AkramHouraniThesisV3.1 201610518 PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 295

Performance Evaluation of Future

Wireless Networks: Node Cooperation


and Aerial Networks

A thesis submitted in fulfilment of the requirements for the degree of


Doctor of Philosophy

Akram Hourani
B. Eng.

School of Electrical and Computer Engineering


College of Science, Engineering and Health
RMIT University

February 2016
Declaration

I certify that except where due acknowledgement has been made, the work is that of the
author alone; the work has not been submitted previously, in whole or in part, to qualify for
any other academic award; the content of the thesis is the result of work which has been
carried out since the official commencement date of the approved research program; any
editorial work, paid or unpaid, carried out by a third party is acknowledged; and, ethics
procedures and guidelines have been followed.

Akram Hourani
February 2016

i
A modest contribution for a better future. . .

and...

To my loving wife and son, and to my loving parents, brother and sister. . . .

iii
Acknowledgements

I would like to express my deepest gratitude to my primary supervisor Dr. Kandeepan


Sithamparanathan for giving me the great opportunity to pursue this PhD, during which he
provided an excellent research atmosphere and technical support. I also would like to thank
him for his patience, guidance, caring and for his experience that considerably added to
my research skills. My special thanks to my second supervisor Prof. Abbas Jamalipour as a
great mentor and advisor during the course of my PhD program. I also would like to express
my acknowledgment to Prof. Ekram Hossain for his great technical advises on developing
the energy efficiency analysis of assisted device-to-device communication manuscript. And
my great acknowledgment to Prof. Bill Moran for his great advise and for the time he spent
during our meetings in developing the manuscript related to CMHC.
My special thanks to all co-authors of the published manuscripts during the course of this
PhD and for the anonymous reviewers that spent considerable time reading and commenting
on my work. I would like to acknowledge my colleagues Sathyanarayanan Chandrasekharan
and Kagiso Magowe for their careful reading and feedback on my manuscripts.
I would like to acknowledge the financial support given by the school of electrical and
computer engineering in RMIT university, and the financial support given by the Australian
government through the Australian postgraduate award, as well as the financial support from
Orange labs. Many thanks for the funds by the ABSOLUTE project from the European
Commission’s Seventh Framework Programme (FP7-2011-8) Grant Agreement FP7-ICT-
318632 and by the TC-RAN project from Orange Labs Research Contract No. 60773972
for partially funding experimentation tools, conference costs and travel expenses.
And last by not least, I would like to thank my beloved wife Tetiana for her great sup-
port and encouragement during the course of my PhD. My endless gratitude to my father
for opening my eyes to the charm of science and for inculcating the love of innovation and
mathematics, and to my mother for her sleepless nights of caring, teaching and the immea-
surable love she gave me.

iv
Contents

Contents v

List of Tables x

List of Figures xi

Abstract 1

1 Introduction 4
1.1 Node Cooperation Technology and D2D . . . . . . . . . . . . . . . . . . . 5
1.2 Aerial Networks Technology . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Summary of Literature Gaps . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Channel Modelling for Aerial Networks and D2D . . . . . . . . . . 12
1.3.2 Performance Estimation and Optimization of Cellular and Aerial
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.3 Performance Enhancement with Node Cooperation . . . . . . . . . 15
1.3.4 Energy Efficiency with Node Cooperation . . . . . . . . . . . . . . 16
1.4 Research Questions and Contributions . . . . . . . . . . . . . . . . . . . . 18
1.4.1 Research Question 1 and Contributions . . . . . . . . . . . . . . . 18
1.4.2 Research Question 2 and Contributions . . . . . . . . . . . . . . . 20
1.4.3 Research Question 3 and Contributions . . . . . . . . . . . . . . . 22
1.4.4 Research Question 4 and Contributions . . . . . . . . . . . . . . . 24
1.5 Publications List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.5.1 Book Chapters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.2 Journal Articles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.3 Conference Proceedings . . . . . . . . . . . . . . . . . . . . . . . 27
1.5.4 Technical Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

v
Contents vi

2 Channel Modelling, Measurements and Experimentations 30


2.1 D2D Channel Modelling in mmWave . . . . . . . . . . . . . . . . . . . . 33
2.1.1 mmWaves Propagation Characteristics . . . . . . . . . . . . . . . . 34
2.1.2 ISM Bands in mmWave Spectrum . . . . . . . . . . . . . . . . . . 34
2.1.3 Atmospheric Attenuation . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.4 D2D Communication Under LTE-Advanced . . . . . . . . . . . . . 37
2.1.5 Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.6 D2D Simulation in mmWave Spectrum . . . . . . . . . . . . . . . 41
2.1.7 path-loss Exponent in Urban Environment . . . . . . . . . . . . . . 42
2.1.8 D2D Link with Beam Steering . . . . . . . . . . . . . . . . . . . . 45
2.2 Vehicle to Vehicle Channel Modelling . . . . . . . . . . . . . . . . . . . . 48
2.2.1 V2V/V2I Physical Channel . . . . . . . . . . . . . . . . . . . . . . 50
2.2.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2.3 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2.4 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2.5 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3 Air-to-Ground Channel Modelling . . . . . . . . . . . . . . . . . . . . . . 60
2.3.1 Low Altitude Platforms . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3.2 Modeling Urban Environment . . . . . . . . . . . . . . . . . . . . 61
2.3.3 Radio Model Development . . . . . . . . . . . . . . . . . . . . . . 64
2.3.4 Understanding Simulation Results . . . . . . . . . . . . . . . . . . 66
2.3.5 Obtaining Radio Model Parameters . . . . . . . . . . . . . . . . . 69
2.3.6 Radio Model Implementation . . . . . . . . . . . . . . . . . . . . 73
2.3.7 Radio Model Verification . . . . . . . . . . . . . . . . . . . . . . . 74
2.4 Identifying Spectrum Opportunities for Node Cooperation . . . . . . . . . 76
2.4.1 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.4.2 Spectrum Occupancy Results . . . . . . . . . . . . . . . . . . . . 82
2.5 Research Question 1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 87

3 Coverage and Rate Analysis and Optimisation 88


3.1 Cellular Networks Performance Analysis Under Generic Channel Fading . 91
3.1.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.1.2 Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.1.3 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.1.4 Modeling Cellular Interference . . . . . . . . . . . . . . . . . . . . 96
Contents vii

3.1.5 Modeling Coverage Probability . . . . . . . . . . . . . . . . . . . 98


3.1.6 Network Performance Analysis . . . . . . . . . . . . . . . . . . . 99
3.1.7 Path-Loss Only Scenario . . . . . . . . . . . . . . . . . . . . . . . 100
3.1.8 Rayleigh Fading Scenario . . . . . . . . . . . . . . . . . . . . . . 100
3.1.9 Rayleigh Interferers with Rician Serving Signal . . . . . . . . . . . 102
3.1.10 Frequency Reuse . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.1.11 Network Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.1.12 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.2 Aerial Communication Performance Analysis . . . . . . . . . . . . . . . . 109
3.2.1 Air-to-Ground Channel Model . . . . . . . . . . . . . . . . . . . . 110
3.2.2 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . 111
3.2.3 Model Implementation . . . . . . . . . . . . . . . . . . . . . . . . 112
3.2.4 Performance Optimization . . . . . . . . . . . . . . . . . . . . . . 115
3.2.5 System Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.3 Aerial Coverage Optimization . . . . . . . . . . . . . . . . . . . . . . . . 118
3.3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.3.2 RF Propagation Model . . . . . . . . . . . . . . . . . . . . . . . . 119
3.3.3 Modeling Line of Sight Probability . . . . . . . . . . . . . . . . . 120
3.3.4 Finding the Optimum Altitude . . . . . . . . . . . . . . . . . . . . 122
3.3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.4 Research Question 2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 128

4 Coverage Enhancement with Node Cooperation and Relaying 129


4.1 Cellular Damage Alleviation with Node Cooperation . . . . . . . . . . . . 132
4.1.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.1.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.1.3 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.1.4 Link-level Success Probability . . . . . . . . . . . . . . . . . . . . 140
4.1.5 Spatial Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.6 Link-level Performance . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.7 Network-Level Performance . . . . . . . . . . . . . . . . . . . . . 144
4.1.8 Chain Relay Network Performance . . . . . . . . . . . . . . . . . 147
4.1.9 Network Performance During a Disaster . . . . . . . . . . . . . . . 150
4.1.10 Monte-Carlo Simulation . . . . . . . . . . . . . . . . . . . . . . . 152
4.1.11 Concentrated Damage Model . . . . . . . . . . . . . . . . . . . . 155
Contents viii

4.2 Temporary Cognitive Femtocell Network . . . . . . . . . . . . . . . . . . 160


4.2.1 Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.2.2 Downlink Interference Mitigation . . . . . . . . . . . . . . . . . . 164
4.2.3 Cognitive Relaying Mechanism . . . . . . . . . . . . . . . . . . . 168
4.3 Cognitive Relay Nodes for Airborne Coverage Enhancement . . . . . . . . 174
4.3.1 Relay Nodes in LTE-Advanced . . . . . . . . . . . . . . . . . . . 175
4.3.2 Rapidly Deployabel Cognitive Relay Nodes . . . . . . . . . . . . . 176
4.3.3 Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.3.4 Cognitive Relay Assisted Emergency Network . . . . . . . . . . . 179
4.3.5 Simulation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 187
4.4 Research Question 3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 193

5 Energy Efficiency Enhancement with Node Cooperation and Relaying 195


5.1 Energy Efficiency for Cellular Networks . . . . . . . . . . . . . . . . . . . 197
5.1.1 Related Work and Novelty of our work . . . . . . . . . . . . . . . 198
5.1.2 System Model and Assumptions . . . . . . . . . . . . . . . . . . . 200
5.1.3 Energy Saving by Relaying . . . . . . . . . . . . . . . . . . . . . . 202
5.1.4 Energy Saving Zone . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.1.5 Spatial Distribution of the Energy Saving Gain . . . . . . . . . . . 207
5.1.6 Energy Saving Gain With Multiple Relays . . . . . . . . . . . . . . 211
5.1.7 Energy Consumption Analysis . . . . . . . . . . . . . . . . . . . . 214
5.1.8 Performance Benchmarking . . . . . . . . . . . . . . . . . . . . . 216
5.1.9 Asymptotic Behaviour of the Distance Threshold . . . . . . . . . . 220
5.1.10 Monte-Carlo Simulations and Discussion . . . . . . . . . . . . . . 221
5.2 Energy Efficiency for Aerial Backhauled Networks . . . . . . . . . . . . . 227
5.2.1 Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.2.2 Energy Consumption Model . . . . . . . . . . . . . . . . . . . . . 232
5.2.3 Energy Consumption Analysis . . . . . . . . . . . . . . . . . . . . 234
5.2.4 Performance Benchmarking . . . . . . . . . . . . . . . . . . . . . 236
5.3 Energy Efficiency with Matérn Hard-Core Point Process . . . . . . . . . . 240
5.3.1 Complementary Matérn Hard-Core Point Process . . . . . . . . . . 241
5.3.2 Points Interaction in CMHC . . . . . . . . . . . . . . . . . . . . . 243
5.3.3 Nearest Neighbour Distribution . . . . . . . . . . . . . . . . . . . 250
5.3.4 Energy Consumption Analysis . . . . . . . . . . . . . . . . . . . . 251
5.4 Research Question 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 256
Contents ix

6 Conclusion and Future Research 257


6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
6.2 Future Research Paradigms . . . . . . . . . . . . . . . . . . . . . . . . . . 260

References 262
List of Tables

2.1 Microwave & mmWave ISM Bands . . . . . . . . . . . . . . . . . . . . . 35


2.2 Unlicensed Local mmWave Regulations . . . . . . . . . . . . . . . . . . . 35
2.3 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4 D2D Mean path-loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 path-loss Deviation from MPL . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 D2D path-loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Experiment Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.8 path-loss Exponent (α) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.9 Estimated Shadowing: Computed Standard Deviation (σ ) . . . . . . . . . 59
2.10 Estimated α, σ and PL(d0 ) from measurement campaign in [1] . . . . . . . 59
2.11 Selected Urban Environments . . . . . . . . . . . . . . . . . . . . . . . . 63
2.12 RF Model Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.13 Measurement Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.1 Notations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114


3.2 Surface polynomial coefficients for a . . . . . . . . . . . . . . . . . . . . . 122
3.3 Surface polynomial coefficients for b . . . . . . . . . . . . . . . . . . . . . 123

4.1 Interference Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137


4.2 Notations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.3 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.4 UE Database formed by the CRN . . . . . . . . . . . . . . . . . . . . . . . 186
4.5 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

5.1 Comparison Among Different Relaying Region Formulation . . . . . . . . 199


5.2 Solutions of S1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
5.3 Notations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

x
List of Figures

1.1 D2D communication scenarios by 3GPP. . . . . . . . . . . . . . . . . . . . 7


1.2 Possible future D2D communication scenarios. . . . . . . . . . . . . . . . 9
1.3 Aerial networks supporting terrestrial cellular coverage. . . . . . . . . . . . 10

2.1 Aerial networks and node cooperation channel modeling. . . . . . . . . . . 31


2.2 Atmospheric Absorbtion due to Water Vapor and Gasses . . . . . . . . . . 36
2.3 Device to Device Communication Concept . . . . . . . . . . . . . . . . . . 38
2.4 Building Layout, and Simulation Setup . . . . . . . . . . . . . . . . . . . . 40
2.5 Proposed Antennas Combined Patterns (13 out 16 antennas are shown only) 41
2.6 RSRP Level due to a D2D transmitter at 60GHz . . . . . . . . . . . . . . . 42
2.7 Device-to-Device path-loss in Urban Environment . . . . . . . . . . . . . . 44
2.8 Device-to-Device Deviation from MPL Distribution . . . . . . . . . . . . . 46
2.9 System Level Throughput at 24GHz and 61GHz . . . . . . . . . . . . . . . 47
2.10 Performance Benchmarking at 5% and 50% Percentile . . . . . . . . . . . 47
2.11 Experiment environments (A)/(B) in Urban area, and (C) in Suburban area
(Map data: Google, Sinclair Knight Merz & Fugro) . . . . . . . . . . . . . 51
2.12 Transmitter and receiver placement, for scenario-1 (no car). . . . . . . . . . 54
2.13 Transmitter and receiver placement, for scenario-2 (with car). . . . . . . . . 55
2.14 Received power samples and fitting curves, from top to bottom:
Environment-A Heights H1 and H2, Environment-B Heights H1 and H2,
and Environment-C Heights H1 and H2 . . . . . . . . . . . . . . . . . . . 58
2.15 Low altitude platform concept . . . . . . . . . . . . . . . . . . . . . . . . 62
2.16 The selected urban layout . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.17 Computer generated city models based on the ITU-R parameters. . . . . . . 65
2.18 The three different types of the simulated rays . . . . . . . . . . . . . . . . 67
2.19 Received Power, obtained at (Frequency = 2,000 MHz, Tx-Height = 200 m,
Tx-Power = 0 dBm) for urban environment. . . . . . . . . . . . . . . . . . 67

xi
List of Figures xii

2.20 Excessive path-loss samples histogram, obtained at frequency = 2,000 MHz


for a dense urban environment. . . . . . . . . . . . . . . . . . . . . . . . . 68
2.21 Excessive path-loss samples organized in terms of the elevation angle θ , and
categorized into three distinct propagation groups, obtained at frequency =
2,000 MHz for a dense urban environment. . . . . . . . . . . . . . . . . . 69
2.22 Excessive path-loss Standard Deviation, obtained at frequency = 2,000 MHz
for an urban environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.23 Propagation Group Occurrence Probability, obtained at frequency = 2,000
MHz for an urban environment. . . . . . . . . . . . . . . . . . . . . . . . 71
2.24 Cumulative Distribution Function comparison between the proposed RF
model vs. ray-tracing simulation data. . . . . . . . . . . . . . . . . . . . . 75
2.25 Illustration of the experiment setup, and the measurement equipment. . . . 78
2.26 The travel route of the experiment in greater Melbourne is divided into 3
classes or regions. The left bar indicates the population density as per the
Australian Bureau of Statistics. [2] . . . . . . . . . . . . . . . . . . . . . . 80
2.27 The discone antenna gain, with/without the effect of the attached cable loss. 81
2.28 Noise analysis of the experiment’s spectrum analyser, represented in terms
of a normalized histogram. Indicating the chosen detection threshold of 5%
CCDF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.29 The spectrogram obtained from the raw samples collected in the three differ-
ent urban classes. From top to bottom (high, medium and low) population
density. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.30 The occupancy graph obtained by comparing the raw samples against the
detection threshold. Dark spots indicate a busy spectral resource (B = 1)
while the white spaces indict free spectral resources (B = 0). From top to
bottom (high, medium and low) population density. . . . . . . . . . . . . . 84
2.31 Duty cycle obtained by comparing the number of occupied samples to the
total number of collected samples for each frequency line. From top to
bottom (high, medium and low) population density. . . . . . . . . . . . . . 85

3.1 Performance estimation of cellular and aerial networks. . . . . . . . . . . . 89


3.2 Cellular Poisson Voronoi tessellation, indicating the typical user located at
the origin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.3 All interfering base stations are located outside the ball b(o, Ro ). . . . . . . 97
3.4 The different dynamics affecting the coverage probability. . . . . . . . . . 100
List of Figures xiii

3.5 The coverage probability, when adopting log-distance path-loss channel


model without considering fading effects. . . . . . . . . . . . . . . . . . . 101
3.6 The coverage probability, for log-distance path-loss and Rayleigh small
scale fading, showing a comparison with the results of [3]. . . . . . . . . . 102
3.7 The coverage probability, for Rayleigh interfering signals and Rician serv-
ing signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.8 The layout of a random frequency reuse scheme in a Poisson Voronoi tes-
sellation cellular network. . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.9 The effect of frequency reuse on the service success probability at an SINR
threshold of T = 10dB. where (S) refers to the serving signal and (I) refers
to the interfering signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.10 The expected network rate obtained using trapezoidal integration of equa-
tion (3.22), versus system’s minimum SINR γmin and γo = 1. (S) refers to
the serving signal and (I) refers to the interfering signals. . . . . . . . . . . 106
3.11 Coverage probability simulation results of the three proposed scenarios,
compared to the analytical integration obtain using the coverage equation.
(S) refers to the serving signal and (I) refers to the interfering signals. . . . 108
3.12 The coverage probability PcovR (αo , βo , γo ) versus the users’ ground dis-
tance R. The utilized SNR threshold is T = 10. The desecrate points repre-
sent the simulation while continuous curves are the analytical performance
in (3.31). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.13 The coverage probability Pcov (αo , βo , γo ) versus the SNR threshold. The
desecrate points represent the simulation while continuous curves are the
analytical performance in (3.29). . . . . . . . . . . . . . . . . . . . . . . . 115
3.14 The expected rate of platform versus a range of minimum SNR γmin . The
desecrate points represent the simulation while continuous curves are the
analytical performance in (3.30). . . . . . . . . . . . . . . . . . . . . . . . 116
3.15 Coverage performance versus platform’s altitude, for a fixed coverage area
Rmax , showing the optimum altitude for a given SNR threshold T = 20dB. . 117
3.16 Low Altitude Platforms radio propagation in urban environment. . . . . . . 119
3.17 The calculated line-of-sight probabilities, with their related S-Curve fitting
for different urban environments. . . . . . . . . . . . . . . . . . . . . . . . 121
3.18 S-Curve parameters 3D-fitting as a relation to the urban environment pa-
rameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.19 The coverage zone by a low altitude platform. . . . . . . . . . . . . . . . . 124
List of Figures xiv

3.20 Cell radius vs. LAP altitude curve for different urban environments. . . . . 126
3.21 Cell radius vs. LAP altitude curve for different maximum pathloss, in urban
environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

