m
long), large linear array (2 m long) and square array 8 users (two antennas per user).
Six 6 schemes with different user locations were designed to achieve various
measurement conditions, such as LOS or NLOS cases, the users being sparsely or
densely distributed, and the user antennas being perpendicular or parallel to the
array. The user/antenna correlation matrix, the singular value and condition number
of the channel, and the power variation over the antenna array was studied. The
measurement results showed that larger antenna aperture could produce more
spatial degrees of freedom (DOF) which was close to the theoretical performance.
The array with the largest aperture was able to approach the performance of an i.i.d
channel since it enabled the channel discrimination among users and between
antennas of a same device.
The researchers in Beijing JiaoTong University (BJTU) recently has launched

indoor and outdoor measurement work. A commercial vector signal generator
(VSG) solely or combined with mixer was used as the transmitter, a commercial
signal analyzer (VSA) or a data recorder was used as the receiver, and there were
two measurement configurations related to antenna and frequency band. In the 1st
configuration, biconical antenna was used at the transmitter to form a virtual large-
scale linear or a virtual uniformly circular array (UCA) both with 128 elements
[59]. One and two omnidirectional antennas were used at the receiver respectively.
The channel was explored at two frequency bands, i.e. 1427~1518 MHz and
4400~4500 MHz. In the 2nd configuration, a single biconical antenna was used at
the transmitter and a virtual uniformly linear array with 64 points was used at the
receiver. The channel was explored in a frequency band of 3.33 GHz with 100 MHz
bandwidth [60]. The non-stationary phenomenon of massive MIMO, the SSPs such
as PDP, PAS, DS and AS, as well as the propagation differences among different
frequencies were analyzed and compared.
3.3.2.2 Measurement of D2D, V2V and HSR
D2D/V2V and high speed rail (HSR) communications are of importance for the
future 5G wireless communications. IEEE 802.11p [61] has been committed to the
standardization of vehicle safety communications and defines three link types such
as V2V, V2I (I refers to Infrastructure) and I2V. IEEE 802.11p adopts 5.9 GHz
frequency band for the V2V communication, and gives a simple SISO channel
model with a tapped-delay line (TDL) structure. The envolope of channel coeffi-
cient of each tap is assumed to be Ricean distributed or Rayleigh distributed. In the
future, MIMO technology will be widely used in vehicle communication system
because it can significantly improve the system capacity and guarantee reliable data
transmission. Therefore, it is essential to strengthen MIMO channel sounding for
D2D, V2V, and HSR scenarios. Extensive channel measurement activities for
D2D/V2V and HSR scenarios have been carried out [1, 6282]. Part valueable
measurement activities and the related details are listed in Table 3.7. The following
is a brief review about these measurement activities.
In the METIS project, one channel measurement for D2D scenario was
conducted by DOCOMO NTT in Japan using self-developed channel sounder
with at 2.225 GHz carrier frequency [42]. A sleeve dipole antenna and a slotted
cylinder antenna were used at the BS side to transmit vertically and horizontally
polarized waves, respectively. A UCA composed of 48 dual polarized antennas
(96 ports) was used at the receiver. The height of antennas at both ends were set to
1.45 m. The UT moved along five designated courses at about 1 m/s. The received
power, PDP, azimuth angular spread at arrival (ASA) and zenith angular spread at
arrival (ZSA) were analyzed. Measurements were carried out during the day and
night time in order to observe the effects of the pedestrians on the channel.
Measurement results showed that at night the received power and DS are larger
because the pedestrians block long-delay path during the daytime, and the ZSA is
Table 3.7 Summary of D2D/V2V/HSR measurement activities
76
Carrier/
Research Channel bandwidth
institutes Scenario Antenna configuration sounder (units: GHz) Measurement parameters
NTT DOCOMO D2D urban,1 m/s Tx: Sleeve, Rx 96 ports, dual DOCOMO 2.225/0.050 Received power, PDP,ASA,
[1] polarization, 1.45 m high ZSA
University of V2V, same/ opposite Tx 1 omnidirectional, 1.6 m high; Propsound 2.3/0.1 PL, DS, SF, K, correlation
Oulu [1, 62] direction,20 km/h Rx:56 ports cylinder array (2.3 GHz), 5.25/0.2 distance(DS, SF, K)
50 ports cylinder array (5.25 GHz),
2.5 m high
Lund University V2V, 2 scenarios, UCA(4 patch), 2.5 m high RUSK 5.2/0.24 TDP(2D), DDP (2D), path
[6366] same/opposite direction power, antenna correlation
matrix
V2V, 3 scenarios ULA( 4 patch), 1.73 m high RUSK 5.6/0.24, AOA,AOD,DS,Doppler,
0.020 analysis ASA
V2V, 10 scenarios bandwidth LSF, time frequency varying
PDP, DSD, DS, Doppler
spread, K
Aalto V2V, 4 scenarios, two Tx/Rx: SPH_5 Aalto 5.3/0.06 PL and exponent, SC
University vehicles, same direction, sounder parameters, DMC
515 km/h [67] parameters, SS fading gain
V2V, 5 scenarios, Tx: ULA 4,Rx: SPH_5,~2 m high PDP,APDP,ACF,CB,CMD,
<40 km/h [68] SD,SC
NEC [70] V2V, 7 scenarios, Single antenna 1.5 m~3 m high Self- 5.9/0.020 Large- and small- scale
<108 km/h developed power
German Aero- V2V speed 30 km/ h Omnidirectional dipole RUSK 5.2/0.12 DDP
space Center
[69]
University of V2V, 2 scenarios, <1 m/s Single antenna Tx:WARP 5.8/0.015 PL,SF, amplitude
Southern Cali- Tx: 1.5 m/3 m high Rx: VSA distribution, correlation
fornia [71] Rx:1.5 m high of DS and SF
BJTU [7477] HSR,multi scenarios Single antenna Tx:GSM-R 0.93/0.001 PL, SF, amplitude distribu-
<350 km/h Rx: VSA tion, DS, K- factor, LCR,
HSR, cutting Single antenna Propsound 2.35/0.05 AFD
BJTU, UPM Subway, <120 km/h Single antenna Tx custom- 2.4/
[7881] ized source, 0.002,0.92/-
Rx:VSA ,2.4/-,5.705/-
WINNER + HSR Tx: monopole, 5 m high; Rx:UCA RUSK 5.25/- PL, DS, K, ASA
[164] 20/100/240 km/h (16 diskcone), 6 m high
BUPT [83] HSR Tx: 2 vertically polarization directional VSG+VSA 2.6/0.02 Correlation Coefficients
370 km/h 45 /135 layout, 40 m high, Rx 2 3 dBi , between subchannels,PL,SF,
~4 m high DS,K-factor
77
small because the pedestrians become the scatterers in the zenith direction).
Whereas, the ASAs in the daytime and night didnt show significant difference.
The statistical parameters obtained according to the measurement results are listed
in details in Table A-14 [1].
Also in the METIS project, a V2V channel measurement was carried out using
Propsound sounder at 2 frequency bands (100 MHz bandwidth in 2.3 GHz band and
200 MHz bandwidth in 5.25 GHz band) by University of Oulu in downtown of
Oulu, Finland [62]. A single omni-directional antenna with 1.6 m high was used at
the transmitter side, an array antenna with 28 dual-polarized elements (56 ports,
2.3 GHz) and an array antenna with 25 dual polarized elements (50 ports and 5.25
GHz) were fixed at 2.5 m high at the receiver side. Two types of V2V scenarios,
moving in the same direction with low traffic density and moving in the opposite
direction with high traffic density, were measured, in which the maximum relative
speed was 20 km/h. The path loss, DS, Maximum Excess Delay (MED), SF
standard deviation, K-factor, as well as their respective correlation distances were
investigated. The results showed that the correlation distance is less than 11 m.
A series of 4-by-4 MIMO channel measurements were carried out by Lund
University in Sweden in many scenarios using a RUSK channel sounder.
In [63], the system configuration for the measurement was 5.2 GHz carrier
frequency and 240 MHz bandwidth. A circular array comprising four microstrip
antennas was placed on the roof of cars, the antenna height was 2.5 m and the
antenna radiation directions were 45 /135 /225 /315 , respectively. A sounding
sequence of 3.2s long was used and the snapshot (tranversing 16 subchannels)
interval was 0.3072 ms which corresponds to maximum Doppler shift of 1.6 kHz
(equivalent to to 338 km/h). The measurement scenarios included one-way rural
expressway and one-way two-lane highway. The movement of vehicles could be in
the same direction or in the opposite direction. Firstly, the Time-Delay Profile
(TDP) and Delay-Doppler Profile (DDP) of the V2V channel were analyzed and the
results showed that the propagation channel was composed of diffuse scattering and
discrete specular scattering. The path loss model was a combination of attenuation
with distance and a slowly varying stochastic processes. Based on the observations,
a GSCM channel model suitable for V2V transmission was propased. The antenna
correlation matrix of the model was verified to be consistent with the measurement
results.
In [64], the system configuration for the measurement was 5.6 GHz carrier
frequency and 240 MHz bandwidth. Four microstrip antennas were placed on the
car roof and along the vehicle axle direction with half wavelength spaced, the
antenna height was 1.73 m and the maximum radiation direction pointed toward
front, back, left and right respectively. Three scenarios were designed as follows.
The 1st scenario was at crossroads, where the transmitter standed statically near an
intersection, while the receiver moved in the perpendicular direction of intersection
with a speed of 30~40 km/h. The 2nd scenario is at expresway with two lanes,
where the transmitting car was blocked in one lane and the receiving car
approached to the transmitter in another lane with a speed of 70 km/h. In the 3rd
scenario, two cars were drived at a speed of 110 km/h in one lane of a two-lane
highway where both cars were obstructed by a tall van. The AOA, AOD, delay, and
Doppler shift of the V2V channel were analyzed. It was found that in 2nd and 3rd
scenarios the channel was mainly composed of one-order reflection paths and has
smaller AS, so that beamforming technique was more suitable than diversity
technique for both scenarios. In the 1st scenario, there were a lot of multiple-
order reflections leading to a larger AS, so that diversity technique could be more
suitable.
In [65], the same sounding system as [64] was adopted to measure in more
scenarios including the following situations: (1) Road crossing (consisting of four
sub-scenarios with respect to suburban or urban, with or without traffic, and single
lane or multiple lanes) where both vehicles approach the crossing from perpendic-
ular directions, driving at 10~50 km/h. (2) General LOS obstruction in highway
where both vehicles drived in the same direction with similar speed of 70~110 km/
h. (3) Ramp merging with partly obstructed main road in rural enviroments where
both cars drived in the same direction at 80~90 km/h. (4) Traffic congestion
including two cases:both vehicles were stuck in a traffic jam and slowly drived at
15~30 km/h; one vehicle was totally stuck in traffic while the other approached at a
relative high-speed of 60~70 km/h. (5) In-tunnel with one way two lans where both
vehicles were driving at 80~110 km/h. (6) On-bridge where both vehicles aparted
from 150 m were drived at about 100 km/h. There were 3~15 measurements carried
out for each scenario. The research showed that V2V channel is non-stationary or
could be considered to be locally stationary. Therefore, the time-varying local
scattering function (LSF) was calculated, based on which the time-varying power
delay profile (PDP) and Doppler power spectral density (DSD) were obtained. The
time varying mean square DS and Doppler spread were further obtained and these
two statistical characteristics could be modeled as a bimodel Gaussian mixture
distribution. Bernado et al. further analysed the measured data to conclulde that the
power of the LOS path is Ricean distributed with with a time-varying K-factor,
while the power of other paths follow Rayleigh distribution [66].
As early as in 2007, some V2V MIMO channel soundings for four scenarios
(campus, six-lane highway, urban, suburban) were carried out by Aalto University
in Finland using self-developed channel sounder with 5.2 GHz carrier frequency
and 60 MHz bandwidth [67]. One snapshot took 8.4 ms and the snapshot repetition
rate was 14.3 Hz allowing a maximum Doppler shift of 7.15 kHz. Two vehicles
were drived in the same direction at 5~15 km/h. Either of the transmitter and
receiver was equipped with one hemispheric antenna array comprised of 15 dual-
polarized patch antennas (30 ports) to obtain 30-by-30 MIMO channel response
data in one measurement. A simplified 2-D geometry of the scattering environment
given the parameter of the scatters including densities, birth/death and locations
was used to build a GSCM for V2V channel, in which the time-varing fading gain
of the specular components (SC) was estimated and modelled as combination of
path loss, large-scale fading being by a Gaussian random process and small-scale
fading being Weibull distributed. Meanwhile, the DMCs were estimated and
modeled as a stochastic process with large-scale power decaying exponentially
(in dB) in the delay domain and small-scale fading being being Weibull distributed.
In another measurement, the same sounding system is used to study the quasi-
stationary region of V2V channel [68]. The same receiving antenna arrys as above
was used, while a four-element uniformly linear array was used at the transmitter
side. One snapshot took 1.632 ms and the snapshot repetition rate was 66.7 Hz,
which limit the maximum speed could not exceed 40 km/h for the five scenarios.
The distance between two cars was between 10 m and 500 m according to the traffic
conditions. The PDP, averaged PDP (APDPs), autocorrelation of small scale
fading, coherence bandwidth (CB), correlation matrix distance (CMD), spectral
divergence (SD) and shadow-fading correlation (SC) were analyzed for different
scenarios and in the LOS condition. It was found that the quasi- stationary interval
is around 3~80 m, which is dramatically affected by the existence of LOS compo-
nent, vehicle speed, and antenna array size and configuration, etc.
One V2V channel measurement was carried out by the German Aerospace
Center using a RUSK channel sounder configured with single antenna, 5.2 GHz
carrier frequency and 120 MHz bandwidth [69]. This measurement mainly aimed to
verify the delay-Doppler PDF proposed for a GSCM. Some measurements in the
LOS and NLOS(due to vehicles or buildings/foliage) case for 7 scenarios were
carried out by NEC laboratory in German using the self-developed IEEE 802.11p
devices with single antenna, 5.9 GHz carrier frequency and 20 MHz bandwidth
(larger than the standardized 10 MHz of IEEE 802.11p) [70]. The measurement
mainly cared about the large-scale and small-scale signal variations. Another V2V
channel sounding was carried out by University of Southern California [71]. -
Software-defined radio platform WARP [72] and a commercial spectrum analyzer
were used as the transmitter and receiver respectively, in which carrier frequency
was set to 5.805 GHz quite close to standardized 5.9 GHz proposed by IEEE
802.11p, and bandwidth was 15 MHz. The channel that the link is blocked by
other trucks was considered for two scenarios, open space and high-rise building
areas. The path loss, SF, amplitude distribution (consistent with the Nakagami
distribution) of small-scale fading, and the correlation of DS and SF was measured
and analyzed.
Professor Bo Ai of BJTU divided the HSR scenarios into 12 categories including
viaduct, cutting, tunnel, and station and further 18 subcategories [73]. Extensive
measurements were done with the GSM-R system as the transmitter and a com-
mercial signal analyzer as the receiver, or with Propsound channel sounder [74
77]. As the train reaches up to 350 km/h, the coherence distance can be reduced to
10 cm [74], which presents challenges to the measurements. Some subway channel
measurement activities were also carried out in Madrid, Spain, and in Shanghai and
Beijing, China, and other cities [7881]. The SISO channel was concerned in these
measurements and several large- and small- scale parameters such as path loss, SF,
amplitude distributions, DS, K-factor, level crossing rate (LCR), average fade
duration (AFD), etc, were mainly analyzed. The HSR channel measurement for
the viaduct scenario were conducted at 2.6 GHz along the Harbin-Dalian
HighSpeed-Railway (HD-HSR) in China by the researchers of Beijing University
of Posts and Telecommunications (BUPT) to obtain 2 2 MIMO channel
measurement data, from which the correlation coefficients between the

subchannels, PL, SF, DS and K-factor were analyzed.
As a part of the WINNER II project, some fast-train channel measurements with
SIMO setup on the express railway between Siegburg and Frankfurt in Germany
was carried out the by Medav, Technical University of Karlsruhe (TUK) and
Technical University of Ilmenau (TUI) in 2006 using RUSK channel sounder
[82]. During the measurement, the train could run at different speeds of 20 km/h,
100 km/h and 240 km/h. Monopole antenna fixed on the top of the train carriage
with 5 m high was used at the transmitter side, and the receiving antenna was a
circular array composed of 16 diskcone antennas (UCA16) placed at 6 m high.
Based on the measurement data, the path loss, K-factor, mean square DS, MED,
ASA, and some other parameters were analyzed. Because the WINNER II D2a
model only supports the LOS case, it is not suitable for the most HSR communi-
cations scenarios. Besides, it frequency band supported is 2 GHz, which is not
suitable for the GSM-R or LTE-R system in nowadays HSR communications.
With the deployment of multi-antenna LTE-R system, it is necessary to carry out
more channel measurements with multiple antennas in various scenarios, such as
HSR, ordinary railway and subway, so as to analyze and better utilize the channel
information in the angle domain [83].
3.3.2.3 Measurement of Millimeter Wave Band
The measurements and researches on high frequency (HF) band channel have been
carried out over 20 years. The main concerns have changed from narrowband
channel characteristics to wideband channel characteristics. The concerned small-
scale characteristics have changed from simple multipath characteristics in delay
domain to joint characteristics in both delay domain and angular domain. In recent
years, with the the gradual progress of 5G R&D, some key frequency bands in 6 ~
100 GHz, such as 11 GHz, 15 GHz, 28 GHz, 38 GHz, 60 GHz and E band, receive
more and more attention, in which the path loss, blockage loss, penetration loss, SF,
DS, AS, and other channel characteristics are being extensively studied. Many
communication enterprises, universities, and research institutes, including Profes-
sor Rappaports team, National Institute of Information and Communications
Technology (NICT) in Japan, Intel Laboratory (in Russia and Germany), Fraunho-
fer HHI laboratory in Germany, Aalto University, Tokyo Institute of Technology,
etc., have carried out extensive HF channel measurements in multiple frequency
bands. They also formed several influential international project teams such as
METIS, MiWEBA, mmMAGIC, 5GCM, etc, to do the channel measurements and
modeling in cooperation. Domestically, HUAWEI, Southeast University, Beijing
University of Posts and Telecommunications, Tongji University, Beijing Jiaotong
University, Shandong University and other research organizations and institutes
have begun the relevant researches and participated in the international coopera-
tion. Some measurements and the related details are listed in Table 3.8.
Table 3.8 Summary of high frequency band measurement activities
82
Research Frequency (GHz)/

institute Bandwidth (GHz) Platform type Antenna and array Scenario Analysis items
NTT 0.8,2.2,4.7,8.45,27,36 Customized nar- Sleeve antenna Outdoors: street canyon PL
DOCOMO [1] rowband, sync with (<1 km), TxH:1.5,6,10 m
cesium clock RxH:1.5,2.5 m
Nokia/ 10,18 Commercial PSG Omni-directional antenna Street canyon, 6 subscenarios PL
Aalborg [12] +PSA (LOS 10~120 m, NLOS
20~270 m);
TxH:7 m, RxH: 1.85 m
TIT [109] 11/0.4 Customized, sync 1. linear array:Tx*6/Rx*2 Indoors: same floor PL, XPR, DS,
with cesium clock with 4 dBi dual polarized coherent band-
dipole width, Ricean
2. circular array, Tx*12/ K- factor
Rx*12, with 6 dBi dual polar-
ized patch
NICT(IEEE 62.5/3 VNA Rectangular or conical horn Indoors: living room, office, All large-/small-
802.15.3c) antennas, rotating in azimuth desktop, library, corridor scale polarized
[47] with 5 step parameters, with-
out zenith angle
INNL(IEEE 60/0.8 Customized VSG, 18 dBi horn antenna Indoors: conference room, All large-/small-
802.11ad) 60/0.8 VSA+ commercial 3D rotating with 10 step office with cubcicles, living scale polarized
[9, 4, 48] up/down-converter room; H:0.7/0.9/1.5/2.9 m parameters
IMC Tx:19.8 dBi horn antenna / Outdoors: campus access, All large/ small
(MiWEBA) 34.5 dBi lens antenna, TxH:6.2 m,RxH:1.5 m scale polarized
[4] Rx:12.3 dBi conical horn parameters
antenna
Rappaport: 28,38,60,73.5/0.2~0.8 Customized PSG+ horn antennas (13.3/15/19/ Indoors; PL and exponent,
VT,Austin U., up/down-converter, 24.5/25 dBi), OEW, 39 dBi outdoors: roof to ground, UMa DS, PAS, AOA,
NYU [9099] sliding correlation parabolic, 2D/3D rotating TxH:8/23/35 m @38 GHz, PDP, outage,
receiving, sync with UMi TxH: 7/17 m, RxH:2/ penetration loss
rubidium clock 4.06 m @28/73 GHz; distance

30~200 m
Fraunhofer 60/0.25 [4] Customized 2 dBi omnidirectional antenna Street canyon, TxH:3.5 m, All large-/small-
HHI baseband,commer- RxH:1.5 m scale polarized
cial RF module parameters,
10/0.25, 60/0.25 [1] Dipole antenna (10 GHz) Street canyon, distance PL and exponent,
5 dBi horizontally omnidirec- 28,142 m; TxH:5 m, DS
tional (60 GHz) RxH:1.5 m
Multi-frequency: 10.25/ omnidirectional Street canyon; Rx 3 LOS -
0.5, 28.5/1, 41.5/1.5, routes(<310 m),4 NLOS
82.5/1.5 [11] routes(<110 m); TxH:5 m,
RxH:1.5 m
Ericsson Ltd. Multi-frequency: 2.44/ VNA+ up/ down- Tx: 7 dBi patch @2.44/ O2I( outdoor 60 m to indoor Peneration loss, DS
[1] [11] 0.08, 5.8/0.15,14.8/ converter 5.8 GHz, OEW@14/58 GHz; 2~15 m); TxH/RxH:1.5 m
0.2,58.68/2 Rx:2 dBi omni-directional; all Indoor(<80 m),Street canyon PL, diffraction and
vertically polarized (~130 m); TxH/RxH:1.5 m diffuse scattering
phenomena
Multi-frequency: 5.8/ TxRx: 2 dBi monopole Indoor(LOS 1.5 m, NLOS AS,DS,PDP
0.15, 14.8/0.2,58.68/2 Virtual cube array 25 [3] 14 m); TxH/RxH:1.5 m
Aalto 62/4 (measurement), 0.2 VNA+ up-/down- Tx: VHUPA(7*7) with Indoors: office; All large/ small
University (analysis) [49] converter biconical TxH:~2.3 m, RxH: ~1 m scale parameters
Rx: VVUPA(7*7) with OEW
61~65 [1] Tx:20 dBi horn antenna, Shopping mall, cafeteria, open All large/ small
rotating in azimuth with 3 square; TxH/RxH:~2 m scale polarized
step; Rx: biconical parameters, with-
out zenith angle
Multi-frequency: Tx: vertically polarized Street canyon (19~121 m); Polarized PDP,
14~14.5,27~27.9,59~63 biconical TxH/RxH:2.57 m PADP,Frequency
[11] Rx:19 dBi dual-polarized horn dependence,DS,
antenna,rotating in azimuth ASA,PL
with 5 step;
27.45/0.9 [11] Antenna same as above,except Open square access;
rotating with 7.2 step TxH:1.6 m, RxH:5 m
83
(continued)
Table 3.8 (continued)
84
Research Frequency (GHz)/

institute Bandwidth (GHz) Platform type Antenna and array Scenario Analysis items
Dual-frequency: Antenna same as above,except Airport check-in area;
15/2,28.5/3 [11] rotating with 5 step TxH:1.6 m, RxH:5 m
81~86 [106108] SFG +VNA+ up/ 45 dBi parabolic antenna, Outdoors: Roof to street, street CIR (CPR/XPR)
down-converter, 20 dBi horn antenna canyon, TxH/RxH:5 m
sync with GPS
Orange,FT at Multi-fre- VNA+up-/down- Tx/Rx: 2 [email protected]/10.5 GHz, O2I(12~25 m); TxH:2.5 m, PDP, Blockage
Belfort [11] quency:3.6~3.725, converter 7.5 dBi@17 GHz vertically RxH:1.5 m loss
10.5~10.625,17.3~17.425 polarized(4 orientations)
CEA-LETI 83.5/6 VNA+up-/down- Tx:10 dBi horn , rotating in Office,conference room(<8 PDP,PADP,angu-
[11] converter azimuth with 45 step; m);TxH/RxH:1.3 m lar distribution
Rx:20 dBi horn,rotating in
azimuth with 10 step and 3*2
planar array with 40 cm
spaced
Tongji Uni- 72/2 VSG+ VSA 25 dBi horn antenna, Tx/Rx Indoors Composite and per
versity/ point to each other(LOS); Cluster ASA, ZSA
Huawei [113] 6 orientations in azimuth and relations with
(NLOS) distance, cluster
numbers
Tongji Uni- 28/0.5 Customized broad- Tx: 24.4 dBi horn fixed orien- Large open office; PADP, DS,AS,per
versity/ETRI band,sliding corre- tation, Rx:9.9 dBi horn, rotat- TxH/RxH:1.5 m cluster DS and AS,
[114] lation processing ing in azimuth with 10 step, cluster numer,ray
3 elevation settings:[0, 10 ] number within a
cluster
BUPT [112] 28/1 VNA+ amplifier 25 dBi horn antenna Indoors(Tx-Rx distance PL, PL index, PDP,
<30 m); TxH:1.93 m, DS
RxH:1.75 m
Southeast 45/1 PSG+VNA 23.7 dBi horn antenna, Conference room, office with PL index, PDP,
University OEW, dipole cubicles, living room, PAS
[51, 110] TxH:1.95 m,RxH:1 m
NCEPU 26/1, 39.5/1 AWG + PSA 24.3 dBi horn(@26 GHz), Dinning hall, TxH/RxH: Blockage loss
27 dBi horn(@39.5 GHz) 1.3/1.3 m,Tx-Rx distance
15 m
Shandong 60/2 PSG+VNA Tx virtual hollow cubic array Office with cubicles PDP,DS,ASD,
University (with monople) /3D rotating a TxH/RxH: 1.6 m ZSD,SV small
[111] 20 dBi horn antenna, Rx 10/20 scale parameters
dBi horn antenna

85
In the early stage, some measurement results for urban and suburban environ-
ments at 9.6/28.8/57.6 GHz [84], 55 GHz [85], 60 GHz [86], 62 GHz [87] and other
frequencies were published, in which the narrowband channel characteristics
including path loss, rain attenuation and oxygen absorption were concerned. At
the end of the 1990s, the transmission and reflection coefficients of some typical
materials including walls, floors, ceilings, windows, etc, at 57.5 GHz, 78.5 GHz and
95.9 GHz bands were measured by NICT [88]. The measurement results were
compared with the reflection model of multi-layer materials. The early measure-
ments were carried out at single frequecy band, till recent the measurements at
multiple frequecy band in the same environment were carried out in order to
establish the fequency dependence of various prarmeters.
The main measurements of IEEE 802.15.3c channel model were carried out by
NICT [47], NICTA in Australia, IMST in Germany, France Telecom (FT), IBM,
University of Massachusetts (UMass) [8]. Except that IBM used a broadband
sounder with 600 MHz bandwidth [89], the other institutes used VNA based
sounder to channel measurement. Measurements mainly concentrated on ten indoor
scenarios, including the living room, office, desktop, library, corridor, aircraft
cabin, etc, in different in LOS and NLOS conditions. The Angle information was
obtained by rotating the standard gain horn antenna with 5 step in the azimuth
direction. Finaly, the cluster parameters (the number, arrival rate, power attenuation
index, AS) and intra cluster ray parameters (arrival rate, average time, power
attenuation rate, etc.) were given by analyzing the measurement data. See Sect.
6 in [8] for details.
Prof. Rappaports team used the separate components and modules to build a
broadband sounder based on the concept of sliding correlation [43]. The chip rate of
PN sequence could be changed to 200 MHz, 400 MHz to 750 MHz (the null-to-null
bandwidth is equal to two times of the chip rate). Some measurements at 28 GHz,
38 GHz, 60 GHz and 73.5 GHz band were successively carried out in VT, AUT, and
NYU [9099], from which the path loss, delay, angle, and other parameters were
analyzed [20]. All measurement activities were designed and targeted to a specific
measurement environment, but the similar equipments and procedures were used.
Directional antennas were adopted at the BS (the transmitter of a channel sounder)
and UT (the receiver of a channel sounder) to send and receive signals respectively.
The directional antenna was installed on a 3D rotating tripod. For every setting of
BS antenna orientation or UT antenna height, the UT antenna was steppedly rotated
in the whole solid angle to scan the channel. The related work are outlined in [20]
and Sect. 2.1.2 of [4]. Based on the measurements, an outdoor mmWave channel
model [100, 101] was established by using a time cluster-spatial lobe modeling
method, in which the path loss, the power and delay parameters of time clusters and
rays within a cluster, and the angular parameters represented by the number and
spread of spatial lobes were included. This model did not support the parametric
modeling of elevation angle i.e., it was a 2D model. Later by adopting the ray
tracing technique, a 3D mmWave channel model under the framework of 3GPP
(WINNER) was proposed [102]. Herein in which the average power in dB scale of
the cluster or intra-cluster rays are Gaussian distributed and decay exponentially
with delay as SV model. For a ray within a time cluster, a spatial lobe is randomly
assigned to it to be used to generate the angle values. The angles of spatial lobes or
rays followed Gaussian or Laplacian distribution. Three path loss models, i.e,
directional, beam combining, and omni-directional, with two model structures of
close-in reference distance and floating intercept, were proposed.
The main measurements to produce IEEE 802.11ad channel model [9] were
carried out by Intel Nizhny Novgorod Lab (INNL) in Russia [48], as well as
Technical University of Braunschweig (TUB) [103], TUI [104], Fraunhofer HHI
[105], MEDAV, and IMST in Germany. A customized broadband sounder com-
posed of a VSG, a VSA and two commercial 60 GHz up/down-converter modules
was used by INNL. Each 18 dBi standard gain antennas that could steppedly rotate
in the whole 3D space was used at the transmitter and receiver. A VNA-based
sounder was used by TUB to measure the human blockage loss. A channel sounder
used by Fraunhofer HHI was comprised of arbitrarily waveform generator (AWG),
oscilloscope and up/down-converter. TUI, MEDAV, and IMST cooperated with
each other to build a 1T2R (one transmitting port and two receiving ports) broad-
band (up to 3 GHz) sounding system composed by an ultra broadband receiver, a
frequency synthesizer and two up/down-converter. Two types of link, access and
D2D, were measured in indoor environments, such as living rooms, meeting rooms
and offices with cubicles. The measurement results showed that obvious clustering
exists in the channel, which matched well with the results of ray tracing simulation.
Fraunhofer HHI laboratory and Intel Mobile Communications (IMC) laboratory in
Germany took part in the MiWEBA project [4]. Fraunhofer HHI used self-
developed baseband sounder with 250 MHz bandwidth, commercial 60 GHz up/
down-converter module and approximate omni-directional (2 dBi) antennas to
measure the access channel for street canyon scenario. The transmit antenna was
3.5 m high and the antenna of the mobile receiver was 1.5 m. The power and delay
of several paths were measured and analyzed. IMC used the same equipment as
INNL to measure the access channel on campus. A 19.8 dBi rectangular horn
antenna or a 34.5 dBi lens antenna was used depending on the Tx-Rx distance
less than or more than 35 m and placed at 6.2 m high. The mobile receive antenna
was a 12.3 dBi circular horn antenna placed at 1.5 m high. The measurement results
showed that two strongest paths were LOS and the ground reflection path, while the
other paths were 15~20 dB lower. The measurement results matching well with the
ray tracing calculation prompted the project team to use the Q-D modeling method
to construct the channel model, which is comprised of deterministric D rays and
random R rays. For more details please refer to the MiWEBA model [4]. The
researchers in Fraunhofer HHI measured the propagation at 10 GHz and 60 GHz
simultaneously and and compared. In the LOS case, they found the propagation loss
at 60 GHz was 15.6 dB higher than that at 10 GHz, which conformed to the Friis
free space path loss equation. These two channels showed similar characteristics,
but at 60 GHz there were less resolvable paths due to smaller dynamic range. In the
NLOS case, some paths simultaneously occured at both frequency bands, while
some paths occured at only one frequency band. In general, similar to the LOS case,
there were less resolvable paths at 60 GHz.
Some measurements at 15 GHz, 28 GHz, 60 GHz, and E band (81~86 GHz) were
carried out by Aalto University in Finland and outlined as follows.
1. The measurements at 60 GHz in conference room were carried out by using a
VNA-based sounder. At the transmitter side, a virtual horizontal uniformly
planar array (VHUPA) with 7 7 elements was formed by translating a
vertically polarized biconical antenna. At the receiver side, a vertically polarized
OEW was used to form a virtual vertical uniformly planar array (VVUPA) with
7 7 elements. The high resolution estimation algorithms were used to obtain
channel parameters, based on which a SV-like model were proposed [49].
2. In the METIS project, multiple measurements at 61~65 GHz (4 GHz bandwidth)
in the indoor shopping mall, indoor cafeteria and outdoor square were carried out
by using a VNA-based sounder [1]. A 20 dBi horn antenna was horizontally
rotated with a step of 3 at the transmitter side, while a 5 dBi biconical antenna
was used at the receiver side, so that only single-directional and 2D channel
parameters were obtained. In indoor cafeteria and outdoor square, more channel
parameters were obtained by adopting the point-cloud field prediction technique
(a laser ranging system was used to build a 3D electronic map called as a point
cloud). After calibrating with the measurement results, a 3D channel model for
these two scenarios were proposed.
3. In the mmMAGIC project, they completed many measurements, e.g., in the
street canyon at multiple frequencies, in the open square at 27 GHz, in the airport
check-in area at 15 GHz and 28 GHz.
4. With Sweep Frequency Generator (SFG) as the transmitter, and the VNA as the
receiver, the wireless channel at 81~86 GHz E-band was measured [106108]. A
24 dBi and a 45 dBi directional antennas were used in the measurement. The
scenarios included street canyon, root-to-street and the maximum measurement
range was up to 1100 m. It was found that there still existed the MPCs even in the
LOS condition and by using high-gain directional antennas, and the first MPC
was 20 dB lower than the LOS component.
The researchers in Ericsson Ltd. used a VNA-based sounder to complete a series
of measurements in participating the METIS, mmMAGIC, and 5GCM projects.
(1) The human blockage measurements in indoors at 57.68-59.68 GHz band with
two 10 dBi (60 elevation beam width, 30 horizontal beam width) antennas were
carried out and found that the blockage loss was as high as 1020 dB. (2) The
measurements in indoor medium range and a long corridor of the office environ-
ment were performed at multiple frequencies to find the path loss exponent and the
diffraction components contributed the main power for indoor mmWwave trans-
mission in the NLOS case. (3) Multi-frequency measurements at 2.44, 14.8 &
58.68 GHz were carried out in street microcell enviroments in the LOS and
NLOS case. The measurements found that in the NLOS case the path loss is less
than expected by knife edge diffraction, and the frequency dependency is not as
obvious as the diffraction, which revealed that there were other reflection or diffuse
components dominating the propagation in outdoor NLOS scenario. (4) The pen-
etration loss through the wall at multiple frequencies were measured. (5) The rich
directional scattering of mmWave channel in indoor scenario was measured at

multiple frequencies using an extreme size virtual cubic array (25 25 25)
antenna formed by means of using a 3D positioning robot to place a dipole antenna
with 0.4 wavelengths space.
Another two measurements at the mmWave band were completed in the METIS
project [1]. (1) The path loss measurements in the city blocks were carried out
narrowband (single frequency) sounding system by NTT DOCOMO at 0.8 and 2.2,
4.7, 8.45, 27, and 36 GHz bands. The cesium clocks were used as the reference
source to ensure the frequency consistency between the transmitter and the receiver.
(2) TIT in Japan used a self-developed modular software radio device to carries out
the sounding of 11 GHz channel [109]. Two antenna schemes were used to meaure
the channel in room and hall scenarios. In the first scheme, the antenna elements
was a 4 dBi dual-polarized dipole antenna. The transmitter used 6 elements to form
a uniform linear array (with 2 spaced) and the receiver used 2 elements. The
system was used to measure a 12 4 dual-polarized MIMO channel. In the first
scheme, the antenna elements was a 6 dBi dual-polarized dipole antenna. Trans-
mitter and receiver used 12 elements respectively to form uniform circular array
antennas (with 0.44 spaced), which was used to measure the 24 24 dual-
polarized MIMO channel. The polarized path loss, DS, cross-polarization power
ratio (XPR), coherent bandwidth and Ricean K-factor, etc. were mainly analyzed.
In the 5GCM project, Nokia and Aalborg University in Denmark completed the
path loss measurement at 10 GHz and 18 GHz [12]. In the mmMAGIC project,
there were other measurement activities. (1) Belfort lab of Orange (a French
telecom operator) completed the measurements for O2I scenarios at multiple
frequencies. (2) CEA-LETI in France completed indoor channel measurements at
83.5 GHz. (3) Bristol University in UK completed channel measurements in a
shopping mall at 60 GHz.
Some measurement campaigns carried out in China are briefly outlined as
follows. The channel measurement at 45 GHz band for three indoor scenarios
defined by IEEE 802.11aj were carried out by Southeast University and China
Telecommunication Technology Labs (CTTL) [51, 110]. A microwave signal
generator and VNA were used as the transmitter and the receiver respectively.
Path loss and DS at 45 GHz were mainly analyzed. The researchers in Shandong
University adopted a similar system to carry out the channel measurements at
18, 28, 38 & 60 GHz in the laboratory with cubicles. Two types of virtual antenna
array formed by rotating horn antenna and translating omni-directional antenna in
3D axes, were used at the transmitter, while the horn antennas were used at the
receiver. The main parameters such as delay, angle, and others were obtained from
the measurement data to be used in the channel modeling [111]. The researchers in
BUPT used a VNA-based sounder to carry out the measurement and analysis the
radio propagation at 28 GHz in the indoor environment [112]. The measurement
results showed that parameters such as path loss exponent, AOA, and others had a
strong correlation with the propagation environment. Xuefeng Yin et al. cooperated
with Huawei Ltd. to conduct channel sounding for the indoor enviroments at
72 GHz [113] and cooperated with ETRI of South Korea to complete the 28 GHz
channel measurements in a large open office in the LOS and NLOS cases [114], in
which the PADP, composite and per cluster DS & AS, cluster numer, ray number
within a cluster were analyzed. Xiongwen Zhao et al. of North China Electric
Power University (NCEPU) measured the blockage attenuation at 26 and 39.5 GHz.
The Voglers multiple KED (Knife Edge Diffration) model was used to model the
blockage effects and found the loss at 26 GHz was smaller than that at
39.5 GHz [115].
3.4 Channel Data Processing
In channel sounding, the measurement data, or the received signals, are superposi-
tion of altered versions of the transmit signal passing through multiple paths with
different delays, gains, and directions. The processing to be applied to the mea-
surement data includes two aspects: path parameters extraction and channel statis-
tical analysis. Path parameters extraction is to extract the parameters of each path
from the received signals using a high resolution parameter estimation algorithm.
These path parameters include complex amplitude, delay, arrival angle, departure
angle, Doppler shift and so on. Channel statistical analysis is to further analyze the
obtained path parameters and find the statistical characteristics and descriptions of
the sounded channel.
3.4.1 Path Parameters Extraction
There are mainly two categories of commonly used channel parameters extraction
algorithms, i.e., subspace-based methods and maximum likelihood methods.
Within the first category are MUltiple Signal Classification (MUSIC) algorithm
[116], Estimation of Signal Parameter via Rotational Invariance Techniques
(ESPRIT) algorithm [117118] and its variant Unitary ESPRIT [119121], etc. In
the second category, there are Expectation Maximization (EM) algorithm
[122, 123], Space Alternating Generalized EM (SAGE) algorithm [124129] and
frequency-domain SAGE algorithm [130], Sparse Variational Bayesian SAGE
(SVB-SAGE) algorithm [131], and Richters Maximum Likelihood Framework
For Parameter Estimation (RIMAX) algorithm [132], etc.
ESPRIT algorithm and its variant are been developed to either jointly estimate
delay and azimuth angle, or jointly estimate azimuth and elevation angles
[120]. MEDAV Companys MIMO channel sounder RUSK took ESPRIT as the
core algorithm at early stage, and later it adopts the RIMAX algorithm. EM
algorithm alternates E (expectation) step and M (maximization) step iteratively to
get the optimal solution but with a relatively slow convergence speed. SAGE
algorithm bases on EM algorithm, separates all parameters into several subsets
and updates a parameters subset each time in the iterative process, which reduces
3.4 Channel Data Processing 91
the computational complexity and expedite the convergence speed significantly. As

an extension of the classical SAGE algorithm, SVB-SAGE algorithm obtains the
distributions of the MPC parameters instead of parameter point estimates by
utilizing the idea of variational bayesian inference. Meanwhile the number of
MPC is jointly estimated using several sparsity priors [131]. RIMAX algorithm
also use the ideal of SAGE to alternatively estimate the parameters of specular
component (SC) and dense multipath component (DMC). Gradient based algorithm
like Gauss-Newton or Levenberg-Marquardt is used to estimate iteratively the
required parameters and achieve a convergence rate even faster than SAGE
algorithm [132].
3.4.1.1 EM Algorithm
Basing on the signal model, one can write the probability density function (PDF) of
the measurement data in terms of signal parameters (aka the likelihood function of
parameters). Maximum likelihood (ML) parameter estimation algorithms choose
the parameter values that can maximize the likelihood function. Usually, the
expression of the likelihood function is too complex to directly obtain the param-
eters maximizing it. Thus, the iterative algorithms are often resorted. EM algorithm
can effectively estimate parameters by iteratively carrying out the E step and M step
[122, 123]. The classical EM algorithm can be described as follows. Firstly, in the E
step, the posterior distribution of the complete (unobservable) data is found with
respect to the observable measurement data (incomplete data) and assumed
(or updated in the last steps) parameters and used to calculate the expectation of
the complete-data log- likelihood. Then the parameters maximize the expectation is
searched in the M step.
For channel sounding, EM algorithm splits the optimization problem that jointly
estimate multiple superimposed paths into multiple separate optimization problems
for estimating a single path. Here we illustrate the EM algorithm used in a SIMO
(single transmitter and M receivers) sounding system. The received signal is
superposition of L paths signals, which can be expressed as
X
L
yty1 t; . . . ; yM t T st; l N t
l1
X
L
cl l expfj2vl tgut l N t 3:6
l1
where u(t) is time-domain sounding signal with transmit power Pu and period Ta.
I snapshots are measured with time intervals Tf. l [l, l, vl, l] denotes the vector
containing the parameters of the l-th path, including path delay l, azimuth AOA l,
Doppler shift vl and amplitude l. c() [c1(), . . ., cM()]T denotes the steering
vector of antenna array reflecting two aspects of information: antenna radiattion
pattern and geometric structure of the array. n(t) [n1(t), . . ., nM(t)]T denotes
M dimensional complex white Gaussian noise. p
Define complete (unobservable) data sets xl tst; l N 0 =2nl t and
X
L
incomplete (observable) data sets yt xl t. First of all, assuming that xl(t)
l1
is known, the log likelihood function of complete data xl(t) in terms of the
parameter vector l can be expressed as follows,

1 T 0 0
0 0 2 0
xl ; l 2 R s t ; l xl t dt ks t ; l k dt 3:7
N0 D0 D0
The parameters that maximize the ML function are simply written as,

bl
x 2 arg max fxl ; l g 3:8
ML l l
E Step: In fact xl(t) is unknown, thus, we need to estimate xl(t) firstly based on
the incomplete data y(t). One direct way is to use the conditional expectation of xl(t)
with repects to Y(t) and the predefined or old parameters. Specifically, we can use
successive interference cancelation method to obtain xl(t) assuming the parameters
about other paths are known,
0 X
L 0
b yt
x^l t; b0
s t; 3:9
l
l0 1
l0 6 l
M Step: The l-th path parameters l can then be re-estimated by computing its
MLE based on the estimation of the l-th signal
0
b 00
bl b
x^l t; 3:10
l ML
In particular, the ML estimator can be expressed as

l , l arg max z ; ; ; x^l
l ,d
, ,
and
I
X 0
z; ; ; xl u t0 ej2t cH ^
x l t0 dt0 3:11
i1 Di
Where Di is observation time interval of the i-th snapshots, i 1,2,... I. The

calculation of z can be seen as correlations (match filtering) in both time domain
and spatial domain. The spatial correlation is just the conventional Bartlett
beamforming. The M step in EM algorithm updates all the parameters at the
same time, which requires a large number of calculations. It is validated by

mathematical proof that the convergence speed of EM algorithm is inversely
proportional to the amount of Fisher information in complete data space, so that a
faster asymptotic convergence can be achived with less Fisher information.
Because of updating all the parameters together in each iteration step, EM algo-
rithm gets a very slow convergence speed.
3.4.1.2 SAGE Algorithm
SAGE algorithm is developed based on the classical EM algorithm. However, in M

step, it divides all parameters of one path into several sets and updates only one set
of parameters at a time while keeping other sets of parameters fixed. The process
continues till all the parameters of one path are updated. Then it run E step and M
step to re-estimate the parameters of next path. The procedure of updating all
parameters of all components in this way constitutes a single update cycle of the
algorithm. The update cycles are repeated anew until convergence or reach a given
maximum cycles. In each step, it updates a parameter subset and hence has less
Fisher information, which not only reduces computation, but also has a faster
convergence speed.
B. H. Fleury, professor of University of Aalborg in Denmark, firstly applied the
SAGE algorithm to the problem of channel parameter extraction [126]. Here we
still consider a SIMO sounder, the M step of SAGE algorithm can be written as
follows:
n 0 o

00l arg max z ;
b b0; b

l l
0
; x^l
b
t;

n 0 o
b 00 arg max z b
00
; ;
b 0
; ^
x l
b
t;
l l l

n 0 o
00 b 00
00l arg max z b
b l ; l ; ; x^l t; b

1 0
b 00l
b 00 z b 00 b 00 00
l ; l ;
b l ; ^
x l
b
t; 3:12
I c l T a Pu
As for the initialization of SAGE algorithm, the initial parameters of each path
can be set all zeros or obtained using the MUSIC algorithm. In the M step, since no
a priori knowledge about the phase of the complex amplitudes, the maximization
procedures for and are replaced by two non-coherent estimation programs.
( )
X
M X I 0 0 0 2
00 0 b
l arg max
b u t ^x l, m t ; dt
Di
m1 i1
( )
I
X 0 H 0 0 0 2
b 00
arg max
u t b 00
l c ^ b dt
xl t ; 3:13
l

i1 Di
Whereas the calculation of Doppler shift still uses the same procedure as the
above iterative process. Correspondingly, since only the l1 paths are estimated, in
the initialization process the E step of estimating l-th path parameters is replaced by
0 Xl1
x^l t; b
yt 0 s t; b 00
3:14
l 1 l
Issues related to the measurement, including the number of sounding channels

(such as SIMO and MIMO), sounding modes (such as parallel sounding and serial
sounding), and sounding sequences (such as PN sequence, frequency sweep in
frequency domain or OFDM signal), the polarization measurement requirements
(such as one or all of the polarization pairs , , , ) will result in greatly
different signal models, which requires to develop corresponding SAGE algorithm.
Based on the traditional SIMO SAGE algorithm, B. H. Fleury, Xuefeng Yin et al.,
pioneered a series of work: extended the SAGE algorithm to estimate 3D angle
(angle of azimuth and zenith) parameters at the receiver [127]; extended to estimate
3D angles at both receiver and transmitter, and proposed Initialization and Search
Improved SAGE (ISI-SAGE) algorithm, whose idea is to reduce the search range
adaptively so as to speed up the calculation [128]; estimated the four polarized
complex gain parameters. Since one complex contains two real values, the real part
and the imaginary part, the algorithm can estimate upto 14 real parameters of each
path [129]. Besides, researchers have considered the parameters estimation problem
for near field communications (supporting spherical wavefront), where the dis-
tances between the transmitter/receiver arrays and the scatterers (the distance from
scatterers to the array center) are added to the signal model, and the corresponding
SAGE algorithm was derived [135, 136].
CRLB Performance Bounds
For the single path signal, Fisher information matrix can be derived from its
likelihood function, Equ (3.5), and then the parameters performance can be got.
SIMO sounding is still taken as an example, i.e., one transmit antenna and uni-
formly linear array comprised of M receive antennas. It is assumed that the
sounding sequence u(t) containing K chips with the chip duration of Tp and the
sequence length of TaKTp. One channel sounding is carried out in each snapshot
and there is a total of I snapshots with snap spacing of Tf. Then Cramer Rao lower
bounds (CRLBs) of the path parameters, AOA , Doppler shift , delay and path
gain , are given by
1 3
CRLB
O 2d=2 sin 2 M2 1
1 3
CRLB
O 2 2 T 2f I 2 1
1 1
CRLB
O 8 2 B2u

1
CRLB 3:15
jj O
p
where Bu 1=T p N s =21 1=K is the Gabor bandwidth of the transmitted
signal u(t). O is the signal-to-noise (SNR) at the output of the correlator and
O MIKNs I. I is the SNR at the input of each antenna port and I Pu||2/(N0/
Ts). d is the space between antenna elements. When d is equal to half a wavelength,
CRLB() coincides with the expression (27) in [126]. It can be seen from the above
formula that each parameter is inversely proportional to O, which means the
estimation error can be reduced by increasing antenna number M, snapshot number
I and sequence length K. Meanwhile, increasing M can further reduce the errors in
AOA estimation. is the angle between the arrival beam direction and the array
axis. The sin2() in the AOA performance bounds shows that estimation error is
minimum when beam impinging along the boresight of the array. CRLB() grad-
ually reaches to infinity when the beam direction gradually changes from boresight
to broadside of the array. Since is always between 0 and , the estimation error
should be bounded. Therefore when is close to 0 or , the estimation ambiguity
exists and CRLB() in the above expression is no longer applicable. Increasing
snapshot number I and snapshot interval Tf will reduce the Doppler shift estimation
error. However, Tf is limited by the Doppler shift according to the Nyquist sampling
theorem, i.e., Tf < 1/(2v).
Separability of Multipath Signals
The resolution of several commonly used estimation techniques for delay, angle
and Doppler shift, such as correlation, beamforming and Fourier methods, is limited
by the intrinsic parameters of the measurement system. These resolutions are
defined to be the half-widths of main lobe of the maganitude of their corresponding
correlation function. For uniform linear arrays (M antenna units), these half-widths
are respectively

360 1
c T p , c , c 3:16
M IT f
Two paths are called separable or resolvable only when the difference of their
parameters satisfy the following conditions
> c 3:17
Shorter chip length (corresponding to higher bandwidth) has higher resolution

capabilities for delay. Increasing the antenna unit number M can provide high
spatial angle resolution. Increasing the measurement snapshot number I and snap-
shot interval Tf can improve the resolution of Doppler shift. The facts is more
optimistic than the claims above. According to the simulation results of the SAGE
algorithm [126], if the parameter difference between two paths satisfies c/5,
or > c/2, or > c/2, then the root mean squared error (RMSE) of parameter
estimation of each path are close to CRLBs of single path, which means the two
paths are resolvable.
3.4.1.3 FD-SAGE Algorithm
The SAGE algorithm mentioned is applicable for channel sounding using time
domain PN excitation signal. In some channel sounding, frequency excitation
signal is commonly used, such as broadband multi-carrier excitation or
VNA-based single frequency stepping excitation. Correspondingly the frequency
domain SAGE (FD-SAGE) algorithm is developed for parameter extraction
[130]. Here we still use a SIMO (single transmitter and multiple receiver) system
as an example, and assume that the entire measurement environment is stationary
(i.e., no need to estimate the Doppler shift). According to the configuration, the
received signal in frequency domain is given by
X
L
Y f cl l ej2f l N f : 3:18
l1
Assuming a total of K+1 (K is an even number) frequency points is soundered,

the signal model at every frequency point can be expressed as
X
L
K K
Y kf cl l ej2kf l N kf , k . . . 3:19
l1
2 2
Using the derivation similar to the time-domain EM algorithm, the complete

(aka admissible hidden) data Xl(kf), i.e., the frequency domain response of the l-th
path signal, can be estimated from the incomplete data Y( f ) by
X
L 0
X^l kf ; b
0 Y f X kf ; b
l0 3:20
0
l 1
0
l 6 l

X^l kf ; b
0 is stacked into a (K+1)-by-M matrix, in which each column repre-
sents the K+1 frequency responses of each antenna element, and each row
represents the frequency response of M antenna elements on a certain frequency

point. The maximum likelihood estimation of parameters is given by

l ,d
l , l arg max fjz; ; ; xl jg
, ,
X X
z ; ; X^l c m ej2kf X^l k; m 3:21
m k
The EM algorithm updates all parameters simultaneously, resulting in slow

convergence speed. Whereas, the SAGE algorithm updates only a subset of param-
eters each time in M step,
n o
00 b 0 ; X^
b l arg max z ; l l

n o
b 00 arg max z b 00 ; ; X^
l l l

1 00 00
l
b
00
b ; X^
z b l ; 3:22
K 1M l l
Assuming the full zero initialization method is used, in the M step, the path delay
is obtained by a non-coherent estimation procedure given by
( )
X X

j2kf ^
b arg max e Xl k; m: 3:23

m
k

The calculation of AOA is the same as the iterative process. The item in braces
in (3.23) can be regarded as summing of the power delay profile of every antenna
element over all antenna elements. Using Fourier transform directly will lead to a
large number of sidelobe components in delay domain. The superposition of multi-
path signals may produce multiple false peaks, which may lead to improper delay.
Usually frequency-domain windowing, such as Kaiser, Hanning or Gaussian win-
dow, can be used to reduce the sidelobe in time domain [130]. That is, X^l k; m is
replaced by X^ l k; m X^l k; mW k, where {W(k), k 1, . . .} is the window
0
function with normalized power.

Accordingly, in the initialization phase, since only the previously estimated path
parameters are available, the E step is changed to
X
l1
0
X^l kf ; b
0 Y t X kf ; b
l0 3:24
0
l 1
As the time-domain version of SAGE algorithm, increasing the number of

antenna M and frequency points (bandwidth) improves the estimation accuracy of
all parameters. Here only a SIMO sounding system is considered, the signal model
and estimation algorithm need to be extended according to the actual configuations
and requirements in specific sounding applications.
3.4.1.4 DMC and Its Estimation
The signal emitting from transmit antenna and arriving the receiving antenna is
composed of line-of-sight (LoS) component, specular components (SCs), and dense
multipath components (DMCs). Each SC corresponds to a discrete and strong
propagation path formed by an independent scatterer reflecting electromagnetic
waves. The propagation paths can be described by simple geometric relationships.
However, the contents of DMCs are far more complex and illustrated in Fig. 3.8.
It is usually known that the DMCs consists of Diffuse components (DC) reflected
off rough surface having wavelength comparable to radio wave, diffraction, reflec-
tion from different layers of scatterers, echoes among scatters, and so on.
The contribution of DMC in the total reveived power varies with the distance
between the transmitter and receiver as well as the environment. Intuitively, the
proportion of DMC is larger in indoor scenarios or in the NLOS case. Recent
studies have indicated that DMCs contribute 20% to 80% to the total receiving
power in indoor environment [137] and Industrial Environment [138]. Usually
DMC has larger angle distribution region than SC. Moreover, DMCs have longer
delay spread than the SCs and the power of DMC cluster approximately decays
exponentially with delay [139]. Ignorning DMCs in the channel modeling will
underestimate the channel capacity. Therefore, DMC has enormous significance for
channel modeling.
The earliest DMC modeling assumed that the DMCs have a white spectrum in
azimuth angle while they have an exponentially decayed Power Delay Profile
(PDP), which can be expressed as [133]
8
h i < 0, < d
E jxj2 d =2, d 3:25
:
d eBd d > d
where Bd is the coherence bandwidth, d is maximum power, and d is base time

delay. The Correlated power spectral density of DMCs is the Fourier transform
( f ) of PDP which is given by
Fig. 3.8 Ilustration of SC

and DMC
d
f ej2f d 3:26
d j2f
where d Bd/(Mf0) is the normalized correlation bandwidth, f0 is the sampling

interval in frequency domain. Equ (3.26) represents the power spectrum in contin-
uous infinite bandwidth. In practice, it is more convenient to use the discrete power
spectrum of finite bandwidth as follows,
" 0 #
d 1 ej2M1d
3:27
M d d j2 M 1=M
0
where d d f 0. The parameters describing DMCs are denoted by
0
d d d d . The covariance matrix of DMCs sampled in frequency
domain is a Toeplitz matrix composed of , i.e. Rd(d) toep(, H). In [133],
Gauss-Newton algorithm is used to iteratively update the DMCs parameters,
while the initial estimation is found by so called global search strategies beginning
with estimated PDP. A simpler least-square based method can be used. First obtain
the residual CIR containing DMC by substracting SC component Hsc() (obtained
by high resolution estimation algorithm such as SAGE) from the total CIR HRx().
Then, averaging the residual CIR over multiple antennas to get the residual PDP
1
res HRx Hsc 2F 3:28
Nt Nr
where k k2F denotes the square of Frobenius norm of a matrix. The optimal DMC
parameters are those that can make parameterized PDP approach to the residual
PDP,
X
M 1
b
d arg min j res m d m; d j2 3:29
d
m0
where d(d) is the PDP of DMCs expressed in discrete time domain, which can be
obtained from Rd(d),

d d diag 1 Rd d , 3:30
where denotes Fourier transform matrix, and diag[] denotes the diagonal ele-
ments of the matrix.
More measurements show that DMC is not white (colored) spectrum in azimuth
angle. In [134], the estimation and analysis of DMC directional features were
presented. Herein we explain its basic ideas as follows. In delay domain, the
power of DMCs is still modeled as exponential decay function. In angular domain,
DMCs are modeled as the sum of multiple DMC clusters and each DMC cluster
corresponds to a SC cluster. The azimuth angle and elevation angle are described by
Fisher-Bingham distribution (or Kent distribution). Some literatures also models

DMCs being von Mises-Fisher distributed [140]. Unlike (3.28), traditional
beamforming algorithm is used to obtain the residual channel power angular
spectrum,
PN t N r
i1 hi Bi ; W i
h; ; q
PN t N r : 3:31
2
i1 j B i ; W i j
Where Bi(, ) is the 3D radiation pattern of the i-th antenna element and Wi is the
window function used to reduce sidelobe. Firstly, DMCs are clustered in angle
domain according to the azimuth anglular spread and elevation anglular spread of
ph i
SC. ; DMC, k 2 ; SC, k 2 2 ;SC, k ; ;SC, k , where (, )SC, k and
;SC, k are the average angles and angular spreads of the k-th SC cluster. Each
DMC cluster has an independent measure of PDP, which is calculated as
maxDMC, k maxDMC, k

res, k jh; ; j2 sindd 3:32
minDMC, k minDMC, k
Then according to (3.29), the parameters of each DMC cluster are calculated.
3.4.1.5 Mobile Channel Estimation and Tracking
In the measurement of a dynamic channel, either the receiver or the transmitter, or

both, are changing over time, so are the channel parameters. To model the dynamic
channel accurately, it is required to estimate and track the number of paths and
multipath parameters. There are several commonly used path tracking algorithms,
e.g. state space method [140, 141] and particle filter method [142]. The state space
method is used to establish state equation and observation equation for the tracked
parameters. Due to the are the nonlinearity of the signal model, Kalman Filter
Extended (EKF) is used to carry out the linearization processing based on the
Taylor series of the observation equation. EKF is only suitable for the situations
where linearization approximates the nonlinear observation equation. When the
channel coefficients are subject to high fluctuations in mobile enviroments, the
linear approximation will be very inaccurate, so that the changes in the channel can
not be catched up in time. There are two approaches to combat the nonlinear
problems. One is Kalman Enhanced Super resolution Tracking algorithm (KEST)
proposed by Thomas Jost et al. , which combines Kalman filter and high resolution
estimation algorithm (such as SAGE) to carry out channel parameters tracking
[141]. Xuefeng Yin et al. apply the Particle Filter (PF, aka Sequential Monte Carlo)
method to track dynamic channel parameters [142]. PF can avoid calculating the
second-order derivatives of the received signal with respect to the path parameters
when be applied to nonlinear observation model. Some researchers use PF to track

the path parameters indirectly. For example, K. Saito used PF to track four
parameters including two-dimensional positions of discrete scatterers, delay from
transmitter to scatterer, and AOD. Therein SAGE algorithm was used to obtain the
path parameters used to update the PF state [143].
3.4.1.6 Automatic Clustering Algorithm
It has been found that the channel MPCs tend to appear in clusters, i.e. in groups of
multi-path components (MPCs) with similar parameters. Meanwhile, describing
and subsequently processing based on clusters will significantly reduce the number
of parameters appearing in the channel modeling. This is also one reason why
clusters are used as the basis for the GSCM and SV modeling methods. In this part,
we will describe the automatic clustering algorithm. It involves two problems, one
is how to cluster all of the MPCs when the number of clusters is given, and the other
is how to determine the optimal number of clusters and the optimal criteria.
Clustering algorithms, such as K Means, standard or fuzzy CMeans and other
algorithms [144, 145], can be used to deal with the first problem. For the second
problem, the commonly used methods at present are to use a variety of indexes to
determine the number of clusters, such as the Kim-Park index [146], Davis-Bouldin
index, Dunn index, and Calinski-Harabasz index [147]. The following gives a
simple description on automatic clustering.
1. Clustering. Assuming there is L MPCs in total, each of which has four param-
eters: delay , AOA , AOD and power P. The parameter set of each path is
denoted by Xl [l, l, l, Pl] and the parameter set of all L MPCs are denoted by
{Xl, l1,. . .,L}. The expected number of clusters is K and each MPC belongs
to one of the K clusters. According to K-Means algorithm, clustering is executed
as follows. Firstly, K MPCs are randomly selected in as the center of
K clusters and denoted as ck 2 . Then the Multipath Component Distance
(MCD), MCD(Xl,ck), between each MPC in to the K cluster centers is
calculated. MCD denotes the distance between two paths. For i-th and j-th
MPC, their MCD can be expressed by [148]
q
MCD Xi, Xj kMCDAOA, ij k2 kMCDAOD, ij k2 MCD, ij 2 3:33
There are three MCD components, two for angle and one for delay. The
calculation of angular and delay MCD component are different. Angular MCD
component is related to the distance between 3D direction vector of the two paths,
and the delay MCD component is the normalized delay difference,
0 1 0 1
sin i cos i
sin j cos j
1
MCDAOA=AOD, ij @ sin i sin i A @ sin j sin j A
2
cos i cos j

i j std
MCD, ij 3:34
max max
Where i and i are elevation angle and azimuth angle respectively, std is the
standard deviation of all path delay, max is the maximum excess delay. is a
proportional factor, usually setting to 1, which is used for adjusting the proportion
of delay in the calculation of distance.
Next, the index of cluster with its center being nearest to each path is taken as
cluster identification of the path, so that can be divided into K clusters. Then new
cluster centers are calculated, which is the power weighted average in delay, AOA
and AOD components of all the MPCs belonging to the same cluster, i.e.,
P
l2cki Pl AOA AOD l
ck
i1 P 3:35
k
l2c i Pl
2. Determinating the optimal number of clusters. The idea of using Kim-Park index
to determine the cluster number is to minimize the sum of normalized intra-
cluster distance and inter-cluster distance [146]. The average intra-cluster dis-
tance is calculated as follows
P
1 X l2ck MCDX l, ck
K
u K 3:36
K k1 jck j
Where |ck| is cardinality of ck, namely the number of MPCs in the k-th cluster. The
average intra-cluster distance reflects the compactness of a cluster. When the
number of clusters is excessive, u(K ) becomes very small. On the contrary,
when the cluster number is too small, many MPC that are distinct or less correlated
are divided into the same cluster, so u(K ) becomes large. The measure of inter-
cluster distance is calculated as follows
K
o K 3:37
min MCD ci ; cj
i6j
When the number of clusters is too small, the minimum inter-cluster distance
mini 6 jMCD(ci, cj) becomes large, so o(K ) becomes very small. On the contrary,
when the number of clusters is excessive, o(K ) becomes very large. Thus, there
will be a optimal point in u(K ) + o(K ) when varing the number of clusters. That is
the key of K-P index. Going through all possible cluster numbers, K2,. . .,Kmax,
the optimal number of clusters Kopt is determined by
8 9
< u K min u K o K min o K =
K K
K opt arg max : 3:38
K :max u K min u K max o K min o K ;
K K K K
Herein normalizd inter-cluster distance u(K ) and inter-cluster distance o(K )

are used.
3.4.2 Channel Statistical Analysis
In this subsection, we will introduce the main channel statistical characteristics that
need to be acquired and the commonly used statistical analysis methods. The main
contents include the basic distribution fitting methods, the basic correlation analy-
sis, and those items calculated from delay, angular and cluster domains. For more
details please refer to WINNER II document [82].
3.4.2.1 Basic Distribution Fitting Methods
The channel parameters, such as amplitude, delay and angle, are usually Random
Variables (RVs) following a certain distribution function expressed in PDF or
cumulative distribution function (CDF). According to the measured data, some
criteria are used to determine the suitable distribution function and statistical
parameters so as to describe the parameters accurately in a sense of probability.
These commonly used criteria include maximum likelihood (ML), least square
(LS), minimum mean squared error (MMSE) or uniformly minimum variance
unbiased estimation (UMVUE). For a channel parameter, RV X, it is assumed
that the PDF of X is known but with unknown parameters, which is denoted by
f(x;). Parameters estimation is to use the known observed points for X to estimate
the unknown parameter . Taking the normal distribution as an example, assuming
there are n observed points xi (1 i n) for a random variable X, the mean and
variance of X is given by
1X n
1X n 2
EX
b xi , b2 Var X xi b
3:39
n i1 n i1
In statistical analysis of channel parameters, the commonly used PDF that are
related with amplitude, angle, time and arrival are listed in Table 3.9, 3.10 and 3.11.
The estimation methods for the parameters and generating methods of RVs are also
given in the three tables.
Table 3.9 Commonly used distribution function for amplitude
104
Distribution
type PDF Parameters estimation Random Variable generation
!
Gaussian 2 b
EX BoxMuller method [149]:
1 x p
or Normal f x; ; p exp d X 2 ln U cos 2V , or
2 2 2 ML: 2 VarX p
X 2 ln U sin 2V
Nd
UMVU: 2 Var X U and V are uniform random variables:
N1 U,V~ Uniform[0,1)
2 p
Rayleigh x x2 d
E X X 2 ln U
f x; exp 2
2 2 2 2
x p
Ricean f x; ; K 2 MoM method [150]:P X2 R X2 Y 2
1 X,Y are Gaussian random variables:
x2 p K^
exp K 2 I 0 2Kx EP X ~ N(2K 2 cos , 2)
2 q 1
EP 2 Var P Y ~ N(2K 2 sin , 2)
is any real number
d
2 EP =2= K^ 1
p
Nakagami 2 m m MoM method X =2mY
f x; m; [151]:
E2 X 2 Y is a Chi distribution random variable Y ~ (2m) with the DOF
m m m^ ,
x2m1 exp x2 Var X2 of 2m;v
b E X
2 u 2m
uX
Yt i X2 ,X ~ N(0, 1)
i
i1
!
Lognormal 1 ln x 2
b Eln X X exp(XG)
f x; ; p exp b2 Varln X XG is Gaussian random variable:
2 x 2 2
XG ~ N(, 2)

Weibull k xk1 x k b
k E Xk X [ ln(U )]1/k
f x; k; exp U is uniform random variable:
E Xk ln X
k^1 k ElnX U~ Uniform[0,1)
E X
Note: I0(x): Zero order modified Bessel function of the first kind.
Table 3.10 Commonly used distribution function for angle, X2[, )
Distribution type PDF Parameters estimation Random variable generation
Uniform 1 None X 2U
f x
2 U is uniform random variable:
U~ Uniform[0,1)
Wrapped Normal 1 Z eiX ,
b argEZ , X mod(XG + , 2)
f ; ; p :
c2
N1 XG is Gaussian random vari-
" 2 # ln
X 1 able:
2k2 N jEZ j2 1
3.4 Channel Data Processing
exp XG ~ N(, 2)
k1
2 2
p
Laplacian 1 2jx j MLE [152]: X p2 sgnU :
f x; ; p exp b E X
2 p ln(1 |U|)
b 2EjX b j U~ Uniform[1, 1)
von Mises 1 Reference [153] Reference [154]
f(x; , ) expcosx
2I 0 Z eiX , b argEZ ,
s
I1 b N jEZ j2 1

I0 b N1
von Mises-Fisher (p2 is von p=21 Reference [155]: Reference [156, 157]
Mises) fp(x; , , p) p=2
exp T x
bE
2 I p=21 X
=EX ,
I p=2 b

: average direction, 1 E X
: focus parameters, 0 Ip=21 b
FisherBingham(Kent) 1 Reference [158] Reference [159]
f x; ; ; 1 ; 2 ; 3 :
n h c ;
2 2 io
exp 1T x 2T x 3T x
c(, ) 2.
X1
j1 2
2j12
2j I 2j12
j0
j 1 2
x is 3d unit vector, denoted by a point on the unit sphere
105
Note: I0(x): v order modified Bessel function of the first kind

Table 3.11 Distribution function related with time and arrival

Distribution Parameters
type PDF estimation Random variable generation
Exponential f(x; ) exp 1 X lnU
ML:b
(x) EX U is uniform random
N 1 variable:
UMVU:b
NEX U~Uniform(0,1]
Poisson f(k; ) Pr(X k) b
EX Reference [160161]
k e

k!
Among many candicates of distribution function, Good of Fitness (GoF) tool is

used to validate the distributions and find the best matching one. The commonly
used GoF tools include Kolmogorov-Smirnov (K-S) test and Anderson-Darling
(A-D) test etc. [162].
3.4.2.2 Basic Correlation Analysis Method
Correlation reflects the similarity between two random variables (RVs) in one or
more dimensions. The correlation calculation does not need to know the distribu-
tion function of the RVs. It is assumed that the RVs are ergodic, so the correlation
can be calculated using ensemble average. Assume there are Nx and Ny collected
values for x and y respectively, the correlation value between x and y can be
calculated as

XN2 k N y 1; N x 1
1
Rxy k xn kyn , N 1 max
k; 0 : 3:40
N 2 N 1 nN 1
1 N 2 min N y ; N x k
This estimation is unbiased, but the estimated variance of Rxy(k) increases with k.
A biased estimation can be used instead when the variance is small
N2 N1
RB k Rxy k 3:41
min N y ; N x
The calculation of covariance is similar to that of correlation, which just

removes the respective average value before calculating
1 XN2

Cxy k xn k x yn y
N 2 N 1 nN 1
1
Ny
1 X Nx
1 X
x xn , y y 3:42
N x n1 N y n1 n
Correlation coefficients can be calculated based on covariance
Cxy k
xy k pp 3:43
Cxx 0 Cyy 0
Correlation distances (time) are symmetrical on both sides of maximum covari-

ance, which are proportional to the area of the covariance function. It can be
calculated as
PNx 1
Cxy k
kN y 1
dR : 3:44
2max Cxy k
When spatial correlation is calculated, if the spatial distance between all snap-
shots is not constant, .e.g., the measured routes are not always along a straight line
or the speed is varying, then the above calculation will be problematic. The
correlation value will not show a certain relationship with snapshot index k.
Therefore, we must group the snapshot by distance, denoted by k(), and the
correlation will be
1 XX
R xn kyn , 3:45
N n k
where N() is the number of valid data pairs [x(n + k()), y(n)] in the distance .
3.4.2.3 Delay Domain Analysis
The items to be analyzed in this part include CIR, time-varying PDP, PDP, MED,
DS, total power, path loss, SF standard deviation, and Ricean K-factor. At this
stage, all calculations are based on the CIR and no angular information is needed.
The CIRs can be got by correlation of the receiving samples with the PN sequence
(known for the receiver) in time domain sounding, or can be got directly by Fourier
transformation in frequency domain excitation including broadband OFDM signal
or single carrier in VNA based system. Of course, we can also use the high
resolution parameter estimation algorithm to extract angular the information, and
get CIRs through accumulating in the angle domain. The former is called direct
method, and the latter is called synthesis method.
1. Channel Impulse Response

CIR, denoted by h(t,,s,u,p), is a function of time (snapshot) t, delay , trans-
mitting antenna element s, receiving antenna element u, and polarization p. If the
channel h() for (t,s,u,p) dimension does not satisfy SNR criterion, it will not be
used in subsequent analysis. This is mainly caused by too large path loss, too small
transmit power, or insufficient receiver sensitivity, etc. The criterion is that if the
peak power of h() exceeds the noise threshold by predefined value, such as 20 dB,
it is considered to be a valid CIR. For such valid CIR, the MPCs with path gain
being lower than noise threshold (sometimes 3 dB above the noise threshold) is set
to 0. There are nt snapshots collected in a stationary interval. The number of CIRs
contained in each snapshot depends on the number of antenna pairs, and there
should be at least one valid CIR per snapshot. It should be noted that all of the
snapshot time t, delay , transmit antenna s and receive antenna u, are discrete. The
snapshot time t depends on snapshot cycle and delay depends on the delay
resolution of the sounding system.
2. Time-variant Power Delay Profile
Averaging the power of CIRs in the antenna pair (u,s) dimension to get time-
variant PDP. Because the propagation environment and Tx-Rx distance change
over time during channel sounding, the delay of peak path will undergo minor but
unignorable drift. Therefore, it is necessary to align highest peaks of all CIRs. The
specific operation is shown as follows
1 X Ns XNu
Pt; 0 ; p jht; u, s t; p; s; u; pj2 3:46
N s N u s1 u1
where u, s(t, p) argmax|h(t, , p)|2 indicates the delay adjustment for each CIR.
0 0
u, s(t, p) is denoted by . Hereafter, is used for delay instead of for clarity.
3. Power Delay Profile
The time-variant PDPs of nt snapshots in stationary interval are averaged to get
the PDP.
1 X nt
P; p Pti ; ; p 3:47
nt i1
4. Maximum Excess Delay (MED).

MED is the maximum delay with non-zero power in PDP during a stationary
interval, i.e.,
max p argmax fP; p > 0g 3:48

5. Mean Square Delay Spread

DS partially reflects frequency selectivity of a channel. If the delay PDF is
replaced by the PDP, then DS is the second-order centeral moment of the delay
distribution. The calculation of DS is performed in a stationary interval with nt
snapshots. If the paths in PDP having power lower than a predefined value (such as
20 dB) below the peak path power will be skipped, i.e.,

P i ; p Pi ; p > maxfPi ; pg=100
P i ; p 3:49
0 other
Where the delay i get discrete value because the measurement system has a
certain delay resolution. The delay distribution function is estimated using PDP
as follows,
P i ; p
p i ; p PN x 3:50
i1 P i ; p
Mean delay , second-order moments of delay 2 and mean square DS DS( p) can
be got from delay distribution function by
p
DS p 2 2
X
Nx
p i pi ; p
i1
X
Nx
2 p 2i pi ; p 3:51
i1
Based on the mean square DS obtained in different positions, the DS correlation

between two locations can be calculated, so can the DS correlation distance. Details
can refer to the subsection of Basic Correlation Analysis Method.
6. Total path power
Total path power is the sum of PDP over all the delay i and given by
X
Nx
P p P i ; p 3:52
i1
7. Path loss
To establish the relations between path loss and other propagation parameters
(such as Tx-Rx distance, carrier frequency, etc.) is a very important task in channel
modeling. The following takes analyzing the dependence of Path Loss on the
distance as an example.The Tx-Rx distance d for every snapshot should be recorded
during measuring. For each total path power there is a corresponding distance,
d
Pp ! Pd; p. Considering the other parameters of the measurement system,
P such
as the transmitting power Ptx, transmitting and receiving antenna gain Gi , the
P i
total cable attenuation Ai , we can obtain the path loss
i
X X
PLd; p Ptx p Gi Ai Pd; p 3:53
i i
where all the parameters are in unit of dB. Multiple measurements at different
distances can obtain more observations of PL values with respect to the distances.
With these observations, linear regression analysis can be used to find slope A( p)
and intercept point B( p) of the path loss model,
2 3 2 3
PLd 1 ; p log10 d1 1
4 54 5 A p
1 3:54
B p
PLd n ; p log10 dn 1
Namely, the path loss model in terms of A( p) and B( p) can be given by,
PLd; p Aplog10 d Bp 3:55
In fact, path loss is also affected by other factors such as carrier frequency,
antenna height and environmental,etc. Therefore, large amounts of measurements
in various frequencies and antenna heights should be conducted to find their
relationships with PL. Multiple regression analysis may be used. Path loss modeling
always is the most important work in the channel modeling, and we will not give
much details due to space limitations.
8. Shadow Fading (SF) Standard Deviation
For N measured data in the path loss analysis, the measured SF values can be got
by substracting the expected path loss PLdn ; p from the measured values PL(dn, p),
SFdn ; p PLdn ; p PLdn ; p. Collecting these values, the standard deviation
of SF can be obtained by
v
u
u 1 X N
SF p t jSFdn ; pj2 3:56
N 1 n1
Using correlation analysis method, we can also calculate the correlation coeffi-
cients of SF at two distant locations, thus the correlation distance of SF.
9. Ricean K-factor
Ricean factor K is the power ratio of the LOS component to all other compo-
nents. In WINNER, IMT-A and 3GPP models, K-factor is calculated through the
method of moments (MoM) [150]. But the MoM method is only suitable for
narrowband channel. For broadband channel, MoM method is required to be used in

each of the coherent bandwidth. Actually the K-factor can be calculated directly by
its definition and given by
q
EfPg2 Var fPg
K q 3:57
EfPg EfPg2 Var fPg
X
nt
where E{P} is the average power, EfPg 1=nt Pi. Var{P} is the power
i1
X
nt
variance, Var fPg 1=nt 1 Pi EfPg2 . The value of K can be used
i1
to estimate the conditional probability of LOS propagation. The obtained K can be
further used to calculate variance and correlation distance.
Angular Domain Analysis
With high-resolution parameter estimation algorithm, the parameters of multiple

MPCs can be extracted, such as delay, the complex gain of each polarization
direction and angle, etc. According to the capabilities of measurement system and
measuring requirements, the angles to be investigated include one or more of the
AOA, AOD, Zenith angle of Arrival (ZOA, or EOA) and Zenith angle of Departure
(ZOD, or EOD). In this part, the analysis on power angle spectrum (PAS) and
angular spread (AS) are given.
PAS ( /, p) is mainly used to describe the power distribution in angular
domain. When focusing on one angular domain, we accumulate the channel
power over other dimensions (including delay) to get corresponding PAS.The
angle wrapping will be involved in the calculation of angle statistics, so that a
special processing method is used here. All angle information is added a small angle
offset to be determined. Taking AOD for example, we have AODi()
AODi + . For azimuth angle, the wrapping range is [,), so the wrapping
operation is given by
0
AODi modAODi ; 2 3:58
For the elevation angle, the wrapping range is [/2, /2). Then the anglular
power distribution function (normalized PAS) is estimated from the PAS,
0

0
PAS AODi

pas AODi PNx 0 : 3:59
i1 PAS AODi
The first-order moment (mean) of angle distribution are calculated and removed
from the individual angle to get an unbiased value, again the wrapping process is
carried out.
! !
00 0
X
Nx
0
0

AODi mod AODi AODi pas AODi ; 2
i1
3:60
Then the second-order moment of the angle distribution are calculated,

v
u Nx
uX 2
t
00 0
AODi pas AODi 3:61
i1
ASD is defined as the minimum second-order moment of the angle () that can
be obtained by varing .
ASD min 3:62

With the ASDs obtained at multiple locations, the mean and variance of ASD as
well as the correlation distance can be obtained. The same processing can be carried
out for other angles.
3.4.2.4 Cluster Analysis
After parameter extraction and cluster classification, some features of the clusters
can be tracked in mobile channel, such as the cluster life cycle, cluster generation
rate, cluster drifting in delay and angular domain. Clusters tracking is often ignored
in the existing models, however, it is particularly important for the new 5G channel
model. In mobile enviroments the parameters of the clusters will change slowly.
The whole life cycle of a cluster can be recorded from its appearance to disappear-
ance, so that it is easy to compute the average survival time and birth and death rate
of the cluster. In a stationary interval, the change of delay and angle of the clusters
from one snapshot to another snapshot is also very important, which can be used to
analyze the time evolution of the channel.
3.5 Existing Channel Models 113
3.5 Existing Channel Models
This section will review several existing channel models, and point out their main
characteristics and restrictions. These channel models include those proposed by
various research project or groups, such as the WINNER famliy [13, 163, 164],
COST 259 [165]/ 273 [166]/ 2100 [167]/ IC1004 [168], METIS [1], MiWEBA [4],
QuaDRiGa [19], mmMAGIC [11], 5GCM [12], and proposed by the standardiza-
tion organizations, such as ITU-R IMT-Advanced [14] and IMT-2020 [178] chan-
nel model, 3GPP SCM [169]/3D [170]/D2D [3]/HF [6], IEEE 802.15.3c [8], IEEE
802.11 TGn [171]/ac [172]/ad [9]/ay [10].
3.5.1 3GPP SCM/3D/D2D/HF
Spatial Channel Model (SCM) is a GSCM developed for cellular MIMO system by
the Third Generation Partnership Project (3GPP) in 2003 [169]. The importance of
SCM lies in that it is convenient for expansion and parameterization and provides
an infrastructure for subsequent channel models. The frequency range and system
bandwidth supported is 1~3 GHz and 5 MHz respectively. The SCM channel model
suits for three scenarios including Suburban Macro (SMa), Urban Macro (UMa),
and Urban Micro (UMi). The path loss (PL) model uses modified COST 231 Hata
urban model (SMa, UMA), COST 231 Walfish-Ikegami NLOS model (UMi NLOS)
and COST 231 Walfish-Ikegami block model (UMi LOS) [173]. In terms of LSPs,
it takes the DS, AS and SF as lognormal distributed or normal distributed and gives
their auto-correlation and cross-correlation coefficient, inter-station SF correlation
coefficient and so on. With respect to the SSPs, a channel is composed of six
clusters, each of which contains 20 rays. The model gives the characteristics of each
cluster with angles being Gaussian distributed, delays being exponential distrib-
uted, and power decaying exponentialy with delay.
Based on the parameters presented by WINNER II and WINNER+, 3GPP
proposed two new channel models, 3D-MIMO and D2D. 3GPP 3D MIMO [170]
is suitable for the scenario of UMi and UMa, both of which include the case of
Outdoor-to-Outdoor (O2O) and Outdoor-to-Indoor (O2I). Its applicable frequency
range, bandwidth and maximum mobility is 1~4 GHz, 10 MHz and 3 km/h
respectively. It gives new path loss parameters (see Table 4.1 in [170]), the
correlation coefficient of intra-station LSPs, and the updated small scale
(SS) parameters (Table 7.3.6, 7.3.7, 7.3.8 in [170]), in which the azimuth angle
follows wrapped Gaussian distribution, zenith angle follows Laplacian distribution,
and the zenith angle parameters partially refer to the WINNER+ channel model.
3GPP D2D Model [3] support the scenarios of UMi (both O2O and O2I) and
Indoor-to-Indoor (I2I). The main contribution is the support for dual mobility and
the core of the model is calculating the Doppler shifts. Two system simulation
scenarios are defined: generic scenarios (2 GHz, mobility speed of 3 km/h) and
public security scenario (700 MHz, maximum speed of 60 km/h). The bandwidth is
set as 10 MHz for uplink and downlink respectively in frequency division
duplexing mode, and 20 MHz in time division duplexing mode. Most of the
parameters and the channel generation process refer to the existing standard
models.
The 3GPP-HF channel model (6-100 GHz) is the first standarderized high
frequency band channel model released to the public. Besides those typical scenar-
ios specified in the 5GCM [12], additional scenarios including backhaul,
D2D/V2V, stadium and gymnasium, are also taken into consideration. The 3GPP-
HF channel model is developed based on the 3GPP-3D MIMO channel model and
adopts both measurement-based and RT modeling methods. Though the modeling
methods adopted and the parameter values are the same as those in the 5GCM,
3GPP-HF gives more detailed implementation. These revisions include multi-
frequency correlation to support carrier aggregation (cluster delays and angles are
the same for all frequence bands, while other parameters are frequency dependent
or frequency independent), finer rays modeling for powers, delays, and angles to
support large bandwidth, and delay drifting over the array for each ray to support
massive MIMO (still not support spherical wavefront). Based on the modeling
method supporting spatial consistency proposed in the 5GCM, the 3GPP-HF pro-
vides the correlation distance required by spatial consistency, and generates the
spatial-domain contineous RVs used to express LOS/NLOS probability, indoor/
outdoor probability, type of building and related SSPs. Time evolution of the
channel was modeled based on locations of clusters. The 3GPP-HF channel
model adoptes the blockage modeling method proposed in the 5GCM and refines
the expression in the polar coordinates, i.e., the blockages are divided into two
catagories. Furthermore, the 3GPP-HF also proposes several simplified channel
models, i.e., five CDL models and five TDL models for link-level simulations.
Moreover, a map-based hybrid model developed by ZTE ltd. is adopted by
combinating the deterministric clusters obtained by ray tracing and the random
clusters obtained by stochastic modeling. The model can be used in the case when
the system performance is desired to be evaluated or predicted with the use of
digital map in order to investigate the impacts of environmental structures and
materials.
3.5.2 WINNER I/II/+
The aim of Working Package 5 (WP5, i.e., the channel model group) of WINNER I
is to present a broadband MIMO channel model in 5 GHz frequency range [163]. At
the beginning of the project, two channel models are considered, i.e. 3GPP SCM for
outdoor scenario and 802.11n IEEE model for indoor scenario. Because SCM only
supports a bandwidth of 5 MHz, WP5 extends the SCM model to the SCME
channel model. But it still cant meet the requirements of simulation. In the year
of 2004~2005, seven partners (Elektrobit, HUT, Nokia, KTH, ETH, TUI)
completed a large number of measurement activities, based on which the WINNER

I channel model was presented. This model supports frequency range of 5 GHz,
maximum bandwidth of 100 MHz, and six scenarios: indoor, UMa, UMi, SMa,
rural macro cell (RMa) and fixed relay. It provides the correlation coefficients for
four LSPs like DS, AS, and SF, in which AS contains angular spread in AOA and
AOD. At the same time, a Cluster Delay Line (CDL) model with fixed delay is
proposed to simplify the simulation. It is worth noting that the model provides
simple 3D parameters (zenith angle) for the A1 scenario.
The Working Group 1 (WP1) of WINNER II continued to extent the WINNER I
Model. Six partners (Elektrobit, Finland Oulu University, Wireless communica-
tions Center of Oulu University, TUI, Nokia and Communications Research Center
of Canada) completed a large number of measurements activities, and ultimately
obtained WINNER II Channel Model [13, 176]. The model supports frequency
range of 2~6 GHz, and includes 13 scenarios like indoor, outdoor, urban, rural,
macro cell, micro area, fixed, mobile, hotspot, etc. All of the scenarios include
NLOS and LOS cases. 3D features are supported in part of these scenarios (A1,
A2/B4/C4) (see Table 4.5, 4.6 in [13]). HSR propagation is supported in D2a
scenario with a maximum speed of 350 km/h. WINNER + Channel Model [164]
further extends the operating frequency down to 450 MHz, in which it refers to
Okumura Hata Model in 450~1500 MHz and refers to COST 231 COST-Hata
Model in 1500~2000 MHz. The model also provides several path loss models for
hexagonal layout in the NLOS case in addition to the Manhattan grid based model,
and improves zenith dimension parameters for the five scenarios including indoor,
O2I (UMi and UMA), UMi, UMA and SMa (see Table 4.3, 4.4, 4.5 and 3.13 in
[164]), which making it a real practicable 3D MIMO channel model.
WINNER family channel Model considers the correlation of intra-station and
inter-station LSPs, and provides the related correlation values based on the mea-
sured data. Intra-station correlation refers to the auto-correlation and cross-
correlation of several LSPs of a UT. Inter-station correlation is the LSP correlation
of two UTs with a certain distance apart, or the correlation of LSP at different
locations in the track of a mobile UT. The model uses a function exponentially
decaying with distance to characterize the inter-station correlation. The number of
clusters varies with the environment, and the number of rays within a cluster is set
to be 20 as SCM. In order to simplify the simulation, a fixed delay CDL model is
also provided. WINNER model also present a conceptual idea to support time
evolution of the channel, which can be realized by an old cluster ramping down
and a new corresponding cluster ramping up when transiting from segment to
segment.
WINNER model is a stochastic model, which embodys its randomness in the
following three aspects. The first is the randomness of the LSPs such as SF, DS and
AS. The second is the random SSPs such as power, arrival angle and departure
angle. The last is the initial random phase of each MPC, after drawing initial phase
randomly, different channel samples can be generated.
In general, the channel models of WINNER family describe the wireless channel
in different environments including outdoor, indoor, and O2I. Its biggest feature is
the use of unified mathematical model with different parameter sets. These param-
eterized models is built from a large number of measurement, which make them
widely be accepted and used. The models support the validation of various tech-
niques, such as multi-antenna, 3D MIMO, polarization, multi-user, multi-cell and
multi-hop network.
3.5.3 ITU IMT-Advanced, IMT-2020
Based on the field measurement by member organization and the WINNER II

channel model, the International Telecommunications Union (ITU) proposed a
IMT Advanced (IMT-A) channel model (aka ITU-R M.2135 channel model for
its document name) [14]. This channel model supports five scenarios, including
indoor hotspot (InH), typical UMa, UMi (including O2I), SMa and RMa. The
model adopts the ideas, methods and stochastic channel generation step from the
WINNER II model, however, it is only a 2D model and its parameter values have
changed a lot (see Table A1-2, A1-3 and A1-7 in [14]). Frequency band range of
2~6 GHz is supported for all scenarios, except frequency down to 450 MHz is
supported for RMa. In addition, another extended Time-Spatial Propagation (TSP)
model is suggested. The TSP model is a geometry-based double directional channel
model and is targeted for up to 100 MHz RF bandwidth. The path loss, PDP, AOD
are modelled by several closed-form functions in terms of the following key
parameters such as city structures (street width, average building height), BS
height, bandwidth and the distance between the BS and the UT. The model lack
the support of spatial consistency of large- and small-scale parameters.
Currently ITU is promoting the standarization of IMT-2020 (5G). In order to
ensure the process of 5G technical standard and rationality and fairness of the
follow-up technical assessment, a draft group (DG) for the IMT-2020 channel
model was established at the 25th meeting of the ITUs Radiocommunication
Sector (ITU-R) Working Party 5D (WP5D). The DG participants agreed to take
the 3GPP related channel models as the reference models in IMT-2020 Evolution
Report. The reference models support frequency band of 0.5~100 GHz and consist
of a Primary Module, an optional Extension Module and an optional Map-based
Hybrid Channel Module [178]. The Primary module inherits from 3GPP models
and is used for evaluation of Radio Interface Techniques (RITs) for the following
test environments, including eMBB(InH, UMi, RMa), URLLC (UMa) and mMTC
(UMa). The proposed Primary Model contains Model A recommended by China
and Model B recommended by Intel, Nokia, NTT DOCOMO, Sumsang, Ericsson
and Telstra. The former inherits the IMT-A model and 3GPP-3D models and
supplements the elevation parameters for InH and RMa in the sub-6 GHz, while
it inherits the 3GPP-HF models in the above 6 GHz. The latter uses unified
parameterization for the full frequency band instead of separate expressions for
different bands. The optional map-based hybrid channel model proposed by ZTE
and also adopted by the 3GPP uses both the ray-tracing-based deterministic and
measurement-based statistical modelling methodologies to better reflect the elec-

tromagnetic propagation features in terms of SC and DC in the actual propagation
environments. This hybrid channel model provides the frequency consistency
between high and low frequency domains, spatial consistency and dual mobility.
The optional Extension Module inherits from the IMT-A TSP model and uses the
elevation angle calculation methods in Primary Module which makes it being a 3D
model. The model is only applicable for the frequency band of 0.4~6 GHz and
bandwidth of 0.5~100 MHz.
3.5.4 COST 259/273/2100/IC1004
The telecommunication related actions supported byEuropean Cooperation in Sci-

ence and Technology (COST) organization have proposed several important chan-
nel models for wireless communication systems. The first proposed channel model
is COST 207 for the second generation mobile communication system GSM, which
is based on a simple TDL structure. The Following COST 231 channel model [173]
concentrates on path loss modeling and provide a COST Walfish-Ikegami model
and a COST Hata model which are still widely used by the WINNER channel
models and 3GPP channel models.
COST 259 Action focuses on the spatial characteristics of the wireless channel
and proposes a directional channel model [165]. The frequency range of the model
is 450 MHz~6 GHz and the bandwidth is less than 10 MHz. COST 259 classify the
radio propagation environment into three levels. The first level is the cellular type,
including the macrocell, microcell and pico cells. The second level is the wireless
environment. Each cellular type contains several communications environment,
such as macro cell consists of four propagation environments like typical urban,
unfavorable urban, rural and hilly environments. The third level is propagation
scenario for practical multipath environment simulation, mainly related to the
distribution of scatterers.
COST 259 assumes that the scatterers are fixed and only the UTs are movable.
All scatterers in the model are classified into two types, i.e., local clusters and
far-end clusters. Local clusters are located uniformly (Gaussian or others
distriubted) around the UT. The location of local scatterers will change with the
UT location. The far clusters are located far away from the BS and UT and can be
seen by them, such as high buildings, mountains, etc. The far-end scatterers will
usually be fixed in a particular area. COST259 presents the concept of visibility
region (VR). Only when the UT enters the VR of a scatterer, the scatterer can
contribute to the channel, which makes the adjacent users see the partially identical
propagation environment. Stochastic CIRs can be generated from the delay-angular
distribution obtained from the measurement, or generated by GSCM based on the
locations of the scatterers. The document of COST 259 give the transformation of
two methods.
COST 259 is only a single directional channel model. COST 273 Working
Group continued to extend this model, and presented a double directional MIMO
channel model [166]. COST 273 introduces three categories of cluster, namely the
local clusters, single-bounce clusters and twin-clusters. The latter two are the
differentiations of far-end scatterers in COST 259. The model provides a method
to calculate the size and the distance of a cluster departing from the BS and UT
based on its delay, delay and angular spread. The supported frequency range is
1~5 GHz, and the bandwidth is less than 20 MHz. The parametric modeling for
different environments is difficult for COST model arising from the hardness to
extract characteristics of scatterers from measurements, which limits the applica-
tion of the COST 273 model. Although COST 273 defines 22 typical environments,
only three of which are parameterized.
COST 2100 channel model [167, 174] continues to adopt the concept of VR
proposed by COST 259 and three clusters proposed by COST 273, and further
extends. Its main contribution are the suppots of polarization characteristics, DMC,
the time evolution and multilink simulink. A SC cluster is modeled as a spheroid
with a certain axis length and orientation according to the delay and angular
information of the cluster, while a DMC cluster is modeled as a biger spheroid
concentric with its corresponding SC. Although the modeling idea is advanced, the
reference code provided for downloading [175] is still very simple, many advanced
ideas have not been implemented, which limits its application.
Differing from the WINNER/IMT-Advanced model, the COST channel model
emphasizes that the scatterers exist objectively in the environment, rather than only
belong to one cluster. COST Action IC1004 focuses on the cooperative radio
communications for green smart environments and ends in 2016. The major
objective of its Working Group 1 (WG1) is to develop an integrated radio channel
model (including mmWave band, D2D/V2V and massive MIMO) and submit the
models to standardization organizations. In the application of massive MIMO,
assigning different VRs for different antenna groups at both BS and UT may
model the antenna non-stationary of the channel. Because the locations of the
scatterers are known, the COST model can be easily extended to support the smooth
time evolution and spherical wavefront [167, 174]. The existing COST model is
also designed for the scenario with one end fixed. It also needs to be further
extended for the D2D/V2V scenario and the extension is relatively easy. Based
on the advanced modeling ideas and methods, COST model is likely to be a
reasonable 5G channel model through appropriate extending, so it is worthwhile
to concern.
3.5.5 IEEE 802.11 TGn/TGac
IEEE 802.11 TGn channel model [171] is a spatial (antenna) correlation-based

stochastic channel model (CSCM). It also adopts the basic ideas of Saleh-
Valenzuela (SV) stochastic channel model. The channel model supports frequency
band of 2.4 GHz and 5 GHz with a maximum bandwidth of 40 MHz and at most
4 antennas. Specifically, it chooses three models A, B, C with small delay spread
among the five channel models presented by HiperLan/2 [177] and addes three
additional models for typical small office, large indoor and outdoor open space, to
form a total of six models A~F. The DSs of six models are 0ns (narrow-band
model), 15 ns and 30 ns, 50 ns, 100 ns and 150 ns respectively. The six models are
composed of 1, 2, 2, 3, 4, 6 clusters respectively. The paths (taps) within a cluster
have different delays but have the same average departure and arrival angle and
AS. The AS and DS of a cluster show strong correlation with correlation coefficient
of 0.7. For each taps, the PAS of a cluster can be assumed as uniformly distributed,
truncated Gaussian distributed, or truncated Laplacian distributed. Based on PASs,
the correlation matrix of the transmit antenna elements and that of the receiver
antenna can be calculated and denoted by RT and RR respectively. The correlation
matrix of the channel is expressed as the Kronecker product of RT and RR. The
model supports UT at a maximum speed of 1.2 km/h, which will lead to a Doppler
shift of 6 Hz at 5.25 GHz and 3 Hz at 2.4 GHz. Doppler power spectrum is a Bell-
shape instead of common U-shape in classic Jakes model. Noise filtering method
instead of Sum of Sinusoids (SoS) is adopted to generate the channel samples.
In order to support higher bandwidth, transmission rate and serve the multi-user
scenarios (MU-MIMO), IEEE 802.11 TGac working group proposed a TGac
channel model based on the TGn model [172], which can support up to 1.28 GHz
bandwidth by interpolating the TGn channel taps. For multi-user case, the angular
parameters of each UT are independently generated, which does not coincide with
the actual situation, so the performance of MU-MIMO may be overestimated. In
addition, according to the actual situation, the smaller moving speed of 0.089 km/h
is proposed, which corresponds to a coherent time of 800 ms or mean square
Doppler spread of 0.414 Hz.
3.5.6 QuaDRiGa, mmMAGIC
QuaDRiGa [19] is a channel model library proposed by Fraunhofer HHI Laboratory

and two partners, which greatly extends the WINNER model by introducing some
new features, including the support for time evolution (adjusted delay, AOA, AOD,
polarization, SF and K-factor according to the UT location in the same channel
segmet), scenario transition (using the idea of WINNER in different channel
segments), and variable speed of UTs (including acceleration and deceleration,
implemented by resampling at equidistant locations followed by interpolation
rather than sampling at equal time instants). It establishes a unified framework for
LOS and NLOS simulation which is convenient for simulation for a moving UT
undergoing both the LOS and NLOS cases. It can calculate the polarization
components for the LOS and NLOS case by using a ray-geometric approach. By
adopting two diagonal filters in addition to the two horizontal/perpendicular filters
used in WINNER II, the model get a more smooth correlated LSPs. The latest
version of this model supports spherical wavefront modeling. Besides the six
scenarios defined in the WINNER II model, namely A1, B1, B4, C1, C2, and C4,
it also adds new UMa measurement scenario in Berlin and 4 scenarios of satellite
ground mobile link. At present, QuaDRiGa does not support precise modeling of
massive MIMO.
The mmMAGIC project, mainly constituted by several European research
institutions including Fraunhofer HHI, Ericsson, and Aalto University, aims to
develop new wireless access technologies for the 5G communications in the
5-100 GHz frequency band. The WG2 focuses on channel modeling and channel
measurement. The mmMAIGC concentrates on the scenarios like street canyon,
open square, indoor office, shopping mall, airport hall, subway station, O2I, and
gymnasium. The mmMAGIC model is developed based on the 3GPP-3D model and
adopts a combination modeling methods composed of measurement, RT and point
cloud field prediction. The model is implemented based on the QuaDRiGa model
library, adopts the KED blockage modeling method and more finer modeling for the
power, delay, and angle of the rays within a cluster. Currently it only synthesizes
the parameters from several existing channel models such as WINNER3GPP-
3DMETIS, and 5GCM and provides some initial model parameters.
3.5.7 IEEE 802.15.3c/ IEEE 802.11ad/aj/ay
IEEE 802.15.3c is the first 60 GHz channel model in the world proposed by IEEE
802.15 (WPAN) working group [8]. It extends the traditional SV channel model to
support arriving angles. However, it only provides the azimuth information, so it is
a 2D channel model. The main measurement to produce the model was completed
by several institutes including Japanese NICT and German IMST, etc. It contains
ten channel models CM1~CM10 for six scenarios including living room, office,
library, conference room, desktop and corridor. The model does not specify the
applicable frequency range and frequency band. According to the configuration of
the measurement equipment, we can infer its applicable frequency range is about
59~64 GHz, and the maximum bandwidth is no more than 3 GHz.
802.11ad IEEE channel model [9] is the channel model proposed for WLAN
system with ultra high data rate operating at unlicensed 60 GHz. It extends the SV
model to support the azimuth and zenith angle at both Tx and Rx, so it is a double-
directional 3D channel model. The model supports three kinds of indoor scenarios
(conference room, office, living room) and two link types of access and D2D. The
modeling methods adopted include RT, measurement, empirical distribution and
theoretical models. RT is used to determine the delay and average azimuth / zenith
angle of the clusters. The main clusters include LOS, one-order and two-order
reflection components. Empirical distribution is used to describe the amplitude and
inter-cluster angle distribution of the reflection paths. Theoretical model is used to
describe polarization characteristics. The main parameters of rays within a cluster
are obtained by measurements. Apart from the classical SV model, in IEEE
802.11ad model the rays within a cluster are divided into pre-cursor and post-cursor
rays. Parameter fitting is done in the two part separately. The azimuth and zenith
angles of the rays within a cluster are independently normal distributed. Different
from the other channel models such as WINNER in which path loss, SF and small-
scale path gain together determine the strength of the clusters and rays, IEEE
802.11ad channel model doesnt distinguish path loss and small scale fading, but
independently generates a path gain for each ray. In general, the model can provide
accurate characteristics of the channel in space and time domain, and support
beamforming, polarization, and consider the blockage loss caused by human body.
In 2012, IEEE 802.11 Working Group established the IEEE 802.11aj Task
Group, which targets the next generation WLAN standards for mmWave band of
45 GHz in China. A series of channel measurements were carried out by Key Lab of
mmWave at Southeast University. A path loss model for three indoor access
scenarios was proposed and the delay spread was analyzed. In March 2015, IEEE
802.11ay Working Group was established to develop a standard for the next
generation 60 GHz transmission system, which intended to extend the application
scope of IEEE 802.11ad with the support for Backhaul and Fronthaul as well as
mobility and the minimum band width of 4 GHz [179]. The channel model
developed has several features: extending the indoor SISO channel models of
IEEE 802.11ad to the MIMO channel models, using Quasi-Deterministic (Q-D)
methodology to build channel models for new scenarios, including open area
outdoor hotspot access, outdoor street canyon hotspot access, large hotel lobby,
ultra short range, and D2D communications. Note that except the D rays and R rays
as like the MiWEBA model, a third type of rays (F-rays) that appear for a short
period of time, e.g., a reflection from the moving cars and other objects, may be
introduced and described with the same way as the R-rays for the special
non-stationary environments.
3.5.8 MiWEBA
Millimetre-Wave Evolution for Backhaul and Access (MiWEBA) [4] is a research

project committes to the promotion and application of mmWave communications.
Its main participants include Fraunhofer HHI, Intel Mobile communications (IMC),
and several other universities and research institutes in Japan, France, and Italy. The
WG5 focused on communications and antenna and released its first version of
channel model in the mmWave band in June 2014 [4]. It models the channel of
mmWave band of 57~66 GHz for three indoor and outdoor scenarios, including
university campus, street canyon and hotel lobby, each of which supports three link
types, i.e., access, Backhaul/Fronthual and D2D. Fraunhofer HHI and IMC have
completed a large number of measurements, which reveal that in the outdoor
environment the diffraction components at 60 GHz is very small and can be
neglected. A Q-D modeling approach is used to represent mmWave channel
being composed of LOS and a few reflection paths (called Quasi-Deterministic
rays) as well as several stochastic clusters (called R rays). The D rays and its
parameters like path delay, power, angle and polarization can be determined
through ray tracking according to the propagation environment. Whereas the
parameters of lower power R rays reflected from far walls or random objects
(cars, lampposts etc.) or second-order reflected can be obtained through measure-
ment and analysis. This approach follows the IEEE 802.11ad model, but there exist
some differences. The reflection coefficient of a Q-D ray is calculated using the
Fresnel equation in addition to considering the roughness of reflection surface.
Moreover, there are only post-cursor rays within a cluster.
3.5.9 METIS
In November 2012, as a pioneer in 5G Research and Development (R&D), EU

launched Mobile and wireless communications Enablers for the Twenty-twenty
Information Society (METIS) project with 29 participating entities including
Huawei Company of China. One of the important tasks of METIS is to propose a
set of channel models suitable for 5G R&D. METIS proposes a flexible and scalable
channel modeling architecture to meet the requirements of accuracy and computa-
tional complexity. The whole channel model consists of a map-based deterministic
channel model, a stochastic channel model or a mixed model of both. The
map-based model, which is based on ray tracing using a simplified 3D geometric
description of the propagation environment, supports the frequency range of
2~100 GHz. The significant propagation mechanisms of diffraction, reflection,
diffuse reflection and blockage are taken into considerations. The model claims to
meet all the requirements of the 5G channel model, including massive MIMO
(pencil beamforming, spherical wavefront, and antenna non-stationary characteris-
tics), D2D double mobility, dynamic modeling, etc. The METIS used the KED
method to model the blockage. The map-based model is verified by comparison
with the measurement results in some specific scenarios. At the project beginning
stage, METIS participants have completed extensive measurements of 2~60 GHz in
the scenarios of dense UMa, UMi, indoors, shopping malls, D2D, and vehicle link.
Based on these measurements and adopting the existing channel models like
WINNER II/+, 3GPP 3D/D2D and IMT-Advanced, METIS present a GSCM
based stochastic channel model and channel parameter list for various propagation
scenarios. Moreover, it describes 60 GHz mmWave channel under the framework
of GSCM model and gives a stochastic channel model for the scenarios of shopping
center, coffee house and square at 50~70 GHz, in which minority of the parameters
are obtained by measurement and most of the parameters are calculated by point
cloud field prediction method [180]. METIS also improves the modeling methods
such as correlated large-scale paremeters generation and angle calculation of rays
within a cluster. The GSCM (GGSCM) proposed in the first version [181] which
supports spatial consistency is adopted by several subsequent channel models.
Mixed model provides a flexible and scalable channel modeling framework,

which balances between simulation complexity and accuracy. For example, SF is
generated based on map-based model while small scale fading is based on a
stochastic model.
3.5.10 5GCM
The 5GCM is a 5G mmWave channel model alliance initiated by the U.S. National
Institute of Standards and technology (NIST) and includes many companies and
universities such as NIST, NYU, AT&T, Qualcomm, CMCC, Huawei, and BUPT.
The typical scenarios in the 5GCM are UMi (urban street canyon and open square),
O2O (outdoor-to-outdoor)/O2I, UMa O2O/O2I, and InH (open or closed indoor
office and shopping mall). The 5GCM is developed based on the 3GPP-3D channel
model and adopt the muti-frequency channel measurement and RT modeling
method. Path loss, LSPs, penetration losses, and blockage models in the LOS and
NLOS cases for several scenarios have been obtained. The path loss is modeled by
the CI, CIF (CI with frequency), and ABG model. Dual-slope CIF model and ABG
model are also provided for InH scenario in the NLOS case. For the O2I penetration
loss, two models were provided, i.e. low loss model and high loss model for the
external wall with different glass. Without doubt the path loss and penetration loss
will depend on the frequency band. However, the LSPs do not show a certain
frequency dependency by analyzing the measurement results. A weak frequency
dependence of LSPs is only discoveried by the RT technique. The correlated
distances and correlation coefficients among multiple LSPs are inherited from the
3GPP-3D model parameters. Three methods are proposed to support spatial con-
sistency, i.e., (1) spatial-time-frequency consistency RVs used to generating the
LOS/NLOS state, path gain, delay, and angles of rays, etc, are obtained by inter-
polating the RVs on the regular grids at the UT location; (2) Dynamic evolution
method similar to the one presented in QuaDriGa; (3) the GGSCM modeling
method proposed by METIS. For the blockage modeling, the KED expressions in
the Cartesian coordinates and polar coordinates are provided. The parameters of
two typical blockages, i.e., human body and vehicle, are suggested. Since the
measurement campaigns have not been finished, the current model is not the final
version.
3.5.11 Comparison of Existing Models
With the progress of 5G research and development, more channel model are built
by the methods combing the deterministric modeling based on Ray Tracing and
stochastic modeling based on channel measurement. The existing channel models
still cannot meet the requirements of 5G mobile communications. Though it is

claimed that the METIS map-based model can support all requirements, it needs to
be further verified and it is too complex to be used in real-time simulation.
Compares and summarizes on the existing wireless channel models meeting the
various requirements of the future 5G channel models were shown in Table 3.12. It
is based on Table 4.1 in [1], as well as some other literatures and our knowledge.
The frequency bands at which the channel are measured and built for those
channel models mainly include sub-6 GHz, 10~11 GHz, 14~15 GHz, 18 GHz,
26 GHz, 28~29 GHz, 38~40 GHz, 45 GHz, E band (71~76 GHz, 81~86GHz) and
60 GHz band. IEEE 802.11ad, MiWEBA and METIS have only measured the
channel at 60 GHz band. The complete measurements and modeling for the full
mmWave bands are ongoing. The frequency dependence of LSPs and SSPs are
being obtained by channel measurements and ray tracing simulations. Higher
bandwidth and carrier aggregation techniques are required for the future wireless
communication. Both the GSCM and map-based model in the 3GPP-HF and its
descendent 3GPP TR 38.901 [182] provide efficient methods to support big band-
width and frequency consistency. With respect to the high mobility, WINNER
model can support HSR environment with maximum speed of 350 km/h. But as the
HSR speed continues to increase (e.g., 500 Km/h) and the scenarios that a train
traveles become more diverse (mountains, hills, etc.), it is required to conduct
enough channel measurements and channel modeling for these new HSR scenarios.
Many models dont support dual mobility (D2D/V2V), spatial consistency and time
evolution (dynamic simulation). For spatial consistency, it is firstly to gurantee the
correlation continuity of LSPs, without which we will get incorrect results [183]. To
guarantee the continuity, WINNER calculates a correlation table in advance at each
location and height (with resolution from a few meters to tens of meters). But with
the diversification of dual locations and antenna heights in D2D, the size of the
correlation table will be extremely large which will lead to the dramatic increase in
computational complexity. The spatial consistency also ask for the spatial continu-
ity of SSPs in dynamic simulations. It is also needed to describe the channel for
close links, e.g, if two UTs are close to each other, they should see the similar
scattering environment, and thus have similar AOA and AOD. If neglecting this, the
performance of multi-antenna technology will be overestimated. We have seen the
achievement in this aspects made by those channel models like 3GPP-HF,
mmMAGIC, 5GCM and IMT-2020. With respect to large antenna arrays, almost
all the models support pencil beamforming, yet the models in the framework of
GSCM dont support spherical wavefront and antenna non-stationarity, however
both of which are supported by the ray tracing (map-based) models and COST
model based on the location of scatterers.
Table 3.12 Comparison of the existing models
Maximum Massive MIMO Frequency
Frequency speed Beamforming Spherical Antenna Dynamic Spatial consistency /
Characteristics band [GHz] Bandwidth [MHz] [km/h] ( 1 ) wavefront non-stationarity 3-D D2D Backhaul simulation consistency dependency
3GPP SCM 1~3 5
3GPP 3D 1~4 100 3 Limited Yes
3GPP D2D 1~4 100 60 Limited Limited
3GPP-HF 6~100 10% fc 500 Yes Not yet Not yet Yes Yes Not yet Yes Yes Yes
(Not yet)
WINNER II/+ 0.45~6 100 350 Limited Yes Yes Very Limited
Limited
IMT-2020 0.5~100 100(<6 GHz) 500 Yes Not yet Not yet Yes Yes Not yet Yes Yes Yes
10% fc (Not yet)
QuaDRiGa/ 100~2 GHz 350 Yes Yes Yes Rx Side Yes Not yet Not yet
mmMagic ( Not yet)
IEEE 802.11n/ac 2.4/5 1.28 GHz Very Low Yes
IEEE 802.11ad/ 57~68 2.16 GHz,2.64 GHz 3~5 Yes Yes Yes Yes Limited
ay (ay)
MiWEBA 57~66 Slow Yes Yes Yes Yes Limited Yes
COST 0.45~100 Yes Yes Yes Yes Yes Yes Yes Yes
5GCM 0.5~100 100 (<6 GHz) 350 Yes Partly Yes Yes Yes Yes
~2 GHz (>6 GHz) (one side fixed)
METIS GSCM 0.45~6,61~65 100 (<6 GHz) 200 Yes Yes Limited <6 GHz
4 GHz (63 GHz)
Map- 100 10% fc 250 Yes Yes Yes Yes Yes Yes Yes Yes Yes
based
Note: fc carrier frequency
3.6 Stochastic Channel Generation
Stochastic channel samples are required in system simulations. Taking 3GPP 3D

channel model as an example, this section gives the general process of stochastic
channel generation, as shown in Fig. 3.9, which is similar to that of WINNER II and
METIS (GSCM). The following channel generation process is suitable for down-
link simulation, however, it can be modified for the case of uplink simulation by
simply exchanging DOA and AOA.The whole generation process is composed of
four parts: definition of geometric parameters of simulation scenarios, generation of
LSPs, generation of SSPs and generation of channel coefficients.
Scenarios parameters Large scale parameters
Set scenarios, Determine

Generate LSPs
network layout propagation Calculate path
DS,AS(ASD,ASA,
and antenna conditions( loss PL
ZSD,ZSA),SF,K
info. LOS/NLOS)
Small scale parameters
Perform Generate arrival &

Generate Generate
Generate random departure angle:
cluster powers multipath
XPRs coupling of AOD,AOA,
Pn delays
rays ZOD,ZOA

Channel coefficients
Apply antenna
Generate Generate
mutual
random initial channel
coupling,
phases coefficients
PL,SF
Fig. 3.9 Channel coefficients generation procedure of 3GPP TR36.873 [170]

3.6 Stochastic Channel Generation 127
3.6.1 Definition of Simulation Scenarios
This part is mainly related to the configuration of simulation environment, network

layout and antenna parameters. The specific steps are as follows.
1. Determining system operating center frequency fc. The effective frequency can
be set to 450 MHz~6 GHz.
2. Choosing simulation scenarios, 3D-Uma or 3D-UMi.
3. Determining number of BSs and UTs in the network layout.
4. Determining the antenna 3D polarimetric radiation patterns Ftx(,) and
Frx(,) of BSs and UTs in the Global Coordinate System (GCS) and array
geometries.
5. Generating 3D geometric locations of BSs. In system level simulation, the entire
network is usually laid out in hexagonal mesh, as shown in Fig. 3.10. The BSs
are placed on mesh points. The inter-site distance (ISD) is typically set to 200 m
(3D-UMi) and 500 m (3D-Uma) [14]. The height of BS, hBS, is usually set to
10 m (3D-UMi) and 25 m (3D-Uma).
6. Generating 3D geometric locations of UT. The horizontal two-dimensional
coordinates of UTs are distributed uniformly in the whole network, and the
heights depend on whether the scenario is indoor or outdoor. In 3GPP TR36.873
[170], it is assumed that 80% of the users are located indoors. The heights of
outdoor users are fixed on 1.5 m above the ground while the heights of indoor
users are highly dependent on the floors that they stay in the buildings. For each
Fig. 3.10 Cellular mesh layout [14]

Fig. 3.11 2D and 3D distances of outdoor (left) and indoor (right) users
building, the number of floors, N, are uniformly distributed in 4~8 with floor
height of 3 m. The floors that the UTs may stay, nfloor, are uniformly distributed
between 1~N so that the heights of indoor users can be written as hUT 3(nfloor-
1)+1.5. The distances between indoor users and the BS can be expressed as
(Fig. 3.11)
q
d 3D d3Dout d3Din d2Dout d2Din 2 hBS hUT 2 : 3:63
It is necessary to ensure that the UT is within the Fraunhofer Region (FR) of the
BS antenna, that is, d3D should be larger than the Fraunhofer distance. In the past, it
was guaranteed by ensuring that the horizontal distances between UT and BS are
more than 35 m (UMa) or 10 m (UMi). If the locations of BS and UT are given,
AoD (LOS,ZOD, LOS,AOD) and AOA (LOS, ZOA, LOS, AOA) of the LOS path can be
determined. In D2D scenario, it is necessary to generate two locations for two UTs
in each link.
7. Generating array orientations of the BSs with respect to the GCS. The BS array
orientations is defined by three angles, i.e., bearing angle BS, downtilt angle
BS and slant angle BS. The bearing angles can be determined in the phase of
network deployment. The mechanical downtilt angles is usually set to
12 degrees, while the slant angle is usually set to 0 degrees. These three angles
constitute the 3D rotation vector (BS, BS, BS) of BS Local Coordinate System
(LCS) relative to the GCS.
8. Generating array orientations of the UTs with respect to the GCS. The array
orientations of each UT is determined by three angles: bearing angle UT,
Table 3.13 LOS transmission probability (3GPP TR36.873 [170])

Transmission
scenarios LOS transmission probability
8
3D UMi < 1, d2D 18 m
Outdoor users:PLOS 18 d2D 18 d2D
: exp , d2D > 18 m
d2D d 2D 36
Indoor users: replacing d2D with d2D_out
3D UMa Outdoor8 users:
< 1C d2D ; hUT
, d2D 18 m
PLOS 18 d 2D 18 d2D
: exp 1 Cd2D ; hUT , d2D > 18 m
d 2D d2D 63

0, hUT < 13 m
Cd 2D ; hUT hUT 131:5
g d 2D , 13 m hUT 23 m
8 10
< 1:25 2 d2D
d2D exp , d 2D > 18 m
gd 2D 6 150
: 10
0, d 2D 18 m
Indoor users: replacing d2D with d2D_out
downtilt angle UT and slant angle UT, which can be randomly generated or
based on the practical situation. These three angles constitute the 3D rotation
vector (UT, UT, UT) of UT LCS relative to the GCS.
9. Determining moving velocity vector of UT in the GCS, including velocity
v and motion direction (v, v). The motion direction is set to be random in
horizontal plane. The typical velocity can be set to 3 km/h, or be set according
to practical situation. In D2D scenario, the velocity vector of both ends are
generated with this method.
10. Determining the LOS/NLOS conditions of a link. For indoor and outdoor users,
the propagation conditions (LOS/NLOS) are determined according to
Table 3.13.
Going through the above steps, a complete GSCM simulation environment can
be built. All the location information and array configuration of BSs and UTs are
defined in this simulation environment, in which all transmission links are deter-
mined as well.
3.6.2 Generation of Large Scale Parameters
This subsction contains two aspects. One is to generate the path loss of all links
according to the definition of simulation environment. The other is to generate LSPs
of all links, such as SF, Ricean K-factor, DS, AS, etc.
Table 3.14 3D-UMi path loss model (3GPP TR36.873 [170])

Path loss model SF SF Applicable distance
fc[ GHz], d [ m ] [ dB ] [m]
LOS PL 22.0log10(d3D) + 28.0 + 20log10( fc) 3 0
10 < d2D < d BP
0
3 dBP < d2D < 5000
PL 40log10d3D 28:0 20log10 f c
2
9log10 d0BP hBS hUT 2
NLOS PL maxPLNLOS ; PLLOS 4 10 < d2D < 2000
PLNLOS 36:7log10 d3D 22:7 26log10 f c
0:3hUT 1:5
O2I 8 PL PLb PLtw PLin 7 10 < d2D < 1000
< PLb PL3DUMi d3Dout d3Din 0 < d2D in < 25
PLtw 20
:
PLin 0:5d2Din
3.6.2.1 Path Loss
The large scale fading models differ with the propagation scenarios and conditions.
Table 3.14 shows the path loss model in 3D-UMi scenario. Unlike IMT-A path loss
model, 3D-UMi scenario only takes the traditional cellular scenarios into account
and does not involve the Manhattan grid deployment. In addition, the SF in 3D-UMi
scenario is lognormal distributed and its variance is also given in Table 3.14.
When the UT is located outdoors, the link can be either LOS or NLOS. In NLOS
conditions, the PLLOS is the path loss assuming the propagation condition is LOS at
0
the same location apart from BS. dBP is the 2D breakpoint distance of PL model.
0 4hBS hE hUT hE
d BP 3:64

where hE is the effective height (in meters) of surrounding environment that the link
exists, which is set to 1 in 3D-UMi scenario. hBS and hUT are the antenna height of
BS and UT, respectively. Generally, hBS is lower than the average height of
surrounding buildings, and hUT is in range of 1.5~22.5 m. When the user is located
indoors, PLb is the basic path loss, which is equal to PL3D-UMi, namely the path loss
assuming the UT is outdoors. PLtw is the penetration loss and set to 20 dB in cellular
cell. PLin is the indoor path loss, which is related to d2D-in, i.e. the perpendicular
distance between the wall and the UT. d2D d2D-in + d2D-out, with d2D-in uniformly
distributed in 0~25 m. The Antenna height can be expressed as hUT 3(nfloor 1) +
1.5, where nfloor is the floor on which the UT locates, with the value usually set to
1~8. nfloor 1 corresponds to the ground floor.
In 3D-UMa scenarios, the path loss model is shown in Table 3.15. Its SF is also a
logarithmic model, and specific fading parameters are also given in the table.
In NLOS transmission, W is the width of the street (in meters), which is generally
set to 5~50 m. h is the average height of buildings (in meters), which is generally set
Table 3.15 3D-Uma path loss model (3GPP TR36.873 [170])

Path loss SF SF Applicable distance
PL [dB], fc [GHz], d [m] [ dB ] [m]
LOS PL 22.0log10(d3D) + 28.0 + 20log10( fc). 4 0
10 < d2D < d BP
0
4 dBP < d2D < 5000
PL 40log10d 3D 28:0 20log10 f c
2
9log10 d0BP hBS hUT 2
NLOS PL max(PLNLOS, PLLOS) 10 < d2D < 5000
PLNLOS 161:04 7:1log10 W 7:5log10 h
24:37 3:7h=hBS 2 log10 hBS
43:42 3:1log 10 hBS log10 d3D 3
20log10 f c 3:2log10 17:6252 4:97
0:6hUT 1:5
O2I PL
8 PLb + PL tw + PLin 7 10 < d2D < 1000
< PLb PL3DUMa d 3Dout d3Din 0 < d2Din < 25
PLtw 20
:
PLin 0:5d2Din
to 5~50 m. The path loss model in LOS condition is similar to that of 3D-UMi
scenario except the value of hE. When the link is LOS transmission, the probability
of hE equaling 1 m is, p(hE 1m) 1/(1 + C(d2D, hUT)), where the definition of
C(d2D, hUT) has been described in Table 3.13. Otherwise hE will get values uni-
formly from the set of {12, 15, ... , (hUT1.5)}. In addition, when the UT is located
inside the building and the link is O2I transmission, the path loss model and the
definition of its parameters are the same as the O2I path loss model in 3D-UMi
scenario.
3.6.2.2 Other Large Scale Parameters
Other LSPs, including SF, Ricean K-factor, DS, and AS (ASA, ASD, ZSA, ZSD)
remain to be generated. Different LSPs of the same UT show certain correlation
between them, which are known as intra-station correlations. In multi-link simula-
tion, different links (generally only the UTs connected to the same BS are consid-
ered) of each LSP show also certain correlation between them, which are known as
inter-station correlations. Assume there are K links corresponding to K user located
at (xk, yk) and each link has M LSPs. There are a total of NM*K large scale
parameters. Usually these parameters are lognormal random variables, and they are
anN-by-N matrix C.
correlated with each other. The correlation matrix is denoted asp
N LSPs random variables can be generated by multiplying C with an N-by-1
column vector formed by N independent standard normal RVs. Obviously, the
computation of LSPs with tens or hundreds of links in system level simulation is
extremely huge. Therefore, WINNER suggests to calculate the intra-station corre-
lations and the inter-station correlations separately.
1. For each LSP parameter, the inter-station correlation is calculated first, which
can be completed through the LSP map. The specific procedure of this method is
as follows. Firstly, generating a two-dimensional mesh based on UT locations.
The region of the mesh is outwardly extention of a rectangle covering all UTs by
double correlation distances. For each mesh grid, M standard normally distrib-
uted RVs (corresponding to M LSPs) are assigned and filtered by their
corresponfing 2D FIR filter separately. Impulse response of the filter for the m-
th LSP can be expressed as:

d
hm d exp 3:65
dm, cor
where d is distance. dm,cor is the correlation distance of the m-th LSPs, which is
dependent on simulation environments and links condition. The specific parameters
are given in Table 3.16 and 3.17. Finally, M filtered LSPs, m (xk, yk), are obtained at
every user locations (grids). The above procedure can only ensure that the inter-
station correlation decreases negatively exponentially with the distance in horizon-
tal and perpendicular directions. QuaDriGa proposes a more reasonable method,
considering the correlation between the two diagonal directions. It uses two sets of
filters. One set is applied in horizontal and perpendicular directions, just as WIN-
NER, and the other is applied in the two diagonal directions. Assuming the grids
spacing is dpx, the filter coefficients at distance kdpx for two filter sets are

1 kd px
am k p exp
d m, cor dm, cor
p
1 k 2dpx
bm k p exp : 3:66
dm, cor d m, cor
This method can discretize the whole two-dimensional plane into finer mesh
grids (less than 1 m), and the LSPs of UTs can be obtained by interpolating the LSPs
at adjacent mesh grids [19]. Whereas in WINNER, the smallest location resolution
of the UT coordinates and mesh grids is one meter.
The above method is only effective for 2D plane. When the UTs are distributed
in 3D space, especially when there exists two-way movement in D2D/V2V scenar-
ios, there will be totally 6D coordinate locations for both ends. Therefore, the
filtering method will be very difficult to use. METIS model introduces a new
method of Sum of Sinusoids (SoS) to describe the inter-station LSP correlations
[184]. Specifically, taking SF as an example, it can be expressed as
r
2 2SF X M
m m

SF sin D 3:67
M m1
Table 3.16 Large-scale parameters 3GPP-3D channel model (3GPP TR36.873 [170])
3D-UMi 3D-UMa
Scenario LOS NLOS O-to-I LOS NLOS O-to-I
DS log10([s]) DS 7.19 6.89 6.62 7.03 6.44 6.62
DS 0.40 0.54 0.32 0.66 0.39 0.32
AOD spread ASD 1.20 1.41 1.25 1.15 1.41 1.25
(ASD) log10([ ]) ASD 0.43 0.17 0.42 0.28 0.28 0.42
AOA spread ASA 1.75 1.84 1.76 1.81 1.87 1.76
(ASA) log10([ ]) ASA 0.19 0.15 0.16 0.20 0.11 0.16
ZOA spread ZSA 0.60 0.88 1.01 0.95 1.26 1.01
(ZSA) log10([ ]) ZSA 0.16 0.16 0.43 0.16 0.16 0.43
SF [dB] SF 3 4 7 4 6 7
Kfactor (KF) K 9 N/A N/A 9 N/A N/A
[dB] K 5 N/A N/A 3.5 N/A N/A
Crosscorrelation ASD vs DS 0.5 0 0.4 0.4 0.4 0.4
ASA vs DS 0.8 0.4 0.4 0.8 0.6 0.4
ASA vs SF 0.4 0.4 0 0.5 0 0
ASD vs SF 0.5 0 0.2 0.5 0.6 0.2
DS vs SF 0.4 0.7 0.5 0.4 0.4 0.5
ASD vs ASA 0.4 0 0 0 0.4 0
ASD vs K 0.2 N/A N/A 0 N/A N/A
ASA vs K 0.3 N/A N/A 0.2 N/A N/A
DS vs K 0.7 N/A N/A 0.4 N/A N/A
SF vs K 0.5 N/A N/A 0 N/A N/A
ZSD vs SF 0 0 0 0 0 0
ZSA vs SF 0 0 0 0.8 0.4 0
ZSD vs K 0 N/A N/A 0 N/A N/A
ZSA vs K 0 N/A N/A 0 N/A N/A
ZSD vs DS 0 0.5 0.6 0.2 0.5 0.6
ZSAvs DS 0.2 0 0.2 0 0 0.2
ZSD vs ASD 0.5 0.5 0.2 0.5 0.5 0.2
ZSA vs ASD 0.3 0.5 0 0 0.1 0
ZSD vs ASA 0 0 0 0.3 0 0
ZSA vs ASA 0 0.2 0.5 0.4 0 0.5
ZSD vs ZSA 0 0 0.5 0 0 0.5
Correlation dis- DS 7 10 10 30 40 10
tance ASD 8 10 11 18 50 11
(in horizontal ASA 8 9 17 15 50 17
plane) [m]
SF 10 13 7 37 50 7
K 15 N/A N/A 12 N/A N/A
ZSA 12 10 25 15 50 25
ZSD 12 10 25 15 50 25
Table 3.17 ZSD and ZOD offset of 3D-UMa and 3D-UMi (3GPP TR36.873 [170])
3D-UMa 3D-UMi
LOS/ O-to-I NLOS/ O-to I LOS/ O-to-I NLOS/ O-to I
Scenario LOS NLOS LOS NLOS
ZOD ZSD max[0.5, 2.1 max[0.5, 2.1 max[0.5, max[0.5, 2.1
spread (d2D/1000) (d2D/1000) 0.01 2.1(d2D/ (d2D/1000)
(ZSD) 0.01 (hUT (hUT 1.5)+0.9] 1000)+0.01| +0.01max(hUT
log10([ ]) 1.5)+0.75] hUT hBS| hBS,0) +0.9]
+0.75]
ZSD 0.40 0.49 0.4 0.6
ZOD offset, 0 10^ 0 10^
offset ZOD {0.62log10(max {0.55log10(max
(10, d2D)) (10, d2D))+1.6}
+1.930.07
(hUT1.5)}
where M is the number of sinusoids. SF is the target standard deviation. D is the 6D

coordination vectors of a transmitter and a receiver. k is the random phase
[0,2). k is the k-th wave vector with arbitrarily direction
uniformly distributedpin
and norm of
7 2=2dcor . Figure 3.12 shows an example that only contains a
3D correlation diagram at one end. Other LSPs can also be obtained by this method.
2. Next, for k-th UT, linear transformation can be used to get M LSPs with intra-
station correlations.
p
~s xk ; yk CMxM 0xk ; yk 3:68
The specific parameters of the elements in correlation matrix CMxM(0) are also
given in Table 3.16. All seven LSPs can be represented as normally distributed RVs
in logarithm domain. But they are somewhat DS and four ASs use log10()
different.
processing method, i.e., log10 X N lgX ; lgX , which is equivalent to
2
X 10N lgX ; lgX . SF and K use 10 log10() processing method, i.e.

2
10 log X N ; 2 , which is equivalent to X 10 N lgX ; lgX =10 . The mean

2
10
and variance of each LSP are also given in Table 3.16. In the above processing,
we start from the standard normally distributed RVs and get the random variables
~s xk ; yk that can reflect the inter-station and intra-station correlations. Since the
diagonal element of correlation matrix is one, ~s xk ; yk is still a standard normally
distributed RV. The final LSPs can be expressed as X 10 ~s xk ;yk (applicable to
DS and four ASs) and X 10~s xk ;yk =10 (applicable to SF and K).
The AS generated randomly is limited. ASA and ASD should not exceed 104 , i.e.,
ASA min(ASA,104 ), ASD min(ASD,104 ). Zenith angular Spread of Arrival
(ZSA, or ESA) and Zenith angular Spread of Departure (ZSD, or ESD) should not
exceed 52 , i.e., ZSA min(ZSA, 52 ), ZSD min(ZSD, 52 ).
Fig. 3.12 An example of 3D SF [1]
3.6.3 Generation of Small Scale Parameters
This subsection will discuss the generation the SSPs, including cluster parameters
(delay, power, angle, XPR) and intra-cluster parameters (delay, power and angle of
rays in clusters).
1. Generating the relative delay of multipath .
In 3GPP 3D channel model, the cluster delays follow exponential distribution,
which can be generated by the methods specified in Table 3.11, i.e.,
0n r ln Xn : 3:69
Where 0n is the absolute delay of the n-th cluster. r is the ratio of standard deviation
of delay distribution delays over DS , i.e. r delays/ . Xn is a RV uniformly
distributed in [0, 1]. A total of N cluster delays are generated. The value of
N depends on the environment, which is given in Table 3.18. The delays are
being normalized and sorted in ascending orde, i.e., 10, n-1n, n 2, . . ., N

n sort 0n min 0n : 3:70
In the case of LOS conditions, additional scaling of delays is required to

compensate for the impact of LOS power on the delay spread,
Table 3.18 Small-scale parameters of 3GPP-3D channel model (3GPP TR36.873 [170])
3D-UMi 3D-UMa
Scenario LOS NLOS O-to-I LOS NLOS O-to-I
Delay distribution Exp Exp Exp Exp Exp Exp
AOD and AOA distribution Wrapped Gaussian Wrapped Gaussian
ZOD and ZOA distribution Laplacian Laplacian
Delay scaling factor r 3.2 3 2.2 2.5 2.3 2.2
XPR[dB] 9 8.0 9 8 7 9
3 3 5 4 3 5
Number of clusters 12 19 12 12 20 12
Number of rays per cluster 20 20 20 20 20 20
Cluster ASD 3 10 5 5 2 5
Cluster ASA 17 22 8 11 15 8
Cluster ZSA 7 7 3 7 7 3
Per cluster shadowing std [dB] 3 3 4 3 3 4
n
nLOS : 3:71
CDS
The scaling factor is related to the Rice factor K, that is,
CDS 0:7705 0:0433KF 0:0002KF2 0:000017KF3 , 3:72
with KF being the Ricean K-factor in dB scale generated in the phase of LSPs
generation.
2. Generating cluster power P.
Cluster powers are calculated assuming a single slope exponential power delay
profile, which is a basic assumption for all GSCMs. The average cluster powers
decay exponentially with the increase of delay, given by

r 1 CSFn
P0n exp n 10 10 , 3:73
r

where CSFeN 0; 2CSF is the per cluster shadow fading (CSF) in dB scale, whose
statistical parameters are also specified in Table 3.18. Note that the delays specified
in Equ. (3.70) are used to generate cluster powers, while not the scaled delays,
i.e. Equ. (3.71), even in the case of LOS conditions. The sum power of all clusters is
normalized to one, i.e.
P0
Pn P N n : 3:74
n1 P0n
In the case of LOS conditions, additional specular component is added to the first
cluster. The cluster powers are given by
1
PnLOS Pn n 1P1, LOS
KF 1 3:75
1 KF
Pn n 1 ,
KF 1 KF 1
where P1, LOS is the power of LOS component and KF is Ricean K-factor in linear
scale. (n) denotes Dirac function. Generally, except the two strongest clusters, the
cluster power is uniformly allocated to every ray within the cluster. Assuming a
cluster contains Mn rays, the power of each ray within a cluster n is given by Pn/Mn.
Some weaker clusters, such as the cluster with power being 25 dB lower than the
maximum cluster power, can be removed.
3. Generating AOA and AOD
The Power Angular Spectrum (PAS) determines the spatial correlation proper-
ties of the channel. It reflects the power distribution in different directions and is
regarded as an important parameters of a MIMO channel model. At both the
transmitter and the receiver, 3GPP 3D channel model assumes that the composite
PAS in azimuth of all clusters is modeled as Wrapped Gaussian distribution, while
the composite PAS in zenith of all clusters is modeled as Laplacian distribution, as
shown in Fig. 3.13. It can be seen that the peak of Laplacian distribution is more
abrupt than that of Wrapped Gaussian distribution, which means that the energy is
more concentrated in zenith, while the energy is more dispersed in azimuth.
0.5
Wrapped Gaussian
0.45 Laplacian
0.4
0.35
probability density
0.3
0.25
0.2
0.15
0.1
0.05
0
-200 -150 -100 -50 0 50 100 150 200
angle(o)
Fig. 3.13 Laplacian distribution and wrapped Gaussian distribution

Table 3.19 Scaling factor for AOA and AOD generation

N 4 5 8 10 11 12 14 15 16 19 20
CAS 0.779 0.860 1.018 1.090 1.123 1.146 1.190 1.211 1.226 1.273 1.289
(a) Generating AOA and AOD

This part takes the generation of AOAs as an example. The generation of AODs
is similar to that of AOAs. The AOAs are determined by applying the inverse
Gaussian function (Ref. to Table 3.10) with cluster average power Pn and RMS
azimuth angle spread of arrival ASA, i.e.,
s

ASA Pn

b n, AOA ln : 3:76
0:7CAS maxPn
The constant CAS is a scaling factor related to the number of clusters N and given
by Table 3.19.
In the case of LOS conditions, additional scaling of angles is required to
compensate for the impact of LOS power on the angular spread, so that constant
LOS
CAS is substituted by Ricean K-factor dependent scaling constant CAS ,

LOS
CAS CAS 1:1035 0:028KF 0:002KF2 0:0001KF3 3:77
with KF being the Ricean K-factor in dB. The AOAs of clusters can be generated
for the case of LOS and NLOS respectively

b n, AOA Y n LOS, AOA
Xn , NLOS
n, AOA 3:78
b n, AOA Y n X1
Xn b 1, AOA Y 1 LOS, AOA , LOS
where random variable Xn is set to 1 or +1 with equal probability,

andYn is a
Gaussian RV with zero mean and variance of (ASA/7)2, Y n eN 0; ASA=72 . LOS,
AOA is the AOA of LOS direction determined in the stage of network layout. In the
case of LOS, the additional items are to enforce the first cluster to the LOS
direction. Finally, the AOA of ray n,m is calculated by adding offset angles to the
cluster AOA,
n, m, AOA n, AOA CASA m : 3:79
Where CASA is the RMS ASA of each cluster and given in Table 3.18. m is the
offset angles of m-th ray within a cluster and given in Table 3.20. With this method,
only rough angular resolution can be obtained, which does not meet the high angle
resolution requirements of massive MIMO or pencil beamforming. METIS
suggested a more precise offset angles generation method, i.e. sampling the Gauss-
ian function directly. For more details please refer to [185].
Table 3.20 Ray offset angles Ray path m Ray path offset angle m [ ]
within a cluster, given for 1
1,2 0.0447
RMS AS
3,4 0.1413
5,6 0.2492
7,8 0.3715
9,10 0.5129
11,12 0.6797
13,14 0.8844
15,16 1.1481
17,18 1.5195
19,20 2.1551
(b) Generating zenith angle of arrival (ZOA)

The ZOAs are determined by applying the inverse Laplacian function (Ref. to
Table 3.10) with cluster average power Pn and RMS zenith angle spread of arrival
ZSA

b ZSA Pn
n, ZoA ln 3:80
CES maxPn
The constant CES is a scaling factor related to the number of clusters N and given by
8
< 1:104, N 12
CES 1:184, N 19 3:81
:
1:178, N 20
In the case of LOS conditions, constant CES is substituted by Ricean K-factor

LOS
dependent scaling constant CES ,

LOS
CES CES 1:3086 0:0339K 0:0077K 2 0:0002K 3 , 3:82
The ZOAs of clusters can be generated for the case of LOS and NLOS
respectively

Xn b
n, ZOA Y n ZOA , NLOS
n, ZOA 3:83
Xn n, ZOA Y n X1 b
b n, ZOA Y 1 ZOA , LOS
where
random variable
Xn is set to 1 or +1 with equal probability and
2
Y n eN 0; ZSA=7 . ZOA is determined dependent on the location of UT. If the

UT locates indoors, ZOA 90 , otherwise ZOA LOS, ZOA, namely the ZOA of the
LOS direction determined at the stage of network layout. Finally, the ZOA of ray
n, m is calculated by adding offset angles to the cluster ZOA,
n, m, ZOA n, ZOA CZSAm 3:84
where CZSA is the RMS ZSA of each cluster and also given in Table 3.18. The
value of m is illustrated in Table 3.20. It should be noted that, the value of n,m,ZOA
is wrapped in [0, 360 ] in the calculation above, while usually n,m,ZOA2[0,180]. If
n,m,ZOA falls within [180 , 360 ], additional processing is applied to set n,m,ZOA
360 n,m,ZOA.
(c) Generating zenith angle of departure (ZOD)
The process of generating the cluster ZOD is similar to the process of generating
ZOA, in which only an extra offset is introduced, i.e.,
(
Xn b n, ZOD Y n offset, ZOD LOS, ZOD , NLOS
n, ZOD 3:85
Xn n, ZOD Y n X1 b
b n, ZOD Y 1 LOS, ZOD , LOS

where Xn is set to 1 or +1 with equal probability, Y n eN 0; ZSD=72 . ZSD is the
RMS zenith angle spread of departure. For its calculation please refer to the section
of LSPs generation. offset,ZOD is a modified factor in the case of NLOS conditions,
which is related to Tx-Rx distance and antenna height, and given in Table 3.17.
Finally, the ZOD of ray n,m is calculated by
3
n, m, ZOD n, ZOD 10ZSD m , 3:86
8
where ZSD is the mean of ZSD with lognormal distribution, which is also given in
Table 3.17. It is also dependent on the Tx-Rx distance and antenna height.
(d) Coupling of rays within a cluster for both azimuth and elevation
Couple randomly n,m,AOA to n,m,AOD within a cluster n (or within the
sub-clusters in the case of two strongest clusters). Couple randomly n,m,ZOA
to n,m,ZOD using the same procedure. Then couple randomly n,m,AOD to n,m,ZOD.
As a result, four angles of each ray within a cluster n are determined (n,m,AOD,
n,m,ZOD) and (n,m,AOA,n,m,ZOA).
4. Generating XPRs
Generating XPR for each ray within each cluster. XPR is lognormal distributed.
Draw XPR values as
n, m 10X=10 , 3:87
where X~N(, 2) is Gaussian distributed with mean and variance 2, which is

given in Table 3.18 for different simulation environment.
Till now, the LSPs and SSPs are generated for each ray m within each cluster n.
Next we will discuss the generation of channel coefficients.
3.6.4 Generation of Channel Coefficient
Firstly, draw random initial phase{ n, m , n, m , n

, m , n, m }for four polarization
pairs (, , , ) of ray n,m. The initial phases are uniformly disturbed in
[,). In the case of LOS conditions, draw a random initial phase LOS for both
and polarization pairs.
Next, by using the array radiation pattern at both ends for every transmission
link, the channel coefficients are to be generated for each cluster n and each receiver
and transmitter antenna element pairs u,s. Since WINNER and 3GPP channel
model seperate the twenty rays of the two strongest clusters into 3 sub-clusters,
the processing are carried out with two ways.
1. For the N2 weakest clusters, the channel coefficients are given by:
r M " #T
Pn X Frx, u, n, m, ZOA , n, m, AOA
u, s, n t;
H NLOS
M m1 Frx, u, n, m, ZOA , n, m, AOA
2
q 3 " #
ejn, m 1
n, m e
jn, m
Ftx, s, n, m, ZOD ;n, m, AOD
4
q 5
Ftx, s, n, m, ZOD ;n, m, AOD
1
n, m e
jn, m
ejn, m

1 T
e j20 rx, n, m drx, u vrx t tx, n, m dtx, s vtx t n, m
T
3:88
Where Ftx, s, and Ftx, s, are the vertical and horizontal polarimetric radiation
pattern of s respectively. Frx, u, and Frx, u, are the vertical and horizontal polari-
metric radiation pattern of u respectively. tx, n, m and rx, n, m are the unit direction
vectors of the ray n,m at transmitter and receiver respectively. dtx, s and drx, u are the
location vectors of s and u respectively. n,m is the cross polarisation power ratio in
linear scale, and 0 is the wavelength of the carrier. n, m is the delay of ray n,m. vtx
and vrx are the velocity vectors of the transmitter and receiver relative to the-first
bounce and last-bounce scatter respectively, so that this model is applicable for
D2D/V2V scenarios.
2. For the two strongest clusters, twenty rays of a cluster are spread in delay into
three sub-clusters, and the relative delays of each sub-cluster is as follows,
Table 3.21 Sub-clusters power and delay information (3GPP TR36.873 [170])
Sub path Ray path Average power ratio Delay offset
1 1,2,3,4,5,6,7,8,19,20 10/20 0 ns
2 9,10,11,12,17,18 6/20 5 ns
3 13,14,15,16 4/20 10 ns
n, 1 n 0 ns
n, 2 n 5 ns 3:89
n, 3 n 10 ns
Where n is the relative delay of the cluster to be seperated. Table 3.21 shows the
power allocation for the rays with different index in the three sub-clusters.
The strongest clusters separating into three sub-clusters may increase the delay
resolution of the channel model, and correspondly support larger bandwidth up to
100 MHz compared with 5 MHz bandwidth of SCM model.
In the case of LOS, an LOS component needs to be added and the power of each
ray is scaled down in terms of Ricean K-factor. The channel coefficients are given
by
r r
1 KF
, s, n t
H uLOS H u, s, n t n 1
NLOS
KF 1 KF 1
" #T
" #
Frx, u, LOS, ZOA ;LOS, AOA ejLOS 0 Ftx, s, LOS, ZOD ;LOS, AOD

Frx, u, LOS, ZOA ;LOS, AOA 0 ejLOS Ftx, s, LOS, ZOD ;LOS, AOD

1 T
e j20 rx, LoS drx, u vrx t tx, LoS dtx, s vtx t
T
3:90
In the applying of massive MIMO technology, the coupling between the antenna
elements will be nonnegligible, so it should be considered in modeling. According
to the method in [31], the channel coefficients are given by
~ u, s, n t 2r 11 R1=2 Z l Zr 1 HLOS=NLOS tR1=2

H t : 3:91
l u, s, n
where Rt is the real part of the impedance matrix of transmitting antennas Zt,
Rt Re {Zt}. Rl is the real part of load impedance Zl, Rl Re {Zl}. Zr is the
impedance matrix of receiving antennas. r11 is t real part of the self-impedance of
single antenna z11, r11 Re {z11}. Finally, apply path loss and shadow fading on the
generated channel coefficients to get
p
H u, s, n t PL SFH ~ u , s , n t 3:92
The delay values of obtained clusters (rays) are continuous, which are not sure to
coincide with the periodically sampling instants of digital communications system.
Therefore, for system simulation in time domain, it is necessary to carry out

interpolation to generate the channel coefficients, which can be used to convolve
with discrete transmitted signal to obtain discrete received signal.
3.6.4.1 Channel Simulation Examples
This subsection gives an example of channel simulation following the above

procedure. The simulation code use a framework provided by WINNER+ and
support four scenarios specified in 3GPP 3D MIMO channel model. Here, we
only show the simulation results for two scenarios, 3D-UMi and 3D-UMi-O2I.
Single BS and four UTs are used in the simulation to form four links. Two of the
UTs locate in the same building but on different floors (1st Floor and 7th Floor), and
the link conditions are NLOS. The other two UTs locate outdoors, with one link
being LOS transmission and another link being NLOS transmission. The BS adopts
a uniformly circular array composed of 8 antennas (UCA-8). All the 4 UTs are
equipped with 2-element linear array antenna (ULA-2). The two-dimensional
layout of the simulation network is shown in Fig. 3.14. Relevant simulation
parameters are listed in Table 3.22. We make some visual observations from one
channel realization produced by following the procedure above. It should be noted
that these observations are only valid for this channel realization. Different simu-
lation conditions and runs will get different observations.
First, we show the distribution of the rays parameters, including the delays, path
powers (gain), and AODs and ZODs at BS.
NETWORK LAYOUT
600
Link 1/2,
500 3D-UMi-O2I
400
Cells area, Y[m]
Link 3,UMi LoS

Link
300
4,
UMi
200
NLoS
100
0
260 280 300 320 340 360 380 400
Cells area, X[m]
Fig. 3.14 2D layout of simulation network

Table 3.22 Simulation parameters

Link No. 1 2 3 4
Scenario 3D-UMi-O2I 3D-Umi(LoS) 3D-Umi(NLoS)
Carrier frequency (GHz) 3.5
BS antenna type UCA-8(Diameter /2)
BS antenna height(m) 10
UT antenna type ULA-2(Interval /2)
UT antenna height(m) 1.5 19.5 1.5 1.5
UT floor 1 7 - -
UT indoor distance 22.5 6.1 - -
UT moving direction ( ) 90 90 195.5 83.6
UT moving speed (m/s) 0.1 0.1 10 10
BS-UT 2D distance (m) 66.2 68.1 217.7 431.9
LOS AoDs ( ) 61.1 60.0 -108.4 -98.6
LOS AoAs ( ) -118.9 -120.0 71.6 81.3
LOS ZoDs ( ) 97.3 82.1 92.1 91.1
LOS ZoAs ( ) 90.0 90.0 87.9 88.9
Straightline propagation delay (us) 0.22 0.24 0.73 1.44
Path loss (dB) -131.8 -118.7 -92.3 -130.6
KF(dB) N/A N/A 8.52 N/A
Figure 3.15 shows the spatial and temporal distribution of these rays. A ray is
marked with tag o, with marker size reflecting its power intensity and color (red,
blue, black, pink for link 1~4 respectively) indicating the link it belongs. There is
the large power difference among rays. In order to show the rays clearly, here we
carry out a simple process.In the LOS case, the rays powers are normalized by
scaling between 2 and 30, where in the NLOS case, the power range is between
2 and 10. The subfigure (a) presents a three-dimensional distribution of the rays.
The subfigure (b) gives the two-dimensional view of ZoDs-AoDs. The subfigure
(c) shows the two-dimensional view of the ZoDs-delays. The subfigure (d) shows
the two-dimensional view of AoDs-delays. Several points can be got obviously
from these figures. (1) Link 3 is LOS transmission, and the LOS components can be
identified clearly from the figure. (2) Link 1 and 2 have shorter communications
distance, so that they have minimum propagation delay. However their DSs are not
small. Link 1 has the largest DS. Link 4 has farthest communications distance, so its
propagation delay is largest while its DS is smallest. (3) All links have obvious
angular spread in the zenith dimension. The span of zenith angles of the four links
are approximately 70, 10, 30 and 50 degrees respectively. The zenith AS are
independent with the propagation conditions, such as the link 3 (LOS) has a midium
zenith AS. (4) Link 1 and 2 have almost the same average azimuth angle and
azimuth AS because the UTs are in the same building. (5) All links have almost the
same azimuth AS which is greater than the zenith AS.
(a) (b)
Delay
(c) (d)
Delay Delay
Fig. 3.15 Rays parameters distribution for one simulation: angles at BS, delays and powers.
(a) 3D view; (b) ZoDs-AoDs view; (c) ZoDs-delay view; (d) AoDs-delay view.
Figure 3.16 shows the channel impulse response of four different links observed
from the selected antenna pair, i.e. the 2nd antenna of BS array and the 1st antenna
of every UT. The path loss and shadow fading are also integrated into the channel
impulse response. In simulation, it is assumed that propagation conditions, delays
and angles of the multipath rays are kept constant. Only the effect of Doppler shift
caused by the movement of UTs is investigated. The figure can reveal: (1) All
channels have obvious time-varying characteristics. Because the UTs in Link 1 and
2 are at a low speed, the two channels change slowly in the observation time span of
0~70 s. The channels of Link 3 and 4 change violently even in a short observation
time span of 0~0.7 s. (2) Link 3 is in the LOS condition. LOS path is dominant in
CIR. (3) Link 4 has more stronger NLOS multipath components and smaller DS,
but the rays scatter in angle domain, which make the channel appear to be more
time-selective.
Fig. 3.16 Downlink time-varing channel from 2nd antenna at BS to 1st antenna at UT
3.7 Chapter Summary
Channel model is crucial for research and development for 5G and future mobile
communications. The 5G channel model needs to face up with a variety of chal-
lenges and requirments, including massive MIMO, mmWave, high mobility and
diversified application scenarios. This chapter begins with the requirement of 5G
channel model, and reviews five modeling methods: measurement-based GSCM,
regular shape-based GSCM, CSCM, extended SV and ray tracing. Especially, ray
tracing has caught more and more attention. With the combination of ray tracing
and other modeling methods, several new channel models are proposed. The
formation and verification of all the channel models cannot be separated from
channel sounding, which can provide real measurement data for channel modeling
and verify the presented channel model. This chapter introduces the measurement
methods and several channel sounders, as well as three types of 5G channel
measurement activities (massive MIMO, mmWave, and high mobility). High
resolution parameter estimation algorithms are used to extract path parameters
from the measured data. These post-processed data are further handled by statistical
analysis to obtain probability distribution functions (PDFs) and the corresponding
References 147
numerical values of statistical parameters. More than ten existing channel models
are introduced and compared according to the requirements of the 5G channel
model. At the lat section, we take 3GPP-3D channel model as an example to gives
the general process of stochastic channel generation, and finally give a simple
instantiation of channel simulation.
References
1. METIS D1.4 V1.0, METIS Channel Models, ICT-317669, METIS project, Feb. 2015.
https://www.metis2020.com/
2. 5G Vision and Requirements, IMT-2020 (5G) Promotion Group. May 2014.
3. 3GPP TR36.843, Study on LTE Device to Device Proximity Services, Radio Aspects, 3rd
Generation Partnership Project, V12.0.1, March 2014. http://www.3gpp.org
4. MiWEBA D5.1, Channel Modeling and Characterization, FP7-ICT-608637, V1.0,June
2014. http://www.miweba.eu
5. METIS D5.1, Intermediate description of the spectrum needs and usage principles, V1.0,
ICT-317669, METIS project, August 2013. https://www.metis2020.com
6. 3GPP TR 38.900, Study on channel model for frequency spectrum above 6 GHz, v14.0.0
2016
7. W. Roh, Ji-yun seol, Jeongho park, et al. Millimeter-wave beamforming as an enabling
technology for 5G cellular communications: theoretical feasibility and prototype results.
IEEE Communications Magazine, vol. 52, no. 2, pp.106-113, 2014.
8. Su-Khiong Yong, IEEE P802.15 Wireless Personal Area Networks - TG3c Channel Model-
ing Sub-committee Final Report, IEEE 15-07-0584-01-003c,March 2007
9. A. Maltsev, V. Erceg, E. Perahia, C. Hansen, R. Maslennikov, A. Lomayev, A. Sevastyanov,
A. Khoryaev, G. Morozov, M. Jacob, S. Priebe, T. Krner, S. Kato, H. Sawada, K. Sato and
H. Harada, Channel Models for 60 GHz WLAN Systems, IEEE 802.11ad 09/0334r8, 2010.
10. A. Maltsev, A. Pudeyev, Y. Gagiev, et.al., Channel Models for IEEE 802.11ay, IEEE
802.11-15/1150r9, 2016
11. H2020-ICT-671650-mmMAGIC/D2.1. Measurement Campaigns and Initial Channel
Models for Preferred Suitable Frequency Ranges, 2016. https://bscw.5g-mmmagic.eu/
pub/bscw.cgi/d94832/ mmMAGIC_D2-1.pdf
12. 5GCM White Paper, 5G Channel Model for bands up to 100 GHz, v2.0, 2016. http://www.
5gworkshops.com/
13. WINNER II D1.1.2, Channel models, IST-4-027756 V1.2, Sept. 2007. Online] Available:
http://www. istwinner.org/deliverables.html
14. ITU-R M.2135-1, Guidelines for evaluation of radio interface technologies for
IMT-Advanced, International Telecommunication Union (ITU), Geneva, Switzerland,
Technical Report, December 2009.
15. K. Zheng, S. Ou, and X. Yin, Massive MIMO channel models:A survey, International
Journal of Antennas and Propagation, vol. 2014, 2014
16. C. A. Balanis, Antenna Thory:Analysis and Design, John Wiley & Sons, Hoboken, NJ, USA,
2012.
17. S. Wu, C.-X. Wang, H. Aggoune, M. M. Alwakeel, and Y. He, A non-stationary 3D
wideband twin-cluster model for 5G massive MIMO channels, IEEE J. Sel. Areas Commun.,
vol. 32, no. 6, pp. 1207-1218, June 2014.
18. X. Gao, F. Tufvesson and O. Edfors, Massive MIMO channels:Measurements and models.
in Proc. of 2013 Asilomar Conference on Signals, Systems and Computers, 2013
19. S. Jaeckel, L. Raschkowski, K. B orner and L. Thiele, QuaDRiGa: A 3.D Multicell Channel
Model with Time Evolution for Enabling Virtual Field Trials, IEEE Transactions on
Antennas Propagation, 2014. http://quadriga-channel-model.de
20. T. S. Rappaport, G. R. Maccartney, M. K. Samimi, et al. Wideband Millimeter-Wave
Propagation Measurements and Channel Models for Future Wireless Communication System
Design.IEEE Transactions on Communications, vol.63, no. 9, pp.3029-3056,2015.
21. G. Calcev, D. Chizhik, B. Goeransson, S. Howard, H. Huang, A. Kogiantis, A. F. Molisch,
A. L. Moustakas, D. Reed and H. Xu, A Wideband Spatial Channel Model for System-Wide
Simulations, IEEE Trans. Vehicular Techn., March 2007.
22. Matthias Patzold, Mobile Radio Channels, 2nd Edition, John Wiley & Sons, Chichester, UK,
2012
23. Y. Yuan, C.-X. Wang, Y. He, M. M. Alwakeel, and H. Aggoune, Novel 3D wideband
non-stationary geometry-based stochastic models for non-isotropic MIMO vehicle-to-vehicle
channels, IEEE Transactions on Wireless Communications, vol 13, no. 1, pp. 298-309, 2014.
24. A. Ghazal, C.-X. Wang, B. Ai, D. Yuan, and H. Haas, A non-stationary wideband MIMO
channel model for high-mobility intelligent transportation systems, IEEE Trans. Intell.
Transp. Syst., vol. 16, no. 2, pp. 885-897, Apr. 2015.
25. A. M. Sayeed, Deconstructing multiantenna fading channels, IEEE Trans. S ignal Process.,
vol. 50, no. 10, pp. 25632579, Oct. 2002.
26. W. Weichselberger, M. Herdin, H. Ozcelik, and E. Bonek, A stochastic MIMO channel
model with joint correlation of both link ends, IEEE Transactions on Wireless Communi-
cations, vol. 5, no. 1, pp. 90-100, Jan. 2006.
27. J. Hoydis, S. ten Brink, and M. Debbah, Massive MIMO in the UL/DL of cellular networks:
how many antennas do we need?IEEE Journal on Selected Areas in Communications, vol.
31, no. 2, pp. 160171, 2013.
28. C. Masouros, M. Sellathurai, and T. Ratnarajah, Large-scale MIMO transmitters in fied
physical spaces:the effct of transmit correlation and mutual coupling, IEEE Transactions on
Communications, vol. 61, no. 7, pp. 27942804, 2013
29. B. Clerckx, C. Craeye, D. Vanhoenacker-Janvier, and C. Oestges, Impact of antenna
coupling on 2 2 MIMO communications, IEEE Transactions on Vehicular Technology,
vol. 56, no. 3, pp. 10091018, 2007.
30. C. Masouros, J. Chen, K. Tong, M. Sellathurai, and T. Ratnarajah, Towards massive-MIMO
transmitters:on the effcts of deploying increasing antennas in fied physical space, in Pro-
ceedings of the Future Network and Mobile Summit, pp. 110, 2013.
31. Y. Fei, Y. Fan, B. K. Lau, and J. S. Thompson, Optimal single-port matching impedance for
capacity maximization in compact MIMO arrays, IEEE Trans. Antennas Propagat., vol.
56, no. 11, pp. 35663575, Nov. 2008.
32. R. Srinivasan, J. Zhuang, et. al., IEEE 802.16m Evaluation Methodology Document (EMD)
, IEEE 802.16m-08/004r2,July 2008.
33. A. Saleh and R. Valenzuela, A Statistical Model for Indoor Multipath Propagation,IEEE
J. Select. Areas Commun., Vol. SAC-5, No. 2, pp. 128-137, Feb. 1987.
34. Wu Zhizhong, Mobile Communications Radio Waves Propagation, Beijing: PeopleS Posts
And Telecommunications Publishing House, 2002. (in Chinese)
35. Nan Wang, Modern Uniform Geometrical Theory of diffraction, Xian: Xian Electronic
Sience &Technology University Press, 2010. (in Chinese)
36. J. Pascual-Garca, M.-T. Martinez-Ingles, J. M. Molina Garcia-Pardo, J. V. Rodrguez, and
L. JuanLlacer, Using tuned diffuse scattering parameters in ray tracing channel modeling,
in Proc. 9th European Conf. Antennas and Propagation (EuCAP 2015), Lisbon, Portugal,
pp. 14.
37. Pekka Kyosti and Tommi Jamsa, Complexity Comparison of MIMO Channel Modelling
Methods, ISWCS07, Trondheim, Norway, October 2007
38. J. Chen, X. Yin, L. Tian and M. D. Kim,"Millimeter-Wave Channel Modeling Based on A
Unified Propagation Graph Theory." IEEE Communications Letters 21(2): 246-249, 2017.
References 149
39. Elektrobit Ltd., Propsound - multi-dimensional radio channel sounder, System specifica-
tions document, Concept and specifications, Technical report, 2004.
40. Channelsounder.de, MEDAV GmbH, [Online]. Available:http://www.channelsounder.de/
41. V. Kolmonen, J. Kivinen, L. Vuokko, and P. Vainikainen, 5.3.GHz MIMO radio channel
sounder, IEEE Trans. Instrum. Meas., vol. 55, no. 4, pp. 12631269, Aug. 2006.
42. K. Kitao, K. Saito, Y. Okano, T. Imai and J. Hagiwara, Basic study on spatio-temporal
dynamic channel properties based on channel sounder measurements,Asia Pacific Micro-
wave Conference (APMC), pp. 1064-1067, 2009.
43. W. Newhall, T. Rappaport, and D. Sweeney, A spread spectrum sliding correlator system for
propagation measurements, RF Design, pp. 4054, Apr. 1996.
44. S. Salous, R. Lewenz, I. Hawkins, N. Razavi-Ghods, and M. Abdallah,Parallel receiver
channel sounder for spatial and MIMO characterization of the mobile radio channel, in Proc.
Inst. Elect. Eng. Commun., vol. 152, no. 6, pp. 912918, Dec. 2005.
45. Y. Konishi, M. Kim, M. Ghoraishi, J. Takada, S. Suyama, and H. Suzuki, Channel sounding
technique using MIMO software radio architecture, in Proc. 5th EuCAP, Rome, Italy, Apr.
2011, pp. 25462550.
46. K. Minseok, J. Takada and Y. Konishi, Novel Scalable MIMO Channel Sounding Technique
and Measurement Accuracy Evaluation With Transceiver Impairments,IEEE Transactions
on Instrumentation and Measurement, 61(12): 3185-3197, 2012.
47. H. Sawada, Y. Shoji, H. Ogawa, NICT propagation data, IEEE 802.15-06/0012-01-003c,
Jan. 2006
48. A. Maltsev, R. Maslennikov, A. Sevastyanov, A. Khoryaev, and A. Lomayev, Experimental
investigations of 60 GHz wireless systems in office environment, IEEE J. Sel. Areas
Commun., vol. 27, no. 8, pp.1488-1499, Oct. 2009.
49. C. Gustafson, K. Haneda, S. Wyne and F. Tufvesson, On mm-wave multi-path clustering
and channelmodeling, IEEE Trans. Antennas Propag., vol. 62, no. 3, pp. 1445 -1455, 2014.
50. J. I. Tamir, T. S. Rappaport, Y. C. Eldar, and A. Aziz, Analog compressed sensing for rf
propagation channel sounding, in 2012 I.E. International Conference on Acoustics, Speech
and Signal Processing (ICASSP), pp. 5317-5320, March 2012.
51. Zhu Jin, Wang Haiming and Hong Wei, Large-Scale Fading Characteristics of Indoor
Channel at 45-GHz Band, IEEE Antennas and Wireless Propagation Letters, 14:735-738,
2015
52. X. Gao, O. Edfors, F. Rusek, and F. Tufvesson, Linear precoding performance in measured
very-large MIMO channels, in Proceedings of the 74th IEEE Vehicular Technology Con-
ference (VTC 11), pp.15, Budapest, Hungary, Sept. 2011.
53. S. Payami and F. Tufvesson, Channel measurements and analysis for very large array
systems at 2.6 GHz, in Proceedings of the 6th European Conference on Antennas and
Propagation (EuCAP 12), pp. 433437, Prague, Czech Republic, March 2012.
54. X. Gao, F. Tufvesson, O. Edfors, and F. Rusek, Measured propagation characteristics for
very-large MIMO at 2.6 GHz, in Proceedings of the 46th IEEE Asilomar Conference on
Signals, Systems and Computers (ASILOMAR 12), pp. 295299, Pacific Grove, Calif, USA,
November 2012.
55. X. Gao, O. Edfors, F. Rusek and F. Tufvesson, Massive MIMO performance evaluation
based on measured propagation data,IEEE Transactions onWireless Communications, PP
(99): 1-1, 2015.
56. J. Flordelis, X. Gao, G. Dahman, F. Rusek, O. Edfors and F. Tufvesson, Spatial Separation
of Closely-Spaced Users in Measured Massive Multi-User MIMO Channels, in Proc. IEEE
International Conference on Communications (ICC), London, June 2015.
57. J. Hoydis, C. Hoek, T. Wild and S. Ten Brink, Channel measurements for large antenna
arrays. 2012 International Symposium on Wireless Communication Systems (ISWCS) , 2012
58. A. O. Martinez, E. De Carvalho and J. O. Nielsen, Towards very large aperture massive
MIMO:A measurement based study. Globecom Workshops (GC Wkshps), 2014
59. L. Liu, C. Tao, D. Matolak, Y. Lu, B. Ai, H. Chen, Stationarity Investigation of a LOS
Massive MIMO Channel in Stadium Scenarios, in Proc. of IEEE 82th Vehicular Technology
Conference (VTC Fall), 2015
60. D. Fei, R. He, B. Ai, B. Zhang, K. Guan, and Z. Zhong,Massive MIMO Channel Measure-
ments and Analysis at 3.33 GHz, ChinaCom, 2015
61. IEEE P802.11p-2010:Part 11:Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications:Amendment 6:Wireless Access in Vehicular Environments, Jul.
15, 2010, DOI:10.1109/IEEESTD.2010. 5514475.
62. A. Roivainen, P. Jayasinghe, J. Meinilau, V. Hovinen and M. Latva-Aho, Vehicle-to-vehicle
radio channel characterization in urban environment at 2.3 GHz and 5.25 GHz,In Proc. of
IEEE 25th Annual International Symposium on Personal, Indoor, and Mobile Radio Com-
munication (PIMRC), 2014
63. J. Karedal, F. Tufvesson, N. Czink, A. Paier, C. Dumard, T. Zemen, C. F. Mecklenbrauker
and A. F. Molisch, A geometry-based stochastic MIMO model for vehicle-to-vehicle
communications,IEEE Transactions on Wireless Communications, 8(7): 3646-3657, 2009.
64. T. Abbas, J. Karedal, F. Tufvesson, A. Paier, L. Bernado and A. F. Molisch, Directional
Analysis of Vehicle-to-Vehicle Propagation Channels. inProc. of IEEE 73rd Vehicular
Technology Conference (VTC Spring),2011
65. L. Bernado, T. Zemen, F. Tufvesson, A. F. Molisch and C. F. Mecklenbrauker, Delay and
Doppler Spreads of Nonstationary Vehicular Channels for Safety-Relevant Scenarios,IEEE
Transactions on Vehicular Technology, 63(1): 82-93, 2014.
66. L. Bernado, T. Zemen, F. Tufvesson, A. F. Molisch and C. F. Mecklenbrauker, Time- and
Frequency-Varying K-Factor of Non-Stationary Vehicular Channels for Safety-Relevant
Scenarios, IEEE Transactions on Intelligent Transportation Systems, 16(2): 1007-1017,
2015.
67. O. Renaudin, V. M. Kolmonen, P. Vainikainen and C. Oestges, Wideband Measurement-
Based Modeling of Inter-Vehicle Channels in the 5-GHz Band,IEEE Transactions on
Vehicular Technology, 62(8): 3531-3540, 2013.
68. He Ruisi, O. Renaudin, V. M. Kolmonen, K. Haneda, Zhong Zhangdui, Ai Bo and C. Oestges,
Characterization of Quasi-Stationarity Regions for Vehicle-to-Vehicle Radio
Channels,IEEE Transactions on Antennas and Propagation, 63(5): 2237-2251, 2015.
69. M. Walter, U. C. Fiebig and A. Zajic, Experimental Verification of the Non-Stationary
Statistical Model for V2V Scatter Channels, in Proc. of IEEE 80th Vehicular Technology
Conference (VTC Fall), 2014
70. M. Boban, J. Barros and O. K. Tonguz, Geometry-Based Vehicle-to-Vehicle Channel
Modeling for Large-Scale Simulation,IEEE Transactions on Vehicular Technology, 63(9):
4146-4164, 2014.
71. He Ruisi, A. F. Molisch, F. Tufvesson, Zhong Zhangdui, Ai Bo and Zhang Tingting,
Vehicle-to-Vehicle Propagation Models With Large Vehicle Obstructions,IEEE Trans-
actions on Intelligent Transportation Systems, 15(5): 2237-2248, 2014.
72. K. Amiri,Y. Sun, P. Murphy,C.Hunter, J.R.Cavallaro, andA. Sabharwal, WARP, a unified
wireless network testbed for education and research, in Proc. IEEE MSE, 2007, pp. 5354.
73. B. Ai, X. Cheng, T. Krner, Z. D. Zhong, K. Guan, R. S. He, L. Xiong, D. W. Matolak, D. G.
Michelson, C. Briso-Rodriguez, Challenges Toward Wireless Communications for High-
Speed Railway. IEEE Transactions on Intelligent Transportation Systems , vol. 15, no. 5:
2143-2158.2014
74. R.S. He, Z. Zhong, Bo Ai, G. Wang, J. Ding, A.F. Molisch , Measurements and Analysis of
Propagation Channels in High-Speed Railway Viaducts,IEEE Transactions on Wireless
Communications, vol.12, no.2, pp.794,805, February 2013
75. Tao Zhou; Cheng Tao; Liu Liu; Zhenhui Tan, Ricean K-Factor Measurements and Analysis
for Wideband Radio Channels in High-Speed Railway U-Shape Cutting
Scenarios,Vehicular Technology Conference (VTC Spring), 2014 I.E. 79th, pp.1,5, 18-21
May 2014
References 151
76. Guan Ke, Zhong Zhangdui, Ai Bo and T. Kurner, Propagation Measurements and Analysis
for Train Stations of High-Speed Railway at 930 MHz,IEEE Transactions on Vehicular
Technology, 63(8): 3499-3516, 2014.
77. Guan Ke, Zhong Zhangdui, Ai Bo and T. Kurner, Propagation Measurements and Modeling
of Crossing Bridges on High-Speed Railway at 930 MHz,IEEE Transactions on Vehicular
Technology, 63(2): 502-517, 2014.
78. Guan Ke, Zhong Zhangdui, J. I. Alonso and C. Briso-Rodriguez, Measurement of Distrib-
uted Antenna Systems at 2.4 GHz in a Realistic Subway Tunnel Environment,IEEE Trans-
actions on Vehicular Technology, 61(2): 834-837, 2012.
79. R. He, Z. Zhong, B. Ai, K. Guan, B. Chen, J. I. AIonso, and C. Briso,Propagation channel
measurements and analysis at 2.4 GHz in subway tunnels, IET Microwaves, Antennas &
Propagation, vol. 7, no. 11, pp. 934941, 2013
80. K. Guan, B. Ai, Z. Zhong, C.F. Lopez, L. Zhang, C. Briso-Rodriguez, A. Hrovat, B. Zhang,
R. He, T. Tang, Measurements and Analysis of Large-Scale Fading Characteristics in
Curved Subway Tunnels at 920 MHz, 2400 MHz, and 5705 MHz,Intelligent Transportation
Systems, IEEE Transactions on , vol.PP, no.99, pp.1,13, 2015
81. J. Li, Y. Zhao, J. Zhang, R. Jiang, C. Tao, Z. Tan, Radio channel measurements and analysis
at 2.4/5GHz in subway tunnels,Communications, China, vol.12, no.1, pp.36,45, Jan. 2015
82. WINNER II D1.1.2 WINNER II Channel Models Part II Radio Channel Measurement and
Analysis Results, IST-4-027756 V1.0, Sept. 2007. http://www. istwinner.org/ deliverables.
html
83. Qian Wang, Chunxiu Xu, Min Zhao and Deshui Yu, Results and analysis for a novel channel
measurement applied in LTE-R at 2.6 GHz,in Proc. of Wireless Communications and
Networking Conference (WCNC), 2014 IEEE,2014
84. E. J. Violette, R. H. Espeland, R. O. DeBolt and F. K. Schwering, Millimeterwave propa-
gation at street level in an urban environment,IEEE Transactions on Geoscience and Remote
Sensing, vol. 26, pp. 368-380, 1988.
85. H. J. Thomas, R. S. Cole and G. L. Siqueira, An experimental study of the propagation of
55 GHz millimeter waves in an urban mobile radio environment,EEE Transactions on
Vehicular Technology, vol. 43, pp. 140- 146, 1994.
86. L. M. Correia, J. J. Reis and P. O. Frances, Analysis of the average power to distance decay
rate at the 60 GHz band, in IEEE in Vehicular Technology Conference, 1997.
87. A. M. Hammoudeh, M. G. Sanchez and E. Grindrod, Experimental analysis of propagation
at 62 GHz in suburban mobile radio microcells, IEEE Transactions on Vehicular Technol-
ogy, vol. 48, pp. 576-588, 1999.
88. K. Sato, H. Kozima, H. Masuzawa, T. Manabe, T. Ihara, Y. Kasashima, K. Yamaki, Mea-
surements of reflection characteristics and refractive indices of interior construction materials
in millimeter-wave bands,in Proc. of IEEE 45th Vehicular Technology Conference(VTC
Spring 1995), vol.1, pp.449,453, 25-28 Jul 1995
89. Thomas Zwick, Troy J. Beukema, and Haewoon Nam, Wideband Channel Sounder With
Measurements and Model for the 60 GHz Indoor Radio Channel,IEEE Trans. On Vehicular
Technology, vol. 54, no. 4, pp.1266-1277, JULY 2005.
90. X. Hao, T. S. Rappaport, R. J. Boyle and J. H. Schaffner, 38-GHz wide-band point-to-
multipoint measurements under different weather conditions,IEEE Communications Letters,
vol. 4, pp. 7-8, 2000.
91. H. Xu, V. Kukshya, and T. S. Rappaport, Spatial and Temporal Characteristics of 60 GHz
Indoor Channels, IEEE J. Sel. Areas Commun., vol. 20, no. 3, pp. 620630, Apr. 2002.
92. T. S. Rappaport, E. Ben-Dor, J. N. Murdock and Q. Yijun, 38 GHz and 60 GHz angle-
dependent propagation for cellular & peer-to-peer wireless communications, in IEEE
International Conference on Communications (ICC), 2012.
93. E. Ben-Dor, T. S. Rappaport, Q. Yijun and S. J. Lauffenburger, MillimeterWave 60 GHz
Outdoor and Vehicle AOA Propagation Measurements Using a Broadband Channel
Sounder, in IEEE Global Telecommunications Conference (GLOBECOM), 2011.
94. T. S. Rappaport, Q. Yijun, J. I. Tamir, J. N. Murdock and E. Ben-Dor, Cellular broadband