4.1 Coverage enhancement with node cooperation and relaying. . . . . . . . . 130


4.2 Random cellular layout according to a PPP, showing cell dominance bound-
aries and base stations’ locations. . . . . . . . . . . . . . . . . . . . . . . . 133
4.3 The relay chain principle, for extending healthy cells coverage. . . . . . . . 137
4.4 The proposed semi-static spectrum allocation. . . . . . . . . . . . . . . . . 138
4.5 Interference on cellular and D2D links. . . . . . . . . . . . . . . . . . . . . 138
4.6 Link-level success probability vs. the distance from the serving transmitter,
for a healthy network with D = 0. . . . . . . . . . . . . . . . . . . . . . . 145
4.7 Nearest eNB distance probability distribution. . . . . . . . . . . . . . . . . 146
4.8 The number of required hops for reaching 95% of the asymptotic service
level κL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.9 Network-level success probability against varying values of the damage ra-
tio D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.10 The damage alleviation buffer due to the introduction of D2D relays. . . . . 152
4.11 Downlink coverage simulation for a healthy network, i.e. damage ratio D =
0% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.12 Serviced UEs, with a network damage ratio = 50% . . . . . . . . . . . . . 155
4.13 Comparison between the analytical approach and Monte-Carlo simulation
runs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.14 Concentrated phase-out model, the (+) sign is the central point of the disaster.157
4.15 Performance comparison between the two models of damage propagation:
(i) The random phase-out, (ii) The concentrated phase-out. . . . . . . . . . 159
4.16 Network model with the three types of networks. . . . . . . . . . . . . . . 162
4.17 TCFN Cluster Concept. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.18 Public Safety Network Operational Stages. . . . . . . . . . . . . . . . . . . 164
4.19 chUE muted subframes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.20 Symbol muting and shifting concept . . . . . . . . . . . . . . . . . . . . . 166
4.21 Soft FFR Concept, Underlined bands shall be used only in conjugation with
time domain interference coordination . . . . . . . . . . . . . . . . . . . . 167
4.22 Simulation model and Target building location . . . . . . . . . . . . . . . . 169
4.23 Spatial service probability . . . . . . . . . . . . . . . . . . . . . . . . . . 169
List of Figures xv

4.24 RSRP for the Baseline System (Parent Network only) . . . . . . . . . . . . 171
4.25 RSRP when uncoordinated TCFN is deployed . . . . . . . . . . . . . . . . 171
4.26 RSRP when enabling the coordinated TCFN . . . . . . . . . . . . . . . . . 172
4.27 Spatial RS-SINR CDF Comparison . . . . . . . . . . . . . . . . . . . . . . 172
4.28 Overview of REN Architecture with CRNs. . . . . . . . . . . . . . . . . . 177
4.29 The proposed AeNB Antenna configuration, adopting the requirements
from the ABSOLUTE project [4]. . . . . . . . . . . . . . . . . . . . . . . 180
4.30 Illustrating the concept of CRN Assisted REN. . . . . . . . . . . . . . . . 181
4.31 The process of CRN intelligently attaching to an AeNB. . . . . . . . . . . 182
4.32 Switched antenna in CRN (with two transceivers). . . . . . . . . . . . . . . 184
4.33 Beam Steering for directing RF power in the network. . . . . . . . . . . . . 185
4.34 Illustrating the concept of geographical pixels service by AeNB. . . . . . . 187
4.35 Simulation results for obtaining the Optimum LAP Altitude for the assumed
urban environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.36 Comparison between the service blockage probability. . . . . . . . . . . . 189
4.37 RSRP Scenario 1 (AeNBs Only). . . . . . . . . . . . . . . . . . . . . . . . 190
4.38 RSRP Scenario 2 (AeNBs and Conventional RNs). . . . . . . . . . . . . . 190
4.39 RSRP Scenario 3 (AeNBs and CRNs). . . . . . . . . . . . . . . . . . . . . 191
4.40 Comparative CDF Plot for the three Scenarios. . . . . . . . . . . . . . . . 191
4.41 Damage alleviation buffer against the number of cascaded relays. . . . . . . 193

5.1 The three different scenarios considered for the benchmarking. . . . . . . . 201
5.2 The energy saving zone for a α = 3, Eµrx1 = 0.4, and different values of β ,
plotted as per inequality (5.4). . . . . . . . . . . . . . . . . . . . . . . . . 203
5.3 The energy saving zone for a path-loss exponent α = 3, β = 1, and different
values of Eµrx1 , plotted as per inequality (5.4). . . . . . . . . . . . . . . . . . 204
5.4 Different realizations of energy saving zones in (5.4) corresponding to var-
ious values of the path-loss exponent α, notice the consistency with the
elliptical approximation in (5.10). . . . . . . . . . . . . . . . . . . . . . . 204
5.5 The relative error in the area of the energy saving zone when approximated
by a perfect ellipse. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.6 Approximating to an ellipse by matching the vertices. . . . . . . . . . . . . 208
5.7 Equi-energy contours (5.13) for α = 3, also showing the scaled approxi-
mated ellipses (5.15). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
List of Figures xvi

5.8 The expected value of the energy saving gain (η) for a fixed number of
relays within the saving zone, as per equation (5.29). . . . . . . . . . . . . 213
5.9 Normalized analytical benchmarking between the three scenarios. . . . . . 218
5.10 Saving distance threshold with respect to the relative density of relays, plot-
ted as equation (5.43). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
5.11 The rate of change in the saving distance threshold with respect to the rel-
ative density of the assisting relays. Obtained by simulation and compared
to equation (5.45). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5.12 Cellular layout indicating simulated calls via the cellular base stations. . . . 222
5.13 Monte-Carlo simulation comparison with the theoretical equation plot of
the distance threshold (5.43). . . . . . . . . . . . . . . . . . . . . . . . . . 223
5.14 Normalized saving distance threshold (θ ) obtained for different values of α
using Monte-Carlo simulation. . . . . . . . . . . . . . . . . . . . . . . . . 224
5.15 Energy cost obtained using Monte-Carlo simulation as compared to the an-
alytical plots of equations (5.39). . . . . . . . . . . . . . . . . . . . . . . . 225
5.16 Achievable gains in the saving distance threshold for an example practi-
cal value of κ = 12, as compared to the theoretical asymptote expressed in
equation (5.46). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
5.17 The layout of an aerial-backhauled network, with Matérn clustering. . . . . 229
5.18 Voronoi tessellations representing the service area of the cluster head. . . . 230
5.19 Fitting the probability density function of the contact distance R between
cluster members and their serving cluster heads, for standard λ = 1. . . . . 232
5.20 The energy-reduction factor ξ for different values of the compression factor
κ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
5.21 The normalized energy difference η for different values of the compression
factor κ, showing Matérn hard-core case, LEACH, HEED and the algorithm
presented in [5]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
5.22 Clustered wireless sensor network using Matérn hard-core point process,
showing the hard core region i.e. b(x, δ ), where no two cluster heads are
allowed within. The leftover points constitute the complementary process
CMHC (cluster member). λ = 1, δ = 2. . . . . . . . . . . . . . . . . . . . 241
5.23 The intersection of the balls b(x1 , δ ) and b(x2 , δ ). . . . . . . . . . . . . . . 244
5.24 The conditional probability of the nth neighbour of a CMHC point to migrate
to the MHC process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
List of Figures xvii

5.25 Pair correlation function of the MHC process and its complement CMHC,
gCMHC (r) and gMHC (r) respectively. . . . . . . . . . . . . . . . . . . . . . 249
5.26 The asymptotic behaviour of the CMHC pair correlation function
gCMHC (r < δ ) formulated in (5.91). . . . . . . . . . . . . . . . . . . . . . 251
5.27 The simulation of the contact distance distribution (normalized histogram),
compared with the empirical formula in (5.94). . . . . . . . . . . . . . . . 252
5.28 The Voronoi tessellation of the three scenarios having equal nodes density
(equal cluster heads density and equal cluster members density), indicating
an increase of regularity from left (PPP) to right (triangular lattice). . . . . 252
5.29 The average energy spent by a cluster member to contact its cluster head,
comparing the three geometrical scenarios with the same cluster heads den-
sity, and for a path-loss exponent α = 3. . . . . . . . . . . . . . . . . . . . 254
Abstract

Perhaps future historians will only refer to this era as the information age, and will recog-
nize it as a paramount milestone in mankind progress. One of the main pillars of this age is
the ability to transmit and communicate information effectively and reliably, where wireless
radio technology became one of the most vital enablers for such communication. A growth
in radio communication demand is notably accelerating in a never-resting pace, pausing a
great challenge not only on service providers but also on researches and innovators to ex-
plore out-of-the-box technologies. These challenges are mainly related to providing faster
data communication over seamless, reliable and cost efficient wireless network, given the
limited availability of physical radio resources, and taking into consideration the environ-
mental impact caused by the increasing energy consumption.
Traditional wireless communication is usually deployed in a cellular manner, where
fixed base stations coordinate radio resources and play the role of an intermediate data han-
dler. The concept of cellular networks and hotspots is widely adopted as the current stable
scheme of wireless communication. However in many situations this fixed infrastructure
could be impaired with severe damages caused by natural disasters, or could suffer conges-
tions and traffic blockage. In addition to the fact that in the current networks any mobile-to-
mobile data sessions should pass through the serving base station that might cause unnec-
essary energy consumption. In order to enhance the performance and reliability of future
wireless networks and to reduce its environmental footprint, we explore two complementary
concepts: the first is node cooperation and the second is aerial networks.
With the ability of wireless nodes to cooperate lays two main possible opportunities;
one is the ability of the direct delivery of information between the communicating nodes
without relaying traffic through the serving base station, thus reducing energy consumption
and alleviating traffic congestion. A second opportunity would be that one of the nodes
helps a farther one by relaying its traffic towards the base station, thus extending network
coverage and reliability. Both schemes can introduce significant energy saving and can
enhance the overall availability of wireless networks in case of natural disasters. In addition
to node cooperation, a complementary technology to explore is the aerial networks where

1
Abstract 2

base stations are airborne on aerial platforms such as airships, UAVs or blimps. Aerial
networks can provide a rapidly deployable coverage for remote areas or regions afflicted
by natural disasters or even to patch surge traffic demand in public events. Where node
cooperation can be implemented to complement both regular terrestrial coverage and to
complement aerial networks.
In this research, we explore these two complementary technologies, from both an ex-
perimental approach and from an analytic approach. From the experimental perspective we
shed the light on the radio channel properties that is hosting terrestrial node cooperation
and air-to-ground communication, namely we utilize both simulation results and practical
measurements to formulate radio propagation models for device-to-device communication
and for air-to-ground links. Furthermore we investigate radio spectrum availability for node
cooperation in different urban environment, by conductive extensive mobile measurement
survey. Within the experimental approach, we also investigate a novel concept of tempo-
rary cognitive femtocell network as an applied solution for public safety communication
networks during the aftermath of a natural disaster.
While from the analytical perspective, we utilize mathematical tools from stochastic ge-
ometry to formulate novel analytical methodologies, explaining some of the most important
theoretical boundaries of the achievable enhancements in network performance promised
by node cooperation. We start by determining the estimated coverage and rate received by
mobile users from convectional cellular networks and from aerial platforms. After that we
optimize this coverage and rate ensuring that relay nodes and users can fully exploit their
coverage efficiently. We continue by analytically quantifying the cellular network perfor-
mance during massive infrastructure failure, where some nodes play the role of low-power
relays forming multi-hop communication links to assist farther nodes outside the reach of
the healthy network coverage. In addition, we lay a mathematical framework for estimat-
ing the energy saving of a mediating relay assisting a pair of wireless devices, where we
derive closed-form expressions for describing the geometrical zone where relaying is en-
ergy efficient. Furthermore, we introduce a novel analytic approach in analyzing the energy
consumption of aerial-backhauled wireless nodes on ground fields through the assistance
of an aerial base station, the novel mathematical framework is based on Matérn hard-core
point process. Then we shed the light on the points interacting of these point processes
quantifying their main properties.
Throughout this thesis we relay on verifying the analytic results and formulas against
computer simulations using Monte-Carlo analysis. We also present practical numerical ex-
amples to reflect the usefulness of the presented methodologies and results in real life sce-
Abstract 3

narios.
Most of the work presented in this dissertation was published in-part or as a whole in
highly ranked peer-reviewed journals, conference proceedings, book chapters, or otherwise
currently undergoing a review process. These publications are highlighted and identified in
the course of this thesis. Finally, we wish the reader to enjoy exploring the journey of this
thesis, and hope it will add more understanding to the promising new technologies of aerial
networks and node cooperation.
Chapter 1

Introduction

Node cooperation and aerial networks are two integral promising technologies that have
gained a recent interest from researchers and industry alike. These technologies are antici-
pated to contribute in solving the tough challenges urged by the overwhelming increase in
wireless network traffic and by the souring number of mobile broadband subscriptions. The
expected growth in mobile traffic alone is around 40% annually as estimated by many of the
main telecom players [6–8], adding a total of 5.4 billion mobile broadband subscribers from
2014 to 2020. Some of the main challenges that are associated with this massive growth are
captured in the following:

• The urgent need for a rapid and reliable backup of wireless infrastructure in case of a
natural disaster, due to the increased dependency on this infrastructure.

• The limited availability of the radio spectrum, as this natural resource is quite scarce
especial in the favoured bands below 6 GHz.

• The emergence of new types of users utilizing broadband wireless communication


such as public safety agencies requiring much higher reliability as their communica-
tion is of a mission-critical nature.

• The increasing interference between network elements due to the dense deployment
of base stations, and due to the increasing density of user equipments.

• The need for cost efficient planning and optimization techniques of wireless infras-
tructure as the market competition is brutal between Telecom operators.

• The need for cost efficient alternative of conventional infrastructure-based service.

4
1. Introduction 5

• Reducing the environmental footprint of the growing network especially in terms of


the energy consumption of both base stations and user equipments.

Many of creative new technologies and concepts are suggested to tackle these chal-
lenges such as; dynamic spectrum access, cognitive radio technologies, small-cells, massive
MIMO, millimetre-wave technology in addition to node cooperation and aerial networks
technology. In dynamic spectrum access [9–11] the allocation of spectrum is resiliently
shared between different systems so that spectrum holes (or opportunities) are efficiently uti-
lized, this key concept is exploited by cognitive radios [12, 13] that actively learn their radio
environment and hence can opportunisticly access radio resources in an autonomous man-
ner without a continuous coordination with the primary user of the spectrum. Small-cells
on the other hand are deployed by licensed telecom operators and are primarily facilitated
by the recent advanced in optical fibre wired infrastructure [14], due to the vast deployment
scale of such cells [15], operators resort to distributed and self-organization techniques [16]
in order to optimize the performance. While massive MIMO technology [17, 18] is based
on the concept of pinpointing electromagnetic energy towards the location of users, utilising
a very large number of radiating antenna elements aiming to achieve a spatial orthogonality
between concurrent radio channels thus increasing the user throughput in a given spectral
band. Thinking out of the box, researchers and industry are currently exploring the use
of Millimetre-wave spectrum band that long was considered as lossy and technologically
channelling [19], the vast vacancy in this band provokes the apatite of telecom operators
and vendors to explore different techniques and tools to utilize this empty spectrum, facili-
tated by advances in microelectronic technologies allowing the piratical exploitation of this
golden band that also favour small radio components and antennas, allowing the realization
massive-MIMO concept in millimetre-wave [20, 21].
On the other hand, in this thesis we focus on node cooperation and aerial networks
technologies aiming to address the aforementioned challenges, where in the following two
sections we provide the necessary background on these two technologies and shed the light
on their most recent advances:

1.1 Node Cooperation Technology and D2D


Node cooperation [22–26, 26–32] is a promising technology that has gained a great interest
from researchers and industry alike. In contrary to the legacy centralized communication
systems, node cooperation concept allows peers to directly communicate together; either for
1. Introduction 6

relaying and assisting cellular traffic or to establish local communication links. The benefits
that can be harvested from such technology are numerous, listing the most important:

1. First benefit is the proximity gain [22, 23, 26], since short distances allow higher
throughput and lower power consumption.

2. Secondly it can enhance cell edge throughput and coverage by performing aggregation
and forwarding of traffic to farther user equipments outside the reach of the base
station [24, 26, 30].

3. The third benefit is the energy saving, since the communication distance between
nodes and base stations can be further broken down into shorter hops, where the total
energy consumption is less that a single long hop [26, 31, 33].

4. Backup means of communication in case of natural disaster, when fixed infrastructure


is severally affected and con not handle and route mobile traffic [32, 34, 35].

In some situations, node cooperation becomes a vital feature that is irreplaceable by any
other means, such situations include the recovery of a cellular network during the after-
math of a natural disaster, where it is expected that the Telecom infrastructure to be severely
affected [32, 36–38]. Other important situations where the network coverage is poor, or
within areas that are completely out of coverage. In fact the lack of node cooperation com-
munication in Long Term Evaluation (LTE) Release 8 and 10 [39], was one of the major
issues that was inhibiting public safety from the wide adoption of LTE technology, due to
the concerns regarding cellular network performance during natural disasters or in remote
and at cell edge areas. The third generation partnership project (3GPP) which is the main
standardization body of LTE, is actively working to develop node cooperation into the stan-
dard. The technology that enables node cooperation in LTE (3GPP) standards is the Device
to Device (D2D) technology, which has been recently introduced into the standard [23, 25].
The previously mentioned benefits of node cooperation are having several interesting
potentials in mobile applications, categorized as the following:

Local Traffic Applications When two nearby devices opt to engage in direct communica-
tion link for applications such as voice/video calls, file sharing, etc... This type of
traffic is very intense inside large offices environments, museums, shopping malls,
and public safety group communication [22, 24, 26].

Coverage Extension whether to provide basic wireless service to black spots or to en-
hance the Cell-edge throughput by allowing devices to relay traffic towards the fixed
1. Introduction 7

infrastructure. In this scenario, a D2D-eneble device is acting as a simple relay node


[23, 25, 32, 40].
Node cooperation can form a temporary ad-hoc network enabling basic communications
services. This particular application is very essential for Public Safety (PS) communication
systems, since that first responders are often operating in areas where the main network is
unreachable, and they are alternatively resorting to Direct Mode of Operation (DMO) in
TETRA system [41] for their team and group calls. DMO has been for long time available
in the second and first generation mobile communication.

Fig. 1.1 D2D communication scenarios by 3GPP.

As stated previously, node cooperation technology allows peer-to-peer communication


amongst mobile users without the need for a cellular base station nor a switching core to
handle and route the traffic, thus boosting the efficiency, speed and reliability of wireless
networks. These enhancements are the results of devices proximity, and their ability to
aggregate and forward the traffic, in addition to their highly efficient spectrum utilization.
Also it was indicated that the D2D is the standardized form of node cooperation, where re-
searchers and standardization bodies have addressed similar concepts to D2D for enabling
direct communication, like Bluetooth and WLAN. However, the advantages of utilizing
D2D technology as controlled/semi-controlled by wireless network provider can add un-
matchable reliability and control [24]. On the other hand, the implementation of D2D tech-
nology involves numerous challenges, for example the authors in [27], emphasize the role
1. Introduction 8

of interference management, otherwise (if interference is not managed properly) the cellular
network throughput and its quality will be severely affected when using D2D. This inter-
ference might result from either between the devices themselves (co-tier interference) or
between D2D devices and the main cellular network (cross-tier interference). In addition to
the interference issues, there are plenty of challenges related to the implementation of D2D
technology, listing below some of these challenges [22, 24–29]:

• Physical channel modelling for D2D communication

• Coverage and rate estimation and optimization towards relay nodes

• Analytical estimation of coverage enhancement using node cooperation

• Quantifying energy efficiency when using node cooperation

• Spectrum availability for D2D communication

• Adaptive and efficient Resource allocation

• Mobility Management and Handover

• Synchronization Issues

• Power control and interference mitigation

• Backward compatibility with previous releases

Due to the wide research paradigm of node cooperation, we opt to focus our research on
the first five challenge areas as listed previously, that are considered the most important en-
ablers for node cooperation, where we identify the literature gaps and formulate our research
questions accordingly.
In fact some of these challenges were highlighted in the initial standardization document
by the Third Generation Partnership (3GPP) under the LTE-A Release 12 [25]. In this
document, 3GPP is mainly setting a common evaluation guidelines including some basic
channel models for LTE D2D proximity services, and defining some of the communication
scenarios that can occur under the umbrella of D2D technology, In Fig. 1.1 the four defined
scenarios (1-A, 1-B, 1-C and 1-D) by 3GPP are all limited to one hop of transmission. As
an initial standardization step, this can be considered as a remarkable development in the
LTE standard. However, several other useful scenarios could leverage D2D communication;
such the cases of two-hop or multi-hop scenarios, as indicated in Fig. 1.2 (scenarios 2 and
1. Introduction 9

3). In these scenarios a helper UE (or relay) indicated as UEr can play the role of a simple
decode and forward relay node [42], thus (scenario II) extending the network coverage and
cell edge performance or (scenario III) facilitating the communication between two UEs
engaged in a D2D link. The former concept has been briefly studied in [43].