millimeter wave propagation and angle of arrival for adaptive beam steering systems, in
IEEE Radio and Wireless Symposium (RWS), 2012.
95. T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Q. Yijun and J. I. Tamir,
Broadband Millimeter-Wave Propagation Measurements and Models Using Adaptive-
Beam Antennas for Outdoor Urban Cellular Communications,IEEE Transactions on Anten-
nas and Propagation, vol. 61, pp. 1850-1859, 2013.
96. J. N. Murdock, E. Ben-Dor, Q. Yijun, J. I. Tamir and T. S. Rappaport, A 38 GHz cellular
outage study for an urban outdoor campus environment, in IEEE Wireless Communications
and Networking Conference (WCNC), 2012.
97. Y. Azar, G. N. Wong, K. Wang, R. Mayzus, J. K. Schulz, Z. Hang, F. Gutierrez, D.Hwang,
T. S. Rappaport, 28 GHz propagation measurements for outdoor cellularcommunications
using steerable beam antennas in New York city, in IEEEInternational Conference on
Communications (ICC), 2013.
98. T. S. Rappaport, S. Shu, Multi-beam antenna combining for 28 GHz Cellular
linkImprovement in urban environments, in IEEE Global Telecommunication Conference
(Globecom)., Atlanta, 2013.
99. M.R. Akdeniz, Y. Liu, M.K. Samimi, S. Sun, S. Rangan, T.S. Rappaport, E. Erkip, Milli-
meter Wave Channel Modeling and Cellular Capacity Evaluation, IEEE J. Sel. Areas on
Comm., Aug. 2014
100. M. K. Samimi and T. S. Rappaport, Ultra-wideband statistical channel model for non line of
sight millimeter-wave urban channels. Global Communications Conference (GLOBECOM),
2014 IEEE,2014
101. T. S. Rappaport, R. W.Heath, Jr., R. C.Daniels, and J. N.Murdock, MillimeterWave
WirelessCommunications.Pearson/Prentice Hall, 2015.
102. T. A. Thomas, Nguyen Huan Cong, G. R. Maccartney and T. S. Rappaport, 3D mmWave
Channel Model Proposal. Vehicular Technology Conference (VTC Fall), 2014 I.E. 80th,2014
103. M. W. M. Peter, W. K. M. Raceala-Motoc, R. Felbecker, M. Jacob, S. Priebe and T.Krner,
Analyzing human body shadowing at 60 GHz:Systematic wideband MIMOmeasurements
and modeling approaches, in Proc. of IEEE 6th European Conference on Antennas and
Propagation, 2012.
104. A.P. Garcia, W. Kotterman, R.S. Thoma, U. Trautwein , D. Brckner,W. Wirnitzer,
J. Kunisch, 60 GHz in-cabin real-time channel sounding,3rd International Workshop on
Broadband MIMO Channel Measurement and Modeling - IWonCMM 2009, Xian, China,
August 25, 2009.
105. M. Peter, W. Keusgen, A. Kortke, M. Schirrmacher, Measurement and Analysis of the
60 GHz In-Vehicular Broadband Radio Channel,Proc. of IEEE Vehicular Technology
Conference, 2007 (VTC-2007), pp. 834 - 838
106. M. Kyro, S. Ranvier, V. Kolmonen, K. Haneda and P. Vainikainen, Long range wideband
channel measurements at 81-86 GHz frequency range, in European Conference Antennas
and Propagation (EuCAP), 2010.
107. M. Kyro, V. Kolmonen and P. Vainikainen, Experimental Propagation Channel Character-
ization of mm-Wave Radio Links in Urban Scenarios,IEEE Antennas and Wireless Propa-
gation Letters, vol. 11, pp. 865-868, 2012.
108. A. V. Raisanen, J. Ala-Laurinaho, K. Haneda, J. Jarvelainen, A. Karttunen, M. Kyro,
V. Semkin, A. Lamminen and J. Saily, Studies on E-band antennas and propagation, in
Antennas and Propagation Conference (LAPC)., Loughborough, 2013.
109. Kim Minseok, Y. Konishi,Yu-yuan Chang, et al., Large Scale Parameters and Double-
Directional Characterization of Indoor Wideband Radio Multipath Channels at 11 GHz.
IEEE Trans. Antennas Propagation, vol. 62, no. 1, pp.430-441, 2014.
110. WANG H M, HONG W, et al. Channel Measurement for IEEE 802.11aj (45 GHz), IEEE
802.11-12/1361r3, 2013.
References 153
111. X. Wu, C.X. Wang, J. Sun, J. Huang, R. Feng,Y. Yang, X. Ge, 60 GHz Millimeter-Wave
Indoor Channel Measurements and Modeling for 5G Systems, IEEE Trans. Antennas Propag,
2017, vol. 65, no. 4, pp. 1912-1924, 2017.
112. Mingyang Lei, Jianhua Zhang, Tian Lei, Detao Du, 28-GHz Indoor Channel Measurements
and Analysis of Propagation Characteristics,IEEE 25th International Symposium on Per-
sonal, Indoor and Mobile Radio Communications (PIMRC 2014),2014
113. Zhang Nan, Yin Xuefeng, S. X. Lu, Du Mingde and Cai Xuesong, Measurement-based
angular characterization for 72 GHz propagation channels in indoor environments,
Globecom Workshops (GC Workshps), 2014
114. YIN Xuefeng, LING Cen and KIM Myung-Don, Experimental Multipath-Cluster Charac-
teristics of 28-GHz Propagation Channel. IEEE Access, 3, 2015. 3138-3150
115. X. Zhao, Q. Wang, S. Li, S. Geng, M. Wang, S. Sun, and Z. Wen, Attenuation by human
bodies at 26 and 39.5 GHz millimeter wave bands, IEEE Antennas Wireless Propag. Lett.,
vol. 19, pp. 1229 1232, Nov. 2016.
116. R. O. Schmidt, Multiple emitter location and signal parameter estimation, IEEE Trans.
Antennas Propagat., vol. AP-34, pp. 276280, Mar.1986.
117. R. Roy and T. Kailath, ESPRITEstimation of signal parameters via rotational invariance
techniques, IEEE Trans. Acoust., Speech, Signal Processing, vol. 37, pp. 984995, July
1989.
118. A. van der Veen, M. Vanderveen, and A. Paulraj, Joint angle and delay estimation using
shift-invariance properties, IEEE Signal Processing Lett., vol. 4, pp. 142145, May 1997.
119. M. Haardt and J. Nossek, Unitary ESPRIT: How to obtain increased estimation accuracy
with a reduced computational burden, IEEE Trans. Signal Processing, vol. SP-43,
pp. 12321242, May 1995.
120. M. Zoltowski, M. Haardt, and C. Mathews, Closed-form 2-D angle estimation with rectan-
gular arrays in element space or beamspace via unitary ESPRIT, IEEE Trans. Signal
Processing, vol. SP-44, pp. 316328, Feb. 1996.
121. J. Fuhl, J.-P. Rossi, and E. Bonek, High-resolution 3.D direction-of-arrival determination for
urban mobile radio, IEEE Trans. Antennas Propagat., vol. AP-45, pp. 672682, Apr. 1997.
122. G. McLachlan and T. Krishnan, The EM Algorithm and Extensions. Probability and Statis-
tics. New York:Wiley, 1996
123. A. P. Dempster, N. M. Laird, and D. B. Rubin, Maximum likelihood from incomplete data
via the EM algorithm, J. Royal Statist. Soc., Ser. B, vol. 39, no. 1, pp. 138, 1977.
124. J. A. Fessler and A. O. Hero, Space-alternating generalized expectation- maximization
algorithm, IEEE Transactions on Signal Processing, Vol 42, no 10, pp 2664-2677, 1994.
125. B. H. Fleury, D. Dahlhaus, R. Heddergott, and M. Tschudin, Wideband angle of arrival
estimation using the SAGE algorithm, in Proc. of the IEEE Fourth Int. Symp. on Spread
Spectrum Techniques and Applications (ISSSTA 96), Mainz, Germany, pp. 79-85, Sept.
1996.
126. B. H. Fleury, M. Tschudin, R. Heddergott, D. Dahlhaus, and K. I. Pedersen, Channel
parameter estimation in mobile radio environments using the SAGE algorithm, IEEE
J. Sel. Areas Commun., vol. 17, no. 3, pp. 434-450, Mar. 1999.
127. M. Tschudin, R. Heddergott and P. Truffer, Validation of a high resolution measurement
technique for estimating the parameters of impinging waves in indoor environments, in
Proc. of The Ninth IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications(PIMRC 1998),1998
128. B. H. Fleury, Yin Xuefeng, K. G. Rohbrandt, P. Jourdan and A. Stucki, Performance of a
high-resolution scheme for joint estimation of delay and bidirection dispersion in the radio
channel,IEEE 55th Vehicular Technology Conference(VTC Spring 2002), 2002
129. Yin Xuefeng, B. H. Fleury, P. Jourdan and A. Stucki, Polarization estimation of individual
propagation paths using the SAGE algorithm,14th IEEE Proceedings on Personal, Indoor
and Mobile Radio Communications(PIMRC 2003), 2003
130. C. C. Chong, D. I. Laurenson, C. M. Tan, S. Mclaughlin, M. A. Beach and A. R. Nix, Joint

detection-estimation of directional channel parameters using the 2-D frequency domain
SAGE algorithm with serial interference cancellation,IEEE International Conference
onCommunications(ICC 2002), 2002
131. D. Shutin, B.H. Fleury, Sparse Variational Bayesian SAGE Algorithm With Application to
the Estimation of Multipath Wireless Channels, IEEE Transactions on Signal Processing,
vol 59, no 8, pp. 3609 3623, Aug. 2011
132. A. Richter, Estimation of Radio Channel Parameters:Models and Algorithms, Ph. D.
dissertation, 2005,nel parameters:Models and algorithms Ph.D. dissertation, Technischen
Universitat Ilmenau, Ilmenua, Germany, May 2005, ISBN:978-3.938843.02-4 [Online].
Available:www.db-thueringen.de.
133. Andreas Richter, Reiner S. Thoma, Parametric Modeling and Estimation of Distributed
Diffuse Scattering Components of Radio Channels, COST273, TD(03)198
134. F. Quitin, C. Oestges, F. Horlin, and P. De Doncker, A Polarized Clustered Channel Model
for Indoor Multiantenna Systems at 3.6 GHz,IEEE Transactions on Vehicular Technology,
vol. 59, no. 8, pp. 3685-3693, Oct. 2010.
135. Zhang Yan, Hu Xinwei, Zhou Shidong ,Wang Jing, Near Field Channel Parameter Estimation
Method Based on the SAGE, Journal of System Simulation, vol. 23, 9 September 2011.
(in Chinese)
136. J. Chen, S. Wang and X. Yin, A Spherical-Wavefront-Based Scatterer Localization Algo-
rithm Using Large-Scale Antenna Arrays. IEEE Communications Letters 20(9): 1796-1799,
2016.
137. F. Quitin, C. Oestges, F. Horlin, and P. De Doncker, Diffuse multipath component charac-
terization for indoor mimo channels, in Antennas and Propagation (EuCAP), 2010
Proceedingsof the Fourth European Conference on, Apr. 2010, pp. 15.
138. Tanghe E., Gaillot D. P., Lienard M. and et al. Experimental Analysis of Dense Multipath
Components in an Industrial Environment. IEEE Transactions on Antennas and Propagation,
62(7): 37973805, 2014.
139. J. Poutanen, F. Tufvesson, K. Haneda, L. Liu, C. Oestges and P. Vainikainen, Adding Dense
Multipath Components to Geometry-Based MIMO Channel Models, [Online] Available:
http://lup.lub.lu.se/record/2218357/file/2218371.pdf
140. J. Salmi, A. Richter and V. Koivunen, Detection and Tracking of MIMO Propagation Path
Parameters Using State-Space Approach, IEEE Transactions on Signal Processing, 57(4):
15381550, 2009.
141. T. Jost, Wei Wang; U. Fiebig, F. Perez-Fontan, Detection and Tracking of Mobile Propa-
gation Channel Paths, IEEE Transactions on Antennas and Propagation, vol 60, no 10, pp.
4875 4883, Oct. 2012
142. Xuefeng Yin, G. Steinbock, G.E. Kirkelund, T. Pedersen, P. Blattnig, A. Jaquier,
B.H. Fleury,Tracking of Time-Variant Radio Propagation Paths Using Particle
Filtering,IEEE ICC 08, pp. 920-924, May 2008.
143. K. Saito, K. Kitao, T. Imai and Y. Okumura, Dynamic MIMO channel modeling in urban
environment using particle filtering. 2013 7th European Conference on Antennas and Prop-
agation (EuCAP), 2013
144. J.C. Bezdek, Pattern recognition with fuzzy objective function algorithms, New York, 1981
145. N. Czink, P. Cera, J. Salo, E. Bonek, J.-P. Nuutinen, J. Ylitalo, A framework for automatic
clustering of parametric MIMO channel data including path powers, in Proc. IEEE 64th
Vehicular Technol. Conf., Sep. 2528, 2006, pp. 15.
146. D.-J. Kim, Y.-W. Park, and D.-C. D.-J. Park, A novel validity index for determination of the
optimal number of clusters, IEICE Trans. Inf. Syst., vol. E84-D, no. 2, pp. 281285, 2001.
147. U. Maulik and S. Bandyopadhyay, Performance evaluation of some clus-tering algorithms
and validity indices, IEEE Trans. Pattern Anal. Mach. Intell., vol. 24, no. 12, pp.
16501654, Dec. 2002.
References 155
148. N. Czink, P. Cera, J. Salo, E. Bonek, J.-P. Nuutinen, J. Ylitalo, Automatic clustering of
MIMO channel parameters using the multi-path component distance measure, WPMC05,
Aalborg, Denmark, Septemter 2005
149. G. E. P. Box and Mervin E. Muller,A Note on the Generation of Random Normal
Deviates,The Annals of Mathematical Statistics (1958), Vol. 29, No. 2 pp. 610611
150. Greenstein et al, Moment method estimation of the Ricean K-factor, IEEE Comm. Lett. Vol
3, Issue 6, p.175, 1999.
151. R. Kolar, R. Jirik, J. Jan (2004) "Estimator Comparison of the Nakagami-m Parameter and Its
Application in Echocardiography", Radioengineering, 13 (1), 812
152. R. M. Norton,The Double Exponential Distribution:Using Calculus to Find a Maximum
Likelihood Estimator,The American Statistician (American Statistical Association) 38 (2):
135136, May 1984. doi:10.2307/2683252.
153. Borradaile, Graham. Statistics of Earth Science Data. Springer. ISBN 978-3.540-43603.4.
Dec 2009.
154. D. J. Best, N. I. Fisher, Efficient Simulation of the von Mises Distribution,Applied
Statistics, vol. 28, No. 2, pp 152-157, 1979
155. Sra, S. A short note on parameter approximation for von Mises-Fisher distributions And a
fast implementation of Is(x), Computational Statistics 27:177190. 2011.doi:10.1007/
s00180-011-0232-x.
156. Andrew T.A Wood, Simulation of the von mises fisher distribution, Communications in
Statistics - Simulation and Computation, 23(1):157-164, 1994.
157. S. Jung, Generating von Mises Fisher distribution on the unit sphere (S2) ,Oct. 2009.
[Online] Available at http://www.stat.pitt.edu/sungkyu/software/randvonMisesFisher3.pdf
158. J. T.Kent, The FisherBingham distribution on the sphere,J. Royal. Stat. Soc., Sr. B,
44:7180, 1982
159. John T. Kent, Asaad M. Ganeiber and Kanti V. Mardia, A new method to simulate the
Bingham and relateddistributions in directional data analysis with applications,Arxiv.
[Online] Available at http://arxiv.org/abs/1310.8110
160. L. Devroye, Non-Uniform Random Variate Generation,Springer-Verlag.1986
161. Donald E. Knuth, Art of Computer Programming, Volume 2:Seminumerical Algorithms (3rd
Edition), Addison-Wesley Professional, November 14, 1997
162. F. Babich and G. Lombardi, Statistical analysis and characterization of the indoor propaga-
tion channel, IEEE Trans. Commun., vol. 48, no.3, pp. 455464, Mar. 2000.
163. WINNER1 WP5:Final Report on Link Level and System Level Channel Models Deliver-
able D5.4, 18.11.2005
164. WINNER+ D5.3, Final channel models, V1.0, CELTIC CP5-026 WINNER+ project.http://
projects.celticinitiative.org/winner+/deliverables_winnerplus.html, 2010.
165. M. Steinbauer, A. F. Molisch, Spatial channel models, in Wireless Flexible Personalized
Communications, L. Correia (ed.), John Wiley & Sons, Chichester, 2001.
166. L. Correia, Ed., Mobile Broadband Multimedia Networks. Academic Press, 2006.
167. L. Liu, J. Poutanen, F. Quitin, K. Haneda, F. Tufvesson, P. D. Doncker, P. Vainikainen, and
C. Oestges, The COST2100 MIMO channel model, IEEE Wireless Commun. , vol. 19, no.
6, pp. 9299, Dec. 2012.
168. C. Oestges,C. Brennan, F. Fuschini, M. L. Jakobsen,S. Salous, C. Schneider and
F. Tufvesson, Radio Channel Measurement and Modelling Techniques, Chapter 9 of
Cooperative Radio Communications for Green Smart Environments, River Publishers 2016
169. 3GPP TR 25.996,Technical Specification Group Radio Access Network; Spatial channel
model for Multiple Input Multiple Output (MIMO) simulations, V11.0.0, 2012-09. http://
www.3gpp.org
170. 3GPP TR36.873, Study on 3d channel model for lte, 3rd Generation Partnership Project,
v12.1.0. 2015. http://www.3gpp.org
171. V. Erceg, et al. IEEE P802.11 Wireless LANs -TGn Channel Models, IEEE 802.11-03/
940r4, May 2004.
172. G. Breit, et al. IEEE P802.11 Wireless LANs - TGac Channel Model Addendum, IEEE
802.11-09/0308r12, March 2010.
173. Cost Final Report, http://www.lx.it.pt/cost231/
174. R. Verdone and A. Zanella, Pervasive mobile and ambient wireless communications,
COST Action 2100, Springer, 2012.
175. L. Liu, Implementation of the COST 2100 model, v2.2.5 [Online]. Available:http://code.
google.com/p/cost2100model/
176. L. Hentila, P. Ky osti, M. Kaske, M. Narandzic, and M. Alatossava. (2007, December)
MATLAB implementation of the WINNER Phase II Channel Model ver1.1 [Online]. Avail-
able:https://www.ist-winner.org/phase_2_model.html
177. J. Medbo and P. Schramm, Channel models for HIPERLAN/2, ETSI/BRAN document
no. 3ERI085B.
178. ITU-R Document 5D/TEMP/332(Rev.1): Preliminary draft new Report ITU-R M.[IMT-
2020.EVAL] Channel model (part), 27th meeting of WP 5D, Niagara Falls, Canada,
June 2017
179. Jian Luo, et. al., Channel Sounding for 802.11ay, IEEE 802.11-15/0631r0, May 2015
180. J. Jarvelainen and K. Haneda, Sixty gigahertz indoor radio wave propagation prediction
method based on full scattering model, Radio Science, vol. 49, no. 4, pp. 293-305, 2014.
181. METIS D1.2, Initial channel models based on measurements, v1.0, 2014. https://www.
metis2020.com/
182. 3GPP TR 38.901. Study on channel model for frequencies from 0.5 to 100 GHz (Release 14).
3rd Generation Partnership Project (3GPP), V14.0.0, Mar. 2017.
183. P. Agrawal and N. Patwari, Correlated link shadow fading in multihop wireless networks,
IEEE Trans. Wireless Commun., vol. 8, no. 8, pp. 40244036, Aug. 2009.
184. T. Jamsa and P. Ky oti, Device-to-device extension to geometry-based stochastic channel
models, EuCAP 2015, Lisbon, Portugal, 2015.
185. W. Fan, T. Jamsa, J. O. Nielsen and G. F. Pedersen, On Angular Sampling Methods for 3.D
Spatial Channel Models, IEEE Antennas and Wireless Propagation Letters, Vol. 14, 2015.
Chapter 4
Software Simulation
Starting from the requirements and technical indicators of 5G network system, this
chapter analyzes the technical challenges in designing and realizing 5G software
simulation system, and introduces the link level simulation and system simulation
techniques with the focus on test evaluation methods, key technologies and appli-
cations, which provides technical reference for the design and realization of 5G
network system software testing.
4.1 Overview on Software Simulation
The modern wireless communications system is rather complicated in terms of

application scenarios, network structure, functions, characteristics, etc. Take 4G
LTE network as an example. A network that can work in reality needs eNodeB,
Mobility Management Entity (MME), Packet data network GateWay (PGW),
Serving GateWay (SGW), Home Subscriber Server (HSS) and many other network
elements as well as related interfaces and protocol stacks. When compatibility and
interoperability with other systems are considered, its network structure and net-
working methods would be more complicated. In addition, the wireless communi-
cations system runs in a real environment which is complicated and volatile. Its
difficulty and complexity further increase with propagation features and user
features like SF, multipath effect, high mobility, strong interference, etc. Due to
plenty of realistic factors above, in reality its difficult to build a complete and
accurate model to analyze the design of wireless communications system. Usually,
its more feasible and effective to set up a software simulation platform via
computer and mathematical modeling, on which software simulation programs
can make performance evaluation for the wireless communications modules or
system.
5G candidate technologies are more abundant and its application scenarios are
more complicated. Accordingly, the software evaluation for the performance of 5G

DOI 10.1007/978-3-319-61869-2_4
158 4 Software Simulation
technical schemes is facing the unprecedented challenges. Therefore, the academia

and industry have been committed to improving the software simulation platforms
computational performance, flexibility and accuracy, and have achieved abundant
research fruits and practical results, providing important reference for building the
5G simulation platform.
GPU is one of the important methods for parallel acceleration. For example, Ben
et al. [1] propose a distributed computation architecture based on CPU-GPU to
solve the problems including the high price, the heavy inter-processor synchroni-
zation overheads, and the huge communications overheads for inter-server distrib-
uted coordination in the multi-core CPU based parallel computing simulation
platform, which aims at meeting the simulation requirements of the large scale
mobile network. This new architecture designs the hardware and software jointly,
decomposes the computational tasks through master-slave node model, and makes
layered design for the whole software based on modular design of application
module, protocol stack, message service, connectivity management, mobility man-
agement and system interface function. The master node deployed on CPU takes
charge of event scheduling, data abstraction design, scenario management and other
functions. It has solved the problem of low parallel synchronization through events
synchronization, data decoupling by functions, memory hierarchical management
and other mechanisms. The slave nodes deployed on GPU execute simulation tasks,
which fully utilize GPUs computing power and reduce the communications data
effectively. Measured results prove that this architecture can improve the operating
efficiency significantly.
Open source software based simulation development and design make full use of
the unique open and sharing characteristics of open source community, and accel-
erates the realization of simulation. For example, Capozzi et al. [2] propose a
LTE-Sim based open source tool, which supports the HetNet scenario with macro
eNodeB and femtocell coexisting. This open source based new model can meet the
simulation requirements of specific scenarios quickly and flexibly. It is one of
trends for future simulation platform development.
Due to the fast development of 5G, the simulation objects, scenarios and
technologies are developing as well, and the design and evaluation methods also
need to develop synchronously. For example, Goga et al. [3] have analyzed the
characteristics of cloud computing, proposed design architecture and simulation
method for cloud computing application, and concluded the key optimization,
modeling and design problems in the follow-up research.
Next, a complete simulation system is introduced briefly, as shown in Fig. 4.1.
Requirements analysis. Traditional simulation process starts with original
requirements. Generally, a networking scenario is needed to be given, including
network type, network element type, network interface function, network con-
figuration parameter, services characteristics, performance indicators, etc. These
original requirements give the simulation objects and corresponding behaviors.
Through analyzing original requirements, the input requirements for simulation
system modeling and network performance evaluation criteria can be obtained.
4.1 Overview on Software Simulation 159
Requirement Analysis
Networking Network Service Performance Other
QoS
Scenarios Characteristics Type Index Requirements
System Modeling
Link Level System Channel Traffic Antenna Other
Model Level Model Model Model Pattern Models
Simulation Realization Evaluation Criteria
Simulation Simulation Simulation Performance Simulation Evaluation

Framework Methodology Procedure Metrics Benchmark Principle
Simulation Results and

Evaluation Report
Fig. 4.1 Composition and implementation process of simulation system
System modeling. System modeling includes abstracting and modeling the

network and system components, and then outputting the simulation model.
These simulation models describe the input/output relations of the network
and system components. The models not only need to maintain the basic
characteristics of the modeled objects, but also cannot be too complicated.
Thus there is a tradeoff between accuracy, complexity and calculation, which
is where the difficulty and emphasis of simulation design lie in. Generally,
complex models design adopts the way that combines top-down with bottom-
up. The former allows the designer to seize the design process line from
requirements to system, subsystem and modules, which helps to reduce the
unnecessary duplicated work and improves work efficiency to the greatest
extent. The latter can guarantee the accuracy of basic simulation objects and
help to break down and implement in parallel the simulation work.
Evaluation criteria. Through the analysis of original requirements, the input
information of simulation evaluation can be obtained, which includes perfor-
mance indicators, simulation benchmark and evaluation criteria. Performance
indicators are used to evaluate the measurable indicators of the system to be
simulated. Simulation benchmark provides benchmark data for evaluation,
which is the reference for comparing different simulation schemes. Evaluation

criteria are used to evaluate the advantages and disadvantages of performance.
Different evaluation criteria may lead to different evaluating conclusion. Com-
mon evaluation criteria include fairness criteria and throughput criteria.
Simulation realization. After the completion of simulation model design is the
simulation realization stage. Simulation realization includes three elements [4],
namely simulation system architecture, simulation methodology and simulation
process.
The design of simulation system architecture defines the simulation systems
logic architecture, development architecture, running architecture, physical archi-
tecture and data architecture [5]. Logic architecture defines software function
partitioning, design of categories and objects, interface definition and other basic
logic functions of simulation system design. Architecture development mainly
considers the scalability, reusability, comprehensibility, testability and other basic
ability during the development. Architecture development also contains the prac-
tical organization mode of software modules. For 5G, it must provide sufficient
scalability to adapt to the change of system architecture. Running architecture
mainly considers the simulation systems performance, usability, and safety,
including the concurrency and synchronization of system modules, and process,
thread and other running concepts. 5G simulation system significantly increases the
computation complexity, which is a big challenge for the design of development
architecture. Physical architecture focuses on how to install or deploy software on
the physical machine. For 5G simulation system, the model selection of physical
computation platform and parallelizing ability of software are the key factors that
influence the computational performance. Data architecture mainly concerns about
the data requirements, including data storage, duplication, transmission,
synchronization, etc.
Simulation methodology is used to guide the definitions of link level and system
level simulation model and relevant parameters.
Simulation processes have defined the processes and steps of simulation. A
typical system simulation process includes: generating network element and net-
work topology, generating users and allocating users to the serving cells randomly,
generating channel model, generating service packets according to service model,
resource scheduling, service packets processing, interference computation, users
Signal to Interference plus Noise Ratio (SINR) computation and outputting simu-
lation results.
Simulation results and report. Simulation system can output simulation data after
it runs as the required process; and based on simulation datas analysis results,
the simulation report can be got eventually. Actually, simulation data needs
further analysis to determine whether the simulation data is reasonable, whether
the output indicators meet the expectation and whether the design and realization
of simulation system need improvement, etc. Generally, the whole simulation
system needs calibration before the authentic report is obtained, that is, observ-
ing whether the output result is same with recognized benchmark system at the
4.2 5G Software Simulation Requirements 161
given input parameters and models so as to check whether the simulation system
realization is right. Due to the introduction of massive MIMO and other new
technologies, 5G simulation system needs to implement plenty of calibration
work according to new simulation parameters and models to guarantee the
correctness of the simulation output result.
From the above analysis, we can know that the software simulation process of
the whole wireless system is very complex. Thus the workload of building simu-
lation system software is huge. To improve the efficiency, simulation designers can
develop the simulation system with the help of proven commercial software
package. Common commercial software tool packages include MATLAB,
OPNET [6], NS2/NS3 [7], QualNet [8], etc. All of these packages provide all
kinds of basic parts that software framework of communications system needs,
including modeler, model library, simulation kernel and postprocessor, but the way
of realizing these parts and focus range of provided model library might be
something different. The design and development of 5G software simulation testing
system inherit and develop from4G and earlier technologies. Early simulation
platforms are of great reference value for building 5G simulation software. Mean-
while, since 5G system will introduce more new functions and new technologies,
designers and developers need to deeply analyze the characteristics and realization
plans of candidate technologies so that they can design and realize 5G software
simulation system efficiently.
4.2 5G Software Simulation Requirements
4.2.1 Simulation Requirements Analysis
As an entirely new network, 5G is the wireless communications system that faces the
information society in 2020. Currently, 5G has entered the key stage of key tech-
nologies breakthrough and standards formulation. Thus corresponding key technol-
ogies simulation requirements will also help the 5G simulation to be promoted to a
new height on system architecture and simulation efficiency. Evaluating the perfor-
mance indicators of 5G network comprehensively, fast and accurately is the impor-
tant requirement for designing and realizing 5G software simulation systems.
Comprehensiveness is the functional requirement of the software simulation
system, which mainly refers to the comprehensive support for 5G performance
indicators and candidate technologies. 5G performance indicators include Key
Performance Indications (KPIs) such as peak data rate, guaranteed minimum user
data rate, connection density, service traffic volume density, wireless delay, end-to-
end delay [9], etc. It also includes secondary indicators that support these KPIs such
as SINR, paging success rate, access success rate, handover success rate, etc.
Software simulation system must provide corresponding modeling, statistic and
evaluation for these new performance indicators. 5G candidate technologies can be
classified into two types, namely air interface technology and network technology.
Air interface technology includes EE-SE co-design, massive antenna, full duplex,
novel multiple access, new waveform, new modulation and coding, software
defined air interface, sparse mining, high frequency band communication, spectrum
sharing and flexible application, etc. Network technologies include C-RAN,
SDN/NFV, Self-Organizing Network (SON), Ultra Dense Network (UDN), multi-
network convergence (multi-RAT) and D2D [9], etc. These new technologies
impact on existing network design and realization can be divided into three types
in terms of architecture, namely architecture level, network element level and
module level. New technologies in architecture level, such as SDN and SON, have
the greatest impact on the design of software simulation system. Because it needs to
model the new network architecture, design new network element types, interfaces
between new network elements and new protocol stacks. All these changes are basic
and we still need to make large-scale changes on the entire network design, which
make the design more difficult and bring software simulation more workload. New
technologies in network element level, such as massive MIMO technology, involve
coordinated design between multiple modules and the influence is mainly limited in
the network elements, so there is no need to make big changes on network architec-
ture and interface between network elements. In comparison, the influence scope of
new technologies in module level is much smaller, which is limited in one function
module within the network element and other related modules can support it with
small modifications, for instance, new modulation coding technologies. The main
work of software simulation system includes comprehensively analyzing every 5G
candidate technologies, proposing corresponding resource models, and designing
and implementing schemes, which serve as significant guarantees for the complete-
ness of 5G software simulation system functions.
Rapidity is the time efficiency requirement for software simulation system. This
requirement is embodied in two aspects. One is that it can substantially reduce
simulation time of 5G network traffic and performance evaluation. The other is that
it can quickly adapt to the change and adjustment of network architecture. Part of
5G technologies [10] have huge simulation objects, for example, the base station
antenna scale of massive MIMO technology can be more than 128, and the number
of cells under UDN reaches hundreds or even more. Some computational overheads
of the simulation characteristics are very large. For example, massive MIMO
precoding calculation involves a lot of more than 128 30 dimensions matrix
operations. Large simulation objects scale and computational complexity require
the computational performance of software simulation system based on existing
system to have great improvement. Only in this way can the requirements of
simulation tasks and timely evaluation be met. Sharp growth of requirements for
computational performance requires the systematic brand-new design and imple-
mentation of the hardware platform, software platform and the simulation program.
For the software simulation system, the key problem is how to complete the
software system concurrent design and coding implementation on the new and
powerful hardware platform with powerful computing power, which require the
simulation program to realize the greatest concurrent on key computation paths,
4.2 5G Software Simulation Requirements 163
such as the concurrent processing of granularity that reaches the subcarrier level.
Rapidity is also reflected in the fast response to the change of 5G network archi-
tecture with a relatively small cost. Different from previous wireless communica-
tions systems, 5G has many network level candidate technologies with great
changes in network architecture and rich supporting scenarios, so the simulation
system needs to be flexible enough. Its architecture design, module design, and
interface design should have characteristics like decoupling, modularity, interface
scalability and ease of integration, etc., which put forward higher requirements on
network resource model design, software function modular design, and the inter-
face design between network elements and modules. In general, the complete cycle
of development and validation of a simulation task should be as short as possible, so
that it can meet the rapid verification and selection requirements of 5G candidate
technologies.
Accuracy is performance requirement for software simulation systems. One of
main factors influencing the simulation accuracy is the simplified method intro-
duced to network modeling, simulation service abstraction, design and realization.
Those simplified methods are usually the results of the compromise between the
simulation authenticity and the simulation calculation cost, i.e., reducing the sim-
ulation calculation cost at the cost of simulation authenticity. Take the widely used
EESM/MIESM link mapping method as an example. This method makes equiva-
lent mapping for link model, in which the simulation process of the link function is
equivalently mapped to several performance system level indicators, so as to
simplify the computational process of link realization in the system simulation
process; yet meanwhile system errors are also introduced. In order to reflect the
impact of link process on system simulation more accurately, hardware and soft-
ware methods can be combined to truly reflect the link process. The link to be
simulated is deployed in hardware (such as FPGA, channel simulator) for real-time
calculation, and the simulation accuracy can be improved through the hardware
simulation calculation. Other simulation processes can do similar processing. When
computational overheads allows, the simulation accuracy can be improved at the
cost of increasing computational resources.
On the basis of summarization of 5G network software simulation systems
requirements, objectives and implementation ideas, the overall technical roadmap
can be obtained in detail, as shown in Fig. 4.2. All the key supporting techniques
involved will be discussed in detail in subsequent chapters.
Based on the above simulation requirements analysis, we will take the require-
ment for rapidity as an example, to briefly introduce some common solutions. All
the present communications networks have complex system architectures. It usu-
ally takes very long time to do the simulation calculation to accomplish a specific
task. The time needed for communications system simulation is at least a few days,
and can be as long as a month or even longer. Aiming at this problem, in addition to
the advanced simulation technology shown in Fig. 4.2, as the old saying goes,
Good tools are prerequisite to the successful execution of a job. We should also
include the following common ways to improve the simulation efficiency from the
perspective of hardware configuration.
Fig. 4.2 The overall technical roadmap of 5G software simulation
1. High configuration desktop. The most simple and direct method to improve the
simulation efficiency is to buy high configuration computer, such as choosing
multi-core high-speed processors, large memory and high speed hard disk,
which is the most common practice to improve the simulation efficiency.
2. Multicore servers. Improvement approach typically refers to the purchase of
multicore servers to improve the simulation efficiency. To achieve the best effect
with the server simulation, it also needs to design parallel program, and make
each CPU run programs respectively.
3. High performance supercomputer. Supercomputers have higher system rate,
there are special parallel tools and job scheduling system, which can be very
convenient for users. Renting a supercomputer is not only convenient and fast
but also cost controllable. It can be charged through the actual number of CPU
cores used and the time in the simulation program.
4.2.2 Technological Impact Analysis
The main factors influencing the system simulation software design and implemen-
tation can be divided into three categories: architecture, function and performance.
Architecture reflects the design of components and interfaces of the system simu-
lation software as well as the definition of the interfaces. Function embodies
simulation objects and behavior scope of the system simulation software. Perfor-
mance shows the computational performance that can be reached by system
simulation software and its platform. As shown in Table 4.1, the effects of main
new 5G candidate technologies on the simulation software design and implemen-
tation can also be analyzed from the above three aspects.
Table 4.1 Analysis on the impact of 5G key technologies on system simulation software
Impact analysis
Key Computational
technologies Overview Software architecture Simulation function performance
Massive In the base station end hundreds of anten- No impact Functional enhancements of Antenna number increases
MIMO nas (128, 256 or more) are installed to channel model and the greatly, interference calcu-
realize the data transmission of massive physical layer are added, lation, precoding, receiver,
MIMO. Specifically, it includes precoding including precoding algo- scheduling and other simu-
design, pilot frequency design, feedback rithm, channel estimation lation modules show
design, receiver algorithm, large-scale and signal detection, etc. sharply increased calcula-
active array design, and channel measure- tion strength
ment and modeling, etc.
EE-SE joint- Energy efficiency and spectrum efficiency No impact New algorithm function Depending on the compu-
design are jointly designed, providing compre- tational complexity
4.2 5G Software Simulation Requirements
hensive instruction and design framework

for the choice of transmission technology,
radio frequency link and antenna configu-
ration, wireless resource management and
network deployment, etc.
Full-duplex improve spectrum efficiency through No impact The physical layer module Calculation intensity is
communication Co-time Co-frequency Full Duplex supports full-duplex mode; enhanced in interference
(CCFD) resource scheduling module calculation, scheduling,
is enhanced receiver modules.
New multiple New multiple access technologies super- No impact The physical layer module Computational complexity
access pose the signals from different users at the supports the new multiple increases at the receiver
technologies same time frequency domain, in order to access function; receiver, module
provide higher spectrum efficiency and resource scheduling and
improve the system access and the capac- other modules are enhanced
ity; the receiver uses the advanced algo-
rithm for demodulation
(continued)
165
166
Impact analysis
Key Computational
technologies Overview Software architecture Simulation function performance
New coding Including LDPC codes of FQAM, Raptor No impact Physical layer module sup- No impact
and modulation type, Polar code, Gray coding of ports new coding and modu-
non-uniform distribution APSK, diversity lation; the link adaptive
technology of joint coding modulation and module is enhanced
other new coding modulation technologies
High frequency Communications are realized through No impact Channel model, Calculate intensity
communication short-wavelength millimeter wave band, beamforming, and some increases with the increase
which is able to provide more than ten other functions are added of system bandwidth
times bandwidth than 4G
C-RAN A green wireless access network architec- Virtualization technology New network types such as Calculation intensity
ture based on centralized processing, col- requires the various modules virtual cell are supported; depends on the complexity
laborative radio and real-time cloud of the simulation system to topology management, wire- of the centralized manage-
computing architecture decouple fully less resource management, ment algorithm
resource, and resource
scheduling modules are
enhanced
UDN The deployment density of low power sta- It is necessary to support the Topology management, Calculation intensity
tion in hot spots in the cellular network is ultra dense nodes deploy- interference calculation, and increases with the increase
increased to improve system capacity and ment, virtual cell and other handover modules should be of number of base stations
network coverage and to reduce the delay characteristics enhanced
and energy consumption
SDN In the SDN architecture, control plane and Control plane and data plane New type network element is Calculation intensity is
data plane are separated; control function is separated; centralization and supported; protocol stack decided by the network
centralized; the bottom network infra- virtualization management module, resource manage- scale and the complexity of
structure is abstracted; and interfaces for for control modules;, unified ment module, resource centralized management
the application and network services are interface is opened scheduling module all need algorithm
provided to be enhanced
4 Software Simulation
4.3 5G Software Link Level Simulation 167
4.3 5G Software Link Level Simulation
4.3.1 Link Technology Overview
Major breakthroughs in basic theories and key technologies of the wireless trans-
mission have been leading the evolution and innovation of the wireless communi-
cations system and its standardization. A series of novel multiple access
technologies and transmission technologies (such as TDMA/FDMA/CDMA/
NOMA, OFDM/MIMO, etc.) have brought about the significant improvement of
the information transmission rate and opened up a new era of the wireless technol-
ogy revolution [11]. Facing the massive growth of the future wireless data and the
rich variety of services and experience requirements, 5G transmission technologies
will explore a series of new types of multiple access and transmission mechanisms,
to greatly improve the spectrum efficiency and the energy efficiency of the wireless
systems [9].
The academia and industry at home and abroad have put forward a number of
candidate technologies for 5G wireless transmissions. Massive MIMO, high-
efficiency modulation and coding, non-orthogonal multiple access, full duplex
and other new transmission technologies have caused great concern in 5G [12]. Fur-
thermore, energy efficiency becomes one of the important performance indicators
of the 5G system [13]. Compared with the 4G technologies, the signal processing
mechanism adopted by the 5G transmission system will be even more complicated,
and performance evaluation indicators would be more multidimensional, which
will challenge the 5G air interface technology standardization, the program system
design and the simulation verification.
As shown in Fig. 4.3, massive MIMO has become one of the important 5G key
transmission technologies [14]. Massive MIMO can bring the huge array gain and
the interference suppression gain through large-scale antenna arrays, thus greatly
improving the system spectrum efficiency and the edge user spectrum efficiency.
Current researches on the massive MIMO transmission mainly focus on system
Fig. 4.3 An example of massive MIMO in heterogeneous wireless networks scenario

capacity, performance analysis, precoding, pilot pollution and other issues. The
evaluated performance results based on the link level simulation for the massive
MIMO technology will present important information for the massive MIMO
system design.
Compared with the traditional multiple access technology, the non-orthogonal
access technology based on the advanced waveform design can achieve much
higher spectral efficiency. Non-orthogonal accesses in the frequency domain, the
time domain as well as the code domain, such as Interleave Division Multiple
Access (IDMA), Low Density Spreading Multiple Access (LDSMA), spatial cou-
pling, and Power Domain Multiple Access (PDMA), have attracted wide research
attentions [15]. The performance of multiple access technologies, as the key
components of the wireless system physical layer transmission, needs comprehen-
sive evaluation and analysis with the help of link level simulation.
In addition, the future 5G network deployment will have distinctive heteroge-
neous characteristics [16], which include heterogeneous connections through var-
ious air interface technologies, the heterogeneous network coordination and
interference management, etc. In the massive MIMO and dense coverage scenario,
the highly-efficient coordinated interference signal processing methods such as
interference alignment will have great expected performance in improving the
systems coverage, the spectrum efficiency and the user service experience, etc.
The complex interference modeling and interference processing mechanism will
also play important roles in improving the precision of the simulation results, which
will be covered in both the 5G link level and system level simulation.
Overall, with the breakthrough in technologies such as massive MIMO, the
mobile communications industry has been in the stage of the 5G era. In terms of
5G transmission technologies, on the one hand, we need to carry out the theoretical
research and algorithm design. On the other hand, we also need to put forward the
corresponding technical evaluation mechanism to provide the detailed performance
evaluation results through constructing the efficient and accurate link level simu-
lation platform.
4.3.2 Link Simulation Realization
Simulation Key Factors
The link level simulation is mostly used to evaluate the physical layer transmission
performance of the wireless communications system. It usually includes the down-
link simulation and the uplink simulation under certain configurations for the simu-
lated wireless channel and the adjacent cell interference environment [11]. Through
modular simulation design, the link level simulation can realize the performance
comparison of different transmission schemes with a variety of transmitter structures
and receiver algorithms. Therefore, it is able to provide the important reference for
choosing the appropriate physical layer design or implementation schemes.
Performance metrics are important in the link level simulation, which mainly
include the Bit Error Rate (BER), the BLock Error Rate (BLER) or Frame Error
Rate (FER) and spectrum efficiency (bps/Hz), etc. As the increasing concern on the
power consumption, the performance evaluation of energy efficiency is also intro-
duced in the 5G system. The above metrics are used to evaluate the reliability
(BER, BLER, FER) and effectiveness (spectrum efficiency, energy efficiency) of
the system transmission. It mainly provides the performance results when the above
performance metrics change with the SNR or the SINR. The provided simulation
results will be used to determine whether the designed physical layer transmission
structure and the receiving algorithm meet the system performance requirements. It
also provides the design directions for the algorithm improvement.
Compared with 4G, the 5G transmission will pose higher requirements for the
computing abilities and the simulation speed of the link level simulation. In
addition, when transmitting wireless signals at high frequency bands, the physical
environment of 5G transmissions will be more complicated. In order to improve the
validity and authenticity of the link level simulation, the researchers will model the
wireless channel and the interference signals more accurately based on the actual
measure data. This will no doubt increase the complexity and the processing time of
the wireless link simulation. For the 5G link level simulation, the 5G software needs
to adopt new simulation methods, such as the high speed parallelization, to cope
with the challenge posed by the technical complexity of 5G new technologies. It
will be able to perform a series of rapid evaluation and verification for candidate
transmission technologies.
Concerning all the above mentioned problems brought by massive MIMO, dense
network nodes, and the more complicated wireless transmission environment, the
link level simulation needs to consider a variety of advanced simulation methods
and implementation mechanisms. Meanwhile, since there are a large number of
candidate technologies for the 5G transmission, it is necessary to make a detailed
analysis on technical features of candidate technologies to determine the
corresponding technologies application scenarios as well as the evaluation process.
The factors that need most consideration in the link level simulation of the main
candidate technologies include the following aspects.
Simulations for the massive MIMO channel
The channel model plays vital important roles in the link level simulations. In
terms of the 5G link level simulation, it is necessary to carefully consider various
characteristics of the massive MIMO channel such as the spatial correlation, the
coupling, the near field effect, etc. [17]. The empirical channel model can also be
constructed through analysis, comparison and fitting of the measured channel
data in combination with the theoretical analysis.
Simulations for the neighboring interference
In 5G scenarios with the ultra-dense nodes, the co-channel interference will
limit the network capacity [18].For the link level simulation, it needs to accu-
rately reflect the influence of the interference links from neighboring cells. On
the one hand, the real interference signals can be used in which the real-time
signals are generated through multiple interference links. As an alternative

method, the interference signals can also be theoretically modeled through
theoretical derivation in order to simplify the simulation complexity.
Simulations for the novel multiple access technology
In order to support the new physical layer schemes such as the
non-orthogonal multiple access, its necessary to build the corresponding link
layer simulation to evaluate the actual performance of the corresponding tech-
nology. It needs a comprehensive evaluation on the impact on the spectral
efficiency and the impact on the robustness of the non-ideal factors such as
frequency offset and channel estimation errors. Meanwhile, the performance
mapping from the link level to the system level corresponding to this physical
layer must be obtained in order to accelerate the execution speed of the system
level simulation.
Simulations with high-performance multi-core parallelization
As mentioned above, the 5G system will be configured with massive numbers
of antennas. Besides, it will also consider more complex channel models,
interference signals, the non-orthogonal signal processing and other more com-
plex transmission environments. Therefore it requires much higher capabilities
to realize the link simulation. The multi-thread parallel simulation, the high-
performance multi-core server, the software and hardware co-simulation and
other advanced simulation methods will be conducive to the fast and accurate
link level simulation evaluation.
Simulation Process Overview
The main role of the physical layer transmission is to ensure the reliability and
effectiveness of the information wireless transmission. The design of each function
module of the physical layer is usually completed after a lot of discussions on the
performance analysis, the algorithm evaluation and the standardization [19]. Take
the downlink system of wireless transmission as an example. The signal processing
modules at the transmitter mainly include channel coding, constellation mapping,
multiple antenna precoding, multiple access, reference signal generation, framing,
etc. The signal processing modules at the receiver mainly include cell search and
synchronization, de-framing, channel estimation, multiple antenna detection,
demodulation, channel decoding, etc. [2021]. In the link level simulation, pro-
gramming is required to realize each of the above mentioned function modules. In
addition, it also needs to implement the modules for the wireless fading channel and
the interference links. During the establishment of the link level simulation plat-
form, in order to guarantee the correctness of the simulation, its also necessary to
go through multiple levels of tests, in turn, including functional module testing,
subsystem testing, and integrated system testing.
The wireless link level simulation system usually includes two sets of simulation
systems for downlink and uplink. In order to improve the efficiency and transport-
ability, it usually needs to complete the software architecture design in advance.
System parameter configuration
Library Main function

functions
(math
functions,
memory Wireless
Downlink Uplink
fuctions, testing channel
functions,
etc.)
Data storage and figure plotting
Fig. 4.4 Example of the link level simulation platform architecture
Figure 4.4 shows an example of the architecture design of the link level simulation.
In it, the library functions provide to the whole simulation system the public library
functions, respectively supporting functions like math calculation, memory alloca-
tion, testing comparison, etc. In addition, the section of the system parameter
configuration covers the system configuration parameters required by simulations.
It may include the system-level parameters such as the frame structure, the system
bandwidth, cell information, resource allocation methods, etc. It can also include
the transmitter configuration parameters such as control channel parameters and
data channel parameters. In addition, the receiver configuration parameters can also
be included such as configurations for the receiving algorithms. The part of the
main function completes the whole link process in accordance with the definitions
and transmission mechanisms defined as the standard protocols. Real-time simula-
tion results will be saved, and the simulation evaluations curves are drawn when
necessary. It is important for the architecture design to be with capabilities of
flexibility, detectability and transportability.
The realization of each function module in uplink and downlink and the interface
definition between modules are the most basic parts of the link level simulation. It
will realize the indispensable functions for the physical layer transmission. They
includes function modules for channel coding and decoding, modulation and
demodulation, multiple antenna precoding and detection, multiple access, reference
signal generation and channel estimation, cell search and synchronization, and
measurement and feedback [2122].
With the introduction of massive MIMO, non-orthogonal multiple access and
other candidate technologies to 5G transmission system, the implementation way of
multiple antenna processing module and multiple access module would undergo big
changes compared with that in 4G transmission system in the link level simulation.
Figure 4.5 gives a general process of the transmitter for the single user transmission.
As shown in this figure, the resulting implementation for massive MIMO and
Reference
signals (RS)
Source
Channel Multiple
CRC Mapping Precoding
encoding access
Massive MIMO, RS design,

Non-orthogonal multiple access,
Resource allocation
Fig. 4.5 Example of a general process of transmitter link simulation
Test for function and performance
Cell search/ Channel MIMO Channel Information

Deframing Demapping
synchronization estimation detector decoder bits
SNR /Frequency Noise/Doppler/ The number of antennas/ Soft Parallization Performance

offset Inter ference multi-mode detector information /Flexibility /High- metric
/noise imbalance speed realization
Received baseband signals

Radio frequency Analog front Digital front
(RF) end end
Fig. 4.6 Example of a general process of the receiver link simulation
non-orthogonal multiple access will affect the simulation modules of the reference
signals, the multi-user resource allocation, etc.
In the link level simulation, the receiver often has higher algorithm complexity
and therefore requires higher computing capability. The receiver usually involves
RF, analog front end, digital front-end, baseband processing and other components.
In the link level simulation, it usually pays more attention to baseband simulation
implementation and evaluation. The non-ideality brought by the RF link, analog
front-end and digital front-end is modeled and introduced to the baseband simula-
tion. Figure 4.6 shows the main processing of a receiver. For simplicity, it simplifies
the example of data extraction modules such as de-frame, resource demapping, etc.
For the receiver, each module has multiple choices for receiving algorithms. And
different algorithms concern different aspects of the system design in terms of
complexity and performance. Theoretical analyses are usually able to provide some
conclusions on the algorithm performance. But due to some ideal assumptions, the
performance results usually have great gap with the actual situations. At this point,
the link level simulation will play an important role in the performance evaluation
of the receiver algorithm.
In addition to the general simulation process of the transmitter and the receiver,
the physical layer transmission also needs to distinguish between the data channel
and the control channel. They have great differences in key performance indicators.
For example, the requirements for the transmission reliability of the control channel
are usually much higher than those of the data channel. Therefore, the link level
simulation also needs simulation verification for a wide variety of physical channels
respectively. Besides, the performance and complexity of the receiver algorithm
will also directly affect the choice of the transmitter for each function module. After
the completion of the transmitter standards establishment, how to design a cost-
effective receiver is also important for chip and equipment manufacturers. There-
fore, detailed and reliable link level simulation results are of great significance for
evaluating the physical transmission technology, transmission system design, stan-
dardization and product design. The link level simulation results of each algorithm
module and the overall link under different scenarios and parameters configurations
will provide important reference basis for the scheme design and system perfor-
mance optimization.
4.3.3 Introduction to Simulation Cases
Case 1: Massive MIMO Performance Simulation
This simulation evaluates the link level performance of the massive MIMO system
and the 88 MIMO system. For simplicity, the ITU multipath channel model is
adopted in the simulation (ITU Vehicular A).The channel state information is
assumed to be known. Turbo 1/3 code rate and Quadrature Phase Shift Keying
(QPSK) modulation are used. In addition, Adaptive Modulation and Coding (AMC)
and HARQ mechanism are not initiated. In the simulation, BER, FER and through-
put are performance evaluation criteria. In the massive MIMO simulation, the
number of the base station antenna is set to 128, and the number of users took
K 20, 30, 40, 50 respectively. When the number of the user changes, the
simulation results of different users can be obtained through evaluation. The total
users throughput under Zero Forcing (ZF) precoding transmitting scheme can be
obtained through simulation. In the 88 MIMO system simulations, the base
station is equipped with 8 antennas, the terminal is configured with single antenna,
and the number of users is 8. Table 4.2 shows the main parameters configuration in
the simulation.
Figures 4.7, 4.8 and 4.9 respectively show the performance of BER, FER, as well
as throughput. Since the massive number of antennas can bring a higher degree of
freedom, the performance of massive MIMO is far superior to the performance of
the existing small 88 MIMO system.
Table 4.3 shows the corresponding required Eb/N0 and the corresponding
throughput of different users under FER 5%.For example, when the number of
single antenna users in a massive MIMO system is 20 and FER 5%, the required
Table 4.2 Simulation parameters configuration

Parameter 88 MIMO system Massive MIMO system
Carrier frequency 2.0GHz 2.0GHz
system bandwidth 5 MHz 5 MHz
sub-carrier number 512 512
CP length 32 32
Sampling rate Fs 7.68 MHz 7.68 MHz
The number of base station antennas 8 128
The number of users 8 single-antenna users 20 ~ 50 single-antenna users
Base station precoder ZF precoding ZF precoding
Channel model ITU Vehicular A ITU Vehicular A
Coding type Turbo 1/3 Turbo 1/3
Modulation type QPSK QPSK
CSI Ideal Ideal
0
10
massive MIMO: (128*1, 50)
-1
10 massive MIMO: (128*1, 30)
conventional MIMO: (8*1, 8)
-2
10
BER
-3
10
-4
10
-5
10
-6
10
-10 -5 0 5 10 15 20
Eb/N0 (dB)
Fig. 4.7 BER performance under different users
Eb/N0 is 13.5 dB, and the corresponding throughput is 280 Mbit. For the 8x8
MIMO system, when FER 5%, the required Eb/N0 is 16.5 dB and the
corresponding throughput is 125Mbit. As it shows, when FER is 5%, the SNR of
massive MIMO system can be reduced to 30 dB and its throughput improves by
280/125 2.24 times. As a result, the massive MIMO system can greatly reduce the
power consumption of the base station and therefore improves energy efficiency
and spectrum efficiency.
0
10
-1
10 massive MIMO: (128*1, 50)
-2
10 massive MIMO: (128*1, 20)
conventional MIMO: (8*1, 8)
FER
-3
10
-4
10
-5
10
-6
10
-10 -5 0 5 10 15 20
Eb/N0 (dB)
Fig. 4.8 FER performance under different users
1000
900 massive MIMO: (128*1, 40)
800
700 conventional MIMO: (8*1, 8)
Throughput (Mbps)
600
500
400
300
200
100
0
-10 -5 0 5 10 15 20
Eb/N0 (dB)
Fig. 4.9 Throughput performance under different users

Table 4.3 The required Eb/N0 and corresponding throughput under different users when
FER 5%
Single antenna UE number 20 30 40 50 88 MIMO system
Eb/N0 (dB) 13.5 9.6 6.8 4.2 16.5
Throughput (Mbit) 280 410 440 700 125
Table 4.4 Parallel acceleration results of link simulation (numerical unit: second)
Computational scheme K 20 K 30 K 40 K 50
Serial (single core CPU) 296,642 356,757 455,895 699,380
Parallel (40-core CPU server) 9215 10,697 11,992 18,137
Acceleration ratio (40-core parallel vs serial) 32 33 38 38
In terms of the simulation time, the general serial simulation method often
requires a long time since the link level evaluation needs to simulate the perfor-
mance of multiple SNR points and each SNR point needs to simulate fading
channel fully with ergodicity. Reasonable parallel simulation can greatly reduce
the simulation time. For the above simulation, the following is a brief experiment
for comparison.
The traditional serial simulation method obtains the data of one point when
running once. The program is running on a single CPU. One point on a massive
MIMO link needs 4 days running time and the complete massive MIMO link
simulation needs as many as 100 sampling points.
To accelerate the process, the parallel optimization is carried on. The link
simulation consists of two layers of cycles, namely the SNR cycle and the frame
cycle. Data in the SNR cycle and the frame cycle are independent. Thus the
simulation programs on the SNR and frame cycles can be parallel optimized. And
the theoretical parallel speedup result of the new parallel simulation method can be
faster in orders of the number of SNRs times by the number of the frames.
Parallelization of the SNR and frame cycles, of course, is the relatively simple
and effective way to improve the simulation timeliness. With additional internal
parallel operations, the simulation time can be further shortened, which will not be
discussed here in detail. The results after the parallel acceleration optimization are
shown in Table 4.4.
Case 2: Heterogeneous Network Energy Efficiency Simulation
Energy efficiency is one of the key performance indicators in 5G systems. And the
wireless heterogeneous network is an effective means to alleviate the contradiction
between the data growth and the energy consumption [23]. In respect of mobile
intelligent terminals, different access methods and network planning schemes will
have different influence on the energy efficiency of mobile intelligent terminals.
K. Wei et al. [24] have proposed an energy efficiency indicator with comprehensive
consideration of the terminal and the network. The indicators have considered the
impact of the new base station layout on the energy consumption increase at the
network side and the energy consumption changes at the terminal side at the same
time. So it can provide a quantitative reference criterion for operators to decide
whether laying out a new base station is a good choice in terms of energy efficiency.
Under this indicator, it further analyzes the changes of energy consumption at the
terminal side as well as the energy efficiency of the entire network after the micro
base station is laid out in the macro cell. And it has obtained the relation between
the terminal energy consumption and the low power station configuration param-
eters. Finally, it also derives the optimal cell radius to obtain the maximum terminal
energy saving and the highest energy efficiency.
In order to verify the rationality and validity of the proposed energy efficiency
indicators, in [24], the energy consumption and energy efficiency indicators have
been simulated after the deployment of different configuration schemes of pico base
stations and micro base stations. In the simulation process, it assumes that the
coverage of the low power consumption base station is always within the coverage
of a macro base station. Table 4.5 provided the detailed simulation parameters.
Figure 4.10 shows the relations between the terminal energy saving with differ-
ent micro cell radius as well as the distance (d ) between the macro base station and
the micro base station. It can be seen that the results of the approximate derivation
and the simulation results are very similar. The figure also shows that with the
increase of the distance between two base stations, the terminal will save more

Macro-cell parameters
Carrier frequency 2.0 GHz
Bandwidth 20 MHz
Pathloss model COST 231-Walfish-Ikegami Model
Cell radius 5 Km
Minimum SNR requirement 10 dB
Noise 160 dBm/Hz
Low power consumption cell parameters
Bandwidth 20 MHz
Pathloss model COST 231-Walfish-Ikegami Model
Minimum SNR requirement 10 dB
Noise 150 dBm/Hz
Micro-cell user parameters
Active user density 10^5 user/m^2
Working state power consumption 1.2 W
Resting state power consumption 0.6 W
Packet size 100 bit
Micromicro-cell user parameters
Background user density 10^5 user/m^2
Hot user ratio 15%
40
theoretical
simulation d = 2900
35
d = 3300 d = 2500
30 d = 3700
UE energy saving(W)
d = 4100
25
20
d = 1700
15
d = 1300
10 d = 2100
d = 900
5
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
micro-cell radius (m)
Fig. 4.10 Relation between terminal energy saving and the micro base station location as well as
the coverage radius
energy under the same micro base station configuration parameters. Under the
simulation parameters, the highest energy saving of the terminal is 35 W, and the
energy consumption of all terminals in the entire network is 860 W. Setting up
micro base stations can bring about 4% of the energy saving for general users. For
users inside the micro cell, the energy saving can be as high as 18%. If active users
need to download more resources, the micro base station can bring more terminal
energy savings.
Figure 4.11 shows energy efficiency value of the network under different micro
base station locations and micro base station radius on the basis of the proposed
energy efficiency indicators. If the energy efficiency value is greater than 0, it is
believed that setting the micro base station is beneficial from the perspective of
energy efficiency. It can be seen from Fig. 4.8 that when the distance between the
micro base station and the macro base station is less than 2100 m, since the
spectrum efficiency of the micro base station is higher when served by macro
station. Even within the optimal microcell radius, the network energy efficiency
value is still less than 0. In this case, the setting of the micro base station is
inappropriate in terms of energy efficiency. The above analysis shows that the
proposed energy efficiency indicator provides a numerical reference criterion for
operators to judge whether a new base station is appropriate in terms energy
efficiency.
10
d = 3300
d = 3700 d = 2900
5
d = 4100
0 d = 2500
5
energy efficiency index
10
d = 1700
15 d = 1300
d = 2100
20 d = 900
25
30
35
energy efficiency index
40
0 200 400 600 800 1000 1200 1400 1600 1800 2000
micro-cell radius (m)
Fig. 4.11 Relation of different micro base station positions and their radius as well as energy
efficiency indicators
Case 3: Tail-Biting Convolution Code Decoder Simulation
The future 5G system not only can greatly improve the traditional cellular
network performance, but also plays an important role in the technology appli-
cations in vertical industries. This trend will become more obvious with the
emergence of the IoT. For example, various communications systems related
with IoT will have much greater demand for short packets. Thus, how to perform
channel encoding and decoding with high efficiency and low complexity has
become an important research direction. In the short packet transmission, short
codes are usually needed for channel coding such as tail-biting convolution codes.
Tail-biting convolution codes usually use Circular Viterbi Algorithm (CVA) for
decoding [25]. However, there is an important problem in the CVA-based itera-
tive sub-optimal decoding algorithm, i.e., the circular trap problem. That is, the
surviving path obtained with the current iteration is the same as the surviving path
obtained before. And as the iterative sequence continues, there is no new tail-
biting path that is closer to the receiving sequence while the decoder cannot
terminate the iteration. The works in [26] and [27] have studied the CVA iteration
process and the circular trap phenomenon existing in CVA, and put forward an
effective circular trap detection scheme. Detection of the circular trap can be used
to help control the CVA decoding process, so as to get the fast-convergent
iterative decoding algorithm.
Table 4.6 BLER performance in the Additive White Gaussian Noise (AWGN) channel for the
suboptimal decoding algorithms (WAVA and S-CTD-II) and the maximum likelihood decoding
algorithms (CTD-ML decoder and two-phase ML decoder) for the tail-biting convolution code
with the generator polynomial of {133,171,165} and (120,40)
SNR
BLER 1.0 dB 2.0 dB 3.0 dB 4.0 dB
WAVA 8.535 102 1.372 102 1.353 103 7.280 105
S-CTD-II 8.598 102 1.377 102 1.352 103 7.280 105
Maximum likelihood decoder 8.466 102 1.363 102 1.352 103 7.280 105
In order to verify the decoding performance and the decoding efficiency, in the
simulation below, the simulation analysis has been made to the performance of
the Circular Trap Detection based ML decoder (CTD-ML decoder)algorithm and
a suboptimal Simplified CTD (SCTD) algorithm. Here SCTD corresponds to the
S-CTD-II algorithm in [27]. Meanwhile, it also compares with suboptimal
decoder WAVA [25] and the hybrid maximum likelihood decoding algorithm
based on Viterbi and the heuristic search [28], etc. The simulation has adopted
the convolution codes with two kinds of lengths respectively. One of the tail-
biting convolution codes adopts the convolution encoder in the LTE control
channel and the broadcasting channel. The generator polynomial is {133,
171, 165} [29] and the information sequence length is 40. This codeword is
denoted by (120, 40).The other code of the tail-biting convolution codes uses the
convolution encoder with the generator polynomial of {345, 237} [25]. The
information sequence length L is 32. This codeword is denoted by (64, 32).In
the simulation, the coded bits perform the QPSK modulation and then enter the
AWGN channel. For convenience, the hybrid maximum likelihood decoding
algorithm based on Viterbi and the heuristic search is denoted by the two-phase
ML algorithm.
Table 4.6 gives the decoding performance of the different decoding algorithm
son the tail biting convolution code (120,40), from which we can find that
suboptimal decoding algorithms, such as WAVA [25] and SCTD decoders, have
close BLER performance to that of optimal decoding algorithms. Maximum like-
lihood decoders refer to the CTD-ML decoder and the two-phase ML decoder.
Since these two decoders are both maximum likelihood decoders on tail-biting
trellis diagrams, their BLER performances are exactly the same.
Figures 4.12 and 4.13 respectively give the relations of the decoding com-
plexity performance of the CTD-ML decoder and the two-phase ML decoder and
the demand for storage space in the decoding process with SNR. From Fig. 4.12
we can see that for the tail-biting convolution codes of any length, the CTD-ML
decoder has higher decoding efficiency than the two-phase ML decoder. Mean-
while, Fig. 4.13 shows the CTD-ML decoder has lower requirements for storage
space. Therefore, the CTD- ML decoder on tail-biting trellis diagrams has
significant advantages in both computational complexity and storage space
reduction.
105
CTD-ML decoder for long tail-biting codes
two phase-ML decoder for long tail-biting codes
average number of states per bit access
CTD-ML decoder for short tail-biting codes

two phase-ML decoder for short tail-biting codes
104
103
102
101
1 1.5 2 2.5 3 3.5 4 4.5 5
Eb /No (dB)
Fig. 4.12 Decoding complexity comparison of the CTD-ML decoder and the two- phase ML
decoder for the long tail-biting convolutional code with (120, 40) and the short tail-biting
convolutional code with (64, 32)
Case 4: Compressive Sensing Simulation
The next generation communications network will produce a large amount of data,
and thus the highly efficient processing of data samples will be one of important
requirements. Compressive Sensing (CS) is an important way of data processing,
which can recover the raw data from extremely few sample values. The algorithm
complexity of the compression process and the recovery process in CS is extremely
uneven. That is to say, the computational complexity of compression sampling is
much lower at the transmitter than that of the receiver. In wireless sensor network
applications, for example, the sensor nodes of sensor networks can compress and
sample data in the form of CS with only low computational cost. Most importantly,
the compressive sensing technology is able to balance the energy consumption of
sensor nodes in the network to solve the bottleneck effect/hot spot effect. So the
CS technology has a great application prospect in the sensor network data aggre-
gation application. The CS technology [30] has aroused great interest among
researchers. Zhao et al. [31] put forward a new type of Treelet-based Compressive
Data Aggregation (T-CDA), and design the corresponding data collection methods.
It is proved that T-CDA is better than traditional CS data aggregation algorithms in
terms of energy efficiency based on the simulation of the real sensor network data.
104
2.5
CTD-ML decoder for long tail-biting codes
average number of required storage space
two phase-ML decoder for long tail-biting codes

CTD-ML decoder for short tail-biting codes
2 two phase-ML decoder for short tail-biting codes
1.5
0.5
0
1 1.5 2 2.5 3 3.5 4 4.5 5
Eb /No (dB)
Fig. 4.13 Storage space requirements comparison of the CTD-ML decoder and the two-phase ML
decoder for the long tail-biting convolutional code with (120, 40) and the short tail-biting
convolutional code with (64, 32)
The simulation uses the data collected from a static sensor network in Ecole
Polytechnique Federale de Lausanne University [32].This sensor network records
the environmental data of 97 test points on the campus every five minutes from July
2006 to May 2007, such as temperature, humidity, illumination, etc. In [33], this
section uses the highly space-time correlated temperature data as the simulation
data. In the simulation, the overall network transmission overheads under different
recovery precisions are used as evaluation indicators for energy consumption.
Simulation has compared the performance of the following methods.
Discrete Cosine Transform-CDA (DCT-CDA) represents the traditional method
which uses DCT to make sparse compressed sensing data aggregation. CS
performance depends on the sparse degree of the original signa. The existing
researches usually use DCT [34] to make the actual signal sparse. However, the
actual signals from the nodes are distributed in the three-dimensional world. It is
difficult to prioritize them and make the data change trend smoothly. Therefore,
DCT can only process smooth signals without fully exploring the actual sensor
signals sparse.
Principal Component Analysis-CDA (PCA-CDA) represents the data aggrega-
tion method with the treelet algorithm in T-CDA being replaced by PCA. PCA is
a typical global data analysis method, and is introduced into the CS data
collection in [35]. PCA can get a sparse transformation matrix of the original
data, which can be used to extract the main ingredients in raw data. But it is a
global operation process, which overwhelms the local properties in the data. In
addition, due to the global properties of the PCA algorithm, noise is also
regarded as useful data. As a result, this method is extremely unstable when
the observation data is small because any change of noise will affect the global
result.
T-CDA represents the proposed treelet-based compressive data aggregation
method. Treelet transformation is an algorithm for data clustering and regression
analysis, which can explore the local correlation characteristics in data. In sensor
networks, to obtain the accurate observation data is a job with large energy
consumption. It is difficult to support such global analysis algorithms like PCA.
And the tree transformation is able to overcome or relieve the situation. In
particular, treelet transformation can be used to utilize the correlation structure
between nodes from the sensor data or training data collected in advance and
then to obtain an orthogonal transformation matrix, which can fully reduce the
training data sparsity. Since sensing data also shows the correlation on time
dimension, the orthogonal transformation matrix can be used to reduce the
sparsity of the sensing data collected within several timing sequences after the
training data.
The simulation results of Figs. 4.14 and 4.15 provide the performance of the
T-CDA algorithm under different system parameters. Figure 4.14 shows the impact
of the selection of training sequence length (L) on the system performance. In the
Fig. 4.14 The impact of the length selection of the training sequence on system performance
Fig. 4.15 The impact of the rounds number of continuous CS data collection on system
performance
figure factor L/, represents the number of rounds for carrying out T-CDA
after a round of training data. As the rough trend shows in Fig. 4.14, the smaller L,
the better system performance. This is because is fixed, which means longer
training sequence corresponds to more rounds of data collection. This suggests that
the major impact of the data recovery error comes from the changes of sensor data
over time, rather than the insufficient information in the training sequence. Fig-
ure 4.15 verifies the above conclusion. From Fig. 4.15, it can be seen that with the
increase of the number of data collection rounds, the accuracy of data recovery
deteriorates. Figures 4.14 and 4.15 result show that L 2 is the optimal value for
the system parameter selection.
The performance comparison of DCT-CDA, PCA-CDA and T-CDA is shown in
Fig. 4.16. By using the actual environment monitoring data, it can be seen that the
performance of DCT- CDA is the worst, and the performance of T-CDA is better
than that of PCA-CDA and DCT-CDA. As a result, we can reach the following
conclusions. Firstly, the method based on the training sequence is better than the
traditional method. Then, the performance of T-CDA is better than PCA transfor-
mation in this application, which means the local correlation of the mining data is
superior to the global correlation.
100
DCT-CDA
PCA-CDA
T-CDA
signal reconstruction error
101
102
2000 4000 6000 8000 10000 12000 14000 16000
transmission overhead
Fig. 4.16 System performance comparison under different data acquisition methods
Case 5: User-Oriented Link Adaption in D2D Network Coding Multicast
With the rapid development of wireless communications technology, smart phones,

tablets and other mobile terminals have become indispensable parts in peoples
lives. Mobile user data requirements increase rapidly, but the available bandwidth
resources of mobile networks are limited. In order to solve the contradiction
between the service growth and limited resources, and to satisfy the proposed
energy-saving green requirements of 5G, D2D has become a key candidate tech-
nology in 5G. As a new kind of short range direct data transmission service based
on cellular mobile networks, D2D communications mainly falls into two scenarios:
the point-to-point D2D communications scenario between a pair of terminals and
the multicast communications scenario of D2D clusters composed of a group of
terminals with the same service requirements.
One big difference between D2D clusters multicast communications and tradi-
tional broadcast communications is the requirements for the reliability in the packet
transmission. So it needs to be guaranteed by technologies like the HARQ. How-
ever, in multicast communications, one users failure of reception will cause the
repeat transmission, leading to the relatively low system efficiency. On the other
hand, in the traditional broadcast/ multicast link adaptation method, the Modulation
and Coding Scheme (MCS) is selected based on the worst link channel quality
between all the receivers and transmitters, leading to the situation that other
receivers with the good channel quality do not make full use of the channel.
Through the research on link adaptive problems in the D2D cluster network coding
multicast scenario, Zhou et al. [36] find that unlike traditional multicast/ broadcast,
the network coding multicast has its special properties. That is, in the network
coding, different information merged via the heterogeneous domain is sent to
different receivers, which brings new possibilities for the adaptive modulation
coding selection for the network coding multicast.
Zhou et al. [36] put forward the user-oriented link adaptive method, in which two
multicast channels whose link quality corresponds to the maximum value of the
modulation type and the minimum value of coding type are chosen. And then
making bit mapping to the terminal with poor link quality at the transmitter to
reduce the equivalent rate. So after the terminals with different purposes make
network coding and decoding to the received multicast data packets, the informa-
tion obtained get different equivalent modulation and coding efficiency to adapt to
the quality of the two different channels. While it should be guaranteed that the
users with poor channel quality can accurately decode the multicast data packets,
the users with good channel quality should be able to obtain more useful informa-
tion in a more efficient modulation and coding type, and eventually the overall
spectral efficiency of multicast communications can be improved.
In order to verify the algorithms BER and spectrum efficiency, link simulations
are carried out to verify the different channel quality combination of two channels
within D2D clusters [36]. Turbo channel coding in 3GPP LTE standards [37] is
used, and the information byte length is set to 512 bits. The combinations of bit
rates and modulation mode are shown in Table 4.7.
Figure 4.17 has simulated the link performance of the terminals BLER and the
overall throughputs of D2D cluster multicast communications. It can be seen that in
the scheme adopted by [36], BLER of the users with poor link quality is basically in
coincidence with the traditional multicast MCS. It shows that the new scheme can
guarantee the users with poor link quality to get the equivalent coding efficiency in
consistency with the traditional way so that they can decode the multicast packets
accurately. Since the users with good link quality won more efficient modulation
and coding types, their overall throughput of D2D clusters improves significantly
by as high as 42%.
Figure 4.18 shows the simulation of the link performance of the overall spectrum
efficiency of D2D cluster multicast communications, with the different combina-
tions of modulation and coding types supported by two channel link qualities in

MCS table of the proposed against conventional scheme
Case MCS of conv link adaptation MCS of the proposed scheme
1 1/2 BPSK 1/2 QPSK
2 1/2 QPSK 1/3 16QAM
3 2/3 QPSK 1/2 16QAM
4 3/4 QPSK 2/3 16QAM
Fig. 4.17 Link performance of 1/2 BPSK and QPSK multicast link quality combination (a) BLER
(b) Throughput
3.5
Conv 2/3
Novel 18% 16QAM
3 3/4 QPSK
1/2
13% 16QAM
2.5
2/3 QPSK
Spectrum efficiency
2 1/3
16QAM
1/2
1.5 QPSK 1/2
QPSK
1/2
1 BPSK 45%
42%
0.5
0
2 1 0 1 2 3 4 5 6
Es/No(dB)
Fig. 4.18 Spectrum efficiency under different MCS combinations
D2Dmulticast communications. With the increase of SNR, new and traditional

schemes adopt higher bit rates and higher modulation methods. Under the same
SNR, the new scheme can support smaller bit rate and higher modulation method.
And the final equivalent modulation coding efficiency is higher than the traditional
scheme. So the overall throughput of D2D clusters has improved significantly. The
simulation results show that the overall throughput has 1345% performance gain
under different modulation encoding combinations.
4.4 5G Software System Level Simulation
4.4.1 Test Evaluation Methods
The first problem of software simulation test is how to evaluate 5G new technol-
ogies. 5G networks have greatly different features from 4G networks. Its evaluation
methods should be studied from the following aspects.
A good evaluation system requires basic features of completeness, simplicity,
better usability, etc. Compared with 4G networks, 5G networks have changed not
only in quantity such as capacity, data rate, and delay, but also in intrinsic
quality, including the basic features of network such as virtualization and
4.4 5G Software System Level Simulation 189
definability. For the comprehensive, accurate and efficient evaluation of 5G net-

work technologies, the study should at least include the following aspects.
Design of new performance indicators
Connection density, traffic volume density, minimum guaranteed rate are
new KPIs introduced by 5G networks. With the deepening of the 5G technology
research, it can be foreseen that there will be new evaluation indicators. The
design of these directly measurable indicators, on the one hand, needs to
combine the characteristics of the new services, and on the other hand, needs
to fully learn from the experience of the previous KPI in network applications.
Redesign of traditional indicators in 5G network
Due to the development of 5G network architecture and services, the conno-
tations of the traditional indicators are changing. For example, the virtual
resources model is introduced after the introduction of the virtual technologies.
How to map traditional indicators to the virtual resources and evaluate network
performance accurately is a new topic in 5G networks.
Compared with 4G networks, there are big breakthroughs and innovations in
some 5G new technologies in terms of the design of the network resources
model, and new concepts and designs are introduced, which are not compatible
with the design in the past. Now end-to-end global design is needed for these
technological models and evaluation methods. For example, the following
technologies:
Network function virtualization
Wireless access network virtualization mainly uses the design idea of SDN
[38] and NFV [39], which is similar to the virtualization of core network and
cable network. There can be three layers, namely infrastructure layer, control
layer and application layer based on the network abstraction of the wireless
access network. This brand new architecture needs to be mapped to the
corresponding network resources model, so that the functions and perfor-
mance of these three layers can be comprehensively evaluated.
Infrastructure layer is supported by various kinds of network equipment nodes,
such as macro eNodeB and micro eNodeB. These equipments provide virtual
management functions through unified interface. Control layer is composed of a
series of distributed management node. As logical nodes, the management nodes
can be deployed flexibly in different types of equipment. The management nodes
are responsible for the management work of the creation, updates, virtual resources
scheduling, allocation and recycling of the virtual network. The application layer is
made up of many different services and applications. These services, scheduled and
allocated by the management nodes, realize specific functions. Control layer shields
the details of the bottom resources to provide top services with unified virtual
resources abstraction through virtual resources unified description. Virtual net-
work construction is based on the services. Different virtual resources are selected
to provide services according to different services and application types, in order to
better adapt to Quality of Service (QoS) requirements of the different services.
For wireless access network side, the main functions of virtual management
layer include: creation and updating of service-centered virtual cell, management of
virtual resources (including backhaul network resources and air interface
resources), connection mobility control, and management and use of context
information. This network architecture introduces the mechanism of the control
and data separation. For example, based on the idea of separation of control plane
and user plane, Ishii et al. present a new network architecture on the basis of the
concept of virtual cell [40].
Two important problems of wireless network resources virtualization are node
mapping that provides different and virtual resource allocation. Node mapping is to
determine what nodes are used for a certain service. In the traditional network, a
service cell is selected for the user according to the signal strength indicators such as
Reference Signal Receiving Power (RSRP) / Reference Signal Receiving Quality
(RSRQ). All the users services are provided by this service cell. Under ultra-dense
network scenario, a large number of irregularly deployed Small Cell will face the
problems of complex interference and load distribution asymmetry. The traditional
way based on signal strength indicators for cell selection will no longer be applica-
ble. For future 5G network virtual cell, the cell can be no longer associated with the
nodes, but instead take user for the center [41]. Reference signals used for data
transmission will decouple with the node ID. Management and control are separated
from data nodes. User-centered virtual cell can be formed at any location in the
network, realizing global virtualization. For management nodes in the virtual man-
agement, virtual resources management is one of its important functions. How to
avoid the interference between each node and make effective utilization of time
domain, frequency domain and space domain resources are the key to the wireless
resource management.
Through the above introduction to the wireless network resource virtualization
technology, we can see the network resources model has undergone great changes.
The virtual cell and virtual resources need to be modeled. Virtual cell needs to
decouple with physical nodes, and provide services with users as the center. So,
different from traditional cells, the performance of virtual cells needs to be evalu-
ated from the users dimension.
4.4.2 Key Simulation Technologies
This chapter describes problems and the corresponding design ideas in the platform
architecture design of the 5G system simulation software. Various kinds of new
technologies simulation design, implementation and evaluation in 5G networks
should be analyzed and carried out separately according to the specific services,
which will not be covered in this chapter. The key technologies of 5G system
simulation design are shown in Table 4.8.
Table 4.8 Key technologies of 5G system simulation design

Key technologies Overview
Dynamic simulation modeling Simulation model library and simulation parameter
technology library are obtained according to simulation scenarios,
services models, service quality requirements, carried
technologies and other kinds of simulation requirement
analysis. Based on the simulation model components
and simulation process of the model and the parameter
dynamic configuration, highly multiplexing of simula-
tion software module and improvement in the simula-
tion development efficiency is realized.
Management technology of Hardware resources, software resources, and commu-
virtualized computational resources nications bus are virtualized into computational
resources, storage resources and transmission resources
for unified management. Meanwhile, users simulation
requirements are converted and broken down into mul-
tiple simulation tasks that can be run in parallel.
Through mapping the simulation tasks dynamically into
different computational resources to implement the
resources dynamic configuration and virtualization
management, so as to improve efficiency and scalability
of parallel computing
Multi-core parallel simulation The software simulation platform design based on
technology multi-core parallel computing includes hardware, oper-
ating systems, software design, model and algorithm
design, and other aspects of the design requirements
Hardware acceleration simulation Hardware module is used to replace the software algo-
technology rithms to take full advantage of the hardwares inherent
fast features, including:
FPGA based high performance hardware acceleration
key technology
Interface and middleware design with combined
hardware acceleration and soft simulation platform
Reconfigurable FPGA hardware acceleration card
design
Real-time transmission technology Mature real-time transmission technologies at present
include:
Real-time signal transmission based on TCP/IP tech-
nology
Real-time signal transmission based on Data socket
Real-time signal transmission based on shared vari-
able engine
Signal transmission with direct optical fiber
Dynamic Simulation Modeling Technology
5G technologies have brought more complex network scenarios and services types,
and also created all kinds of new technologies. It makes the simulation model and
process more complicated and changeable, and leads to the exponential rise in the
number of the simulation scenarios. It is embodied in the following aspects.
Networking scenario is complicated and changeable

The 5G network architecture becomes more flexible after adopting SDN, NFV,
and SON. More combinations of different network function entities come into
being. And network scenarios are also more complicated and changeable. In
order to adapt to these changes, flexible network oriented simulation modeling
technology is needed.
Network model is evolving
The design concept of no more cell has weakened the independent concepts of
traditional link level and system level simulations. The design concept of
virtualization requires network resource model to completely decouple in three
dimensions, namely control plane, data plane, and physical resources. The channel
modeling of massive MIMO and High Frequencies Band (HFB) communications
derive new channel models. All these key technologies have been promoting the
new network model design, and the system simulation also needs to develop
synchronously with the abstraction and design of the network models.
The efficiency of the traditional simulation design pattern which is realized by
encoding for a particular scene is very low, which is far from being able to meet the
increasing requirements. The high multiplexing modeling technique becomes a
must, and thus the dynamic simulation modeling technology is proposed.
The core idea of dynamic simulation modeling technology is network layering
and modeling, and modular design of different layers of simulation object models
[42]. Meanwhile, simulation model components and simulation parameters are
obtained based on the simulation scenarios and services model mapping, and then
through dynamic configuration, the specific simulation process is combined. As the
simulation object model has realized the modular design, the main simulation
design implementation can be fully multiplexed, which on the one hand, has
improved the simulation design and development efficiency, and on the other
hand, enhanced the scalability of the simulation platform.
Dynamic simulation modeling techniques include two key technologies.
(a) Generation function library and parameter library. The basic function modules
of the simulation platform are outputted. Modeling is made according to the
requirements of the simulation. Public library and characteristic library are
abstracted and separated. The well-matched and practical functions are realized
through intelligent interfaces, and the function scalability is met at the same
time. The purpose for separating design of function library from parameter
library is to ensure that the simulation model can adapt to different simulation
scenarios and requirements, and be fully decoupled. The positions of the
function library and parameter library in the system are shown in Fig. 4.19.
(b) Dynamic analysis and configuration mechanism. In the whole running process
of the simulation, this technology provides analysis and configuration mecha-
nism. It includes the decomposition of simulation requirements, mapping into
different functional library and parameter library, and generating simulation
process according to the specific simulation requirement configurations.
Fig. 4.19 Function library and parameter library
Fig. 4.20 Process of dynamic analysis and configuration mechanism
As shown in Fig. 4.20, dynamic simulation modeling process is composed of the

following steps.
According to simulation requirements, the corresponding simulation model is
decomposed. According to the OSI layered architecture, the simulation model is
decomposed into five layers, namely application layer, transmission layer,
network layer, MAC layer and physical layer, each of which has a corresponding
simulation model. In addition to the basic requirements of simulation model
design according to the OSI layered architecture, each new technology of 5G has
put forward more new requirements to the design of simulation model. For
example, SDN requires that control plane and user plane are separated, NFV
requires network functions to be decoupled from dedicated hardware devices,
and the concept of virtual cell is put forward in the UDN, etc. These new
technologies have put forward model design, decoupling, abstraction, resource
virtualization and centralized management, and other design requirements for
simulation model design, which are the key design requirements for the entire
simulation platform.
Simulation parameter library is generated according to the model and the
requirements, including the system specifications and the scenario parameters,
as well as configuration parameters of the key technologies. Since 5G network
models are more complex, and network modes are more abundant, the network
scenarios and configuration parameters involved increase dramatically. There-
fore, when designing the simulation parameters library, we should take the
simulation model as the center. Parameterized templates of networking scenario,
network function and system specification are established based on simulation
model, and with the reasonable combination of these parameter templates the
complexity of the parameter library is reduced.
The corresponding function libraries are mapped by the model. These functions
libraries consist of different functional components. Each component has real-
ized the corresponding model function. Communication interactions between
components are realized through specific function interfaces. Each component
can be allocated to the computational resources and can be managed as an
independent computational unit. Function libraries can realize decoupling and
scalability through the flexible interface design. For example, massive MIMO
technology will use different precoding techniques, and each precoding tech-
nique corresponds to a precoding matrix design algorithm. The input variables
and output variables of different precoding techniques have the same form, with
the input of channel matrix and the output of precoding matrix. Therefore,
different precoding functions can provide a unified functional interface. Realiz-
ing the separation of interface and implementation through the concept of virtual
function is good for the extension new precoding algorithm and the realization of
independent allocation of computational resources. For example, global
precoding matrix algorithm involves the big dimension matrix inversion and
multiplication calculation and it is the key path of the simulation platform
calculation. On the basis of precoding algorithm decoupling design, this function
can be flexibly deployed on GPU or FPGA for acceleration.
According to the simulation requirements, the mapped function library and
parameter library are organically organized to become a complete simulation
process. The design of the simulation process has ensured the consistency of
every step of the simulation on the whole. Simulation processes of different
types are realized through the rational allocation of scenario parameters, fea-
tures, and service model parameters. In UDN scenario, for example, the user
handover process has different implementation scheme, in which different
handover simulation process can be completed by configuring different hand-
over algorithm parameters.
By dynamically configuring the parameter library, function library, and the
simulation process, we can get the specific simulation tasks, which directly
face the users and need to provide friendly configuration management interface.
It can be seen from the above process that the key to dynamic simulation
modeling lies in the design of model, library components and parameters. When
the designs of stratification, encapsulation, and interface decoupling are used to
solve the coupling between the conceptual model and implementation model, the
goal of the minimal impact of the technological change on implementation can be
achieved.
Virtual Computational Resources Management Technology
The computational efficiency in single CPU simulation environment mainly

depends on the frequency of the CPU. However, the current CPU frequency is
close to its limit and cannot be greatly increased, so the main means of enhancing
the simulation calculation efficiency is the parallel calculation, which means to use
multi-core technology to execute the computational tasks in parallel. Since the
computational resources can be distributed on different physical equipment, how to
distribute management resources reasonably has become the core issue.
To solve this problem, the virtual computational resources management tech-
nology is proposed, which can be broken down into three parts.
The simulation requirements are mapped into computational tasks that can be
deployed independently.
Simulation requirements are a systematic working process, which needs to
map the requirements into computational tasks and complete two parts of work:
process decomposition and resources allocation.
Process decomposition: The simulation requirements are decomposed into
several serial tasks. Then these serial tasks are further decomposed into inde-
pendently deployed parallel subtasks.
Resources allocation: According to the characteristics of the parallel com-
puting tasks decomposed by process, corresponding virtual resources are con-
figured. Computational resources and storage resources are distributed to the
locally deployed parallel computing tasks. For computational tasks deployed in
slave nodes we also need to make sure to allocate enough communications
resources to avoid any delay caused by data transmission.
Various kinds of hardware resources can be virtualized into three kinds of virtual
resources.
Computational resources: Undertaking detailed numerical calculation work.
Related hardware devices are CPU, GPU, FPGA.
Storage resources: Undertaking data cache work, completing the calculation
process by cooperating with computational resources. Related hardware
includes memory, hard disk, etc.
Communications resources: Undertaking the transmission of control data and
service data between the computational nodes. Related hardware includes
high-speed bus, shared memory, etc.
Binding the virtual resource dynamically to computational tasks.

Simulation subtasks can only run after bound with the virtual resources.
Dynamic binding can make full use of resources. The virtual resources needed
by simulation sub-tasks need to be assessed in advance. Different simulation
subtasks have different computational complexities and different requirements
for virtual resources. The running performance of the simulation subtasks can be
obtained through the simulation code static analysis and statistical analysis
methods. The resource requirements of each simulation subtasks can be identi-
fied according to the simulation goals. For example, in massive MIMO charac-
teristics, precoding module and interference module have high computational
intensity, which require sufficient CPU or GPU processor resources, while 3D
channel model requires more storage resources.
With the development of the bottom software and hardware platforms, the
available parallel virtualization technologies are relatively abundant [43]. For
example, MATLAB offers parallel tools such as parfor, SPMD and MDCE, and
parallel programming mechanisms which are applicable to multiple develop-
ment languages, such as MPI and OpenMP. For specific simulation implemen-
tation, not only the bottom parallel technology is needed, but also the parallel
design is needed for simulation applications. Its hard to give a universal method
to the parallel design of simulation applications. We need to design special
parallel algorithm according to specific service characteristics, so it is the key
path to realize parallel virtualization for the simulation system.
Multi-core Parallel Simulation Technology
Simulation platform design based on multi-core parallel computation covers hard-

ware, operating systems, parallel technology, simulation software, models and
algorithm design, etc.
From the hardware level, parallel server scheme or high performance host
system can be chosen to support higher computing power. Parallel server generally
uses the following architecture types: Parallel Vector Processor (PVP), Symmetric
MultiProcessor (SMP), Massively Parallel Processor (MPP), Distributed Shared
Memory multiprocessor (DSM), and the Cluster of Workstation/PC (COW). Par-
allel server has strong parallel processing capability, but the cost is high. The high
performance host system based on high-speed Internet network is able to realize
high performance computing with relatively small cost. The system can consist of
several multi-core servers and high-speed infiniband switch.
Operating system allocates process, storage and other hardware resources for
parallel tasks, realizing the inter-process communications.
Parallel technologies generally include messaging, shared storage, and data
parallel.
Messaging: The asynchronous parallel operation mode and the distributed data
storage mode are adopted with good scalability. Typical examples include MPI,
PVM, etc.
Shared storage: The asynchronous parallel operation mode and shared data
storage mode are adopted with poor scalability. Typical examples include
programming models like OpenMP.
Data parallel: The loose synchronous parallel operation model and shared data
storage mode are adopted with medium scalability. A typical example is HPF.
HPF provides annotation-like instructions to extend the variable types so that it
can control the data layout of the array in detail.
The parallelization of simulation software is the key work of multi-core parallel
design of simulation platform, which needs to consider the following design
requirements.
Simulation software is decomposed in parallel from the aspects of function,
algorithm and operands. Serial simulation tasks are decomposed into fully
decoupled and independently-processed subtasks. Since different functions,
algorithms, and operands have different computational intensity and character-
istics, in parallel design we need to analyze and process them according to actual
situation.
The reasonable division design of simulation function modules can reduce the
communications data between parallel subtasks. It tries to ensure that the
equivalent calculation amount of each parallel subtask. And it can reduce the
waiting time needed for synchronous processing.
As the evolution of the multi-core CPU parallel processing, the heterogeneous
scheme of CPU + GPU has made it possible to improve the speed of the simulation
calculation [1]. This is because that the CPU, as a general-purpose processor that
has complex control logic, is good at complex logic operations. However, GPU is
graphics processor, which often has hundreds of stream processor cores. Its design
goal is to realize large throughput of data parallel computation with a large number
of threads. Its single precision floating points computing power can reach more
than 10 times of CPU, which is suitable for processing parallel computing of large
scale data.
Therefore, adopting the heterogeneous parallel architecture of CPU + GPU, in
which the two are coordinated and multi-core parallel CPU makes complex logic
calculation while GPU does data parallel tasks, can perform the maximum com-
puter parallel processing capabilities. The worlds fastest computer Tianhe-II
currently adopts CPU + GPU heterogeneous polymorphic architecture.
Programming of GPU generally adopts CUDA architecture, under which the
computational tasks are mapped to a large number of threads that can be executed in
parallel. Its organization form is the Grid-Block-Thread 3 levels. And it has the
hardware dynamic scheduling and execution.
Figure 4.21 shows a typical heterogeneous multi-core architecture. It can be seen
that multi-core CPU uses OpenMP while GPU uses Compute Unified Device
Architecture (CUDA) for processing. And task division is specified by program
and operating system level. The two parts are interconnected with PCIee bus.
OpenMP CUDA
Muti-core CPU GPU
core core core core SM SM SM

L1 cache L1 cache L1 cache L1 cache SP SP SP SP SP SP
L2 cache L2 cache L2 cache L2 cache SP SP SP SP SP SP
SP SP SP SP SP SP
SP SP SP SP SP SP
L3 cache Shared Shared Shared

memory memory memory
Main memory Global memory
PCI-E
Remark: 1. SM (stream multiprocessor)

2. SP (stream processor)
Fig. 4.21 Heterogeneous multi-core architecture
Multicore CPU-GPU heterogeneous platform has multiple layers of parallel

execution ability in task level and data level. In load distribution, the
computing power should be fully utilized. The simulation model structure shown
in Fig. 4.22 can be considered.
The models mapping process has the following three levels.
(a) Mapping from simulation model instance to logical process
The map divides the simulation examples into several groups, with each
group assigned a logical process. Through this division, the load of each logic
process should be balanced, and the communications overheads between each
logical process should be as small as possible.
(b) Mapping from logical process to thread
Thread is the execution unit of the software hierarchy while the GPU thread
organization is performed by cores. The computational load of each logical
process is firstly divided into the CPU thread. Then the CPU thread takes the
responsibility to assign the tasks that are suitable to be executed on GPU to
GPU. Since GPU allows performing multiple cores at the same time, the cores
under different logical process mappings have distinct task parallelism. The
process can adopt the way of 1 to 1 or 1 to more.
Group1 Group K
Simulation
model
instance
1
Logical Logical Logical

process 1 process K process
2

L L
Thread
GPU core CPU thread CPU thread GPU core
1 m
3

Processor
GPU SMs CPU core 1 CPU core p GPU SMs
core
Fig. 4.22 Simulation model structure
(c) Mapping from thread to processor cores

This mapping is about how software execution unit is scheduled to hardware
for execution. In order not to disrupt the operating systems optimization
scheduling strategy, the operating system can take full charge of the scheduling
from CPU thread to CPU execution core. Multiprocessor scheduling from GPU
core to GPU stream is charged by CUDA bottom drive.
It can be seen from the above analysis that the final acceleration capability of
multi-core parallel simulation technology is influenced by hardware, parallel
processing mechanism, system transmission bandwidth, parallel granularity of
the simulation application and many other factors. Evaluating parallel comput-
ing power of the whole simulation system is a system level problem. Due to the
rapid development of the 5G technology, the simulation requirements change
quickly, and the simulation application programs will also be constantly
updated. How to quickly and accurately assess the overall simulation systems
computing bottlenecks and provide the corresponding parallel solutions
become the key issues in the multi-core parallel simulation technology. Wang
et al. [44] present a parallel computing performance evaluation method based
on equivalent statistical model. They propose the equivalent statistical model
with double transform domain, in which the target program is continuously
transformed from serial calculation domain to the equivalent serial computing
domain and equivalent parallel computing domain. Then a few comprehensive
indicators are derived based on Principal Component Analysis (PCA) and
statistical methods so as to construct the equivalent calculation model. Thus,
the optimal parallel scheme is obtained by using this model to achieve the
Parallel computing evaluation system

Serial Parallel
Performance Performance
application Data analysis application
modeling evaluation
system module system
module module
Output performance statistics
Analyse performance statistics and

extract performance characteristics to get a
unified comprehensive performance
indicators
Based on the comprehensive

performance index, the equivalent
serial program is constructed.
According to the parallel

optimization program, the serial
equivalent program is transformed
into parallel equivalent program
Run parallel equivalent

program
Output performance statistics
Performance evaluation based on performance

statistics, optimization objectives, etc.
Select other parallel optimization schemes, jump

to step complete assessment of all required
parallel optimization schemes.
Output evaluation report
Fig. 4.23 Performance evaluation process for parallel computing
performance optimization goals. After measurement, with this method the

prediction errors for simulation resource and time consumption are less than
10%. The basic working process of this method is shown in Fig. 4.23.
Hardware Acceleration Simulation Technology
Hardware acceleration simulation technology uses hardware modules rather than

the software modules to make full use of the inherent fast hardware features. The
hardware acceleration simulation technology has different focus from the
hardware-in-loop simulating technology, with the latter focusing on the real-time
implementation of function features and the former aiming at solving system
bottlenecks and accelerating the overall calculation speed. Generally, hardware
uses high performance FPGA board which has strong computing power and logical
processing ability. FPGA board has stronger floating point computing power than
CPU server and stronger task management, resource scheduling and other logic
handling abilities than GPU server. With the rapid development of FPGA develop-
ment environment and compiler in recent years, the development difficulty for
FPGA board is reduced greatly. The main research points of hardware acceleration
simulation technology can be divided into the following three aspects.
1. Studies on key technologies of FPGA based high performance hardware
acceleration
With the development of integrated circuit technology, the computing power
and storage resources of FPGA chip grow rapidly. Meanwhile, the FPGA has
become the ideal algorithm accelerating implementation platform due to its
programmability, powerful parallel ability, rich hardware resources, flexible
algorithm adaptability and lower power consumption. FPGA can easily realize
the distributed algorithm structure, which is very useful for acceleration scien-
tific computing. For example, most matrix decomposition algorithms in scien-
tific computation require a lot of Multiply and ACcumulate (MAC) operation.
FPGA platform can effectively realize the MAC operating through distributed
arithmetic structure. Todays FPGA products have entered the era of 20 nm
\14 nm. Major companies are working on developing logic products with more
logical units and higher performance.
Research contents include high speed parallel processing, hardware and
software simulation task partition and mapping, high precision signal
processing, etc.
2. Interface and middleware design with the combination of hardware acceleration
and software simulation platform
For some calculations that are hard to complete by FPGA, we can transfer it
to C language. At the end of the design, the intermediate results from the
hardware are imported into C for visual analysis. Through drawing graphical
waveform and eye diagram, etc., we can check whether each part of the system
design is correct. Introducing the simulation based on C/C++ to the design of
FPGA has solved the FPGA debugging difficulty, as well as the frequent
inconsistency with the system simulation logic.
Research contents include virtual adaptation mechanism, middleware design,
and reusable calculation design.
3. The design of reconfigurable FPGA hardware accelerator board
Reusable programming is an important characteristic of the hardware accel-
eration based on FPGA platform. The FPGA based system can be altered or
reconfigured in the development and application phase arbitrarily, which will
increases the overall gain.
Research contents include: high speed PCIe interface design, data interaction
between high-speed USB 3.0 interface and the host.
The implementation process of hardware acceleration simulation is as
follows.
The first is the key technology research. Starting from the simulation design
requirements, we should fully combine the characteristics of the hardware
parallel processing and software system architecture to reasonably divide and

design hardware acceleration module, so as to significantly reduce the time
overheads of simulation core module calculation.
Then the FPGA based hardware accelerator card system is designed, involv-
ing the C simulation code based adapter, the design of middleware on
interface layer (including board support package) and the design of
reconfigurable calculation.
When the host receives the signals indicating that the configuration data is
loaded, the initial data is sent into memory via PCIe interface. When the
initial data is sent, launch signal is sent to accelerator through PCIe interface.
When the accelerator completes the processing, accelerated processing com-
pletion signal is sent to the host through PCIe interface. When the host
receives the acceleration processing completion signals, the final results are
read from memory.
As shown in Fig. 4.24, on system simulation platform, some links use
hardware implementation. The actual test results of the delay of this links
physical bus transmission are within 10 ms. Physical bus transmission delay is
defined as the delay of the whole process when data is distributed from system
System simulation
platform
Control plane: parameter Data plane: business
configuration, scheduling, flow generation,
command interface transmission, feedback
High speed bus

Link level processing module for hardware
implementation
Cell N
TX RX UE N
Cell 1 UE 1
User Up
baseband DA RF/down
conversion AD decoding
data conversion
to antenna
Baseband synchroniz demodulati

framing ation on
FPGA resource set1 FPGA resource set2

Air interface
Air interface
Channel
UE N
UE 1
Cell N UE N
User time-domain UE 1
Cell 1 channel
parameters convolution Up conversion to
Down conversion to
baseband RF
User channel
factor
generation
FPGAresource set3
Fig. 4.24 System implementation diagram of software and hardware co-simulation

simulation platform, sent to the air interface through the link transmission
module, and then through the link receiving module, received, demodulated,
and decoded, and then sent back to the system simulation platform.
This kind of software and hardware co-simulation method can fully use the
hardware high-speed processing capabilities, and enable some links system
simulation performance to be close to real-time. Combined with relatively
perfect system function of the system simulation platform, it can better simulate
the system application scenarios with high indicator requirements for system
transmission delay. The implementation scheme of patent CN101924751B [45],
for example, is the entire network solution scheme across the core network and
access network, ranging from the physical layer of base station to every layer of
core network protocol stack. The handover process is at millisecond-level scale,
which is difficult to verify on the old system simulation platforms. Adopting the
method of software and hardware co-simulation platform can well support
system function simulation test at millisecond-level small scale.
Real-time Transmission Technology
Network architecture can extend hardware processing ability of the simulation

platform through the way of extending server node. But in parallel computing
solution scheme of multi servers networking, the data exchange delay between
the servers becomes the key bottleneck to affect the simulation calculation speed
[1], which requires the corresponding solutions.
After the high-strength computational tasks are parallelized, the computational
time of each independent parallel subtask becomes shorter, which is usually within
hundreds of microseconds. Calculation results of each subtask in each calculation
cycle need to be collected to the center module for centralized processing. After
completing the previous cycle, the calculation of the next cycle is carried out. The
collection process needs the data interaction between the servers. If data interacts
via IP switches,, the data exchange delay between the servers is commonly in
millisecond due to the bus mechanism as well as the TCP/IP protocol stack
processing overheads, which has exceeded the calculation time of subtasks. This
will greatly reduce the overall efficiency, and directly influence the final parallel
effect, so it is necessary to provide low latency server data sharing interactive
solution. In order to reduce the data interaction delay between servers brought by
server cluster and make sure that data between the multiple simulation nodes can
share with each other in real-time, reflective memory technology can be used to
realize memory sharing mechanism between multiple servers or multiple servers
transmission networking based on high-speed InfiniBand switches. Reflective
memory network is a special type of memory shared system, which aims at sharing
common data sets between multiple independent computers. Reflective memory
networks can store the independent backups of the whole shared memory in each
subsystem. Each subsystem shall have full and unrestricted access, and can also
modify local data sets with extremely high local memory writing speed. The
transmission delay between reflective memory cards is up to 400 nanoseconds.
4.4.3 Introduction to Simulation Cases
Case 1: Massive MIMO System Level Simulation
This section uses massive MIMO 5G key techniques as an example, and suggests
that how to apply the key technologies described before to complete the design and
implementation of massive MIMO technology in simulation system, so as to reduce
the simulation calculation complexity and accelerate the simulation calculation
speed.
1. Simulation scenario description
Simulation parameter description
Massive MIMO uses the MU-MIMO model to simulate LTE downlink
system performance. The channel matrix of MU-MIMO is formed with
128 base station transmitting antennas, 1 UE receiving antenna, 15 users
being scheduled in a single cell at the same time. Detailed simulation
parameters can be seen in Table 4.9.
The statistics indicators of simulation output mainly include the average
cell spectrum efficiency, 5% cell edge spectrum efficiency, UE downlink
SINR and some other indicators.
Simulation environment description
Hardware: GPU server XR-4802GK4
CPU configuration: two Intel Xeon Ivy Bridg E5 (3.0 G, single 10 cores,
20 threads)
GPU configuration: 8 pieces of TESLA K20
CPU memory: 256GB
Bus: PCIe 3.0 16
Software:
WINDOWS SERVER 2008R2
MATLAB R2014a
2. Simulation design analysis
(a) Functional procedure
Simulation process of MU-MIMO is shown in Fig. 4.25. The main
features of MU-MIMO are embodied in the following function modules.
Transmitter module: It includes CSI feedback, pilot pollution
suppression, antenna resource allocation, user scheduling, and transmitter
precoding.
Transmitting: Wireless channel modeling, including 3DUma, 3DUmi,
3DUMa-H channel.
Receiver processing: It includes interference calculation, SINR calcula-
tion, and receiving detection algorithm.
Table 4.9 Simulation parameter settings

Parameter Value
Base station antenna numbers 128
Base station antenna polarization 45 double polarization
modes
Terminal antenna numbers 1
Terminal polarization modes Perpendicular horizontal dual polarization
Access user numbers of each sector 30
Networking type Same frequency in three sectors, 19 stations, 57 cells,
30 users in each cell
Channel model 3DUMa
Interval between horizontal 0.5 wavelength
antennas
Interval between perpendicular 0.5 wavelength
antennas
Moving speed 3 km/h
Precoding algorithm sent by multi- Zero Forcing
ple users
Terminal receiving algorithm LMMSE
Enabling of ideal uplink channel Yes
estimation
Enabling of ideal downlink channel Yes
estimation
Constraint type of transmitting Consistency of total transmitting power
power
Pilot pollution modeling No pilot pollution
Scheduling method Proportional fairness
Transmitting power Total transmitting power of the base station is 40w
Simulation bandwidth 10 M
Terminal antenna gain 0dBi
Single array gain 8dBi
Antenna gain Determined by the mapping type from the antenna
elementto the port
Thermal noise level -174 dBm/Hz
eNB noise figure 5 dB
UE noise figure 7 dB
(b) Computation analysis

According to the simulation parameter settings and MU-MIMO features,
calculations are mainly distributed in the channel calculation, precoding
calculation, and receiving SINR computation. Set subcarrier number to be
Nc, OFDM symbols M, the number of base station antennas Nt, users (single
antenna) number Nr, receiving antenna number Nr, and the cell number in
the system C.
System parameter configuration
Generate network topology
Place terminal
Initialize the frequency planning
Initialize each system function

module
Calculate channel
Generate service message packet
Schedule cell resources

TTI + 1
Transmitter precoding
Calculate signal to noise ratio
Receiver processing
Protocol stack state migration
Current TTI is the end of the NO

simulation time?
YES
End of simulation
Fig. 4.25 MU-MIMO simulation process

3D channel
It includes the path loss calculation of large scale, shadow and small
scale. Because the number of transmitting and receiving antennas
increases significantly compared with 4G, and the antenna and the channel
parameter models are more complex, the computation has a huge increase
compared with the commonly used TU channel in 4G. Lets analyze only
the times of FFT transformation calculation from time domain channel to
frequency domain channel. If 3D channel FFT transformation calculation
in a cell is about M Nt Nr, then the calculation amount ratio is
(128 15)/(2 1) 960 compared with the scenario that antenna scale
of 4G is 2 1in the case that downlink antenna scale is 128 15. The
calculation amount ratio increases linearly with the product of transmitting
antenna number and receiving antenna number.
According to the simulation parameter settings, the transmitter
precoding scheme is forced zero algorithm, and precoding matrix calcu-
lation formula is WZF H(HHH)1 , H 2 CNtxNr. Precoding computa-
tional complexity mainly lies in the matrix multiplication and inverse.
Under the condition of forced zero algorithm, the main calculation
includes two parts. The first part is C Nc times Nr Nr dimensional
matrix inversion. The second part is multiplication of C Nc times
Nt Nr dimensional matrix and Nr Nr dimension matrix. The algorithm
complexity of different kinds of matrix calculation is usually O (n ^ 3).
Compared with antenna scale with 2 1 scenario in 4G, the calculation
ratio of matrix inversion is 15^3/1^3 3375. The calculation ratio of
matrix multiplication is (128 15 15)/(2 1 1) 14,400, which
increases along with the three times power of antenna number.
SINR calculation
For the calculation of this part, the users received signal power, the
interference power in the cell, interference power between cells are calcu-
lated based on the channel matrix and precoding matrix, and then the
users SINR is obtained. According to the following MIMO signal
model, the size of calculation can be roughly obtained.
pXp H X
L K
dl
yjm dl l hljm wlk xlkdl njm
dl
l1 k1
p H p H X
K
p X L p H X K
dl j hjjm wjm xjm
dl
dl j hjjm wjk xjkdl dl l hljm wlk xlkdl
k1, k6m l1, l6j k1
|{z} |{z}
intracell interference intercell interference
dl
njm
Calculating the signal power of an UE in a cell requires two times Nt

dimensional vector multiplication; the number of multiplication is 2Nt + 1;
the number of multiplication getting the interference power (including

interference in the cell and between cells) needs is C(2Nt + 1); the
total number of multiplication is (C + 1) (2Nt + 1). Compared with
antenna scale 2 1 scenario in 4G, the calculation amount ratio is
(2 128 + 1)/(2 2 + 1) 51.4.
It can be known from the above analysis that there are big differences of
MIMOs calculation amount in different modules, which mainly lie in
channel calculation, transmitter precoding and receiving SINR calculation
modules. Among the modules, transmitter precoding has the largest com-
putational amount of 3 power of antenna number, followed by channel
calculation. According to the law of Amdahl, these two modules are also
the key optimization goals of computational acceleration.
(c) Optimization scheme
According to the characteristics of the different modules, acceleration
optimization scheme can be made combining the previous key technologies.
Channel computational
Given simulation parameters, wireless link channel coefficients have
nothing to do with behaviors like system scheduling, so the channel
calculation can be done in advance, and the results can be stored in hard
disk. When simulation system is initialized, the channel matrix stored in
hard disk can be read directly, and then be stored in memory. There is no
need to calculate the channel matrix in the process of the simulation. Using
the pre-computed results directly saves the channel calculation time. If
memory is big enough, the actual time overheads only depend on the
memory reading time, and the channel computational time can be
neglected.
Transmitter precoding mainly involves large matrix multiplication and
inverse calculation, during which multicore computing capability of CPU
and GPU can be fully utilized to compute in parallel in the subcarrier level.
First CPU parallel computing capability is used to do parallel processing in
the subcarrier granularity. Different sub-carrier precoding calculation
work is allocated to different GPU boards for parallel computing Matrix
inversion and multiplication calculation are completed by each GPU. As a
result of the combined parallel computing of CPU and GPU, and parallel
computing on the granularity of subcarrier, it can parallel to C Nc paths
in maximum. In this test example, it can parallel 57 300 17,100 roads.
For enough GPU cores, considering the influence of the transmission
bandwidth, it can meet the requirements of acceleration optimization
requirements of the precoding in transmitter. In this test example, since
only one GPU server is adopted and CPU and GPU hardware computa-
tional resources are limited, the actual acceleration effect is limited by the
hardware resources. In multi servers networking and super computer
environment, the acceleration effect can be further improved.
SINR calculation
This part of the calculation is mainly the vector multiplication. Its
calculation amount is much smaller than channel calculation and
precoding module of the transmitter. So, it is able to obtain better results
with the help of CPU acceleration.
3. Simulation measurement results and analysis
The simulation measurement results are shown in Tables 4.10 and 4.11.
According to the calculation characteristics of different modules, the final
acceleration effects are also different when using different acceleration program.
SINR calculation module and message processing module adopt the
CPU parallel computing scheme. Precoding module adopts CPU + GPU
co-acceleration scheme. The acceleration ratio of the interference module is
less than that of message processing module from the perspective of acceleration
ratio. Its because interfere module needs to transfer large amounts of data
between the parallel computing tasks, including signal power, interference
power, SINR, channel allocation, scheduling information and other data, most
of which are vast data of subcarrier granularity. The time overheads on data
transmission is greater than that on message processing module, so its acceler-
ation ratio is smaller than that of the message processing module. The further
optimization of SINR calculation module includes increasing the number of
parallel CPU cores, compressing transmitted data, increasing the transmission
bandwidth (high-speed optical fiber transmission and memory reflection tech-
nology, etc.) and other schemes. Precoding module adopts CPU + GPU
co-acceleration scheme, whose acceleration ratio can reach 127 times. Due to
the limitation of hardware resources, the acceleration effect of this part is far
from upper limit. The above acceleration effect is measured on a single GPU
server. Because the parallel granularity of various software modules are much
more than the number of the server processor units, the parallel acceleration can
also still be greatly improved after improving the capacity of the hardware
Table 4.10 Computational time overheads of each MU-MIMO module in serial scheme and
parallel scheme
Other modules
Total length SINR Message (with no parallel
Computational of one TTI calculation processing Precoding optimization)
scheme (second) (second) (second) (second) (second)
Serial 9338.72 772.64 124.88 8411.95 29.25
Parallel 153.92 50.8 4.12 66.03 /
Table 4.11 Acceleration ratio under different modules of MU-MIMO

Measurement results Total acceleration SINR Message
of acceleration ratio ratio (serial/parallel) calculation processing Precoding
Acceleration ratio 60.67 15.2 30.31 127.4
computational platform. For example, the acceleration ratio of precoding mod-

ule, SINR module and other modules can be further enhanced in super computer
environment.
Case 2: Radio Resource Optimization of Ultra Dense Network
With the increasing number of base stations near the small base station in ultra-
dense network scenario, the interference will also be more serious. In particular, the
site number to be coordinated will also increase in resources allocation, making
resource allocation more difficult. The site deployment in hot spot areas shows a
trend of high density and no programming. With more and more deployed sites,
how the system performance changes in hot spot area is rather important for both
users and network operators. Liu et al. [46] simplify the resource allocation under
the ultra-dense deployment scenario through network clustering. Assuming that
intra-cluster channel is orthogonalized, it puts forward a semi-distributed resource
allocation algorithm based on clustering, which reduces the resource allocation
problem under the dense deployed scenarios. A semi-distributed resource allocation
algorithm based on clustering mainly includes the following three stages.
The above optimization problem is a nonlinear mixed integer planning problem.
The resource allocation in ultra-dense deployment scenarios is simplified based on
the assumption of network clustering and intra-cluster channel orthogonalization.
Then a semi-distributed resource allocation algorithm based on clustering is pro-
posed to reduce the resource allocation problem under the dense deployment
scenarios. A semi-distributed resource allocation algorithm process based on clus-
tering is shown in Fig. 4.26, which mainly includes the following three stages:
In the first stage, the network clustering algorithm based on Breadth-First Search
algorithm (BFS) is proposed, where the performance loss brought by the interfer-
ence is taken as the similarity parameters of the site, network clustering is carried
out based on BFS and the cluster heads are selected;
In the second stage, it introduces the concept of reference cluster to approxi-
mately estimate the received interference of each cluster for clustered network. In
the information interaction between cluster heads, it selects the cluster with the
strongest interference as a reference for a given cluster, and only takes into
consideration the interference of the reference cluster in the next phase of resource
allocation. Introducing the concept of reference cluster can reduce the information
interaction of channel state information during the channel power control and
reduce the complexity of the resource allocation problem by obtaining reasonable
suboptimal solution. In addition, the clustering algorithm in first stage assigns the
cells with strong interference to the same cluster. Combining the assumption of
orthogonal channel within the cluster, it can guarantee that the reference cluster
with the strongest interference is a reasonable approximation for all interference
sources.
In the third stage, solving the sub-problems of resource allocation for cluster and
reference cluster, and two-step iterative heuristic resource allocation algorithm is
Fig. 4.26 Process diagram

of half distributed resource The first stage: network
allocation algorithm based clustering
on clustering
The second stage: select the

reference cluster
Third stage: cluster

resource allocation
Step1: Distribution of intra cluster
channel
Step2: adjust the power of the

intra cluster channel
NO
Convergence?
YES
Output
proposed. The algorithm uses heuristic Algorithm 3 to obtain channel allocation,

and then uses the power allocation algorithm based on binary search to adjust
power. It is iterated repeatedly until the system convergence.
In the simulation scenario, the system contains 16 small Base Stations (sBS) in
the 400 m 400 m indoor hot spot area. The radius of sBS is 100 m; each small
base station has 16 available sub-channels; each of these small stations has 4 users
UE randomly distributed in the coverage area of sBS. Channel gain includes large
scale fading caused by path fading and shadow effect. Path fading model uses the
ITU PED-B path loss model with PL 37 + 32log10(d ) [47], where d is the distance
between sBS and user UE. To ensure that the solutions of the problems do exist,
each user uses the smaller QoS constraints. Except special instructions, other
related simulation parameters are shown in Table 4.12.
Figure 4.27 has compared the system performance in scenarios with different
cluster sizes considering all the interference (Interference Case 1) and estimating
interference with reference cluster (Interference Case 2). The result of Fig. 4.27 is
the site average simulation results for many times. In each simulation, the clusters
with different sizes are realized through selecting different and r for 16 sBS within
the area. The approximate simulation results of reference cluster interference are
obtained by estimating the interference in accordance with the reference cluster
concept in this chapter; and then subchannel allocation and power control are made.
Table 4.12 System Parameter Value

simulation parameters of
resource allocation based on
clustering Sub-channel bandwidth 180 KHz
Available channel number 16
Minimum user speed 10kdps
Noise spectral density 174 dBm/Hz
Indoor shadow fading standard deviation 4 dB
Base station maximum transmitting power 15 dBm
4
Interference Case1
3.5 Interference Case2
Interference Case1
Interference Case2
3 Uncoordinated
Average Data Rate (Mbps)
Cluster Size=16
2.5
Cluster Size=2
2
1.5 Cluster Size=4
0.5
0
1 1.5 2 2.5 3 2.5 4
Number of UEs per sBS
Fig. 4.27 System performance comparison between interference estimation through reference
cluster and including all the interference
After multiple iteration, the solution of power and channel allocation is obtained.
And then the actual average rates of each user in the computational system are
obtained through computing all the interference. Simulation results including all the
interference have considered all the interference of the site to solve the power
allocation in iteration. The average rate of each user within the computational
system is also the actual average rate of each user including all the interference.
Simulation results show that reference cluster approximation and the way including
all interference have a very similar system performance. But the way using refer-
ence clusters to estimate the interference is better in approximation than including
all interference with the increase of the cluster size,. This is because the site number
within the cell is fixed, and the cluster number decreases within the cell when the
cluster size increases. In accordance with the concept of reference cluster, the
cluster including the strongest interference has a better interference approximation

for a cluster. In particular, when the cluster size is 16, the way of including all the
interference and the way of estimating interference through reference clusters have
the same result. Its because all the base stations are in the same cluster, and the way
of estimating interference through reference cluster is equivalent to that including
all the interference, thus the results are the same.
In the simulation shown in Fig. 4.28, the 16 sBS sites within the area are in two
areas with different sizes respectively: 100 m 100 m and 60 m 60 m. The user
number changes within in sBS from 1 to 5. When the number of in-cluster users is
more than the number of subchannels within the cluster, simple Round Robin
Scheduling is adopted. Fig. 4.28 has compared the performance of the proposed
algorithms in this chapter, the Semi-Definite Programming Based algorithm (SDP
Based) and each sBS uncoordinated resource allocation algorithm. In this figure,
user average rates are all obtained by calculating all the system interference in the
channel allocation and power allocation schemes obtained by using each algorithm.
The interference estimation of reference cluster is only used in solving power
allocation.
Figure 4.28 shows that the average rate of each user decreases with the increase
of users within sBS site. This is because that the resource can be allocated in Round
Robin Scheduling decreases with more users. Further, it can be seen from the figure
that when the algorithm of allocating resources independently in each cell is
adopted, the system performance is the worst with severe interference since the
20
SDP Based
18 Proposed Algorithm
Uncoordinated
16 SDP Based
Average Data Rate (Mbps)
Proposed Algorithm
14 Uncoordinated
12
10
2
1 1.5 2 2.5 3 3.5 4 4.5 5
Number of UEs per sBS
Fig. 4.28 The user performance variation when adopting different ultra-dense network resource
allocation algorithms in different site densities
interference from other sites is not considered in resource allocation. Besides, with
the increases in the site density, compared with 100 m 100 m (the lines with
hollow markers) and 60 m 60 m (the lines with solid markers) areas, the system
performance is even worse. The algorithm proposed in [48] can get better system
performance, but with high complexity. The algorithm proposed in this chapter has
the similar performance with that proposed in [48]. Through the above theoretical
analysis, we can know that our algorithm has lower complexity. Figure 4.28 in fact
verifies that the proposed algorithm can guarantee approximate optimal system
performance with lower complexity.
Case 3: Statistical Modeling Simulation of the Uplink Interference [49]
In the Orthogonal Frequency Division Multiple Access (OFDMA) -based fre-

quency multiplexing network, inter-cell interference has become one of the key
factors that restrict the improvement of system performance. Research shows that
the uplink power control for the user terminals in the cell, on the one hand, can
reduce the uplink interference between cells and improve the system uplink fre-
quency spectrum efficiency; on the other hand, can reduce the energy consumption
of user terminals and bring power saving effect for terminals. However, in the past
network deployment, it is often difficult to choose appropriate control parameters
and only conservative settings can be made according to limited experience. In fact,
in order to make parametric statistical modeling analysis on the interference
characteristics between cells, we can analyze the impact of the system parameter
setting on the interference between cells and the system performance according to
the constructed model. Therefore, in the uplink open loop power control scenario, it
is necessary to build an uplink interference parametric statistical model so as to
provide theoretical guidance for the design and optimization of power control
scheme.
After the parametric statistical modeling of PDF for the uplink interference
between cells is obtained through theoretical analysis, firstly the statistics is carried
out through a large number of samples produced by Monte Carlo simulation. Then
the samples are compared with the results of the theoretical derivation to verify
whether the theoretical derivation is correct. Secondly, the simulation results can be
used to display the impact of system parameters on the statistical distribution of the
interference between the uplink cells.
System Model
Considering the open-loop uplink power control in the terminals, the system model
of the uplink interference between cells is shown in Fig. 4.29. The model is
established through the location relationship between the nodes in polar coordinate
system. eNB1is located in the origin point of coordinates system, which is (0, 0).The
cell coverage area is a circle with the radius R. UEs are distributed randomly
UE (r , q )
eNB0 D
eNB1
( 0, 0)
( 3R, p ) Cell radius
R
Uplink signal Uplink interference

Fig. 4.29 Open loop power control uplink interference scenario
uniformly in the eNB1 service region, with the coordinates of the location as (r, ).
eNB0 ispthe
interfered adjacent cell base station;
p the
distance between eNB0 and
eNB1 is 3R; the position in coordinates is 3R; .The distance between UE and
the interfered base station eNB0 is D.
Base station location is fixed with users distributed randomly and uniformly,
r
f r; 2 :
R r min 2
The transmitting power of UE is Pt min(Pmax, P0 + PL), where P0 denotes the

nominal power, and denotes the open loop path loss compensation factor.
PL represents the UE uplink signal link propagating loss, including path loss and
shadow shading: PL A Blog10 r XUE-eNB1 , where A and B are path loss
model parameters; r is the distance between UE and eNB1. XUE eNB1 is the
shadow shading that UE uplink experienced.
Interference link propagation loss is PLUE-eNB0 A Blog10 Dr; XUE-eNB0 ,
where the distance of interference link is denoted as
p q
q p
Dr; r 3R 2 3Rr cos r 2 3R2 2 3Rr cos :
2 2
Shadow shading: XUE-eNB1 aXUE bXeNB1 ; XUE-eNB0 aXUE bXeNB0 ,

where a2 b2 1/2.
Assume that the terminals maximal transmitter power threshold is set high, and
then the uplink interference in Log-Domain is:
1
I P0 1A Blog10 r Blog10 Dr; p XUE p XeNB1
2 2
1
p XeNB0
2
Interference Statistic Model and Simulation Verification
After derivation, the uplink interference link can be simplified as I P0 + ( 1)A

+ Y(r, ) + X, and thus the PDF of uplink interference can be obtained through the
expression below,
f I I f X I 1A P0 yf Y ydy,
1
where the distribution function of X related to shadow fading and the distribution
function of intermediate variable Y related to UEs position. They can be given
respectively as follows.

1 x2
f X x p exp 2 ,
2 2 1 2 1 2
dFY y
f Y y
dy
8
> 2
>
> 2 r y; r y;
>
> 2 d, Y min ; r min y Y 0
>
> R r 2 y
>
> min
>
< 0
:
>
> r 2 y;
>
> d
>
> 2 y
>
>
> R r y; 0 y d0 y 0 y
2
>
: , Y 0 < y Y max ; R
R2 r min 2 dy R2 r min 2
In order to verify the above theoretical derivation, the theoretical interference
distribution and Monte Carlo simulation of intermediate variable Y and interference
I are shown in Figs. 4.30 and 4.31 respectively. The system parameters involved in
theoretical calculation and simulation are shown in Table 4.13. Clearly, there is
high consistency between simulation results and theoretical results according to the
results in Figs. 4.30 and 4.31, which prove the correctness of the theoretical
derivation.
Further, based on the derived theoretical model, the impact of the power control
parameters (path loss compensation factor and nominal power P0) setting on the
uplink interference distribution is explored. The corresponding results are shown in
Figs. 4.32 and 4.33.
0.2
simulation
0.18 theoretical
0.16
0.14
0.12
PDF
0.1
0.08
0.06
0.04
0.02
0
110 100 90 80 70 60 50
Y (dB)
Fig. 4.30 PDF of intermediate variable Y
a=0.5
0.07
simulation
theoretical
0.06
0.05
P0=60dBm, =4
0.04
PDF
0.03
0.02
0.01
0
180 170 160 150 140 130 120 110 100
Uplink inter cell interference I (dBm)
Fig. 4.31 PDF of uplink interference I

Table 4.13 Simulation parameter setting

System parameter Value
Cell diameter R 1000 m
Nominal power P0 60 dBm
Minimal distance between UE and eNB rmin 3 m
Antenna model of UE and eNB Omnidirectional antenna
Antenna gain of UE and eNB 0 dBi
Path loss model A 15.3, B 37.6
UE distribution Uniform distribution
Standard deviation of shadow shading 4 dB
Path loss compensation factor 0.5
Monte Carlo simulation times 500,000 times
a=0.5
0.07
simulation
theoretical
0.06
P0=50dBm, =4dB
0.05
P0=60dBm, =4dB
0.04
PDF
P0=50dBm, =8dB
0.03
P0=60dBm, =8dB
0.02
0.01
0
170 160 150 140 130 120 110 100
Fig. 4.32 The impact of nominal power P0 on interference distribution
Through the comparison in Fig. 4.32, we can see that the increase in P0 will
directly lead to the increase in the uplink interference power. The uplink interfer-
ence PDF curve will translate to the region with larger interference. The translation
scale is the increase scale of P0. The shape of the distribution curve will not change
with the change of P0. For different shadow fading scenarios, the impact of the
change of P0 is consistent with the interference distribution function.
0.07
simulation
theoretical
0.06
a=0.5, =4dB,
P0=60dBm
0.05
a=1, =4dB,
P0=100dBm
0.04
PDF
0.03 a=0.5, =8dB, a=1, =8dB,

P0=60dBm P0=100dBm
0.02
0.01
0
170 160 150 140 130 120 110 100 90 80
Fig. 4.33 The impact of path loss compensation factor on interference distribution
Intuitively, for the same nominal power setting, the larger the path loss com-
pensation factor is, the greater the user transmitting power is, then the greater the
corresponding uplink interference for the adjacent cell is. It is not hard to find from
Fig. 4.33 that in the scenario with shadow fading standard deviation 4 dB,
although P0 is lower when 1 than that when 0.5 by 40 dB, the uplink
interference when 1 is still far greater than the uplink interference when
0.5. On the other hand, from the perspective of interference distribution,
when increases, the distribution is more dispersed, and the variance distribution
is bigger. In contrast, when is smaller, the uplink interference distribution is more
concentrated. For the scenario with shadow fading standard deviation 8 dB, the
changes of also impact the distribution of the uplink interference. From
the analysis conclusion in Fig. 4.32, the change of P0 does not affect the shape of
the uplink interference PDF, but only affects the translation of distribution curve on
the horizontal axis (interference). Therefore, the increase in path loss compensation
factors will not only introduce more uplink interference, but also lead to greater
volatility in the uplink interference.
Case 4: Local Mobile Cloud Assisted Computation Offloading
Computation offloading technology, as an important application of mobile cloud

computing, has drawn a lot of attention in recent years. Computation offloading
technology offloads the users local computational tasks to the cloud with rich
resources and extends the computing power of the mobile terminal with limited
resources, so as to use the new computation-intensive applications. Hu et al. [50]
research the computation offloading in Heterogeneous Cloud Radio Access
Network(H-CRAN) scenario, and put forward User-Centric Local Mobile Cloud
(UC-LMC) based on the D2D communications. The basic idea of this framework is:
suppose that the users reach an agreement with the network, and based on some
kind of mechanism, the users can exchange for the traffic/service priority through
providing free computational resources so as to help network provide services to
other users. In the framework of H-CRAN, base station function modules in
baseband resource pool collect the appropriate computational resources, and build
and maintain its mobile local cloud for particular requesting users. Suppose that a
mobile application is divided into a series of subtasks. UC-LMC executes each
subtask in serial order; at the same time, it will send back the calculation results on
each auxiliary user to the requesting user equipment in the short distance commu-
nications link. For battery-powered mobile equipment, energy consumption is an
important factor needed to be considered in the process of offloading. The article
studies the subtask allocation problem of minimum energy consumption. Given that
different services require different network abilities for front haul network, the front
haul link loading limitation is added to the subtask allocation algorithm.
In order to simulate and evaluate the subtask allocation algorithm put forward in
[50], this section simulates with the simulation tools of Matlab. Simulation uses
small cell dense deployment model, and the simulation parameters are listed in
Table 4.14. In the simulation, 25 small cells are set up with each small cell serving
for 5 user equipment. Assuming that each small cell is square of 10 m 10 m. RRH
is located in the center of the small cell. And the mobile user equipment is
uniformly distributed in the small cell.
As shown in Fig. 4.34, this research simulates the influence of different data
input rates on the average energy consumption. The figure compares four kinds of

Small cell number 5*5
Number of mobile user equipment 5
in each small cell
Simulation times 1000
Path loss model (dB) PL
=
{ 30 * log10 (distance) + k * PL_wall + 37)
knumber of walls
PLwall penetrationloss
PL_ wall (dB) 0,5,10,15,20,25,30

0.05
n 3
D (kbits) 1 30
0.2, 0.5, 0.8,1
N 10
12
energy consumption/kbits (lg mJ)
11
10
9
8
7
6
5 offload to UC-LMC
offload to macro-cell BS/BBU pool
4 offload to micromicro-cell BS
offload to UC-LMC with forward link load balancing
3
3 5 10 15 20 25 30
The total amount of sub task data (kbits)
Fig. 4.34 Comparison of 4 types of computation offloading in average energy consumption
computation offloading strategies: (i) offloading to macro station (or through macro
station to the network side server); (ii) offloading to a small cell station (or via a
small cell station/RRH to be connected to network side server); (iii) offloading to
local mobile cloud without considering the effect of the fronthaul network;
(iv) offloading to local mobile cloud and considering load balancing of the prequel
network in the meanwhile. It can be seen from the simulation results that the energy
consumption of offloading the computational tasks to the macro station is the
highest. It is because that the requesting user equipment is usually far from the
macro station, its transmission energy attenuation is bigger on the air interface, and
thus the energy efficiency is minimal. For the same reason, the energy consumption
of offloading the computational tasks to the cell RRH is in the second place. In the
strategy proposed in [50], computational tasks is offloaded to a local mobile cloud
composed of the nearby auxiliary user equipment, which consumes the least energy.
When considering load balancing in fronthaul network, since its main purpose is to
scatter data, the performance of this strategy when the data rate is small is no better
than the situation that offloads the computational task to the small base station or
local mobile cloud. When the amount of data is large, channel differences between
auxiliary user equipment cannot play a key role, so the fronthaul network load
balancing algorithm is also able to reach a good performance.
The simulation result given by Fig. 4.35 is the impact of the set value of
different fronthaul load balancing factors on mobile equipments subtask alloca-
tion. Four different values are selected: 0.2, 0.5, 0.8, 1. When putting the
auxiliary user equipment in descending order according to the channel quality, we
can see when a fronthaul load balancing factor is bigger, the allocated computa-
tional data of auxiliary user equipment is mainly decided by the channel quality.
The channel with better channel quality would be allocated to more data. When
1, i.e., not considering fronthaul network capacity, it will try to allocate the
computational tasks to the auxiliary user equipment with good channel quality
until it reaches the upper limit of its computational power, which will lead to large
2
1 w = 0.2
0
2 1 2 3 4 5 6 7 8 9 10
Forward link data rate
1 w = 0.5
0
1 2 3 4 5 6 7 8 9 10
2
w = 0.8
1
0
1 2 3 4 5 6 7 8 9 10
2
w=1
1
0
1 2 3 4 5 6 7 8 9 10
Number of mobile devices
Fig. 4.35 The subtask allocation considering fronthaul network load balance
Table 4.15 The variance of 0.2 0.5 0.8 1

energy consumption and
E(J/kB) 43,400 29,429 21,537 20,547
fronthaul network load.
V 1 10 25 34
load variance, as shown in Table 4.15. When the fronthaul load balancing factor is
small, the impact of channel quality will be weakened and the fronthaul capacity
restriction factors will be highlighted. The user with relatively good channel
quality, may be allocated with relatively less computational data than the auxiliary
user equipment with poorer channel quality because the fronthaul link capacity
of its correlated Remote Radio Head (RRH) is limited, but less data is assigned to
the auxiliary user equipment with abundant fronthaul link capacity of its corre-
lated RRH.
Case 5: Cross Operator Flexible Spectrum Management Simulation
ITU-R WP5D spectrum requirements report points out that the world frequency
spectrum resource shortfall will reach thousands of MHZ in 2020. To meet the
requirements for high traffic and high rate brought by the rapid development of
intelligent terminals, its necessary not only to develop LTE available spectrum
(such as high frequency band and unauthorized frequency band), but also to
continuously explore the efficient use of spectrum. The network service require-
ments of different operators are unevenly distributed in terms of time, space and
frequency. The QoS requirements are diversified with great differences, and autho-
rized spectrum of different operators has different center frequency, bandwidth,
propagation characteristics, interference level and other characters. Cross-operator

spectrum sharing can match the spectrum resources and services distribution well,
which traditional spectrum planning models cant do. Besides, it enables the
operators to use and manage the frequency spectrum more flexibly and dynami-
cally. Yu et al. [51] propose a cross-operator interference coordination scheme
based on peer-to-peer spectrum sharing in ultra-dense network. Each operator has
the same priority for the resource in the shared spectrum pool. The shared spectrum
pool is divided dynamically and flexibly through network statistical information.
And asymmetric power allocation is used to coordinate between the Inter-Operator
Interference (IOI). Finally, the network performance gain brought by cross-operator
spectrum sharing is analyzed through numerical simulation. The simulation results
show that, when the small cell is in dense deployment, the network spectrum
efficiency of cross-operator spectrum sharing is promoted significantly compared
with the scenario where there is no inter-operator sharing.
The simulation has set two scenarios. One is small cell dense deployment
scenario. The other one is the small cell sparse deployment scenario. The density
of small cell deployment is represented by the activation rate of small cell so as to
verify the applicable scenarios of cross-operator spectrum sharing. Besides, several
schemes are compared in the simulation performance.
Proposed represents the scheme proposed in this paper. Spectrum is shared
between operators. With the spectrum requirements obtained by network statis-
tical information, the usable spectrum resources are dynamically assigned to the
operators, and the inference between operators is coordinated through asymmet-
ric power.
Baseline1 represents the first comparison scheme, that operators use their own
spectrum rather than the spectrum sharing.
Baseline2 represents the second comparison scheme. Spectrum sharing is
adopted between the operators, but the resources allocated to each operator are
fixed. And the asymmetric power is used to coordinate the interference between
the operators.
High power represents that the spectrum sharing is adopted between the
operators, but all the operators use the high power transmission. And the high
power is the base station power in simulation parameters.
Middle power represents that the spectrum sharing is adopted between the
operators, but all the operators use the medium power for transmission. And the
middle power in the simulation is lower than the high power by 3 dBm.
Low power represents that the spectrum sharing is adopted between the
operators, but all the operators use the low power for transmission. And the
low power in the simulation is lower than the middle power by 3 dBm.
Specific simulation parameters are shown in Table 4.16.
Figure 4.36 is the simulation results of small cell in dense deployment scenarios;
the activation rate of small cell is 1, i.e., all the rooms in the small cell are active.
The line with triangle markers represents cross-operator spectrum sharing scheme
using power control in this proposal. The line with dot markers represents the

Parameter Value
Deployment model Dual-stripe model, single layer
Room size 10 m * 10 m
Activation rate Dense 1; sparse 0.5
Operator number 2
The minimum distance between users and base stations 0.3 m
User distribution One BS serve for one UE
Base station power 23 dBm
System bandwidth 80 MHz
Shared resource pool 8 CCs
Each carrier bandwidth 10 MHz
Received noise indicator 9 dB
Path loss model TR 36.872
Seepage loss Outer wall 23 dB; interior wall 5 dB
Shadow fading standard deviation 10 dB
0.9
0.8
0.7
0.6
CDF
0.5
0.4
0.3
Proposed
0.2 Baseline 1
Baseline 2
High power
0.1 Middle power
Low power
0
1 1.5 2 2.5 3 3.5 4
SE(bps/Hz)
Fig. 4.36 The performance in dense deployment scenarios (small cell activation rate is 1)
spectrum sharing scheme without power control in this proposal. The line with short
line markers represents the scheme without spectrum sharing between operators. It
can be seen from the figure that, in a small cell dense deployment scenarios, the
scheme with spectrum sharing between operators have a significant performance
gain compared with the scheme without sharing. Meanwhile, with the use of power
0.9
0.8
0.7
0.6
CDF
0.5
0.4
0.3 Proposed
Baseline 1
0.2 Baseline 2
High power
Middle power
0.1 Low power
0
1.6 1.8 2 2.2 2.4 2.6 2.8 3
SE(bps/Hz)
Fig. 4.37 Performance in sparse deployment scenario (small cell activation rate is 0.5)
control, interference between the operators is reduced, which makes the perfor-
mance of the scheme with power control improved compared with the scheme with
no power control (using the same power, such as high power, middle power and low
power). The simulation shows that in small cell dense deployment scenario, the
inter-operator spectrum sharing mechanism can bring certain performance gain to
network performance. And the inter-operator interference can be eliminated by the
rough information of operator interaction.
Figure 4.37 is the simulation results of small cell sparse deployment scenario.
The activation rate of small cell is 0.5, i.e., only half of each operators small cells
in the rooms are active. In the small cell sparse deployment scenario, due to the
little spectrum requirement of each station in the simulation, it can still meet all the
spectrum access requirements of small cell even without cross-operator sharing
scheme. As a result, the performance curves of the three schemes are basically
overlapping. Meanwhile, the line with hollow circle markers and the line with
hollow rectangle markers adopt lower power, which directly reduce the network
spectrum efficiency. The simulation results show that in the sparse deployment
scenario, there is no need for inter-operator spectrum sharing. Of course, due to the
relatively low services traffic in the simulation, the intra-operator shortage of
spectrum resources may also appear if the services traffic of each small cell is
big. Thus the inter-operator spectrum sharing is needed.
4.5 Software Visualization of 5G Network Simulation
4.5.1 Architecture Summary
Here we introduce a kind of new simulation platform for 5G based on the universal
software and hardware platform. The platform adopts the distributed master-slave
parallel processing architecture, including the master node of the simulation plat-
form, the slave node of the simulation platform, the client end, and the communi-
cations interface. As shown in Fig. 4.38, the functions of each part are as follows.
The master computational node. It is the management center of the simulation
platform, which manages the simulation task requirement from the clients,
decomposes the simulation tasks, schedules the simulation task to multiple
core computational resource of the master node or slave node for parallel
processing, collects the simulation results of each slave node in real time, and
then presents the summary results to the clients. Master nodes are also the
computational nodes at the same time, which undertakes the specific simulation
task. The master node also undertakes master-slave node synchronization man-
agement to ensure that the master-slave node simulation task executed synchro-
nously. The multi-core parallel processing capabilities of master node and slave
nodes have greatly increased the simulation calculation performance. The mas-
ter node and slave nodes support both multi-core servers and hardware board/
FPGA.
The slave node. The slave nodes are managed by master node, take on compu-
tational tasks of the master node, and report the simulation results. A master
Distributed computing service Slave

computing
node1
Client1
Master Slave
Client2 computing computing
node node 2
Clientm Slave
computing
node n
Fig. 4.38 The logical architecture of the new simulation architecture

4.5 Software Visualization of 5G Network Simulation 227
Fig. 4.39 A function deployment of the simulation platform
node can manage multiple slave nodes, so increasing the number of slave nodes
can improve the computing capability of the simulation system.
The client end. It provides human-machine interface, supports the starting up
and issuing of simulation tasks and the real-time display of the simulation results
on the interface.
The communications interface. Considering the high-speed and real-time simu-
lation services requirements, it requires the use of high speed bus and the
communications protocol with higher real-time property.
Figure 4.39 gives a deployment configuration of master-slave node and the client
end. Each software module of the simulation can be flexibly deployed on the
master-slave nodes according to the requirements and doesnt have to be in a
particular form.
The software functions of simulation system are organized according to
Fig. 4.40.
4.5.2 Interface Demonstration

Demonstration of the Simulation Results of 3D MIMO Scenario
Simulation platform supports the simulation calculation of 5G key technology in

massive MIMO technology and the dynamic display of the simulation. As shown
below, the map shows massive MIMO area coverage effect after the user opens and
executes massive MIMO project. QoS effect is shown through the comparison
Fig. 4.40 Simulation system function hierarchy
between video display effect of massive MIMO cell and common cell, as shown in
Fig. 4.41. The video on the left side is the video display effect of the users in the
massive MIMO cell, where antenna configuration number of the base station is 32.
The right side is the video display effect of the common cell, where the base station
antenna configuration number is 8. The trend graph of the cell throughput rate in
lower left shows the real-time changes in the cell throughput indicator of massive
MIMO cell and common cell. It can be seen from the throughput and video that
massive MIMO performance is superior to the common cells. By switching the
number of massive MIMO antennas to 64, the cell beamforming ability is
improved, which further enhances the performance as shown in Fig. 4.42.
Demonstration of the Simulation Results of UDN Scenario
The simulation platform supports the dynamic demonstration of the simulation

calculation and simulation effects of 5G key technology and UDN technology. As
Fig. 4.41 The coverage effect of the massive MIMO cell with 32 base station transmitting
antennas
Fig. 4.42 The coverage effect of the massive MIMO cell with 64 base station transmitting
antennas
Fig. 4.43 UDN dynamic demonstration
shown in the figure below, the map shows the coverage effect of the dense network
with 121 cells after the user opens and executes UDN project. As shown in
Fig. 4.43, each circle represents a cell. The closed circle with no filled color
represents the cell is closed and services is not available. The cell with color
represent that it is open and can provide services. The deeper the color is, the
greater services volume can be provided. The change in the user number in the
whole simulation and the cell throughput trend graph are given below the main
interface. The simulation task has simulated the whole process of a football match
held in the Bird Nest Stadium, which has experienced UDN cell from the closed
state to the open state from before to during the game, and the changing process of
improving the throughput through the interference coordination algorithm, as well
as the process from the UDN cells open state gradually to the closed state after the
match. It intuitively presented the simulation effect of UDN network in this
scenario.
Through analyzing the characteristics of the 5G network and the key technolo-
gies of system simulation software, we can see the future development trend of
software simulation test.
1. Rapidity. It can be predicted that within a period of time the parallel calculation
based on multi-core and many-core will become the major technology means to
improve the computing efficiency. Cloud computing and supercomputers will
References 231
gradually become the main stream calculation way of the software system
simulation.
2. Flexibility. Influenced by factors like the flexible network architecture, network
resource virtualization management, and parallel computational requirements,
each module of the system simulation software needs full decoupling, flexible
expansion. The requirements for the flexibility of software design become
higher.
3. Comprehensiveness. Out of the pursuit for higher QoS, simulation evaluation
indicators have expanded from the cell performance indicators to the user level
QoS indicators. It requires that the simulation should be more comprehensive, be
able to establish the services model closer to the real scenarios, design more
refined and comprehensive statistical methods and indicators, and achieve the
algorithm with more comprehensive performance.
References
1. B. R. Bilel, N. Navid, M. S. M. Bouksiaa. Hybrid CPU-GPU Distributed Framework for Large

Scale Mobile Networks Simulation. IEEE/ACM International Symposium on Distributed
Simulation and Real Time Applications. IEEE, 2012:4453.
2. F. Capozzi, G. Piro, L. A. Grieco and et al. On accurate simulations of LTE femtocellsusing an
open source simulator. EURASIP Journal on Wireless Communications and Networking, 2012
(1):328
3. K. Goga, O. Terzo, P. Ruiu and F. Xhafa. Simulation, Modeling and Performance Evaluation
Tools for Cloud Applications. IEEE 2014 Eighth International Conference on Complex,
Intelligent and Software Intensive Systems, 2014:226232.
4. IEEE 802.16m-08/004r5: IEEE 802.16m Evaluation Methodology Document.
5. P. Kruchten. The 4+1 View Model of architecture. IEEE Software, 1995, 12(6):4550.
6. Y. Zaki, T. Weerawardane, C. G org and et al. Long Term Evolution (LTE) model development
within OPNET simulation environment. OPNET workshop, 2011:15.
7. N. Baldo, M. Miozzo, M. Requena-Esteso and et al. An open source product-oriented LTE
network simulator based on ns-3. Proceedings of the 14th ACM international conference on
Modeling, analysis and simulation of wireless and mobile systems. ACM, 2011:293298.
8. M. Suresh, C. Kamalanathan and S. Valarmathy. Avoiding congestion in IEEE 802.16 and
IEEE 802.11b using Qualnet. 2013 Fifth International Conference on Advanced Computing
(ICoAC), 2013:488493.
9. White paper. 5G Vision and Requirements. IMT-2020 (5G) Promotion Group, China, May,
2014.
10. T. Xie, Q. Han, H. Xu, Z. Qi and W. Shen. A Low-Complexity Linear Precoding Scheme
Based on SOR Method for Massive MIMO Systems, IEEE Vehicular Technology Conference
(VTC Spring), 2015:15.
11. TR 36.201. Evolved Universal Terrestrial Radio Accesss (E-UTRA); LTE Physical layer;
General Description, version 12.2.0, Mar. 2015.
12. C.-X. Wang, F. Haider, X. Gao and et al. Cellular architecture and key technologies for 5G
wireless communication networks. IEEE Communications Magazine, 2014, 52(2):122130.
13. Y. Yang, W. Zhang, K. Wei and X. Yang. Power Reduction for Mobile Devices by Low-power
Base Stations Planning. IET Communications, 2014, 8(18):33723380.
14. S. Jin, X. Liang, K. Wong, X. Gao and Q. Zhu. Ergodic Rate Analysis for Multipair Massive
MIMO Two-Way Relay Networks. IEEE Transactions on Wireless Communications, 2015, 14
(3):14801491.
15. D. Truhachev. Universal multiple access via spatially coupling data transmission. in Proc. of
IEEE International Symposium on Information Theory Proceedings (ISIT), 2013:18841888.
16. X. Chu, D. Lopez-Perez, Y. Yang and F. Gunnarsson. Heterogeneous Cellular Networks:
Theory, Simulation and Deployment. ISBN:9781107023093, Cambridge University Press,
2013.
17. S. Wu, C.-X. Wang, H. Haas, H. Aggoune, M. Alwakeel and B. Ai. A Non-stationary
Wideband Channel Model for Massive MIMO Communication Systems. IEEE Transactions
on Wireless Communications, 2015, 14(3):14341446.
18. R. Zhang, X. Cheng, Q. Yao, C-X. Wang, Y. Yang, and B. Jiao. Interference Graph Based
Resource Sharing Schemes for Vehicular Networks. IEEE Transactions on Vehicular Tech-
nology, 2013, 62(8):40284039.
19. S. Sesia, I. Toufik and M. Baker. LTE-The UMTS Long Term Evolution; From Theory to
Practice. ISBN:9780470697160, John Wiley & Sons, 2011.
20. TR 36.211. Evolved Universal Terrestrial Radio Accesss (E-UTRA); Physical channels and
modulation, version 12.5.0, Mar. 2015.
21. TR 36.212. Evolved Universal Terrestrial Radio Accesss (E-UTRA); Multiplexing and chan-
nel coding, version12.4.0,Mar. 2015.
22. J. C. Ikuno, M. Wrulich and M. Rupp. System level simulation of LTE networks. IEEE
Vehicular Technology Conference (VTC 2010-Spring), 2010, 11(18):15.
23. R. Q. Hu, Y. Qian, S. Kota, G. Giambene. Hetnets - a new paradigm for increasing cellular
capacity and coverage. IEEE Wireless Communications, 2011, 18(3):89.
24. K. Wei, W. Zhang and Y. Yang. Optimal Microcell Deployment for Effective Mobile Device
Power Saving in Heterogeneous Networks. Proceedings of IEEE International Conference on
Communications (ICC 2014), 2014:40484053.
25. R. Shao, S. Lin, and M. Fossorier. Two Decoding Algorithms for Tailbiting Codes. IEEE
Transactions on Communications, 2003, 51(10):16581665.
26. X. Wang, H. Qian, W. Xiang, J. Xu and H. Huang. An Efficient ML Decoder for Tail-biting
Codes Based on Circular Trap Detection. IEEE Transactions on Communications, 2013, 61
(4):12121221.
27. X. Wang, H. Qian, J. Xu, Y. Yang. Trap detection based tail-biting convolution code decoding
algorithm. Electronics and Information Technology Journal, 2011, 33(10):23002305.
28. H. Pai, Y. Han, T. Wu, P. Chen and S. Shieh. Low-complexity ML Decoding for
Convolutional Tail-biting Codes. IEEE Communications Letters, 2008, 12(12):883885.
29. 3GPP TS 36.2122007; 3rd Generation Partnership Project; Technical pecification group radio
access network; Evolved Universal Terrestrial Radio Access (EUTRA); Multiplexing and
Channel Coding (Release 8), Sep. 2007.
30. E. J. Candes, J. Romberg and T. Tao. Robust uncertainty principles:Exact signal reconstruction
from highly incomplete frequency information. IEEE Transactions on Information Theory,
2006, 52(2): 489509.
31. C. Zhao, W. Zhang, Y. Yang and S. Yao. Treelet-Based Clustered Compressive Data Aggre-
gation for Wireless Sensor Networks. IEEE Transactions on Vehicular Technology, 2015, 64
(9):42574267.
32. EPFL LUCE SensorScope WSN. http://lcav.epfl.ch/cms/lang/en/pid/86035.
33. G. Quer, R. Masiero, G. Pillonetto and et al. Sensing, Compression and Recovery forWireless
Sensor Networks: Sparse Signal Modelling and Monitoring Framework Design. IEEE Trans-
actions on Wireless Communications, 2012, 11(10): 34473461
34. N. Ahmed, T. Natarajan and K. R Rao. Discrete cosine transform. IEEE Transactions on
Computers, 1974, c-23(1):9093.
35. G. Quer, R. Masiero, G. Pillonetto and et al. Sensing, Compression and Recovery forWireless
Sensor Networks:Sparse Signal Modelling and Monitoring Framework Design. IEEE Trans-
actions on Wireless Communications, 2012, 11(10): 34473461.
36. T. Zhou, B. Xu, T. Xu, H. Hu and L. Xiong. User-specific link adaptation scheme for device-
to-device network coding multicast. IET Communications, 2015, 9(3):367374.
References 233
37. TS 36.212: Multiplexing and channel coding.