Fig. 1.2 Possible future D2D communication scenarios.

1.2 Aerial Networks Technology


Aerial networks have recently gained a wide popularity as key enablers for rapid deploy-
able wireless networks where coverage is provided by on-board radio heads [36, 44–50].
Aerial networks are capable of delivering essential wireless communications for public
safety agencies during the aftermath of natural disasters [51, 52], or to provide a rapid
coverage patch in remote areas out of reach of the main public safety network or the com-
mercial cellular networks. A concept has been recently endorsed by the Homeland Security
Bureau in USA as the Deployable Aerial Communications Architecture [53] envisioning
the recovery of critical communications for first responders within 12-18 hours of the disas-
ter. An example of related development efforts is the European Commission funded project
ABSOLUTE (Aerial Base Stations with Opportunistic Links for Unexpected and Tempo-
rary Events) [4] focusing on Low Altitude Platforms (LAP) that are increasingly believed to
be as one of the ultimate answers for emergency networks. In Fig. 1.3 we depict the concept
1. Introduction 10

of aerial networks supplementing damaged terrestrial cellular infrastructure, where we also


show how node cooperation (indicated as relays) can extend the coverage of both aerial and
terrestrial base stations to farther users that are out-of-reach of the main coverage.

Fig. 1.3 Aerial networks supporting terrestrial cellular coverage.

Aerial networks can play an important role in supporting critical communication and in
supporting commercial telecom operators as elaborated in the following:

Public Safety, and Critical Communications An accelerating pace of emerging new


broadband critical applications, is driven by Public Safety (PS) organizations. PS are
actively involved in the developments of mission-critical requirements of the Long
Term Evolution-Advanced (LTE-A) communication system. In fact LTE has been en-
dorsed by US major PS organizations as the technology of choice for the 700 MHz
band [54], and accordingly a 20 MHz spectrum bandwidth chunk has been allocated
for this purpose [40]. Mission-critical communication for applications such as dig-
ital dispatcher, first responder deployments, fire fighters missions and police forces
operations might sometimes occur in off-network locations or during the aftermath
of natural disasters, where the main radio access network (RAN) could be partially
damaged, or out of reach.
1. Introduction 11

Commercial Cellular Networks Telecom operator can opt to deploy aerial networks in
remote arias as temporary solution for patching coverage gaps [46]. Many recent
ambitious projects are aiming to provide a regional coverage through high altitude
platforms, providing internet access to remote communities. An example of such
projects is Google loon [55]. Emphasizing that this thesis is focusing on low altitude
platforms with some work applicable to high altitude platform as well.

Aerial platforms can be categorized according to their altitude capability, where plat-
forms operating above the stratosphere are referred as high altitude platforms (HAPs)
[44, 45], mostly ranging in 17 km to 22 km. On the other hand low altitude platforms
operating in the troposphere and are usually less technologically challenging. Another cat-
egorizing of aerial platforms is according to their mean of lift where some utilize lighter-
than-air gases to float such as airships, balloons and blimps having slow and limited mo-
bility. On the other hand some platforms utilize active thrust such as UAVs and helicopter
giving them high level of maneuverability and mobility. In this thesis we focus into low
altitude platforms with limited mobility. Aside from the engineering challenges involved
in the physical construction of a reliable and cost effective aerial network, there are several
research challenges related to the network performance. We summarize this challenges in
the list below:

• Air-to-ground channel modelling for low altitude platforms

• The analytical estimation of the coverage of aerial platforms

• The analytical optimization of aerial platform altitude

• Spectrum availability for aerial networks

• Co-exitance with terrestrial cellular networks

• Interference mitigation and analysis

• Network architecture and design

• Platform mobility effect on coverage

In this thesis we limit our research scope to the first four challenge areas, where we identify
literature gaps and formulate our research questions accordingly.
1. Introduction 12

1.3 Summary of Literature Gaps


In this section we summarize the main literature gaps within the challenge areas as listed
previously under both aerial networks and node cooperation, the aim of this summary is to
develop and formulate the research questions. Detailed literature review and methodology
comparison follows within each of the specific chapters (2,3,4,5) allowing more focused
and insight analysis of each of the research questions topic. The challenges listed under
aerial networks and node cooperation are investigated in the state-of-the-art literature were
we identified research gaps as the following:

1.3.1 Channel Modelling for Aerial Networks and D2D


Radio propagations models varies from the simple path-loss exponent models, up to Clusters
Delay Lines (CDL) that extend the famous Tapped Delay Line (TDL) model in the spatial
domain [56, 57]. Some of these models are even endorsed by the International Telecom-
munication Union-Recommendation section (ITU-R) in order to provide a common and
fare evaluation environment for equipment vendors and researchers for simulating wireless
communication technologies [57]. This ITU-R document [57] source most of its mathe-
matical models from the extensive channel modelling research project “WINNER” [58].
ITU-R models cover variant propagation scenarios such as indoor, indoor-to-outdoor, urban
micro-cell, outdoor-to-indoor, suburban macro-cell, urban macro-cell and rural macro-cell.
However, two main drawbacks in this model when trying to apply it to typical D2D sce-
narios, (i) the first, it does not concentrate on D2D specific environment, such as the low
transmitter and receiver heights. (ii) The second, the modelled frequencies are limited be-
tween 2 and 6 GHz, understanding that most of today’s wireless communications happen
within this band. On the other hand, D2D communication can certainly appear in other
frequency bands. In fact, recent researches in [19, 59, 60] suggest the feasibility of cellular
communication in Millimetric bands (mmWaves). Communicating using mmWaves band
is a challenging task due to the losses by atmospheric absorption peaking near certain fre-
quencies such as 24 GHz, 60 GHz and 120 GHz [61], also the very high shadowing loss
and sharp diffraction behaviour. While on the other hand, the benefits that could be brought
up by utilizing this band are substantial, since mmWaves band is potentially underutilized,
allowing a wide band spectrum allocation for D2D service.
Referring to the proposed scenario in Fig. 1.3 we recall that in addition to D2D (or node
cooperation) there is another physical media that need to be carefully studies, this media is
the air-to-ground (AtG) communication. As we are proposing an integrated terrestrial-aerial
1. Introduction 13

network to assisting damaged cellular network. Understanding the propagation behaviour of


the AtG media is an essential step for further research in this aria, since it allows the proper
estimation of the key performance indicators using both simulation approaches and ana-
lytic approaches. The main issue here is that AtG channel largely differ from conventional
channel models that for long being used for terrestrial communications [62] such as cellular
macro cell, broadcasting, cellular indoor, etc... Few are the available models that address
AtG channel in careful analysis, the most common methods is to (i) either derive a simula-
tion based model based on a certain city layout [63, 64], such as the work in [63] focusing on
Bristol city in UK, or (ii) otherwise to conduct practical measurements for AtG coverage,
such approach is quite common for Land-Mobile-Satellite (LMS) where a mobile station
(or a mobile user) could freely roam in an urban environment, rather than using a fixed and
highly directional terminals. This capability comes at the cost of heavy radio losses due
to the non-line-of-sight nature of the mobile link [65] especially when compared to legacy
immobile VSAT systems. In the case of a LMS system, the service can be provided by
a geostationary satellite (e.g Thuraya), or a constellation of satellites at low/medium earth
orbits (e.g. Iridium, Globalstar) [66]. Due to the specific nature of the satellite system itself,
the previous studies on LMS radio channel largely depends on the system-specific nature
[67–69] of the particular satellite under study. A generic path-loss modelling attempt of the
AtG link is presented in [70] where it primarily addresses high-altitude-platforms utilizing
tools from the uniform theory of diffraction [71]. Thus, the the main gap here is that their
is no AtG model for LAPs that combines simplicity and generalization, also that can be
repeatedly and independently verified.
Under the topic of the physical understanding of the communication media, it is impor-
tant to understand the current spectrum occupancy in order to utilize spectrum opportunities
and holes for both aerial networks and terrestrial node cooperation. As spectrum utiliza-
tion varies with respect to the geographic location, several studies around the world has
been conducted such as in Germany, Netherland, Singapore, Spain, New Zealand and China
[72–76], however no field measurements have been collected in Melbourne metropolitan
area.
In the light of the previous literature summary, we identify below the main gaps in this
area that we focus on:

• The lack of device-to-device radio propagation model in sub 6Ghz spectrum

• The lack of device-to-device radio propagation model in mmwave spectrum

• The lack of air-to-ground propagation model for low altitude platforms that is both
1. Introduction 14

generic and easy to use

• The lack of spectrum occupance measurement in Melbourne metropolitan area

1.3.2 Performance Estimation and Optimization of Cellular and


Aerial Networks
The analytical estimation of the performance of the cellular and aerial networks is quite
essential for the proper strategic planning of the networks parameters. It is important to
quantify the expected end-user experience as based on certain network parameters. Surely
the most accurate way to do so, is to tune these parameters on live networks, however this
might cause unexpected degradation impact on the performance and productivity (loss of
profit) and it is usually time and resource consuming. The industry norm of estimating the
key performance indicators (KPIs) is achieved by running case specific extensive simula-
tions. A more efficient way is to analytically estimate these KPIs based on mathematical
models that can capture to a good degree of accuracy the expected performance even before
the actual deployment of the networks. Many types of KPIs affect the end-user experience
of which we limit our study to the most important KPIs, namely the coverage probability of
the network and the throughput.
The recent advances in the field of stochastic geometry [77–84] allowed an extended
modelling for the different parameters affecting the network performance such as the density
of base stations [3] the effect of multi tier base stations such as macro-cells, micro-cells and
femto-cells [85], or the interference mitigation parameters [86, 87]. The most common
assumption in the literature is to use Poisson point process to model the distribution of
base stations, very few literature papers take the effect of the mutual repulsion of base
stations such as the work in [88, 89]. Another very common assumption is to use Rayleigh
propagation fading model to capture the random effect in the radio channel, however very
few works realised this gap and tried to model the performance of a generic channel fading
such as the work in [90] which obtains the expected cellular network throughput without
the need to obtain the coverage probability utilizing a novel method based on the moment
generating function (MGF). Stochastic geometry analysis is applicable to terrestrial cellular
network and other scenarios that have a vast number of base stations, however when it
comes to the performance estimation of aerial networks we can’t find a common analytic
evaluation methodology that is followed in the literature, but rather some simulation based
works [91, 92]. One of the significant benefits of the analytical estimation of aerial coverage
is the ability to optimize the performance by fine-tuning the key parameters that affects this
1. Introduction 15

performance such as the altitude of the platform. The main influencer on the selection
of the optimum altitude is the likelihood model of having a line-of-sight (LoS) condition
between the arial platform and terrestrial users. A geometric approach is followed in [93]
to determine the theoretical likelihood of the LoS in built-up areas, however the study was
not based on the ITU statistical parameters [94]. On the other hand, practical measurements
were presented in [95] for satellite to ground LoS estimations. Overall, there is a gap in the
literature of how to determine the optimum altitude of an aerial platform based on analytical
methods. The main gaps that we focus on in this thesis are identified in the following list:

• The lack of an analytical approach in determining the expectancy of the coverage and
rate of cellular networks based on generic channel fading conditions.

• The lack of an analytical methodology for calculating the performance of an aerial


platform.

• The lack of an analytical optimization approach for selecting the best altitude of an
aerial platform

1.3.3 Performance Enhancement with Node Cooperation


It is quite important to quantify the introduced impacts by node cooperation on the cellu-
lar network performance. These impacts might potentially harm the cellular network if the
deployment strategies and techniques are not adequately quantified either by extensive sim-
ulations, or by deriving stochastic spatial models. D2D cannot be easily inculcated in the
current 4th generation (4G) cellular system due to several challenges such as interference
and resource allocation since the current 4G utilizes the Orthogonal Frequency Multiple
Access (OFDM) technology which allows a frequency-band reuse factor of unity [96–98],
meaning that all neighbouring cells can operate on the same carrier frequency [99]. The
introduction of D2D within the same frequency band poses a great challenge in manag-
ing cross-tier interference between the cellular network and the underlying D2D [28, 100].
Since D2D communication depends on fully distributed terminals, a better way to explore
the impact of this technology is by utilizing stochastic geometry [101, 102] that study the
average behaviour of network performance where nodes are positioned according to a cer-
tain probabilistic distribution. The stochastic geometry approach is much more effective
in analysing modern communication networks, as the conventional hexagonal cellular lay-
out is not applicable in most of the actual deployments. For example, a remarkable work
has been done on characterizing the interference in wireless communication networks using
1. Introduction 16

stochastic geometry in [103], however the study does not address D2D scenarios in particu-
lar. The authors in [85] utilizes stochastic geometry to mathematically express the Signal to
Interference and Noise Ratio (SINR) in a multi-tier cellular network that are facilitated by
several types of base stations (Macro, Femto, Pico). However, this work did not take into
consideration the particular conditions of D2D. By properly quantifying and understanding
the interference effect of D2D on the cellular network, we can then exploit this technology
to expand the cellular coverage [104, 105] and capacity [106] by mitigating the identified
interference. Both cross-tier and co-tier interference have a random nature both in time and
space, and is difficult to quantify the effect of D2D on the cellular network without extensive
simulations or involving advanced methods from stochastic geometry, an approach that is
not deeply investigated by researches for the D2D application yet. For example the authors
in [28] utilizes stochastic geometry to analyse UE-to-UE interference in D2D enabled net-
works, focusing only on Rayleigh fading channel model, without a comprehensive cross-tier
and co-tier interference study that involved different realistic cellular channel models.
Node cooperation and relaying can also play a potential role in enhancing the coverage
and rate of aerial networks, as airborne base stations are limited in power due to the me-
chanical limitations of aerial platforms that have a tight weight payload budget [45]. In this
scenario, some terrestrial nodes (user equipments) can play the role of relays and form D2D
links with farther nodes [36] thus expanding the coverage and providing better quality of
service. Many works consider the enhancement effects of D2D and relays in conventional
cellular communication [28, 107, 108], however very few are targeting relay and D2D ap-
plications underlaying aerial networks such as the analytical work in [48]. The summary of
gaps that we focus on in this thesis are listed below:

• The lack of the analytical modeling of the performance enhancement of single and
multihop D2D as a disaster alleviation solution

• The lack of the practical analysis of node cooperation under the specific conditions of
natural disaster

• The lack of the practical analysis of the effect of terrestrial relays underlaying aerial
networks.

1.3.4 Energy Efficiency with Node Cooperation


Having efficient energy utilization in a wireless network becomes of an utmost significance
when terminals are energy-limited due to the reliance on batteries [109, 110]. Efficient
1. Introduction 17

battery-energy utilization can substantially increase the lifetime of the deployed terminals
[5, 111] and reduce the cross-tier/co-tier interference temperature generated by the commu-
nicating devices. One of the mechanisms of achieving such efficiency can be accomplished
by breaking long wireless distances into smaller multi-hop chains that are assisted by me-
diating devices (such as D2D relays). This topic is usually one of the focus research areas
of sensor networks, and has not been widely investigated for cellular communication net-
works. For example the authors in [112] are suggesting a packet routing scheme that tries
to balance energy consumption in the deployed network. Similarly the authors in [113] are
concluding that in order to maximize the network lifetime, a balanced energy routing should
be adopted. While in cellular networks, the research lean more towards the power control
of terminals that is to lessen the overall interference temperature of the communicating de-
vices, and accordingly achieve a better overall network throughput. Generally speaking
in cellular networks we can have two power control schemes: (i) In the first scheme, the
overlaying cellular network is managing the transmit power assignment of D2D devices,
aiming to reach an optimum power allocation, that enhance the overall cellular throughput
[114]. The (ii) second scheme is to allow D2D devices to adjust their power levels in an
autonomously manner. On the other hand Relay placement and selection is an important
issue in most of wireless networks, having applications in cellular systems, ad-hoc and in
sensor networks. Relay selection criteria varies according to the imposed constraints on
the network performance. Mainly, the selection optimization aims to enhance the following
network metrics: (i) energy-saving [115, 116] (ii) capacity-enhancement [117–119], and
(iii) QoS/connectivity of the nodes [120, 121]. Optimizing relay selection based on mul-
tiple criteria is also an open research topic. The general approaches in investigating relay
selection mechanisms and performance enhancement ranges from studying the fundamental
performance limits to studying the detailed application-specific protocols, where the con-
cept of having a relaying region is investigated in the literature as part of either exploring
the theoretical limits or to reduce the complexity of relay selection protocols [115].
Node cooperation can also introduce a significant energy saving to aerial communica-
tion [122] where ground relays can help to breakdown a lengthy AtG link into two shorter
ones or a ground relays can aggregate the traffic from several terrestrial nodes then perform
compression and upload to the aerial platform. The main gaps that we focus on in this thesis
are listed below:

• The lack of an analytic approach to quantify the energy saving that can be introduced
by node cooperation to the cellular communication.
1. Introduction 18

• The lack of the analytic approach to model the energy consumption and thus to model
the saving introduced when using node cooperation to assist aerial networks.

1.4 Research Questions and Contributions

1.4.1 Research Question 1 and Contributions


What is the Physical Channel Properties of AtG and D2D links?

Under this research question, we investigate the physical channel characteristics required
for node cooperation, mainly focusing on the path-loss and the large scale shadowing (slow
fading). The investigation is conducted under three different applications:

1. The first application residing under this research question is the air-to-ground (AtG)
media, which is considered as an essential channel for future wireless networks, and
for disaster relief aerial coverage, modelling AtG channel includes the investigation of
the statistical behaviour of radio signals in terms of the path loss and the shadowing.
The target frequency band is 400-6,000 MHz, and the target environments are urban
built up areas.

2. The second application related to this research question is focussing on investigating


the propagation properties of Millimetric wave band in different urban environments.
These properties include the statistical modelling of the expected path loss i.e. to
find a relation between the statistical parameters of the propagation and the urban
environment properties (as defined by ITU-R [94]). Furthermore, we investigate the
utilization of some techniques and methods that can overcome the specific propaga-
tion impairments in the mmWave band, techniques as beam forming.

3. The third application area is the current wireless networks were we focus on node
cooperation experimental channel measurements in realistic circumstances in differ-
ent urban environments. That is, the investigation of the path-loss exponent and the
standard deviation of the node cooperation channel.

4. The fourth application is to identify spectrum opportunities for node cooperation,


since spectrum scarcity is increasingly becoming an urgent issue, imposing a tough
impediment on both wireless service providers, and on regulatory authorities alike,
leading to an increasing spectrum fees and inhibiting wireless technology from cop-
ing with the growing users-demand. The main driver of this scarcity is the increasing
1. Introduction 19

vast reliance on wireless communication as a ubiquitous means of communication,


coupled with low utilization efficiency of the electromagnetic spectrum. Several mea-
surement campaigns observed a low spectrum utilization [72, 76], albeit the current
scarcity and the soaring spectrum fees. This under-utilization is mainly referred to
legacy wireless systems, where spectrum right-of-use is restricted to a single opera-
tor (or user) running a low efficiency wireless system. Alternatively, more efficient
spectrum utilization can be achieved by exploiting one or more of the radio resource
domains in a non-static manner, these domains span over frequency, time and space.

Contributions Under Research Question 1

The key deliverables by the author under this research question formed the following pub-
lished papers:

1. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, "Modeling Air-to-Ground


Path Loss for Low Altitude Platforms in Urban Environments", in IEEE Global Com-
munications Conference (GLOBECOM), Austin, USA, 2014, pp. 2898-2904. DOI:
10.1109/GLOCOM.2014.7037248 (Published) [123]
This main contribution here is a novel propagation model for air-to-ground coverage
in urban environments, the model quantifies the statistical parameters of the path-loss
and the shadowing. These parameters are obtained as a function of the elevation angle
and the ITU urban environment statistics [94].

2. Akram Al-Hourani, S. Chandrasekharan, and S. Kandeepan, "Path Loss Study for


Millimeter Wave Device-to-Device Communications in Urban Environment," in IEEE
International Conference on Communications (ICC), Sydney, Australia, 2014, pp.
102-107, DOI:10.1109/ICCW.2014.6881180 (Published) [124]
In this contribution we provide a simulation-based propagation model for D2D
mmWave in urban environments. Namely we determine the path-loss exponent (α),
and the shadowing standard deviation. These parameters are obtained as a function of
the distance and the ITU urban environment statistics [94].

3. Akram Al-Hourani, S. Chandrasekharan, G. Baldini, and S. Kandeepan, "Propa-


gation Measurements in 5.8GHz and Pathloss Study for CEN-DSRC," IEEE Inter-
national Conference on Connected Vehicles, Vienna, Austria, 2014, pp. 1086-1091,
DOI:10.1109/ICCVE.2014.7297518 (Published) [125]
1. Introduction 20

Under this contribution we conduct extensive field measurement for node cooperation
channel taking the effect of a standard vehicle on the path-loss. We determine the
path-loss exponent and the standard deviation of the shadowing in different urban
environments in the city of Melbourne.

4. Akram Al-Hourani, V. Trajkovic, S. Chandrasekharan, and S. Kandeepan, “Spec-


trum Occupancy Measurements for Different Urban Environments”, IEEE European
Conference on Networks and Communications, Paris, France, 2015, pp. 97-102,
DOI:10.1109/EuCNC.2015.7194048 (Published) [126]
In this contributions, we present the results of spectrum monitoring experiment con-
ducted in three different urban environments, characterized with different population
density and degree of urbanization, for the greater Melbourne area in Australia. The
spectrum occupancy is determined by comparing the received signal level to a certain
detection threshold based on the noise power of the measurement tool. The occupancy
results are categorized with respect to the surrounding population density as obtained
from the Australian Bureau of Statistics [2].

5. S. Chandrasekharan, Akram Al-Hourani, K. Magowe, L. Reynaud and S. Kan-


deepan, “Propagation Measurements for D2D in Rural Areas”, in International
Conference on Communications (ICC), London UK, IEEE, 2014, pp. 639-645,
DOI:10.1109/ICCW.2015.7247253 (Published) [1]
In this contribution we present field measurements for D2D path-loss and shadowing
standard deviation in rural environments in Victoria, Australia.

1.4.2 Research Question 2 and Contributions


What is The Expected Performance of Aerial and Terrestrial Networks without Node
Cooperation, and How to Optimize This Performance?

The locations of the deployed base stations, are usually constrained by many factors such as
economical, urban planning codes, and the availability of land/utility etc... these factors, are
very difficult to control and to predict, which leads to an increasing randomness in the BS
locations, where the theoretical hexagonal model is no longer feasible [3, 89]. Due to the
increasing complexity of cellular network, designers and researchers utilize simulation tools
for predicting network coverage and performance. Such approach is widely accepted in the
industry, however it cannot give an analytical insight of the influence contributed by the
1. Introduction 21

vast simulation parameters. Rather, it provides a detailed case-specific solution with neither
tractability nor flexibility. Analytical insight of network dynamics is an essential enabler for
strategic planning and long-term economical modelling [127]. Under this research question
we investigate this research question under two main application areas:

1. Current and future cellular networks, under generic channel fading conditions rather
than Rayleigh fading limited. The performance of these networks is quantified in
terms of the service success probability and in terms of the throughput measured in
bits per seconds.

2. Aerial networks coverage and throughput estimations, quantified similar to the previ-
ous application area.

3. Aerial network altitude optimization, in order to maximize the coverage plot of an


arial platform and exploit its best capabilities.

Contributions Under Research Question 2

The key deliverables of the author under this research question formed the following pub-
lished / under review papers:

1. Akram Al-Hourani, and S. Kandeepan,” On Modeling Coverage and Rate of


Random Cellular Networks under Generic Channel Fading“, Wireless Networks,
Springer, 2015. DOI: 10.1007/s11276-015-1114-x. (Published) [128]
This contribution lays a novel mathematical framework and provides a novel generic
formula (the coverage equation) for calculating the expected service success probabil-
ity (or the coverage probability) in random cellular networks, under general channel
fading conditions. This coverage equation is flexible to allow different fading models
for the serving signal from one side and the interfering signals from the other side.
We also provide a practical method to compute the expected rate in a random cellular
network, without resorting to complicated simulations.

2. Akram Al-Hourani, S. Kandeepan, Geoff Kaandorp , William Glenn and Abbas


Jamalipour, ”Coverage and Rate Analysis of Aerial Base Stations”, IEEE Wireless
Communications Letters (Submitted)
In this contribution we provide a novel analytic formula to determine the expected
service success probability given a certain signal to interference and noise (SINR)
1. Introduction 22

threshold. Also, we provide another formula that describes the expected through-
put. This approach is based on the air-to-ground model developed previously under
research question 1.

3. Akram Al-Hourani, S. Kandeepan, and S. Lardner "Optimal LAP Altitude for Max-
imum Coverage," IEEE Letters on Wireless Communications, vol.3, no.6, pp.569 -
572, Dec. 2014. DOI: 10.1109/LWC.2014.2342736. (Published) [129]
This contribution lays a novel mathematical framework and provides a novel opti-
mization methodology to optimize the altitude of an aerial platform in order to max-
imize the coverage patch area. This approach is based on the air-to-ground model
developed previously under research question 1.

4. Akram Al-Hourani, S. Chandrasekharan, and S. Kandeepan “Aerial Platforms for


Public Safety Networks, Performance Optimization” Wiley/ISRE, United Kingdom.
(In Press)
This chapter provides a broader range of contributions in maximizing the performance
of aerial networks under the context of public safety networks.

5. Karina Gomez, Akram Al-Hourani, L. Goratti, R. Riggio, S. Kandeepan and I. Bu-


caille, “Capacity Evaluation of Aerial LTE Base-Stations”, European Conference on
Networks and Communications, Paris, France, 2015 IEEE. (Published) [91]
This paper provide a simulation-based analysis of the expected throughput of aerial
platforms operating with LTE-Advanced systems, my contribution here was the chan-
nel model to analyse the performance.

1.4.3 Research Question 3 and Contributions


What is The Expected Enhancement of Aerial and Cellular Networks When Utilizing
Node Cooperation/Relaying?

The starting point in this research question is the quantification and the mathematical mod-
elling of the interference caused by cooperative nodes. A proper modelling of the interfer-
ence directly leads to the visualization of the SINR in the network, and to the modelling
of the expected enhancement using node cooperation/relaying. This research question is
investigated under three different applications areas:
1. The first application is the current 4th generation mobile systems, where D2D is ex-
pected to play a vital role in expanding network coverage by exploiting D2D devices
1. Introduction 23

as decode and forward nodes assisting farther UEs outside the coverage reach of the
RAN base stations. An important example application is to relief sudden disruption
in mobile networks caused by a natural disaster.

2. In the second application area, we investigate future wireless application scenarios for
efficient D2D deployments, namely the multihop D2D.

3. The third application area focuses on the Air-to-Ground communication media, where
investigations is continued on how to enhance aerial coverage using relay technology.

Contributions Under Research Question 3

The key deliverables by the author under this research question formed the following pub-
lished papers:

1. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, “Stochastic Geometry Study


on Device to Device Communication as a Disaster Relief Solution”, IEEE Trans-
actions of Vehicular Technology, vol.65, no.5, pp.3005 - 3017, May. 2016. DOI:
10.1109/TVT.2015.2450223. (Published) [130]
Under this contribution we investigate the alleviation that can be introduced by D2D to
the damage caused by a natural disaster affecting a cellular networks. We apply tools
from stochastic geometry in the special context of D2D communication to understand
the amount of damage that a cellular network can take before degrading below its
nominal performance level.

2. Akram Al-Hourani and S. Kandeepan, "Cognitive Relay Nodes for Airborne LTE
Emergency Networks," IEEE 7th International Conference on Signal Processing and
Communication Systems (ICSPCS), Gold Coast, Australia, 2013, pp. 1-9. DOI:
10.1109/ICSPCS.2013.6723940 (Published) [131]
In this contribution we present a novel algorithm that can be implemented in terrestrial
cognitive relay nodes in order to assist an overlaying aerial coverage. These relays
helps in expanding the coverage patch and enhance the users’ throughput.

3. Akram Al-Hourani and S. Kandeepan, "Temporary Cognitive Femtocell Network


for Public Safety LTE” IEEE 18th International Workshop on Computer Aided Mod-
eling and Design of Communication Links and Networks (CAMAD), Berlin, Ger-
many, 2013, pp. 190-195. DOI: 10.1109/ICSPCS.2013.6723940 (Published) [132]
1. Introduction 24

In this contribution we present a novel operational framework for enabling practi-


cal node cooperation underlaying an aerial platform, namely we propose a prac-
tical mechanism to deploy node cooperation under the specific properties of LTE-
Advanced

4. Akram Al-Hourani, S. Kandeepan, and S. Arunthavanathan, "Temporary Infrastruc-


ture with Cognitive Radios for Small Cell Networks," in the book of “Design and
Deployment of Small Cell Networks”, by Cambridge University Press, 2015, ISBN-
13: 9781107056718 (Published) [133]
In this contribution we provide the large picture of how temporary infrastructure can
be deployed in order to overcome the damaged occurred to cellular network, where
cognitive radios having self organization techniques are the central building blocks of
such scenario.

1.4.4 Research Question 4 and Contributions


What Is The Expected Energy Consumption Reduction When Utilizing Node Cooper-
ation/Relaying?

Under this research question we investigate both cellular and aerial networks, where node
cooperation / relaying can enhance network-level energy saving, by breaking long distance
links into smaller multi-hop links. The application areas under this research question are:

1. Future cellular systems where D2D can help a farther node to reduce its energy con-
sumption by breaking its lengthy node link towards the base station into two segments.

2. Aerial networks where terrestrial D2D-enabled nodes can form a clustered wireless
network, where a cluster head can play the role of aggregating the traffic generating
from multiple other nodes and then perform compression and upload this traffic to-
wards an aerial platform. This scheme can introduce significant energy saving in the
network.

Contributions Under Research Question 4

The key deliverables by the author under this research question formed the following pub-
lished / under review papers:
1. Introduction 25

1. Akram Al-Hourani, S. Kandeepan, and Ekram Hossain, “Relay-Assisted Device-to-


Device Communication: A Stochastic Analysis of Energy Saving”, IEEE Transac-
tions on Mobile Computing, DOI:10.1109/TMC.2016.2519343. (Accepted) [134]
Under this contribution we provide a novel mathematical framework to identify the
geometrical region where a relay can be energy-efficient if located inside this region
between a transmitter and a receiver. Furthermore, we identify the expected statis-
tical behaviour of the energy saving when utilizing a relay between this transmitter-
receiver pair. To quantify the energy reduction of node cooperation, we develop a
novel comparison methodology between conventional cellular communication and
D2D-enabled networks.

2. Akram Al-Hourani, Sathyanarayanan Chandrasekharan, Abbas Jamalipour, Lau-


rent Reynaud and Sithamparanathan Kandeepan, “Optimal Cluster Head Spacing for
Energy-Efficient Communication in Aerial-Backhauled Networks”, in Global Com-
munications Conference (GLOBECOM), 2015 IEEE. (Published).
In this contribution we provide a novel analytic method to calculate cluster head spac-
ing in aerial-backhauled networks in order to minimize energy consumption in terres-
trial nodes.

3. Akram Al-Hourani, Bill Moran and Sithamparanathan Kandeepan, “Regularizing


random points: Complimentary Matérn Point Process “, IEEE Transactions on Wire-
less Communications (Submitted)
In this contribution we deduce a novel mathematical tool to regularize random wire-
less nodes into energy-efficient clusters, we call this tool as the “complementary
Matérn hard-core point process” that can regularize randomly placed communication
nodes for best energy-reduction strategy.

1.5 Publications List


We list below the author’s publications conducted during the course of his PhD program
classified into four categories: book chapters, journal articles, conference proceedings and
technical reports. We also indicate the status of each of these publications as of the date of
writing this thesis.
1. Introduction 26

1.5.1 Book Chapters


1. Akram Al-Hourani, S. Kandeepan, and S. Arunthavanathan, "Temporary Infrastruc-
ture with Cognitive Radios for Small Cell Networks," in the book “Design and De-
ployment of Small Cell Networks”, by Cambridge University Press, 2015, ISBN-13:
9781107056718 (Published) [133]

2. Akram Al-Hourani, S. Chandrasekharan, and S. Kandeepan “Aerial Platforms for


Public Safety Networks, Performance Optimization” Wiley/ISTE, United Kingdom
(In Press)

1.5.2 Journal Articles


1. Akram Al-Hourani and S. Kandeepan, S. Lardner "Optimal LAP Altitude for Max-
imum Coverage," IEEE Letters on Wireless Communications, vol.3, no.6, pp.569 -
572, Dec. 2014. DOI: 10.1109/LWC.2014.2342736. (Published) [129]

2. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, “Stochastic Geometry Study


on Device to Device Communication as a Disaster Relief Solution”, IEEE Trans-
actions of Vehicular Technology, vol.65, no.5, pp.3005 - 3017, May. 2016. DOI:
10.1109/TVT.2015.2450223. (Published) [130]

3. Akram Al-Hourani, Sithamparanathan Kandeepan, and Ekram Hossain, “Relay-


Assisted Device-to-Device Communication: A Stochastic Analysis of Energy Sav-
ing”, IEEE Transactions on Mobile Computing, DOI:10.1109/TMC.2016.2519343.
(Accepted) [134]

4. Akram Al-Hourani, and S. Kandeepan, "On Modeling Coverage and Rate of


Random Cellular Networks under Generic Channel Fading", Wireless Networks,
Springer, 2015. DOI: 10.1007/s11276-015-1114-x. (Published) [128]

5. Akram Al-Hourani, Bill Moran and Sithamparanathan Kandeepan, “Regularizing


random points: Complimentary Matérn Point Process “, IEEE Transactions on Wire-
less Communications (Submitted)

6. Akram Al-Hourani, S. Chandrasekharan, Geoff Kaandorp , William Glenn and Ab-


bas Jamalipour, S. Kandeepan, "Coverage and Rate Analysis of Aerial Base Stations",
IEEE Transactions on Aerospace and Electronic Systems (Submitted)
1. Introduction 27

7. S. Chandrasekharan, K. Gomez, Akram Al-Hourani, S. Kandeepan, T. Rasheed,


L. Goratti, L. Reynaud, D. Grace, I, Bucaille, T. Wirth, and S. Allsopp "Designing
and Implementing Future Aerial Communication Networks", IEEE Communications
(Accepted)

1.5.3 Conference Proceedings


1. Akram Al-Hourani, S. Chandrasekharan, and S. Kandeepan, "Path Loss Study for
Millimeter Wave Device-to-Device Communications in Urban Environment" in Inter-
national Conference on Communications (ICC), Sydney, Australia, 2014 IEEE, pp.
102-107. DOI: 10.1109/CAMAD.2013.6708115 (Published)[124]

2. Akram Al-Hourani and S. Kandeepan, "Temporary Cognitive Femtocell Network


for Public Safety LTE” 18th International Workshop on Computer Aided Modeling
and Design of Communication Links and Networks (CAMAD), Berlin, Germany,
2013 IEEE, pp. 190-195. DOI: 10.1109/ICSPCS.2013.6723940 (Published) [132]

3. Akram Al-Hourani and S. Kandeepan, "Cognitive Relay Nodes for Airborne LTE
Emergency Networks," 7th International Conference on Signal Processing and Com-
munication Systems (ICSPCS), Gold Coast, Australia, 2013 IEEE, pp. 1-9. DOI:
10.1109/ICSPCS.2013.6723940 (Published) [131]

4. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, "Modeling Air-to-Ground


Path Loss for Low Altitude Platforms in Urban Environments" in Global Communi-
cations Conference (GLOBECOM), Austin, USA, 2014 IEEE, 2014, pp. 2898-2904.
DOI: 10.1109/GLOCOM.2014.7037248 (Published) [123]

5. Akram Al-Hourani, S. Chandrasekharan, G. Baldini, and S. Kandeepan, "Propa-


gation Measurements in 5.8GHz and Pathloss Study for CEN-DSRC," IEEE Inter-
national Conference on Connected Vehicles, Vienna, Austria, 2014, pp. 1086-1091,
DOI:10.1109/ICCVE.2014.7297518 (Published) [125]

6. Akram Al-Hourani, V. Trajkovic, S. Chandrasekharan, and S. Kandeepan, “Spec-


trum Occupancy Measurements for Different Urban Environments”, IEEE European
Conference on Networks and Communications, Paris, France, 2015, pp. 97-102,
DOI:10.1109/EuCNC.2015.7194048 (Published) [126]

7. Akram Al-Hourani, Sathyanarayanan Chandrasekharan, Abbas Jamalipour, Lau-


rent Reynaud and Sithamparanathan Kandeepan, “Optimal Cluster Head Spacing for
1. Introduction 28

Energy-Efficient Communication in Aerial-Backhauled Networks”, in Global Com-


munications Conference (GLOBECOM), 2015 IEEE. (Published).

8. S. Chandrasekharan, Akram Al-Hourani, K. Magowe, L. Reynaud and S. Kan-


deepan, “Propagation Measurements for D2D in Rural Areas”, in International Con-
ference on Communications (ICC), London UK, IEEE, 2014. (Published) [1]

9. I. Belski, Akram Al-Hourani, A. Valentine, A. Belski, "Can Simple Ideation


Techniques Enhance Idea Generation?", The 25th Annual Conference of the Aus-
tralasian Association for Engineering Education, Wellington, New Zealand; IEEE
2014. (Published) [135]

10. T. Chen, S. Kandeepan, Akram Al-Hourani, A. Munari, and M. Berioli, "Spectrum


sensing for cognitive maritime VHF networks," in OCEANS 2014 - TAIPEI, 2014
IEEE, pp. 1-7. DOI: 10.1109/OCEANS-TAIPEI.2014.6964419 (Published) [136]

11. Karina Gomez, Akram Al-Hourani, L. Goratti, R. Riggio, S. Kandeepan and I. Bu-
caille, “Capacity Evaluation of Aerial LTE Base-Stations”, European Conference on
Networks and Communications, Paris, France, 2015 IEEE. (Published) [91]

1.5.4 Technical Reports


1. Karina Gomez, et al.. , Kandeepan Sithamparanathan, Akram Al-Hourani. “FP7-
ICT-2011-8-318632-ABSOLUTE/D4.2.3 QoS Optimization Techniques: Final is-
sue”, FP-7 Project ABSOLUTE-Aerial Base Stations with Opportunistic Links for
Unexpected and Temporary Events [4]

2. Wei Jiang, et al.. Akram Al-Hourani “FP7-ICT-2011-8-318632-


ABSOLUTE/D3.2.2 Evaluation of Opportunistic Relaying for Disaster Relief
and Temporary Events”, FP-7 Project ABSOLUTE-Aerial Base Stations with
Opportunistic Links for Unexpected and Temporary Events [4]

1.6 Thesis Structure


This thesis is primely structured around the four research questions identified in Section
1.4, where the introduction presented in this chapter provide the essential background on
the fields of Node Cooperation and Aerial Networks identifying the main literature gaps in
1. Introduction 29

this area and formulating the four research questions. It also identifies the main contribu-
tions of the author. The rest of the thesis is structured as the following; in Chapter 2 we
address research question 1 where we provide our contribution in the physical channel char-
acterization of D2D links and AtG links. Followed by Chapter 3 that addresses research
question 2 by investigating the performance of both aerial networks and cellular networks,
providing novel mathematical tools to estimate and optimize this performance. Leading to
Chapter 4 that takes the effect of node cooperation in enhancing the performance of cellular
networks and aerial networks, this addresses research question 3. The last research ques-
tion 4 is addressed in Chapter 5, laying a novel mathematical methodology in analysing the
energy efficiency introduced by node cooperation in both cellular and aerial networks. Sum-
marizing in Chapter 6 we provide our conclusion remarks and future research paradigms.
Chapter 2

Channel Modelling, Measurements and


Experimentations

The essence of radio wireless communication is the ability of transmitting and receive elec-
tromagnetic energy in the radio spectrum. This signal is propagating in stochastic envi-
ronment where it encounters absorption, multiple reflections and diffractions causing signal
attenuation and producing multiple copies of the main signal. It is essential to understand
the statistical properties of the radio channel in order to estimate the received power level,
which will indicate the goodness of the wireless service. This chapter addresses research
question 1 through four key contributions: (i) D2D radio channel modeling in mmWave in
Sec. 2.1, (ii) cooperative radio channel modelling in for vehicular scenarios in Sec. 2.2, (iii)
Air-to-Ground radio channel modeling in Sec. 2.3, and (iv) determining the radio spectrum
utilization in Melbourne metropolitan area in Sec. 2.4. We depict in Fig 2.1 the highlight of
where this chapter is focusing, compared to the general layout presented previously in Fig
1.3.
In the recent years the millimeter wave spectrum is being explored as a prospective band
for the next generation (5G) cellular communications. In this chapter we study the propaga-
tion of the millimetre wave spectrum using ray-tracing model for an urban environment. We
consider the ISM bands in 24GHz and 61GHz in particular and conduct ray-tracing simula-
tions to study the path-loss behaviour in terms of the path-loss exponent and the shadowing
variance for both line of sight and non line of sight conditions. As a potential application we
examine the device to device (D2D) communication, which is currently being developed for
LTE-A standard. The resulting path-loss exponents and the shadowing variances are pre-
sented here based on ray-tracing simulations for an ITU-R statistical urban model, moreover
this chapter shows that intelligent beam steering can significantly improve the throughput