38. SDN architectureIssue 1, OPEN NETWORKING FOUNDATION, June, 2014
39. R. Kokku, R. Mahindra, H. Zhang and S. Rangarajan. NVS:A Substrate for Virtualizing
Wireless Resourcesin Cellular Networks. IEEE/ACM Transactions on Networking, 2012, 20
(5):13331346.
40. H. Ishii, Y. Kishiyama and H. Takahashi. A Novel Architecture for LTE-BC-plane/U-plane
Split and Phantom Cell Concept. IEEE Globecom Workshops, 2012, 48(11):624630.
41. Y. Zhang and Y. J. Zhang. User-Centric Virtual Cell Design forCloud Radio Access Networks.
IEEE International Workshop on Signal Processing Advances in Wireless Communications
(SPAWC), 2014:249253.
42. G. Piro, L. A. Grieco, G. Boggia, F. Capozzi and P. Camarda. Simulating LTE Cellular
Systems: an Open-Source Framework. IEEE Transactions on Vehicular Technology, 2011,
60(2):498513.
43. J. Dongarra,I. Foster,G. Fox,W. Gropp and et al. Sourcebook of parallel computing. Computer
Science Computational Sciences, 2003.
44. H. Wang, K. Li, Y. Ouyang. Shanghai Research Center for Wireless Communications. A
Parallel Computing Performance Test System and Method for Testing of Communications.
Application No. 20150508961.8, Application date: August 18, 2015
45. Z. Xie, Z. Hao, Q. Tao. ZTE Corporation CN101924751B, Single-Mode Service Continuity
Implementation Method and Single-Mode Service Continuity System, Patent publication
(announcement) date 201003-12.
46. H. Liu, Y. Yang, X. Yang and Z. Zhang. Semi-distributed Resource Allocation for Dense
Small Cell Networks, IEICE Transactions on Fundamentals of Electronics, Communications
and Computer Sciences, 2015, E98.A(5):11401143.
47. K. Son, S. Lee, Y. Yi and et al. REFIM: a Practical Interference Management in Heteroge-
neous Wireless Access Networks. IEEE Journal on Selected Areas in Communications, 2011
(29):12601272.
48. A. Abdelnasser, E. Hossain and D. Kim. Clustering and Resource Allocation for Dense
Femtocells in a Two-Tier Cellular OFDMA Network. IEEE Transactions on Wireless Com-
munications, 2014(13):16281641.
49. Y. Zhu, J. Xu, Z. Hu, J. Wang and Y. Yang. Distribution of Uplink Inter-Cell Interference in
OFDMA Networks with Power Control. Proceedings of IEEE Conference on Communications
(IEEE ICC 2014), 2014:57295734.
50. H. Hu and R. Wang. User-centric Local Mobile Cloud-Assisted D2D Communications in
Heterogeneous Cloud-RANs. IEEE Wireless Communications, 2015, 22(3):5965.
51. Q. Yu, J. Wang, X. Yang, Y. Zhu, Y. Teng and H. Kari. Inter-operator Interference Coordi-
nation for Co-primary Spectrum Sharing in UDN. China Communications, 2015, 12
(s1):104112.
Chapter 5
Evaluation Test of Software and Hardware
Co-simulation
Software simulation has high scalability and flexibility. However, its authenticity
and efficiency have a certain gap to the hardware simulation. This chapter will
introduce the technique of co-simulation evaluation on software and hardware, so
as to make up for weakness of software simulation in authenticity and efficiency. As
a systematic statement, firstly, we will introduce the requirements, forms and
applications of the co-simulation evaluation test with software and hardware.
Hardware-In-the-Loop (HIL) simulation evaluation is an important and effective
method forthenew technology test evaluation of wireless communications. Next,
we will elaborate the co-simulefation evaluation test of software and hardware
forthelink-level and system-level simulations, and present the real implementations
of system test cases. Finally, we summarize the co-simulation test evaluation of
software and hardware.
5.1 Overview of Software and Hardware Co-Simulation

Evaluation and Test
The concept of software and hardware co-simulation was proposed as early as the
establishment of the Hardware Description Language (HDL) language. It offered a
reliable software and hardware co-simulation technology basis through the inter-
action between the driver realized by the C language and various hardware devices.
Valderramaet al. [1] proposeda unified model of software and hardware
co-simulation, and the authors in [24] presented different hardware acceleration
designs. The main thought of software and hardware co-simulation evaluation test
is [5]: dividing a system into two main parts. One is realized by using software
modules. The other is achieved by using hardware like actual equipments, devices,
or actual channels. Through the integration of two parts, a minimum system which
needs a technical evaluation will be achieved. It is impossible in the past that

DOI 10.1007/978-3-319-61869-2_5
236 5 Evaluation Test of Software and Hardware Co-simulation
software algorithms or measured objects can be fully evaluated and tested in such a
system.
As a technological means, software and hardware co-simulation evaluation and
test reflect the idea of authenticity and rapidness, which is an effective testing and
evaluating method to deal with the rapid development of 5G communications,
dispersing error detection in the whole design process and reducing the error risks
in the final product.
5.1.1 Requirements of 5G Software and Hardware

Simulation Evaluation and Test
Traditionally, the design for wireless systems in mobile communications mainly

uses software simulation to make algorithm verification and evaluation. However,
there is a certain gap between the system environment under the software simula-
tion and the real situation. They are mainly reflected in:
1. Authenticity
First, the software simulation can not introduce the impact of hardware
environment on communication systems. Software simulation believes that the
hardware design can perfectly realize the software algorithm. But in fact, the
algorithm design is often limited by the hardware conditions. And the software
algorithm is often compromised in hardware implementations. For example, in
the actual hardware environment, the algorithm with particularly huge compu-
tations and extremely high calculation accuracy requirements are often unable to
obtain the optimal performance. The precisely designed algorithm isoften vul-
nerable to the outside interference. The algorithm at the front of transmitting and
receiving hasto consider the power, the impact on other hardware components
and many other factors. If the software simulation is only used for system
verification and evaluation, it is relatively easy to neglect the above problems,
which leads to a result that designed algorithms will only stay in the theoretical
stage and are unable to meet the needs of the real situation. Second, the analog of
the software simulation for the wireless channel has a certain gap with the real
situation. The research purpose of 5G mobile communications is to realize the
transmission of ubiquitous, high-quality and high data rate mobile multimedia.
In high-rate data transmission, the mobile communication system is facing a
very poor wireless channel environment. How to deal with the complex wireless
channel has become a crucial problem in the design of mobile communications
product. Especially in current software emulation conditions, the analog for
wireless channel is realized by abstractly extracting it via a large number of
mathematical methods, which often greatly simplifies the wireless channel. The
difference from the real situations is still very obvious. Because the analog
methods of the software simulation for the wireless channel are insufficient, it
may result in that the designed algorithm is too simple and can not handle the
real complex environment.
5.1 Overview of Software and Hardware Co-Simulation Evaluation and Test 237
2. Rapidness
In general, the software simulation system is a complex computing system
based on operating systems, which needs a non-strictly determined period to
complete each operation. And the entire simulation is composed by tens of
millions or even billions of tiny calculation units. Therefore, the complex simula-
tion will not only take much time, but also have no definite period. For example, in
the LTE system simulation, simulation operations of the analog for one Transmis-
sion Time Interval (TTI) include 798,474,240 times Fast Fourier Transform (FFT)
calculations, 817,152 times matrix inversions, and 13,074,432 times matrix mul-
tiplications. If using software and hardware co-simulation and replacing the
software calculation by the real channel environment and hardware emulation,
we can greatly improve the efficiency of simulation, testing and evaluation.
The development of the modern technology and the market demand has
caused the increasing reduction of product development cycles, while the
needs for product testing and the integrity requirements are constantly growing.
This requires the effective evaluation in the algorithm or code stage, which
advances the original evaluation for the prototype to the technology research and
the development stage, while the methods of software and hardware
co-simulation evaluation test can effectively evaluate the result of technology
implementation in advance in the design stage.
In the system of mobile communication verification and evaluation, it is
significant to use software and hardware co-simulation test and evaluation to
test the key technologies under the real environment. Users realize the designed
algorithms quickly in the wireless link verification test platform based on the
hardware system. Then they can test and evaluate in a real wireless link
environment with many advantages. It can, provide a sufficient and effective
support for universities and research institutes to develop new communication
algorithms. It can alsoshortenthetechnology development cycle toa large extent
and reducing R&D costs and risks of new technologies, algorithms and standard
investment product. In addition, it canrealize the goal of effectively promoting
mobile communication technology development and overcome the shortcom-
ings of excessive codes and the simulation time brought by the pure software
simulation.
5.1.2 Forms and Applications of Software and Hardware

Co-Simulation Testing and Evaluation
Forms of Software and Hardware-in-the-Loop Simulation Testing

and Evaluation
The forms of software and hardware co-simulation in this chapter are described
based on bidirectional uplink and downlinkcommunication applications. The
one-way communication can be used as a special case.
Figure 5.1 shows one form of software and hardware co-simulation with base
station side as the test and evaluation object. It can be divided into two conduction
test evaluations direct connection and through the channel. In this application, the
sample to be tested is base station physical hardware, and the test and evaluation
device at the terminal side is the joint platform of hardware and software.
Figure 5.2 shows another form of software and hardware co-simulation with
terminal side as the test and evaluation object. It can be divided into two conduction
test evaluations direct connection and through the channel. In this application, the
test sample is terminal physical hardware, and test evaluation device measured by
the base station is a joint platform of hardware and software.
Figure 5.3 does not specify the terminal side or the base station side as the test
and evaluation object. Instead it adopts the form of joint software and hardware
platforms on both sides, which extends the scope of test and evaluation to base
station and terminal sides. In this application, the conduction test and evaluation
can also be divided into two partsthrough and not through the channel. This
application form combines the base station with uplink and downlink terminals,
as well as channels to form a closed loop by using the software and hardware
combination. Thus, this kind of software and hardware co-simulation testing and
evaluation is called HIL simulation testing and evaluation.
BS UE
Software
downlink Realtime downlink
DUT uplink channel uplink Hardware
emulator
Fig. 5.1 Software and hardware co-simulation with base station side as the test and evaluation
object
BS UE
Software
Hardware uplink Channel uplink DUT
emulator
Fig. 5.2 Software and hardware co-simulation evaluation with terminal side as the test and
evaluation object
BS UE
Software Software
Hardware uplink Channel uplink Hardware
emulator
Fig. 5.3 No specification of terminal side or base station side as test and evaluation object
5.1 Overview of Software and Hardware Co-Simulation Evaluation and Test 239
Superiority of HIL
HIL simulation refers that during the system test, the tested system uses a real
control system while the rest parts use actual products if they can. If not, they will
use a real-time digital model to simulate the external environment of the controller
to test the entire system. In the HIL simulation, the actual controller and the
simulation model used to replace the real environment or equipment together
form a closed loop test system. The components which are difficult to establish
the mathematical simulation model can be retained in closed loop, so that the test
and initial matching work can be completed in the lab environment, which can
greatly reduce development costs and shorten the development cycle [6].
HIL simulation test evaluation is a semi-physical simulation developed on
the basis of the physical simulation and the digital simulation [7]. It is a typical
semi-physical simulation method based on DUT and environment [8, 9], which
realizes the function of a particular device or the external environment. In the
HIL simulation test, the simulation model replaces the actual equipment or the
environment. The model and real controllers constitute a closed loop test system
through the interface. For the components which are difficult to establish the
mathematical model (such as the inverter system), they can stay in the closed
loop system. Thus, the test for the controller can be completed in the lab
environment. Limit testing, fault testing and high-cost or the testing impossible
in the real environment can also be carried out. The HIL simulation technology
makes full use of computers modeling convenience and simplicity and reduces
the costs. It is easy to make fast and flexible changes to the system input,
by which the changes of system performance can be simultaneously observed
while changing the parameters. For the complex links of unessential investiga-
tions in systems, hardware can be connected directly with the simulation
system. There is no need to make mathematical modeling for all details of the
system [10, 11].
As a real-time simulation, since the HIL simulation technology can incorporate
some physical objects into the simulation loop, it mainly has the following
advantages. Real-time simulation takes the same time as the natural operation of
the system. It increases the reliability of hardware. By using the HIL technology,
the system function test can be carried out at the beginning of the design, which
can effectively reduce errors and defects possibly existing in the process of
development and design. It reduces test evaluation risks. In the process of using
the simulation system environment to simulate the actual testing, the high-risk
control functions, like security operations of verification systems, alarms and
emergency treatment, effectively reduce the testing risks. It reduces testing
costs. By using the simulation system environment to simulate the actual testing
process, it can avoid the procurement configuration of various ancillary equipment
in the early system design, reducing the system testing time and costs. It meets
testing requirements of different application environments. Using flexible soft-
ware configurations, different system environments can be simulated to meet
specific test requirements.
In communication applications, the most significant advantage of HIL simula-

tion is that it can verify the algorithm in the product R&D stage. And it can also
simulate and test the hardware and software environment and work status of the
future products. Meanwhile, it is able to simulate real situations of the whole
equipment without producing real risks or high costs. For example, we can simulate
various channel environments in the laboratory, rather than going to the
realenvironment, and develop the multi-channel RF test environment more rapidly,
rather than waiting for the appearance of MIMO communication products.
In addition, the HIL simulation can be used to test the communication equipment
when the extreme conditions can not be achieved in a real world. Even in the strong
interference environment, it can also test and simulate the ability of communication
and anti-interference for communication products. HIL can help developers iden-
tify deficiencies in the equipment design, even if they only happen under certain
circumstances. The HIL simulation can be used to take the output as a function of
the past or current input for calculations. And the HIL simulation can also be used to
test the hardware subsystem when the entire system is not well prepared, making
the test become an effective component of a development process (from design to
operation). The HIL simulation also helps developers make an early decision for the
special design alternative scheme so as to guarantee its effective operation in the
future application environment. The powerful high-fidelity real-time HIL simula-
tion not only accelerates the time to market by shortening development cycles, but
also reduces the equipment costs and the associated maintenance costs because
there is no need for real hardware during the test.
5.2 Evaluation and Test of Software and Hardware

Link-Level Co-Simulation
5.2.1 HIL-Based Link Simulation Composition
Currently, the relatively mature and rapid prototyping method to realize HIL
function is to design and develop the system architecture based on the current
Commercial Off The Shelf (COTS) module, the FPGA and the advanced multi-core
microprocessor. By making use of its functions and the prototype development
speed, we can accelerate the test process of the design and verification and can
verify and test the new technology algorithms in the analog environment. Because
the COTS system reduces the requirements for the specific customized hardware,
we can get rid of difficulties of the specific customized hardware in design,
maintenance and expansion.
As shown in Fig. 5.4, the HIL verification system in communication systemis
similar to the real communication system. It can provide a full-duplex communi-
cation loop and usethesoftware simulation code platform to intervene the real-time
hardware to simulate, verify and test the rest of system.
5.2 Evaluation and Test of Software and Hardware Link-Level Co-Simulation 241
Realtime information interaction
Simulation and interface Realtime Simulation and

Realtime interface
management software and management
software and
software of PC Synchroniza Synchroniza hardware model software of PC
hardware model tion
(data pool) tion C (data pool)
H
A
N
N
E
Realtime interface
l interface Realtime
software and software and
hardware model Synchroniza Synchroniza hardware model
tion tion
Tx&Rx Tx&Rx
Fig. 5.4 Link simulation system configuration diagram of communication system hardware
in Loop
The HIL simulation combines the physical simulation with the mathematical
simulation. During simulation, it connects the computer with one part of the real
system and builds a mathematical model in calculation to simulate those that are not
easy to test or non-existing parts. This simulation model utilizes the computer
modeling features, featuring simple modeling, low cost, easy parameters modifica-
tion and flexible use. For the part that is difficult to establish the mathematical
model in the system, the actual system or physical model is accessed. Thus, the
operation of the entire system can be ensured, achieving the simulation of the
overall system. The HIL simulation has a higher authenticity, so it is generally
used to verify the validity and feasibility of the system scheme, and also to simulate
the products failure mode and have a dynamic simulation for the closed loop test of
communications systems. The HIL simulation makes simulation conditions closer
to the real situation. In pre-research, commissioning and testing of the product, it
can reflect the performance of products more accurately and objectively.
The HIL test system uses real-time processor operation simulation models to
simulate operating status of evaluated objects. It conducts the all-round and sys-
tematic test to the tested system by connecting to it via the I/O interface. Although
there are HIL tests in different fields with different testing emphases, the overall
simulation architecture is similar. Generally speaking, a HIL test system can be
divided into three main parts, namely, the real-time processor, I/O interfaces and
the operation interface.
(1) The real-time processor unit is the core of the HIL test system. Itis used to run
the real-time model of tested and evaluated objects, board driver and commu-
nication exchange information of up and low position, such as hardware I/O
communications, data records, stimulus generation and model execution. The
real-time system is essential for non-existing physical parts in anaccurate
simulation test system.
(2) The I/O interface is the analog, digital and bus signals interacting with the
measured components and connecting the HIL-based communications. Signal
analog,
digital, Bus
internet Real time interface
GUI I/O DUT
processor
Fig. 5.5 Universal HIL test architecture
Fig. 5.6 HIL-based

Wireless link test
HIL verification
architecture software
MIMO MIMO
Channel
Tx/Rx Tx/Rx
Emulator
system system
Test result
verification
stimulus is produced through the I/O interface to obtain the data for recording
and analysis, and the sensor/actuator interaction is provided between the tested
Electronic Control Unit (ECU) and the virtual environment of the model
simulation. The operation interface can be provided for users to set parameters
and control the test platform. The HIL simulation is a powerful test method,
supporting the direct conversion from algorithms to RF signals (Fig. 5.5).
(3) The operation interface is similar to a real-time processor, which provides test
the command and visualization. It provides configuration managements, test
automation, analysis and reporting tasks. The HIL test system can simulate the
real system. It passes the hardware interface and generates physical signals to
the tested object. At the same time, the system will collect control signals from
output of the tested object and convert it into numbers to calculate, and then
combine the HIL system with the tested object to form a closed loop test
system. Figure 5.6 gives an HIL case architecture in the communication
domain. It mainly includes the signal transmitter, the signal receiver, channel
simulator between transmitter and receiver as well as the software module.
Through the real hardware, it can avoid defects of the pure software simulation
and accelerate the simulation speed.
(1) Multi-channel transmitting/receiving systems: achieving the function of the

signaltransmitter and the signal receiver, being used for testing data that the
testing system sends and receives. Transmitting module includes the embedded
controller used for the remote access and control the data generation as well as
the data transmission between modules, the arbitrary waveform generator used
for generating modulation baseband data in the test, and the frequency con-
verter used for up-convert a IF signal into a RF signal. Receiving module
includes the embedded controller used to remote access and control data
receiving and transmission between modules, the IF digital processor used to
receive, synchronize and demodulates down-converted digital signals, and the
frequency converter used to down-convert RF signal into IF signal or baseband
signal.
(2) Channel simulator: the real channel transmission environment used to simulate
different scenarios.
(3) HIL verification software module: including the HIL verification software, the
test result verification module and other system software.
Users can use the operation interface in the HIL verification software to control
the testing platform to make algorithm replacement test verification as well as data
processing. Adopting the HIL mode can avoid virtuality and limitations of software
simulation verification and increase the accuracy and validity of testing and veri-
fication. This new testing and verification method can meet requirements of com-
plex algorithm RF in-the-loop verification in the 5G wireless communication.
5.2.2 Realization of Link Simulation Testing and Evaluation
Key Factors of Software and Hardware Simulation Testing

and Evaluation
1. Hardware coordination simulation

A software and hardware coordination simulation verification system which
contains a hardware accelerator board generally includes three main parts [12]:
system management tools, simulation software, and hardware platform used by
hardware accelerators, that is, hardware accelerator boards. Its composition is
shown in Fig. 5.7.
System management software is responsible for coordinating the job assign-
ment of software and hardware as well as completing the operation control of the
Fig. 5.7 Typical

System
composition of software and
hardware coordination
management tool
simulation system
Simulation hardware
software accelerator card
PCI interface memory cell
PCI controller
daugh
ter
board
FPGA module FPGA chipset
Fig. 5.8 Structureofhardware acceleration board
hardware acceleration board. Thehardware accelerator board completes the

algorithm simulation verification of the specified pending acceleration module.
Simulation software completes the algorithm in software, and different simula-
tion software can be used according to different processor platforms. Based on a
universal processor, C/C ++, Simulink and LabVIEW can be chosen. And based
on a dedicated processor, ModelSim, VCS and NC-Verilog will be suitable.
Their role is to produce the compiling data files needed by the hardware
acceleration board, and to complete simulation of each module in addition to
acceleration the simulation module in the hardware acceleration board.
2. Hardware acceleration board
A major part of the software and hardware coordination simulation system is
the development of the hardware acceleration board. This is also the heteroge-
neous section in software simulation. In general, choosing PCI/PCI-E as the
connection bus is a conventional manner more common and easier to achieve.
The typical structure of the hardware acceleration board based on this mode is
shown in Fig. 5.8.
The hardware accelerator generally comprises at least two FPGA chips,
respectively for user programming and downloading and for system control.
The pending acceleration module is configured in the main FPGA chip in a
hardware acceleration board, while the chip which controls the FPGA chip is to
control dynamic compilation of main FPGA chip. Main FPGA chip is connected
with the PCI or PCI-E interface through the PCI controller to achieve data
communications with the host. The hardware acceleration board also typically
contains a storage module, such as Read Only Memory (ROM) and Ramdom
Access Memory (RAM). Therefore, the simulation model is no longer needed as
in traditional HDL simulators, which greatly improves verification ability. In
order to realize the simulation of large-scale circuits on the hardware accelera-
tion board, general hardware acceleration boards all contain sub-board connec-
tion interfaces. The sub-board can be an FPGA or a storage unit. It also can be
Advanced RISC Machines (ARM) or other micro-processing chip. If necessary,
adding the sub-board can be a way to increase simulation capabilities.
3. Software and hardware data exchange

Software and hardware coordination simulation verification uses the real-
time hardware acceleration and the software emulation platform simultaneously.
Therefore, it is a key requirement for development on how to realize communi-
cations and data exchange between the hardware acceleration board and the
software platform. If the PCI/PCIe interface is used in design, we can design to
use DMA or FIFO for data exchange.
The general operation principle and data exchange process of hardware
acceleration verification can be expressed asfollows.
The hardware acceleration verification system first uses the system manage-
ment tools to transmit HDL code compilation files to the driver of the hardware
acceleration board through interface tasks, such as Programming Language
Interface (PLI), FLI and VHPI, etc. Then the driver will convert these data
into a specific format that the hardware acceleration board can receive, so the
pending accelerated verification module will be configured to the main FPGA
chip in the hardware acceleration board. Thus according to simulation require-
ments set in system management tools, the accelerated simulation will be
realized. After the acceleration simulation, similar to the above-mentioned
process, the hardware acceleration board driver reads acceleration simulation
results, then transmits it to the software simulation platform through interface
task for the follow-up software simulation, observation and debugging.
Since the simulation task that was originally finished in the software platform
is downloaded in the real hardware, such hardware-acceleration verification
system can speed up the simulation to a large extent and improve the simulation
efficiency. Especially for the large-scale MIMO link simulation or multi-user
simulation, acceleration effect will be particularly significant.
4. Unified development tools of software and hardware
Compared with the traditional software simulation platform, software and
hardware coordinated simulation verification incorporates the real-time hard-
ware. So it brings new development tasks and challenges todevelopers of the
traditional software simulation platform Therefore, in developing the software
and hardware coordinated simulation verification environment, providing a
unified development tool is a demand, and also a task and challenge.
The advanced languages, such as C language, have advantages over Verilog
language in the process control, so C program can be used to generate test
excitation and read the signal value. Taking Windows platform as an example,
users write the interface function in C language and Verilog PLI, compile codes
and generate dynamic link library, and then call these functions in TestBench
written by Verilog language. Thus, designers use the Verilog PLI interface to
create their own system call tasks and system functions. C language can be
programmed to help simulate DUT to achieve the function that Verilog grammar
can not make.
In current market, there are some commercial cross-platform software and
hardware unified development tools, such as System or HandleC, which can help
complete the development of the software and hardware coordination simulation
verification system.
Main bottleneck for system implementation
Applicationswith HIL test systems in the communications domain have become

popular. But it is not difficult to find that in designing and implementing the test
scheme, we need to consider technical bottlenecks faced by the closed loop in the
realizing HIL test. These bottlenecks are closely associated with the increase in
wireless communications requirements. New wireless communication require-
ments are growing and new technical standards requirements continue to increase,
such as the growth of bandwidth, increase of RF channels and In-phase/Quadrature
(I/Q) baseband channels, as well as the development of the digital simulation
technology and the data rate improvement of digital interface, which are all
technical bottlenecks faced by system implementations.
Closed loop method based on HIL link simulation
Closed loop methods based on the HIL link simulation can be divided into the
following categories:
1. Signal real-time feedback based on the TCP/IP technology.
2. Signal real-time feedback based on Datasocket.
3. Signal real-time feedback based on the shared variable engine.
4. Signal real-time feedback adopting other high-speed bus (Fig. 5.9).
In a HIL verification systemforcommunicationsbased on the software defined
radio platform, the design of feedback signal implementation methods should
consider the following factors: software and hardware architecture of transceiver;
reduction of heterogeneous components of system software and hardware; using the
most reasonable, practical and efficient way to achieve feedback signal while
meeting the minimum system requirements. Considering all the factorsabove,
signal real-time feedback based on the TCP/IP technology and the shared variable
engine may be the best scheme.
Signal
Transmit channel Receiver data
resouce
(1)TCP/IP
(2)Data socket
(3)shared variable
(4)optical fiber
(5)Other high speed bus
Fig. 5.9 Feedback signal implementation methods in HIL verification system of communications
system
Signal real-time feedback based on the TCP/IP technology is realized by using

the TCP/IP protocol. The characteristics of the TCP/IP mutual access and shared
resources can be used to achieve signals real-time feedback and variable sharing.
Signal real-time feedback based on the shared variable engine is to use shared
variables to share data in different circulations of the same program block diagram
or different software in the network. Three types of shared variables can be created:
single-process, web publishing, and time-triggered shared variables, similar with
UDP/TCP queue, real-time FIFO and other existing data sharing methods on the
software platform. Wherein the network-released shared variables can be read and
written on Ethernet. Network application processing is completed entirely by
network-released variables. In addition to making data available on the network,
the network-released shared variables also add many functions that single-process
shared variables can not provide. Due to the need for a variety of additional
functions, internal implementations of network-released shared variables are
much more complex than the single-process shared variables. In single-process
shared variables, all writing and reading operations share a single real-time FIFO.
Yet for network-released shared variables, this is not the case. Every reading
operation of network-released shared variables has its own real-time FIFO in a
single element and multiple elements. Temporary fluctuations in a variables
reading/writing speed can be solved through the FIFO buffer in order to achieve
signal real-time feedback based on the shared variable engine.
5.2.3 Case Introduction of Software and Hardware

Simulation Evaluation Test
Case I: Test of Spectral Domain Channel Multiplexing Technology Based

on Cyclic Delay Modulation
I. Tested technology principle and test requirements

1. Spectral domain communications technology principle
As a standard multiple-antenna diversity technology with good compat-
ibility, Cyclic Delay Diversity (CDD) has greatly enhanced the existing
standard OFDM technology to enable them to get enough space diversity
gain in rich scattering wireless environment [13, 14]. Cycle delay processing
can convert spatial diversity to frequency diversity, which adds redundancy
in frequency domain of OFDM systems. Different from Space-Time Block
Coding (STBC) [15] and Space-Time Trellis Codes (STTC) technology
[16], the CDD technology can be realized only at transmitter, so the system
which uses this enhancement technology maintains the compatibility for
standards. Thus, CDD technology can be integrated into some existing
broadcast standards (e.g., Digital Audio Broadcasting (DAB), Video Broad-
casting (DVB) [17] and Digital Video Broadcasting Handheld (DVB-H) [18]
and next generation of mobile communications (3GPP-LTE) [19], and also

applicable to wireless Metropolitan Area Network (MAN) and LAN stan-
dards such as IEEE 802.11a [20] and HIPERLAN 2 [21]. However, the CDD
technology can not provide Space Division Multiple Access (SDMA) func-
tions and spatial multiplexing functions simultaneously.
Cyclostationarity of Cyclic Delay Diversity Orthogonal Frequency Divi-
sion Multiplexing (CDD-OFDM) signals, in general, is caused by two
different inherent cyclic processes, namely Cyclic Prefix (CP) and CDD
process. Specifically, CP and CDD processes induce different, separate
cyclostationary components separated from cyclic frequency and delay
parameters on a 2D plane. In particular, the position and size of
cyclostationary components induced by CDD may vary with changes of
cycle delay parameters, and will be mutually linearly independent
[22]. Cyclostationarity shown in the CDD-OFDM signal has been applied
to the following two types of spectral domain communication system. They
are spectral domain channel multiplexing transmission for singleusers based
on CDD [23] and spectrum division multiple access for multi-users based on
the cyclic delay channelization vector [24].
In a spectral domain channel multiplexing transmission system based on
Cyclic Delay Modulation (CDM), the transmitting device of the system
embeds the CDM module and the CDD-OFDM transmitter module. The
CDM module is used to map the sub-information bit stream as a cyclic delay
vector. The CDD-OFDM transmitter module is used to have a cyclic delay
operation on the diversity OFDM symbol according to the described cyclic
delay vector to achieve the multiplexing of spectral domain channel. The
system embeds the CDM module in the existing CDD-OFDM standard
system through the multi-antenna device and modulates the size and position
of the cyclic autocorrelation function of CDD-OFDM signals. So it achieves
the multiplexing of OFDM modulation channel and CDM channel and
solves the problems of multiplexing spectral domain channel in multi-carrier
frequency domain channel. While the system gains a CDD, it does not
consume extra power and bandwidth and increases the transmission data
rate of the system.
2. Implementation of the spectral domain channel multiplexing technology
based on CDM
In a spectrum division multiple access system based on the cyclic delay
channelization vector, the transmitting device of the system comprises the
first spectrum division multiple access scheduling entity based on a cyclic
delay channelization vector, and the first spectral division multiple access
physical layer entity based on at least one of cyclic delay channelization
vectors. Its first spectrum division multiple access scheduling entity based on
the cyclic delay channelization vector comprises the first dispensing unit of
cyclic delay channelization vector and the first adaptive modulation unit
based on spectral division multiple access channel. The first spectral division
multiple access physical layer entity based on at least one of cyclic delay
channelization vectors comprises a cache unit connected in sequence, a

coding and data rate matching unit, an orthogonal amplitude modulating
unit, and a spectrum division multiple access processing unit. By using the
multiple-antenna transmitting device, the system makes cyclic autocorrela-
tion functions of CDD-OFDM signals become linearly independent, so as to
solve the problems of ASU multiple access that the CDD-OFDM technology
can not achieve.
(1) Implementation of CDD-OFDM transmitting scheme
The system block diagram of the transmitting device of a spectral
domain channel multiplexing transmission system based on CDM is
shown as Fig. 5.10.
As shown in Fig. 5.11, where the CDD-OFDM transmitting module
generates module modification, adds a processing unit and realizes the
generation of CDD-OFDM signals based on standard OFDM signals.
The CDM module is used to control the signals generated by the cyclic
delay diversity OFDM transmitter module. The block diagram of added
a processing unit in the CDD-OFDM transmitter module is shown in
Fig. 5.11.
Figure 5.12 contains a processing unit, a power normalization
processing unit and a cyclic shift processing unit based on the cyclic
delay vector. The processing unit is the component of the CDD-OFDM
transmitter module, which is used to deal with the signal of CDM vector
mapping unit. The cyclic shift processing unit based on the modulated
cyclic delay vector is used to generate a shift signal.
(2) Implementation of CDD-OFDM receiving scheme
Cyclic delay
1st FEC modual modulation vector
mapping unit
Input
bit
stream
S/P
Quadrature
amplitude
2nd FEC modual Process unit
modulation ..
unit .
Fig. 5.10 System block diagram of transmitting device of CDD-OFDM system

Cyclic delay
modulation
vector
c mapping unit
d
sl ,1[n]
Quadrature cyclic prefix
amplitude sl ,2 [n]
modulation cyclic shift D 2 cyclic prefix
{cl , k }kN=-01 sl [k ] 1
unit IFFT unit
NT sl , nT [n]
cyclic shift D n
T
cyclic prefix

sl , NT [n]
cyclic shift D N T
cyclic prefix
Fig. 5.11 Block diagram of processing unit in CDD-OFDM transmitter module
Cyclic
Cyclic delay Cyclic delay
autocorrelati
modulation modulation deco
on function
vector vector mapping der
estimation Output
decision unit unit
unit bit
stream
P/S
OFDM demodulation
Fig. 5.12 System block diagram of transmitting device of CDD-OFDM system
The system block diagram of the transmitting device of a spectral

domain channel multiplexing transmission system based on CDM is
shown in Fig. 5.12.
The transmitting device of a CDD-OFDM system comprises an
OFDM demodulation module and a CDM module, which are respec-
tively used for the standard OFDM signal module demodulation and
CDD-OFDM signal demodulation.
3. Testing requirements analysis
At present, the spectral domain channel multiplexing technology based
on CDM is still in the theoretical research stage. The principle is usually
realized by software simulation, which still has a certain gap with the actual
hardware. For this, we need to use semi-physical simulation functions of the
testing platform to complete the joint testing and verification of software and
hardware of key technologies, providing a more objective reference for the

final hardware implementation.
To complete the test verification of the CDD-OFDM technology, the test
verification platform should meet the following requirements:
(1) The CDD-OFDM transmitter and receiver modules in line with the
3GPP LTE standard.
(2) Antenna configuration of transmitting and receiving devices support
2 1 and 2 2, and support 4 2 in the future.
(3) Meeting the requirements of configuring cyclic delay vectors in different
OFDM symbol lengths.
(4) Able to call the baseband digital signals gained according to sampling
periods in the receiving device. The minimum requirement of data
storage is to support at least ten OFDM symbols once (including CP).
(5) The instrument platform can support digital signal processing modules
extended by users in the transmitting and receiving devices (achieved by
using MATLAB program).
(6) The instrument must support the transmission performance that two data
streams can be separately analyzed.
II. Scheme design of software and hardware integration testing
1. Overview of scheme
In order to achieve the verification of the CDM spectral domain channel
multiplexing technology, the semi-physical testing scheme based on
SystemVue is used in this test case. On the one hand, SystemVues powerful
wireless link simulation functions can be used to achieve CCD-OFDM signal
generation and reception analysis. Meanwhile the connection functions of
SystemVue with instrument realize signal download and capture. On the other
hand, the vector signal generator and vector signal analyzers are used to
generate and receive real signals. The overall scheme is shown as in Fig. 5.13.
In order to meet the verification requirements of the CDM spectral
domain channel multiplexing technology (CDD-OFDM), this scheme adopts
software and hardware integration test program to achieve the HILstest.
Considering that the uplink of an LTE system uses Single-Carrier Fre-
quency-Division Multiple Access (SC-FDMA) modulation, the scheme
selects a modified scheme based on the LTE downlink.
The test verification platform scheme of the CDM spectral domain
channel multiplexing technology includes CDD-OFDM signal generation,
CDD-OFDM signal reception and MIMO system test verification platform,
which will be described respectively in the following.
(1) CDD-OFDM signal generation scheme based on SystemVue
(1) LTE OFDM signal generation Scheme
According to standard specification, the OFDM signal modula-
tion scheme is used in 3GPP the LTE signal downlink [25]. This test
Fig. 5.13 System block diagram of semi-physical simulation testing of CDD-OFDM system
case uses SystemVue LTE OFDM signals to generate module ver-

sion as shown in Table 5.1.
The schematic diagram of the 3GPP LTE downlink OFDM signal
generation based on System Vue, is shown in Fig. 5.14.
(2) CDD-OFDM signal generation scheme
A CDD-OFDM signal generation scheme used in this test case is
modified on the basis of the LTE downlink OFDM signal generation
schematics as shown in Fig. 5.14, which mainly modifies the
[LTE_DL_MIMO_2Ant_Src] module (red circle marked).The mod-
ified block diagram of CDD-OFDM generation module is shown in
the blue marked circle in Fig. 5.15.
In the CDD-OFDM modulation signal generation scheme, users
can set their own CDD shifting parameter. SystemVue will have a
cyclic shift according to the value set by users, so as to generate
CDD-OFDM signal required by users.
Table 5.1 LTE downlink signal (OFDM) generation module parameters

Name Note Remark
Software platform System Vue 2008.09
version
LTE signal In line with 3GPP LTE v8.6
standards
LTE OFDM signal Duplex mode FDD Optional: TDD
settings Signal 5MHz Optional: 1.4MHz, 3MHz, 10MHz,
bandwidth 15MHz, 20MHz, etc.
Frame struc- Type 1 Optional: Type 2
ture type
Frame length 10ms
Sampling rate 15.36MHz
Oversampling Ratio 2
Fig. 5.14 Schematic diagram of LTE downlink OFDM signal generation based on SystemVue
(2) CDD-OFDM signal receiving scheme based on SystemVue

The CDD-OFDM signal generation scheme used in this test case is
modified on the basis of 3GPP LTE downlink OFDM signal generating
schematics, which mainly modifies [LTE_DL_MIMO_2Ant_Rev] mod-
ule; block diagram of the modified CDD-OFDM reception module is
shown in the blue marked circle in Fig. 5.16.
Fig. 5.15 CDD-OFDM signal generation schematics diagram based on SystemVue
In this CDD-OFDM modulation signal reception case, in order to

simplify the receiver, we use the CDD shifted sequence which is
assumed as known to demodulate the received CDD-OFDM signal,
thus achieving CDD-OFDM signal reception analysis.
2. MIMO system experimental verification platform scheme
In order to meet the experimental verification needs, the physical map of
the test verification platform that supports a 2 2 MIMO system is shown in
Fig. 5.17. The test verification platform equipment consists of two
ESG/MXG vector signal generators, two PSA/MXA spectrum analyzers
(for down-converted to IF signal), and a one multi-channel digital oscillo-
scope. In addition, the synchronization wire, the RF connecting wire and the
network wire and other materials are also needed.
Fig. 5.16 CDD-OFDM signal reception schematic diagram based on SystemVue
Fig. 5.17 MIMO experimental verification platform based on SystemVue
The working principle of this test platform is to use the CDD-OFDM

signal generation module of SystemVue to generate user-defined waveform
files, separately download them into two vector signal generators and trans-
fer waveform files to RF modulation through the I/Q modulation function of
the vector signal generatorOnthe other hand, SystemVue can send two-way
time domain waveform data selected by the multi-channel digital
oscilloscope to the SystemVues CDD-OFDM signal reception module for

analysis, and finally obtain test results. In this platform, the synchronization
wire is used to guarantee the synchronized output of two vector signal
generators, and to ensure the synchronized trigger of receiving and sending
signals.
3. Results and analysis of experimental verification
(1) CDD-OFDM signal generation verification
Through the building of the experimental verification platform, we
obtain the OFDM signal without CDD modulation and the OFDM signal
with CDD modulation, which are respectively shown in Figs. 5.18 and
5.19. From comparison of the following two figures, we can see that
CDD modulation causes shift to OFDM signals.
(1) Testing results of CDD-OFDM received signal constellation graph
The SystemVue simulation platform is used to build the
CDD-OFDM system. Constellation of received signals can be
obtained through simulation tests, as shown in Fig. 5.20.
(2) CDD-OFDM system BER simulation test results
The SystemVue simulation platform is used to build the CDD-OFDM
system. By adding the white Gaussian noise in the system, BER and
SNR relationship graphs can be obtained through the simulation test, as
shown in Fig. 5.21.
Fig. 5.18 Time domain graph of OFDM signal without CDD modulation
Fig. 5.19 Time domain graph of OFDM signal after CDD modulation
Fig. 5.20 Constellation graph of received signal from CDD-OFDM system simulation test
(16QAM)
The scheme uses the software-hardware integrated test scheme to

achieve the semi-physical objects simulation test, better solving the
testing verification of the CDD-OFDM spectral domain multiplexing
transmission technology in hardware systems. This test case uses the
SystemVues powerful simulation test function to modify the downlink
Fig. 5.21 CDD-OFDM system BER simulation test results
OFDM simulation link based on LTE MIMO configurations and quickly

realize the CDD-OFDM signal generation and reception analysis by
increasing the CDD software module.
On the other hand, the LTE MIMO test verification platform adopted in
this scheme has good scalability, which supports the generation and
reception of signals whose carrier frequency is up to 6GHz and bandwidth
up to 20 MHz. It can be extended to 4 2 MIMO test platform, or even
higher according to MIMO configurations. In addition, the scheme sup-
ports the addition of the channel simulator, so as to achieve key technical
test verification under a typical MIMO wireless channel environment.
Case 2: Verficationof Frequency Offset Correction Algorithm

in Centralized SC-FDMA
I. Overview
The LTE uplink multiple access mode is SC-FDMA, whereas the downlink
mode is OFDMA. Both have higher spectral efficiency and use the frequency
domain balancing technology with low complexity to suppress multipath
fading. However, the SC-FDMA is a single carrier technology with even
lowerpeak rate, so it is used as an uplink multiple access technology to reduce
the costs of the mobile station. SC-FDMA has become an uplink transmission
technology of the LTE physical layer. The technology has achieved great
success in commercial and standardization activities. The reason is that
SC-FDMA system has the following advantages:
(1) Good performance of resisting multipath fading.

(2) High spectrum utilization efficiency.
(3) Low implementation complexity.
(4) Flexible bandwidth configuration.
(5) High frequency diversity gain.
SC-FDMA modulation signal is in the time domain. In the frequency
domain, each modulation can be extended to the entire frequency band.
Therefore, we can take advantage of multipath channel frequency diversity
gain to suppress narrowband interference.
(6) PAPR is lower than OFDMA system.
The SC-FDMA system is a single carrier system with lower PAPR.
From another perspective, the SC-FDMA system is an OFDMA system
extended from Discrete Fourier Transform (DFT). The partly offset com-
pensation function of DFT transform and Inverse Discrete Fourier Trans-
form (IDFT) transform of OFDMA leads to lower PAPR of SC-FDMA
signals. Wherein the IFDMA system is a constant envelope system and
PAPR is around 0 dB. PAPR of LFDMA signals is also lower than
OFDMA signals.
In summary, compared with OFDM systems, SC-FDMA has the advantages
like low PAPR, full utilization of frequency diversity of multipath channel, etc.
1. Tested algorithm
Assume that SC-FDMA uplink totally has N subcarriers and M users,
with each user occupying P N/M subcarriers and all users communicating
with base station through independent multipath channel, as shown in
Fig. 5.22. Assume that i denotes subcarriers assigned to i, and then
[Mi1 i f0; 1; ; N 1g, i \ j , for i 6 j. The subcarrier distribu-
tion scheme is as follows,
i fi 1P rjr 0; ; P 1g 5:1
Assume that the length of the cyclic prefix Ng is larger than the maximal
channel delay length and the largest time shift among users.
Fig. 5.22 SC-FDMA uplink model

After passing through multipath fading channel, the signal of user i is
yni xni * hni 5:2

where * presents the linear convolution. xni is the transmitted signal of the
users i hni is the unit impulse of channel between user i and base station.
Only for n 0 , , L 1, hni is not zero, where L is the maximum channel
delay spread (DS).
Taking into account of the influences of carrier frequency deviation and
additive noise, the received signal of the base station can be denoted as
X
M j2i n
rn yn i e N zn , Ng n N 1 5:3
i1
where i , i 1 , , M denotes user is normalized carrier frequency

offset value, and zn denotes the additive white Gaussian noise.
(Eq. 5.3) can be rewritten into the following form:
1 X M
R p Yi Ci Z 5:4
N i1
h iT
i i i
where R [R0, R1, , RN 1]T, Yi Y 0 ; Y 1 ; ; Y N1 ,
h iT
i i i
Ci C0 ; C1 ; ; CN1 , Z [Z0, Z1, , ZN 1]T, []Tdenotes the
i i j2i n
matrix transpose, Rk , Y k , Ck denote the FFT transform of r n , yni , e N .
The receiver detection algorithm is shown as below in Fig. 5.23.
To compensate for the carrier frequency deviation, before the FFT
processing, the received sequence rn is multiplied by the time domain
j2o n
sequence e N ,
j2o n
b
y n rn e N , 0nN1
Fig. 5.23 Receiver test block diagram

The goal of algorithm is to find the optimal0value, which maximizes the

average Signal Interference Ratio (SIR), as shown in Fig. 5.23. It is assumed
that carrier frequency deviation value has been accurately estimated.
After FFT for b y n in expression,
b p1 R C0
Y o
N
1X M
0 1 0
Yi Ci Co p Z Co
N i1 N
1 X M
0 0
p Yi Doi Z
N i1
M
X
0 0 0
Hm Xm Dom Hi Xi Doi Z
i1
i 6 m
where Co Co, 0 ; Co, 1 ; ; Co, N1 T , Y [Y0, Y1, , YN 1]T,

0
h 0 iT h iT
0 i i0 i0 i i i
Doi Do, 0 ; Do, 1 ; ; Do, N1 , Xi X0 ; X1 ; ; XN1 ,
n o h iT
i i i b Y b 1 ; ; Y
b0; Y b N1 , Co , k, Yk,
Hi diag H 0 ; H 1 ; ; H N1 , Y
j2 i o n
i0 i i j2o n
b k are the FFT transform of e N , yn, e N ,xi , hi , b
Do, k , Xk , Hk , Y n n yn.
In the expression, the first term includes the signal term of the k-th
sub-carrier and Self Interference (SI); the second is MultiUser Interference
(MUI); the third is additive noise. Received data on the k-th subcarrier is
given by the expression.
j N1m o sin m o
b k p1 Xm Hm e
Y N
k k
N m o
sin
N
X m o qm k sin m o q
1 j N1
m k
p Xqmm H qmm e N
N q 2 m o qm k
m m sin
N
qm 6 k
1 X M X j N1i o qi k sin q k
i o i
p Xqii Hqii e N Noise
N i 1 qi 2i i o qi k
sin
N
i 6 m
h i h i
m
Assume Xqii ,Xqmm ,Xq0 are not correlated,E Xqii E Xqmm 0, where
m
0 0
qi 2 i , qm , qm 2 m , qm 6 qm , E[ ] represents the expectation operator. Set
2
2
E Xqii 2X , E Hqii i , and then the powers on received signal of
user m on the k-th sub-carrier, SI and MUI are as follows:
m 2X sin 2 m o
2Signal
N sin 2 mNo
m 2X X sin 2 m o q k
m
2SI, k
N sin 2 m oNqm k
qm 2 m
qm 6 k
2X XM X sin 2 i o q k
i
2MUI, k i
N qi 2i sin 2 i o qi k
i1 N
i 6 m
According to the expression -, SIR of user m on the k-th sub-carrier is

2
m o
m sin
m sin 2 mN o
SIRk o
P
M P sin 2 i o qi k
i
2 i o i
q k
i1 qi 2i qi 6k sin N
It is noted that the optimal carrier frequency offset compensation value

should be between the minimum and maximum values of user CFO. The
optimal o value of user m in SC-FDMA system is given by the following
expression:
(
m
o, opt arg max SIRm o
o
s:t: minf1 ; ; M g o maxf1 ; ; M g
m
where SIRm o P1 SIRk o , P N/M.
2. Test environment and setup
(1) Characteristics of input data
The input signal is the SC-FDMA signal, with the characteristics
including:
(1) Total sub-carrier number (256).
(2) Cyclic prefix length is 40.
(3) Modulation mode (QPSK)
(4) Subcarrier spacing (15 kHz)
(5) Moving speed (39 km/h)
(6) Carrier frequency (400 MHz)
(2) Characteristics of radio channel
The wireless channel is assumed to be a multipath Rayleigh fading
channel.
3. Test indicators
BER: defined as the ratio of the number of received error bits and the
total number of sent bits.
4. Expression forms and accuracy
For tested data, the following expression can be used:
(1) When other parameters are unchanged, frequency deviation is fixed.
And BER performance is compared with traditional methods.
(2) When other parameters are unchanged, frequency deviation is ran-
domly and uniformly distributed. The range of random frequency
deviation is controlled, and BER performance is compared with the
value from the traditional methods.
II. Scheme design
According to test case requirements, the design uses NI PXIe-5673E, NI
PXIe-5663E and C8 to complete testing tasks (Table 5.2).
Hardware connections in testing scheme are shown as below:
The device connection is shown as in Fig. 5.24 According to the test algorithm,
5673E device generates the transmission signal waveform. It then passes
through the channel machine C8 to receive signals, and finally enters into
5663E. The stored data is processed by PC and then BRE is analyzed.
III. Test results
Because of equipment limitations, when in single-user situations, the user
number M 1, SC-FDMA system degenerates into a Single Carrier Frequency
Domain Equalization (SC-FDE) system. Test channels are AWGN and
Extended Vehicular A model (EVA) channels. BER performance of measure-
ments and numerical simulation are indicated as in Fig. 5.25.
In AWGN channel, the measured and numerical simulated BER perfor-
mance is basically consistent. For EVA channel, the measured BER perfor-
mance is worse than numerical simulation performance. When BER 0.1,
there is a loss of about 8 dB. The reasons for the poor measured BER
performance in EVA channel are as follows:
Table 5.2 Instrument list Instrument Description

used in table design
NI PXIe-5673E Vector signal generator
NI PXIe-5663E Vector signal analyzer
C8 Channel simulator
Fig. 5.24 Device

connection diagram 5673E
C8
5663E
0
10
-1
10
-2
10
BER
-3
10
-4
10
AWGN PXI
-5
10 EVA PXI
EVA Simulated
AWGN Simulated
-6
10
0 2 4 6 8 10 12 14 16 18 20
SNR(dB)
Fig. 5.25 BER performance
1. The frequency offset estimation algorithm in Labview codes does not

consider the impact of multipath channel;
2. Because there is no AWGN module in the hardware platform, in the process
of adding Gaussian noise in Matlab script,
It is considered that the received data does not include noise. But in fact,
the received data has already contained noise caused by hardware devices.
Thus, SNR in the figure is bigger than real SNR.
Therefore, in the single-user testing, the correctness of time synchroni-
zation, channel estimation algorithms and the basic system module building
is verified.
Case 3: Laboratory Integrated Verification Platform Supporting

Remote Use
I. The main functions and difficulties of the laboratory integrated verification

platform supporting remote use
1. 4 4 channels wireless link platform based on modular instrument
Multichannel is also called MIMO technology. It can multiply increasing
the channel capacity and spectrum efficiency of communication system
without increasing the channel bandwidth and the total transmitting power.
Theoretically speaking, channel capacity will increase linearly with the
growth of the number of antennas. It has become the key technology and
hot research topic of IMT-A mobile communications systems. Therefore,
the test for multi-channel RF and wireless channels is a necessary and
important part of the key IMT-A technology testing and verification
platform.
The previous test solutions in the industry mostly concentrate on 2 2
channels. Their application scope is limited. Therefore, we can say that by
far there is no really mature commercial solution suitable for the multi-
channel (4 46 6, or even more number of channels) test.
German MEDAV has channel sounder products, and Finnish
EleKtrobithas ittoo. They both can support the multi-channel channel tester.
However, in their implementations, a single signal source transmits in
different time slices from the same source through the high-speed RF
switch. Therefore, they can not complete the test of diversity technology
and are different from the proposed implementation method of the multi-
channel adaptive and self-updating test platform. Besides, the value of each
companys single product not including software is more than 5 million.
1.1 Difficulties and challenges in the system development
The multichannel adaptive and self-updating test platform will
directly face the challenge of multi-channel testing. The typical chal-
lenges it brings to traditional test equipment are as follows:
1. It needs multiple signal sources to precisely synchronize generating
source signals and self-correct.
2. It needs multiple analyzers to precisely and synchronically analyze
the signals and self-correct.
3. Data sample transmission rate and memory depth of multi-channel
testing are far beyond the support ability of the traditional test
equipment.
4. It needs to develop complex channel matrix algorithms and multi-
path test algorithms.
5. It supports a channel model as the foundation of performance eval-
uation and comparison.
6. Therefore, the challenge and complexity faced by the development
are enormous. And with the increase of channels, the complexity
will increase dramatically. In this case, the design object is 6 6
channel RF and channel testing system.
2. Shared inter-modulated interfaces and mechanisms with software
simulation
In this case, the integrated verification platform fully uses the research
results of subtasks in the case software simulation platform. It truly fulfills
the reusability of project results. This is consistent with the initial target of
the verification platform design. The verification platform is a HIL verifi-

cation system. By the sharing and inter-modulated interface with the soft-
ware simulation platform as well as the design and implementation of
mechanisms, algorithms of the simulation platform can be moved to the
hardware environment for an actual verification.
In order to ensure that all algorithms can be embedded into the HIL
system, the innovative concept that the algorithm can replace the interface
specification is proposed in the case, and the specification is developed.
Therefore, in any algorithm, as long as users follow this specification to
write software algorithms and upload them to the site database in the test
task, it can be replaced in the verification platform, so as to verify the HIL
simulation in hardware environment. Therefore, it can truly have perfor-
mance evaluation and indicator verification for the users software algo-
rithm. It greatly improves the efficiency of users software development
testing and verification and greatly increases the flexibility of the software
algorithm verification.
The interface specification of replacable algorithm is the innovation of
this case, which is the first one in China. On the one hand, if users can just
follow the algorithms defined on the interface specification, they can make
the HIL simulation and verification on the integrated verification platform,
verifying their own software algorithms fast, flexibly and efficiently. On the
other hand, users can effectively segment the needed verification software
according to the interface specification, dividing chunks of codes into
several small pieces for verification, so as to position errors more rapidly.
It greatly reduces the workload of software development and improves its
working efficiency.
3. Multiple automatic testing and verification systems of different
manufacturers
The key techniques of IMT-A include various aspects, covering from the
physical layer to the application layer. Such techniques without standards
cannot be accomplished by using only one test equipment or one scheme.
Currently in 2G and 3G test schemes, the mixed operation systems with
multiple test equipment have been adopted. As the IMT-A technology is
still under development, so there isnt a single equipment, instrument or
scheme covering all technical standards and indicators. Test and verifica-
tion of IMT-A is bound to be a mixed system that integrates the existing test
instrument, effectively uses resources and complement one another, which
must also be an ultra large mixed system.
In terms of multiple equipment, during mixed operations, an integrated
development progress is needed. Otherwise they cannot work orderly and
the resources cannot be shared. Moreover, there may exist technology and
resourcescollisions. The developed platform for the IMT-A key technology
test and verification is an open platform for sharing, so a lot of work such as
control, coordination and maintenance should be done in the following
facets: all equipment used in order and simultaneously, their using
rightusage specification and operation procedure, remote openness, test

data storage, test data playback, etc. Therefore, if the test equipment is
not integrated with the system, it is just a group of disordered equipment, on
the basis of which the system development becomes impossible, and the
effective resources can not be used to finish what single equipment cant
accomplish and the orderly multi-task tests. If so, it cannot be called a test
and verification platform.
Therefore, it is necessary to make an overall integrated development for
purchased and developed equipment, unifying a platform that specifically
provides solutions and effective management for testing the key IMT-A
technologies. Then via the interface of the simulation platform and the
services platform, the test and verification of key IMT-A technologies are
completed in a coordinated way. Meanwhile, an open website shared for
log-in and use via the network is designed and developed, thus making an
active remote shared lab for the long-term use.
Such integration of equipment with various manufacturers and functions
is unique in China. The established platform for test and verification of key
IMT-A technologies is also the first large-scale remote and combined test
platform in the field of mobile communications in China.
Based on our research, at present, there isnt a similar large-scale pubic
integrated test platform in the IMT-A technology domain in international
mobile communications. The testing and verification platform of IMT-A
key technologies is similar in concept with the integrated products group of
Next Generation Automatic test system (NxTest) brought up by the
United States Department of Defense (DoD). NxTest program is still in
implementation, which is substantially at the same progress stage as our key
IMT-A technology test and verification platform. NxTest program explic-
itly supports the system-level RF test but does not support the code-level,
module-level and subsystem-level tests. That means the purpose of NxTest
is merely the product-level and the system-level integrated test, rather than
supporting the whole stage of R&D as the key IMT-A technology test and
verification platform. Thus, the established key IMT-A technology test and
verification platform will be a global-leading large-scale integrated test
platform which supports IMT-A R&D.
4. Remote test configuration and operation technology that based on the
Browser/Server (B/S) architecture.
On-site equipment for measurement and control undergoes the bottom-
up development, with functions gradually extended to networking, open-
ness and distributedness. However, computer networks permeate the Inter-
net from top to bottom, till it makes a direct communications with the
bottom on-site equipment. Therefore, the Internet-based remote monitoring
system in this test case came into being accordingly. It connects the
equipment that is distributed at different sites and can independently
accomplish specific tasks through the on-site control network (or the
on-site bus), enterprises the network and Internet. Thus it can be an
all-round distributed equipment status monitoring and fault diagnosis sys-

tem for the purpose of resources sharing, coordinated work and centralized
management. This system is the product of the development of Internet, the
TCP/IP network communications technology, the field bus technology, the
equipment fault diagnosis technology, the browser technology and the
network database technology.
The Internet remote measurement and control system includes the fol-
lowing functions: (1) collecting the test equipment data; (2) transmitting the
test data; (3) analyzing and processing the test data; (4) controlling the
equipment status. Collecting the test equipment data means to choose the
signals that can best embody the test objectivesaccording to different
objectives. The signals are collected and stored by the on-site test equip-
ment. Transmitting the test data means to send the collected signals to the
remote test analytic system via Internet. Analyzing and processing the test
data means that the test and analysis center analyzes and processes the
received data, extracts the error signals, and makes comparison with fault
documents in the system database. Corresponding solutions are proposed
and the on-site equipment is controlled by Internet and the on-site bus.
In this Internet-based remote monitoring and control system, the equip-
ment and parameters on the key IMT-Advanced technology verification
platform are configured by software which can also make intellectualized
judgment and correction. The B/S architecture in the system can support
many users and terminals for real-time and non-real-time operations via
various network remote control and verification platforms.
The B/S architecture-based technology of the test server, remote testing
configuration and operation of the Internet-based remote measurement and
control system in this case project is the first practitionersincommunications
test platform.
5. Developing test configurators, multi-test task operation managers that sup-
port remote operation
During the project development process, we focus on not only the
advancement of the platform verification technology but also the platform
overall architecture and future sustainability. For this purpose, test
configurators, multi-task operation managers are included in the design of
the platform architecture.
As a bridge for customers to realize the test through managers, the
configurator also provides the test system with a configuration platform.
Clients can choose the test scheme through the local general control inter-
face or remote websites. And managers will choose configurators for
equipment according to the clients scheme. As the bottom level of the
whole test platform, the configurator is the most basic platform to execute
the automated test.
This makes it easier for managers to configurate the client scheme. By
controlling test instruments, configurators collect and store the data.
Configurators are a important and indispensable parts of the IMT-Advanced

test platform.
Managers integrate the deployment and management functions with the
software development platform, providing the standard interface to make
the data exchange. Around the core of the multitask sequence, managers
develop the architecture of the whole test system, define the implementation
process, collect results and report the recording database to communicate
results.
Compared to the previous management module, the module in this
project has a great advantage. On the one side, it supports the test module
that developed by development tools of various application programs and
can independently customize the test sequence in a specific tested part so as
to meet the requirements of the customers independent test. On the other
side, this manager module possesses fully customizable the user interface
which can offer the humanized interface for better human-computer inter-
action measurement and control. Moreover, it can control various instru-
ments and equipment with the help of the software control test system so as
to adapt to flexible and diverse test requirements.
The test platform of the configurator and manager applied in this case is
the first practitioners in communications.
6. Developing large-scale test database that supports fast indexing with all
kinds of configurator interface.
Database is the core of the system, whose design is directly related to the
system implementation efficiency and stability. Only when a rational data-
base module is designed can the difficulty of system program and mainte-
nance be reduced, and the actual operating efficiency increased.
Established in the R&D stage of the testing and verification platform of
key IMT-A technologies, the test database of IMT-A covers plenty of data
parameters of mobile communications and all the frame informations of the
project. It will be the first large-scale shared database in the future mobile
communications R&D domain in China. This is of great significance and
value for carrying out research, development, experiment, test and educa-
tion on mobile communications in China in the future.
The large-scale IMT-A R&D stage test database built in this project is in
accordance with the principles of naming, concurrent search, and reference
of repeated table. And it minimizes redundancy to ensure the completeness
and accuracy of the data. The database and configurators have various
message interfaces to finish task receiving, data processing and playback
for configurators. Substituting the original traditional polling mode with the
message mode greatly improves the efficiency of the system. When devel-
oping and designing the database, we took every possible conditions of the
messenger and designed a rational universal message passing model which
greatly simplified the programming procedures of client and server as well
as the maintenance of the entire large-scale test database.
The large-scale database is used to store the test instrumentation, the test
task information, the test log data, etc. It also provides data service for the
manager and the adaptor to accomplish all kind of interfaces of the
configurator in the message mode, which realize a complete series of the
process from the remote user input, configuration, testing and data
playback.
II. Architecture design and implementation
The overall architecture design is shown as below and the system is divided
into seven parts.
1. 4-transmitting and 6-receiving types of the wireless link platform which is
based on modular instruments, have achieved synchronization among mul-
tiplex transmitters and among multiplex receivers as well as between the
multiplex transmitters and receivers.
2. Common inter-modulation interface and mechanism of software simulation
platform in which its algorithm can replace the interface specification.
3. Automation testing and verification systems of all the five manufacturers
and seven types of instruments are supported.
4. Test server technology, remote test configuration and implementation tech-
nology based on the B/S architecture.
5. Test configurators supporting remote operation, multi-task implementation
managers and task executors are developed.
6. A large-scale database that includes all kinds of test configurator interfaces
and supports fast indexing is developed.
7. Websites that users can use via network and the general control interface
supporting local and network use are developed (Fig. 5.26).
The verification platform of key IMT-Advanced technologies has com-
pleted the substitutable HIL key technology algorithm, 4-transmitting and
4-receiving, and the verification environment for channel real-time simulation.
As an open and sharable platform, the verification platform allows different
users to make remote verification configuration via network and browser and
to use it both online and offline. Therefore, the verification platform has
developed and accomplished the following function components:
1. Testing and verification configurator
It is designed to input and proofread multi-users verification parameters,
analyze their validity and store the data to the database.
2. Multi-task manager
It is designed to manage and schedule the testing and verification tasks
requested by multiple users and to interact with the database.
3. Multi-task actuator
According to managers scheduling, it actuates orthogonal multi-tasks
and manages the execution state. Composed of several standardized test
modules, it can get the test results directly and store them in the database.
platform remote GUI (website)

Parameter configuration
XML script
Test item platform GUI

Test item
Parameter
configurator configuration
XMLscript
Test status Manager Data

base
(write)
test module
Test sequence Local bin
Test data file
data file
test module
Parameter configration
test module Data base(read) Test execution Storage

(Configration information) (Call instrument to execute) (Test data)
Fig. 5.26 Blueprint of overall architecture
4. Test database
It used to record the relevant user and the test data.
5. Database and configurator interface
It is used to finish the conversion from the XML-based user request to the
storage data.
6. General interface
It enables users to login the verification platform through local login or
remote login.
7. SWAN website
It is designed for remote users, supporting online and offline use via
browser.
III. Applications of integrated test
1. Examples of standardized test scheme
(1) General parameters measurement module
The function of this module is to control the spectrum analyzer
E4440A, by which to test indicators of wireless devices, mainly includ-
ing channel power, power spectrum, etc. The E4440A configuration
interface is shown in Fig. 5.27.
Fig. 5.27 Configuration

interface of E4440A
(2) Control interface of 53,132 and E4438C

The function of this module is to control the vector signal generator
E4438C and digital frequency meter 53,132. By controlling E4438C,
the vector signal is generated. By controlling the digital frequency
meter 53,132, the central frequency and other parameters are automat-
ically measured. The configuration interface of E4438C and 53,132 is
shown in Fig. 5.28.
(3) PXI LTE RF phase noise test module
The phase noise module based on the modular instrument. It uses NI
PXI to make the PXI LTE RF phase noise test. Automatic measurement
of the phase noise at different frequency bands is performed by means
of transmitting and receiving LTE RF of PXI board of NI 5663 and NI
56732. Figure 5.29 LTE RF input.
(4) Control interface of E4440A and E4438C
This module aims to control two vector signal generators E4438C
and the spectrum analyzer E4440A. Controlling two vector signal
generators E4438C can produce signals of 3.01G and 3.00G while the
spectrum analyzer E4440A can make automatic measurements for the
third order intermodulation etc. Figure 5.30 is the parameter configu-
ration of the third order intermodulation.
Transmitting and receiving of 1 by 1
This module aims to control PXI to verify theexperiment of
1 by 1 wireless communication. Transmitter PXI 5611 and receiver
Fig. 5.28 Configuration

interface of E4438C and
53,132
Fig. 5.29 LTE RF input

Fig. 5.30 Parameter

configuration of third order
intermodulation
Fig. 5.31 Path configuration interface of 1 1 package
PXI 5601 are used for parameter configurations. The data of transmitter
and receiver can be seen and can be used to verify the transmitting and
receiving of algorithms withsingle antenna, as shown in Figs. 5.31,
5.32, 5.33, and 5.34.
Fig. 5.32 User self-designed non-package interface
Fig. 5.33 Transmitter and receiver of 1 1 algorithm verification

Fig. 5.34 Parameter configuration of QAM transmitter and receiver
2. Customizable test scheme

The non-package scheme is a user self-designed test scheme which
allows the user to choose the instrument and the link mode while users
only need to fill in basic parameters. Then the website will enter
configurators according to the information that users filled in. The
configurator popups the parameter interface that users need based on the
sole verification code put in by the users. Users can fill the parameter
according to their own needs. Then they should complete the parameter
configuration and store the results that need to be stored in the database.
Test order is arranged by manager on the basis of users information. The
final result is stored in the database. When choosing the test link, users can
upload files via the upload button on the platform to replace the existing file,
which means uploading their own algorithm to complete the test link. Users
can choose apparatus of different manufacturers to complete the test. They
can also choose the connection mode of instrument links. As shown in
Fig. 5.32, the user chooses Agilent and PXI for 2 2 transmitter and
receiver, and meanwhile chooses C8 channel for both routes.
(1) Algorithm verification of 1 1 transmitting and receiving
In this module, verification experiment of 1 1 algorithm is done by
controlling PXI. Transmitter PXI 5611 and receiver PXI 5601 are used
for parameter configurations. The data of transmitter and receiver can
be seen in actuation and can be used to verify the transmitting and
5.3 Evaluation and Test on Hardware and Software System-level Co-simulation 277
receiving of algorithms with single antenna, just as shown in the

Fig. 5.33.
(2) Quadrature Amplitude Modulation (QAM) transmitting and receiving.
This module is to conduct experiment on the QAM transmitter and
receiver by controlling PXI. The parameter configuration is made via
the transmitter PXI 5611 and the receiver PXI 5601. The data of
transmitter and receiver can be seen in actuation and can be used to
verify the performance of the QAM transmitter and receiver in real
channels, as shown in Fig. 5.34.
5.3 Evaluation and Test on Hardware and Software

System-level Co-simulation
5.3.1 Composition of Hardware and Software System-level

Co-simulation (Fig. 5.35)
In this hardware and software system-level co-simulation, the physical layer of the
simulation platform is partially substituted by the real physical layer and the
transmission network, which can increase the reality and the instantaneity of the
system.
Design and implementation of the simulation software platform should refer to
Chapter 4 and the design and implementation of the real physical layer adopts the
above-mentioned HIL software and hardware link system. Mapping relation from
the system-level simulation software to the real physical layer should be custom-
ized in accordance with the specific requirements of the simulation evaluation. It
will be introduced in detail in the real case in Sect. 5.3.3 of this chapter.
System software simulation platform
Interface layer
Real time
channel
emulation/
PHY hardware platform Real PHY hardware platform
channel
Fig. 5.35 Hardware and software system-level co-simulation

5.3.2 Key Technologies of Hardware and Software System-

Level Co-simulation
Hardware acceleration based on USRP-RIO
The hardware accelerated simulation technology substitutes the software module

with the hardware module to make the most of its speediness. This hardware
accelerated simulation technology focuses on real-time functions to solve the
bottleneck of system calculations and to accelerate the overall simulation effi-
ciency. As a commercial collect-and-use device that can complete analog of single
base station and single terminal, USRP-RIO provides a high-performance FPGA
board with strong computing power and logic processing capabilities. Compared to
the GPU server, the FPGA board has stronger logic capability for task management
and resource scheduling. In addition, with the fast development in its development
environment and compiler, the difficulty of developing FPGA has been greatly
reduced. To accomplish the hardware accelerated simulation, the key link is divided
into the following three aspects.
(1) Model selection of hardware accelerator board based on FPGA
With the development of the integrated circuit technology, the computing
power and internal storage resources of the FPGA chip has undergone rapid
growth. Meanwhile, the FPGA device has become an ideal algorithm accelera-
tion implementation platform due to its programmability, granularity parallel
ability, rich hardware resources, flexible algorithm adaptability and lower power
consumption. FPGA can easily realize the distributed algorithm structure, which
is very useful for acceleration scientific computing. For instance, most matrix
factorization algorithms in scientific computing requireslarge quantities of mul-
tiplication and accumulation, which can be effectively realized by distributed
arithmetic structures of FPGA. Todays FPGA products have achieved millions
of gate-level design in 65 nm process. And all major companies are seeking to
develop more logical units and more high-performance logic products.
(2) Design of interface and middleware that combines hardware accelerator board
and software simulation platform.
For some computations difficult for FPGA to fulfill, C language can help. At
the end of the design, the intermediate result running on hardware is imported
into C for visual analysis. Through drawing the graphical waveform and the eye
diagram and so no, we can check whether each part of the system design is
correct. Introducing the simulation based on C/C++ to the design of FPGA has
solved the FPGA debugging difficulty, as well as the frequent inconsistency
with the system simulation logic. The concrete elements of the FPGA design
includes a virtual adaptation mechanism, middleware design and re-design
calculation.
(3) Reconfigurable FPGA logic design
Reprogrammable is an important feature of FPGA. The FPGA-based system
can be changed and reconfigured easily in the development stage and in the
using process, which increases the overall gain.
The implementation process of hardware accelerated simulation is as

follows.
First, considering simulation design requirements, the hardware parallelism
and the software system architecture, divide and design rationally hardware
acceleration modules, in oder to significantly reduce the calculation time over-
heads of the simulation core modules.
Next, design of the FPGA-based hardware acceleration board system,
involves design of adaptation based on C simulation codes, design of
middleware (including board level support package) of interface layer, and
design of the reconfigurable computing.
When the host receives the signal that configuration data loading is finished,
the initial data will be sent to memory via PCIe interface. When it is transmit-
ted, activating signal is sent to the accelerator through the PCI-E interface.
After that, the accelerator returns the completion signal to the host via the PCIe
interface. When the host receives the completion, the final results can be read in
memory (Fig. 5.36).
As shown in Fig. 5.37, parts of the link in the simulation system platform
adopt hardware implementation. The actual test results of physical bus trans-
mission delay is less than 10 ms. Physical bus transmission delay is defined as
System simulation
platform
Control panel : parameter Data panel Service
configuration, Scheduling, Flow, sending, feed
Command interface back receive
High speed bus
Link level hardware module

UE N
TX RX UE N
UE 1 UE 1
UE
baseband DA upconvert downconvert AD decode
data
constructing synchroniz demodulati

frame in
base-band ation on
FPGA resource 1 FPGA resource 2
Air interface Air interface

Channel
UE N
UE 1
UE N UE N
User time domain
UE 1 UE 1
channel convolution
parameter Up convert to
RF
User
channel
factor
FPGA resource 3
Fig. 5.36 A block diagram of system implementation of software and hardware co-simulation
System software simulation platform
Interface layer
Real time
channel
emulation/
PHY hardware platform Real PHY hardware platform
channel
Fig. 5.37 Software implementation link
the delay of the whole process in which the data is sent from the simulation
platform to the air interface via the link transmitter module, and then goes back
to the platform after receiving and demodulating decode through the link
receiving module.
This kind of software and hardware co-simulation methods can make full
use of the hardware high-speed processing capabilities, and enables some links
system simulation performance to be close to the real-time level. Combined
with the relatively perfect system function of the system simulation platform, it
can better simulate the system application scenarios with high indicator
requirements for system transmission delay.
Distributed data sharing technology
1. Reflective memory technology

Reflective memory is a distributed data sharing system that enables multiple
independent computers to share a universal dataset. The reflective memory
network can store the independent backups of the whole shared memory in
each subsystem. Each subsystem is entitled to full and unrestricted access,
which allows them to write and modify the local datasets with a high-speed
local memory. When data is written to the local reflective memory to backup,
high-speed logic synchronization transfers it to the next node of the loop
network. Each subsequent node will write new data to local backup, and then
send it to the next node in the loop network. When the information goes back to
the initial node, it will be removed from the network, and then, according to
specific hardware and node quantities, all the computers on the network will
have the same data at different addresses in a microsecond. Thus, the local
processor can read the data at any time without access to the network. In this
way, each computer can always have the latest local backup of the shared
memory.
2. Advantages of reflective memory technology

(1) Time-delay stability
A significant advantage of the reflective memory is the stability of
timedelay. As long as the network bandwidth isnt overrun unexpectedly,
the timedelay wont increase significantly. The reflective memory board
(node) includes a local memory, embedded interfaces and an arbitrary
logic that provides the access channel for the host and the reflective memory.
In addition, the reflective memory board can be installed in a physical way or
be connected to a variety of computer buses which include Versa Module
Eurocard (VME), PCI/PCI-X, Compact PCI, PCI Express or other standard/
special systems that can integrate PMC slots. This method enables popular
workstations to be connected with one-board computer via the reflective
memory.
(2) Standard-exceeding LAN
The reflective memory board provides several features beyond standard
network, such as the global memory, high-speed data transmission and
software transparency, all of which make the reflective memory the most
attractive multi-machine communication solution. Compared with the cost
produced by additional development time, testing, maintenance, documen-
tation compiling and additional CPU of traditional communication means,
the reflective memory is more cost-effective.
PXI multi-computing specification
In November 2009, the PXI multi-computing (PXImc) specification released by

PXI Systems Alliance defined a Non-Transparent Bridge (NTB) of PCI Express
with low-cost and readily available technologies, which is the technical require-
ment of software and hardware for remotely connecting two or more intelligent
systems based on the PCI or PCI Express interface. The PCI Express bus provides
actual data throughput of multi-gigabit per second and stable microsecond delay.
Therefore, it meets the requirements for applications of software and hardware
co-simulation of multiple physical terminals.
1. PXImc principle
The following figure illustrates this concept. Systems A and B can fully
control the resources allocation in their own domain. Meanwhile, NTB does
not affect the algorithm of resource allocation of any system.
Using a NTB, two systems on the above figure are connected together via PCI
Express. Response of NTB to resources from the root complex systems is similar
to other PCI terminal request which is achieved through requesting a certain
amount of physical address space. Then the system BIOS allocates a specific
range of physical addresses to NTB. When this resource allocation occurs
simultaneously on the System A and B, NTB will obtain resources in both PCI
domains. As shown in Fig. 5.38, the address space obtained by NTB within PCI
Fig. 5.38 NTB wont affect the algorithm of resource allocation of any system
Fig. 5.39 Communication mechanism within two PCI domains using NTB
domain of System A will serve as a window into the physical address space in
the PCI domain of system B while the address space of System B will be a
window into the physical address space within PCI domain of System A.
After finishing resources allocation of System A and B, NTB transfers data
between the two systems according to memory mechanisms. These mechanisms
include memo registers for transferring data and doorbell registers for interrupt
request, transferring large amount of the address space to the address space
by NTB.
Figure 5.39 uses the communication mechanism within two PCI domains
using NTB. The PXImc specification developed by PXISA defined specific
Table 5.3 Bandwidth value of various PCI Express links

The theoretical The theoretical
value of unidirectional value of bi-directional
PCI Express link Versions transmission rate transmission rate
x4 1st generation 1 GB/s 1 GB/s 2
x16 1st generation 4 GB/s 4 GB/s 2
x4 2nd generation 2 GB/s 2 GB/s 2
x16 2nd generation 8 GB/s 8 GB/s 2
requirements for hardware and software components, therefore providing a

standardized protocol for the communication between PCI or PCI Express
systems. In term of hardware, some issues have been resolved to allow two
independent systems to communicate directly via the PCI or PCI Express.
Whats more, in terms of software, a communications architecture has been
established, allowing any system to detect and configure their own resources in
order to realize communications with other systems.
2. PXImc bandwidth and delay performance
Since PXImc uses PCI Express as the physical communication layer, the
performance of PXImc link depends on the type of PCI Express interface.
Table 5.3 lists the theoretical bandwidth value of various PCI Express links.
In PXISAs experiment, the performance of a typical PXImc link is compared
with other alternatives. In a fixed configuration, a 6 s one-way delay and a
670 MB/s throughput were measured. Compared with Gigabit Ethernet, these
indicators show a ten times increase in bandwidth and 100-fold reduction in
delay. This confirms the fact that PXImc is an ideal interface to build the
multicomputer test and control systems of high-performance.
3. Multi-core processor is supported for distributed process
In the hardware and software co-simulation test and evaluation, there must be
a communication interface with highbandwidth and lowlatency that can assign
and process tasks at different partitioned nodes. In this distributed processing
system, FPGAand Digital Signal Processors (DSP) can be used to meet the
requirements for acceleration of the real physical layer. However, to use the
system simulation software platform, the existing 86 software IP should be
applied and computing should be accomplished in a fixed mode instead of the
floating-point mode. For these cases, the PXImc technology can use the PC with
the latest multi-core CPU to be the external computing node, thus to create a
distributed processing system. The figure below shows an example of a feasible
distributed computing system that uses the PXI Express system and the PXImc
interface board (Fig. 5.40).
In this example, the master controller is responsible for collecting data from
the I/O module, and subsequently allocating it to four 86 compute nodes by the
PXImc link. Depending on the requirements of processing performance, the
computing nodes can be either an ordinary PC or a high-end workstation.
Fig. 5.40 An example of a distributed computing system that uses PXI Express system and
PXImc interface board
Therefore, the PXImc specification made by PXISA expands the distributed

computing function of the PXI platform, providing an interoperable scheme for
the connection of multiple intelligent systems via a communication interface
with high throughput and low delay, so as to meet application requirements for
software and hardware co-simulation test and evaluation.
5.3.3 Case Study of Simulation

Case1: System Simulation PlatformwithPhysical Layer Acceleration
Based on USRP-RIO (Fig. 5.41)
To simulate the interference of multiple users for adjacent cells to a valid user, the
PXI bus architecture is used as the core, and USRP-RIO is used to simulate the real
physical channel so as to complete the system configuration of upper level param-
eters and the downlink simulation for the LTE physical layer.
I. System architecture
The system simulation platform architecture based on USRP-RIO physical
layer acceleration is shown as Fig. 5.42. By integrating the link-level test into a
complete system simulation platform, the multi-cell HIL test and simulation is
realized.
While the traditional test scheme only considers the single link or a limited
number of interference, this HIL scheme added the HIL performance verifica-
tion of the transceiver to be tested in the scenario of real multi-cell multi-user
scheduling. This scheme aims especially to verify the performance of algo-
rithms that involves multiple cells, for example the ComP, synchronization of
carrier aggregation, etc.
Fig. 5.41 Composition ofsystem simulation platform with physical layer accelerationbased on
USRP-RIO
Multi user multi-cell

systems
Simulation platform
System simulation
interface
parameter passing through PXI-mc
USRP-RIO RF USRP-RIO
RF Air interface or channel
Fig. 5.42 System simulation platform architecture based on USRP-RIO physical layer
acceleration
Compared with the traditional system, this scheme improves the hardware
acceleration and authenticity of the system simulation platform. As for the
implementation of traditional system simulation platform, the way equivalent
SNR mapping into BLER will result in a loss in accuracy. Especially for
complex transmission (equilibrium) scheme, the instantaneous SNR greatly
varies in each time-frequency location. Besides, it is difficult to evaluate
System simulation
platform
8381
8381
PXImc model8381, 8383, 8384
1085 PXImc PXImc PXImc PXImc

8383 8384 8383 Channel emulator 8383
CPU adapter adapter adapter adapter
MXI
USRP-RIO(Tx) MXI USRP-RIO(Rx)
MXI USRP-RIO(Tx) MXI USRP-RIO(Rx)
Fig. 5.43 Hardware connection frame
system performance under a non-steady-state, such as initial synchronization

and frame format switching process. However the verification method of this
platform can effectively simulate the above scenario.
Both the HIL and system simulation have considered the influential factors
such as the actual channel environment and medium RF.
The hardware connection frame of the platform is shown below (Fig. 5.43):
Specifications of system simulation platform are as follows:
1. System simulation platform supports 19 cells.
2. Each cell supports 50 users.
3. It supports LTE DL physical layer MAC protocol (data channel).
4. SISO supported currently.
5. Standard channel model (or actual scenarios playback) generates chan-
nel factors.
A user takes the real-time test on the actual physical channel. During
the process, the system simulation platform will transfer related user configu-
ration parameters to the transmitter, the receiver and the channel simulation
terminal via a high-bandwidth data board. And physical signals will be sent
by the USRP-RIO (Tx) through the air interface or the channel simulator.
After it is received and demodulated, the feedback information will return to
the system simulation platform via the high-bandwidth data transmission
board.
II. Scheme design

Firstly, setparameters of system-level cells and usersandsetthe links. The
initial parameters of all users can be obtained.
Secondly, for real-time test level links, the system simulation platform will
generate the MAC schedule information of the user in real time according to
the scheduling algorithm. On the basis of user feedback Channel Quality
Indicator (CQI) and other measurements, for real-time test level users, the
observed values such as measured CQI are reported, and for other users, the
simulated CQI and other measurements are used for feedback.
Thirdly, for real-time test level link, the system sends the downlink sched-
uling information and the channel parameter of the base station to the hardware
platform, and then the platform will generate downlink signals at cell level
according to the real-time scheduling information. The channel generation part
in the hardware platform will generate real-time channel information on the
basis of channel parameters, and then adding the actual air interface signal
after channel fading to the RF signals.
Fourthly, the simulation user link calculates Estimated Signal-to-Noise
Ratio (ESNR) according to each carrier signal and interference power. Based
on the ESNR look-up table, the link gets BLER probability to judge the current
ACKnowledgement (ACK). Real-time test level users obtain each carriers
received signal through demodulating actual transmitted data frame and esti-
mate the signal power. Meanwhile, they will also make OFDM demodulation
to downlink signals in other cells (non-target cells), and calculate actual SNR
based on RB-level scheduling information from the base station, and also
measure actual information such as CQI and decode them to get ACK
feedback.
Fifthly, the base station calculates scheduling results of the next TTI in line
with all scheduled users feedback of ACK and CQI measurement value and
makes iteration of the next scheduling period.
III. System implementation
1. Interface mode
Both of the universal processor servers and the heterogeneous platform
of FPGA adopt the high bandwidth data transmission card for connection.
The interactive throughput and delay can meet industrial requirements.
Specific data parameter configuration is shown in the subsequent section
of the interface agreement implementation.
2. Scheduling mode
(1) The total program of system platform is operated on universal CPU, and
the bottom level calls the processing section of FPGA.
(2) Matlab is used to write the processing part of the system simulation.
And the interface of the high bandwidth data transmission board is used
to read and write in the agreed interface format.
(3) The processing section of the test link is divided into the upper com-
puter (CPU) and the lower computer (FPGA). Wherein the upper
computer is software and is responsible for reading and writing the

high-bandwidth data transmission board in accordance with the agreed
interface format as well as initializing transceiver with the read user
link information. The lower computer uses hardware programming
language to write the FPGA program, completes the physical layer
transmitting, receiving and channel passing, as well as giving imple-
mentation results feedback to the upper computer. Eventually the upper
computer writes the feedback into the high bandwidth data transmission
board according to the agreed interface.
3. Interface agreement (Tables 5.4 and 5.5)
The following is the actual received parameters clusters, and transmited
parameters clusters.
(1) Receiving cluster of data (Fig. 5.44)
(2) Sending cluster of data (Fig. 5.45)
The mapping relation of MCS and modulation is shown in the following
table (Table 5.6).
Table 5.4 MAC- > PHY (receiving)

Parameter name Range/value/remarks Data type/size
Device number INT 1(I32 bits)
Subframe ID 1 11,000 (range) INT 1
User ID (RNTE) 1 INT 1
Cell ID 1 INT 1
Base station ID 1 INT 1
Channel type INT 1
RB resource allocation of target users. 4 values (1bit-1RB from MAC, INT 4
represented by four INT)
RB resource allocation of other users 36 values (1bit-1RB from INT(4 9)
MAC, represented by four INT)
MCS of target users. One enumeration type is: INT 1
MCS 0(QPSK,rate_0.12)
MCS 10(16-QAM,rate_0.33)
(continued)

Parameter name Range/value/remarks Data type/size
Modulation typeof other users The same as the above MCS of INT(1 9)
target users.
Carrier Frequency Float 1
(32 bits)
Transmit power Float 1
The path loss value Float 1
Doppler fd Float 1
Target users power offset 10 10(0.1 dB, Float 1
U11 < 2,9>)
Other users power offset 10 10(0.1 dB, Float 1 9
U11 < 2,9>)
RB-level interference from adjacent cell 1 50(0.1 dB, U18 < 9,9>) Float 100
Table 5.5 PHY- > MAC Parameter name Range/value/remarks Data type/size
(transmitting)
Device No. INT 1
TTI No INT 1
UE ID INT 1
Cell ID INT 1
BTS ID INT 1
Feedback CRC INT 1
Feedback CQI FLOAT 1
Feedback ESNR FLOAT 1
Fig. 5.44 Receiving cluster

of data
Fig. 5.45 Sending cluster

of data
Table 5.6 The mapping No. MCS Modulation mode Code_rate

relation between MCS and
1 0 QPSK 0.12
modulation
2 1 QPSK 0.15
3 2 QPSK 0.19
4 3 QPSK 0.25
5 4 QPSK 0.30
6 5 QPSK 0.37
7 6 QPSK 0.44
8 7 QPSK 0.51
9 8 QPSK 0.59
10 9 QPSK 0.66
11 10 16-QAM 0.33
12 11 16-QAM 0.37
13 12 16-QAM 0.42
14 13 16-QAM 0.48
15 14 16-QAM 0.54
16 15 16-QAM 0.60
17 16 16-QAM 0.64
18 17 64-QAM 0.43
19 18 64-QAM 0.46
20 19 64-QAM 0.50
21 20 64-QAM 0.55
22 21 64-QAM 0.60
23 22 64-QAM 0.65
24 23 64-QAM 0.70
25 24 64-QAM 0.75
26 25 64-QAM 0.80
27 26 64-QAM 0.85
28 27 64-QAM 0.89
29 28 64-QAM 0.93
4. Implementation of real-time receiving and transmitting links

SISO implementation is considered now.
The physical layer acceleration system of software and hardware
co-simulation platform is mainly used to transmit and receive the downlink
data of the LTE physical layer. The operational process of the signal flow is
shown as below (Fig. 5.46).
The following is the system block diagram of the implementation of
transmitters and receivers of downlink in the physical layer (Fig. 5.47).
USRP UE PC
Parameter
configuration
eNB UE MAC data
stream
Measurement
PHY Tx PHY Rx parameter
RF front end
Fig. 5.46 Operational process of signal flow
Fig. 5.47 The system block diagram of the implementation of transmitters and receivers of
downlink in physical layer
(1) The transmitter:

Firstly, the transmitter generates baseband data signals according to
the users initial scheduling information input by MAC, which
includes the random signal generation of data channel, coding, data
rate match, QAM mapping, antenna port mapping, insertion of refer-
ence signal (synchronous signal), framing and OFDM modulation.
Secondly, it converts the baseband signal to IF through the interpola-
tion filter and the digital up-converter Direct Digital Controller
(DDC), and turns to analog domain by calling DAC. Next, the IF
analog signal is modulated to carrier frequency point via amplifying
and frequency mixing.
(2) Channel generation section:
Channel simulator is configured according to the information input
by MAC such as antenna matching, transmitting power, path loss, and
channel parameters, with which increasing valid channel fading for the
input RF signal. In addition, the channel can be bypass of channel
emulator, instead to direct connection or via air interface or playback
with the channel factor acquired.
(3) The receiver:
Firstly, the receiver transforms the attenuated RF to IF by processes
like frequency mixing and the low noise amplifier.
Next, after IF ADC is adopted, through multistage downsampling,
filtering and orthogonal demodulation, it is transformed to baseband
signals.
Thirdly, in baseband processing, the receiver makes cell search and
time-frequency passing in the frequency band of 1.4 M on the basis of
the synchronization signal. After synchronization, the receiver begins
to modulate the OFDM, decode frame, estimate channel, test MIMO
and decode Turbo, and finally completes CRC decision output. In this
process, the channel factors after channel estimation are used for related
measurements including the CQI measurement.
The interference of the user can adopt the following two ways.
In the first way, the user only has one real-time link, while other
carrier-level interference generated by the high bandwidth data trans-
mission card. The hardware computes the total RB-level interference
power according to the powers of each interference signal and generates
the random equal-value frequency-domain variables to superpose it to
the received signal. The second method is to run all the links of the
target user (including user links in the cell and other base stations) on
the real-time link. In this method, the target link demodulates each base
station link in time division, and superposes the interference directly to
the useful signal according to RB.
In receiver RB-level interference implementation algorithm, AWGN
is implemented at FPGA side and Host side.
Fig. 5.48 Receiver RB-level interference
Fig. 5.49 DMA channel read
Host-side can produce the AWGN noise on each Resource Element