30
2. Aerial and D2D Physical Channel 31

Fig. 2.1 Aerial networks and node cooperation channel modeling.

of the considered D2D scenarios. For better exploring practical measurements for node
cooperation channel, we take the specific application area of Dedicated Short-Range Com-
munications (DSRC) technology in the 5.8GHz band, which is a key enabler to support a
safer and more efficient vehicular transportation in the near future. The chosen 5.8GHz fre-
quency band is therefore of a great interest for us to study the propagation of signals under
various environments. Thus, in this chapter we also investigate and record the propagation
effects of 5.8GHz radio signals in an urban and sub-urban like environments with experi-
ments conducted in the city of Melbourne for an infrastructure to vehicle (I2V) use case.
The experiments were conducted for a TX-RX separation from 20m to 150m with line of
sight conditions for the three chosen environments with and without the vehicle for com-
parisons as described in the chapter. Based on the measurements we estimate the path-loss
exponent and the shadowing standard deviation for the chosen scenarios. More interest-
ingly our results show a constant path-loss difference between the measurements with the
car when compared to the measurements without the car for different scenarios.
As emphasized previously, the reliable prediction of coverage footprint resulting from
an airborne wireless radio base station is at utmost importance, when it comes to the new
2. Aerial and D2D Physical Channel 32

emerging applications of air-to-ground wireless services. These applications include the


rapid recovery of damaged terrestrial wireless infrastructure due to a natural disaster, as
well as the fulfilment of sudden wireless traffic overload in certain spots due to massive
movement of crowds. In this chapter, we also propose a statistical propagation model for
predicting the air-to-ground path-loss between a low altitude platform and a terrestrial ter-
minal. The prediction is based on the urban environment properties, and is dependent on the
elevation angle between the terminal and the platform. The model shows that air-to-ground
path-loss is following two main propagation groups, characterised by two different path-loss
profiles. In this chapter we illustrate the methodology of which the model was deduced, as
well as we present the different path-loss profiles including the occurrence probability of
each.
This chapter is based on the following publications contributing towards sections 2.1,
2.2, 2.3 and 2.4 respectively:

1. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, "Modeling air-to-ground


path-loss for low altitude platforms in urban environments," in Global Communi-
cations Conference (GLOBECOM), Austin, USA, 2014 IEEE, 2014, pp. 2898-2904.
DOI: 10.1109/GLOCOM.2014.7037248 (Published).

2. Akram Al-Hourani, S. Chandrasekharan, and S. Kandeepan, "path-loss study for


millimeter wave device-to-device communications in urban environment," in In-
ternational Conference on Communications (ICC), Sydney, Australia, 2014 IEEE
(Published).

3. Akram Al-Hourani, S. Chandrasekharan, G. Baldini, and S. Kandeepan, "Propaga-


tion Measurements in 5.8GHz and path-loss Study for CEN-DSRC," IEEE Interna-
tional Conference on Connected Vehicles, Vienna, Austria, 2014 IEEE. (Published).

4. Akram Al-Hourani, V. Trajkovic, S. Chandrasekharan, and S. Kandeepan, “Spec-


trum Occupancy Measurements for Different Urban Environments”, European Con-
ference on Networks and Communications, Paris, France, 2015 IEEE. (Published).
2. Aerial and D2D Physical Channel 33

2.1 D2D Channel Modelling in mmWave


A recent report on key ICT facts and figures by the International Telecommunication Union
(ITU) [137] suggests that the year 2013 holds an interesting new event in which the number
of mobile-cellular subscriptions has almost equaled the population of the world. Since, the
mobile network is no longer limited to simple mobile phones but rather includes tablets and
other smart devices, there is an expectation that the mobile-cellular subscriptions will far
exceed the human population.
Also it is anticipated that in 2017 an overwhelming increase of mobile data traffic of
over than 7 times will occur [138], an alarming figure that is urging research centers and
standardization bodies to find new and creative methods in enhancing wireless networks
efficiency and performance. In fact, the Third Generation Partnership Project (3GPP) with
its recent Release 10 of the Long Term Evolution (LTE-Advanced) is trying to target these
challenges by introducing novel technologies such as Relay Nodes, Enhanced Intercell In-
terference Coordination (eICIC), Carrier Aggregation and several other improvements in the
Heterogeneous Network concept (HetNet). More advanced techniques such as Device-to-
Device (D2D) communication are still under development. Some of our previous work on
D2D assume common wireless channel models known in the literature but are not specific
to D2D [132][131].
The challenge of implementing D2D resides in mitigating the harm caused by interfer-
ence to the cellular spectrum by the communicating devices due to spectrum reuse. Espe-
cially with the recent advancements in the areas of cognitive radio related wireless com-
munications [12] the deployment of D2D is becoming more and more feasible. In order
to resolve this issue, we are examining the utilization of millimetric-waves (mmWaves) for
the D2D link as an outband means of communication, since mmWave spectrum has a num-
ber of key advantages such as spectrum availability, low interference and small antenna
dimensions that allows the implementation of highly directive antenna arrays with small
form factor. The interesting properties of mmWaves are discussed in more detail in section
2.1.1 illustrating the different impairments of such band, and how to turn such properties
into advantages for Device-to-Device communication. In order to investigate the propa-
gation behavior of mmWave spectrum, a ray-tracing simulation was preformed inside an
urban environment model that is based on ITU-R recommended statistical parameters, tak-
ing into consideration the D2D’s devices heights conditions. These simulation results were
further processed and analysed, so that both the path-loss exponent and the shadowing vari-
ation distribution were obtained for two different cases (i) Line of Sight (LoS) and (ii) Non
2. Aerial and D2D Physical Channel 34

Line of Sight (NLoS). Also this section presents D2D system level performance simulation
when devices are deployed in the same urban environment. The obtained results show that
mmWave D2D is feasible only when utilizing beam forming or beam switching techniques.

2.1.1 mmWaves Propagation Characteristics


The spectrum range between 30 GHz and 300 GHz is referred as the mmWave spectrum
since the wavelengths for these frequencies are in order of millimeters (less than 10mm)
[139]. In general, the utilization of spectrum above 6 GHz has been greatly limited to highly
directive Point-to-Point fixed communication systems, due to the fact that only longer wave-
lengths (i.e. lower frequencies) can diffract around terrestrial terrain and obstacles more
smoothly, and can penetrate more easily though buildings, in contrary to the shorter wave-
lengths (i.e. higher frequencies including mmWaves) that are usually impaired with strong
reflections, refractions and scattering. Also mmWave radios were usually bulky and unsuit-
able for mobile or handheld communication. However there is recently an increasing interest
in exploiting the vast available bandwidth in mmWave spectrum for NLoS communication
and for mobile and cellular systems by exploiting mmWaves ability of strong reflections.
Applications like Point-to-Multipoint PtMP are already in the market today, while other
wireless applications are currently being developed. An example of the promising next
generation mmWave wireless communication systems is the IEEE 802.11ad standard that
is currently under development, proposing mmWaves as the carrier for the future Wireless
LAN Technology (WLAN), along with several other standards that are already in place [60]
such as WirelessHD, ECMA-387. These advancements are mainly facilitated by the new
RFIC technology allowing low cost, low power electronics and antenna solutions [20]. This
section is addressing the general propagation characteristics of mmWaves focusing on the
newly allocated ISM bands within mmWaves spectrum.

2.1.2 ISM Bands in mmWave Spectrum


In spite of the original intention to use the Industrial Scientific Medical (ISM) spectrum
band for non-telecommunication purposes, such as the “operation of equipment and appli-
ances locally generating and using radio frequency energy” [140], this band has acquired
a major focus and interest due to its availability for non-licensed and free use (with some
minor exceptions). In fact, Wi-Fi/WLAN is a very famous example of ISM band utilization
installed in most of the houses and enterprises around the world. Additionally, many other
technologies like Zigbee and Bluetooth utilize ISM bands.
2. Aerial and D2D Physical Channel 35

According to the ITU-R Radio Regulation Articles 5.138 and 5.150, both 24GHz and
61GHz bands are now defined for unlicensed used as ISM [140], if local radio regulation
does not contradict. Table 2.1 is listing the most common applications of microwave ISM
bands in addition to the two mmWave bands.
Table 2.1 Microwave & mmWave ISM Bands

Frequency Range Available Bandwidth Main Applications

Wi-Fi IEEE 802.11b/g/n


2,400-2.500 GHz 100 MHz Bluetooth IEEE 802.15.1
Zegbee IEEE 802.15
5.725-5.875 GHz 150 MHz WLAN IEEE802.11a/n
24.00-24.25 GHz 250 MHz Point-to-Point
Point-to-MultiPoint
WirelessHD (Overlapping)
61.0-61.5 GHz 500 MHz Point-to-Point
IEEE 802.15.3c

In addition to the globally defined ISM bands, local and regional regulators have en-
dorsed the use of mmWave spectrum for unlicensed communication as listed in table 2.2
which summarizes the available 60 GHz band for unlicensed use in different countries [141].

Table 2.2 Unlicensed Local mmWave Regulations

Country Frequency Range Available Bandwidth

Australia⋆ 59.4-62.9 GHz 3.5 GHz


USA 57.0-64.0 GHz 7.0 GHz
Japan 59.0-66.0 GHz 7.0 GHz
Canada 59.0-64.0 GHz 5.0 GHz
⋆ Under discussion

In this section, we are proposing the use of the mmWave ISM bands for D2D commu-
nication that have their pros and cons; on one hand, since ISM is a license-free spectrum,
commercial cellular operators could favor free-of-charge bands comparing to the tremen-
dous spectrum license fees currently in place, that will allow the rapid growth of D2D
market and the alleviation of the congested cellular networks. While on the other hand,
special care and consideration needs to be taken when operating in unlicensed bands due
2. Aerial and D2D Physical Channel 36

to the major interference introduced by other unlicensed users and systems in the future,
taking into consideration that the utilization of mmWaves is currently, to a far extent, very
minimal. An additional merit of the mmWave spectrum is that high antenna directivity can
be achieved using small form factor printed antennas [142]. This high directivity will al-
low the mitigation of interference received from other users sharing the same spectrum by
steering the receiving beam towards the direction of the intended signal only, compared to
omnidirectional antenna that receives power from all directions (including the interfering
signal). On the other side of the link, the transmitter will focus the entire power towards the
intended receiver. This important mechanism will help in avoiding UEs to become neither
aggressors nor victims of co-tier interference.

2.1.3 Atmospheric Attenuation

Fig. 2.2 Atmospheric Absorbtion due to Water Vapor and Gasses

Radio Frequency (RF) propagation in earth atmosphere is usually impaired with mi-
nor energy losses due to the atmospheric absorbtion caused by the resonance of water vapor
2. Aerial and D2D Physical Channel 37

molecules and oxygen gas molecules [61], peaking near certain frequencies such as 24 GHz,
60 GHz and 120 GHz as depicted in Figure 2.2, that was regenerated by curve approxima-
tion according to ITU guidelines in [61], showing a total atmospheric absorbtion of 0.23
dB/km and 13.55 dB/km for frequencies 24 GHz and 61 GHz respectively calculated at
temperature 25◦ C , relative humidity 50% and pressure of 101.3 kPa. In general the signals
using mmWave spectrum endure higher atmospheric absorbtion than signals using lower
frequencies. Also, heavy rain can cause relatively higher attenuation in mmWave spectrum
due to the comparable raindrop size with the mmWave wavelength. These impairments, for
long time, limited the mmWaves communication to only LoS scenario, while NLoS was
avoided. However for short range communication applications, as in the cellular networks,
atmospheric and rain absorbtion are not playing a limiting factor since at distances around
250 meters these absorbtions are summing up to a less than 4dB in case of 61GHz links
and 0.5dB in the case of 24GHz, which is rather considered as an advantage, since it can
help limiting the signal propagating beyond its intended target UE(s) and causing a co-tier
interference.

2.1.4 D2D Communication Under LTE-Advanced


Device-to-Device Communication is a promising technology that has gained a great interest
from researchers and industry alike. This technique will allow peer-to-peer communication,
adding numerous gains to LTE-Advanced; the (i) first gain is proximity gain [22], since
short distance allows higher throughput and lower power consumption, (ii) secondly D2D
can enhance cell edge throughput [43] by performing aggregation and forwarding i.e. acting
as a decode and forward mobile Relay Node, the (iii) third benefit is the reuse gain, since the
transmission range of D2D is distance-limited, a very short spectrum reuse-distance can be
implemented, while utilizing mmWaves will further boost this feature. Figure 2 illustrates
the general concept of D2D communication.
In some situations D2D is considered as a favoured or even a vital means of communi-
cation, these cases are listed below:

• Local source-sink type traffic between two nearby devices (such as local voice/video
calls, file transfer, etc...)

• Extending Cell-edge coverage, by allowing devices to relay traffic towards eNB.

• Backup connection in case of poor cellular network coverage or during natural disas-
ters during which the wireless cellular infrastructure is expected to be severely dam-
2. Aerial and D2D Physical Channel 38

Fig. 2.3 Device to Device Communication Concept

aged. D2D can form a temporary ad-hoc network enabling basic communications.

In fact the lack of D2D communication in LTE Release 10, is one of the major issues
that is inhibiting public safety from adopting LTE technology, due to the concerns regarding
cellular network performance during natural disasters or in remote and at cell edge areas. A
prime demand for D2D technology by public safety authorities is being currently observed,
while 3GPP is striving to include the basic guidelines of this feature in LTE Release 12 [25].
The PHY of D2D link could be modulated by either using Orthogonal Frequency Di-
vision Multiplexing (OFDM) or Single Carrier modulation (SC), the former will give the
ability of coordinated resource sharing with the cellular network and will allow the pos-
sibility of multiDevice-to-multiDevice (mD2mD) by using OFDMA. However SC, on the
other hand, will allow power saving on terminals and lighter signalling overhead, but gives
limited multiplexing capabilities.

2.1.5 Simulation Model


Modeling Urban Environment

When studying a wireless system performance, it is required to develop an accurate defini-


tion of the radio propagation conditions and constraints; one of the most important condi-
tions inside a city is the layout of the buildings that determines the behavior of the propagat-
ing radio signals. ITU-R in its recommendation document [94], is suggesting a standardized
model for urban areas, based on three statistical parameters αo , β and γ, that describe to a
2. Aerial and D2D Physical Channel 39

fair extend the general geometrical statistics of a certain area. These parameters are ex-
plained below:

• Parameter αo : Represents the ratio of the built-up land area to the total land area
(dimensionless).

• Parameter β : Represents the mean number of buildings per unit area (buildings/km2).

• Parameter γ: A statistical variable that describes the building heights distribution ac-
cording to Rayleigh probability density function:

−h2
e 2γ 2
P(h) = h (2.1)
γ2

where (h) is the building height in meters.

Accordingly for simulating urban environment we have selected αo = 0.3, β = 500 and γ =
15, [70] although the latter is irrelevant for our study since the assumed User Equipments’
(UE) heights is limited to 1.5m. In this section it has been selected to reproduce the virtual-
city environment according to Figure 2.4; an array of structures (buildings or houses) of
an assumed square plot of width (W), and inter-building spacing (S), that are linked to the
ITU-R statistical parameters as per the following formulas:
r
αo
W = 1000 (2.2)
β

1000
S = p −W (2.3)
β
While the buildings material is considered concrete, and the ground waves reflections were
neglected. The selected transmitter location is shown in the same figure, as well as the
locations of the distributed receivers (a total of 114 receivers). As it is clear from the figure,
some of the receivers can favor an LoS condition while others do not. The simulated patch
area is 275m X 275m that includes 36 building blocks.

Antenna Beam Switching

Directional antennas transform the propagation impairments associated with mmWave links
into a great advantage by exploiting the strong reflections and refractions caused by urban
structures as an alternative to the LoS communication. Current mobile systems (working
2. Aerial and D2D Physical Channel 40

Fig. 2.4 Building Layout, and Simulation Setup

in the range of 300MHz-3GHz) depend largely on omni-directional antennas for the UE,
while the base station usually exploits directive antennas (of half power beam width 60◦ to
70◦ ) forming the conventional three sectors cellular site. The uplink of the UE is suffering
a huge energy waste, since the RF power is transmitted in all directions, not only draining
the battery energy, but also raising the level of interference in the neighboring cells, and
accordingly reducing the cellular spectral efficiency. The reason of adopting omni anten-
nas in the current UEs is mainly related to antenna dimensions, since the latter is inversely
proportional to system’s frequency, and implementing directional antennas would occupy
a huge space at the current standard cellular frequencies, in the contrary to mmWave fre-
quencies, very small, highly directional antennas could be fabricated [142]. Beam switching
(i.e. using RF switches) is selected for the simulation setup, rather than beam forming (i.e.
using RF phase shifters), as the Switched Beam Array would be more feasible and easier to
implement in future UEs [143]. The utilized antenna array in this section is based on the
generic formula in [144] which relate the Directivity (G) of a planar antenna array to the
2. Aerial and D2D Physical Channel 41

half power beamwidth as per the following:

32400
G= (2.4)
θE θH

where θE and θH are the half power beamwidth in E-Plane and H-Plane respectively. In the
performed simulation we have selected the following for 24GHz: (G=24dBi, θE = 22.5◦ ,
θH = 60◦ ). And for 61GHz: (G=32dBi, θE = 22.5◦ , θH = 45◦ ). Aiming to form a set of 16
antennas, noticing that only one antenna at a time will be active.

Fig. 2.5 Proposed Antennas Combined Patterns (13 out 16 antennas are shown only)

2.1.6 D2D Simulation in mmWave Spectrum


Two main simulation scenarios are presented in this section; the (i) first scenario is for il-
lustrating the path-loss (PL) exponent behavior in urban environment assuming isotropic
antennas on both the transmitter and the receiver, while in the (ii) second scenario is per-
formed using a limited number of receivers distributed according to Figure 2.4. The second
2. Aerial and D2D Physical Channel 42

scenario is addressing three situations; (a) where all antennas are isotropic, (b) using di-
rectional antennas for the receivers (c) using directional antennas for both the transmitter
and the receivers, so the results of these three cases will allow benchmarking the possi-
ble throughput enhancement when embracing the antenna beam switching technique. ray-

Fig. 2.6 RSRP Level due to a D2D transmitter at 60GHz

tracing simulation were performed using Wireless InSite®, and the postprocessing of results
were performed using MATLAB®, including the system level throughput estimation. The
main simulation and the post processing parameters are listed in Table 2.3.

2.1.7 path-loss Exponent in Urban Environment


The first scenario simulation was performed in order to get a general sense of the D2D
path-loss behavior in built-up areas when utilizing mmWave ISM bands. Large number
2. Aerial and D2D Physical Channel 43

Table 2.3 Simulation Parameters

Parameter Value

Simulation Frequency 24 GHz, 61GHz


Channel Bandwidth 20 MHz
Number of Simulated Reflections 2
Number of Simulated Diffractions 1
Number of Simulated Transmissions Not Simulated (No penetration)
Ambient Temperature 25◦ C
Atmospheric Pressure 101.3 kpa
Water Vapor Density 11.5 g/m3
TX/RX Hight from Ground 1.5 m
Transmitter TX Power 23 dBm[145]
Receiver Noise Figure 9 dB [145]
Noise Additive White Gaussian Noise

of receivers were deployed in the simulation environment (above 20,000 Receivers) allow-
ing more accurate path-loss Exponent estimation. The Reference Signal Received Power
(RSRP) is shown Figure 2.6, where it can be clearly noticed that the signal level deterio-
rates rapidly for NLoS receivers.
The received power due to an isotropic transmitter can be calculated as:

Pt
Pr = Ar (2.5)
4πDα

where Pt is the transmitted power, α is the path-loss exponent (which is equal to 2 in case
of free space transmission), D is the distance between the transmitter and the receiver, and
Ar is the aperture area of the receiver defined as:

λ2
Ar = Gr (2.6)

where λ = C/ f is the wavelength and Gr is the receiver antenna gain and it is equal to 1 in
case of an isotropic receiver. Accordingly equation (2.5) can be rewritten to represent the
2. Aerial and D2D Physical Channel 44

incurred path-loss in decibel form as:


 
4π f
PL = Pt [dB] − Pr [dB] = 20log + 10αlog(D) (2.7)
C

As it can be noticed from Figure 2.7 that the simulated path-loss samples have a clear ten-
dency towards two specular modes; one corresponds to LoS and another corresponds to
NLoS. Accordingly the Mean path-loss (MPL), representing the highest PL expectancy,
was obtained using curve fitting (for LoS and NLoS samples separately) following equation
2.7. The samples with large deviation from the specular modes were ignored during the
fitting process, that is because of two reasons; (i) In order to simplify the path-loss model,
(ii) the contribution of these samples is less than 15% of the total samples set. The resulting

Fig. 2.7 Device-to-Device path-loss in Urban Environment

MPL(s) are listed in Table 2.4

Table 2.4 D2D Mean path-loss

Frequency Condition path-loss

24 GHz LoS 60.05 + 1.95 * 10log(D)


NLoS 60.05 + 4.32 * 10log(D)
61 GHz LoS 68.15 + 1.88 * 10log(D)
NLoS 68.15 + 4.49 * 10log(D)

Another important path-loss behavior aspect is the samples deviation from the MPL
2. Aerial and D2D Physical Channel 45

value. This deviation is usually caused by multiple effects; such as small-scale fading,
doppler effect and large-scale fading. However, since the simulation is performed for static
receiver locations with interspacing of 2m (much larger than λ ), the only relevant fading
deviation that can be obtained from this simulation is the large-scale fading. For a receiver
(n) we define the PL deviation for each specular mode as Dn = PLn − MPL, where PLn
is receiver(n) incurred path-loss. Accordingly we can obtain PL deviation distribution as
plotted in Figure 2.8. Multiple statistical distribution fitting has been compared with respect
to their Bayesian Information Criteria (BIC) and it has been found that the best fit for this
variation distribution is the General Extreme Value (GEV) as depicted in the same figure.
However Gaussian Distribution showed a very close BIC score to GEV, and since it is a
much easier way to represent a distribution, we have opted to use it for modeling the path-
loss variation distribution. Accordingly the best fit gaussian standard deviation values are
listed Table 2.5.
Table 2.5 path-loss Deviation from MPL

Frequency Condition σ

24 GHz LoS 4.3 dB


NLoS 7.8 dB
61 GHz LoS 1.2 dB
NLoS 4.0 dB

2.1.8 D2D Link with Beam Steering

Table 2.6 D2D path-loss

Correction Parameter Value

BWe f f 0.57
η 0.9
SNRe f f 1.25 dB

The second scenario simulation is mainly aiming to illustrate the enhancement in sys-
tem level (MAC) throughput that can be achieved when utilizing antenna beam switching.
The D2D transmitter is assumed to behave in a similar manner to an LTE eNodeB, and ac-
cordingly the link level and system level overhead were utilised inline with the bandwidth
2. Aerial and D2D Physical Channel 46

Fig. 2.8 Device-to-Device Deviation from MPL Distribution

efficiency assumptions obtained in [146]. The maximum Modulation and Coding Scheme is
assumed to be QAM512, 4/5, to show the allowed additional gains of the setup, and hence
the data throughput was linked to the SNR using the following:
 SNR−SNRe f f

T P = BW.BWe f f .η.log2 1 + 10 10 (2.8)

where BW is the system bandwidth 20MHz, BWe f f is the system level Bandwidth Effi-
ciency, η is a correction factor and SNRe f f is the SNR implementation efficiency. These
2. Aerial and D2D Physical Channel 47

Fig. 2.9 System Level Throughput at 24GHz and 61GHz

Fig. 2.10 Performance Benchmarking at 5% and 50% Percentile

parameters are given in [146], and listed in table 2.6.


The Cumulative Distribution Function of the D2D throughputs are depicted in Figure
2.9, showing limited enhancement when adopting the one sided beam switching, while a
major improvement can be witnessed when both the transmitter and the receivers utilize
the beam switching technique giving a significant throughput increase in the lower (5%)
percentiles and in also the median (50%) percentile , these enhancements are depicted in
Figure 2.10, showing that for both 24GHz and 61GHz bands, more than 95% of the receivers
in the simulated patch were able to attain a usable OFDM link.
2. Aerial and D2D Physical Channel 48

2.2 Vehicle to Vehicle Channel Modelling


The road transportation sector is undergoing a significant evolution in recent years. Cars
and commercial vehicles are increasingly equipped with sensors and telematics equipment,
which are used to support various Intelligent Transport System (ITS) applications to im-
prove fuel consumption efficiency, safety, traffic management, fleet management and so on.
This evolution is based on the adoption of technologies, which are already deployed in the
consumer market but they are also finding their use in the road transportation sector. The
application of Global Navigation Satellite Systems (GNSS) technology was instrumental
to support drivers in traffic management, while new types of sensors can improve vehicle
fuel consumption, yet another significant progress will come from the adoption of wire-
less communication technologies. While cellular communication technologies can support
the extension of smartphone-based applications to road transportation, manufacturers have
been working on the design and implementation of short-range communications technolo-
gies, which are specifically designed for road transportation.
Dedicated Short Range Communications (DSRC) is the wireless communication tech-
nology widely supported by the industry in the road transportation sector to support Vehicle-
to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication. The term short range
is derived from the advanced vehicle safety or collision avoidance scenarios where vehicles
exchange information at a short distance (e.g., the road) and communication systems can be
designed with this requirement. There is not a single consistent DSRC standard around the
world; different DSRC standards are based on different protocols, uses different frequency
bands and they are not compatible among each other. In fact, these standards have been de-
signed for different purposes and applications. The two most common DSRC standards are
the IEEE and SAE Standards for Wireless Access in Vehicular Environments (WAVE) in the
5.9 GHz band, mostly adopted in USA [147] and the CEN-DSRC standard mostly adopted
in Europe. In this section, we will focus on the CEN-DSRC standard. CEN-DSRC is a
communication standard developed by the European Standardisation Organisation (CEN)
specifically for applications in the Intelligent Transport Systems (ITS) domain.
From this perspective, V2V/V2I are essential for node cooperation in pubic safety where
emergency vehicles communicating with each others and with the fixed infrastructure, in
this section we characterize the V2V and V2I channels in the context of node cooperation.
2. Aerial and D2D Physical Channel 49

Main Applications of CEN-DSRC

The main application of CEN-DSRC has been the electronic toll fees collection and in fact,
CEN-DSRC has been designed for this purpose. Currently, CEN-DSRC is accepted in all
European countries and has found widespread deployment for tolling applications. More-
over, CEN-DSRC is one of the three technologies supported by the European Electronic
Toll Service (EETS) directive number 2004/52/EC. The EETS directive requires members
of the European Union to make their electronic toll fee collection systems interoperable
such that users can move throughout Europe using just one “toll collection box” and only
one contract. This interoperable Europe-wide service is known as “European Electronic
Tolling Service”, EETS.
Beyond electronic tolling, which is a typical V2I application, CEN-DSRC has also been
considered for new ITS applications, one of them is the digital tachograph. A digital tacho-
graph is a mandatory equipment that records the driving time of drivers in commercial
vehicles above 3.5 tons. The tachograph is currently specified by European Economic Com-
munity (EEC) No 3821/85 of 20 December 1985, which was modified at several occasions
and more recently in 2006, the Digital Tachograph (DT) was introduced, and in 2009, the
DT was updated to technical progress to avoid fraud and reduce administrative burden. In
July 2011, the Commission made a proposal (COM(2011) 451 final) for a fully new regula-
tion introducing a new generation of tachograph. The new regulation has been published in
February 2014.
According to the new regulation, the new digital tachograph shall be equipped with a re-
mote communication functionality that will allow law enforcers to read the DT information
from passing vehicles at a roadside control site. A correct implementation of the remote
communication functionality is critical for the successful deployment of the new Digital
Tachograph. The remote communication functionality has a high probability to be imple-
mented based on the existing CEN-DSRC standard. Even if a definitive decision has not
been made at the time of writing this section, the remote communication function will use
the 5.8 GHz band and the physical layer specification will be very similar to the current
CEN-DSRC standard.

Main Objective:

The main objective of this section is to investigate the signal propagation effects in the
5.8GHz frequency band for vehicular communications. In particular we select three distinct
propagation environments in the city of Melbourne and conduct propagation measurements
2. Aerial and D2D Physical Channel 50

for two different antenna heights for the cases of (1) with car and (2) without a car. We
then use the well-known log-normal path-loss propagation model to estimate and report
the path-loss exponent values and the shadowing standard deviation values for the chosen
scenarios. We finally compare and contrast the estimated values with previously conducted
experiments in vehicular networks noting that the vehicular propagation environment varies
from region to region and country to country depending the urbanization and the geographic
structure of the environments.

2.2.1 V2V/V2I Physical Channel


Unlike conventional cellular channel models, vehicular channel has an inherited nonstation-
ary nature due to the high mobility of scatterers (mostly vehicles). This nonstationary nature
can be spotted in both Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) scenar-
ios [148]. In addition to that, V2V communication usually takes place between peers that
are below 2 metres in height, whereas outdoor cellular communication takes place between
a user equipment (UE) with a height of 1-2 meters and a much higher Base Station (BS) at
10-60 metres. Thus, cellular channel models largely deviate from V2X ones leading to the
need for the investigation of the particular nature of the vehicular channel. The main factors
that affect the propagation of radio signals in V2X, are the following [149]:

Dynamic Multi-Path Environment:

Scatter rich environment causes multiple copies (echoes) of the transmitted signal to arrive
to a receiver with different propagation delays, carrying different amounts of power. In
the case of high bit rate communication, these extended copies might overlap with one an-
other, creating Inter Symbol Interference (ISI). An effective way of tackling this issue is to
utilize Orthogonal Frequency-Division Multiplexing (OFDM) which concurrently transmit
the data on narrow adjacent sub-carriers. OFDMA is used in DSRC to effectively mitigate
the ISI issue by properly setting guard intervals. Also, the nature of the V2X channel is
dynamic (time-varying) caused by the rapidly changing scattering environment due to the
mobility of the scatterers (vehicles). This might results in the loss of orthogonality between
the subcarriers causing in Interchannel Interference (ICI) [150]. Multi-path environment
leads to the fast-fading behavior of the received signal that can be characterised by its sta-
tistical distribution such as Rayleigh, Rican or m-Nakagami [151].
2. Aerial and D2D Physical Channel 51

Fig. 2.11 Experiment environments (A)/(B) in Urban area, and (C) in Suburban area (Map
data: Google, Sinclair Knight Merz & Fugro)

Dynamic Shadowing:
The slow-fading component of the received signal results from the changes in the attenua-
tion. However, unlike static environment (such as cellular), the dynamic shadowing in V2X
2. Aerial and D2D Physical Channel 52

caused by the relative movement of vehicles and objects, introduces higher rate slow-fading
than conventional cellular channels. This type of environment continuously fluctuates the
average received power, and causes gray-zone phenomenon as it is called in [149], indicating
that good packet reception is not always guaranteed over V2X channel.

Doppler Shift:
The high mobility nature of the transmitter, receiver, and scatterers causes doppler frequency
shift of the received signal coming from different directions. The amount of frequency shift
varies between the multi-path components, thus when combined at the receiver side causes
what is called doppler spread or frequency dispersion [56]. This spread leads to additional
ICI [152] between the subcarriers. In order to compensate such spread a spectral guard gap
is provided in the CEN-DSRC.
The impairments of the radio signal resulting from the above listed factors can be quanti-
fied using the following metrics [148]: (i) path-loss, representing the average attenuation in
the signal, (ii) Fast-fading, representing the rapid changes in the signal level, (iii) Doppler
spread, as explained previously, and (iv) Delay spread, caused by multi-path component
arriving at different times.
In our experimental setup we focus on the path-loss characteristic of infrastructure to
vehicle (I2V)1 and adopt the widely accepted log distance model [57], where the average
received signal is a function of the base-10 logarithm of the distance d, according to the
following equation:
 
d
Pr (d) = Pr (do ) − 10α log10 (2.9)
do
where Pr (do ) is the reference received power at a distance d0 and α is the path-loss
exponent. The selection of the reference distance do is depended on the specific propagation
environment [153], typically do is chosen to be 1m for indoor scenarios, 100m for outdoor
micro-cellular scenarios and 1600m for outdoor macro cellular scenarios. In our application
of course the scenario is different where the communication usually takes place within a
street over a range up to 100 to 200 meters. Therefore we propose to use a reference distance
around 10m to 20m.
Furthermore, the log-distance model can be extended to adopt the random shadowing
component by incorporating a Gaussian distributed random variable in (2.9) in the dB scale
1 The I2V model can also be adopted for V2I scenarios as well.
2. Aerial and D2D Physical Channel 53

with a standard deviation σ (dB). Therefore the log-normal shadowing model is given by
[153]:
 
d
Pr (d) = Pr (d0 ) − 10α log10 + XdB (2.10)
d0
where, XdB ∼ σ N (0, 1) is a zero mean Gaussian distributed random variable in dB with
a standard deviation of σ (dB). Note that the models presented in (2.9) and (2.10) are well
known in the literature, in this section we use these models to study and understand the
5.8GHz propagation in a vehicular environment.

2.2.2 Experimental Setup


The major aim of the conducted experiment is to provide a generic sense of how the path-
loss exponent behaves in different V2X vehicular environments, as well as to understand if
any difference can be spotted when using stand-alone receiving antenna and when mounting
the same antenna on the roof of the car. The measurement focuses mainly on line of sight
(LoS) situations.

Environments

We have selected three different road environments in the city of Melbourne for this exper-
iments as depicted in Figure 2.11: (A) Urban type-A (B) Urban type-B and (C) Suburban
Type-C.
Urban Type-A Environment: As indicated in Figure 2.11 consists of 6m wide lane,
1-3m wide walk surrounded by high-rise buildings from both sides. The selected test path
is around 120m line, no inclination in the street level. This road is usually busy with pedes-
trians, and has limited vehicular traffic. The transmitter is placed at one end of the test path,
as shown in the same figure.
Urban Type-B Environment: This environment consists of two parallel 6m wide lanes
with a 5.5m wide island in the middle for car parking, also two side-walks about 3m wide
for pedestrians. This road is surrounded by multi-storey buildings on both sides and is busy
with many cars in both directions with some cars parked on the sides as well as in the
middle island. The transmitter is placed on the middle of the road at one end of the test path
as indicated in Figure 2.11. The road is flat with no inclination. Sparse vegetation is present
around the road.
Suburban Type-C Environment: Although this environment is located in a suburban
2. Aerial and D2D Physical Channel 54

district, however it can be seen as a transition between urban environment towards suburban.
It consists of two parallel 6m wide lanes with a 5m wide island in the middle for parking
and 3m side-walks for pedestrians. This road is surrounded by one and two-storey buildings
on both sides. Sparse vegetation can be observed on the sides of the road. The transmitter
is placed on the middle of the road at one end of the test path as shown in Figure 2.11. The
terrain of the environment is flat with no inclination.

2.2.3 Scenarios
In order to spot the effect of the car body on the path-loss exponent, we have performed the
measurements for two scenarios:
Scenario-1 No Vehicle: A transmitter is placed on top of a telescopic mast, while the
receiving antenna is placed on another mast, as indicated in Figure 2.12.
Scenario-2 With Vehicle: Similar to scenario-1 but the receiving antenna is placed on
the roof of the car near to the rear window as indicated in Figure 2.13. The same antenna is
used for both scenario-1 and scenario-2. The used car for the measurements was a Toyota
Corolla Sedan 2013 (note that the car in the Figure does not exactly resemble the actual car
that was used).

Fig. 2.12 Transmitter and receiver placement, for scenario-1 (no car).
2. Aerial and D2D Physical Channel 55

Fig. 2.13 Transmitter and receiver placement, for scenario-2 (with car).

2.2.4 Procedure
All measurements are conducted for two different antenna heights of the transmitter (H1 =
1.5 m and H2 = 3 m), while the receiving antenna is always kept at 1.5m. The aim of this is
to mimic the V2V and V2I situations when using H1 and H2 respectively.
The used transmitter is a commercial analog video extender having 14dBm EIRP power,
with selectable carrier frequency, our experiment is conducted at frequency (f = 5.82GHz),
while at the receiver side we used an omni directional antenna mounted on a 1.5m mast con-
nected to a handheld spectrum analyzer (Agilent FieldFox) capable of performing accurate
channel power measurement. The spectrum analyzer was connected to a laptop to record
the readings along with the location of the receiver which was measured using a GPS device
connected to the same laptop. For each reading, we have averaged 15 samples of the channel
power. We have used the internal pre-amplifier of the spectrum analyzer for increasing the
Signal to Noise Ratio (SNR), thus enhancing the power accuracy. Experiment parameters
are listed in table 2.7
In each environment, the location of the transmitter was recorded using the same GPS
device. Field measurements raw data include two key pieces of information. For each
measurement point we collect: (i) The average received power of the channel, (ii) The
GPS coordinates of the receiver. The distance between the transmitter and the receiver is
2. Aerial and D2D Physical Channel 56

Table 2.7 Experiment Parameters

Parameter Value

Frequency 5.80 GHz


Measurement Bandwidth 200 kHz
Internal Pre-Amplifier Gain 25 dB
Transmitter EIRP 14 dBm
Receiver Antenna Gain 12 dBi
Distance 20-150 m
Transmitter Antenna Height H1=1.5m & H2=3.0m
Receiver Antenna Height 1.5m

accordingly calculated using the logged coordinates. Raw data along with the calculated
distances are then fitted using Levenberg–Marquardt algorithm to a linear equation of the
following form:
y = P1 x + P0 (2.11)

where in our case y is the averaged received power at distance di , x = log10 (di ). Therefore
by comparing with the log-normal path-loss model described in (2.9) an estimate of the
path-loss exponent α is given by α̂ = −P1 /10.

2.2.5 Measurement Results


Following the procedure in section, we have obtained samples as indicated in Figure 2.14,
where the results of environment-A are indicated in the upper graph and results from
environment-B and C follow below. Accordingly, we have obtained the values for the path-
loss exponents which are mentioned in table 2.8.
It can be observed from Figure 2.14 and Table 2.8 that different environments have dif-
ferent path-loss profiles. Also, that the transmitter antenna height has a significant influence
on the path-loss profile of a given environment. One consistent commonality that can be
observed from Figure 2.14 is that the samples collected with car are 2-10dB lower than
the samples collected using the stand-alone antenna. Also, that the fitting curve for with
car is always below the fitting curve for without car for all scenarios. We believe the main
reason for this difference is the drop in the receiving antenna directivity caused by the fi-
2. Aerial and D2D Physical Channel 57

nite metallic rooftop of the vehicle. As explained in [148], the finite conductive surface of
the car causes the horizontal omni pattern of a monopole antenna to be shifted upward by
around 20◦ degrees, causing a degradation of the directivity of the receiving antenna in the
horizontal plane.
Table 2.8 shows that the variation in path-loss exponents for urban environments A and
B for a given transmitter antenna height is high whereas for suburban environment C it is
low. In other words, for suburban environment C the path-loss exponent does not vary much
for no vehicle and with vehicle scenarios for a given transmitter antenna height.