(RE) required by ten subframes at one time and then send it to the
FPGA via DMA. The noise level can be adjusted by inputting standard
deviation of parameters where the upper and lower paths are I path and
Q path noises. 15,600 represents RE (100 * 13 * 12) of 13 symbols
(no noise is added to the first one) (Fig. 5.48).
FPGA, first in a single-cycle timing cycle, reads 15,600 data from
DMA channel, and writes the data sequentially in the predefined mem-
ory for later noise superposition. The cycle will stop after writing, as
shown below (Fig. 5.49):
Then AWGN Module is added between Demapper and Channel
Estimation path, as shown below (Fig. 5.50):
Fourthly, lower computer transmits the feedback to the reflective
memory. Currently, the CRC, ESNR and CQI are feed back.
Fig. 5.50 AWGN adding

module
Fig. 5.51 The estimated value of original band noise channel obtained from LS algorithm
The CQI and ESNR algorithms of the downlink receiverare

implemented. The algorithm is shown as below:
Y = S*H +N
H_est = Y/S = H +N/S
H_filt
H
CQI = S*H/N = H/(N/S)
H_filt/(N/S)
Whereas
N/S = H_est H_filt = H + N/S H = N/S
so
SINR = H_filt/(H_est H_filt)
In the above expression, H_est refers to the value transmitted by

FPGA while H_filt represents the value of upper computer after
filtering.
Algorithm details are as follows:
The estimated value of original band noise channel is obtained
through LS algorithm in the bottom layer of FPGA, and the value is
sent to the upper computer in DMA mode, just as shown below
(Fig. 5.51):
In the upper computer, a total number of 800 (100RB * 8CRS/RB)
estimated channel values at CRS in the subframe of index 5 are
extracted, as shown below (Fig. 5.52).
Fig. 5.52 The estimated value of CRS channel
Fig. 5.53 The estimated channel values of each sub-band are extracted and then sent to filter
module
From the data of a subframe, in the order of index symbol first, and
then index sub-band, the estimated channel values of each sub-band
are extracted and then sent to filter module (as two steps below), as
shown below (Fig. 5.53).
In the upper computer, after across a low-pass filter, an approximate
value of denoising channel is obtained.
The difference of H_est obtained through the LS algorithm and
H_filt after filtering serves as the original estimated value of the noise.
The above two steps are shown as below:
Fig. 5.54 The original estimated value of noise
Fig. 5.55 Averaged CQI value on sub-frame
LTE Host DL.gvi- > LTE Read Channel Estimates.gvi- > LTE Calcu-
late CQI.gvi- > LTE Calculate SubframeCQI.gvi- > LTE Channel
Estimation subband.gvi (Fig. 5.54)
N/S and H obtained in the previous stage are used to calculate CQI and
ESNR
LTE Host DL.gvi- > LTE Read Channel Estimates.gvi- > LTE Calcu-
late CQI.gvi- > LTE Calculate SubframeCQI.gvi->
LTE Calculate SNR for CQI
For CQI, as shown below, in Part 1 the average noise of a RB is
calculated; in Part II, the CQI values in each CRS position is calculated;
finally the average is got on sub-frame (Fig. 5.55).
For ESNR, each bit of RB allocation corresponds to 4 RBs. Each RB
includes 8 CRS, so every 32 CRS is a group (Fig. 5.56).
Since there is disparity between the estimated original channel
power and original noise power, it is likely to have disparity between
the obtained CQI and ESNR. Hence, calibration is needed. The cali-
bration equation is shown below:
CQI/dB 1.8 * CQIraw/dB 10.2 6 CQIraw/dB < 6
Fig. 5.56 SNR calculation
Fig. 5.57 CQI calibration
CQI/dB 1.1 * CQIraw/dB 6 6 CQIraw/dB < 30

As shown in the below figure (Fig. 5.57):
In order to avoid initial synchronization process and solve the
problem that implementation speed of the physical layer is faster than
the MAC layer, the following controls are taken:
(1) Transceiver of the physical layer transmits and receives in real time
and synchronization is implemented once at the beginning and
tracked in follow-up steps.
(2) When MAC layer data packets of different links arrive, they are
inserted into the closed downlink sub-frame.
(3) With the help of the controller, the physical layer coordinates the
switch between transmitter, channel simulator and receiver.
In addition, the SNR can be changed from the parameter
configration in the channel emulator.
5. Implementation of parameters transmission based on PXI-mx.
System simulation parameters are transmitted from the simulation
server. The transmission parameters include channel configuration infor-
mation, physical layer data configuration information, and the handshaking
signals for time sequence processing. Among them, the channel
References 299
configuration information is used to generate the physical layer configura-

tion parameters to be used to transmit the LTE baseband. The details are in
the above agreed interface table. The produced LTE baseband signal,
after up-conversion, is simulated via air interface or channel. After AD
sampling at the receiver and then the down-conversion, the signal is
demodulated and decoded. Then the handshaking signal of ACK/Negative
ACKnowledgement (NACK) is returned to the simulation server for the
next data transmission.
The actual hardware need to be considered. Now is the interface verifi-
cation of USRP-RIO and VST and PXI-mx.
5.4 Summary
This chapter describes the classification, methods and applications of software and
hardware co-simulation test and evaluation. It puts emphasis on the HIL link
simulation technology and its applications. With the HIL technology, the software
and hardware co-simulation platform can be developed on the basis of existing
commercial equipment. The technology evaluation and test at the algorithm stage
can give us earlier test and evaluation results without increasing the product
development cycle. Then this chapter further introduces the methods and applica-
tions of system-level software and hardware co-simulation evaluation and test. In
the cases of link-level software and hardware co-simulation test and evaluation, we
not only introduce the cases where the hardware platform and the software platform
are in the same computing environment, but also introduce the cases of cross-
network remote tests.
References
1. C. A. Valderrama, A. Changuel and et al. A unified model for co-simulation and co-synthesis
of mixed hardware/software systems. European Conference on Design & Test, 1995:180184.
2. R. Ruelland,G. Gateau, T. A. Meynard and J. C. Hapiot. Design of FPGA-based emulator for
series multicell converters using cosimulation tools. IEEE Transactions on Power Electronics,
2003, 18(1):455463.
3. J. Ou, V. K. Prasanna, MATLAB/ Simulink based hardware/software cosimulation for design-
ing Using FPGA configured soft processors. International Proceeding of parallel and distrib-
uted processing symposium, 2005:148b-148b.
4. A. Hoffman, T. Kogel and H. Meyr. A framework for fast hardware-software co-simulation.
Proceeding of the conference on design,utomation and test in Europe, 2001.
5. K. Wei. Software and hardware co-simulation platform. E-World, 2012:128131.
6. W. Zheng. Research and development on new generation of hardware in the loop simulation
platform. Tsinghua University, 2009.
7. B. Heiming. Hagen Haupt. Hardware-in-the-Loop Testing oflutions for Networked Electronics
at Ford[C]. SAE Paper,2005.
8. R. Isermann, J. Schaffint. Hardware-in-loop simulation for design and testing of engine-control

systems. Control EngineeringPractice, 1999, 7(5):643653.
9. dSPACE GmbH, Experiment guide documents for release 4.0[K]. Germany: dSPACE Gnbh.
2003.
10. X. Liu. ECU hardware-in-the-loop real-time simulation test platform[D]. Tianjin: Tianjin
University, 2009.
11. W. Tan. Research on simulation of hardware-in-the-loop and its application in wind power
generation. Qingdao University of Sicence and Technology, 2014.
12. H. Jia and et al. Hardware acceleration verification technology in digital integrated circuit
design[J]. Industrial technology, 2007.
13. Dammann and S. Kaiser, Standard conformable antenna diversity techniques for OFDM and
its application to the DVB-T system. Proceedings of the IEEE Globecom, 2001, 5:31003105.
14. J. Tan and G. L. Stuber. Multicarrier delay diversity modulation for MIMO systems. I IEEE
Transactions on Wireless Communications, 2004, 3(5):17561763.
15. V. Tarokh, N. Seshadri, and A. R. Calderbank. Space-time codes for high data rate wireless
communication: Performance criterion and code construction. IEEE Transactions on Informa-
tion Theory, 1998, 44(2):744765.
16. V. Tarokh, H. Jafarkhani, and A. R. Calderbank. Space-time block codes from orthogonal
designs. IEEE Transactions on Information Theory, 1999, 45(5):14561467.
17. Z. Hong, L. Zhang, and L. Thibaut. Performance of cyclic delay diversity in DAB/DMB. IEEE
Transactions on Broadcasting, 2006, 52(3):318324.
18. Y. Zhang, J. Cosmas, and M. Bard, and Y.-H. Song. Diversity gain for DVB-H by using
transmitter/receiver cyclic delay diversity. IEEE Transactions on Broadcasting, 2006, 52
(4):464474.
19. 3GPP TSG RAN WG1 #46, R1062566. Link evaluation of DL SUMIMO: Impact of
generalized CDD. Seoul, Korea, Oct., 2006.
20. IEEE standard for local and metropolitan area networks part 16: Air interface for fixed
broadband wireless access systems, IEEE 802.162004.
21. R. Van Nee, G. Awater, M. Morikura, H. Takanashi, M. Webster, and K. W. Halford. New
high data rate wireless LAN standards. IEEE Communications Magazine, 1999, 37(12):8288.
22. H. Guo, H. Hu, and Y. Yang. Cyclostationary signatures in OFDM- based cognitive radios
with cyclic delay diversity. IEEE International Conference on Communications, 2009:16.
23. H. Guo and H. Hu. Transmission and receiving apparatus and method of spectral domain
channel multiplexing transmission system (Invention patent application number: 2009 1005
4524.8).
24. H. Guo and Hu Honglin. Transmission and receiving apparatus of spectrum multiple access
system and uplink and downlink access system Invention application number: 2009 1005
6374.4).
25. 3GPP TS 36.201: Evolved Universal Terrestrial Radio Access (E-UTRA); Physical Layer
General Description.
Chapter 6
5G Hardware Test Evaluation Platform
Co-simulation of software and hardware provides a more realistic testing and

evaluation environment to some extent, improving the efficiency of test and evalu-
ation. But it is still different from the real test environment. Therefore, a real
hardware test platform and the environment are essential to the process of develop-
ing new technology. On July 15, 2016, the US government announced that 400 mil-
lion US dollars will be invested to the Advanced Wireless Research Initiative
(AWRI) led by National Science Foundation (NSF), which will build four city-
scale experimental test platforms and deploy advanced future wireless research
projects. The program will use two models: (1) 85 million US dollars will be
invested through public-private partnership to deploy and establish an advanced
wireless experimental test platform. Each platform will deploy software defined
wireless networks in the urban area, allowing academics, entrepreneurs and wireless
companies to test, evaluate and improve their technology and software algorithms in
real-world environment. (2) The NSF will invest US$ 350 million to support
advanced wireless communications theory and technology research. These wireless
experimental test platforms and fundamental researches will support academic and
business to develop and test advanced wireless communication technologies.
As the worlds major telecommunications equipment suppliers started R&D
work of 5G mobile communications system, the mainstream standardization orga-
nizations in the world all have seen the urgency of 5G technology development, and
started corresponding R&D programs to promote the relative standardization work.
Since 5G introduces new scenarios and new frequency bands, the requirements for
5G channel testing and modeling, as the basis for wireless communication protocol
development, are becoming more and more important. In this context, National
Science and Technology Major Project and 863 National High Technology
Research and Development Program (HTRDP) in China also place the R&D of
5G test platform in priority. In 2013, the HTRDP, Early Stage R&D of 5G Mobile
Communication System (Phase I), facing the requirements of mobile communi-
cations in 2020, proposed to complete the technology research of 5G mobile
communications technology evaluation and testing and verification. It includes

DOI 10.1007/978-3-319-61869-2_6
302 6 5G Hardware Test Evaluation Platform
two aspects of works contents: (1) studying evaluation and testing methods of 5G
mobile communication network and wireless transmission technology;
(2) establishing the corresponding 5G technology simulation testing and evaluation
platform, and finishing the evaluation and testing for 5G mobile communication
network and wireless transmission technology. In 2014, National Science and
Technology Major Project, in China wireless innovative technology test platform,
created public trials, verification and testing platform for wireless innovation tech-
nology of the 3GPP R12 and its follow-up standards. The project proposed to further
establish the multi-scenario real channel base with the characteristics of classic
Chinese environment to realize field trial in lab, and to develop the real channel
scenario system-level hardware and software co-simulation system with multi-cell
multi-user to support the research and innovation of new wireless technology. It also
proposed to develop and construct an open and shared, flexibly configured, new
technology R&D-faced, standardized and new technology applied, and multi-cell
and multi-user supported co-simulation, experiment and test platform of system-level
software and hardware. In 2015, National Science and Technology Major Project in
China further constructed the international standard evaluation environment of
IMT-2020 for 5Gcandidate technologies and international standardization. It
includes the simulation evaluation platform as well as test and verification platform
for IMT-2020 candidate technologies. The two platforms can complete feasible and
practical performance evaluation of potential network architecture, key technologies,
algorithms, protocols of IMT-2020, etc., and support IMT-2020 candidate technol-
ogy research, international standard-setting and follow-up product R&D. Based on
the above platforms and facing the requirements of IMT-2020 technology,
researchers can complete the evaluation and verification of massive MIMO array,
UDN, HFB communications, M2 M enhancement, D2D, C-RAN, SDN, NFV,
Content Delivery Network (CDN) and other IMT-2020 candidate technologies.
Overall, the current test platform presents the technology features of universal-
ization, platform-based and software-based. This chapter will introduce a typical
hardware test platform used for 5G evaluation. It includes the first parallel channel
sounder platform in industry used for channel measurement and modeling and
MIMO OTA platform based on a specified channel model, the first platform of
hardware and software supporting software open source community of 5G terminal
and base station.
6.1 Overview of Typical Hardware Platform Used

for 5G Evaluation
The constitution of typical hardware test platform of 5G evaluation is shown in

Fig. 6.1, which includes a parallel channel sounder platform of channel measure-
ment and modeling, a MIMO OTA platform of designated channel model, a
platform of software and hardware of open source community, and terminal and
base station system based on the general purpose processor.
6.2 Test Platform of MIMO Parallel Channel 303
Base Terminal
station
Tested Baseband Baseband

RF module RF module
system: module module
The parallel channel

Sounder platform of
channel measurement
and modeling
Testing
system
The MIMO OTA platform of designated
channel model
The joint platform of software and

hardware of open source community
Fig. 6.1 Typical hardware platform constitution used for 5G evaluation
6.2 Test Platform of MIMO Parallel Channel
6.2.1 5G Channel Test Requirements
As mobile communication progresses towards 5G, the physical layer characteris-

tics, such as HFB (millimeter wave) and multiple antenna (Massive MIMO), will
bring the evolution of its new air interface protocols. Wireless channel measure-
ment and modeling have always been the fundamental R&D topic of every gener-
ation of mobile communication systems. For example, the WINNER, the European
Unions sixth framework project, uses the channel sounder from Elektrobit com-
pany to perform a wide range of channel test of wireless scenarios, and proposes
WINNER channel model.
The main performance requirements of 5G channel test are as follows:
1. Abundant application scenarios
IMT-2020 defines a wealth of application scenarios. These scenarios can be
summarized into four characteristics: (1) seamless wide-area coverage scenario
which requires the experienced rate of users can reach 100 Mbps; (2) high-
capacity hot-spots scenario which satisfies the users experience in centralized
area, such as large-scale concerts, stations and other areas with high population
density and traffic volume density; (3) high mobility scenario, supporting the
mobility speed of over 350 km/h; (4) 3D space measurement and modeling,
covering the scenarios of high buildings and dense space.
2. Coverage of a new spectrum range

5G candidate frequency band will go beyond 6GHz and is seeking the
applications of a higher frequency band and millimeter wave band. The Federal
Communications Commission (FCC) in the US has announced in July 2016 that
the millimeter wave bands, such as 28 GHz, 37 GHz, 39 GHz, and 6471 GHz,
would be used for the future 5G mobile communication systems and applications
[1]. So it is necessary to explore the propagation features of unknown electro-
magnetic waves of HFB and millimeter wave.
3. Multi-channel, high-precision and large bandwidth
With the introduction of new 5G spectrum and new technologies, such as
massive MIMO and millimeter wave technologies, the requirements for techni-
cal parameters of channel test equipment are becoming higher and higher, which
must support the real-time generation and reception of multi-channel RF data
stream; signal bandwidth up to 2GHz; center frequency up to millimeter wave
band; high-precision synchronization among multiple channels.
4. Raw data acquisition of channel
The large bandwidth and high mobility of 5G cause a large amount of channel
measurement data captured at the time of channel measurement. Traditional
channel measurement device collects channel impulse response files, while the
5G channel test needs to capture the raw data of channel, so as to make deep data
mining to support the fine channel modeling requirements such as channel
fingerprint.
6.2.2 Status and Shortcomings
Many universities and research institutes have made researches on channel mea-
surement and modeling of some scenarios, but have not covered the new scenarios
defined by IMT-2020. Moreover, the majority of previous channel model studies
are focused on the derivation of the 2D channel model, not for 3D spatial features. It
is very necessary to systematically carry out channel measurement and modeling of
new scenarios. At the same time, the channel measurement equipment used in
previous researches mainly adopted serial channel test equipment. For 5G specific
scenarios, this equipment cannot be satisfied with 5G MIMO test and measurement
requirement.
In general, classic serial channel sounder consists of a pair of transceivers and
multiple antenna arrays and high-speed RF switches configured at both of trans-
mission and reception, as shown in Fig. 6.2 . In measurement, one transmitter sends
predefined waveform, such as time domain sequence. The transmitter and receiver
periodically and synchronouslyswitch antenna channels to capture air interface
signals. The receiver can correlate the received signal with local sequence and
obtain CIR of each transmitting and receiving antenna pair. The channel parameters
can be extracted by deriving CIR data using the parameter extraction processing
algorithm. The processes are shown in Fig. 6.3.
Transmitting channels Receiving channels

and antenna arrays and antenna arrays
Fig. 6.2 Schematic diagram of serial channel sounder (single-channel antenna array)
The
Predefined The capture Channel
transmission
transmission of received parameter
of predefined
sequence signal extraction
waveform
Fig. 6.3 Flow chart of channel measurement of serial channel sounder
Tx switching seq Rx switching seq
1
2
3
1
2
3
RX 4
time
Fig. 6.4 Time-division multi-channel
As shown in Fig. 6.4, the channel sounder of this architecture has advantages
that the channels are orthogonal in different ways. The calibration is convenient
and the post-processing is relatively simple. This scheme, however, is restricted
with the measurement period of all ergodic channels, so the channel coherence time
is short. The scheme can only test channel characteristics when it is unchanged or in
slowly changing static and quasi-static scenarios. This means it is not suitable for
channel test under high mobility scenario. In addition, traditional serial channel
sounder is limited by early equipment development conditions. It preserves CIR
files, rather than raw channel test data, limiting the accuracy of channel Doppler
measurement.
In other words, using the channel sounder with this architecture cannot measure
time-varying channel when there are a large number of transmitting and receiving
antennas. In the millimeter wave band, the coherence time of channel decreases as
the carrier frequency increases. And when massive MIMO scenarios are consid-
ered, traditional serial channel sounder fails to meet the test requirements.
6.2.3 Key Technical Challenges
5G channel testing and measurement is urgently needed with the parallel channel
sounder. However, the development of parallel channel sounder has to face the
following difficulties and challenges.
(1) Synchronization across multiple channels
For the parallel channel sounder, picoseconds level synchronization
accuracy among the MIMO channels both in Tx and Rx sides must be achieved
in order to guarantee high spatial and temporal resolutions for channel
parameters estimation. This goal is much challenging because of the differ-
ences between clock Phase-Locked Loop (PLL) circuits across multiple RF
channels.
(2) Real-time storage of massive raw measurement data
For the parallel channel sounder, multi-channels raw data with a high
sampling rate per channel, must be simultaneously stored. Dozens of Gigabits
data streaming require data transmission interface/bus with very high through-
put and very efficient storage mechanisms design.
(3) Parallel channel calibration
Compared with TDM-based channel sounder which has only one Tx and Rx
pair, multiple RF channels in parallel channel sounder can be considered as a
multi-channel time-varying complex system. The non-ideal and different
responses among every Tx/Rx pairs rooted from the multipath clock Phase-
Locked Loop (PLL) circuits and RF devices. A new sophisticated parallel
calibration algorithm must be designed to carefully compensate the non-ideal
and difference in channel response as to guarantee the estimation accuracy of
channel estimation.
(4) High-speed continuous storage of raw data
The serial channel sounder first correlates the received signal with local code
to obtain the CIR, and then stores the CIR for each channel. However, this
correlation yields the average impulse response over a long period. Accurate
channel Doppler characteristics cannot be obtained based on this impulse
response when the channel time-varying is fast. Therefore, the most effective
way is to collect and store the original data. However, in the case of simulta-
neous storage of multi-channel with high sampling rate per channel, the archiv-
ing system must be able to achieve the high-speed transmission throughput with
an efficient storage mechanism.
6.2.4 Architecture of MIMO Parallel Channel Sounder
Faced with these channel modeling requirements and technical challenges, Shang-
hai Research Center for Wireless Communications has developed a sophisticated
parallel MIMO channel sounder. Compared with a serial channel sounder, the
biggest difference is the use of multiple parallel RF channels based on code
division, i.e., multiple parallel transmitter transmit simultaneously time domain
sequences, and multiple parallel receiver receive the air interface signal at the same
time. The use of such kind of architecture can fundamentally solve the problem of
measuring fast time-varying channels in multiple antenna scenarios. Figure 6.5 is
the picture of the channel sounder.
The system architecture of parallel channel sounder is shown in Fig. 6.6, mainly
including the following subsystems:
RF/baseband subsystem
Clock / synchronization trigger subsystem
Fig. 6.5 Picture of MIMO parallel channel sounder
Synchronization Synchronization
trigger system trigger system
Transmitti Multi-path Multi-path Receiving High-

ng-end Antenna Antenna -end speed
baseband RF RF baseband steam disk
subsystem array array subsystem
subsystem subsystem subsystem
GPS/local GPS/local
reference clock Calibration reference clock
Calibration Parameter
subsystem subsystem extraction
Fig. 6.6 System architecture of MIMO parallel channel sounder

Calibration subsystem
Antenna array
High-speed stream disk subsystem
Parameter extraction
1. RF/baseband subsystem
Parallel RF/baseband subsystem is the core of the entire parallel channel
sounder. Its performance and stability directly determine the accuracy of mea-
surement. The measurement system is realized by software defined radio plat-
form based on PXI bus.
At the transmitter, the parallel channel sounder generates a set of PN with
good cross-correlation characteristics as the transmission sequence of each
channel. Then through up sampling and shaping filter, the signal spectrum
shape is improved. After digital-to-analog conversion, the signal is converted
to a high-frequency analog signal by up-conversion modules and then radiate
through an antenna. In the receiver, down-conversion of parallel channel
sounder and the signal after AD sampling are sent into FPGA. In FPGA, the
data can be pre-processed accordingly. Then through Peer to Peer (P2P) FIFO, it
is transferred to disk array for storage at a high speed (Fig. 6.7).
(1) Test signal screening
A major difficulty of MIMO parallel measurement scheme is that multi-
ple concurrent detection signals will interfere with each other. Even with
orthogonal sequences, the orthogonality cannot be guaranteed for different
delay sequences, which will still produce interference. To this end, it is
needed to fully consider the requirements of parallel channel to screen a set
of eight signal sequences. The formula is shown in Table 6.1. These signal
Calibration Calibration
RF path (Low
RF path(DA
noise/down-
RF subsystem conversion/up-
conversion/AD
conversion/Amplifier)
conversion)
Transmitting-end
Trigger signal Receiving-end baseband Trigger signal
baseband processing
processing
Baseband (PN sequence
(digital filtering/digital
generation/RRC shaping
subsystem stream disk)
filter)
Reference Reference
clock clock
Fig. 6.7 RF and baseband subsystem

Table 6.1 PN sequence Sequence type of transmitter PN sequence

generation.
Sequence length 2048
Primitive polynomial D11+D2+1
4500
4000
X= 1 X= 2049
Y= 4096 Y= 4096
3500
3000
2500
2000
1500
1000
X= 278
500 Y= 261.8396
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Fig. 6.8 Autocorrelation characteristic curve graph of eight test sequences
sequences are generated based on PN sequence optimization with the length

of 2048. Figure. 6.8 and 6.9 respectively, show the autocorrelation and
cross-correlation characteristics of the curve of eight sequences under mod-
ulation of orthogonal QPSK.
(a) Autocorrelation characteristic curve
Figure 6.8 shows the simulation results for the autocorrelation per-
formance of one of the test sequences, and the other seven are similar to
this result. From the curve, these eight sequences all have very good
autocorrelation properties.
(b) Cross-correlation curve
Figure 6.9 shows the simulation results of seven cross-correlation
curves of one test sequence, and the other seven are similar to this result.
From the curve, these eight sequences all have good orthogonal charac-
teristics, which can meet the requirements of channel measurement.
2. Clock/synchronization trigger subsystem
Parallel channel sounder system requires strict synchronization and triggering
mechanisms. It includes the following aspects: the strict synchronization of each
transmitting channel, the strict synchronization of each receiving channel, and
the strict synchronization between the transmitting and receiving channels.
4500
4000
3500
3000
2500
2000
1500
1000
X= 1431
500 Y= 280
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Fig. 6.9 Cross-correlation curves of eight test sequences
In order to achieve strict synchronization of the parallel channel sounder, the

following synchronization mechanisms have been designed. First, each channel
of receiver and transmitter of parallel channel sounder shares the same time-
based reference signal and the same local oscillator. Second, the clock at both
ends of transmitter and receiver must strictly implement the synchronization
cycle and phase. In the system, two rubidium clocks are used at both ends of
transceiver. And the jitter of two phases can be controlled to a fixed phase error
after a period of recension of two rubidium clocks. Parallel channel sounder will
send 10Mhz reference clock signal to both ends of transceiver, and make the
reference clock of each channel at both ends of transceiver share these two 10Mhz
reference signals, so as to achieve the precise clock synchronization of each
transceiver channel. Both ends of transceiver adopt GPS receiver, to provide one
pulse per-second signal (pps) as the initial synchronization trigger source. After
triggering at the same time, the initial trigger signal, as a reference signal, is sent
to both ends of transceiver to produce the periodic trigger signal. The periodic
signal be send through PXI bus, sending to each channel of transmitting/receiving
and triggering the FPGA transmitting and receiving signals simultaneously.
Although each channel of transmitter/receiver shares the same clock source,
it is difficult to ensure that the phase of reference clock input to each channel is
strictly synchronized. Also, considering more general scenarios, it is particularly
desirable to have a common reference clock to align the respective sampling
clocks when each channel is configured with a different sampling clock fre-
quency. When the sampling time of each channel is aligned, synchronization
trigger mechanism is adopted to achieve simultaneous transmission and
reception. Parallel channel sounder increases the synchronization accuracy of

channels to the picosecond level by the following strategies.
(1) Unified bus architecture of multi-channel channel test system. The same
reference clock and local oscillator are used to ensure the equal clock and
triggering signal path in each channel, which reduces the phase error intro-
duced additionally.
(2) Shared trigger reference clock signal. During the initialization phase,
according to different sampling clock frequencies and channel numbers of
each channel, the Shared Trigger Reference Clock (STRC) signal is auto-
matically calculated and generated, which is used for multi-channel phase-
to-phase calibration.
(3) Real-time phase alignment based on FPGA. Real-time acquisition of phase
difference between STRC signal and sampling clock of each channel is
achieved by FPGA hardware programming for fine-tuning and aligning the
phase between multi-channel sampling clocks.
(4) Regenerated trigger signal and delay sampling strategy. After the channel
sampling clock completes the phase alignment, a short pulse trigger signal is
regenerated. There is subtle difference in the time delay of receiving the
trigger signal in each channel. After detecting the trigger signal, each
channel samples at the next rising edge of STRC simultaneously. Thus,
the synchronization accuracy improves further.
3. Calibrate subsystem
As described above, since parallel channel sounder can be considered as a
multi-channel parallel time-varying system. Therefore, it is necessary to obtain
the response of hardware system and the response of antenna system accurately
before data post-processing, so as to formulate an accurate calibration scheme to
deal with the time-varying characteristic.
In actual use, we design an 8-input 8-output calibration device to connect
both ends of transceiver. For ease of operation and speed, an RF switch is
designed at both ends of transceiver for the quick switch between calibration
and test.
First, the passive device in the whole system performs a system response test,
getting the system response of calibration equipment and RF switch and then
substituting it into parameter extraction post-processing software. At the same
time, 3D pattern of antenna array is also been substituted into parameter extrac-
tion post-processing software.
Second, calibration is made. RF switch is turned to calibration state. The
originator begins to launch, and then the receiver obtains the calibration data.
Third, the measurement starts. RF switch is turned to the test state. The
calibrated RF path is turned off. And the receiver records measurements data at
each test point.
Finally, parameter extraction is made. The system response of transmitting
and receiving hardware and antenna pattern data are first removed from the
measurements data, and then the extraction of channel parameters can be
achieved.
4. Antenna array
As an indispensable part of channel measurement, MIMO antennas play an
important role in MIMO channel measurement. Shanghai Research Center for
Wireless Communications has developed an 8-antenna omni-directional antenna
array as the measurement antenna.
The basic requirements for measuring antenna are:
(1) Transmitting and receiving antennas support 8-transmission and 8-reception
respectively.
(2) Standing Wave Ratio (SWR) is not greater than 1.5.
(3) Antenna gain is greater than 6dBi.
(4) Antenna bandwidth is greater than 200 MHz.
To meet the technical requirements, the antenna design uses a monopole with
1/4 wavelength. Then, according to center frequency fc 3.5 GHz, 3 108/
3.5 109 85.71 mm. The antenna monopole length is 20.2 mm and diameter is
0.4 mm, with 1 pole in the center and 7 poles evenly distributing at the periphery.
The distance between central monopole and any edge monopole is 40.3 mm. The
antenna design is shown in Fig. 6.10. The antenna array using this scheme can
effectively estimate 3D channel characteristics.
5. High-speed stream disk subsystem
The stream disk scheme of measurement system is based on the combination
of zero-copy technology and asynchronous storage technology, and uses Tech-
nical Data Management Streaming (TDMS) to achieve the ultra high-speed and
low-latency data storage scheme thus greatly reducing the CPU consumption.
In this measurement system, the bandwidth of PXI bus provided by case A
backplane is 24 GB/s. For this high-speed bandwidth, the data transmission is
achieved through the establishment of Direct Memory Access FIFO
(DMA-FIFO). Another problem that affects the transmission delay is additional
data copies and state transitions brought by the operating system kernel. Using
Fig. 6.10 Antenna array design

before read before write
user context user context
kernel context kernel context
syscall read syscall write
1
0 1 2 3 4
CPU copy
DMA copy
user space user buffer
kernel space kernel buffer CPU copy socket buffer
DMA copy hard drive protocol engine
0 1 2 3 4
steps
Fig. 6.11 Traditional I/O operation of a file
the zero-copy technique (Fig. 6.12) can effectively reduce the additional copy
and state switch caused by the operating system kernel.
Figure 6.11 is the traditional I/O operation of file.
It can be seen that in order to complete the process from reading data in
hardware to write it in disk, four copying processes are needed. First, in the
system call of reading the file, the hardware FPGA transmits data to the
kernel buffer in CPU through DMA. Then, CPU performs copy operation to
copy the data in its user cache. In system call of writing file, CPU copies the
data in user cache to socket cache, and finally transfers to disk through
DMA-FIFO.
Figure 6.12 is the I/O operation based on the zero-copy technology. After the
zero-copy technology is used, only the operations that hardware FPGA transmits
to the kernel buffer through DMA and the socket buffer transfers to disk through
DMA are retained, eliminating the need for two data copies in kernel space and
reducing delay and liberating CPU. At this point, the CPU-triggered disk writing
is an asynchronous operation. That is, it can execute other processes in parallel
without waiting for the completion of DMA transmission.
As shown in Fig. 6.13, TDMS file storage format is a binary file storage form,
which has an advantage of small space usage. In addition, it can classify stored
data and abstract the composition of data into three categories: top-level file
group, channel group, and channel. A file group may contain several channel
groups, and a channel group can contain many channels.
before sendfile
user context user context
kernel context
syscall read
0 1 2 3
user space
spend dscr DMA gather copy
kernel space kernel buffer socket buffer
DMA copy hard drive protocol engine
0 1 2 3
steps
Fig. 6.12 I/O operation based on zero copy technology
Data Set (File)

Description
Property 1 Title
Author
Property n etc.
Channel Group(s)
UUT
Procedure
Property 1
Test Fixture
Property 2 etc.
Property n
Channel(s) Name
Comment
Property 1 Unit
Sensor Info
Property n etc.
Fig. 6.13 Organization mechanism of TDMS file

Actual stream disk program includes the GPS location information, IQ data or
other relevant information. These data need to be precisely synchronized. Using
TMDS can easily combine these data together and package them in a file as a
different data set or channel. This greatly simplifies the difficulty of different
data synchronization in data post-processing.
6. Parameter Extraction
SAGE algorithm is an iterative method of maximum likelihood estimation. It
can realize the optimization of multidimensional parameters by allocating the
observed data in multiple subspaces and then estimating the multipath parame-
ters in each subspace.
In order to obtain the results of the maximum likelihood estimation, the
SAGE algorithm uses iterative methods to carry out the interference cancellation
on the observed values in subspace, so as to ensure the monotonic rise of overall
likelihood of parameter estimation with increasing iterations. The optimal esti-
mation of output parameter set is achieved when the requirements for conver-
gence are satisfied.
The measurement platform of parallel transmitting and receiving channel is
optimized based on the multi-dimensional parameter estimation SAGE algo-
rithm in the development of feature extraction software. The parallel sounding
SAGE (P-SAGE) algorithm is proposed. This algorithm has the following
functions and features:
The received signal data analysis in baseband based on parallel M N
MIMO system;
Good orthogonality based on the baseband spread spectrum code;
The non-uniformity among RF channels is fully considered;
The calibration is done separately at Tx and Rx;
The complexity of algorithm is well reduced to achieve fast estimation;
The algorithm is optimized for the presence of phase noise and can be applied
to scenarios where phase noise exists.
Figure 6.14 shows a block diagram of the parameter estimation algorithm
with multi-dimensional channel characteristic based on the PS-SAGE algorithm.
The innovations of PS-SAGE algorithm are:
(1) Suppression of multi-channel phase noise. Parameter extraction and data
post-processing algorithm use multi-channel phase noise sample data to
construct the maximum entropy statistical model. The maximum likelihood
Transmitting channel IR
Baseband measurement data
calibration
SAGE algorithm improvement-based

parameter estimation algorithm
Test antennas calibration data Receiving channel IR

calibration
Fig. 6.14 PS-SAGE algorithm block diagram

method is used to estimate model parameters. The multi-channel phase noise

model and space multipath propagation channel model are combined. Multi-
channel phase noise covariance matrix and channel multipath parameter
estimation perform joint iterative updating. The influence of multi-channel
phase noise is suppressed by using sub-space received signal to finish phase
noise bleaching processing.
(2) Multi-channel interference cancellation. The initialization parameters are set
to reconstruct the interference modes, such as multi-channel multipath self-
interference, orthogonal interference among transmitted signals and mutual
interference among received signals. The interference is canceled in the
iterative process of estimating parameter to effectively guarantee the accu-
racy of multi-channel multi-path parameter estimation.
(3) High-precision channel parameter estimation. The expected function itera-
tion and maximum estimation iteration are made for multi-channel signals
after canceling the interference. A more accurate channel estimation param-
eter is obtained and then the multi-channel interference is canceled (Step 2)
to achieve rapid convergence through the cycle iteration to obtain high-
precision channel estimation parameters.
6.2.5 Test Results
The test results of the parallel channel sounder in anechoic chamber are shown in
Fig. 6.15:
For the static scenario in the chamber, we choose PN sequence with length of
4096 and an I/Q chip with rate of100 M/s. Each snapshot collected 100 cycles. The
total length is the measurement data within 10 ns 4096 100 2 8.192 ms.
We first use the 8-input 8-output connector to make calibration and measure-
ments, and then test via air interfaces. The test results are shown as below.
Fig. 6.15 System

verification test in
microwave anechoic
chamber
Fig. 6.16 Multi-path delay spectrum before and after calibration
Fig. 6.17 PS-SAGE algorithm estimates the spatial parameters and reconstructs the signal spatial
spectrum
Figure 6.16 is pair wise multi-path delay spectrum of 64 groups for In it, they are
respectively, from left to right, air interface collection, calibration data, and the
results after using calibration data compensation. It can be seen from the figure that
after using 64 calibration files to calibrate original signals, the response difference
of each transmitting and receiving channel can be well compensated to obtain the
consistent delay as shown in the right figure.
Then here are the test results of channel parameter estimation. In Fig. 6.17, the
elements from left to right, respectively are air interface data, reconstructed signal
based on parameter estimation, and the power spectrum of Azimuth of Departure
(AOD)/ EOD of signal after cancellation. In it, the horizontal and vertical coordi-
nates are AOD and EOD respectively. The color represents the intensity of spectral
component. It can be seen from the figure that the reconstructed signal has a strong
similarity with the original signal spectrum. The energy of reconstructed signal
compared to the energy the original signal is reduced by 37 dB. The chamber
verification shows that the spatial parameters of wireless channels which are
obtained from the post-processing algorithm estimation are very close to actual
ones.
40
original
45 estimated
residual
50
55
power in dB
60
65
70
75
80
0 10 20 30 40 50
delay samples
Fig. 6.18 PS-SAGE algorithm estimates the time domain parameters and reconstructs the signal
time-domain spectrum
Figure 6.18 is a view of the time domain characteristic of wireless channel

characteristics. The estimated delay spectrum is consistent with the morphological
trend of original time-delay spectrum. And the original signal power can be reduced
by more than 10 dB after cancellation.
From the above time domain and spatial parameter estimation results and the
results of reconstruction and elimination, it can be seen that the channel measure-
ment data and estimation results are reliable, which can have a more accurate
correspondence with actual chamber environment.
6.2.6 Channel Measurement and Test
Scenario Test Planning and Measurement
After finishing the development and chamber verification of parallel channel

sounder, Shanghai Research Center for Wireless Communications begins the plan-
ning for the channel scenario test with typical geographic characteristics defined by
IMT-2020, which includes 12 scenarios with hot spots, high capacity and seamless
wide-area coverage.
The parallel channel sounder has now completed the following scenario test,
collecting the preserving mass data of parallel measurement channel:
Anechoic chamber environmental verification
Office scenarios
Fig. 6.19 Main scenario graphs of progressed channel tests of parallel channel sounder test
Large-scale indoor scenarios

Stadium outdoor scenarios
Rural outdoor scenarios
Shanghai Circuit V2 V scenarios
City hot spot regional scenarios
High mobility scenarios (Fig. 6.19)
Channel Model Library
Based on the parallel MIMO channel sounder and the definition of 5G application
scenarios in IMT-2020 (5G) promotion group white paper, Shanghai Research
Center for Wireless Communications began to construct shared data channel
model based on practical measured data in 2014. The designed channel model
library will basically cover the following two main application scenarios.
(1) 5G mobile Internet application scenarios
The future 5G mobile communication system will meet diverse service
requirements in different areas, such as residence, workplaces, leisure venues,
and transportation venues. It could provide ultra-high definition video, virtual
reality, real-world enhancements, cloud desktop, online gaming and other
excellent service experience for customers in different scenarios. The scenarios
include the places of dense residential areas, office, stadiums, open-air gather-
ings, subways, highways, high-speed rail, and wide-area coverage and other
application scenarios with ultra-high mobile data traffic density, ultra-high
mobile connection density and ultra-high mobility.
All above application scenarios can be summarized into two types: contin-
uous wide-area coverage scenario and high-capacity scenario with hot spots.
(2) 5G IoT application scenarios
The future 5G mobile communications system will penetrate to IoT and
various (vertical) industry fields, deeply integrating with industrial facilities,
medical equipment, vehicles, etc., so as to effectively meet the service
requirements of diversity of many vertical industries, such as industry, medi-

cine, transportation, (ultimately) to achieve true all things interconnected.
All of the above application scenarios can be summarized into two types:
high connection scenarios with low power consumption and high reliability
scenarios with low delay. We can see that these two applications are newly
explored by 5G, which can be used to solve the wireless IoT applications and
vertical segmentation industry applications which have not been well supported
by traditional mobile communication systems.
Based on the existing channel test database of scenarios, Shanghai Research
Center for Wireless Communications shares the data acquired by the parallel
channel sounder with the industry peers through big data open-source platform
of wireless channel under construction. And according to the requirements, the
test data are developed to promote the development of channel measurement
and modeling work in the industry. There are already Tongji University,
Shandong University and other universities making related research and coop-
eration on open-source platform data.
The eight units firstly signed up for wireless channel big data open-source
platform under construction are: Beijing University of Posts and Telecommu-
nications, Tsinghua University, University of Electronic Science and Technol-
ogy, Southeast University, Tongji University, Forty-first Study Institute of
China Electronics Technology Group Corporation, First Study Institute of
Telecommunications Science and Technology, Shanghai Advanced Studies
Institute of Chinese Academy of Sciences.
6.3 OTA Test Platform
6.3.1 5G Requirements for OTA Test
The method to evaluate radiation performance of traditional SISO is now more

mature. Its evaluation mainly aims for two indicators, making OTA test for Total
Radiated Power (TRP) and Total Radiated Sensitivity (TRS). The two indicators
respectively characterize the ability of a mobile terminal to transmit as a whole,
including an antenna device and to receive signals from a base station, which can
effectively evaluate the transceiver performance of a single antenna terminal in the
real world. As early as in 2001, the Cellular Telecommunications and Internet
Association (CTIA) began to study the performance evaluation methods of con-
ventional SISO antennas.
Many performance parameters of passive MIMO antenna, such as efficiency,
gain, are not fundamentally different from traditional SISO antennas. Since MIMO
antennas contain many antenna elements, passive parameters are introduced to
describe the relationships among several antenna elements. Therefore the existing
SISO OTA testing techniques cannot well adapt to the requirements of MIMO
application testing. Since the end of 2007, relevant organizations have begun to
6.3 OTA Test Platform 321
follow up the applications and extensions of the SISO testing technology in MIMO
OTA test. Many standardization organizations and agencies including the 3GPP,
COST and CTIA named the test method of evaluating the performance of multiple
antennas as MIMO OTA.
The MIMO OTA test scheme is essentially a different simulation method of
multipath in space propagation environment, producing wireless communication
environment close to the reality to deal with the key challenge in massive antenna
MIMO OTA testing techniques. The challenge is how to generate an RF channel in
anechoic chamber that is closest to the real-world spatial, angular and polarization
behavior. Therefore, the main requirements can be divided into three types.
1. Emulator of the real wireless propagation environment.
This requirement is currently addressed by a channel simulator, which sim-
ulates the channel characteristics of test scenarios through a real-time channel
model. These features are also reflected in time domain, frequency domain and
spatial domain. Only the exact channel model can approximate the test
scenarios.
2. The close-to-reality wireless propagation playback environment.
The channel environment is played back by using different configurations of
chambers and antenna arrangements to combine with the channel emulator. In
particular, the spatial characteristics of signal should be considered.
3. Calibration of OTA test system.
In MIMO systems, spatial correlation is a very important parameter, which
contains the characteristics of antenna and transmission channel. In fact, the
unknown antenna characteristics of known antenna model cannot get the corre-
lation. Likewise, the unknown channel model with known antenna characteris-
tics cannot get the correlation. Therefore, the characteristics of antenna and
transmission channel to test multiple antenna terminals must be considered.
6.3.2 Status and Shortcomings of OTA Test Scheme
The Technical Report (TR) of 3GPP organization of mobile communication part-

nership program 37.976 summarizes seven MIMO OTA test schemes. These
schemes can be divided into three categories according to chamber features: OTA
test scheme based on Anechoic Chambers, OTA test scheme based on Reverbera-
tion Chambers and test scheme based on Multi-stage Method.
In the summary of 3GPP TR 37.976 candidate test scheme, OTA candidate
testing schemes based on anechoic chambers include:
(1) Multi-probe method;
(2) Symmetrical ring of the probe method;
(3) Two channel method;
(4) Spatial fading emulator method.
Likewise, OTA candidate test scheme based on the reverberation chambers

include:
(1) Basic or cascaded reverberation chamber;
(2) Reverberation chamber with channel emulator.
Test scheme based on multi-stage method is a two-stage method.
OTA Test Scheme Based on Anechoic Chambers
The test scheme based on anechoic chambers will connect RF channel simulator to
the probe array in an anechoic chamber environment. The probe array surrounds the
DUT to reproducibly simulate a wireless environment that generates complex
multipath fading at the location of DUT.
(1) The multi-probe OTA test block diagram based on anechoic chamber is shown
in Fig. 6.20, which mainly includes anechoic chamber, system simulator,
MIMO channel simulator and test probes, etc. The multi-probe OTA test
based on anechoic chamber is the most intuitive test method. The test system
simulates the downlink signal of a base station with channel simulator, launches
it through many test probes in anechoic chamber, and simulates the channel
model with a specific AOA in chamber center, and then sets terminal under test
in chamber center and tests the MIMO antenna performance in channel sce-
narios gotten from simulation.
Tests for multiple antenna wireless devices are typically conducted in a
conduction mode, and proper fading is typically simulated using a channel
simulator. The channel model used in the current test includes the signal AOD
and AOA information, as well as antenna patterns on both sides of the trans-
mitter and receiver of channels. These spatial channel models can be better
adapted to environmental simulation requirements of OTA test by modifying
the channel model in MIMO channel simulator.
test anechoic chamber
wave-
absorbing
materials tested object
probe
antenna
Base station simulator Channel simulator
Signal generator Power
Fig. 6.20 Multi-probe test scheme based on anechoic chamber

In such a scheme, the DUT is free to use the number and position of probes
to combine different test scenarios. For example, the probe can be placed in the
same orientation to simulate the 2D scenario, and the antenna can be placed in
different planes to simulate the 3D scenario. The system configuration can be
adjusted to obtain a signal similar to that received by the device in the real
world, thereby making it easier to evaluate the position, direction and impact of
MIMO antenna on system. The advantages of the multi-probe method are that
the concept is simple, configuration is flexible, and the testing accuracy is good.
However, the disadvantage is that multi-probe may cause high cost. Another
challenge of the multi-probe approach is that it can only simulate parts of the
3D channel, making it difficult to simulate a complete 3D channel.
(2) Ring-shaped probe test methods based on anechoic chamber symmetrically
distribute the probe antenna around the DUT equidistantly, and place the
measured object in the center of chamber, as shown in Fig. 6.21. The system
consists of anechoic chamber, multidimensional fading emulator, communica-
tion tester / BS emulator, OTA chamber antenna, etc.
Similar to the multi-probe approach, each probe antenna transmits signal
with a specific time domain after the processing of the channel simulator based
on the ring probe test method of an anechoic chamber. Unlike the multi-probe
scheme, there is no correspondence between signal AOD and the position of the
probe antenna. Based on it, it is possible to simulate any 2D spatial channel
model without readjusting the position of the probe antenna.
(3) The test method based on Spatial Fading Emulator (SFE) of darkroom was
initially proposed by Panasonic Company in Japan. The main feature of SFE is
the manufacture of spatial fading characteristics through the antenna probe and
associated RF devices surrounding the DUT. The amplitude of the signal
emitted by each probe is directly determined by the sampling of target PAS.
Doppler frequency shift depends on the angle between the probe position and
the direction of the virtual motion of DUT. Since there is no pre-fading, the
tested
object
communication tester
station simulator
Loss simulator
anechoic chamber
Fig. 6.21 Schematic diagram of a ring probe test method based on anechoic chamber
number of probes determines the pros and cons of SFE for the reconstruction of
synthesized channel model. A theoretical study on the number of full-ring
antenna probes made by NTT DoCoMo has shown that for a single-cluster,
the test method of SFE based on chamber generally requires 10 probes to satisfy
1.5 times of wavelength terminal test. In the multi-cluster channel, at least
1115 antenna probes are needed to meet the requirements of the test domain
with 1.5 times of wavelength.
The SFE test method based on the anechoic chamber is essentially regarded
as a variation of the ring probe test method. The test method of SFE is to use a
programmable attenuator and an RF phase shift device to adjust the amplitude
and phase of each antenna to replace the channel simulator in the ring probe
method. The block diagram of its system is shown in Fig. 6.22. The RF signal
from a base station simulator is fed into a power divider, which provides the
same RF signal for each output. The number at the output end of power divider
is the same as the number of probe antennas. And each output is connected to an
RF phase shifter, which carries out phase offset adjustment by receiving a
control signal from the digital-to-analog converter. Through the attenuator,
the signal passing through the RF phase shifter is output to a horizontally or
vertically polarized probe antenna. The DUT measures the signals from each
antenna probe and outputs the data to a computer control terminal for further
UECVVGT
0 r WPKV
N-1 1
D h
2
5
R
4 3
4GEGKXGT
wave-absorbing anechoic chamber
2QYGT
#PVGPPC 2JCUGUJKHVGT 6TCPUOKVVGT
FKXKFGT
&#
%QORWVGT
EQPXGTVGT
Fig. 6.22 Test method for SFE based on anechoic chamber

processing. At the same time, the computer control side also provides the
configuration and interaction of simulation parameters.
The SFE test method based on the anechoic chamber can generate multipath
fading of a particular distribution by controlling and adjusting the amplitude
and phase of RF signal in real time, such as Rayleigh distribution. Because the
scheme uses an attenuator to control the amplitude, the RF phase-shifting
device adjusts the phase, which is less expensive than a commercially available
channel simulator. But its flexibility has also been greatly limited. In addition,
the method cannot simulate the characteristics of the transmitter.
(4) The dual-channel measurement method based on anechoic chamber is a direct
and valid method to test OTA performance of MIMO equipment. Its principle
is shown in Fig. 6.23. At the same distance from the UE, two polarized test
antennas with a rotatable incident angle are placed, transmitting different
MIMO downlink signals respectively. The overall characteristics of UE
antenna are obtained by a combination of various azimuth and polarization.
As shown in the figure below, the two-channel chamber contains four angular
positioning devices: , 1, 2, angle positioners. The two test antennas A1
and A2 are distributed as 10 or 90 angle (simulate rural environment) and a
communications antenna ANTUL (simulate urban environment). An optional
angle positioner on turntable controls the tilt angle of antenna under test. All
of these angle controllers can be used to implement any combination of angles
for MIMO test. External devices include a base station simulator and a
switching matrix. The dual-channel approach can be viewed as a special
case of a multi-probe method using only two probe antennas without a channel
simulator.
The advantage of the dual-channel test method is the easy updating on the
basis of the original SISO test system. It only needs to add the second angular
positioner control system and a second test antenna, which greatly reduces the
cost. At the same time, it can also be used to verify the pattern, load and
impede the smart antenna which can be adaptive with the environment
changing.
Fig. 6.23 Dual-channel

test method based on
anechoic chamber
OTA Test Scheme Based on Reverberation Chamber
The basic principle based on reverberation chamber OTA test scheme is actually to
provide the characteristics that the reverberation chamber is rich in reflection and
the specific channel environment created by mixer for DUT testing. The purpose of
the reverberation chamber is to produce a statistically uniform power distribution
around DUT. And the antenna and channel simulator can be used to generate the
desired delay characteristics.
Depending on whether the reverberation chamber is connected to a channel
simulator, the OTA test scheme based on reverberation chamber can be divided into
two types. The first is to use a separate reverberation chamber or a cascade
reverberation chamber, as shown in Figs. 6.24 and 6.25. And the second is the
reverberation chamber to connect the channel simulator. For example, the partic-
ipation of channel simulator in the second case realizes time diversity through
inputting a fading signal at different time steps, overcoming the limitations in the
first case.
In general, the reverberation chamber can provide a subclass of multipath
environments that can do frequency fading and time diversity simulation. The
reverberation room test scheme is limited by limited analog capabilities of different
fading environment, so it can only provide a limited performance evaluation for
terminals.
Limited by the statistical isotropy at receiver in spatial domain of channel
environment, i.e., its evenly distributed AOA, the spatial diversity cannot be
simulated. In addition, the statistical isotropy in reverberation chamber also deter-
mines that vertical polarization and horizontal polarization of the channel model
station matrix switch

simulator 2*T
1 T
blender
tested plate with

object holes
Fig. 6.24 Test method of basic reverberation chamber

channel
wall
simulator blender
antenna
(e)nodeB
simulator
tested
object
return path
turning disc
Fig. 6.25 Reverberation chamber test method with channel simulato
can only be equal. For the DUT using polarization diversity method, reverberation
chamber method cannot be effectively differentiated.
The scheme based on the reverberation chamber does not specify the device. The
same result can be obtained with any device. The size of test domain is not strictly
required, which is very convenient for the actual test operation.
Two-Stage OTA Test Scheme Based on Anechoic Chamber
The two-stage test method based on the anechoic chamber is shown in Fig. 6.26.
The two-phase test scheme consists of two test phases. The first stage, in an
isotropic environment, uses a conventional anechoic chamber as a basic test system
and an integrated tester to measure a complex active antenna array. The second
stage combines the information of antenna array with the channel model by the
following two means: using channel simulator to make conduction test or using
antenna array information obtained from test to calculate a theoretical channel
capacity performance by theoretical calculation. Therefore, at this point, the
two-stage test method can only obtain limited data, and further research is needed
to obtain accurate performance indicators.
In the two-stage OTA test method, the absolute accuracy of DUT power mea-
surement has little effect on the accuracy of final results, since the correction is
made at the second stage. The accuracy of relative phase measurement has great
influence on the result. But the existing DUT, usually mobile terminal, has strong
phase measurement ability, so it has limited influence on measurement result.
Because 3D antenna pattern can be measured more easily, a two-stage OTA test
method can simulate any 3D channel transmission condition. The quality factors
test box
wave-absorbing
materials
MIMO
tested object
Base station simulator reference antenna
antenna
pattern
wire
MIMO
tested object
BER, FER
H, R
Channel simulator Base station simulator
Fig. 6.26 Two-stage test method based on anechoic chamber
that the two-stage OTA test method can measure include TRF, TRS, throughput,
block error rate, MIMO channel capacity, antenna correlation coefficient, etc.
However, another feature of two-stage approach is that it cannot obtain the effects
of self-interference.
Compared with the conventional multi-probe test scheme, two-stage OTA test
only simplifies the receiving diversity performance rather than the channel-related
characteristic, which is a fast, accurate, economical and efficient MIMO OTA test
method. Secondly, the two-stage OTA test method can re-use the antenna pattern
obtained from test to simulate a 2D or 3D channel model without re-using an
anechoic chamber test, and thus improving its flexibility, and taking full advantage
of test platform resources constructed in the LTE phase to rapidly expand and
achieve MIMO OTA test, which is a fast and economical test solution.
In summary, the existing schemes focus more on two requirements of building
and playing back wireless propagation environments close to the reality and
OTA test system calibration. While wireless propagation environment close
to reality obtained from emulator all relies on ready-made commercial channel
simulators, so the pertinence and applicability of channel environment have
common flaws.
6.3.3 Key Technical Difficulties and Challenges
Compared with traditional single-antenna OTA testing, the MIMO OTA testing
technology adds multiple antenna broadband testing. The tests must use multi-
dimensional RF channel parameters, such as fading, delay, Doppler, AOA and
polarization and other evaluations.
The key challenge in the development of MIMO OTA test platform is how to
generate an RF channel model in anechoic chamber that is closest to the real-world
spatial, angular and polarized behavior. Therefore, further study is needed for the
test combining channel simulator and chamber, and antenna characteristics must be
taken into account.
The key challenge in massive MIMO OTA test techniques is how to generate an
RF channel model in the anechoic chamber that is closest to the real-world spatial,
angular and polarized behavior. This complexity requires a lot of space and
equipment investment for the R&D of MIMO OTA test platform. Its cost is too
high for the majority of terminal equipment manufacturers. The following is the
discussion of key technologies and difficulties of OTA test.
1. Spatial fading simulation technology
Since wireless channels play a key role in MIMO performance, wireless
channel simulator is an important part in MIMO OTA air interface test system.
A test signal generated by a transmitter or a base station simulator passes through
a wireless channel simulator, which simulates a wireless channel according to a
predefined channel model. The signal is then separated in simulator and distrib-
uted to each probe in the chamber, each independently radiating into the
chamber. Its result is that the multiplexed radiated signal is synthesized in the
central space of the chamber and produces the desired wireless channel envi-
ronment around DUT.
The advanced SFE technology must be adopted to restore the real living
environment in the chamber. The most typical analog parameters include path
loss, multipath fading, delay spread, Doppler spread, polarization, and, of
course, spatial parameters such as AOA and AS.
In order to obtain valuable results from MIMO OTA test, wireless channel
simulator must have excellent RF performance. Its Error Vector Magnitude
(EVM) and internal noise level must be very low in order to minimize errors
that affect the measurement results.
Moreover, in order to obtain consistent test results in multiple measurements,
the fading process must be repeatable. This is very important when
benchmarking different DUT.
2. Stochastic channel model based on geometry
The channel model for MIMO OTA test is a GSCM, in which the wireless
channel is defined by the following parameters:
The location and array of transmitting antennas;
Propagation characteristics (delay, Doppler, AOD, AOA, angle spread of
transmitted signal, angular spread of received signal and polarization
information);
Movement speed and travel direction;

The location and array of receiving antenna;
And multiple large-size parameters.
These channel models based on measurement include all parameters of
wireless channel (time, frequency, space and polarization). Since both space
and polarization are key parameters of spatial correlation (MIMO performance
is strongly correlated with it), these two parameters are very important. Obvi-
ously, this method accurately and vividly simulates the environment required by
MIMO device test. The signal received by the receiver is obtained by transmitted
signals of the transmitter through a multipath environment with the spatial
characteristic. Therefore, the spatial characteristic, including receiving antenna,
becomes the main factor of testing receiver performance. Stochastic channel
model (ie, geometric-based stochastic channel model) series must also include
the channel models of Spatial Channel Model (SCM) of 3GPP, Spatial Channel
Model Extension (SCME), ITU, and International Mobile Telecommunications-
Advanced (IMT-Advanced).
3. MIMO OTA channel model mapping
In MIMO OTA test, the receiving antenna is integrated directly on DUT and
becomes an integral part of it. Therefore, it is not necessary to load its antenna
array in the channel model of simulation. The actual impact on the performance
of DUT can be directly obtained through testing. In this way, the key challenge
in MIMO OTA testing is to generate actual propagation characteristics in the
chamber, especially AOA and its ASA. The information based on geometric
parameters, such as in SCM, should create appropriate correlation parameters
for the antenna of DUT. In addition, the information of the transmitting antenna
array (base station) is also needed, including geometric information of the array
and the array of antenna fields, as well as the speed component or Doppler
frequency component of each path or cluster for terminal.
These clusters are then mapped at the same time to the corresponding OTA
antenna, so that the spatial synthesis result of transmitted signals in the center of
chamber is consistent with the defined model. The mapping is performed by a
spatial wireless channel simulator. In order to make the angle expansion more
precise, each cluster of signals is generated by the several OTA antennas. As a
result, a wireless channel environment based on geometric information can be
accurately generated in the chamber.
OTA systems must also support complex full-3D systems expanding from the
simple single-cluster system. The high degree of correlation caused by
extremely narrow AS and the requirement for the rotation of DUT will signif-
icantly affect performance results. Wider AS makes the correlation low. The
channel is more easily to achieve spatial multiplexing. All 2D channel models
(such as SCM, SCME, WINNER, IMT-Advanced) need to be simulated using a
2D full-circle system. A complete 3D application allows the test results to take
into account not only the azimuth, but also the pitch propagation.
6.3.4 OTA Platform Example
OTA Platform Architecture
As shown in Fig. 6.27, the MIMO OTA platform developed by Shanghai Research
Center for Wireless Communications is a scheme based on the anechoic chamber as
a whole. The principle of the scheme has been described in Sect. 6.3.2.1, and the
different points of the platform are highlighted here. The biggest difference lies in
the following two aspects.
I. An on-line channel model library is added to test platform as a support for
channel model. The on-line channel model not only contains the standard
channel model of the traditional channel simulator, but also has a typical
channel model of China, which increases the applicability of the channel
model in China.
II. In addition to using the traditional channel simulator based on channel models,
this platform can combine the parallel channel sounder to perform real-time
playback for the captured signal under appointed locations (areas). The evalu-
ation reliability of DUT for the suitability of appointed channel environmental
capabilities is improved.
The channel model library has been described in Sect. 6.2.6 in details. The
channel analog/playback device is highlighted below.
Channel Analog/Playback Device
As shown in Fig. 6.24, the channel analog/playback device of the OTA test platform
developed by Shanghai Research Center for Wireless Communications uses two
forms: the commercial channel simulator and the self-developed channel simulator.
tested
base station/ object
base station
simulator
channel model parallel channel

sounder playback/
databases channel simulation
wave-absorbing anechoic
chamber
Fig. 6.27 OTA platform architecture

Channel models can be imported into commercial channel simulators to perform

real-time channel simulations and then reproduce real signals through anechoic
chamber based on channel model library collection, parameter extraction and
modeling.
It is worth mentioning that the parallel channel test probe described in Sect. 6.2
also has the function of channel playback. As a parallel 8-channel channel detector,
it can be taken as a 16-channel channel simulator in channel playback. So the same
set of equipment plays the role of two sets of instruments, saving costs for the
construction of OTA platform. At the same time, the system calibration scheme, as
a channel detector, can also be applied to the OTA test.
6.4 5G Open Source Community
6.4.1 Introduction to Requirements
In order to meet future service requirements, mobile operators expect to be able to

introduce the latest technology into existing systems in a timely manner. As a result,
communications equipment is subject to ongoing system upgrades. The traditional
base station will face a greater challenge. Base station is one of the most important
subsystems in mobile communication systems. The mobile communication systems
with different modes need the base stations with different modes for support.
Traditional mobile communication base stations are designed for a single commu-
nication standard based entirely on dedicated hardware elements, such as
FPGA, DSP and Application-Specific Integrated Circuit (ASIC). Hardware
implementations solidify their independence of each other. There are problems of
poor compatibility, cumbersome upgrades and lack of flexibility in resource
between equipment. As the actual problems brought by continuous development
of mobile services, the coexistence, convergence, maintenance and upgrading of
many communication mode standards are becoming more and more obvious. The
costs of operation and solution are rising. Therefore, designing a universal and
flexible base station system which can configure various modes is the key to solve
these problems.
On the other hand, the convergence of Internet and traditional telecom networks
has become an irresistible trend. And standard servers have been widely used in the
core network domain. As early as in 2010, in the global technology outlook released
by IBM, the convergence technology of wireless and Information Technology
(IT) was taken as one of the important technological trends. Intel also takes real-
time signal processing of communication base stations based on CPU as a key
research direction [1] For decades, semiconductor industry has been moving for-
ward with Moores Law. DSP and General Purpose Processor (GPP) have made
great progress in architecture, performance and power consumption, providing
more options for software-based mobile communication base stations. However,
6.4 5G Open Source Community 333
because the various DSP of different manufacturers or even the same manufacturers
are inconsistent in backward compatibility, and the supported real-time operating
systems are not the same, the industry currently lacks a unified platform and
standard. In addition, the BaseBand Unit (BBU) of mobile communication base
stations developed based on DSP platform system is generally a non-open propri-
etary platform, which still has many deficiencies in smooth upgrading and
virtualization of networks. With the rapid development of relative technologies,
GPP can gradually meet the requirements of high data load operations, such as
digital signal processing, and provide a new choice for software to realize digital
signal processing. The new technologies of these GPPs include multi-core, Single
Instruction Multiple Data (SIMD) supporting fixed and floating point arithmetic,
high capacity on-chip cache and low latency off-chip memory. With these new
technologies, GPP can finish digital signal processing performed by a DSP, espe-
cially baseband processing functions in base station devices. The use of GPP for
baseband signal processing has the following advantages:
(1) Simplified design process and shortened development cycle.
The communication system design process of traditional scheme is often
based on architecture characteristics, uses a simulation platform to achieve new
algorithm design, and then optimizes the corresponding codes according to the
characteristics of programming. After constant optimization and correction to
ensure the performance of the entire communication system, the fixed point
code is reconstructed and under the premise of maintaining system perfor-
mance. Finally, it is moved to the platform for corresponding programming,
optimization and testing, etc. R&D process is long and inefficient, and the
invested human labor costs are also rising. The migration between the platform
and fixed-point design of codes all give the smooth development of the project a
great deal of risks. On the GPP platform, the algorithm can be directly opti-
mized, thus effectively improving the efficiency of programming and greatly
reducing development cycle.
(2) Easy to realize multi-mode base station and achieve resource sharing.
The multi-mode base station based on the unified platform is based on a
universal platform composed of modularized and standardized hardware
units to realize the partial communication function of wireless equipment.
It has good scalability, which can effectively extend the life cycle of base
stations, save costs and seamlessly converge different communication
modes. The network upgrade and smooth evolution are realized through
software configuration without changing the hardware. In addition, using a
multi-mode base station enables the use of a set of base station equipment to
achieve multi-mode network coverage to save space, improve power effi-
ciency and reduce power consumption. In short, the fully software-based
baseband processing of base station can support multiple standards on a
single system, which is conducive to resource multiplexing, improving
resource utilization.
(3) Conducive to realize resource virtualization and build cloud platform.

In a centralized scenario, the operating characteristics of the GPP platform
are fully compatible with the requirements of the virtualization technology.
Based on universal computing platforms virtual realization of the base station
system, it can maximize the advantages of hardware systems and support the
protocol layer processing of high-speed communication systems. The GPP
platform can perform tasks in a hardware-friendly manner, which mainly
depends on computing, storage and interface resources provided by the proces-
sor. Computing resource refers to the instruction processing capacity per unit
time, which is usually decided by the processors main frequency and its
arithmetic and logic computing unit. The storage resource is related to the
internal and external storage capacity and cache capacity of the processor.
While interface resources correspond to different levels of inter-server interac-
tion, such as the inter-processor, the inter-processor or processor cores. By
initially abstracting and equivalently calculating key capabilities of hardware as
processing resource, the details of specific hardware are ignored and too-close
ties with underlying hardware are eliminated, so as to implement universal and
efficient virtualization. Since different processing resources of hardware pro-
cessors in the centralized platform have different functions and manifestations
in system implementations, they are relatively independent but also connected.
Therefore, a processing resource can correspond to one dimension of
virtualization to construct joint relationship of multiple processing resources
on hardware platforms, and thus effectively playing the hardware virtualization
capabilities.
(4) Building an open platform for easy management and maintenance.
Under the traditional communication system, a vendor usually provides a
complete set of solutions, so the cost and dependence of system maintenance or
upgrade are very high. Based on the centralized shared virtual base station, the
multi-standard is unified, and the communications interfaces at each level are
standardized, which can form an open baseband pool platform of coordinated
wireless signal process, allowing access to solutions provided by any hardware
vendors to achieve the smooth upgrading and the expansion of the system, so as
to achieve healthy competition of industry, which is conducive to sustainable
development of operators. Thus, a universal and open 5G network platform
must be built. Open 5G network is conducive to work with other industries for
coordinated innovation. The GPP platform has the advantage of openness in
baseband signal processing of base stations, which provides a new way for base
station baseband processing, and is expected to become a new generation of
unified and open mobile communication systems and schemes of multi-mode
baseband processing platforms, which has sustainable competition in future-
oriented applications and services.
6.4.2 Development Status of Communications Platform

Based on General Purpose Processor
Open Air Interface (hereinafter referred to as OAI), based on GPP, releases

hardware and software version supporting Evolved Packet Core (EPC) and the
base station. OAI is the most complete development platform of open source LTE.
The platform uses the design idea of software defined radio. Architecture is in the
Intel GPP and Linux operating system. The entire platform, including the baseband,
is realized by complete software. Base station includes baseband, MAC and Radio
Resource Control (RRC) signaling. The core network comprises MME, SGW,
PGW and HSS. OAIs base station and core network use the standard S1 interface,
which can be used alone. OAI is established and managed by EURECOM, an
educational and research institution in the communication domain of French Rivi-
era, whose partners include Agilent, China Mobile, IBM, Alcatel-Lucent, Thales
and Orange.
Universal Software Radio Peripheral (USRP) [3] is a widely used software-
defined radio platform based on GPP designed by the Massachusetts Institute of
Technology. USRP includes hardware front end and GUN Radio of corresponding
software development kit [4] USRP can connect a personal computer to RF. It is
essentially equivalent to a digital baseband or IF portion of a wireless communica-
tion system. GNU Radio provides real-time signal processing software and
low-cost software radio hardware. GNU radio can be used to implement software
defined radio on low-cost RF hardware and GPP.
Sora system [5] developed by Microsoft Research Institute Asia, is a software
defined radio system based on multi-core GPP. It uses both hardware and software
technologies to address the challenges of real-time processing of wireless digital
signals. A new RF control interface board is redesigned. PCIe is used to transmit
broadband wireless signal sampling, which can support transmission rate of more
than 10Gbps and meet the needs of most wireless technologies.
IBM Research Institute has proposed a virtual base station pool based on cloud
computing, which realizes the TDD WiMAX standard on multi-core universal
processing platform and a newly designed RRH, and completes the prototype
design of the virtual base station pool.
Intel China Research Institute has developed the LTE mobile communication
base station based on Intel architecture, and made prototype verification and
reference design for wireless signal processing of physical layer in LTE standard,
MAC layer and RLC layer.
The white paper of C-RAN released by China Mobile Communication Institute
in October 2011 has developed centralized C-RAN prototypes based on the uni-
versal IT platform [6]
Facebook announced the new OpenCellular [2] open source platform in June
2016. It is a software-defined wireless access platform, a subsystem calculated by
General purpose Baseband Computing (GBC) of integrated power supply, synchro-
nization and other systems, and a design combined by front-end video subsystem of
integrated analog to support several mobile communication standards from 2G to

LTE. By providing devices and open source software needed to connect cellular
networks, OpenCellular allows telecom operators, device manufacturers,
researchers and entrepreneurs to develop, implement, deploy and operate wireless
infrastructure based on this platform, share and develop new telecommunication
hardware freely, and help normal users in remote areas with poor environmental
conditions realize Internet access.
As a result, we can see that the software-defined wireless communication system
has high flexibility and openness. It can realize the backward smooth upgrade of
mobile communication system, which has become a hot research topic at home and
abroad and achieved a lot of research results. But the realization of hardware and
software architecture and composition is a complex digital signal processing
system, which still lacks in-depth systematic study. 5G communications system is
completed based on GPP. Achieving all-digital and softening communication
signal processing is a task has not been solved well.
6.4.3 Introduction to Open 5G Universal Platform
For5G test and universal platform R&D based on GPP, Shanghai Research Center
for Wireless Communications has built and completed the open LTE network
platform based on universal server, which can well support the commercial termi-
nals access to Internet. The platform uses software defined radio design. The
processor uses Intel GPP. And the operating system uses open source Ubuntu
Linux. The development language uses C language. The entire platform (including
baseband) is totally implemented by software. The software architecture of this
platform is as shown in Fig. 6.28. The high-performance GPP is connected to
optical fibers via the PCIe bus and then to Remote Radio Unit (RRU) unit to realize
the complete functionality of the LTE system, including the base station, the core
network and the terminal. The core network includes the realization of software,
like MME, SGW, PGW and HSS and other network elements. The base station
includes software implementation of physical layer, MAC layer and RRC layer.
The core network may implement a software-defined full-function soft base station
on the same server, or may be connected to a plurality of third-party base stations to
achieve multi-user access.
The hardware architecture of Open 5G platform currently uses universal com-
mercial server hardware architecture. The key hardware involved in wireless signal
processing, transmitting and receiving includes CPU, Synchronous Dynamic Ran-
dom Access Memory (SDRAM), accelerator card and interface control board.
Far-end also includes Analogy Digital/Digital Analogy converter (AD/DA), RF,
and FPGA chips for front-end digital signal processing and interface control.
Multi-core CPU is used to allocate the signals to be processed to the kernel
which is used for baseband digital signal processing to process the transmission.
And some of the large computations, such as turbo decoding, is assigned to the
+LJKSHUIRUPDQFH*33 5HPRWH5DGLR8QLW 558
3&,HEXV 2SWLFDO
ILEHU 5)
6'5$0 &35, )3*$
&38 &38 $''$
6'5$0
3&,H
6'5$0 VORW
&38 &38 6'5$0 5)

&35, )3*$
$''$
6'5$0
)3*$ &35,
6'5$0
6'5$0
)3*$ &35,
6'5$0
FDOFXODWLRQDFFHOHUDWRUFDUGLQWHUIDFH
FRQWUROFDUG
Fig. 6.28 Architecture figure of Open 5G hardware
calculation acceleration unit by task scheduler module for calculation to reduce the
baseband processing load of CPU.
The interface unit is an extremely critical unit whose role is to provide high-
speed data rate and low-latency data channel between BBU and RRU. One interface
unit is connected to the universal server memory with the PCIe standard interface,
reads and writes digital baseband signals in the universal server memory with DMA
technologies. And the other end is connected to the remote RRU by optical fibers.
This interface uses a Common Public Radio Interface (CPRI) protocol, including
Path I and Path Q data for transmission, as well as system operation, maintenance
and control signals.
The realization of real-time processing of high-speed baseband signals by
universal server requires solving two technical challenges. One is real-time signal
processing capability of the universal server, and the other is data interaction
bandwidth of the internal interface.
With the development of multi-core GPP, the new architecture continues to
emerge, so that the baseband signal processing capability has reached a high
performance and has basically met the requirements of high-speed baseband signal
processing. Moreover, by increasing the calculation accelerator card, CPU and
FPGA/DSP form a heterogeneous computing platform, which can greatly improve
platforms baseband signal processing capabilities and basically satisfy the

baseband signal processing required by real-time communications. At the same
time, selecting appropriate high-speed interface can increase the data interaction
ability between the platform and the system and enhance the communication
throughput of the universal platform. The PCIe standard interface, which is used
in a universal platform, provides up to 16 Gbps transmission capacity and shares the
data bus directly. The transmission delay and bus delay are in the same order of
magnitude, which can ensure high-speed data rate exchange between the common
server platform and the computing accelerator card.
The hardware acceleration architecture shown in Figure 6.28 can meet the real-
time processing requirements of existing communication systems. As shown in
Fig. 6.29, a core network architecture based on the universal server. The joint
networking architecture of the core network and the access network based on the
universal server is shown in Fig. 6.30.
Fig. 6.29 EPC network

based on universal server
Fig. 6.30 Networking of

core network and access
network based on universal
server
Fig. 6.31 Software-defined mobile network architecture with the key functions in core network
and base stations
(1) The functions of the LTE core network and the base station are achieved based
on the universal server with versatility, openness and function customization
capacity of the core network equipment. Based on OAI open source EPC and
through increasing the functions, such as mobility, multi-user multi-base sta-
tion, stability, the open network based on universal platform EPC is
constructed. As shown in Fig. 6.31, the core network of LTE is deployed in
the universal server of Intel Linux. CPU uses Core i7-5557 U, 3.1GHz, dual-
core and four threads, and supports 16GB of memory with 5.0GT/s of bus
frequency and 4 MB L3 cache.
(2) Core network and soft base station joint network based on GPP are supported.
The functions of the core network (MME, SGW, PGW, HSS, etc.) and the base
station (RRC, Packet Data Convergence Protocol (PDCP)/RLC/MAC and
PHY, etc.) are realized. Data transmission and reception are realized through
USRP B210. Self-made SIM is used to support the commercial terminal access
network, as shown in Fig. 6.32, Frequency Division Duplex (FDD) and TDD
modes as well as 5/10/20 M bandwidth. Both the core network (MME, SGW,
PGW, HSS, etc.) and the base station (coder and decoder and other physical
layer functions, scheduling, power control, link adaptation, etc.) can be
enhanced and updated.
Although some progress has been made in the GPP-based Open 5G platform,
there are still some shortcomings in the existing technology solutions, which are far
from practical applications. Considering all kinds of technical challenges, core
technologies of baseband signal real-time processing, resource and network func-
tion virtualization, resource scheduling and task matching need to be R&D and
verified.
Fig. 6.32 Simultaneous access of multiple terminals and smooth switches among base stations
6.4.4 Key Technologies of Universal Platform Based

on General Purpose Processor
Real-Time Processing of Baseband Signal
In wireless communication systems, the real-time processing of baseband signal

possesses the characteristics of compute-intensive, and is faced with severe real-
time requirements. In order to ensure the real-time processing of base stations and
support the baseband data processing to guarantee the system performance, tradi-
tional communication systems often realize the baseband signal processing with
special hardware, such as DSP and FPGA. However, using dedicated hardware to
implement different carrier frequency bands, different coding, modulation, sam-
pling rates and different multi-user accesses will further aggravate the indepen-
dence of each other. Dedicated hardware is not conducive to the compatibility and
upgrading of many heterogeneous systems with different standard modes and
different networks. This not only reduces the resource utilization efficiency of
base station equipment, but also directly brings operators increasing costs of
maintenance and upgrading. Taking China Mobile for example, as shown in the
left figure in Fig. 6.33, the expenditure of main equipment and ancillary equipment
of base stations has reached 59% of CAPital EXpenditure (CAPEX) of cell site,
while the figure on the right of Fig. 6.33 shows the total energy consumption
constitution of China Mobile, of which 72% of energy consumption is from the
base station site of the wireless network access.
Therefore, the design of a universal, flexible, low energy consumption base
station system compatible with various modes is an urgent need to address the
problem. Using GPP to combine with universal operating systems (such as
Windows, Linux) and its convenient development environment and strong
CAPEX of cell site Base station energy

corollary
consumption analysis
equipment
base station power
6% 6% civil engineering consumption
7% 6%
24% transmitting 15% communication
machine room power
35% 16% equilibrium
consumption
base station 72%
management power
primary device consumption
13%
prospecting station
channel power
consumption
network planning
Fig. 6.33 CAPEX and energy consumption analysis of base station
reconfigurable capabilities are increasingly concerned. GPP has the advantages of

flexibility and openness in baseband signal processing of a base station, which can
integrate different mobile communications systems into one, providing a new
realization way for wireless signal processing. Facing future applications and
services, GPP real-time processing baseband signals have sustainable competitive-
ness, which is expected to be a solution for 5G communication systems and multi-
mode baseband processing platforms.
Although using GPP to perform baseband signal processing has shown advan-
tages in comparison with traditional dedicated hardware in terms of system cost,
capacity and flexibility, the traditional mobile communications base stations have
their own merits. Firstly, GPP is difficult to well support the computer-intensive
processing, which cannot meet the real-time requirements of mobile communica-
tions. Second, limited by the architecture and processing ability of GPP, commu-
nication algorithms are inefficient in GPP with poor real-time, stable and effective
features. Third, the power consumption of GPP chip largely hinders the develop-
ment of a GPP for baseband signal processing. Therefore, the construction of
mobile communication base station based on GPP must solve the bottleneck
problem of real-time digital signal processing based on GPP under the consider-
ation of time delay and overheads of transmission.
Software Acceleration Method
At present, the industry has not had the recognized implementation standard of
wireless baseband processing based on GPP. Various architectures and algorithms
are still under investigation. When dealing with network-intensive and computa-
tionally demanding baseband algorithms, the processing ability of GPP is not as
good as that of a dedicated chip. Performance bottlenecks may occur, and there may
be some distance from large-scale practical applications. First, because GPP usually
faces many tasks, operating systems (such as Linux) mainly aim for time-division
operation design. And baseband signal processing is computing intensive with high
real-time requirements. Therefore, the interrupt response and delay waiting process
should be avoided in the design of the signal processing algorithm, reducing the
complexity of algorithm. Efficient algorithm should be used in subsequent appli-
cation development as much as possible. Second, GPP has many defects in high-
speed data access. However, the high-speed cache mechanism can be used to
improve the storage efficiency. For example, control logic is specified in advance
to determine which data and instructions are stored in the on-chip cache or memory.
Finally, additional CPU resources are needed for GPP to manage multiple threads/
multiple processes. However, using computer multi-threading/multiple processes to
deal with and optimize big data and repetitive computations can increase the data
rate exponentially. In addition, with the instruction set to optimize the CPU
operation, the purpose of accelerating the baseband signal processing can be
achieved.
Instruction set-based software acceleration methods include the following types:
(1) Look Up Tables (LUT): LUT [7] is a compromise operation after examining
computation complexity and space complexity. The LUT operation can greatly
reduce on-line processing delay by replacing conventional bit operations.
(2) SIMD: Intel [8] CPU has a special multi-data instructions, and SIMD instruction
set can accelerate signal-level computing signal processing. SIMD repeatedly
performs the same operation for symbol-level data. A single instruction of SIMD
can handle several operations with low computational cost (computational
resources) and fully use bit bandwidth to significantly increase CPU efficiency.
(3) Sample level-Intel Integrated Performance Primitives (IPP): IPP [9] developed
by Intel is a set of software library of cross-platform and cross-operating systems
that can realize signal processing, image processing, multimedia and vector manip-
ulation and other operations. In [7], IPP is used to implement Fast Fourier Trans-
form (FFT)/Inverse Fast Fourier Transform (IFFT). The test results show that
FFT/IFFT is accelerated by IPP, which can improve the performance significantly.
Actual Test of Physical Layer of OAI Open Source Platform
Based on the OAI open source platform, the bandwidth to test each 5 MHz system
of FDD uplink/downlink and each module of processing delay performance of
physical layer of single-base station single-user system with SISO modes are given.
Each module data of physical layer on this platform is floating-point type, using
SIMD instructions to achieve the acceleration of data processing.
System performance test environment of the OAI open source platform includes
that EPC, base station and user equipment run on the same host. And the base
station and user equipment are directly connected through the UDP. The program
runs on an IBM System 3400 M3 server. CPU uses 2.13GHz and 4-core Intel
Xeon E5606 processor; 4G memory; 256G hard drive. The operating system Linux
Debian 7 uses 64-bit Ubuntu 14.04 DeskTop version and installs 64-bit low-delay
core. CPU uses the highest operating frequency.
Uplink
shared Multi-
channel Segmen- Turbo Data rate
CRC Interleaving channel Modulation
tation coding matching interleaving
IFFT & CP
SR /
CQI / RI / acknowledgement
acknowledgement PUCCH
UCI generation generation
Fig. 6.34 Uplink transmitting process [8]
PUSCH
Channel Log Data rate receiving
Data Channel Deinterlea Turbo
compensati IFFT likelihood de-
extraction estimation ving decoding
on estimation matching
FFT
SR / CQI/
aknowledgement
PUCCH signal Uplink control RI/acknowledgement
information
receiving receiving
Fig. 6.35 Uplink receiving process [8]
The uplink transmitting procedure of the OAI platform is as shown in Fig. 6.34.
First, the uplink shared channel information makes 24-bit CRC and segmentation.
Second, turbo channel coding is performed. Then, bit interleaving and data rate
matching are conducted. CQI, Rank Indication (RI) and acknowledgement infor-
mation are added, and then multi-channel interleaving and modulation are carried
out. Then Scheduling Request (SR) and acknowledgement information and other
Physical Uplink Control CHannels (PUCCH) are combined to complete IFFT. And
CP is added and finally the frame is formed. Uplink receiving flow is shown in
Fig. 6.35. PUCCH can extract scheduling request and acknowledge indication
information after FFT. While the acquisition of Physical Uplink Shared CHannel
(PUSCH) information is subject to data extraction, channel estimation, channel
compensation, IFFT, log likelihood estimation, data rate matching (at this moment,
uplink control information can be solved, such as CQI, RI and acknowledgement
information), de-interleaving and turbo decoding and other procedures.
Figures 6.36 and 6.37 describe the transmitting and receiving flows of downlink
data respectively. The former downlink shared channel needs to complete CRC,
segmentation and turbo coding first. The broadcast channel information and down-
link control information respectively complete convolutional coding, interleaving
and data rate matching, scrambling and modulation. Physical HARQ Indicator
CHannel (PHICH) generates HARQ indication information and Physical Control
Format Indicator CHannel (PCFICH) generates control format indication informa-
tion. Furthermore, these channels are multiplexed with Primary Synchronization
Signal (PSS), Secondary Synchronization Signal (SSS) and Reference Signal (RS).
After that, FFT and CP is added, and finally physical baseband signal is generated.
The latter can output HARQ indication and control format indication information
respectively by FFT, channel estimation, frequency compensation and physical
layer measurement. And the downlink shared channel information, broadcast
channel information and Downlink Control Information (DCI) are received through
Downlink
shared Segmentati Turbo Interleaving Data rate
channel CRC Scrambling Modulation
on coding matching
Broadcast channel
information Convolutio Data rate
Interleaving Scrambling Modulation
nal coding matching
Downlink control
information Convolutio Data rate
Interleaving Scrambling Modulation
nal coding matching
HARQ indicator FFT