Table 2.8 path-loss Exponent (α)

Height = 1.5m Height = 3.0m


Without Car With Car Without Car With Car
Environment A 2.7 2.3 1.9 2.2
Environment B 2.9 3.7 3.3 2.8
Environment C 1.5 1.7 1.9 1.7

In order to get a general sense of the variation of the collected received-power samples
with respect to the mean received-power fitting line, we have obtained the standard devia-
tion σ of the difference between the samples values and the corresponding expected mean
received-power at a certain distance. Below is the related mathematical expression:
s
1 N
σ= ∑ [yi(di) − ȳ(di)]2
N i=1
(2.12)

where N is the number of collected samples, yi (di ) is a received power sample collected at
a certain distance di in dB, and ȳ(di ) is the corresponding mean received-power in dB as
obtained from the curve fitting process.
The results of each of the scenarios are listed in table 2.9. It is noticeable that environ-
ment B has the highest standard deviation of other environments. An expected result that
can be primely referred to the high car traffic in this environment. This high traffic increases
the dynamic shadowing as explained previously.
Note that, considering the fact that we have a high time varying channel environment,
the shadowing standard deviation values can be improved in accuracy by time averaging the
results which will statistically provide a higher confidence in the estimates when the same
experiments are repeated over time.
2. Aerial and D2D Physical Channel 58

Fig. 2.14 Received power samples and fitting curves, from top to bottom: Environment-A
Heights H1 and H2, Environment-B Heights H1 and H2, and Environment-C Heights H1
and H2
2. Aerial and D2D Physical Channel 59

Table 2.9 Estimated Shadowing: Computed Standard Deviation (σ )

Height = 1.5m Height = 3.0m


Without Car With Car Without Car With Car
Environment A 2.2 dB 1.8 dB 2.0 dB 1.7 dB
Environment B 4.4 dB 3.4 dB 3.2 dB 1.9 dB
Environment C 2.2 dB 1.3 dB 2.2 dB 1.3 dB

Propagation Measurement Results for D2D in Rural Areas

Following the same procedure presented here, we have also conduced propagation measure-
ments for D2D channel in rural areas, aiming to get an insight on the path-loss exponent and
the shadowing standard deviation. The selected environments were as the following: (i) Ru-
ral street with structures on both sides (indicated as Hewson and Grundy), (ii) rural street
with structures on one side (indicated as Follet and Watts), and (iii) forest environment (in-
dicated as ReserveS01 and ReserveS02 ), as shown in Fig. 2,3 and 4 in [1]. The results of the
measurements are listed in Table 2.10, where α is the path-loss exponent and σ dB is the
standard deviation of zero-mean Gaussian distributed random variable Xσ representing the
shadowing in the following equation:
 
d
PL(d) = PL(d0 ) + 10α log10 + Xσ , (2.13)
d0

where PL(d) is the path-loss in dB at a distance of d metres from the transmitter, PL(d0 ) is
the reference path-loss in dB at a distance of d0 .

Table 2.10 Estimated α, σ and PL(d0 ) from measurement campaign in [1]

Environments 922 MHz 2,466 MHz


PL(d0 ) α σ (dB) PL(d0 ) α σ (dB)
Hewson 46.81 3.55 4.32 45.70 4.95 7.10
Grundy 41.99 4.09 4.36 46.07 4.89 7.14
ReserveS01 44.49 3.67 4.89 61.83 4.02 5.63
ReserveS02 48.43 5.29 4.86 69.82 6.59 5.81
Follet 52.96 2.97 5.39 61.44 2.64 6.62
Watts 48.40 3.31 4.70 54.63 3.54 5.11
2. Aerial and D2D Physical Channel 60

2.3 Air-to-Ground Channel Modelling


The recent developments in broadband wireless communication technology in terms of ca-
pacity and reliability has led an emerging rapid adoption by a wide sector of mission critical
users, such as the public safety agencies including police forces and fire fighters. However,
with the increasing dependency on such broadband networks, the total failure of public ser-
vices would be massive if networks are disrupted due to a natural disaster such as flood,
earthquake or tsunami, making the need for finding a rapid and cost-effective temporary
recovery solution an important necessity.
One of the prospective feasible solutions for realizing wireless recovery networks, or
as it is called in [131, 132] the Emergency Supplementary Networks (ESN) is by utilizing
airborne base stations [36]. A concept that has been endorsed by the Homeland Security
Bureau in USA as the Deployable Aerial Communications Architecture
When deploying an aerial network it is important to properly estimate the infrastructure
required to establish the service [154], especially that radio resource and energy is quite
limited in such networks [155, 156]. Accordingly, Radio Frequency (RF) planning should
be carried out for the target area that can produce an estimation for the (i) the required num-
ber of Aerial Base Stations, (ii) the optimum altitude of the platforms [129] (see Sec. 3.3),
and (iii) the expected service level. In this section, we propose a statistical RF propagation
model for estimating the coverage of an Air-to-Ground wireless service provided by Low
Altitude Platforms (LAP). The proposed radio model in this section will allow aerial net-
work operators to perform accurate RF planning without the need to rely on full site-specific
ray-tracing simulation. Since during the rush of the aftermath of a disaster it is unlikely that
the city buildings 3D-model to be easily available. In this case the RF planning could be
based on the urban statistical parameters that describes the city structures. These parameters
are possible to obtain.
The call for proposing this generalized RF model was encouraged by two main mo-
tivations; (i) the increasing need for LAP-based wireless services [157], and (ii) the lack
of a simple and generic statistical model that can easily link the urban statistical parame-
ters with the radio propagation conditions. For example, authors in [158] are suggesting a
method for finding the probability of a Line of Sight (LoS) between an aerial base station
and a ground receiver as a function of the elevation angle, however no full shadowing model
has been proposed. A step further was taken in [63] by building a path-loss model based
on three classes of links, Line of Sight (LoS), Obstructed Line of Sight (OLoS) and None
Line of Sight (NLoS), however the study was based on a single model city, and cannot be
2. Aerial and D2D Physical Channel 61

widely generalized for different types of urban environments. While in [70] the approach
was toward deducing a generic statistical model for air-to-ground path-loss that applies for
a wide range of urban environment types, but the results were obtained for High Altitude
Platforms only. In [159] a model based on Lonely-Rice propagation program was discussed,
however it only accounts for large-scale terrain irregularities rather than addressing small
scale urban environment properties. To the authors best knowledge there is no published
work that is specifically addressing the path-loss of low altitude platforms in different urban
environments.

2.3.1 Low Altitude Platforms


Low Altitude Platforms (LAP) [160] [161] are quasi-stationary aerial platforms (Quad-
copters, UAVs or Balloons) with an altitude below the stratosphere (10,000 m), in contrary
to High Altitude Platforms (Above 10,000 m) [162]. LAPs are much easier to deploy, and
are more in line with the broadband cellular concept, since low altitude combines both cov-
erage superiority and confined cell radius. For the RF model proposed here, we are focusing
on the lower part of the stratosphere, i.e. for heights between 200m and 3,000m. The tech-
nology carried by the LAP depends on the end-user’s application, budget and bandwidth
requirements. Applications could be as advanced as LTE, Wi-Fi, WiMAX or as legacy as
TETRA or P-25 systems.
Doppler Effect resulting from the possible high velocity of an airborne transmitter is
not taken into consideration in this analysis, since the assumption is focusing on the quasi-
stationary LAPs as represented in Figure 2.15.

2.3.2 Modeling Urban Environment


Developing an RF model requires an accurate definition of the study conditions and con-
straints; one of the most important conditions in an urban environment is the layout and
characteristics of the buildings. The International Telecommunication Union (ITU-R) in its
recommendation document

• Parameter αo : Represents the ratio of built-up land area to the total land area (dimen-
sionless).

• Parameter βo : Represents the mean number of buildings per unit area (buildings/km2 ).
2. Aerial and D2D Physical Channel 62

Fig. 2.15 Low altitude platform concept

• Parameter γo : A scale parameter that describes the buildings heights distribution ac-
cording to Rayleigh probability density function:
 2
h −h
P(h) = 2 exp (2.14)
γo 2γo2

where (h) is the building height in meters.

In order to cover a wide range of possible applications for this model, four simulation en-
vironments were selected, similar to [70]: (i) Suburban Environment that also covers the
rural areas, (ii) Urban Environment which is the most common situation representing aver-
age European cities, (iii) Dense Urban Environment representing some types of cities where
buildings are in close proximity to each other, (iv) Urban Environment with highrise build-
ings, representing modern cities with skyscrapers style. ITU-R statistical parameters are
relatively straightforward to obtain from a certain city’s urban plan, and are likely to be well
documented by the city urban planning authorities. Table 2.11, summarizes the selected
ITU-R parameters for these environments.
The challenge lies in finding a generic geometrical model that satisfies these parame-
ters, and at the same time represents an acceptable layout of which a certain city might be
structured. It is well known that for every city or suburb a certain urban planning code and
2. Aerial and D2D Physical Channel 63

Table 2.11 Selected Urban Environments

Environment αo βo γo

Suburban 0.1 750 8


Urban 0.3 500 15
Dense Urban 0.5 300 20
Highrise Urban 0.5 300 50

style exists, so even if two cities share the same statistical parameters, that does not imply
sharing the same buildings layout. Accordingly a “standard-city” plan should be modeled
for allowing a valid general mathematical approximation. We select to shape the virtual-city
environment similar to Manhattan grid depicted in Figure 2.16; an array of structures (build-
ings or houses) of an assumed square plot of width W , and inter-building spacing of S. where

Fig. 2.16 The selected urban layout

W and S are measured in meters, and can be linked to the ITU-R statistical parameters, since
W 2 Nb
by definition: αo = (1000D) 2 , where D is the map side measured in kilometers (for a square
Nb
patch), and Nb is the number of building inside the patch. On the other hand βo = D2
. Ac-
2. Aerial and D2D Physical Channel 64

q
cordingly we can write the width expression W as the following: W = 1000 αβoo . Looking

at the map patch, we can also notice that D = S+W
1000 Nb , accordingly, the spacing S can
1000
be rewritten as the following: S = √ −W , where S includes all structure-free areas such
βo
as: pavements, roads, open gardens, open car parkings. Corresponding to the urban envi-
ronment parameters listed in Table 2.11, four virtual-cities were generated (using a script
writing in MATLAB® ), a city patch size of (1,000m × 1,000m) was selected for perform-
ing the radio propagation simulation. As depicted in Figure 2.17, building heights were
randomly generated based on equation (1), while the width and spacing were kept constant
for each of the generated environments. The plotted area in Figure 2.17 is limited to (250m
× 250m) for a clearer representation.
Modeling urban environments does not only involve the geometry of the buildings but
also should consider the materials and surface behavior of all structures, that will interact
with electromagnetic (EM) waves generated by the LAP transmitter and will cause reflec-
tion, diffraction and scattering. Starting with reflection, the assumed material constituting
the outer layer of the buildings was concrete, that has a non-negligible reflection and dielec-
tric parameters
Additional possible urban geometry effects of trees, lampposts and mobile objects re-
flections were neglected, assuming that the large-scale building geometry and its EM char-
acteristics will dominate the average path-loss. Buildings were considered as solid blocks,
since it is unlikely that the propagation by penetration through buildings will have a signifi-
cant effect on the model.

2.3.3 Radio Model Development


When considering urban environments, RF signals are unlikely to travel via a single type
of propagation method [163], since manmade structures introduce significant interactions
with the radio path; such as signal blockage, diffraction, transmission and reflection. A
ray-tracing simulation (using 0.25◦ rays spacing) was conducted using Wireless InSite®
ray-tracing software for three types of rays (Direct, Reflected and Diffracted), while trans-
mitted rays (through walls) were neglected in order to simplify the calculations. As depicted
in Figure 2.18, showing the direct rays, which are defined as signals traveled freely between
the transmitter and the receiver with no obstacles’ interaction. Receivers that can favor this
type of rays are considered having full Line of Sight (LoS) condition with the transmitter. It
is important to note that the availability of the geometrical line of sight does not necessarily
mean that the RF line of sight condition is satisfied, since RF signals require much wider
2. Aerial and D2D Physical Channel 65

ellipsoids than the optical light, according to Fresnel Zones Concept [151]. The second
treated rays type is the diffracted ones, which are considered a principle propagation mech-
anism in urban environments. In our simulation setup, we are considering the buildings as
the only diffractive obstacles, where all edges are approximated to ideal knife-edges. In this
case, the EM wave will follow a well-determined diffraction behavior. The third type of
rays that might reach a receiver is by the means of reflection from buildings’ walls, where
each reflection causes a magnitude reduction in the electric field and a certain phase shift
depending on the angle of incidence.

Fig. 2.17 Computer generated city models based on the ITU-R parameters.

In order to obtain the coverage of the LAP over the target area (1,000m × 1,000m)
we have simulated the received power of more than 37,000 uniformly distributed receivers,
where the electric field of the all captured rays is summed (complex summation) and the
received power is calculated from the resulting electric field. The total power loss impairing
a signal transmitted from the LAP to nth receiver can be written in decibel form as:

PLn = 10 log(PTX ) − 10 log(PRXn ) (2.15)

where PTX is the LAP transmitted power and PRXn is the received power at the nth re-
ceiver. The simulated received power level is depicted in Figure 2.19, assuming a reference
isotropic transmitter power of (0 dBm) placed 200m above the mean ground level. The ray-
tracing simulator will yield a list of all receivers including the corresponding path-loss of
each (we call these results as samples), and the samples are then further processed in order
2. Aerial and D2D Physical Channel 66

to obtain the excessive path-loss component as we define it in the following formula:

ηn = PLn − FSPLn (2.16)

where FSPLn is the free space path-loss for a certain receiver, obtained from Friis transmis-
sion equation [151] and represented as the following:

FSPLn = 20 log(dn ) + 20 log( fMHz ) − 27.55 (2.17)


 
∆h
where dn = is the distance between the LAP and a receiver n measured in meters,
sin θn
and fMHz is the system center frequency in MHz, accordingly:
 
∆h
FSPLn = 20 log + 20 log( fMHz ) − 27.55 (2.18)
sin θn

where ∆h = hLAP − hRX ; hLAP is the LAP altitude and hRX is the receivers’ common height.
θn is the elevation angle (measured in degrees) as the LAP is seen from a certain receiver
point. It is worthy to note that the LAP altitude itself has a minimal effect on the excessive
part of the path-loss, where the main effect of the altitude resides in increasing the FSPL
part.
The simulation is performed for three different frequencies (700 MHz, 2,000 MHz and
5,800 MHz) that are believed to cover a wide range of applications. Each frequency is
simulated over the four different urban environments. The resulting data sets are 12 (3
frequencies × 4 environments).

2.3.4 Understanding Simulation Results


In order to understand the simulation results and deduce from them a radio model, we first
obtain the data set histogram as depicted in Figure 2.20 showing the excessive path-loss η.
The histogram is showing a clear tendency towards three propagation groups, denoted as G1,
G2 and G3, where (G1) the first group correspond to receivers favoring Line-of-Sight con-
dition or near-Line-of-Sight condition, while (G2) the second group generally corresponds
to receivers with no LAP Line-of-Sight but still receiving coverage via strong reflection and
refraction, (G3) the third group has a very limited contribution to the total sample set, and
corresponds to receivers suffering deep fading resulting from consecutive reflections and
diffractions. In our model, we disregard the effect resulting from G3 as it constitutes less
than 3% of the total samples set.
2. Aerial and D2D Physical Channel 67

Fig. 2.18 The three different types of the simulated rays

Fig. 2.19 Received Power, obtained at (Frequency = 2,000 MHz, Tx-Height = 200 m, Tx-
Power = 0 dBm) for urban environment.
2. Aerial and D2D Physical Channel 68

Samples clustering is then performed in order to study the statistical behavior separately
for each propagation group by deducing the mean value and a standard deviation trend
versus the available range of elevation angles, an example of samples clustering is depicted
in Figure 2.21 showing the three propagation groups, while the density of samples was
reduced by a factor of 10 for clearer depiction. It can be noticed from the figure that the
samples tends to have a constant mean value (per group) for a certain frequency and urban
environment, however samples distribution around the mean is taking more disperse values
for lower elevation angles, and it is noticed that the standard deviation is angle dependent.

Fig. 2.20 Excessive path-loss samples histogram, obtained at frequency = 2,000 MHz for a
dense urban environment.

Finding parameters for the model requires fitting experimental data to explicit mathe-
matical formulas. In order to do so, we define the group occurrence probability that repre-
sents the probability of a certain propagation group to occur at a certain elevation angle, and
we denote it as pθ (ξ ), where ξ is the propagation group that a receiver could have, with
two discrete values of either (1 or 2) as the third propagation group is ignored. Accordingly,
at a particular elevation angle the excessive path-loss will have a joint probability density
function density function (PDF) of fθ (η, ξ ) which can be expressed as:

fθ (η, ξ ) = fθ (η|ξ ).pθ (ξ ) (2.19)


2. Aerial and D2D Physical Channel 69

Fig. 2.21 Excessive path-loss samples organized in terms of the elevation angle θ , and
categorized into three distinct propagation groups, obtained at frequency = 2,000 MHz for
a dense urban environment.

where fθ (η|ξ ) is the conditional PDF of η given ξ . An illustrative representation of the


excessive path-loss PDF could be visualized by obtaining the integration of fθ (η, ξ ) for all
values of θ , given by: Z
fθ (η, ξ ).dθ = f (η, ξ ) (2.20)
θ

In order to simplify the mathematical modeling of the propagation group distribution


fθ (η|ξ ) we propose the use of Gaussian distribution model. Hence, fθ (η|ξ ) can be ex-
pressed as the following:
fθ (η|ξ ) = N (µξ , σξ2 (θ )) (2.21)

where N is the normal distribution of a mean µξ and a standard deviation σξ (θ ). The


remaining task is to find the model equations for pξ (θ ) , σξ (θ ) and µξ .

2.3.5 Obtaining Radio Model Parameters


Starting with µξ , as stated previously the mean excessive path-loss did not show a clear
dependency on the elevation angle θ but rather a constant value, that can be obtained by
2. Aerial and D2D Physical Channel 70

averaging all samples in a certain propagation group, the results are listed in Table 2.12.
While for obtaining the general trend of the standard deviation of G1 and G2. We first
obtain the standard deviation of each subset belonging to a common elevation angle and
a common propagation group as discrete points depicted in Figure 2.22 as (triangles and
squares). After that, the samples are fitted to the following formula:

σξ (θ ) = aξ exp(−bξ .θ ) (2.22)

where aξ and bξ are frequency and environment dependent parameters obtained by curve
fitting using Damped Least-Squares (DLS) method, and are listed in Table 2.12. As depicted
in the same figure, the fitting of the standard deviation shows very good approximation.

Fig. 2.22 Excessive path-loss Standard Deviation, obtained at frequency = 2,000 MHz for
an urban environment.

For modeling pθ (ξ ) we first count the number of samples occurred in a certain propa-
gation group at a certain elevation angle Kξ (θ ), also we count the total number of samples
having the same elevation angle KTotal (θ ), then we calculate the following ratio:

Kξ (θ )
p̂θ (ξ ) = (2.23)
KTotal (θ )
2. Aerial and D2D Physical Channel 71

Equation (2.23) gives an estimation of the group occurrence probability pθ (ξ ), the estima-
tion accuracy enhances for large sample sets. In this study we assume pθ (ξ ) = p̂θ (ξ ) and
accordingly we obtain the explicit form of pθ (ξ ) using curve fitting of p̂θ (ξ ) points. In
Figure 2.23 an example of a group occurrence curve fitting is illustrated for urban environ-
ment at frequency of 2,000 MHz for the two main distinct groups of propagation. The solid
curves in the figure represent the probability of a certain propagation group to occur at a
certain angle, that can be represented as:
Z +∞
pθ (ξ ) = fθ (η, ξ ).dη (2.24)
−∞

while the discrete data points (triangles and squares) represent the measured ratio p̂θ (ξ ) as
explained before.

Fig. 2.23 Propagation Group Occurrence Probability, obtained at frequency = 2,000 MHz
for an urban environment.

It is important to note that the proposed model here is valid only for elevation angles
above 15◦ , since low elevation angles have a very limited probability of receiving any signal
from the LAP, another reason is that the selected simulation setup can produce results for
down to about 15◦ only.
The resulting curve explicit equation for group occurrence probability is chosen to bal-
2. Aerial and D2D Physical Channel 72

ance simplicity and accuracy, and it is a function of the elevation angle as the following:

pθ (1) = c.(θ − θo )d (2.25)

where θo is selected as 15◦ corresponding to the minimum angle allowed by the model. c
and d are frequency and environment dependent parameters obtained by curve fitting using
(DLS) method. The results are listed in Table 2.12. pθ (1) represents the probability of
a signal to be received obeying group 1 path-loss profile. On the other hand group 2 oc-
currence probability pθ (2) can be simply calculated as it is the complimentary of group 1
probability:
pθ (2) = 1 − pθ (1) (2.26)

The three model parameters are now explicitly defined, the following section will demon-
strate how to implement and utilize the proposed radio model.