PHICH channel generation
& CP
Control format
indicator PCFICH generation
PSS generation
SSS signal generation
RS generation
Fig. 6.36 Downlink transmitting flow [8]
data extraction, channel compensation, log likelihood estimation, descrambling and

data rate de-matching, de-interleaving, etc., and then respectively through turbo
decoding or Viterbi decoding.
As it can be seen from the above test results, for both the user side and the base
station side, Turbo decoding module is the most time-intensive course. Although
using GPP to realize real-time processing of baseband signals can reach a high
performance, when GPP handles some real-time modules with high computational
requirements, it may not meet the functional requirements of networks limited by
the core and architecture of GPP. In order to fully develop the effectiveness and
flexibility of the universal processing platform, using the hardware acceleration
method, such as combining GPP with FPGA/DSP is an essential means. As a result,
it can realize baseband signal real-time processing to break the bottleneck of real-
time processing of baseband signals (Tables 6.2, 6.3, and 6.4).
Resource and Network Function Virtualization
In recent years, communications, as a communication tool, is undergoing

significant changes. In order to bear the ever-increasing demand of users, operators
are constantly optimizing their own network. At the same time, challenged by
Internet services, telecommunication operators are trying to speed up the process
of intelligent management to avoid becoming a pipeline. However, now the net-
work has many problems, which hinders the network innovation and is embodied as
follows.[10]
Downlink
Log Data rate shared channel
Data Channel Deinterleav Turbo receiving
likelihood Scrambling de-
extraction compensation ing decoding
estimation matching
Broadcast
6.4 5G Open Source Community
Channel Log Data rate channel

Data Deinterleav Viterbi receiving
compensation likelihood Scrambling de-
extraction ing decoding
estimation matching
Channel Log Data rate DCI receiving

Frequency Data Deinterleav Viterbi
Channel PHY layer
FFT compensation compensation likelihood Scrambling de-
estimation measurement extraction ing decoding
estimation estimation matching
HARQ indicator
PHICH receiving receiving
Control format
indicator receiving
PCFICH receiving
Fig. 6.37 Downlink receiving flow [8]

345
346
Table 6.2 Turbo decoding test results of user-side downlink shared channel
Data rate 2.152Mbps 4.392Mbps 8.76Mbps 11.832Mbps 13.536Mbps 17.56Mbps
Number of code blocks 1 1 2 2 3 3
Average number of iterations 2 2 4 4.08 6.25 6.54
Average value 113.44(s) 234.68(s) 465.01(s) 680.27(s) 734.86(s) 1047.61(s)
Variance 28.52 74.27 494.92 1796.38 3765.47 17459.83
Mean squared error 5.34 8.62 22.25 42.06 61.36 132.14
Maximum value 127(s) 253(s) 533(s) 850(s) 952(s) 1514(s)
Minimum value 106(s) 221(s) 429(s) 644(s) 667(s) 923(s)
6 5G Hardware Test Evaluation Platform
Table 6.3 DFT module test results of base station side

Data rate 2.216Mbps 4.392 Mbps 8.76 Mbps 11.064 Mbps 14.688 Mbps
Average value 35(s) 35.03(s) 35.03(s) 36.08(s) 36.28(s)
Variance 0 0.02 0.03 0.1 0.22
Mean squared error 0 0.16 0.17 0.31 0.47
Maximum value 35(s) 36(s) 36(s) 37(s) 37(s)
Minimum value 35(s) 35(s) 35(s) 36(s) 36(s)
Table 6.4 Turbo test results of uplink shared channel of base station side
Data rate 2.216Mbps 4.392 Mbps 8.76 Mbps 11.064 Mbps 14.688 Mbps
Number of code 1 1 2 2 3
blocks
Average number 2.21 2.19 5.62 5.38 7.51
of iterations
Average value 151.99(s) 307.47(s) 642.04(s) 822.57(s) 975.09(s)
Variance 213.85 753.29 3122.41 4504.32 3462.69
Mean squared error 14.62 27.45 55.79 67.11 58.84
Maximum value 211(s) 459(s) 841(s) 1145(s) 1256(s)
Minimum value 107(s) 212(s) 448(s) 583(s) 735(s)
(1) Complex network function and poor flexibility. Mobile communication net-
work is formed by a large number of single-function network equipment.
Hardware resources are specialized and fragmented, and there are various
types and large quantity in equipment. Once the network is completed, it will
be difficult to change, expand and update, and resources cannot be rapidly
allocated on demand within the whole network.
(2) Hardware and software of network element equipment is vertically integrated
with closed architecture. In order to provide different services and multiple
access modes, and meet various requirements of QoS and security, the com-
munication network introduces a large number of control protocols and is
bound to a specific forwarding protocol. The code is directly or indirectly
written in hardware, constituting a closed architecture with integrated control
and forwarding. In the long run, the equipment is becoming more and more
bloated. The space for improving performance is small. Technical innovation
and upgrading are difficult, and the scalability is limited. The service develop-
ment cycle is long.
(3) The network and business form the chimney group phenomenon. The provi-
sion of new services often leads to proliferation of new equipment types and
quantities. The division between departments is easy to form a large number of
independent and closed network and business chimney groups. In addition, the
cost of infrastructure construction is high, and resources cannot be shared with
each other. Network and services cannot be coordinated and converged and
cannot adapt quickly to new services and new models.
(4) CAPEX and OPerating EXpense (OPEX) are still high. Because the special
nature of network devices results in the coexistence of equipment in different
manufacturers, it requires a lot of manpower and resources for different ages
and different standards, the procurement, integration, testing, deployment and
maintenance of equipment. At the same time, the insufficient competition
among different equipment manufacturers causes relatively high costs of
equipment, operation and maintenance management and upgrade. According
to 5G White Paper [16] 5G network architecture poses higher requirements in
the aspects of access speed, green power saving, cost-reliance, etc. Facing with
2020 and the future, the popularity of ultra-high definition, 3D graphics and
immersive video will drive a substantial increase in data rate, augmented
reality, cloud desktop, online gaming and other services, which will not only
make challenge for uplink and downlink data transmission rate, but also make
the harsh requirements of no perception on delay. In the future, a large
amount of personal and office data will be stored in the cloud. Massive real-
time data exchange will be comparable to the transmission rate of optical fibers
and make traffic pressure for mobile communication networks in the hot spot
area. Therefore, the new network architecture is imperative. The new network-
ing approach must address various problems of current networks so as to meet
the changing requirements of users.
First, as people have higher bandwidth requirements of wireless networks, for
mobile operators, the bulk of the increase comes from expenditures base stations
construction, operation management and network infrastructure upgrades. Besides,
enterprises are facing more and more intense competitive environment. However,
revenue may not grow at the same rate. The traffic of mobile Internet services is on
the rise due to competition, but the average revenue of single user has grown slowly
and sometimes reduced rapidly, which has seriously weakened the profit of mobile
operators. In order to maintain the ability to continue to make profits, mobile
operators must look for ways to increase the network capacity at a low cost, thus
providing better wireless service for users.
Traditional wireless base stations have the following characteristics. First, it is
difficult for each independently operating base station to increase the spectral
efficiency due to the capacity interference of system. Second, each base station
only covers a small area and connects a fixed number of sector antennas and can
only process the signal reception and transmission of the cell. Third, the base
stations are usually based on a vertical solution developed by the proprietary
platform. Compared with traditional base stations deployed by these above opera-
tors by connecting special line or optical fibers and core networks, the small base
stations in the future 5G access network are causally deployed by third parties or
users on the basis of requirements (such as deploying in a commercial and office
area or users home). This brings a huge challenge for operators. A large number of
small base stations mean high site matching and leasing, construction investment
and maintenance costs. Moreover, the greater number of small base stations means
that operators will have to pay more capital and operating expenses. The average
load of network of small base stations is generally much lower than the busy-time
load, and the actual utilization efficiency is very low. Meanwhile, different small
base stations are difficult to share the processing power, making the spectral
efficiency difficult to improve. Finally, the specified platform used by small base
stations means that the operator needs to maintain multiple incompatible platforms
at the same time, and will require higher costs when expanding or upgrading.
At present, with the rapid development of the virtualization technology, given its
characteristics like lightweight management and optimization resources, etc., build-
ing a virtual management platform for small base stations with intensive deploy-
ment is becoming more possible. Virtualization is a technology to make abstract
analog for computer and communication resources. The virtualization technology,
on the basis of the existing computer and communication hardware resources,
simulates all or parts of virtual hardware resources, such as baseband, CPU,
memory, input and output devices. These virtual hardware resources can share
the same platform with local real hardware resources, which is called virtual
machine. In general, from the software point of view, virtual machine and real
machine have no difference. That is, the realization and operation of virtual
machine are transparent for software programs.
Considering the evolution of mobile communication networks, IMT2020 has put
forward new requirements in order to complete the evolution of old network
architecture and adapt to 5G evolution: NFV, Cooperative Communications (CC),
Automated Network Organization (ANO), Flexible Backhauling (FB) network and
advanced traffic management and offloading and other key technologies [11]
Among them, NFV is the next generation of network building scheme proposed
and dominated by operators, with the purpose to bear more and more mobile
network function software through the use of 86 and other universal hardware
and the virtualization technology, thereby reducing the high cost of network
equipment. At the same time, with the hardware and software decoupling and
functional abstraction, the network equipment function no longer depends on the
dedicated hardware. The resources can be shared flexibly, so as to realize the rapid
development and deployment of new services. The automatic deployment, flexible
scalability, fault isolation and self-healing are made based on actual business
requirements. The architecture is as shown in Fig. 6.38.
The nature of the virtualization technology is the division and abstraction of
computing resources. The advantages, such as isolation, consolidation and migra-
tion, make it possible to integrate applications on different platforms safely and
securely to the same server, and migrate an application on a server quickly to other
servers, thereby improving server utilization, reducing hardware procurement and
operating costs, as well as simplifying system management and maintenance. The
virtualization technology has had nearly 50 years of history [12, 13] The first virtual
machine was System/360 Model 40 VM developed by IBM Corporation in 1965. In
recent years, with the rapid development of the computer hardware technology and
continuous innovation of the computer architecture, especially in the late 1990s,
desktop computer performance has been able to simultaneously support multiple
systems to run. At this point, the contradiction between increasingly powerful
Fig. 6.38 NFV virtualization architecture [13]
computing power of the computer system and relative backwardness of the com-
puting model has become increasingly prominent. The own characteristics of the
virtualization technology can find a balance between hardware systems with rapid
development and application requirements with complex changes, which provides
more advantages for enterprise-class applications, such as improving resource
utilization, reducing management costs, increasing the use flexibility, improving
system security, enhancing the availability, scalability and maneuverability of
system. Therefore, the virtualization technology has attracted the attention of
academia and industry both at home and abroad, and become one of the hotspots
in current research [14]
From 1990s to today, virtualization technology has made considerable progress,
and a variety of technologies become mature. In addition to VMware, Denali and
Xen, there are also many emerging virtualization software, such as KVM, Virtual
Box, Microsoft virtualization series products (Virtual PC, Hyper-V), Paralles
virtualization series products (Virtuozzo, Parallels Desktop for Mac), Citrixs
XenServer, Suns xVM, Oracle VM and VirtualIron. The main applications of
virtualization today involve the following domains: (1) server domains, including
data centers with high requirements for hardware platforms, cloud computing,
distributed computing, Virtual servers, etc.; (2) enterprise management software,
such as trusted desktop based on virtual machine, easily and effectively managing
and supporting employee desktop computer; (3) individual users, including
antivirus technologies based on virtual machine, program development and
debugging, operating system kernel learning, server consolidation, cloud
computing and data centers, virtual execution environments, sandboxes, system

debug testing and quality assessment, etc. With the further development of the
wireless communication domain and virtualization technology, the application
space will be further expanded. In the future, 5G is promising for applications in
many fields. At present, the main existed researches focus on the followings.
Peng et al. [15] present a 5G H-CRANs that is compatible with the software-
defined network architecture. The architecture converges virtual cloud computing
to achieve large-scale centralized coordinated processing of user data, involve the
traditional base station into a new communication entity Node C, and achieve the
integration of multiple RANs.
Chinas IMT-2020 5G Promotion Group released White Paper On 5G Network
Technology Architecture in 2015 [16] and proposed a 5G network cloud archi-
tecture based on SDN. The architecture is mainly composed of three parts: wireless
access cloud, control cloud and forward cloud. The access cloud supports the access
of multiple wireless modes, including Wi-Fi, LTE and mesh networks. It can realize
flexible network deployment and efficient resource management. Control cloud
uses virtualization technologies to realize the centralized control of network func-
tions logically. It can support control and open network capacity, satisfy the
requirements of different services and improve the service deployment efficiency.
Forward cloud realizes high reliable and low delay transmission of service data flow
based on unified resource scheduling under the control of control cloud.
Agyapong et al. [17] propose a two-layer 5G network architecture consisting of
the wireless network and the network cloud to support the massive MIMO system
and the separation of the control plane and the user plane. The architecture consists
of two layers of virtual logical networks that support wireless networks with the
simplest physical layer and MAC layer functions and all virtual network cloud with
higher layer functions.
Li et al. [18] propose a dual-virtualized 5G network architecture based on a small
base station and user level. In this architecture, the virtualization technology is
applied to small base stations with intensive deployment and users with different
requirements respectively. In the virtualization of small base stations, a primary
base station is responsible for the control signaling and mobility management, such
as broadcasting and paging, and for communicating with and managing other base
stations of a virtual cell. Virtualization of users virtualizes multiple users in one
cell, reduces the communication overhead between small base station and users,
and enhances the communication among different users in a virtual cell.
Through the analysis of existing research work, it can be seen that the
virtualization of current 5G networks focuses on the research of the overall network
architecture of the core network. The technology used mainly comes from cloud
computing and SDN, emphasizing the separation of control and forwarding, thus
realizing flexible network deployment and efficient resource scheduling. It is still
lacking in the extensive and in-depth researches in realization of virtualization
technologies of the access network and resource scheduling. In order to adapt to the
low delay and high efficiency of transmission in the future 5G, deeply studying the
virtualization technology of the core network and the access network can produce
enormous value in the future.
Dynamic Task Allocation and Resource Matching Scheduling
Task Segmentation and Parallelization Mode
The segmentation and parallelization of tasks has two main advantages. First, it is
conducive to the increase in data rate. GPP has the advantages in multi-core
processing, large storage and fast cache. Using task segmentation to make parallel
processing of multiple threads can increase the operating rate of processor and
reduce time-consuming. Second, it improves the program structure and is easy to
understand and maintain the program [19] In the case of multi-user multi-base
stations, the processing of some very complex communication processes is divided
into several threads, and the individual user processes can be independently run by
multi-threading, which can reduce the interference of multiple users.
The parallelization improvements are made for communication procedures.
First, it is needed to segment the task, dividing the large computational processes
of coarse granularity into a number of fine-granularity subtasks. And then parallel-
ism should be found according to the logic of sub-tasks. If there is a data depen-
dency among sub-tasks, it can be vertically converged. Otherwise, the horizontal
convergence will be made [20] Common methods of task segmentation of commu-
nication systems are as follows.
(1) Data segmentation. Data are divided into several independent data blocks,
creating several threads to deal with respectively, so as to complete the
processing of all data.
(2) Process segmentation. Complex process can be divided into a number of simple
sub-processes with a certain degree of independence, using multi-thread to
implement sub-processes to speed up processing speed.
(3) Problem segmentation. Complex problem is divided into several independent
sub-problems. Various sub-problems are solved through multiple threads, and
finally the solution of original problem is gotten.
Task parallel processing methods of communication systems, according to the
amount of calculation, mutual independence, real-time requirements and other
factors, can be divided into two types:
(1) Distributed parallel approach. Large tasks can be divided into several small
tasks which can make parallel execution to process. As shown in Fig. 6.39 [20]
the large tasks A-B-C-D are equally divided into four threads, thread 14. Each
thread processes 1/4 of the task. This parallel approach can load in balance,
which is suitable for the large tasks easy to be divided into a number of separate
strong parallel subtasks [21]
Taking turbo decoding as an example (as shown in Fig. 6.40), the turbo
decoder repeats the same operation for each code block. The turbo decoder has
strong independence among code blocks, which is suitable for parallel
processing and can effectively reduce the time consumption of processing.
However, the creation and scheduling of threads all need certain extra
thread1 A1 B1 C1 D1
thread2 A2 B2 C2 D2
Input data Output data
thread3 A3 B3 C3 D3
thread4 A4 B4 C4 D4
Time
Fig. 6.39 Task distributed parallel approach [20]
thread1
Codebook1 Codebook4 codebook7
Turbo decoder 1 Turbo decoder 4 Turbo decoder 7
thread2
Resource block Resource block
data input Turbo decoder 2 Turbo decoder 5 Turbo decoder 8 data output
thread3
Turbo decoder 3 Turbo decoder 6 Turbo decoder 9
Fig. 6.40 Turbo parallel decoding structure [22]
overheads, so the number of threads is comprehensively determined depending

on scheduling overhead, CPU resources, load balancing, delay and throughput
requirements and other factors.
(2) Pipelined approach. A large task is divided into a series of functional modules.
In parallel implementation process, each functional module is executed by
independent thread, and each task module operates at the same time, inputting
and outputting for each other. As shown in Fig. 6.41, thread1, thread2, thread3,
thread4 are responsible for the processing at A, B, C, D stages respectively. The
previous stage outputs enough data to trigger the execution at a later stage. This
parallel approach is suitable for dealing with the task that data have dependence
on a serial form. For data-related parts at different stages, it is needed to take
synchronization measures to protect it, and then reduce tasks time consumption
through the pipeline and improve the CPU utilization.
As shown in Fig. 6.42, the uplink receiving flow may be handled in a
pipelined fashion. The uplink receiving flow includes symbol level signal
processing, bit-level pre-decoding signal processing, turbo decoding and CRC
Input data
thread1 A1 A2 A3 A4
thread2 B1 B2 B3 B4
thread3 C1 C2 C3 C4
thread4 D1 D2 D3 D4 Output data
Time
Fig. 6.41 Task parallel pipelined method [22]
bit-level pre-
Uplink Symbol-level
decoding signal Turbo decoding CRC de-check
receiving signal processing processing
Multiple
threads Symbol-level bit-level pre- Decoding
decoding signal CRC de-thread
pipelined signal processing processing thread
processing
processing
Fig. 6.42 Schematic diagram of pipeline parallel processing of uplink receiving flow [23]
de-check and other function modules. Each function module is implemented

with different threads. Only when sufficient complete input data is obtained can
processing operation be executed without waiting for one sub-frame to com-
plete all the processing and execute the processing of next subframe, which
greatly reduces the idle wait time of each processing module. However, in order
to avoid pipeline errors, it is needed to take synchronization communication
mechanism to protect the data in different modules, so that the process before
and after the two steps can be connected.
Scheduling Strategy and Real-Time
Resource scheduling methods of threads have two types: dynamic and static. The
former means that the scheduler is responsible for dynamically mapping the
processing module needed to schedule to currently available CPU core perform
task processing. Due to the flexibility of dynamic scheduling, it is not suitable for
physical layer signal processing module with high data dependency. The latter
divides data flow into several sub-data streams, and then statically assigns different
sub-streams together with related task processing modules to different CPU core
task processing. Static schedule is suitable for dealing with the physical layer data
flow with a substantially constant data structure, which can improve the utilization
rate of high-speed cache and reduce the overheads of data synchronization, so as to
improve the integrity of communication system [23]
The existing scheduling strategies based on multi-thread programming support
both real-time and non-real-time processes, and the priority of real-time processes
is higher than that of any non-real-time processes. The schedule policy of real-time
process is divided into the strategy of first come first served and the time-slice
rotation strategy. The former is queued according to the priority, and the system
schedules the process group with highest priority. At the same priority level,
priority schedule first arrives at queues process, and the current executing process
can continue to consume system resources until a higher-priority process arrives or
it exits. The latter is similar to the former. The process group with the highest
priority is scheduled first. The process group with the same priority is scheduled in
turn by time slice. After the time slice of current execution process is exhausted, it is
placed at the end of queue.
Non-real-time process uses the time-division scheduling strategy. The system
schedules each thread in a time-slicing fashion. Each process defines its priority at
the time of creation. The higher the priority, the longer the time slices. After the
time slice allocated by current execution process is finished, it is placed at the end of
queue and waits for the next schedule. This schedule is essentially a scheduling way
to share the proportion.
Hardware Resource Scheduling Allocation Method
Communications system has high real-time requirements for real-time. In order to

meet the real-time requirements, the system often needs to use multi-core, multi-
CPU or even multi-processor to make coordination processing. Effectively sched-
uling universal hardware resources to meet real-time requirements is one of the
main indicators for building an Open 5G system.
A reasonable resource allocation scheduling method can make full use of the
ability of hardware resources to achieve the performance limit, while unreasonable
resource allocation scheduling method will seriously reduce the performance of
system. Therefore, in the improvement of system performance, it is very important
to optimize the allocation and schedule hardware resources. Hardware resources
include computing resources, storage resources, network resources, etc. Consider-
ations for hardware resource scheduling in setting up the Open 5G Universal
platform include:
(1) It is needed to take full account of the correlation among processes when
scheduler assigns the processor to process tasks. For the process with more
shared data or frequent data exchange, it is needed to be allocated on the same
processor, thereby increasing the hit rate of processor cache. This is because the
shared data of two processes has a greater chance to be stored in cache area,
which can then be retrieved directly from the cache, and the data interaction
between two processes can also be made directly through the shared cache. This
acquisition speed will be much faster than memory.
(2) Scheduler should ensure the balance scheduling of load. On the one hand, it can
prevent overload and shorten average response time of task. On the other hand,
the full use of resources in the entire system can improve the efficiency of the
resource utilization. The current scheduling methods to realize static load
balance mainly are various heuristic algorithms and methods based on graph
theory. The methods of implementing dynamic load balance include scheduling
algorithms based on the gradient model and the probabilistic scheduling algo-
rithm based on random selection [24] etc.
(3) Scheduler should try to minimize the migration costs in migration process. To
achieve load balance, schedulers are often required to make inter-core migra-
tion of task processes in different cores. When the task is scheduled, the
performance of the system is greatly influenced by the scheduling mode.
Because the costs in migrating different processes among processors are dif-
ferent, schedulers must take full account of memory amount occupied by the
process, CPU occupancy, restricted types (CPU-bound, input/output-bound),
etc., and fully consider the load balance of system and the costs of dealing with
load balance.
Based on software defined radio, open 5G platform can achieve a full-featured
communication system with GPP. Due to the large data dependency among mul-
tiple processing modules of the communication system, it is needed to comprehen-
sively use task segmentation and parallel processing. Allocating corresponding
hardware resources to the task threads with scientific scheduling strategy, the
platform can meet the system requirements of delay, timing and isolation in
wireless communication protocol. Besides, by effectively scheduling the comput-
ing resources of the universal platform and designing a reasonable real-time
scheduling algorithm, the utilization efficiency of computing resources can be
improved, and the real-time processing of baseband signals can be realized.
6.5 Summary
This chapter mainly describes the MIMO parallel channel test platform, the channel
model built on the parallel channel test platform, and the MIMO OTA platform and
the 5G open source community platform developed by Shanghai Research Center
for Wireless Communications. These three sets of platforms are the first platform in
China currently, and are being put into the 5G R&D process. These three sets of
platforms will continue to evolve, such as adding the parallel channel test function
of millimeter wave, sharing millimeter wave channel test data and providing OTA
test prototypes of millimeter wave equipment, etc.
References 357
References
1. https://www.fcc.gov/document/fcc-promotes-higher-frequency-spectrum-future-wireless-tech
nology-0
2. NI TDMS format http://www.ni.com/white-paper/3727/zhs/
3. USRP[EB/OL]. http://www.ettus.com/.
4. GNU Radio[EB/OL]. http://gnuradio.org/.
5. K. Tan, H. Liu, J. Zhang, J. Fang and et al. Sora: High performance software radio using
general purpose multi-core processors. Communications of the Acm, 2011, 54(1):99107.
6. China Mobile Communications Research Institute. White Paper of C-RAN Access Network
Green Evolution. Version 2.5. October 2011. (Chinese)
7. K. Niu, J. Sun, K. Chen, and K. Chai. TD-LTE eNodeB prototype using general purpose
processor. Equine Veterinary Journalii, 2004, 36(3):248254.
8. Intel. Intel 64 and IA-32 Architectures Optimization Reference Manual. May, 2007.
9. Integrated Performance Primitives for Intel Architecture Reference Manual.
10. PlanetLab: An open platform for developing, deploying, and accessing planetary-scale ser-
vices. http://www.planet-lab.org.
11. GENI: Global Environment for Network Innovations. http://www.geni.net/.
12. L. Peterson, Anderson T, Blumenthal D and et al. GENI design principles. IEEE Computer,
2006, 39(9): 102105.
13. VINI: A virtual network infrastructure. http://www.vini-veritas.net/.
14. N. Feamster, L. Gao and J. Rexford. How to lease the Internet in your spare time. ACM
SIGCOMM Computer Communication Review, 2007, 37(1): 6164.
15. M. Peng, Y. Li, Z. Zhao and et al. System architecture and key technologies for 5G hetero-
geneous cloud radio access networks. IEEE Network, 2015, 29(2):614.
16. IMT-2020 (5G) Promotion Group. White Paper on 5G Concept. Beijing: Release Conference
of White Paper on 5G Concept, 2015.
17. P. Agyapong, M. Iwamura, D. Staehle and et al. Design considerations for a 5G network
architecture. IEEE Communications Magazine, 2014, 52(11):6575.
18. Y. Li, P. Hao, X. Feng and et al. Cell and user virtualization for ultra dense network. 2015 I.E.
26th Annual International Symposium on Personal, Indoor, and Mobile Radio Communica-
tions (PIMRC), 2015:23592363.
19. T. Wang. The implementation of sending and receiving LTE PUGG and SRS based on GPP.
Beijing University of Posts and Telecommunications, 2013.
20. Q. Zhang. Research on the key technology of multi-core processor. Fudan University, 2014.
21. H. He. The Research and realization of LTE real-time communication link on GPP-based SDR
platform.Beijing University of Posts and Telecommunications, 2013.
22. W. Shi. Universal multi-thread parallel processing technology of wireless signal on GPP
platform. Beijing University of Posts and Telecommunications, 2014.
23. X. Zhang. Study on key algorithms and optimization of uplink receiver for LTE-A based on
GPP platform. Beijing University of Posts and Telecommunications, 2014.
24. Z. Tan. Thread schedule based on multi-core system. University of Electronic Science and
Technology of China, 2009.
Chapter 7
Field Trial Network
The ultimate goal of mobile communications technology research and development

and evaluation is application. The field trial is the last and also the most important
part of the technology from research to the application. During the R&D process of
5G key technology, the technological feasibility must also be verified by the field
trial, providing experimental basis for standardization.
This chapter combines with domestic and international 5G research vision,
analyzes the technical challenges of 5G system application scenarios and the field
trial, introduces typical cases of the field trial and intelligent analytic methods of
network testing data, and describes the feasible scheme for the R&D of the future
field trial environment. In Sect. 7.1, we analyze the requirements definition and
technical challenges of 5G application scenarios by EU METIS project and Chinas
5G Promotion Group. In Sect. 7.3, we give the corresponding test case explanation
in terms of the test methods of mobile communications wireless channel, key
physical layer technology verification, and network coverage performance, etc. In
Sect. 7.4, based on the existing network analysis methods, we propose an intelligent
network data analysis method. In the section of future test network planning and
outlook, we combine the 5G field trial challenges and analyze the field R&D
planning from aspects of application scenarios diversity, heterogeneous network
convergence, and large-scale users, etc.
7.1 Requirements and Technical Challenges
7.1.1 Various Application Scenarios
Mobile communications technology R&D and test and verification cannot be

separated from the application scenarios. The goal of R&D for key technologies
is to apply to the application scenarios. With the explosive growth of

DOI 10.1007/978-3-319-61869-2_7
360 7 Field Trial Network
communications needs, in 2020 and years to come, mobile Internet and IoT will
become the main driving force of the development of mobile communications.
However, with various scenarios and the extreme difference in performance
requirements, its unlikely that 5G system can form a solution of all scenarios
based on single technology just like before. Therefore, at the beginning of 5G
technology research, we must define the future application scenarios classifications
and the corresponding technical challenges.
Back in November 2012, EU launched 5G METIS project, in which the follow-
ing five characteristics of 5G application scenarios are summarized [1].
1. Fast, i.e., amazingly fast: 5G will guarantee a higher data rate for future
mobile broadband users.
2. Dense, i.e., Great in a crowd: 5G will guarantee that the densely populated
areas can get high quality mobile broadband access.
3. Complete, i.e., Ubiquitous things communicating: 5G is committed to efficient
handling of various types of terminal equipment.
4. Best, i.e., Best experience follows you: 5G is dedicated to providing mobile
users better user experience.
5. Real, i.e., Super real-time and reliable connections: 5G will support new
applications with more stringent requirements on latency and reliability.
In White Paper on 5G Concept (201502) [2] by Chinas 5G Promotion
Group, 5G technology scenarios are described. From the main application scenar-
ios, service requirements and challenges of the mobile Internet and IoT, four main
technology scenarios can be summarized, namely seamless wide-area coverage
scenario, high-capacity hot-spot scenario, low-power massive-connection scenario,
and low-latency high-reliability scenario.
Seamless wide-area coverage and high-capacity hot-spot scenarios mainly target
the demand for mobile Internet in 2020 and the years to come, which is also the
main technical scenario of the traditional 4G.
1. Seamless wide-area coverage is the most basic coverage means of
mobile communications, which provide users with seamless high-speed
service experience with ensuring the users mobility and service continuity as
its goal. The main challenge for this scenario is to provide more than 100 Mbps
user experienced data rate with guaranteed service continuity anytime and
anywhere (including harsh environment like coverage edge and high-speed
moving).
2. High-capacity hot-spot scenarios mainly target local hot-spot areas where ultra-
high data rates should be provided to users and ultra-high traffic volume density
needs to be handled. The main challenges include 1 Gbps user experienced data
rate, tens of Gbps peak data rate, and tens of Tbps/km2 traffic volume density.
Low-power massive-connection and low-latency high-reliability scenarios
mainly target IoT service, which is a newly expanded 5G scenario with the
key focus on IoT and vertical industry applications that traditional mobile
communications cannot provide good supports.
7.1 Requirements and Technical Challenges 361
Table 7.1 5G typical scenarios classification

No. Scenario features Typical scenarios
1 Tens of Tbps/km2 traffic volume density Office
2 Gbps user experienced data rate Dense residence
3 1 million connections /km2 Stadium, open-air gathering
5 6 persons/m2 super high density Subways
6 Millisecond level end-to-end latency Highway
7 Higher than 500 km/h mobility High-speed train
8 100 Mbps user experienced data rate Wide-area coverage
3. Low-power massive-connection scenarios mainly target sensor and data

collecting use cases, such as intelligent city, environmental monitoring, intelli-
gent agriculture, forest fire prevention, etc., which features small data packet,
low power and massive connection. This type of terminals are widely distributed
in large quantity, which should be able to not only support ultra hundreds of
billion connections and meet the 100 million/km2 connection density require-
ments, but also must ensure the ultra low power consumption and ultra low cost
of terminals.
4. Low-latency high-reliability scenarios mainly target special application require-
ments of IoT and vertical industries such as Internet of Vehicles (IoV) and
industrial control, which have extremely high indicator requirements on latency
and reliability. The ms-level end-to-end latency and nearly 100% reliability need
to be guaranteed in this scenario.
Specifically, 5G typical application scenarios include the following categories
(Table 7.1).
7.1.2 Technical Challenges in Field Trial
Chinas 5G Promotion Groups White Paper on 5G Vision and Requirements

(201405) [3] describes the 5G application scenarios and performance challenges.
5G typical scenario will touch many aspects of life in the future, such as
residence, work, leisure, and transportation. The 5G scenarios include at least
dense residential areas, office towers, stadiums, open-air gatherings, subways,
highways, high-speed railways, and wide-area coverage. These scenarios, which
are characterized by ultra-high traffic volume density or ultra-high connection
density or ultra-high mobility, may be quite challenging for 5G.
Some typical services, such as augmented reality, virtual reality, ultra-high-
definition videos, cloud storage, IOV, smart home, and Over The Top (OTT)
services, will take place in these scenarios. The performance requirements for 5G
are derived for each scenario, according to the predicted distribution of users,
percentage of different services, and service requirements such as data rate and
latency. The key performance indicators for 5G include user experienced data rate,
connection density, end-to-end latency, traffic volume density, mobility, and user
peak date rate.
From the mobile Internet and IoT main application scenarios, we can sum up that
5G will meet peoples various service requirements in residence, work, leisure, and
transportation. Even in dense residential areas, office towers, stadiums, open-air
gatherings, subways, highways, high-speed railways, and wide-area coverage
which are characterized by ultra-high traffic volume density or ultra-high connec-
tion density and ultra-high mobility. The extreme service experience like ultra-
high-definition videos, virtual reality, augmented reality, cloud desktop, and online
games can be provided to users. Meanwhile, 5G will also penetrate into and deeply
converges with IoT and a variety of industries such as industrial facilities, medical
equipment, and transportation tools, effectively meeting the needs of diversified
services in vertical industries like industry, medical service, and transportation to
realize a real everything connected world.
5G will respond to the challenges brought by diversified performance indicators
in the various application scenarios. Different application scenarios are faced with
different performance challenges in user experienced data rate, traffic volume
density, latency, energy efficiency and connections, all of which may be challeng-
ing to different scenarios.
METIS project converts five characteristics of the 5G application scenarios into
a series of highly challenging requirements in the form of numbers.
In the densely populated city environment, 5G will provide the rate of
101000 Gbps, 10100 times of the current situation.
The unit area or single user mobile data will increase by 1000 times, or exceed
100Gbps/km2 or 500GB/user/month.
The number of the interconnected terminal equipment will increase by 10100
times.
Battery service life of large low-power communications equipment will be
extended by 10 times. And that of terminals such as sensors or pagers will
reach 10 years.
Ultra-fast response applications such as touch Internet will be supported. The
end-to-end latency within five microseconds and the very high reliability will be
realized.
7.1.3 Development Status of 5G Test Bed in Foreign

Countries
In May 2014, Japanese operator NTT DoCoMo announced the start of the 5G
network experiment. It was reported that the companies that participated in NTT
DoCoMo 6 GHz-above spectrum 5G tests are Alcatel Lucent, Ericsson, Fujitsu,
NEC, Nokia and Samsung. NTT DoCoMo announced that 5G network commercial
application was expected to be launched in 2020. Compared with the 100Mbps to

150 Mbps speed in the existing LTE network, 5G network speed can be as high as
10 Gbps at maximum. NTT Docomo plans to do 5G experiments firstly in the
company lab in Yokosuka, Kanagawa Prefecture, and then do the outdoor exper-
iment. The company says that it will share its research results with peers in the
industry, and strive to promote the development of the 5G network standard starting
in 2016. In the 5G experimental project, NTT DoCoMo will cooperate with other
communications companies to carry out research, testing and verification. The tests
with Nokia mainly concentrate in millimeter wave technology and 70 GHz spec-
trum of beamforming. The cooperation with Samsung is mainly on 28 GHz spec-
trum tests. 15GHz spectrum of outdoor tests with Ericsson study new antenna
technology which can support the large-scale multi input multi output (massive
MIMO).In the tests on 36GHz spectrum with Fujitsu, NEC focus on time domain
transmit beam forming in the 5GHz spectrum. The cooperation with Alcatel
Lucent is mainly oriented to existing below 3GHz spectrum to mobile broadband
and M2 M applications.
In October 2014, SamSung carried out the 5G technology field experiment, in
which the 1.2 Gbps high-speed breakpoint-free data transmission speed in a car at
the speed of 110 km/h and the 7.5 Gbps peak transmission rate in stop state were
reached. The company has announced a partnership with SK Telecom in joint R&D
of 5G technology and scheduled to put the 5G network in trial during the
Pyeongchang Olympic Winter Games in 2018, in South Korea.
In September 2015, 5G Innovation Center (5GIC) of University of Surrey, UK
was officially founded. Its goal is to carry out 5G technology test in the real world,
rather than in the laboratory. It hopes to build a 5G network testbed, which is
expected to be completed in three phases. The first phase is expected to be
completed around April 2015. The first phase of the testbed will focus on the
construction and validation of cloud wireless access network in UDN, which will
display unprecedented network capacity to users. With the continuous improve-
ment of the testbed, a variety of 5G candidate waveforms, including Sparse Code
Multiple Access (SCMA) technology will be verified in the 5GIC testbed. When the
entire test platform is ready for operation, in 2016 or 2017 it can begin to be
deployed on campus of University of Surrey, covering the campus composed of
17,000 faculty members and students. As one of the founder members of the 5GIC,
Huawei plays an irreplaceable role in the testbed construction. All of the wireless
access equipment in testbed is provided by Huawei. Huawei has also set up a 5G
key technical verification experts team to provide on-site support to test bed
construction. In addition, Huawei will continue to invest 5 million pounds of
funds in the 5GIC joint R&D.
In September 2015, the largest U. S. mobile network operator Verizon
announced that it would start 5G wireless service field trials in 2016, and would
be committed to the commercial use of 5G services as soon as possible. Verizon has
begun to work with Alcatel Lucent, Cisco, Ericsson, Nokia, Qualcomm and
Samsung to carry out 5G tests, and constructed a 5G test site in San Francisco
innovation center.
Overall, globally people have begun to prepare and plan the 5G technology test
network for testing and validation. However, by far these test networks are mainly
for the cellular mobile technology test verification.
7.1.4 Development and Evolution of HetNet Convergence
Different from the traditional ways, 5G will no longer feature the single multiple
access technology. Instead, 5G should include cellular mobile communications
technology, new-generation WLAN technology and network technology. Since
the simple cell division technology is unlikely to improve the data capacity and
cell edge spectrum efficiency, Heterogeneous Network (HetNet) has gradually
caught attention [4]. Different communications technologies, such as cellular
communications 2G/3G/4G, broadband wireless access IEEE802.16/20 and short
distance communications WLAN, Bluetooth and UWB, offer a variety of services
to users. Using heterogeneous structure to build a mobile network is an effective
way to ease the hot-spot data traffic, which makes the HetNet will be the long-term
trend of the mobile network.
HetNet, in a broad sense, refers to the integration of a variety of wireless access
network technology, networking architecture, transmission mode and a variety of
transmission power of the base station types, such as adding WLAN hot-spots in the
mobile network. HetNet, in a narrow sense, means to add low-power nodes of the
same mode under the coverage of Macro eNodeB, such as micro cell, RRH and pico
cell, HeNB, relay node, etc [5].
Coordination and convergence between HetNets have become the focus of the
industry attention. Through the convergence of HetNet, we can fully utilize the
advantages of different types of network technologies and get various benefits. We
can greatly enhance the performance of a single network, support traditional service
and meanwhile create conditions for introducing new services. We can expand the
overall network coverage, so that the network has better scalability. We can balance
the network service load and increase system capacity. We can make full use of
existing network resources and reduce the cost of network operators and service
providers, thus making them more competitive. We can provide all kinds of needed
services to different users, so as to better meet the diverse needs of customers, and
improve customer satisfaction. We can also improve the usability and reliability of
the network and enhance the system survivability.
The concept of network convergence can be traced back to the 1970s, when
communications circles proposed the concept of network and service convergence,
such as the famous Integrated Services Digital Network (ISDN) and Broadband
Integrated Services Digital Network (B-ISDN) [6]. But limited by service and
technology development, ISDN failed. Until the 90s of last century, with the
development of mobile communications technology, it was proposed to develop
IMT-2000 global unified mobile communications standards, but also failed. Mean-
while, with the rapid development of Internet at the end of the last century, the
industry put forward the concept of the Next Generation Network (NGN). And the
research idea changed from network integration to network convergence, which for
the first time displayed the prospect of the convergence of the information and
communications network on basis of unified IP technology. The research fruits of
NGN on network convergence are embodied in IP Multimedia Subsystem (IMS)
technology proposed by 3GPP, which integrates telecom network, Internet tech-
nology fixed network and mobile network technology. Network convergence inte-
grates Internet IP technology, soft switch technology and cellular core network
technology. IMS technology itself has some limitations. But as a core network
convergence technology, it is widely recognized by the industry [7].
In fact, since 1990s, the convergence and development of telecom network,
radio and television network and Internet have been put on the agenda, that is, what
we often call triple play. Triple plays goal is to achieve the interconnection of the
three major networks, resource sharing and service convergence, to provide users
with a variety of services including language, data, radio and television and
multimedia services. Triple play is the product of constant information technology
innovation, but also the inevitable requirement of the development of information
technology. However, until now the triple play in its real sense has not been
completed. Since triple play has entered a substantive implementation stage, this
book focuses on the convergence and interconnection between heterogeneous
communications networks (especially heterogeneous wireless networks).
At present, there are two main directions for the development of multi network
convergence. One is based on the IP backbone network and the other is based on
hoc Ad. Multi network convergence system based on Ad Hoc can extend the
coverage of wireless communications, improving resource utilization, improving
the system throughput, balancing the traffic, and reducing the power consumption
of mobile terminals. It has become a research hot-spot in recent years. Especially a
series of research fruits have been achieved in terms of the combination of wireless
ad hoc network and cellular mobile communications system. And many practical
network models have been proposed, such as Ad hoc assisted GSM (A2GSM),
Multi-hop Cellular Network (MCN), Self-Organizing Packet Radio Ad hoc Net-
works with Overlay (SOPRANO), integrated Cellular and Ad hoc Relaying System
(iCAR), Unified Cellular and Ad hoc Network architecture (UCAN) [812].
A2GSM in the traditional cellular network can support mobile terminal multi-hop
relay in order to increase the system capacity, enhance cellular network coverage
and solve the haunting problem of cellular coverage blind areas. MCN allows
packets move via multi-hop transmission from mobile nodes to the integrated
communications network model of cellular communications system base station.
SOPRANO combines cellular network with wireless ad hoc network, and intro-
duces the dynamically-distributed wireless router with routing relay function in a
cellular network, so as to provide wireless Internet and multimedia services. The
basic idea of iCAR is to set up a certain number of Ad hoc Relay network Stations
(ARS), so as to achieve traffic load balancing between cells and to control network
congestion, relieve hot spots and avoid line dropping etc.
Globally, EU is leading in the study of HetNet convergence, and has carried out
a series of research projects. The BRAIN project proposes an open system with
WLAN and Universal Mobile Telecommunications System (UMTS) converged.
DRIVE project studies the convergence of cellular network and TV and radio
broadcasting networks. MOBilitY and DIfferentiated serviCes in a future IP net-
worK (MOBYDICK) project discusses the convergence of mobile network and
WLAN in IPv6 network system [13]. My personal Adaptive Global NET (MAG-
NET) project provides mobile users with ubiquitous and secure personal service via
design, R&D, and realization of PN in a HetNet environment. End-to-end Quality
Of Service Support over Heterogeneous Networks (EuQoS) focuses on end-to-end
QoS technology of HetNets [14]. WINNER project [15] hopes to use a ubiquitous
wireless communications system to replace the current coexistence pattern of many
systems (cellular, WLAN and short distance wireless access, etc.) and to improve
the flexibility and scalability of the system, so as to adaptively provide various
services in a variety of wireless environments. These projects, covering the aspects
of access, network and services that are both competing and cooperating with each
other, have made meaningful researches from a number of aspects and perspectives
on the HetNets convergence. Although they put forward different ideas and
methods on convergence of different nets, there is still a certain distance to the
convergence of different HetNets. Recently, the concept of network convergence
based on cognitive networks and wireless ad hoc networks has been proposed by
Network Ambient. It can provide a more effective way to realize the convergence
of HetNets.
7.2 5G Test Field Design
7.2.1 Network Architecture of HetNets Convergence
R&D of the 5G test field fully considers the HetNets convergence requirements of
3GPP and WLAN. And a unified control platform for HetNets convergence was
designed at the beginning of the design. Through the unified control platform and
the information interaction between the interfaces of the future 3GPP core network
and WLAN controller, HetNets convergence can be realized in the network side.
Basic architecture of the HetNets convergence network is shown in Fig. 7.1.
7.2.2 Abundant 5G Typical Scenarios
Similar to the under-construction 5G test field in U.K., our 5G test field is also
chosen in the campus environment, located in the campus of ShanghaiTech Uni-
versity. ShanghaiTech University is located in Pudong Science and Technology
7.2 5G Test Field Design 367
Fig. 7.1 Architecture of HetNets convergence network
Metro
ShanghaiTech
Meglev
International
Airport
Fig. 7.2 Geographical location diagram of ShanghaiTech University
Park (central Shanghai Zhangjiang High Tech Industrial Development Zone. Its
specific location is to the south of Chuanyang River, north of Middle Huaxia Road
of Middle Ring, East of Luoshan Road, and west of Jinke Road) which is
constructed by the Chinese Academy of Sciences and the Shanghai Municipal
Government. The specific location is shown in Fig. 7.2.
Fig. 7.3 Campus aerial view of ShanghaiTech University test field
The campus covers an area of about 900 mu (about 148.26 acres). And the
construction scale is 701,500 square meters of the total construction area, including
587,8000 square meters of Phase I new campus construction area (about 280,000
square meters of teaching and experiment buildings, sports facilities and auxiliary
buildings; about 160,000 square meters of student dormitories and teachers apart-
ments; about 150,000 square meters of underground construction). Universitys
industry-university-research base (Phase II Science and Technology Park) project
has a total construction area of 113,700 square meters.
Campus aerial view is shown in Fig. 7.3.
According to the above introduction, 5G mobile communications test field has
the following typical scenarios.
(1) High mobility scenario
The field is near the worlds only in-commercial-operation Shanghai maglev
line with the maximum speed of 430 km/h, suitable for high mobility scenario
test. Meanwhile it is close to the middle ring elevated road, suitable for 120 km/
h mid-to-high-speed mobility scenario test. Campus peripheral roads are suit-
able for below 60 km/h in slow mobility scenario test.
(2) A variety of indoor and outdoor typical scenarios
Compared with the single function park, this test field has abundant indoor
and outdoor typical scenarios. The field is divided into the area of teaching and
living by function. Among them, teaching area is mainly composed of teaching
buildings, research buildings, activity center, library, gymnasium and other
larger volume buildings, most of which are as high as 2030 meters. Living
area is mainly in high-rises, including apartment layoutthree apartments on
each floor, dormitory layout--rooms on both sides of the passway, as well as a
hotel. The outdoor scenarios include the ways between buildings, the open
areas like square and stadium, and the areas with landscape of river, lawn,
trees, etc.
(3) Large enclosed underground indoor scenarios
This field has a 150,000 square meters of underground scenario, which is not
just a single underground garage, but also a planned underground space with
living functions, which is very suitable for indoor coverage, D2D communica-
tions and IoT and other related technology tests.
7.2.3 3GPP Wireless Network Design
In order to realize the interconnection between LTE and non-3GPP access net-
works, the evolution of LTE system on the system network architecture is also
studied. In December 2004 in the 26th plenary session, 3GPP was officially
approved as a research project, which studies UTRA&UTRAN LTE feasibility.
High data rate, low latency, IP packet-based service and becoming competitive in
the next 10 years are the direction of the 3GPP system evolution research. In order
to achieve this goal, in addition to considering the evolution of wireless access
system, we also need to study the System Architecture Evolution (SAE) in order to
support the new LTE access network system. SAEs work objectives and research
direction are to achieve the full IP network architecture and provide real-time
services, and realize the interconnection between the network in evolution and
the existing 3GPP network or non 3GPP access networks (such as Worldwide
Interoperability for Microwave Access (WiMAX), WLAN). Supporting multiple
access systems is one of the basic principles of SAE, which is very important for the
competitiveness of SAE.
The main features of the SAE network include:
(1) Guaranteeing to support end-to-end QoS.
(2) Overall grouping. Providing pure packet access in the true sense, while no
longer providing a circuit domain service.
(3) Supporting multi-access technology. Supporting the interoperability with
existing 3GPP system and the access of non 3GPP networks (e.g., WLAN,
WiMAX). And supporting network roaming and switching between the 3GPP
network and non 3GPP.
(4) Adding support for real time services. Simplifying network architecture and
user service connection signaling process; reducing the service connection
latency. The time required for connection is less than 200 ms.
(5) Flat network hierarchy. Nodes of user plane are compressed as much as
possible. RNC is canceled in the access network. The user side nodes of the
core network are merged into one.
SAEs objectives are consistent with LTE. The first is to improve the perfor-
mance, reduce the latency, provide higher user data rates, improve system capacity
and coverage, and reduce the operation cost. The second is to achieve mobility
flexible configuration and implementation of an IP network-based existing or new
access technology. And the third is to optimize IP transmission network. However,
different from LTE, SAE tends to consider the trend and characteristics of future
mobile communications from the perspective of the whole system and to determine
the future direction of mobile communication from the angle of network architec-
ture. Under the condition that the wireless network interface technology is diver-
sified and homogeneous, the network architecture that can meet the future trend will
make operators more competitive in the future. And the users changing service
needs will also be well met.
In SAE system, non-3GPP access network is defined as IP access network whose
access technology is beyond 3GPP, such as WLAN, WiMAX, CDMA2000, etc.
According to the trust relationship between EPC and non-3GPP access network, the
access network can be divided into the non 3GPP access network and the
not-trusted non-3GPP access network. The criteria of trust for access network are
not decided by the characteristics of the access network, but by the operators
various strategies. The trusted non-3GPP access network indicates that the com-
munications between UE and EPC is secure. And all the communications between
the access network and the EPC is transmitted through the pre-established secure
link. The non-trusted non-3GPP access network indicates that the communications
between UE and EPC is not secure. And when UE joins the non-trusted non-3GPP
network, Internet Protocol Security (IPSec) tunnel between the access network and
EPC should be established.
3GPP Core Network Design
According to different trust relations between non-3GPP access network and EPC,
and different protocols used for connection of non-3GPP access network and EPC,
six communications scenarios are defined.
(1) Realize the billing information connection between WLAN and mobile cellular
network.
(2) Realize user WLAN access authentication and billing functions via mobile
network.
(3) Support users to get access to mobile network packet domain service
via WLAN.
(4) Support users to switch between WLAN and mobile networks and achieve
service continuity, though during the switch there may be brief interruptions.
(5) Support users to seamlessly switch between WLAN and mobile networks,
while ensuring service continuity and non-interruption.
(6) Support users to get access to the mobile network circuit domain via WLAN.
These six kinds of connected scenarios in order provide users with more and
more perfect convergent functions. At present, in WLAN and 2G/3G network, the
connection of the first three access scenarios are realized. With the development of
service model and technology, as well as the large-scale deployment of LTE

network, users want to be able to freely switch between LTE and WLAN to achieve
service switch, which requires the implementation of the interconnection between
four and five scenarios. For example, when a user is in a WLAN-covered area, he
hopes to get PS services via WLAN access operators. And when he leaves the
WLAN-covered area, he wants to be switched to the LTE network, while in order to
obtain better experience, switching between systems should be smooth. Con-
versely, the user in the LTE network is able to find the suitable WLAN around
automatically, and be seamlessly switched to WLAN.
Core Network Design Process
Core network is the core of the entire networks control and service functions. The
driving force of its development and evolution is mainly from two aspects: service
requirements and wireless access network evolution requirements, which are the
external causes of the evolution of the core network. The major changes in the
wireless network have brought the changes in the support system. For example, the
change from voice services to data services has brought the evolution from core
network to circuit domain and packet domain. With the demand for high-speed data
services, 2G network evolved into 3G network, and core network into soft exchange
time. With the demand for differentiated data services, core network further
evolved into SAE network. Network becomes flat and Policy Control and Charging
(PCC) is introduced for differentiated services. On the other hand, the internal
requirement for the core network, which mainly means to enhance network secu-
rity, improves the resources utilization and optimizes the network structure. For
example, the separation between call control and bearing has improved the net-
working flexibility and equipment utilization. The all-IP flat structure introduced by
SAE expanded the transmission bandwidth and reduced the end-to-end latency.
Core network EPC system network planning is the upgrading and evolution of
soft switching network. In principle, their network structures and soft switching
networks are similar in system planning, both of which need present network data
analysis, users and traffic prediction, preliminary exploration, network topology,
system design (including service test) and other steps. However, since SAE is based
on flat, full-IP system, in the network planning we must consider its unique system
characteristics, so as to give full play to SAEs technical advantages in conver-
gence, full-IP bearing, and high transmission efficiency. Also in the SAE network
planning and construction, we also need to consider the actual deployment of 2G,
3G network, landline network and WLAN network. Not only should we learn from
the planning scheme of the existing systems and cater to local conditions, but also
we should take into account the system coexistence and balance operation, striving
to achieve a perfect combination of excellent performance and low cost in the
design stage.
Network planning is the key work before the network construction. According to
factors like service demand, social and economic development, regional conditions,
and supporting facilitates, the capacity, QoS, network structure layout are config-
ured. And the construction principles of overall planning and step-by-step imple-
mentation are adopted. Overall planning means, in accordance with the principle
of all-process all-network, smooth expansion, and overall stability of network
structure, to develop a plan for the SAE network (short-to-middle term rolling
planning for small-range amendments). Step-by-step implementation means to
ensure the synchronization of network construction and market demand, avoid
resource waste, match each projects investment plans, and ensure the indepen-
dence of planning results as well as the sound construction and development of the
network.
Network Element Planning and Design
EPC core network includes S-GW, MME, P-GW, HSS, PCC, DNS, CG, BG. Full-
IP construction is used.
MME is the only control plane equipment of the core network. Its main functions
are access control, mobility management, session management, network elements
selection and user bearing information storage.
S-GW is located on the user plane. Each of the EPS-related UE has a S-GW
serving it at a point of time.
P-GW is located on the user plane, which is the gateway targeting Public Data
Network (PDN) and ending at Short Guard interval (SGi) interface.
HSS is a database used to store user contract information. The home network can
contain one or more HSS. HSS can also generate user security information for
authentication, integrity protection and encryption. HSS is responsible for the
connection between call control and session management entities in different
domains and subsystems.
Policy and Charging Rules Function (PCRF) is the policy and billing
control unit.
Bandwidth Requirements Bearing
For capacity allocation of the base station, we should mainly consider two aspects.
The first is the system bandwidth selection, and the second is the base station carrier
configuration.
Considering from the perspective of improving the spectrum efficiency, we
should use the large bandwidth, such as 20 MHz, 10 MHz. However, from the
perspective of improving coverage, we should choose the lower frequency bands as
much as possible. But most of the frequency bands below 2GHz in the present
3GPP-defined LTE frequency bands are currently occupied by other systems. These
bands may be released gradually in the future, but they are relatively dispersed.
While the variable bandwidth characteristic of LTE makes it possible to utilize the
dispersed bands.
When we make network plans in reality, we should consider the frequency bands
division of the operators and properly choose the biggest possible bandwidth
configuration.
In LTE capacity estimation, we need to consider two aspects: control channel
and service channel. Among them, downlink capacity should be estimated from the
maximal simultaneously scheduled users supported by downlink control channel
Physical Downlink Control CHannel (PDCCH) and the average throughput of a
single cell Physical Downlink Shared CHannel (PDSCH). Uplink capacity is
estimated mainly from the user numbers that can be borne on one Physical
Resource Block (PRB) in PDCCH and the average throughput of a single cell
PUSCH.
The carrier configuration of the base station should be considered from two
aspects: the number of users that can be scheduled by the control channel and the
traffic that can be borne by the service channel. The overall estimation formula is
shown in (Eq. 7.1).

Requirement of data traffic Number of users
Number of carriers max ;
Througput=Carrier Scheduled users=Carrier
7:1
Generally speaking, sector service demand is the main determinant.

Here are two ways to convert the cell throughput rate to the maximal number of
supported broadband users. One is based on the traffic, and the other is based on the
single user data rate. Suppose 20 MHz bandwidth, S111 configuration, 2 2
MIMO, and the cells with the frequency efficiency of 1.74bits/s/Hz/are used.Busy
time gets 15% of the whole day traffic, with the average load of 50%.
[Method One]
Single cell data service capacity is:
20MHz 1:74bits=s=Hz 35Mbit=s 7:2
When the unit is converted to GByte, then the single cell per hour processable
data traffic is:
35Mbit=s
3600s 15:38GB=h 7:3
8192
The busy-time processable data traffic when the cell average load is above
50% is:
15:38GByte=h 50% 7:69GB=h 7:4
Suppose the traffic per user per month is 5GB, then the average busy-time data
traffic per user is:
5=30 5% 0:025 7:5

Number of users can be borne by S111 base station:
3 7:69=0:025 921 7:6
[Method Two]
Single cell data traffic capacity: 35Mbits/s
The busy-time cell data traffic capacity when the cell average load is above 50%
is: 17.5Mbits/s
Required per activated user data rate: 1Mbits/s
Busy time user activation ratio: 1/20
Average busy-time throughput per user: 50kbits/s
The number of users that the three sector base stations can accommodate:
3 17:5 1000=50 1050 7:7
Through the above calculation, it shows that the LTE base station can accom-
modate a considerable number of broadband data users.
The transmission requirements of LTE base station mainly consider the trans-
mission bandwidth requirement of S1 interface and X2 interface. At the early stage
of network construction, when both the overall network traffic level and the average
busy-time throughput requirement are low, in order to give full play to the advan-
tages of LTE technology, the busy-time traffic requirements and the minimum peak
rate requirements should be considered in the interface transmission bandwidth
configuration.
LTE single-station transmission bandwidth requirements can be estimated based
on the service throughput and transmission overhead of each sector.
S1 interface BW Requirement of traffic

1 Transmission overhead 7:8
According to the test results of FDD test network, S1 interface user plane
transmission overhead is between 2% and 10%. At the early stage of network
construction, if we consider meeting the single-sector (20 MHz bandwidth,
2 2MIMO) peak data rate requirements, the S1 interface bandwidth configuration
should be:
172Mbit=s 1 10% 189Mbit=s 7:9
According to the test results of TDD test network, S1 interface user plane
transmission overhead is generally less than 10%.
At the early stage of network construction, if we consider meeting the single-
sector (20 MHz bandwidth) peak data rate requirements, the S1 interface bandwidth
configuration should be:
104:54Mbit=s 1 10% 115Mbit=s: 7:10

It should be noted that the peak data rate achieved here is the downlink peak data
rate with 3:1 time slot ratio. If the time slot ratio is 2:2, the downlink peak data rate
should be reduced by 1/3.
3GPP Wireless Access Network Design
Wireless network design is a systematic project covering from wireless propagation

theory research to the antenna equipment indicator analysis, from the network
capability prediction to detailed engineering design, from the network performance
test to the system parameter adjustment and optimization. It runs through the whole
process of the entire network construction, from the macro overall design ideas to
the micro parameters of every cell.
Network planning means, according to the needs of network objectives and
evolution and the cost requirements, to choose suitable network element equipment
for planning, finally output element number and element configuration, and deter-
mine the connection between the network elements, providing basis for the next-
step.
Wireless Network Design Process
The objectives of network planning and design can be summarized as 3C1Q,

i.e., with the minimal Cost, to obtain the best possible Coverage; with the
biggest possible system Capacity, to serve as many as possible users; meanwhile
ensure certain communications Quality. In order to achieve the above objectives,
we must follow a certain planning and design process to make scientific network
planning and design. The whole wireless network planning is divided into four
steps.
1. Preparation stage
At this stage, the main work involves collecting the market information,
geography, demographic information, customer demand, and other related infor-
mation needed for the planning and design; acquiring the preliminary informa-
tion of the network size; carrying out Continuous Wave (CW) test and
propagation model correction, sweep frequency test, indoor penetration loss
test, etc.; mastering the wireless environment of the service area, so as to provide
the basis for the following coverage and capacity planning.
2. Pre-planning stage
It mainly includes estimating and analyzing service model, network perfor-
mance, network size and equipment demand, determining the network construc-
tion strategy and specific construction objectives, making budget for coverage
and capacity, and determining the overall size of the network construction,
making investment estimation, etc.
3. Detailed planning and design phase

It includes the specific site selection, top surface layout, equipment selection,
code programming, parametric planning, etc. The effects of planning and
designing can be verified via the simulation and prediction software. If the
construction objectives remain unmet, the planning and designing schemes can
be adjusted until they do.
4. Project implementation phase
On the basis of the previous work, the plans are implemented in reality and
some necessary adjustments are made.
Propagation Model Selection
Wireless communications model plays a key role in the link budget. And the
coverage radius of the base station is determined by the maximal path loss allowed
in the link budget. Wireless communications model falls into two types, namely,
indoor and outdoor. The difference between the two models lies in the different
parameters. In the outdoor wireless propagation model, the landforms and buildings
in the propagation path are the influencing factors that must be considered, because
the signal fading is different in different environments. The wireless propagation in
free space has the minimal signal fading. The fading in open area/suburb is greater
than that in free space. The fading in general urban area is greater than that in open
area/suburb. And the signal fading in dense urban areas is greater than that in
general urban area. The features of the indoor propagation model are low transmit-
ting power, small coverage, and complex surrounding environment. In the follow-
ing part we are going to introduce several common communications models.
1. Free space model
Free space represents an ideal space or a space composed of isotropic media.
When the electromagnetic wave propagates in this space, there is reflection,
diffraction or absorption. The cause of the propagation loss is only the electro-
magnetic wave propagation. Satellite communications and LOS communica-
tions of microwave lines are typical examples of free space propagation. Under
certain conditions, the base station and the terminal antenna can be installed at
any height. In such a case, the communications between the base station and the
terminal is LOS communications. If the clear LOS communications exist
between the transmitting antenna and receiving antenna, the path loss follows
the free space model. The propagation loss model in free space is as follows:
PL 32:4 20lgd 20lgf 7:11
In it, d is the distance between the terminal and the base station, the unit is
km. f represents the frequency, the unit is MHz. This model is applicable when
the base station antenna and the terminal are installed at a certain height and
there is LOS between the base station and the terminal.
Table 7.2 Okumura Hata Model application range

Frequency range 1501500 MHz
Base station height 30200 M (The height of the base station
must be higher than the surrounding buildings.)
Terminal antenna height 110 m
Distance between transmitter and receiver 120 km
Table 7.3 Cost231-Hata Model application range

Frequency range 15002000 MHz
Base station height 30200 M (The height of the base station
must be higher than the surrounding buildings.)
Terminal antenna height 110 m
Distance between transmitter and receiver 120 km
2. Okumura-Hata model
As an evolutionary version of the Okumura model, Hata model is mainly used
in urban areas, whose application range is shown in Table 7.2.
Okumura-Hata Model can be expressed as (Eq. 7.12).
L 69:55 26:16lgf 13:82lghb ahre 44:9 6:55lghb lgd

7:12
For small and medium cities:
ahre 1:11lgf 0:7hre 1:56lgf 0:8dB 7:13
For big cities:
ahre 8:29lg1:54hre 2 1:1dB f < 300MHz 7:14

2
ahre 3:2lg11:75hre 4:97dB f > 300MHz 7:15
Among them, L is the path loss in urban areas with dB as its unit, f is the
systems operating frequency with MHz as its unit, hb represents the height of
the base station antenna with m as its unit, hre indicates the height of the terminal
antenna with m as its unit, a(hre) is the antenna height correction factor, d
represents the distance between the terminal and the base station with km as
its unit.
3. Cost231-Hata
Cost231-Hata can be used as the propagation model of the macro cell base
station, and its application range is shown in Table 7.3.
Cost231-Hata Model can be expressed as (Eq. 7.16).
Total L ahm Cm 7:16

Where
L 46:3 33:9lgf 13:82lghb 44:9 6:55lghb lgd 7:17
Among them, f is the systems operating frequency with MHz as its unit, hb
represents the height of the base station antenna with m as its unit, hm indicates
the height of the terminal antenna with m as its unit, d represents the distance
between the terminal and the base station with km as its unit, a(hm) is the
terminal gain capability. This function is related to antenna height, terminal
operating frequency and environment.
The values of Cm is determined by the landscape type. The values of Cm in the
standard Cost231-Hata model are as follows:
Big cities: Cm 3
Medium and small cities: Cm 0
Suburb: Cm 2[lg( f/28)]2 5.4
Rural open area: Cm 4.78(lgf )2 18.33 lg f 40.98.
Since the frequency of some of the mobile communications networks is more
than 2GHz, such as 2.3 GHz, 2.6 GLTE Hz and 3.5GHz, it has exceeded the
application range of the standard Cost231-Hata model. Therefore, in the actual
network planning and design, Cost231-Hata model must be corrected based on
the results of the CW test.
4. Standard Propagation Model (SPM)
SPM is particularly suitable for predicting the model of 150 ~ 3500 MHz
frequency band and long-distance communications (1 km<d<20 km).It is very
suitable for a variety of cellular mobile communications technologies. The
model is based on the terrain profile, diffraction mechanism and has considered
the clutter and effective antenna height to calculate the path loss.
This model can be used in any technology, and it can be calculated in
accordance with (Eq. 7.18).

LSPM K 1 K 2 lgd K
3 lg H Txeff K 4 DiffractionLoss 7:18
K 5 lgd lg H Txeff K 6 H Rxeff K clutter f clutter
The meaning of the parameters in the formula is shown in Table 7.4.

By using the simulation tool, the CW test is completed, which can be used to
calibrate the standard propagation model.
Coverage Design
1. TD-LTE coverage characteristics

If the target service is data service with certain data rate, then determining the
reasonable target rate is the basis of coverage planning.
Table 7.4 The meaning of parameters of SPM model

Parameter Meaning
K1 Offset constant (dB)
K2 lg(d ) constant factor
K3 lg (HTxeff) constant factor
K4 Constant factor of diffraction loss; K4 must be positive
K5 lg(d ) lg(HTxeff) constant factor
K6 HRxeff constant factor
d Distance from transmitter to receiver (m)
HTxeff Effective height of transmitting antenna (m)
HRxeff Effective height of receiving antenna (m)
DiffractionLoss Diffraction loss caused by obstacles. (dB)
Kclutter f(clutter) constant factor
f(clutter) Average weight loss caused by ground objects
In TD-LTE, there is no circuit domain service but only PS domain service.

The coverage capability of different PS data rates is different. In coverage
planning, we must first determine the data rate target of the edge users, such as
500kbits/s, 1Mbits/s and 2Mbits/s. Different target data rates have different
demodulation thresholds, and thus different coverage radius. So determining
the reasonable target data rate is the basis of coverage planning.
(1) LTE resource scheduling is more complex, and the coverage characteristics
and resource allocation are closely related.
TD-LTE network can flexibly select users Resource Block
(RB) resources and modulation coding modes and combine them so as to
cope with different coverage environments and planning requirements. In
the actual network, the user data rate is related with MCS and RB occupied
while MCS depends on the SINR value and RB occupied will affect SINR
value. So MCS, RB occupied, SINR value and user data rate will affect each
other, making LTE network scheduling algorithm more complex. In cover-
age planning, it is difficult to simulate the complex scheduling algorithm of
the real network. Therefore, how to reasonably determine the RB resources,
modulation and coding modes to make the selection more suitable for the
actual network conditions is a difficult problem in coverage planning.
(2) Selection of transmission mode and antenna type influences coverage
planning.
Multi-antenna technology is one of the most important key technologies
of LTE. The LTE network with multiple antennas has many kinds of
transmission modes (there are eight transmission modes at present) and a
variety of antenna types (two antennas, eight antennas, and so on at the base
station side). Selection of transmission mode and antenna type has great
impact on coverage performance.
(3) Inter-cell interference affects TD-LTE coverage.

OFDMA technology is introduced to TD-LTE system. Due to orthogonal
frequency between different sub carriers, the interference between different
users in the same cell can be ignored. But there still exists same-frequency
interference between cells of TD-LTE system. With increase of network
load, inter-cell interference level will increase, drawing down the user SINR
value and data transmission rate accordingly, showing certain respiratory
effects. In addition, different interference cancellation technology will pro-
duce different inter-cell interference suppression effects, which will affect
the TD-LTE edge coverage effect. Therefore, how to evaluate the inter-cell
interference is also a difficult problem in TD-LTE network coverage
planning.
2. LTE wireless network coverage strategy
The main objectives of LTE system coverage planning is based on the
requirements of the actual cell edge coverage. Under certain system parameter
setting, we can estimate the coverage distance that the base station can achieve,
so as to get the network size requirements. According to the application scenar-
ios and the actual planning requirements, the LTE systems coverage planning
strategy is generally divided into three categories.
(1) Network size estimation based on the requirements for uplink edge data rate
The first strategy is mainly used for the coverage requirement that only
the uplink edge data rate is limited. Based on the uplink data rate, with the
input of certain link budget parameters, the uplink coverage radius is
calculated. Based on the obtained uplink coverage radius, the uplink achiev-
able edge data rate is predicted.
(2) Network size estimation based on the requirements for downlink edge data
rate
The second strategy is mainly used for the coverage requirement that only
the downlink edge data rate is limited. Based on the downlink data rate, with
the input of certain link budget parameters, the downlink coverage radius is
calculated. Based on the obtained downlink coverage radius, the uplink
achievable edge data rate is predicted.
(3) Network size estimation based on the requirements for uplink and downlink
edge data rate
The third strategy is mainly used for the coverage requirement that both
the uplink and downlink edge data rate is limited. Based on the uplink and
downlink data rate, with the input of certain link budget parameters, both the
uplink and downlink coverage radii are calculated. And by comparison the
limited coverage radius can be obtained.
In the actual network planning, according to different requirements and
application scenarios, we need to select the appropriate coverage planning
strategy and flexibly respond to the problems in network planning.
Capacity Design
TD-LTE system capacity is determined by many factors. First of all is the fixed
configuration and algorithm performance, including single sector frequency band-
width, time slot configuration mode, antenna technology, frequency usage, inter-
cell interference cancellation technology, resource scheduling algorithm, etc. Next,
the actual networks overall channel environment and link quality will affect
TD-LTE networks resource allocation and modulation and coding mode selection,
so the network structure also has a crucial impact on TD-LTE capacity.
(1) Single sector frequency bandwidth. TD-LTE supports the flexible bandwidth
configuration of 1.4 MHz, 3 MHz, 5 MHz, 10 MHz, 15 MHz, 20 MHz.
Obviously, with greater bandwidth, network resources will be more available
and the system capacity will be greater.
(2) Time slot configuration mode. TD-LTE uses TDD, which can, according to the
different uplink-downlink proportions in an area, flexibly configure the uplink-
downlink time slot ratio. The present protocol defined seven kinds of uplink-
downlink time slot configuration modes, the special time slot of which have
nine modes to choose from. Different configurations have distinct differences in
uplink and downlink throughput.
(3) Antenna technology. TD-LTE uses a multi-antenna technology, so that the
network can, based on the actual network requirements and antenna resources,
realize single stream diversity, multi stream multiplexing, multiplexing and
diversity adaptation, single stream beamforming, multi stream beamforming,
etc. They have different usage scenarios, but all will affect the user capacity to
certain extent.
(4) Frequency usage. Currently, the analysis shows that TD-LTE network can use
the same frequency network. But the capacity performance of single cell with
the same frequency network system of the same bandwidth will be poorer than
that of the different frequency network systems. So in the actual operation we
should comprehensively consider the frequency resource, capacity require-
ments and other factors to determine the frequency usage.
(5) Inter-cell interference cancellation technology. For TD-LTE system, due to the
characteristics of OFDMA, the intra-system interference is mainly from other
cells with the same frequency. The co-channel interference will reduce the
users SNR and thus affect the user capacity. So the effect of interference
cancellation technology will affect the overall system capacity and the cell
edge users data rate.
(6) Resource scheduling algorithm. TD-LTE adopts the adaptive modulation cod-
ing method, so that the network can detect the test feedback in real time
according to the channel quality, and dynamically adjust user data encoding
modes and occupancy resources, achieving the optimal performance. There-
fore, TD-LTE overall capacity performance and resource scheduling algorithm
is closely related. Good resource scheduling algorithm can significantly
improve the system capacity and user speed.
(7) Network structure. TD-LTE user throughput depends on the quality of the
users wireless channel environment. Cell throughput is determined by the
overall cell channel environments. And the most critical factors affecting the
overall cell channel environment are the network structure and the cell cover-
age radius. In TD-LTE planning, we should pay more attention to network
structure than the 2G/3G systems, choose site strictly in accordance with station
distance principle, and avoid high stations and the sites deviating greatly from
the cellular structure.
TD-LTE capacity evaluation indicators.
The TD-LTE system capacity analysis can be separated into control plane and
user plane. Control plane capacity indicators include simultaneous scheduled user
number and simultaneous online user number. The simultaneous scheduled user
number is the basic indicator for system evaluation, specifically including uplink
and downlink control channel capacity, which is limited by the air interface
resources and channel configuration. Meanwhile the simultaneous online user
number is not only limited by the channel resources of the air interface control
plane, but also closely related to the equipment hardware processing capacity. User
face capacity indicators can be divided according to service types, including Voice
over Internet Protocol (VoIP) services and non-VoIP data services. Non-VoIP data
service indicators have cell peak throughput, cell average throughput and cell edge
data rate, etc. The VoIP service indicator is the number of VoIP users. The basic
indicators of the user plane include the average cell throughput and the number of
VoIP users. The above indicators are briefly discussed in the following part.
(1) Simultaneous scheduled user number: the number of users that can be sched-
uled by every system TTI.
(2) Simultaneous online (activated) user number: the number of users that maintain
the system connected status.
(3) Cell average throughput: When distributed in certain rules, the average
throughput of the whole cell equals to the sum of all cell throughput / cell
number.
(4) Cell edge throughput: throughput of the users distributed at the cell edge. In the
simulation system, the edge user is defined as the user at 5% when all the users
in the network are ranked in descending order according to the user throughput.
(5) The number of VoIP users: the total number of VoIP users contained in the cell.
VoIP user number is related with bandwidth configuration, control channel
resources and VoIP scheduling algorithm.
Site Resources Design
Base station site selection is an important link in mobile communications network

planning. Reasonable site location can use the network construction funds with
maximum efficiency, can avoid coverage blind area and reduce capacity
insufficiency to a certain extent, greatly improving the capacity, coverage and

quality of service. Generally speaking, a good site should follow the following
principles.
(1) Try to choose the site with regular cell structure that can meet the coverage
target of single edge users uplink and downlink throughput requirements.
Regular cellular structure can ensure the effective system coverage in the
planned area, avoid weak signal area, better achieve higher edge data rate,
and meanwhile facilitate future cell division. However, in the actual network
construction, due to various reasons, sometimes the site cannot be built in the
cellular center and the alternative suboptimal site around the ideal site has to be
found. Therefore, it is necessary to provide a search circle around each site. The
actual site can be found within the search coil. Generally speaking, search circle
radius deviation should be less than 1/4 of the radius of the base station cell.
Meanwhile, taking into account the yearly increasing requirements of the
commercial network, in the area where conditions allow, it can be considered
to reduce the minimum base station spacing.
(2) The site distribution should correspond to traffic distribution density. In site
selection, the areas with more users like government agencies, airports, railway
stations, news centers, important hotels and the main hot-spot regions, etc.
should be given priority to ensure good communication and avoid the
overlapping coverage. Meanwhile, the over-concentration of cell edge users
should be prevented, so that the site distribution matches the traffic density
distribution and the cell service users are concentrated in the cell center as much
as possible.
(3) The antenna of the macro cellular base station should be higher than the average
height of the surrounding objects. Exceeding values have different require-
ments for different terrains, landforms and environments. The antennas of
general urban site should be higher than the surrounding average by 5~10 m
(about 5 m in dense urban areas), and that of suburban and rural areas can be
decided according to the coverage requirements. For site planning, the site
antenna height in implementation should not be too different to the require-
ments of the planning, in order to avoid the emergence of the cross-area
coverage. The base station site should be selected in the main coverage area
with relatively high terrain or where there are high-rises or towers that can be
used. If the height of high-rises cannot meet the base station antenna height
requirements, there should be conditions for setting tower on roof or on ground,
so as to ensure a broad vision around the base station and there is no tall
buildings higher than the base station antenna, and most of the cell can be
covered by base station antenna via straight line propagation.
(4) Avoid strong interference sources near the base station. The base station site
should avoid the high power radio transmitting station, high power television
transmitting station and high power radar station. Strong interference will
greatly reduce the capacity of base stations and voice quality. Generally, in
site investigation, a spectrum analyzer must be used to scan the uplink and
downlink frequency near the base station so as to ensure that there is no strong
interference source. The isolation of LTE site and other wireless equipment
with similar frequency should be given full consideration. The sites where the
interference cannot be solved should be avoided, such as paging, microwave,
and other equipment with similar frequency.
(5) The site should be selected in the locations with convenient transportation,
good power supply and safe environment. Base station is the equipment that
needs long-term stable operation and regular maintenance. Any failure is likely
to reduce the users trust to the network operator and reduce the users support
for the network, causing users and income reduction and even the operation
failure. Transportation and city power supply must ensure the base station a
long-period high-power operation and the timely maintenance by maintenance
staff. Site should not be selected near the flammable and explosive buildings
and stacks, or near the industrial enterprises which emit harmful gases, heavy
smoke, dust, hazardous substances in the production process.
(6) Investment restrictions. In the site selection, the site with lower cost should be
preferred. For those high-cost sites, other alternative lower-cost sites should be
considered. In addition, while not affecting the overall layout, the existing
equipment room, power supply and other facilities should be utilized as much
as possible. At the early stage of network construction, in the case of insuffi-
cient funds, the coverage of important users and high density users should be
guaranteed as much as possible.
Indoor Wireless Network Design
Indoor distribution is a successful solution for indoor user groups to improve the
mobile communications environment in buildings. In recent years, it has been
widely used in mobile communications operators all over the country.
Indoor distribution system provides a good solution for indoor signal coverage.
Its principle is to use the indoor antenna distribution system to evenly distribute the
signals of the mobile base station in every corner of the room, so as to ensure the
indoor area an ideal signal coverage.
Construction of indoor distribution system can comprehensively improve the
call quality within buildings, enhance the mobile phone connection rate, and open
up a high quality indoor mobile communications area. Meanwhile, micro cellular
systems can share the outdoor macro cell traffic and expand network capacity, so as
to improve the overall service level of mobile network.
Indoor distribution system is to introduce the base station signal into the build-
ing, make a reasonable distribution via the power distribution device, and then
transmit signals via indoor antennas.
With the increasing number of high-rise buildings in the city, the user satisfac-
tion is also rising. These buildings are large in size and good in quality, and have a
strong shielding effect on the signal of mobile telephone. Especially in the lower
floors of large buildings, underground shopping malls, underground parking and
other environment, mobile communications signals are very weak. And mobile
phones cannot be used normally, forming the mobile communications blind areas
and shadow areas. In the middle floors, due to interference from surrounding base
stations, ping pong effect occurs, where mobile phones frequently switch or drop
off, seriously affecting the normal use of mobile phones. In higher floors of the
building, due to the height limitation of the base station antenna, normal coverage
cannot be achieved. Its also a blind area of mobile communications. Besides, in
some of the office buildings, although the mobile phones can be used for normal
phone calls, the mobile phones are difficult to go online because of the high user
density and the base station channel congestion. Under the fierce competition
environment, the indoor mobile communications network coverage, capacity,
quality are key factors for operators to gain the competitive advantage. It funda-
mentally reflects the service level of mobile network and remains the priority of
mobile operators in recent years.
7.2.4 WLAN Wireless Network Design
Analysis on New Generation WLAN Technology
The Latest Development of WLAN Technology
WLAN, featuring high throughput and low cost, can be well combined with the
Internet service to provide users with convenient high-speed wireless Internet
access service. With the rapid development of the Internet, WLAN equipment is
rapidly becoming popular all over the world, offering great convenience for peo-
ples work and life. According to Wi-Fi Union statistics, currently there are over
1 billion WLAN users globally. Telecom operators also attach great importance to
WLAN and see it as an important complement and extension of fixed network and
cellular network. They deploy WLAN hot spots in large scale to provide services to
the public, which further promotes the WLAN development.
In order to meet the growing market demand, WLAN technology and standards
are constantly developing and improving, and data transmission capabilities con-
tinue to improve. After more than 20 years of development, the maximum infor-
mation transmission rate of WLAN equipment in the current IEEE 802.11n device
reach up to 600Mbits/s. In addition, the 802.11n device also uses two channel
access mechanisms, static 40 MHz and dynamic 20/40, significantly improving the
system throughput. IEEE has launched the next generation IEEE 802.11 ac/ad
WLAN technical standards, whose data throughput will reach 7Gbits/s at maxi-
mum, better meeting the market demand for high throughput wireless data services
such as wireless High Definition (HD) video transmission. IEEE 802.11 ac project
has already begun as early as the first half of 2008, when it was known as very high
throughput with the goal directly reaching 1Gbits/s. In the face of the requirements
of multiple HD video and lossless audio for over 1Gbits/s code rate, IEEE 802.11 ac
was helpless, especially in the indoor high speed data transmission environment.
Thus, IEEE 802.11ad was proposed, which would be used to achieve the transmis-
sion of domestic wireless HD audio and video signals, and bring more complete HD
video solutions for home multimedia applications.
In order to achieve higher wireless transmission rate, IEEE 802.11ad has aban-
doned the crowded 2.4GHz and 5GHz frequency bands, but rather uses 60GHz
frequency spectrum high frequency carrier with 57 ~ 66GHz unallocated frequency
band. With such a wide bandwidth, the data transmission rate can be greatly
improved. Since 60GHz spectrum in most countries (including the United States)
has large quantities of available frequencies, IEEE 802.11ad can realize the simul-
taneous multi-channel transmission with the support of MIMO technology with the
bandwidth of each channel exceeding 1Gbits/s. On the basis of integration of IEEE
802.11 s and IEEE 802.11z, it totally can used to realize file transmission and data
synchronization between equipment, and the speed will be faster than the second
generation Bluetooth technology by more than 1000 times. Of course, its main
purpose is to achieve HD signal transmission.
IEEE802.11 ac
802.11 ac is specifically designed for the 5GHz band. The unique characteristics of
the new radio frequency improve the performance of existing wireless LAN until it
can be comparable with the level of wired Gigabit network. 802.11 ac is a new
standard of IEEE wireless technology. It draws on the advantages of 802.11n and
further optimization. In addition to the most obvious characteristics of high
throughput, it has a lot of improvements in many aspects.
Improvement on the 1802.11 ac standard physical layer
802.11 standards include physical layer and MAC protocols. Since the first
release, the physical layer has made a number of important additions and amend-
ments, while most of the MAC basic features remain unchanged. Here we focus on
the changes in the physical layer of 802.11 ac.
(1) A wider channel bandwidth
802.11 ac supports 80 MHz bandwidth, and choose to use continuous
160 MHz frequency band or discontinuous 80+80 MHz frequency band. Its
channel distribution is shown in Fig. 7.4
Fig. 7.4 802.11 ac channel distribution

The increase in bandwidth brings more available data subcarriers. 80 MHz

has 234 available subcarriers while 40 MHz has only 108, so 80 MHz can bring
about 2.16 times of the growth rate. The accompanying side effect is that the
same transmission power has to be separated on the added sub carriers,
resulting in a slightly reduced signal coverage.
(2) Choosing 5 GHz band
The most important reason for the significant improvement in 802.11 ac
performance is technological innovation, and most crucially 5GHz band is
used. Previous 802.11 devices always have to face the problem that the network
equipment based on these standards have to compete for channels with other
non 802.11 devices, such as baby monitors, Bluetooth earphones, or even
microwave ovens, etc. These devices are also using 2.4GHz frequency band.
When all the equipment is competing for the 2.4GHz band, its just like a road
crowded with cars and everyone slows down. The 802.11 ac works on the not-
that-crowded or cleaner 5GHz band with fewer wireless devices competing
for bandwidth, so the speed is also guaranteed. Of course, the 802.11 ac
standard has downward compatibility, ensuring that the 802.11 ac device can
be used on the existing Wi-Fi network.
(3) Higher order modulation
802.11 ac uses OFDM technology to modulate the data bits in the wireless
medium transmission. According to conditions, 802.11 ac can choose
256 QAM, which makes each subcarriers data bits increase from six to eight
so that the throughput increases by 33%. But 256 QAM is only applicable in
high SNR environment, that is, in good channel conditions.
(4) More spatial streams and MU-MIMO
802.11 ac supports up to eight channels of spatial streams. Supporting
multiple spatial streams is optional, but the increase in spatial streams is the
most effective when combined with the 802.11 ac MU-MIMO new features.
802.11 ac technology, in single user and multi-user MIMO mode, supports up
to eight channels of spatial streams, up to four users. And in multi-user mode,
each wireless terminal has no more than four channels of spatial streams.
(5) Carrier interception technology enhancement
802.11 ac standards also apply a number of MAC layer enhancements to
further enhance the performance of RF and MU-MIMO characteristics. In
802.11n, Request to Send/Clear to Send (RTS/CTS) undertakes clearing
tasks, stopping the transmission of 802.11a/g devices during the transmitting
to avoid collision. In 802.11 ac, since 80 MHz uses more channels, RTS/CTS
mechanism needs to be enhanced to handle communications collision on the
secondary channel. The improved RTS/CTS also support the dynamic
bandwidth mode.
In MAC layer, the 802.11n device announces its transmission intention via
sending RTS/CTS frame. These frames allow 802.11a/g devices nearby to
sense the channel being used in order to avoid collision.
Due to 802.11 acs wider bandwidth and limited channels when 80 MHz
bandwidth is in use, the hidden nodes on the second channel have become an
important problem to be solved. The RTS/CTS mechanism has been updated to

better detect whether or not a non-main channel is occupied by different
transmissions.
Therefore, RTS and CTS (optional) support the dynamic bandwidth mode.
In this mode, if the partial band is occupied, the CTS frame is transmitted only
on the main channel. The client (STA) that sends RTS frame can be dropped to
a lower bandwidth mode, which will help to reduce the impact of the hidden
nodes. Anyway, the final transmission bandwidth always includes the main
channel.
(6) Message Aggregation Enhancement
In 802.11 acs basic MAC protocol, in order to ensure that each station can
achieve fair access to media usage and avoid collision, a series of control
mechanisms are used. These mechanisms can improve the system performance
but also bring the fixed overhead, which limits the increase in system through-
put. In order to increase the size of Aggregated MAC Protocol Data Unit
(A-MPDU) and reduce the communications overhead, 802.11 ac introduces
two kinds of frame aggregation methods: MAC Service Data Unit (MSDU)
aggregation and Message Protocol Data Unit (MPDU) aggregation. These two
frame aggregation methods reduce the overhead of the single channel RF lead
code during the transmission of each aggregated frame.
IEEE802.11ad
In the face of the requirements of multiple HD video and lossless audio for over
1Gbits/s code rate, 802.11 ac was helpless. Therefore, 802.11ad came into being,
which will achieve the ultra-high data rate of 7Gbits/s and be mainly used for
domestic wireless HD audio and video signal transmission, bringing more complete
HD video solutions for family multimedia applications .
After 60GHzVHT project got approval in December 2009, 802.11VHT working
group passed the 802.11ad Draft 2.0 draft in March 2011, and the official standard is
expected to be published by the end of 2012.
In order to achieve a higher rate of wireless transmission, 802.11ad abandoned
the crowded 2.4GHz and 5GHz band. Instead it uses high frequency carrier 60GHz
spectrum (57GHz66GHz).Since 60 GHz spectrum in most countries have large
quantities of available frequencies, 802.11ad bandwidth of each channel can reach
2.16GHz, which will be 50 times of the 802.11n channel. In addition, 802.11ad also
uses adaptive beamforming, a variety of physical layer types, PBSS network
architecture, mm Wave channel access, fast session migration and other enhance-
ment technologies to improve the system throughput and coverage. However,
802.11ad is also faced with its technical limitations. For example, 60GHz carrier
has very poor penetration ability, and its signal attenuation is very serious in the air,
greatly affecting its transmission distance and signal coverage. So its valid connec-
tion can only be limited in a small range.
To maximize the system performance and minimize the system implementation

complexity and cost and also support the interconnection with the existing 802.11
standard, 802.11ad at the physical layer and MAC layer have been enhanced via the
following key technologies.
(1) Greater channel bandwidth
802.11ad uses 60GHz frequency band, which is license free and has very
large available bandwidth. Chinas frequency allocation of this frequency band
is 59G-64GHz. Compared with the 83.5 MHz available bandwidth on the
2.4GHz frequency band, 60GHz frequency band has the 5G-9GHz available
bandwidth on it. This bandwidth can be divided into multiple channels, each of
which has the bandwidth of 2.16GHz. With such a wide channel, 802.11ad does
not need to use the channel binding technology. And it can support the
application with high bandwidth requirement such as the non-compressed
video stream transmission.
(2) Different physical layer types are supported for different application scenarios
802.11ad supports a variety of physical layer types, including the PHYsical
Control layer (PHY Control), Single Carrier PHYsical layer (SCPHY), Low
Power Single Carrier PHYsical layer (Low Power SCPHY), OFDM PHYsical
layer (OFDMPHY).
Control PHY is designed for low SNR operation, in which single carrier
mode is mandatory and the maximum data rate is 27.5Mbits/s. With /2-DBIT/
SK modulation, 1/2 data rate Low Density Parity Check (LDPC) code, and the
maximum packet length limit of 1024 bytes, it can achieve shorter data block
and get better robustness in the high noise environment.
SCPHY allows for low power, low complexity transceiver and supports the
data rate as high as 4.6Gbits/s.
Low Power SCPHY reduces the processing power of the transceiver by a
simplified encoding method and a shorter symbol structure. Its maximum data
rate is 2.5Gbits/s.
OFDMPHY can achieve better performance in frequency selective channels.
It uses 512 point FFT, including 336 data sub carriers, 16 pilot sub carriers, and
three Direct Current (DC) carriers. With 64QAM modulation, in 13/16 data rate
and LDPC coding, it supports the data rate as high as 7Gbits/s.
SCPHY and OFDMPHY have different application scopes. OFDMPHY,
which is suitable for transmission in long distance and large scalability latency,
can deal with the multi-path propagated signals. SCPHY generally can get
lower power consumption and is suitable for small, low-power handheld
devices.
(3) Adaptive beamforming
Due to the huge transmission loss in the 60GHz band, 802.11ad uses
adaptive beamforming technology to achieve data transmission over long
distance above ten meters.
Adaptive beamforming, by adaptively adjusting the antenna directions and
reducing the beam width, obtains a higher antenna gain, reduces interference,
and expands the signal coverage. Besides, if there is an obstacle in the line of
sight of the transceiver, the transceiver can quickly evade the obstacle and
rebuild a new link for communication. Beamforming can be realized by
different techniques such as beam switching, phase weighted antenna array,
multi antenna array, etc.
Beam shaping protocol consists of three stages: Sector Level Sweep (SLS),
Beam Refinement Phase (BRP), and Tracking phase (Tracking). SLS is divided
into a transmitting SLS and receiving SLS. The former is used to determine the
optimal transmitting direction of the transmitting device and the latter is used to
determine the optimal receiving direction of the receiving device. BRP further
optimizes the beam direction through the joint adjustment of transmitting and
receiving beam. Tracking stage is used to dynamically adjust the beam
according to the channel changes in the process of data transmission.
(4) New network architecture PBSS
802.11ad defines a new network architecture Personal Basic Service Set
(PBSS), which allows direct communications between two devices. PBSS
network architecture falls into the ad hoc network, similar to Independent
Basic Service Set (IBSS).But the difference between themis that in PBSS
architecture, one partys STAtion (STA) bears the PBSS Control Point (PCP)
function. Only the PCP can transmit beacon frames.
(5) mmWave channel access mechanism
The medium time in PBSS system is divided into Beacon Intervals
(BI) structure, and the sub time interval in BI is the access time. Different
channel access time follows different access rules. As shown in Fig. 7.5, one BI
contains the following four types of access time. First, Beacon Time (BT): PCP
sends beacon frames in different directions and discovers new STA. Second,
the Associated BeamForming Training time (A-BFT): Performing the
beamforming training between PCP and STA, establish beamforming link.
Third, Announcement Time (AT): It is used to transfer control frames between
PCP and STA. Fourth, Data Transmission Time (DTT): It is used to accomplish
data frame exchange between STAs, including scheduling-based Service
Period (SP) and Competition-Based access Period (CBP), or a combination of
any number of the two. During the SP, the scheduled STA can access the
channel. During the CBP, all STA can access the channel based on the
802.11DCF and HCF mechanisms. DTT scheduling information is transmitted
from STA to PCP through beacon frame and announcement flame.
BI
DTT
BT A-BFT AT CBP 1 SP 1 SP 2 CBP 2
Time
Fig. 7.5 BI structure schematic diagram

(6) Multi-band inter-operable Fast Session Transfer (FST)

In order to achieve the interoperability of 802.11ad and 802.11a/b/g/n/ac,
and support the seamless switching between Wi-Fi communications in the three
frequency bands, 802.11ad uses FST technique. In such a case, as for user
experience, the users with multiple frequencies device can enjoy uninterrupted
continuous communication between the Wi-Fi networks of different modes.
When covered by 802.11ad signals, they can experience high-speed network
connection, and when 802.11ad signal is poor or does not exist, they can
seamlessly switch to Wi-Fi networks.
(7) Power saving technology
802.11ad devices can reduce power consumption by utilizing scheduling
access mechanism. The two devices using directional link communications can
access the channel at the scheduled time, and begin hibernation at the schedule
interval so as to achieve the purpose of power saving. This technology can
enable the device to manage the power overhead based on the actual opera-
tional load, which is very important for mobile phones and other handheld
devices. 802.11ad defines the non-PCP power saving mechanism and the PCP
power saving mechanism, allowing non-PCP STA and PCP to hibernate for
several BI times.
In non-PCPSTA power save mode, after STA and STA negotiation, STA enters
the hibernation mode in a specific channel access interval, and selects the wake-up
interval according to STAs own service needs. PCP tracks the hibernation interval
of the STA associated with it, stores MSDU cache for the STA in hibernation mode,
and passes on the cache service to STA when STA is woken up.
In PCP Power Save (PPS) mode, PCP can make one or more continuous BI enter
the hibernation state, so as to reduce its power consumption. PCP running in PPS
mode will no longer send beacons. Before entering hibernation, PCP will notify the
non-PCPSTA in PBSS about the necessary information through beacon frame or
announcement frame, so that PCP is still able to communicate with STA in
hibernation.
WLAN System Architecture
WLAN consists of an end STA, an AP, an Access Controller (AC), an Authentica-

tion Authorization Accounting (AAA) server and a network management unit.
AAA server is an entity that provides AAA services. In the model, the AAA server
supports the Radius protocol, and the portal server is the entity that is used to push
the portal website. In the WEB certification, it assists to complete the certification
function. Figure 7.6 is the WLAN network model.
Since IEEE802.11 standard was put forward in 1997, WLAN access speed has
increased from original 1 Mb/s to todays 300 Mb/s. And IEEE802.a/b/g/n stan-
dards have all greatly promoted the expansion of WLAN.WLAN is not only a
supplement to the wired network, but also gradually developing towards large-scale
Network Element Portal server

Air
Interface L L LP
STA M M
LT Internet
AP ACU LU
STA LB
LT LA LS
LI AAA server
Network Element
Mobile Network
Fig. 7.6 WLAN network model
Fig. 7.7 Autonomous

WLAN network Internet
architecture
L2 / L3
Equipment
AP1 AP2 AP3
deployment and independent network, and even has replaced the wired network in
some places. Traditional WLAN architecture has been unable to meet the require-
ments of large-scale network. Therefore, the Internet Engineering Task Force
(IETF) established Control And Provision of Wireless Access Points (CAPWAP)
working group, studying the large-scale WLAN solutions. After working on the
current mainstream WLAN solutions, CAPWAP working group divides the WLAN
system into three types: autonomous, centralized and distributed network
architecture.
1. Autonomous architecture
The early WLAN structure is a kind of autonomous system structure. In this
architecture, all 802.11 functions namely the 802.11 PHY Layer and MAC
functions are completed by AP. In addition, some complex functions, such as
the 802.11idefined security function, 802.11-defined QoS function, and even
Radius client functions are implemented by AP. As shown in Fig. 7.7 is the
autonomous WLAN network architecture.
In the autonomous WLAN system structure, fat AP mode is mainly used.

The so-called fat AP means that one access device has most of the network
intelligence, whose function is to control most protocols including roaming,
encryption, management, user authentication, etc. In this WLAN architecture,
the network intelligence is put in the AP equipment, which reduces the workload
of the central switch and makes the deployment more convenient. However, this
system also has the following main defects.
(1) AP must be managed one by one, i.e., the control strategy can only manage
one AP at a time, which increases the management costs, and thus more
management resources are needed.
(2) Since the possible attack and interference cannot be seen in the whole
system, the service refusal of the whole WLAN cannot be alleviated.
(3) Since the system cannot predict and analyze the enterprise activities, it is not
easy to implement real-time load balancing management. Also, users cannot
execute the fast switch based on real-time applications such as voice.
(4) If one AP encounters theft or destruction, security will not be guaranteed.
2. Centralized architecture
In order to solve the problems existing in the autonomous architecture, the
centralized WLAN architecture was put forward, which has gradually become a
hot research topic recently. The CAMP working group has developed a set of
standards based on the centralized WLAN architecture to solve the problems in
large scale WLAN such as network management, security, resource manage-
ment and interoperability. As shown in Fig. 7.8 is the centralized WLAN
architecture.
It can be seen from Fig. 7.8, compared with autonomous architecture, cen-
tralized WLAN architecture increases the factor of AC, which can be regarded as
a set of logical devices to realize the management, monitoring, dynamic config-
uration, AAA and other functions of the whole network. Wireless Termination
Fig. 7.8 Centralized

WLAN architecture Internet
AC
Switch
WTP1 WTP2 WTP3 WTP4

Point (WTP) is different from AP defined in IEEE802.11 standard. AP can

achieve all of the functions of 802.11 while WTP only needs to achieve part
of the functions. So WTP can be seen as a lightweight AP. AC and WTP have
three kinds of connection modes: direct connection, layer 2switching connection
and layer three routing connection. Among them, the layer three routing con-
nection enables WTP to connect AC through IP network, which not only makes
the WTP deployment more flexible, but also solves the wireless STAs seamless
layer three roaming problem. It is a widely adopted connection mode. The
centralized WLAN architecture has the following advantages.
(1) It reduces the required traffic filtering and strategy processing in AP and
realizes the centralized control over traffic filtering and strategy execution.
Any change in traffic will be announced to AP, so as to reduce the traffic
processed in AP and to concentrate the limited resources in WLAN system
on wireless access.
(2) For the whole WLAN system, a centralized management and traffic man-
agement are adopted so that the traffic control, authentication, encryption
and execution strategies are realized (to ensure QoS and security).
(3) Through the network infrastructure or IP routing network, universal encap-
sulation and transferring mechanism is provided for multi-user AP interop-
erability, enabling the enterprise users LWAPP console to better perform
LWAPP AP storage and processing, and making it more convenient for the
system to implement access strategy or service management.
(4) Functions of 802.11 are realized between AP and the centralized device,
giving it good management and scalability. The centralized WLAN archi-
tecture adopts the thin AP, which means most intelligent configuration is
concentrated in the centralized equipment, thus reducing the WLAN execu-
tion management function and the total operating cost.
The above characteristics of the centralized WLAN architecture determine
that it is suitable for large scale deployment.
3. Distributed WLAN architecture
The distributed WLAN architecture is a distributed network composed of
wireless AP which depends on the wired or wireless medium. In this structure,
wireless APs are connected through wireless links, similar to the mesh structure.
Some of these APs are connected to the external network via a wired link and
they can be used as gateway nodes in charge of network management. The
distributed WLAN architecture is shown in Fig. 7.9
The distributed WLAN architecture has the advantages of strong robustness
and easy maintenance It can also draw on the research results on mesh to
optimize the deployment. Distributed WLAN architecture is currently a new
research direction. IEEE 802.11 working group is developing the 802.11 mesh
standard 802.11 s.Researches on its application scenarios, service management,
and execution strategies are still in its infancy.
Fig. 7.9 Distributed

network architecture Internet
L2 / L3
Equipment
AP
WLAN Coverage Design
WLAN Indoor Propagation Model
WLAN indoor coverage is mainly characterized by smaller coverage, and larger

environmental changes. In general, we select the following two models that are
suitable for WLAN analysis. Because the indoor wireless environments vary
greatly, in planning the reference model and the model coefficients should be
chosen according to the actual situation.
1. Devasirvatham model
Devasirvatham model is also known as linear path attenuation model, of
which the formula is shown in (Eq. 7.19).
PLd; f dB PLFS d; f a 7:19
In it, PL(d, f ) is the indoor path loss, PLFS(d, f ) is the free space loss, which is
shown in the expression (Eq. 7.20):
PLFS d; f 32:4 20lgf MHz 20lgdkm 7:20
In it, d is the propagation path, f is the radio frequency, a is the model

coefficient, the value is 0.47.
2. Attenuation factor model
In terms of the spatial propagation loss of the radio wave, the electromagnetic
wave in 2.4G frequency band has the approximate path propagation loss
formula.
Table 7.5 Theoretical value Propagation loss (dB)

of WLAN typical path
Distance (m) n 2.5 n 3.0 n 3.5
propagation loss
10 63.47 66.97 70.46
50 80.95 87.94 94.93
100 88.47 96.97 105.46
Table 7.6 Maximum distance of WLAN theoretical indoor propagation

Indoor maximum transmission distance (m)
Data rate (Mbps) Sensitivity (dBm) Devasirvatham model Attenuation factor model
11 88 53 48.9
5.5 91 60 63.1
2 93 64 72.4
1 96 69 94.8
PLdB 46 10 n lgDm 7:21
In it, D is the propagation path, nis the attenuation factor. In different wireless
environments, attenuation factors nhave different values. In free space, the path
attenuation is proportional to the squared distance, i.e., the attenuation factor is
2. Within the building, the effect of distance on path loss will be greater than that
of free space. Generally speaking, the value of n in the open environment is
2.0 ~ 2.5, the value of n in the semi open environment is 2.5 ~ 3.0, and the value
of n in the closed environment is 3.0 ~ 3.5. The theoretical calculated value of
the typical path propagation loss is shown in Table 7.5.
3. AP signal link loss calculation
According to the model, the indoor path loss is equal to the free space loss and
the additional loss factor. Moreover it grows exponentially with distance. The
receiving level estimation formula is shown in (Eq. 7.22).
PrdB PtdB GtdB PLdB Gr dB 7:22
In it, Pr[dB] is the minimum receiving level, which is the receiving sensitivity
of AP at different transmission rates. Pt[dB] is the maximum transmit power.
Gt[dB] is the transmitting antenna gain. Gr[dB] is the receiving antenna gain.
PL[dB] is the path loss. We can make the theoretical calculation of the limit
propagation distance of AP signal as follows.
Assuming that the antenna transmitting and receiving gain is zero, when AP
transmitting power is 16.2 dBm, the maximum distance of theoretical indoor
propagation is shown in Table 7.6.
4. AP signal penetration loss
The experience values of the current 2.4G electromagnetic waves penetra-
tion loss for a variety of building materials are as follows.
(1) Block by partition walls (brick wall thickness 100300 mm): 2040 dB
(2) Block by floors: above30dB
(3) Block by wooden furniture, doors and other wooden partition: 215 dB
(4) Thick glass (12 mm): 10 dB
In addition, when measuring the penetration loss of AP signals, we need to
consider the incident angle of the AP signal. For a 0.5 m thick wall, when the
linear connection of AP signal and the coverage area forms a 45 degrees
incident angle, its equivalent to a 1 m thick wall. With 2 degrees angle, its
equivalent to a wall of more than 14 m thick. Therefore, in order to get a better
acceptance effect, we should try to make the AP signal through the wall or
ceiling vertically (90-degree angle).
WLAN Indoor Coverage Planning Scheme
When planning WLAN network, we should firstly consider the interaction between
AP and wireless network adapter signal, as well as the users effective access to the
network. Therefore, how to ensure the wireless signal coverage is a factor that must
be considered in AP selection. Since WLAN has high work frequency band, low
sensitivity (compared with mobile base station/mobile phones), great signal reflec-
tion and diffraction, there are several different indoor coverage solutions that the
planning personnel can choose according to the actual situation on site.
On-site investigation must be carried out before designing to find out the
following points.
(1) Getting to know the coverage area and signal coverage quality requirements
since different locations have different coverage requirements.
(2) Investigating the existing signal distribution of the coverage area, to grasp the
blind spots, hot spots and signal collision areas.
(3) Investigating the composition of the buildings in the coverage area, and the
signal blocking.
(4) Signal access position and mode.
(5) Investigating the location where the equipment can be installed.
After the on-site investigation of the environment, we can choose three different
plans according to the actual situation, respectively, co-cellular network indoor
distribution system coverage scheme, independent AP distribution coverage
scheme and cross patch coverage scheme. Here are brief introductions to the
three schemes.
1. Co-cellular network indoor distribution system coverage scheme
(1) Application scope and usage requirements of the scheme
Application scope: mid-to-large scale indoor coverage. The system struc-
ture is complex. Mainly used in medium blind area coverage or important
public places to meet the coverage requirements of places such as hotels,
airports, conference centers, etc., but not suitable for the network with high
capacity requirements.
Usage requirements: the system is an indoor coverage system, and the
equipment is required to be installed indoors.
(2) Engineering design experience
Generally, if the antenna radiation power is 10 dBmW and a 2dBi small
omnidirectional indoor antenna is used, then within 30 meters from the
antenna, when in spacious conference center and without brick wall, the
coverage level can reach -75 dBm/WLAN.
For the hotel room with dense wall structure, if the antenna tip radiation
power is 10 dBmW, and a 2dBi small omnidirectional indoor antenna is
used, then within 8 to 10 meters from the antenna, the room signal can reach
70~85 dBm.In general, the signal to the room door should be controlled
within 4 m.
In the energy distribution, signals should be distributed as evenly as
possible. In design, in the paths from the base station to each antenna, the
emergence of more than two power dividers (or coupler) should be avoided,
so as to ensure the effective access of uplink signals.
The use of this scheme is suitable for indoor coverage of large scale hotel
lobby, airport, and conference centers with lower capacity requirements.
2. Independent AP distribution coverage scheme
Application scope: Its suitable for small and medium scale indoor
coverage where there is no indoor distribution system, small area coverage
or important public places, such as hotels, conference centers, etc. The
system has simple structure and covered by independent APs.
Usage requirements: indoor coverage, applicable to the design and instal-
lation of indoor equipment.
If only one AP is installed in a hall, then the AP would be better placed in
a central position in the hall. And it is better to be placed on the ceiling of the
hall. If two APs are installed within the same space, they can be put on two
diagonals.
The number of signals penetrating walls and ceilings should be kept
minimal. 2.4G signal can penetrate walls and ceilings, but each wall and
ceiling will reduce the AP signal coverage by 130 meters. AP and the
computer should be placed in a suitable position so that the signal block path
of the wall and ceiling can be kept the shortest and the loss minimum.
Linear connection between the AP and the coverage area should be
considered. The location of AP should be chosen in the way that the signal
can go through walls or ceilings vertically (90-degree angle).
Different building materials have different transmission effects. Building
composed of metal frame or door will shorten the WLAN signal transmis-
sion distance. Therefore, AP should be placed where signals go through dry
walls or open doors, instead of where signals must penetrate the metal
material.
AP antenna direction is adjustable, so AP should be installed where the
antenna main beam directly faces the coverage target area to ensure good
coverage effect.
AP should stay far away from electronic equipment (1~2 meters), such as
microwave ovens, monitors, motors, etc.
3. Cross patch coverage scheme
Scope of application: for large-scale indoor coverage, the system struc-
tural design is relatively complex and the renovation project is smaller.
WLAN signal source connected to the indoor distribution system,
complemented by independent separate AP layout covering a few blind
spots, and ultimately seamless coverage with good effect can be achieved.
It can meet the coverage needs of public places and open areas like halls and
airports, and can also cover the medium capacity needs of hotels and
conference centers.
Usage requirements: indoor coverage, applicable to the design and instal-
lation of indoor equipment.
AP signal source access is similar to co-cellular network distributed
antenna system coverage scheme in terms of distribution system, design
requirements and hardware transformation requirements. It needs to be
noted that when the original indoor systems antenna location does not
meet the WLAN coverage/range requirements, we may not consider moving
the antenna or adding terminal antennas by changing the power divider. But
rather, we can use independent AP planning method to cover a small amount
of hot spots for coverage complement.
In cross distribution coverage planning scheme, when using the original
indoor distributed system of WLAN coverage, for the problems of massive
renovation project, more antenna positions changing, and difficult power
divider load matching, we can moderately use independent APs to cover
blind spots and hot spots, as supplement to the original indoor distribution
system, realizing good overlapping coverage effect. This scheme has the
characteristics of flexible planning. Before design, we should make
pre-evaluation on the renovation project scale and the expected coverage
effect, and then after comprehensive consideration, we develop an ideal
project plan.
WLAN Capacity Design
In WLAN network planning, the impact of the whole mesh capacity on the system
performance is greater than that of the AP coverage. To avoid the situation that a
single user access slows down the whole cells transmission rate, an AP transmit-
ting power threshold should be set up. By adjusting the AP transmission power to
alter the cell size, we can make the cell smaller in design. In such a way, since there
arent many users in a cell small, each users high transmission rate can be
guaranteed.
Network Capacity Calculation
Capacity in communications can be studied from two aspects: theory and practice.
Theological concept of capacity is based on the amount of information on unit
bandwidth (time or area). For voice services: Erl/unit bandwidth/unit area; for data
services: bit/unit bandwidth (time or area). The above theological concepts of
capacity are difficult to be applied in the actual measurement and comparison of
information. Practical concept of capacity is based on the amount of charges
(or users) on unit bandwidth (time or area). For voice services: Erl/unit bandwidth
or Erl/unit bandwidth/unit area; for data services: Channel/unit bandwidth. This
practical capacity concept is applicable to engineering communications capacity
measurement.
Network capacity, without a standard definition in industry, can be characterized
by different parameters from different angles, and the results are also different.
From the users point of view, by using mathematical basic theories to study the
network capacity, we can get the following definitions.
Definition 1: Network capacity C is defined as the sum of the communications
probabilities of all users, and its expression is as shown in (Eq. 7.23).
X
C Qi 7:23
i
where Qi is the communication probability of the user i and means the average
number of users communicating simultaneously at any given moment supported by
the network.
Definition 2: Communications probability Qi is the probability that the user can
normally send the data when the expected user does not transmit data. Assume that
user is communications status Zi is subject to two points, that is, Zi (0, 1), in
which Zi 0 means the user is not communicating and Zi 1 means that the user is
in a state of communications. Assume user is interference user number is Ni, and
then the Zi distribution rate is:
8
> Ni 1
>
< N , Zi 0
i
PZ i 7:24
>
> 1
: , Zi 1
Ni
When Zi 1, i.e., when the user is in a state of communications, the network

capacity is:
XM
1
C 7:25
N
i1 i
Network capacity needs to be calculated for actual network. According to the

definition of network capacity, as shown in (Eq. 7.23), to calculate the WLAN
network capacity, we need to know the number of AP, AP coverage results etc., and
also the distribution of users in the network. In reality, with constantly user leaving,
coming and changing position, the actual network is also always in constant change.
Therefore, the calculation of network capacity is only for the capacity of an actual
network in a given state at a given time. It does not represent the evaluation results
of the network capacity for a period of time.
Let A {1, . . . , n} denote all APs and APj denote the jth AP, then A [ APj .
j
Let B {1, . . . , m} represent all users in the coverage and Bi denote the ith user,
then B [ Bi .Dj is the set of the user covered by the jth AP, then Dj B, and
i
[ Dj B. In order to facilitate the calculation of network capacity, for A and B, we
j
introduce a coverage matrix G. If the operator ! is defined as coverage, then
G A ! B. This matrix represents the coverage result of AP set A to user set B. Its
mathematical expression is
n o
G gij ji 1; . . . ; n; j 1; . . . ; m 7:26
In it,
(
1, Ai 2 A; Bj 2 Di
gij 7:27
0, else
Based on the coverage matrix G, the network capacitys calculation steps are as
follows.
(1) Initialize parameters. Let i 0 , j 0, and all elements in Array flag[n] be 0.
(2) If gij 1, then Array Element flag[i] gets the value 1.
(3) If i < n, then i i + 1, jump to (Eq. 7.27). Or enter the next step.
(4) k 1 , Ni 0.
Pm
(5) If flag[k] 1, then N j N j gkj 1.
j1
(6) If k < m, then k k + 1, repeat Step (5). Or enter the next step.
(7) Nj Nj + 1; Qj N1j .
(8) If j < m, then j j + 1, return to Step (2).Otherwise, jump out of the program.
After the above process, we get each users communications probability vector
Q {Qi| i 1, . . . , m}, and then get the capacity value by the calculation of net-
work capacity (Eq. 7.23).
Discussion on Optimal Network Capacity Based on Computational Results
According to the definition of network capacity in (Eq. 7.23), the conditions for the
optimal network capacity are
!
X X
MAXC MAX Qi MAXQi 7:28
i i
Put Qi N1i in (Eq. 7.28) and get
X
1
MAXQi MAX MINN i 7:29
i
Ni
The above model shows that user is maximum communications probability is

equivalent to the number of user is minimum interference users Ni. Since Ni 1,
MAX(Qi) MIN(Ni) 1. Assume the number of users is n, and then
X
n
MAXC MAXQi n 7:30
i1
(Eq. 7.30) shows that in the network when the number of users is n, deploy n APs,
and each AP covers and only covers one user. When all users are covered, the
network has the largest capacity, and the maximum capacity value is n.
The Guiding Significance of Network Capacity in Actual Network Planning
The goal of the actual network construction is that in the premise of ensuring the
user network throughput rate, using APs as less as possible to obtain the users
coverage as much as possible, and meanwhile pursuing good balanced coverage
results. The mathematical significance of AP is
MAXC [ MINAP [ MAXUser [ MAXQi 7:31
Obviously, the network capacity is proportional to the number of AP. After

considering the cost of network construction and operation and the actual use of the
environment, we cannot simply increase the amount of AP to improve network
7.3 Typical Case Analysis of Field Trials 403
capacity. That is to say, the network deployment plan in the maximum capacity
calculated in (Eq. 7.30) has no practical application value. Therefore, it is necessary
to comprehensively consider different factors and find a relatively optimal network
deployment scheme with a good overall cost and effectiveness.
7.2.5 Large-Scale Test User Application Scenarios
Large scale user application test is the last step to carry out the network perfor-
mance evaluation of the communications system. The traditional method is to carry
out the relevant work by issuing the test user number in the pre-commercial stage.
For 5G systems, since technologies of various modes will be interconnected, it is
absolutely necessary to carry out the relevant tests and verification at the R&D
stage, which requires the test environment to provide the appropriate test
conditions.
ShanghaiTech University has 6000 students. Together with faculty and staff, the
overall size is more than 10 thousand people, which is very suitable for a dense
crowd massively connected scenario test. In particular, the characteristics of stu-
dents needs for new services make this test field very suitable for the new service
applications test.
In the preliminary experiment, it is clear that students characteristics for
network use make them very suitable for large-scale user application test. Different
from traditional voice services, students curiosity about new types of services
makes them more willing to cooperate with the application test of new services.
In addition, the tidal effect of the student population movement is also very suitable
for carrying out load balancing technology test.
7.3 Typical Case Analysis of Field Trials
Wireless communication field trials involve all aspects of the wireless communi-
cation system, covering from radio wave propagation characteristics research to the
test of air interface key technology, to network performance evaluation and vali-
dation. Every link needs field trial to confirm whether the technology meets the
design requirements.
A typical field trial procedure is shown in Fig. 7.10.
This chapter gives typical test cases from three aspects, namely wireless channel
measurement and modeling, wireless communications key technology test and
performance test of wireless network.
START
Requirement Analysis
N
Field is meet ? Reconstruction
Parameter
configuration
Data collection
Data analysis
N
Modify &
Results conform?
Retest
END
Fig. 7.10 Outfield trial flow chart

7.3.1 Case One: Research on the Irregularity of Wireless

Cellular Networks
Test Purpose
In order to ensure seamless coverage of wireless network, we need to consider the

coverage shape of wireless cell. Traditionally we consider that the hexagonal
boundary shape is close to a circle, which is in a honeycomb shape, to achieve
the continuous coverage. The research on cellular wireless coverage shape has been
going on for decades. For the free space propagation environment, with a single
transmitters signals radiating evenly in all directions, the coverage of a wireless
cell in two-dimensional plane should be a circle centering around the transmitter. It
is assumed that the service area of the wireless cellular network can be divided into
a seamless coverage service area of a plurality of normal triangles, squares, or
regular hexagon, when the plurality of transmitters are deployed in an equal
distance in accordance with the regulations. Taking into account the regular
hexagon which is the most close to the shape of the circular shape, it is widely
used as a conventional wireless cellular network coverage shape. However, in the
actual network deployment process, the increase of the base station makes the cell
divide continuously. Irregular terrain landforms make wireless cell coverage shape
can really achieve the effect of positive hexagon, but there are irregular boundary
characteristics.
An important challenge for wireless cellular network operators is to ensure the
mobile users seamlessly from adjacent cells to achieve continuous coverage,
especially those located in the residential area of the boundary of the user. In
addition, the switching signal of the user in the adjacent cell is decided by the
cell boundary in the wireless cellular network. Therefore, the cell boundary shape
of wireless cellular network is a key indicator to design, deploy and optimize the
wireless cellular network. How to evaluate the boundary features of a cell is a
problem worthy of studying.
Test Environment
In order to achieve the purpose of studying the irregularity characteristics of cell

boundary shape, we need to build the test environment in typical scenarios so as to
avoid the impact of the adjacent base stations signals on the test results. Of course,
the position of the transmitter should be effective to evaluate the theory analysis
results. Therefore, we have built temporary test environments in several locations in
Shanghai, carried out a large number of field trials, obtained the real test data, and
analyzed the test results. Figure 7.11 shows the transmitter location and test
roadmap of a dense city scenario, in which the test radius reaches 1.5 km.
Fig. 7.11 Shanghai Xuhui intensive city scenario test roadmap
Test Results
Through the field data collection in three test scenarios and the analysis on the
statistical fractal characteristic for the stochastic process with three methods, we
find that although there are slight differences between three different typical
methods in Hurst parameters estimation, the final estimations for the wireless
cellular coverage boundaries are basically the same. That is, the estimated average
Hurst parameters are approximately 0.9, which shows the real wireless cellular
coverage boundary exhibits a statistical fractal feature.
7.3.2 Case Two: Characteristics Analysis on the Radio Wave

Propagation Between Multiple Frequency Bands
Test Purpose
Spectrum, as a valuable and non-renewable resource, is the fundamental of the

survival of mobile communications technology, and the basic resource of the
wireless communications system. Mobile spectrum resources are currently
managed by the nation, and its management mode is the fixed spectrum manage-
ment mode. World Radio Conferences in 1992, 2000, and 2007 has allocated many
frequency bands in succession for the wireless mobile communications system,
including 450 MHz, 700 MHz, 900 MHz, 1800 MHz, 2GHz, 2.3GHz, 2.6GHz,
3.5GHz, etc., which are the main frequency bands that support the current 2G, 3G,
4G mobile communications systems of GSM, TD-SCDMA, CDMA2000,
WCDMA, LTE, etc.
Wireless channel measurement, based on the feature analysis of radio wave
propagation, analyzes the large scale characteristics of radio wave propagation as
its basic content, that is, the path loss.
Most of the current wireless networks design has multi-mode, shared network
and shared antenna. Usually there is current network data of a certain network
frequency band. As for how to infer from the coverage characteristics of one
network to another networks frequency band propagation coverage, we lack the
experiential data of objective testing. So we cant guide the front line to use the
current network data in other networks with shared network. This requires the
multi-band radio wave propagation characteristic analysis. Through the actual test
and the test data analysis, we can find out the rules for the differences in propagation
characteristics between different frequency bands, so that we can maximize the
utilization of the current network data, support the use of the coverage data from
different networks in the coverage prediction of other networks, and improve the
coverage prediction accuracy.
Test Environment
To analyze the propagation characteristics of different frequency bands in the

scenarios with the same characteristics and in different scenarios, and to analyze
the influence of different data processing methods on the propagation characteristic
analysis, we have selected several test environments in Shanghai, including dense
urban scenario, ordinary urban scenario and suburban scenario. And for each
typical scenario, we have chosen 2~3 test areas.
Figure 7.12 shows the transmitter location and test roadmap of the suburban
scenario in Zhangjiang, Shanghai.
Test Results
For this test, 11 typical frequency bands that are commonly used in wireless cellular
network within 700 MHz~3500 MHz range are selected for all the test scenarios.
Through data and processing, we have obtained the comparison results of path
loss indicator and shadow fading factor between same scenarios different frequency
and different scenarios same frequency. In data processing, we have studied the
geographical average method of the test sample points in the mobility test, and
Fig. 7.12 Test roadmap of the suburban scenario in Zhangjiang, Shanghai
compared the impact of the different processing methods on the results. The
following are part of the test conclusions.
(1) Comparing the test results of the suburban, dense urban and ordinary urban
environments, we find that the dense urban environment, with the most obsta-
cles, has greater path loss attenuation value than the other two communications
environments.
(2) Comparing the standard deviations of the test results and the theoretical
propagation model, we can see that the standard deviation decreases with
the increase of the frequency point, and the standard deviation is less
than 9 dB, which meets the requirement of the network test to the standard
deviation.
(3) Comparing three data processing methods of 10 meters, 10 meters grid and
17 meters grid, we find that the 17 meters grid can simultaneous meet the data
processing requirements from low frequency band to high frequency band, and
the smaller standard deviation can be obtained.
7.3.3 Case Three: CoMP Key Technologies Test
Test purpose
Coordinated Multiple Points (CoMP) Transmission/Reception refers to the tech-

nique that the geographically separated multiple transmission points coordinate to
transmit data for one terminal (PDSCH) transmission or jointly receive the data
from one terminal (PUSCH).
According to the relations between coordinated nodes, CoMP can be divided
into two types, namely Intra-site CoMP and Inter-site CoMP. The coordination of
IntraiteCoMP occurs within one site, so there is no limit for backhaul feedback
capacity and massive data interaction can take place in one site. Inter-site CoMP
occurs between multiple sites, so there are more stringent requirements for
backhaul capacity and latency. That is to say, the performance of Inter-site CoMP
is restricted by the current backhaul capacity and latency tolerance. According to
whether the data can be obtained at multiple coordination points, it can also be
divided into Joint Processing (JP) and Coordinated Scheduling/BeamForming
(CS/CBF).
The systematic diagram of CoMP key technologies is shown in Fig. 7.13.
The goal of CoMP key technology verification is to carry out the system
performance tests of Inter-site CoMP and Intra-site CoMP operating modes in the
real indoor and outdoor scenarios, and to evaluate the performance gain with the
conventional single station transmission.
Inter-site
Intra-site CoMP
CoMP
Fig. 7.13 Schematic diagram of CoMP key technologies

Fig. 7.14 Schematic diagram of CoMP key technologies verification test environment
Test Environment
For CoMP key technologies test environment, we build a multi-layer LTE wireless
coverage network from outdoor to indoor according to the test requirements, which
is shown in Fig. 7.14.
As shown in the above figure, macro-0 and macro-1 are two sectors (cells)
of the same LTE outdoor macro cell base station (eNodeB). IA-2 is a sector of
the LTE outdoor micro cellular base station. In the teaching building we also
constructed more than 10 LTE indoor coverage stations, thus forming a macro
cell -- micro cellular --- indoor multi-layer coverage of wireless network test
environment.
In the test configuration, we can choose an appropriate position to achieve
different CoMP configuration.
(1) The outdoor CoMP Inter-site test environment composed of macro-0 and
macro-1;
(2) The Intra-site CoMP test environment composed of IA-2 and indoor sites;
(3) The Intra-site CoMP test environment composed of multiple indoor sites.
Test Results
This test verified the effect of CoMP in the system capacity performance improve-
ment in multi-layer network environment. More detailed test results will not be
presented here.
7.3.4 Case Four: Active Antenna System (AAS) Key

Technologies Test
Test Purpose
AAS solution is to integrate the base station RF into the antenna and to use the
multi-channel RF and antenna array in coordination, so as to realize the space
beamforming and complete the RF signal transceiving. Active antenna is a new
form of base station architecture while BBU similarly brings baseband signals to
the active antenna unit. Different from BBU+RRU architecture, active antenna
divides the transceiving channels into the sub antenna oscillator level with more
refined particles size. Through different configurations of active antenna oscillators,
we can achieve the functions like flexible beam control and MIMO, and more
flexible and dynamic resource allocation and sharing, so as to achieve the goal of
the optimal performance and lower cost in the whole network.
AAS technology uses the adjustable angle design and thus improves the network
performance. It mainly includes the following several key technologies.
(1) Vertical sector splitting
AAS system can form multiple beams in horizontal and vertical dimensions,
achieving the multiplexing of the resource with the same time frequency. For
the AAS vertical sector splitting performance test, we need to evaluate the split
sector performance on the same horizontal plane.
(2) 3D beamforming
Through the 3D beamforming, AAS can achieve very good spatial resolu-
tion, making MU-MIMO and spatial interference suppression field have a
certain performance improvement.
(3) Proactive cell shaping
By flexibly adjusting the lower angle and the shape of the beam, AAS can be
carried out in advance, in order to promote the performance balance in the
macro cell and the low power node, especially in the non-equilibrium network
configuration scenario.
Test Environment
The test field for AAS key technologies verification is composed of four stations
(one center station, three interference stations), among which the center station has
three sectors. This network topology consists of six sectors (1 3 + 3 1). The
topological graph is shown in Fig. 7.15.
To facilitate the test of the network configuration, the core network EPC adopts
the simulated core network and builds the core network elements with servers,
including MME, SGW, and PGW.
Site 3
Site 2
S
AA
AA
S
SimEPC BBU2_1Cell BBU3_1Cell
Optic
AA
S
Switch
S
BBU1_3Cell
AA
ODF
AAS
Site 1
AA
S
BBU4_1Cell
Site 4
Fig. 7.15 AAS key technologies test network topology
Test Results
For AAS key technologies verification field trial, we compare the network perfor-
mance in three modes, namely conventional RRU, vertical sector splitting and
virtual sector. The test results are summarized as follows.
(1) Conventional RRU mode, under the same test conditions in comparison test,
will change with the inclination angle. And RSRP, SINR, DownLink (DL),
UpLink (UL) and other data change regularly. In the planed road test area, there
will be the optimal inclination angle.
(2) For vertical sector splitting mode, under the same test conditions in comparison
test, with combination of different inclination angles, representative areas (far
end and near end) are selected in the planed road test area. And RSRP, SINR,
DL, UL and other data change regularly. There will be the optimal inclination
angle combination.
At the near end, the downlink test indicators will be worse than other modes,
mainly because of the same frequency interference by the side lobe of the outer
cell. At the far end, the downlink test indicators show similar performance with
the other modes.
(3) For virtual sector mode, under the same test conditions in comparison test, with
combination of different inclination angles, representative areas (far end and
near end) are selected in the planed road test area. And RSRP, SINR, DL, UL
and other data change regularly. There will be the optimal inclination angle
combination.
At the near end, the downlink test indicator is slightly worse than the
conventional RRU model but significantly better than vertical sector splitting.
The main reason is that avoidance scheduling is done in the base band side
according to the downlink interference detection, which significantly lowers the
near-end interference. At the far end, the downlink test indicators show similar
performance with the other modes.
(4) As for the uplink tests of the three modes, the conventional RRU mode is better
than the other two modes, while the virtual sector and vertical sector splitting
are similar. The main reason is that the same frequency interference between
UEs uplinks is at a very low level, the impact on performance is very small.
(5) In terms of network coverage evenness, vertical multi sector and virtual sector
are better than conventional RRU mode, with the coverage performance
improved.
7.3.5 Case Five: Inter-Cell Interference Coordination (ICIC)

Key Technologies Test
Test purpose
Inter-Cell Interference (ICI) is an inherent problem of the cellular mobile commu-

nications system which severely affects the system performance. The reason is that
the users of the same frequency resource in each cell can interfere with each other.
Severe inter-cell interference will directly affect the system coverage and capacity.
Particularly for cell edge users, it will directly cause their frequent connection lost
or disability to access the network.
According to the existing research results and the research projects of standards
organizations such as 3GPP LTE, IEEE 802.16 m, 3GPP2 Ultra Mobile Broadband
(UMB) for inter-cell interference control processing modes, there are three types of
inter-cell interference control technologies.
(1) Interference randomization technique
Interference randomization is a common technique to reduce the inter-cell
interference by the interference whitening of adjacent cells. Its main advantage
is that it does not affect the complexity of the receiving end scheduling and
receiving process. But when the system is fully loaded, the interference ran-
domization technology is limited in improving the system performance. The
representative technologies of interference randomization are base station-
based scrambling codes and various frequency hopping technologies.
(2) Interference coordination technology
Interference coordination technology mainly means to effectively coordi-
nate the channel resources and power of space, time and frequency domain of a
number of cells, and to avoid or reduce the interference to adjacent cells
through limiting the use of some resources in this cell (such as frequency,
power and time). The main technology includes Fractional Frequency Reuse
(FFR), multi station MIMO and power control technology, etc.
(3) Interference cancellation technology
The interference cancellation technology is to decode and copy the signals
from the interfering cells, and then subtract the interference signal from the cell
in the received signals. The advantage of interference cancellation technology
is that there is no limit to the use of cell frequency resources. But its limitation
is that the target cell must also know the pilot frequency structure of interfer-
ence cell to make channel estimation of the interference sources. As a result, the
signaling cost and implementation complexity of the interference cancellation
technology are relatively high.
Test Environment
The test field of ICIC key technology test is composed of two stations (one cell in
one station, and the two cells have overlapping areas).The topological graph is
shown in Fig. 7.16.
The test will take the parameters like the target cell RSRP, adjacent cell RSRP,
Cell Reference Signal (CRS) SINR, CQI, etc., which are obtained based on the
location of the test terminals, as the input variables of the cell interference coordi-
nation algorithm. Then it will calculate the optimized power parameters, configure
the base station transmitting parameters according to the optimized parameters, and
retest the validity of UE network performance verification algorithm.
RB1
RB2
Cell_1
UE11 UE12
UE32
UE21
UE22
Cell_3 Cell_2
UE31
Fig. 7.16 Schematic diagram of UICIC key technologies validation test environment
The test selects two scenarios for two tests. One is that the terminal is located
near the cell center, and the other is that the terminal is located in the overlapping
area of the cell.
Test Results
(1) Test results of the near-station position

Both the two users are close to the base station. One users service cell is
Cell 1, while the other users service cell is Cell 2. Table 7.7 and Table 7.8 have
shown the two tests results.
(2) Test results of the far-away-from-station position
Both the two users are far away from the base station and located in the
overlapping area of the two cells. One users service cell is Cell 1, while the
other users service cell is Cell 2. Table 7.9 and Table 7.10 have shown the two
tests results.
Table 7.7 The first test result when users are near the base station
Cell throughput Total Power
(Mbps) throughput Proportional efficiency
Power configuration (W) Cell 1 Cell 2 (Mbps) fairness and value (Mbps/W)
P117.31 20.06 27.34 47.4 6.307 EP11.159
P217.31 EP21.579
P16.76 19.33 24.84 44.17 6.174 EP13.222
P26.00 EP23.675
Table 7.8 The second test result when users are near the base station
Cell throughput (Mbps) Total Power
Power throughput Proportional efficiency
configuration (W) Cell 1 Cell 2 (Mbps) fairness and value (Mbps/W)
P117.31 27.986 12.293 40.279 5.841 EP11.617
P217.31 EP20.71
P14.38 15.916 17.232 33.148 5.614 EP13.634
P25.51 EP23.127
Table 7.9 The first test result when users are far away from the base station
P117.31 9.882 6.668 16.55 4.188 EP10.571
P217.31 EP20.385
P14.38 10.509 7.556 18.065 4.375 EP12.399
P25.51 EP21.371
Table 7.10 The second test result when users are far away from the base station
P117.31 11.606 4.708 16.314 4.000 EP10.670
P217.31 EP20.272
P14.38 11.685 7.309 18.994 4.447 EP12.668
P25.51 EP21.326
(3) Test result analysis

According to the test results analysis of the above data, when the user is in the
cell center, the base stations power optimization ensures the network throughput
and proportional fairness summation value. Its power is reduced by 70%, greatly
improving the resource utilization efficiency. When the users is at the cell edge
location, after the base stations power optimization, not only the network
throughput and proportional fairness summation value increase slightly (network
throughput by 9%, proportional fairness summation value by 4.5%), but also the
base station power is reduced by 70%, greatly improving the efficiency of
resource use. Meanwhile, it can be seen that the edge user scenario power
optimization algorithm has more effectively improved the resources utilization.
The field trial results well verified the algorithm performance.
7.3.6 Case Six: Public WLAN Network Coverage

Performance Test
Test Purpose
Because WLAN system uses the unlicensed Industrial Scientific Medical (ISM)
frequency band and most of the terminals support the WLAN function, WLAN
network is widely used and brings users a good wireless connection experience.
However, due to the publics insufficient awareness for the unlicensed ISM band
using provisions and the irrational use of network overall plan and channel, massive
co-channel interference and adjacent channel interference occur, eventually making
the actual user experience cannot reach the expected performance. As for how to
evaluate the WLAN network coverage performance, a joint test of network trans-
mission performance and air interface signal quality is required to avoid the
incomplete evaluation for network performance by a single test.
Test Environment
This test chose the heavy-traffic public places in Shanghai, including the railway
station and the airport area. Figure 7.17 shows the test roadmap schematic diagram
Fig. 7.17 Schematic diagram of WLAN test environment of Hongqiao Railway Station
Fig. 7.18 Schematic diagram of WLAN test environment of Pudong Airport
of Hongqiao Railway Station area. Figure 7.18 shows the test roadmap schematic
diagram of Pudong Airport area.
Test Results
For WLAN network, its KPI indicators system includes the following four categories.
(1) Signal optimization indicators
(2) AP performance optimization indicators
(3) AC performance optimization indicators

(4) Data side optimization indicators
For the network coverage, the signal optimization indicator is the most important
indicator system, which specifically, includes the following categories.
(1) Signal coverage ratio
In target coverage area, the proportion of the test points whose signal value
>75 dBm.Target value 95%.
(2) Signal coverage range
Control the AP signal coverage to avoid the too large signal leakage range
threatening user data security. Target value: within 50 m.
(3) Signal strength
Signal strength refers to the signal value in the target coverage area. A signal
strength that is greater than -75 dBm is the basic requirements for WLAN
services. If the signal strength is too weak, it will affect the service access and
quality of service. If there should be a signal, we should add points to patch the
coverage or adjust the AP transmitting power, so as to meet the requirement for
signal strength. Target value: greater than -75 dBm (key areas greater than
-70 dBm).
(4) Leakage signal strength
Leakage signal strength refers to the co-channel leakage signal intensity
within the coverage area. Target value: less than -75 dBm.
(5) Signal to Noise Ratio (SNR)
It refers to the Signal to Noise Ratio (S/N value) of the downlink signal
received by the wireless network adapter of the user terminal in the 95% or
more areas of the target coverage area. SNR that is greater than 20 dB is the
basic requirements for WLAN services. Poor SNR will affect the service access
and quality of service. If the SNR is too low, we should inspect the on-site
wireless environment and re-plan the AP transmitting power and channel.
Target value: greater than 20 dB.
(6) Air interface low-rate proportion
It refers to the proportion of 1 M, 2 M low-rate air interface rate when users
transmit data. Target value: 30%.
(7) Air interface error message proportion
It refers to the proportion of channel CRC error message. Target value:
40%.
(8) Air interface packet loss rate
It refers to the average data packet loss ratio of wireless air interface. Target
value: 3%.
The test results of Hongqiao Railway Station and Pudong Airport show that
since WLAN network operators do not make unified planning for channels and
some private APs use channels irregularly, the actual public WLAN coverage
performance is far below the user expectation. Reasonable planning and use of
the public WLAN network is very essential.
7.4 Wireless Network Data Intelligence Analysis 419
7.4 Wireless Network Data Intelligence Analysis
7.4.1 Background and Necessity
With the gradual opening up of China telecommunications market, mobile com-

munications has become the main competition focus. In the context of the financial
crisis, mobile communications has become a rare investment highlight. In the new
competitive environment, the competition mode of the telecommunications indus-
try has changed from quantity to quality competition. The former is often
manifested as hard resource expansion like base station expansion, telephone
numbers increase, while the latter often seeks to enhance the network quality
and user service satisfaction through increasing investment in software.
An efficient and stable wireless network is composed of wireless devices
(wireless hardware products), network performance support system (wireless soft-
ware products) and operation and maintenance personnel (people) together, all of
which are indispensable. Among them, the wireless software products are playing
the key value of coordinating and connecting human and hardware. Restricted
by network architecture, monitoring scope, control cycle, processing speed and
other factors, the automatic balance capability that the wireless device itself can
provide is limited. In face of complicatedly entangled technical problems, without
the assistance of professional software, even the maintenance personnel with rich
experience can hardly guarantee the efficient and reliable operation of the wireless
networks.
After entering 3G and 4G era, mobile operators will seek more advanced and
intelligent software support tools. Due to the introduction of a large number of new
technologies, new mobile phone services are springing up. The complexity of a new
generation of wireless network will be far beyond the traditional 2G network, and
the system theory method is needed to update the network optimization technol-
ogy system. For example, in 3G network Radio Network Controller (RNC) devices,
the operating parameters of different types that impact each other has reached
several thousands, which has been more than ten times of the similar Base Station
Controller (BSC) devices in 2G network. Even in the telecommunications equip-
ment developers, it is difficult to have enough technical staff that can fully grasp so
much complex technical information. Frequent and fast network upgrade is pushing
high the telecom enterprises cost in learning. Telecom operators at present are in
extreme lack of the operation and maintenance personnel with knowledge of the
characteristics of the new-generation of network technology. The traditional model
that mainly relies on personnel experience for network maintenance will be very
difficult to adapt to the new situation requirements. Instead, they must rely on more
professional and intelligent software solutions.
However, at the present stage, the software support is far behind the develop-
ment of telecommunications equipment. In the process of wireless system devel-
opment, telecom equipment manufacturers product development focus is the most
important part of the network construction investment. Wireless performance
support system software is limited to verification conditions and other factors, its
accumulated technology and mature products are generally later than the develop-
ment of hardware devices. At this stage, in order to respond to the external policies
and market situation, the immature 3G, 4G and other new mobile communications
equipment has entered the large-scale network construction phase, but its
supporting software technologies have failed to develop synchronously.
In fact, this imbalance has led to the low efficiency of the current network
construction and operation. For example, the mobile phone user capacity and
mobile Internet access rate are much lower than the network design capabilities.
The signal blind area improvement is inefficient. The imbalanced development is
also suggesting the important future direction of the communications technology
market.
In recent years, both inside and outside of the telecommunications industry have
been actively exploring the issue of self-determined choosing or even dynamically
choosing the wireless network by mobile phone users. If the service model develops
well, it will have very significant impact on the competition mode of the existing
telecommunications industry. Telecommunication operators are no longer able to
lock the users by controlling the phone numbers or phone types, and the users will
be able to choose the right wireless network according to the quality of the network.
This new feature will force the operators to pay more attention to the network
quality assurance means to consolidate and develop the users.
Key Technologies
When the mobile communications network fully enter 3G and 4G era, network
characteristics will undergo great changes. Its wireless performance optimization
must also transform from 2G eras troubleshooting mode into equilibrium
mode, and the analytical method of complex giant system in the system theory
must be used to reconstruct the network optimization technology system.
The new generation wireless network intelligent analysis needs to provide the
following key technological means.
(1) The technological means to make an in-depth automatic comprehensive
analysis on the wireless system data
(2) The technological means to translate the wireless system data into the man-
agement information needed for the operational decision-making
(3) The intelligent correlation analysis method that can meet the requirements of
system equilibrium optimization
(4) The method to carry out overall monitoring and integrated optimization for
service performance and network indicators in all network protocol layers
(5) The dynamic planning and configuration management method that can adapt
to the frequent adjustment characteristics of the actual operating networks
(6) The dynamic network management method that can do real-time network
performance monitoring and system adjustment
(7) Collaborative performance analysis method covering both network side and
terminal side
(8) Application and verification method for the mid-to-long cycle radio resources
management (RRM) algorithm in the system implementation
(9) The network performance monitoring method that meet the characteristics of
massive data and fast evaluation in the balanced network
(10) The systematic analysis and management method that fully cover all kinds of
service performance and network indicators
(11) Data sharing and analysis method that links configuration, monitoring, plan-
ning, drive test, and customer service.
Meanwhile, the specific requirements in network operation and maintenance
need to be met both in breadth and depth. Its wireless analysis capabilities must
have the 6 alls features:
(1) All targets
Be able to meet both the targets in professionalism and management of
wireless network operation and maintenance.
(2) All networks
Be able to analyze the whole network, including special network element
equipment, such as repeater, home base station.
(3) All indicators
Be able to comprehensively analyze various types of service quality and
network performance indicators, and even the equipment reliability.
(4) All links
Be able to make balanced coordination analysis of uplink and downlink
wireless links.
(5) All time
Be able to make network performance analysis of various temporal granu-
larity, including second, minute, day, week, month, etc.
(6) All segments
Be able to make unified analysis on the data collected in configuration,
monitoring, planning, drive test, customer service, etc.
Technical Roadmap
Guided by the methods of system theory, combined with the advanced Busi-
ness Intelligence (BI) software technologies, realize wireless professional
in-depth analysis and verification means, and provide fast performance guar-
antee capability for the mobile communications network. Its important key
technical points are as follows.
(1) Wireless network information quantitative analysis and evaluation model
Various kinds of parameters, key events, and analytical algorithms are
summarized into the tree structure. Information analysis model, independent
from software platform, can realize dynamic upgrade and loading. In the model,
all kinds of data from the network side and the terminal side have unified
definitions. Quantitative data association attributes are provided. The profes-
sional wireless performance analysis framework is established. It is compatible
with various types of wireless networks.
(2) Multi-dimensional cube database
Multi-dimensional cube data model is designed to facilitate correlation
search and association analysis, which provides a logical basis for the informa-
tion analysis in the dimensions of physics, space, object, time, etc.
(3) OnLine Analysis & Process (OLAP)
The intelligent analysis on network performance is realized based on corre-
lation search technology. The advanced parallel processing technology is used
to realize the ultra-high speed real-time data correlation analysis.
(4) on-line transaction processing (OLTP)
It solves the problem of data acquisition and processing efficiency in the
means of trading space for time.
(5) Real-time display technology (including dynamic display based on GIS and
line graph)
It completely solves the problem of function and performance of the information
release, and ensures the high concurrency in the system application, which benefits
collaborative work.
Uniqueness
(1) It has the complete mobile network performance management functions.

It provides innovative automatic positioning and diagnosis ability of net-
work performance, provides element level, network level, and service level
performance intelligent analysis ability, provides the comprehensive target
management of coverage, capacity, quality of service, quality of the equipment
etc., and comprehensively processes network side and terminal side data, with
point and surface combined and uplink and downlink balanced.
Multi-dimensional correlation search query with performance data is pro-
vided. GIS region correlation analysis function can effectively and directly
reflect the true situation of network. Synchronous adjustment tracking record of
key configuration parameter in wireless network is also provided.
It supports dynamic network element management, including repeater sta-
tion, home base station and other special network. Network management,
network measurement, instrumentation, and other performance data get unified
model definition and storage. User-defined function of the evaluation model is
defined to meet the needs of the development of network management.
(2) Strong quantitative management performance
Mass data association rules processing engine is embedded to adapt to large
telecommunications network applications. The complete network information
quantitative analysis and evaluation model can effectively reduce the
complexity of the operation and maintenance. It adapts to multi service char-

acteristics, and is close to the different service characteristics and quality level.
(3) Easy to deploy and upgrade, with extensive adaptability
Expert experience database, separated from software platform, makes it
convenient for rapid upgrade and experience expansion. Its compatible with
various network test tools, instruments files and data interfaces. It adapts to
multiple network interfaces, data file types, and its easy to integrate with other
systems. It can be seamlessly connected and completely integrated with the
vendor systems. It targets at provincial and regional multi-level mobile network
application modes. And it can be integrated into the process management tools.
(4) Able to meet the needs of daily operation and maintenance of mobile network
Online linkage, and interactive management. According to different types
and levels of users, provide differentiated information customization service.
Provide multi-level multi-perspective network evaluation information, power-
ful information release and report output function, which is conducive to the
integrated management of properties and configuration data related with net-
work performance.
Provide the process management tools interfaces, so as to facilitate the work
process control and management. Provide a sound data storage, management
functions, in order to easily trace the original data.
Requirements to Be Achieved
(1) The unified framework of complex network analysis algorithms: providing the
core theory of abstraction and integration of different types of analysis
algorithms.
(2) Quantitative evaluation model of wireless network performance: providing the
theoretical basis for the analysis and optimization of wireless network.
(3) Correlation analysis multi-dimensional database: providing the data model
meeting the requirements of the logic association mining.
(4) Correlation association analysis engine: providing the software means for
adaptive automatic analysis.
(5) Complex network visualization technology: providing the data presentation and
interactive method of complex relationship information.
Through loading expert experience algorithm library, the embedded intelligent
analysis engine can make automated correlation analysis and processing for system
operation information and test data of various kinds of communications networks.,
It can provide detailed diagnostic and statistical summary information for various
types of network anomaly, faults and causes in multiple forms, and provide the
specific rectification measures and suggestions for operation and maintenance
personnel to choose from.
It features three aspects efficiency, expertise, and ease of use.
Efficiency is reflected in the support for the information correlation processing
of multiple data sources. With massive data processing, effective information can
be fully mined and utilized. Network anomaly analysis and processing can be done
in a fully automated way, greatly reducing the time of manual work.
Expertise is embodied in the process that this technology fully integrates and
cherishes the expertise of communications experts. The algorithm library is flexi-
ble. The analysis process is close to the requirements of the network optimization
site. All the algorithms and solutions are derived from the practice of the network
optimization project. The analysis results are reliable and can be used. Meanwhile,
it outputs the software analysis conclusions in cutting-edge graphics, intuitively
reveals the generation mechanism of the complex network anomaly, makes hard
things simple, strengthens the operators imaginal thinking, and reduces the under-
standing difficulty, making it easy for operators to quickly capture and memorize
key information and solve complex technical problems.
Ease of use means the application and operation process is clear, convenient,
instructive and intuitive. The interface is always reasonably simple, and the interactive
process is comfortable, ensuring the operators work efficiency in massive data
analysis. All kinds of information in the interface are presented in a clear way with
refined visual effect, improving work pleasure to the greatest extent so that the operator
will not get tired easily. Meanwhile it provides a variety of online map automatic
update function, and there is no need for manual installation of the map data, easy for
the operator to carry out network analysis activities at any time any place.
7.4.2 Status
At present, there are many kinds of wireless mobile communications network data
analysis systems in the industry. According to analysis angles (terminal side and
network side) and functions (technical and management), product can be effec-
tively categorized.
Different product functions have different focuses, roughly shown in Fig. 7.19.
As the core of network performance analysis, network optimization software and
other analysis tools require very powerful expertise and comprehensiveness, and
need to achieve the above-mentioned 6 alls analysis ability and other profes-
sional requirements. However, seen from the development course of the commu-
nications system, restricted by the lack actual network verification environment, the
development of network analysis technology is often lagging behind the develop-
ment of hardware and equipment. And thus the development of related products
lags behind the development of network technology.
The development course of the system is shown in Fig. 7.20.
The first generation analysis system provides basic data display function, while
the extraction, sorting, statistical handling of the drive test recorded data need
manual work and the interpretation of data is completely based on the individual
experience. The second generation analysis system provides rich data display and
statistical functions, and can partially replace manual work for drive test data
sorting and statistics, but the data analysis process still needs manual interpretation.
Horizontal axis Professional Ability

Analysis at Phone Side
Vertical axis Analysis Propsect
Driving test tools

Optimization tools
Planning tools
Technical Analysis Management Analysis
Signaling analysis tools Optimization

Management Platform
Equipment OMC
Analysis at Network Side
Fig. 7.19 Data analysis system for wireless mobile communication network
Fig. 7.20 The development course of wireless mobile communication network optimization tools
DT Tools OMC
Performance Report
Configuration Report Network
Testing Phone Testing Manage Configuration
DT Log Equipment
Data Collection Warning
Network
GPS Receiver Data Playback Performance
Equipment
Automatic DT DT Log Signaling Log Signaling

Module Instrument
scanner Log
Scanner Data Playback
Signaling Analysis Post Analysis Signaling Analysis
Statistical Form
Signaling Collection Signaling Collection
Signaling Log
Equipment Analysis Equipment Analysis
User Analysis User Analysis
Analysis On the Phone Analysis On the Network
Fig. 7.21 Composition of intelligent network optimization analysis system
The third generation analysis system introduces the thematic analysis function,
which while replacing human for data sorting and statistics also provides some
thematic analysis process guide, but the manual interpretation remains the core of
each sub process of analysis. The current industry lacks of the fourth generation
analysis system, which, with wide coverage, is able to maximize the integration of
expert analysis experience, and provide the whole-process automated association
analysis function, while human only review and make decisions for the analysis
results, realizing the intelligent software analysis system in the real sense. Wide
coverage requires the system to be compatible with a variety of data sources and
data formats. Integration experts need to convert the complex thinking process into
machine language according to certain models. The automated process requires
each link and each module of the system work together for execution in accordance
with the established sequence in an orderly way. Association function requires
mining the existing data to the greatest possible extent and drain data to the last
drop. In such a way, we can maximize the role of machines and minimize manual
work while ensuring certain degree of accuracy, and the real intelligence can be
reflected.
As the core means of background analysis of the wireless network optimization
department, it deals with all kinds of data in wireless network, including the test
data of mobile phone test and automatic road test, sweeper test data, parameter data
of planned software base station, network management system report data, signal-
ing data, other internal data etc (Fig. 7.21).
7.4.3 Application Scenarios
The specific application scenarios of the intelligent analysis system are shown in
Fig. 7.22.
Network external environment Algorithm development

environment
UE
Algorithm
development
eNodeB eNodeB platform
Configuration Algorithm
DT Log
parameter library
Model
Model Intelligent analysis correction
correction system
Coverage
simulation
System operating environment
Fig. 7.22 Application scenarios of intelligent network optimization analysis system
Application process: start up field operation test ! acquire network test data !
import intelligent acquisition analysis system ! load auxiliary analysis data !
automatically detect anomalies ! automatically analyze network fault ! automat-
ically analyze fault correction measures ! output adjustment measures, model
correction results, coverage simulation results ! implement field network adjust-
ment ! restart field service test !re-acquire test data ! verify network problem
correction.
Innovation Points
Automatic solution generation: It fully integrates and cherishes the expertise of

communications experts, and automatically generates solutions. The analysis algo-
rithm library is flexible, and the analysis process is close to the requirements of the
network optimization site. All the algorithms and solutions are derived from the
practice of the network optimization project. So the analysis results are reliable and
can be used.
Visualization of network problems: It outputs the software analysis conclusions
in cutting-edge graphics, intuitively reveals the generation mechanism of the
complex network anomaly, makes hard things simple, strengthens the operators
imaginal thinking, reduces the understanding difficulty, making it easy for opera-
tors to quickly capture and memorize key information and solve complex technical
problems.
Coordinated analysis of multiple data sources: It supports the coordinated
analysis of multiple networks and multiple modes, supports the information
correlation processing of multiple data sources. With massive data processing,

effective information can be fully mined and utilized.
Application Model: Complex Network Visualization Technology
It provides data display and interaction means of complex relational information,

provides rich information view with progressive levels, so as to realize the brand-
new network optimization modes tracking, auditing, reviewing, and meet the
application requirements of different types of users (Fig. 7.23).
As for the management evaluation level, generally it uses control management
view and overall view, focuses on the overall situation, uses the analysis conclusion
navigational function between views, and focuses on problem tracking.
As for the optimization and supervision level, generally it uses overall view and
local view, uses the analysis conclusion drilling function between views, focuses on
determining the overall optimization decision, and guides the optimization.
As for the audit and authentication level, it generally uses local view and manual
analysis view, uses the raw data drilling function between views, focuses on the
specific issues positioning, analysis and settlement, as well as the accuracy of the
audit conclusion.
In interaction, a large number of views can be summarized into three levels and
five dimensions, in which three levels refer to the three-level structure of the
overall view, local view and manual analysis view, and the five dimensions refer to
the management dimension, feature dimension, element dimension, geographical
Business scenario Application model
Control management
view
Management Analysis conclusion navigation
Problem
evaluation tracking
Overall view
Optimizing
supervision Optimization
Conclusion drill
decision
Audit Local view

verification
Conclusion
The original data of drilling
audit
Manual analysis view
Fig. 7.23 Working mode of intelligent network optimization analysis system

Analysis conclusion navigation portal

Overall view
Conclusion drill
Management Feature NE Geographical Time
dimension dimension dimension dimension dimension
Local view
The original data of drilling Distribution Event sequence

TopN view Geographic view
network diagram
Manual analysis view
Summary
Statistics
Special guide Global / filtering

decision map
alarm
management
defect
management
Cross dimension
inspection Anomaly analysis
guide
Trajectory
Process analysis Signaling view Trend chart Parameter view
diagram
Fig. 7.24 The brand new three levels and five dimensions information interaction mode
dimension and time dimension. The relationship of the views is shown in the
following figure (Fig. 7.24).
Application Mode: Building the Macro Optimization Intelligent System
Based on the in-depth mining of the intelligent diagnosis conclusions, it provides

the connected information management and control objects of multi-level network
optimization of objects, which can effectively connect the management and pro-
duction tasks and meet the different levels of information needs of the network
optimization operation and maintenance department (Fig. 7.25).
Risk area: It is automatically generated based on the characteristics like the
distribution density of abnormal points. It is used for the active identification and
analysis on the service quality high risk areas in the process of network operation.
Alarm points: According to the current network optimization goals, the abnor-
mal points of specific types that require high attention are automatically extracted
and gathered, used to determine the key rectification projects of the current network
optimization.
Abnormal points: They are automatically generated in the anomalies detection
process, providing detailed anomalies information, causes and rectification recom-
mendations, used for the decisions in specific network optimization rectification
program.
Based on multi-level control mode, the network optimization personnel
of different functions can respectively deal with the operation and
Fig. 7.25 The three-level network optimization management and control system that covers both
points and surfaces
maintenance tasks with different urgency and importance levels, achieving the
seamless docking of the active optimization and ensuring optimization goals
between the upper and lower levels personnel, and completing the network opti-
mization coordinately.
Application Mode: Closed Loop Intelligent Management

of Network Problems
As the participants and managers of network optimization operation and mainte-

nance, we need to grasp the overall performance of the wireless test. And the
formation of abnormal events and the life cycle of the causes behind, including
generation, maintaining, and elimination, which not only is necessary but also
serves as a kind of effective management on the network optimization process.
The function of the closed loop intelligent management of network problems, by
referring to the mature network management application model and based on
intelligent analysis function, is the first in the industry to create the road test
alarm/fault detection and management, effectively monitoring the tracking of
problem loop.
Based on the road test alarm function, it can track the multiple batches of data
and get in-depth understanding of the alarm points status, grade, cause, distribution
and other information. Flexible management can be realized on site through alarm
type, road location, alarm status, date, etc. Meanwhile, for each problem, it can
view the causes of the current and past problems, solutions and other relevant
information (Fig. 7.26).
Analysis set automatic analysis conclusion

Geographic clustering
Alarm
abnormal point
According to the
origin collection element Defective
item
Fig. 7.26 Closed loop management of the problem points in the daily optimization stage
eliminate hidden danger

maintain Newly added
118 83
300
31
67
68
200 58
25
Alarm 63 30
26 38
number
46 18
16 34 116
21 18 104
100 36 89
82 47 80
0
20
33 8
0
2013-3-22 2013-4-23 2013-5-7 2013-6-27 2013-7-26 2013-8-7 2013-9-12 2013-10-10
Fig. 7.27 Road test alarm history information
Clustering by Geography. Road test abnormal events, based on the configurable

generation rules, alarm objects can be automatically generated. Alarm points
belong to the geographic area and the tracking history can be set up according to
the road test batches.
Categorizing by Network Elements. Based on the fault causes of the anomalies
of the road test alarm, the defect objects are generated. Defect item falls into
network elements and the tracking history can be set up according to the road test
batches. Defect items support the setting of functional departments, conducive to
differentiated input and coordinated rectification.
By tracking the road test alarm statistics of the multiple data, it can accurately
grasp the changes in the number of network obstacles (Fig. 7.27).
Application Mode: Intelligent Single Station Batch Submission For

Acceptance
The verification and optimization of single stations is the base of the whole
wireless optimization of mobile network. It is the unavoidable initial stage of
network optimization. The solution of single stations can help eliminate the
hidden troubles in the project construction stage, and effectively reduce the
pressure of the follow-up optimization work, so as to lay a solid foundation for
the higher level of the whole network optimization. Network construction and
maintenance department are generally in urgent need of professional and highly-
efficient technical means to replace manual analysis so that they can automati-
cally, accurately and quickly make optimization analysis and acceptance review to
the new base station inspection test data, quickly master each base stations
running status, performance and fault causes. In such a way, they can give
effective coordinated rectification to the existing problems so that the new stations
can quickly meet the acceptance criteria and enter into formal operation as soon as
possible.
Intelligent single station batch submission for acceptance function can make
fast, in batch and rolling judgment about whether the base station can fully meet
the preset quality standards, and provide detailed information on the fault
positioning. So that the various functional departments can accurately,
rapidly and collaboratively make rectification and eliminate the hidden troubles
of network construction to the greatest extent. Coordinated rectification can
only be implemented when the network problem causes are positioned
(Fig. 7.28).
1 eNodeB intersection dimension 2 Inspection road survey 3 automatic analysis
anomaly
detection
fault
4 Submit inspection report analysis
B C
A
Inspection and
inspection
Fig. 7.28 Base station batch submission for inspection function

7.5 Summary 433
Batch 1 Results
1st OK
Base
Test log for
Inspection mission Analysis set station
each batch batch 1 1st NOK
1st NULL
Legacy base Has passed the

base station
Batch 2 Results
1st NOK
st
1 NULL 1st OK
Base 2nd OK
station 1st NOK
batch 2
2nd NOK
1st NULL
2nd NULL
Fig. 7.29 Multi-batch rolling base station inspection
It supports the checking and analysis of multi-batch data, makes it easy to track
failed base stations. The inspection results are summarized in rolling way. Based on
the submission for inspection function, the inspection road test data analysis is
carried out automatically. The progress and problems of the base station optimiza-
tion project can be mastered synchronically (Fig. 7.29).
7.5 Summary
As stated at the beginning of this chapter, the ultimate goal of the R&D and
evaluation of mobile communications technology is application. The field trial is
the last and also the most important part of the process from the technology R&D to
application. In particular, the acceleration of the new technology update makes it
more important to carry out the field trial and verification in the R&D stage. This
chapter, based on the analysis on the requirements and application scenarios of 5G
mobile communications technology, gives the field trial evaluation cases in the
LTE-A phase from the wireless channel measurement to the key technology
validation and then to the network performance validation, providing references
for the later 5G technology research. In addition, with the development of big data
technology, intelligent wireless network data analysis tools will greatly enhance the
network performances self-optimization ability. At the end of this chapter, we
present the architecture that is still in planning stage of the multi-scenario hetero-
geneous test network that supports large scale test users, providing an ideal test site
for the field trial and evaluation in R&D stage for relevant research institutions.
References
1. Scenarios, Requirements and KPIs for 5G Mobile and Wireless System. METIS Deliverable
D1.1, Apr. 2013, https://www.metis2020.com/wp-content/uploads/deliverables/METIS_D1.
1_v1.pdf
2. White Paper on 5G Concept. IMT-2020 (5G) Promotion Group. Feb 2015. http://www.imt-
2020.cn/zh/documents/download/23
3. 5G Vision and Requirements. IMT-2020 (5G) Promotion Group. May 2014. http://www.imt-
2020.org.cn/zh/documents/download/1
4. A. Ghosh, N. Mangalvedhe, R. Ratasuk and et al. Heterogeneous cellular networks:From
theory to practice[J]. Communications Magazine, IEEE, 2012, 50(6): 5464.
5. B. Soret, H. Wang, K. Pedersen and et al. Multicell cooperation for LTE-advanced heteroge-
neous network scenarios. IEEE Wireless Communications, 2013, 20(1): 2734.
6. W. Stallings. ISDN and broadband ISDN. Macmillan Publishing Co., Inc., 1992.
7. F. Xu, L. Zhang and Z. Zhou. Interworking of Wimax and 3GPP networks based on IMS
[IP Multimedia Systems (IMS) Infrastructure and Services]. IEEE Communications Magazine,
2007, 45(3): 144150.
8. M. He, X. Wang, T. D. Todd and et al. Ad hoc assisted handoff for real-time voice in IEEE
802.11 infrastructure WLANs. International Journal of Wireless and Mobile Computing, 2007,
2(4): 324336.
9. R. Ananthapadmanabha, B. S. Manoj and C. Murthy. Multi-hop cellular networks:the archi-
tecture and routing protocols. Personal, IEEE International Symposium on Indoor and Mobile
Radio Communications, 2001, 2:G-78-G-82 vol. 2.
10. A. N. Zadeh, B. Jabbari, R. Pickholtz R and et al. Self-organizing packet radio ad hoc networks
with overlay (SOPRANO). IEEE Communications Magazine, 2002, 40(6): 149157.
11. C. Qiao, H. Wu and O. Tonguz. Integrated cellular and ad hoc relaying system: U.S. Patent
Application 10/213,058[P]. 20028-6.
12. H. Luo, R. Ramjee, P. Sinha and et al. UCAN: a unified cellular and ad-hoc network
architecture. Proceedings of the 9th annual international conference on Mobile computing
and networking. ACM, 2003:353367.
13. M. Dick. Mobility and differentiated services in a future IP network. IST Project. http://www.
ist-mobydick.org, 2003.
14. T. Braun, M. Diaz, J. E. Gabeiras and et al. End-to-end quality of service over heterogeneous
networks. Springer Science & Business Media, 2008.
15. W. Mohr. The WINNER (Wireless World Initiative New Radio) projectdevelopment of a
radio interface for systems beyond 3G. International Journal of Wireless Information Net-
works, 2007, 14(2): 6778.

Yang Yang-5G Wireless Systems. Simulation and Evaluation Techniques-Springer (2017)

Uploaded by

Copyright:

Available Formats

Yang Yang-5G Wireless Systems. Simulation and Evaluation Techniques-Springer (2017)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Yang Yang-5G Wireless Systems. Simulation and Evaluation Techniques-Springer (2017)

Uploaded by

Copyright:

Available Formats

What are the main topics covered in the document?

What are the main topics covered in the document?

What are some of the references cited in the document?

What are some of the references cited in the document?

Wireless Networks

Guang Shi Cheng-Xiang Wang

ISSN 2366-1186 ISSN 2366-1445 (electronic)

Library of Congress Control Number: 2017947041

Springer International Publishing AG 2018

Printed on acid-free paper

This Springer imprint is published by Springer Nature

and testing cases, which can be used as technical references by 5G researchers,

Shanghai, China Yang Yang

1 Candidate Technologies and Evaluation Challenges for 5G . . . . . . . 1

4.3.2 Link Simulation Realization . . . . . . . . . . . . . . . . . . . . . . 168

6.2.5 Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

7.4 Wireless Network Data Intelligence Analysis . . . . . . . . . . . . . . . . 419

5G The Fifth Generation mobile communications system

CPU Central Processing Unit

H-CRAN Heterogeneous Cloud Radio Access Network

MMSE Minimum Mean Squared Error

PCFICH Physical Control Format Indicator CHannel

Rma Rural Macro

SLS Sector Level Sweep

UMass University of Massachusetts

With several decades of booming development, mobile communications technol-

1.1 5G Requirements Analysis

Mobile communications industry has been developing at an amazing speed, and at

Springer International Publishing AG 2018 1

Fig. 1.1 Usage scenarios of IMT for 2020 and beyond

1.1.1 5G Application Types

numerous refreshing applications, reaching an effect of flowers blooming

1.1.2 5G Application Scenarios and Requirements Indicators

Table 1.1 5G performance requirements

1.2 5G Research and Development Process

Twenty-twenty (2020) Information Society (METIS). There were 29 project mem-

1.3 5G Candidate Technologies

1.3.1 Ultra-Dense Wireless Network

1.3.2 Large Scale Antenna Technology

In practical scenarios, the point-to-point multiple antenna technology needs multi-

1.3.3 Millimeter Wave Communications Technology

Services (LMDS), cellular communications backhaul link and wireless transmis-

or outage states. Of course, line-of-sight and non-line-of-sight transmissions have

1.3.4 Flexible Spectrum Usage

In the process of the rapid development of mobile communications, one of the

Finally, in peer spectrum sharing between operators, operators can jointly

1.3.5 Waveform and Multiple Access

1.3.6 Device to Device

synchronization time, resulting in multiple synchronization sources. Alterna-

1.4 Challenges in 5G Testing and Evaluation

1.4.1 Challenges Posed by New Technologies to Testing

According to the network requirements and technical indicators defined in 5G white

1.4.2 Four Elements of Testing and Evaluation

Affected by flexible network architecture, network resource virtualization manage-

Realness Comprehen- Rapidity Flexibility

Chapter 3 Chapter 5 Chapter 6 Chapter 7

2.1 The Importance of Testing Technology

Testing means measurement, inspection and test. In different application fields,

Springer International Publishing AG 2018 21

software. The process of compiling software is the process of testing, debugging

i;k i;k1 i;k i;k1