Table 2.12 RF Model Parameters

700 MHz
Suburban Urban Dense Urban Highrise Urban
µ1 0.0 0.6 1.0 1.5
µ2 18 17 20 29
(a1 ,b1 ) (11.53, 0.06) (10.98, 0.05) (9.64, 0.04) (9.16, 0.03)
(a2 ,b2 ) (26.53, 0.03) (23.31, 0.03) (30.83, 0.04) (32.13, 0.03)
(c,d) (0.77, 0.05) (0.63, 0.09) (0.37, 0.21) (0.06, 0.58)

2,000 MHz
Suburban Urban Dense Urban Highrise Urban
µ1 0.1 1.0 1.6 2.3
µ2 21 20 23 34
(a1 ,b1 ) (11.25, 0.06) (10.39, 0.05) (8.96, 0.04) (7.37, 0.03)
(a2 ,b2 ) (32.17, 0.03) (29.6, 0.03) (35.97, 0.04) (37.08, 0.03)
(c,d) (0.76, 0.06) (0.6, 0.11) (0.36 , 0.21) (0.05, 0.61)

5,800 MHz
Suburban Urban Dense Urban Highrise Urban
µ1 0.2 1.2 1.8 2.5
µ2 24 23 26 41
(a1 ,b1 ) (11.04, 0.06) (10.67, 0.05) (9.21, 0.04) (7.15, 0.03)
(a2 ,b2 ) (39.56, 0.04) (35.85, 0.04) (40.86, 0.04) (40.96, 0.03)
(c,d) (0.75, 0.06) (0.56, 0.13) (0.33, 0.23) (0.05, 0.64)
2. Aerial and D2D Physical Channel 73

2.3.6 Radio Model Implementation


The proposed RF model here allows designers to calculate the expected path-loss for re-
ceivers in a certain geographical area. path-loss information can be used to produce nu-
merous information, such as the Signal to Interference and Noise Ratio (SINR), and the
expected throughput. The design steps to implement this RF model are illustrated below:

1. Get System Parameters (Center Frequency, LAP Altitude) and Urban Statistical Pa-
rameters αo , βo and γo .

2. Select a receiver n.

3. Calculate the Elevation Angle θ of receiver n.

4. Calculate the Free Space path-loss of receiver n according to equation (2.18).

5. Pick a propagation group randomly corresponding to the occurrence probability, using


the Group Randomization method explained later.

6. Generate the excessive path-loss η of receiver n as a random number according to the


mean and standard deviation of the corresponding propagation group.

7. Calculate the total path-loss of receiver n according to equation (2.15).

8. Repeat steps 2 to 7 for all receiving points.

It is important to implement the Group Randomization correctly, the randomization should


deduce the same probability of occurrence as per equations (2.25) and (2.26). And in order
to facilitate the model implementation we suggest the following pseudocode for propagation
groups randomization:

1. Calculate groups occurrence probability as per equations (2.25) and (2.26): pθ (1) ,
pθ (2).

2. Generate a random (RAND) number between 0 and 1, with uniform distribution.

3. If pθ (1) ≥ pθ (2) then If RAND ≥ pθ (2) the propagation group is 1 otherwise it is


group 2.

4. Else if pθ (1) < pθ (2) then If RAND ≥ pθ (1) the propagation group is 2 otherwise it
is group 1.
2. Aerial and D2D Physical Channel 74

If the reader opt to use MATLAB® for implementing the RF model, then the following func-
tion can be utilized for creating the randomization in the values of η as: ETA=normrnd(MU,
STD), where (MU) and (STD) are the mean and standard deviation corresponding to the sim-
ulated propagation group, urban environment and to the system frequency.
Noticing that patch areas with dimensions larger than 1km × 1km can be split up into
smaller divisions if needed, i.e. if the urban environment statistical parameters αo , βo and
γo noticeably varies for each geographic sub-patch, then each sub-patch can be treated in-
dependently. Another point that should be taken into consideration is the directivity of the
implemented antenna system, because the provided model is assuming isotropic transmitters
and receivers.

2.3.7 Radio Model Verification


The advancement of computer-aided propagation tools that are based on ray-tracing simu-
lation allows the prediction of signal strength in a very accurate manner [164]. However
as any other RF model, the accuracy lies in the stipulated mathematical assumptions, for
example if the target area seems to follow an entirely different buildings layout than the
one assumed here, then the accuracy of the model predictions will decrease. The aim of an
RF model is to simplify simulation calculations and to speed up radio coverage estimation;
however it cannot match the accuracy of site-specific ray-tracing simulations. In order to
verify the proposed RF model in this section, we have compared the simulation data for a
patch area of (1,000m × 1,000m) obtained from Wireless InSite® ray-tracing simulator on
one hand, versus the model parameters listed in Table 2.12 implemented using MATLAB®
on the other hand. The obtained results are presented in terms of the cumulative distribution
function versus the path-loss (in dB). Figure 2.24, depicts the comparison results, showing
high level of matching between the CDF curves. This matching indicates that the prediction
resulting from the mathematical model reproduces the path-loss estimations in an accurate
manner with respect to the ray-tracing simulation. Verification performed for the thee model
frequencies, while only F=2,000 Mhz is depicted in the figure.
2. Aerial and D2D Physical Channel 75

Fig. 2.24 Cumulative Distribution Function comparison between the proposed RF model vs.
ray-tracing simulation data.
2. Aerial and D2D Physical Channel 76

2.4 Identifying Spectrum Opportunities for Node Cooper-


ation
Spectrum scarcity is increasingly becoming an urgent issue, imposing a tough barrier for
both wireless service providers, and for regulating authorities alike, leading to an increas-
ing spectrum fees and inhibiting wireless technology from coping with the growing users-
demand. The main driver of this scarcity is the increasing vast reliance on wireless commu-
nication as a ubiquitous means of communication, coupled with low utilization efficiency
of the electromagnetic spectrum.
Several measurement campaigns observed a low spectrum utilization [72–76], albeit the
current scarcity and the soaring spectrum fees. This under-utilization is mainly referred
to legacy wireless systems, where spectrum right-of-use is restricted to a single operator
(or user) running a low efficiency wireless system. Alternatively, more efficient spectrum
utilization can be achieved by exploiting one or more of the radio resource domains in a
non-static manner, these domains span over frequency, time and space.
In the non-static method, a spectrum chunk can be assigned to a group of operators
rather than a single exclusive user. This type of spectrum management is referred as Dy-
namic Spectrum Access (DSA) [10], allowing several users to share a spectrum band while
adhering to a set of rules and regulations formed to guarantee some different contractual
rights. These rights do not have to be in necessary equal. In fact, one of the promising so-
lutions for increasing the spectrum utilization, is offered by Cognitive Radio (CR) systems
[165], where Secondary Users (SU) equipped with CRs can access the spectrum allocated to
Primary User(s) (PU) in an opportunistic manner, without causing harmful interference or
service degradation below a pre-defined level. In this manner, the PU has a guaranteed Qual-
ity of Service (QoS) level, while SUs might have no guarantee for the QoS performance.
CRs exploit the under-utilized spectral gaps, that might be available in any of the three do-
mains (frequency, time and space). Understanding these advantages, telecom authorities are
recently trying to regulate dynamic spectrum access [166] facilitated by cognitive radios.
From this perspective, a solid understanding of the current spectrum utilization is required,
in order to provide an essential input to academic research, industry and regulators.
It is anticipated that the next generation wireless networks (5G) [167] will largely rely on
small cells deployment and Device-to-Device communication, that are characterized with
highly dynamic traffic load, where DSA techniques could cope with this rapid nature of
traffic. In addition, recent broadband disaster recovery solutions (such as the EU funded
project ABSOLUTE [4]), are investigating the possibility to employ DSA concept to effi-
2. Aerial and D2D Physical Channel 77

ciently access the damaged network’s spectrum while recovering from the disaster.
Another prospective implementation of DSA is investigated from the vehicular networks
perspective in [168, 169], as vehicular communication can facilitate safer and more efficient
transportation system (see Sec 2.2 by allowing the exchange of safety and information mes-
sages between vehicle-vehicle and between fixed infrastructure-vehicles. In addition, vehic-
ular radars are essential sensories for future cars, where autonomous decisions can be made
related to dynamic cruise-control and collision-avoidance system. Given the density of ve-
hicles in some areas, a considerable radio resources are required to enable these prospective
technologies. Where, DSA and CRs can play a significant facilitator in the implementation
of vehicular communication.
A handful of research works have been conducted to determine the spectrum occupancy
in different locations in the world (Germany, Netherland, Singapore, Spain, New Zealand
and China) [72–76, 170], most of which have focused on static measurements of the spec-
trum, only taking frequency and time dimensions into account. While some other research
papers [168, 169] have focused on Vehicular Dynamic Spectrum Access (VDSA), address-
ing spectrum occupancy of TV channels and trying to identify the behaviour of TV white
spaces from a vehicular perspective.
In this section, we present the results of spectrum monitoring experiment conducted in
three different urban environments, characterized with different population density and de-
gree of urbanization, for the greater Melbourne area in Australia. The spectrum occupancy
is then determined by comparing the received signal level to a certain detection threshold
based on the noise power of the measurement tool. The occupancy results are categorized
with respect to the surrounding population density as obtained from the Australian Bureau
of Statistics [2]. Comparing to the previous work, where spectrum measurements are rather
case specific to the particular locations of measurement. The main contribution of this sec-
tion can be seen in two main points:

• It provides an initial understanding of the wideband spectrum occupancy, as seen from


the prospective of frequency and spatial dimensions.

• The measurements are conducted in three different urban environments with varying
range of population density, and the results are categorized based on the geo-density
of population.

The results from the spectrum measurements reported here are expected to guide spectral
allocation for cognitive radios especially targeting Device-to-Device communications and
small cell networks.
2. Aerial and D2D Physical Channel 78

Fig. 2.25 Illustration of the experiment setup, and the measurement equipment.

2.4.1 Measurement Setup


The main tool in our measurement setup is the hand-held FieldFox Spectrum Analyser (SA)
characterized with a low average noise level around -150 dBm/Hz (when the internal pre-
amplifier is active). The spectrum analyser is connected to a custom made discone antenna
[171] mounted on top of an SUV car. The setup is interfaced via Ethernet port to a laptop
equipment with MATLAB® . A capturing script configures the parameters of the SA, then
periodically requests spectrum sweeps for the range 400 to 6000 MHz with 1.0 MHz reso-
lution bandwidth. In order to capture the geographic effect on the measurements, we utilize
a standard USB-GPS receiver, and stamp each sweep with the corresponding location and
time. In Fig. 2.25, the diagram of the setup is illustrated.
We summarize the measurement parameters in Table 2.13, noticing that the resolution
bandwidth is selected as 1.0 MHz in order to speed up the sweep cycle, that is to allow more
location-dependent analysis, and to capture any rapid changes in the environment [72]. This
wide resolution bandwidth comes at the cost of higher noise power. The total travel route as
obtained from the measurement log was 248 km, performed over 3.8 hours.
2. Aerial and D2D Physical Channel 79

Table 2.13 Measurement Parameters

Parameter Value

Frequency Span 400 - 6000 MHz


Resolution Bandwidth 1.00 MHz
Sweep Time (Average) 3,950 ms
Noise floor (Average) -149.55 dBm/Hz
Antenna System Gain (Average) -7.35 dBi
Antenna Polarization Vertical
Total Route Length 248 km
Total Measurement Duration 3.8 hours
Preamplifier Gain 20 dB

Measurement Regions

The main scope of our experiment is to measure the spectrum occupancy from the per-
spective of a mobile vehicle experienced in different classes of urban environments. Ac-
cordingly, we first accurately describe the sensed environment in terms of the surrounding
population density, utilizing the Australian population geographic maps, published by the
Australian Bureau of Statistics [2], then we segregate the raw measurement data into three
different groups corresponding to high, medium and low population densities. The overlay
of the travelled route is depicted in Fig. 2.26 on top of the population density geographic
map of greater Melbourne area.
As it can be noticed from the figure, the travel route is divided into three distinct groups,
according to the degree (the class) of urbanization:

• Class 1: High population density class (over than 8,000 persons/km2 ), corresponds
to Melbourne Central Business District (CBD). We refer to this class as dense urban
environment.

• Class 2: Medium population density class (from 500 to 5,000 persons/km2 ), corre-
sponds to a typical suburban environment.

• Class 3: Low population density class (less than 500 persons/km2 ), corresponds to a
typical rural environment.
2. Aerial and D2D Physical Channel 80

Fig. 2.26 The travel route of the experiment in greater Melbourne is divided into 3 classes
or regions. The left bar indicates the population density as per the Australian Bureau of
Statistics. [2]

Antenna Calibration

In this experiment we have utilized a wideband discone antenna [171]. In order to guarantee
a homogeneous detection of the sensed signal, we perform gain calibration inside an ane-
choic chamber with the aid of a network analyser and a calibrated horn antenna acting as a
reference gain antenna, where the gain of the discone antenna can be calculated using the
following expression:

GDis. = S12Case1 − S12Case2 + GRef. , (2.27)

where S12Case1 and S12Case2 are the insertion loss measured by the network analyser cor-
responding to the following cases: (Case 1) one port is connected to the reference antenna,
and the second port to the discone antenna, (Case 2) both ports are connected to identical
reference horn antennas. The term GRef. represents the gain of the reference horn antenna as
per the factory calibration report. This calculation is performed of each measurement point
corresponding to a certain frequency. The resulting gain-frequency plot of the antenna is
depicted in Fig. 2.27, noticing that the measurement includes the effect of the cable loss.
2. Aerial and D2D Physical Channel 81

Fig. 2.27 The discone antenna gain, with/without the effect of the attached cable loss.

Noise Analysis and Detection Threshold

From the perspective of cognitive radios, spectrum sensing is the key element in understand-
ing the surrounding radio environment, where diverse sensing methods are proposed in the
literature [172]. However, when utilizing a spectrum analyser, the only feasible choice is to
use the energy detector method, which requires no prior knowledge about the signal being
measured.
In energy detector method, the collected signal energy is compared to a predefined
threshold, we call it θ , where the channel occupancy is simply identified when the sig-
nal energy is above this threshold. In a cognitive radio, determining the optimum value of
the detection threshold depends on both internal receiver’s noise and the monitored signal’s
strength. However, in our particular case we can only obtain knowledge about the internal
noise of the spectrum analyser.
In order to understand the impairing noise statistics, we record 1,000 samples for each
frequency point (400 to 6000 MHz) while connecting a calibrated matched load (50Ω). Then
we obtain the histogram distribution for these samples, as depicted in the Fig. 2.28, where
the colour indicates the normalized intensity of the histogram, equivalent to the Probability
Density Function PDF of the noise power.
At a certain frequency line (a vertical line) the probability that the noise power N to be
2. Aerial and D2D Physical Channel 82

greater than a threshold θ ( f ) is equivalent to the Complimentary Cumulative Distribution


Function (CCDF) obtained at θ ( f ), expressed as:

CCDFN( f ) (θ ( f )) = P[N( f ) > θ ( f )]. (2.28)

Accordingly we can obtain the threshold that corresponds to a certain level of a false alarm
value. The latter is defined as the probability of judging a channel as occupied, while in fact
it is not. That is due to the contribution of the noise power in raising the sensed signal level.
If x is the actual received signal and n is the noise signal, we can write the total apparent
sensed signal energy Es seen by the spectrum analyser during the integration interval ∆Ts as
the following:
Es = ∑ [x(i) + n(i)]2 , (2.29)
i∈∆Ts

where i is merely the collected samples index within the integration interval ∆Ts (sensing
interval). Then the false alarm probability can be expressed as:

p f = P[Es > θ × ∆Ts |Ho ], (2.30)

where Ho is the hypothesis of having an unoccupied channel. We can notice that in the
case of Ho , the false alarm probability is simply equal to the noise CCDF, because the total
apparent sensed signal energy will be Es = ∑i∈∆Ts [0 + n(i)]2 = N∆Ts .
In our measurement we select a false alarm value at p f = 5%, to minimize the errors
in the occupancy evaluation, where a moving average is employed to smooth the inherited
fluctuations in the threshold level. The window of the moving average is selected to cover
20 frequency samples, i.e. 20 MHz. The resulting detection threshold (the fifth-percentile)
is depicted in Fig. 2.28.

2.4.2 Spectrum Occupancy Results


In this section we present the results of the spectrum measurements. We analyze the raw
samples collected over all sweeps in Fig. 2.29 as a spectrogram segregated into three sub-
plots corresponding to the urban environment class.
Raw power-level samples fed from the spectrum analyser are compared to the detection
threshold θ ( f ), accordingly the channel occupancy at a certain frequency f in a certain
2. Aerial and D2D Physical Channel 83

Fig. 2.28 Noise analysis of the experiment’s spectrum analyser, represented in terms of a
normalized histogram. Indicating the chosen detection threshold of 5% CCDF.

sweep number (spectrum sample) j is determined according to the following rule:


(
1 (Busy) : Ps ( f , j) > θ ( f )
B( f , j) = (2.31)
0 (Free) : Ps ( f , j) ≤ θ ( f ),

where Ps ( f , j) is the sensed signal power at frequency f and a sweep number j. The rep-
resentation of (2.31) is a two dimensional Boolean matrix as depicted in Fig. 2.30, as an
occupancy graph, where the figure only shows the results up to 3GHz since no significant
spectrum utilization was detected above this frequency (the setup is 400 to 6000 MHz).
The null utilization appearing above 3 GHz is referred to the fact that this band is al-
located to multiple special applications such as aeronautical, satellite and radio astronomy
[173]. We suspect that the transmission over this band might be occurring below the detec-
tion threshold of our setup, especially the satellite downlink, requiring a high gain antenna
to receive an interpretable signal.
In order to quantify the level of spectrum occupancy, we calculate the duty cycle of the
measured signal, that is the number of the occupied samples divided by the total number of
collected samples. In other words, the duty cycle represents the percentage of time when
a channel is busy. For a binary occupancy analysis, the calculation is performed as the
2. Aerial and D2D Physical Channel 84

Fig. 2.29 The spectrogram obtained from the raw samples collected in the three different
urban classes. From top to bottom (high, medium and low) population density.

Fig. 2.30 The occupancy graph obtained by comparing the raw samples against the detection
threshold. Dark spots indicate a busy spectral resource (B = 1) while the white spaces indict
free spectral resources (B = 0). From top to bottom (high, medium and low) population
density.

following:
M
Occupied Samples ∑ j=1 B( f , j)
D( f ) = = , (2.32)
Total Samples M
where M is the total number of sweeps. The duty cycle can take values within the range
D( f ) ∈ [0%, 100%], indicating the level of occupancy of a certain channel. We plot the duty
cycle in Fig. 2.31 for the three classes of urban environment. It can be easily noticed that
the duty cycle is different between the three classes, concluding that if a channel has a high
utilization in the dense urban environment does not imply that it can’t be opportunistically
accessed in another urban environment such as rural or suburban.
The frequency range 402 - 460 MHz is allocated to multiple types of users, mainly Mete-
orological satellite uplink, radio astronomy, and mobile UHF narrow band radios (licensed
and amateurs) [173]. This band could be considered as a candidate for DSA by oppor-
2. Aerial and D2D Physical Channel 85

Fig. 2.31 Duty cycle obtained by comparing the number of occupied samples to the total
number of collected samples for each frequency line. From top to bottom (high, medium
and low) population density.

tunistically accessing multiple narrow band spectrum holes left by licensed and/or amateur
users. A special care is required for radio astronomy and space research bands, since radio
telescopes demands extremely low terrestrial interference.
The band 520 - 820 MHz is originally allocated to analog and digital TV broadcasting,
however, new allocation to LTE (Long Term Evolution) has taken place in 700 MHz band
(694 - 820 MHz). The apparent occupancy variation in the 520 - 694 MHz band can be re-
ferred to the low signal-to-noise ratio when receiving a TV signal with a street-level antenna
inside a dense urban environment. Identifying spectrum holes in this band requires further
investigation with smaller bandwidth measurements.
Cellular networks in Australia originally operates in three main bands 900 MHz, 1800
MHz and 2100 MHz [174], later the bands 700 MHz and 2500 GHz were allocated in 2013
(under the Digital Dividend Act [175]). We can notice the large difference in spectrum uti-
lization of LTE bands between the three classes of environments, where network operators
are naturally deploying the eNBs (LTE base stations) in locations expected to provide higher
2. Aerial and D2D Physical Channel 86

return on investment. These bands might be considered as potential for location-dependent


dynamic spectrum access. In addition, uplink sub-bands might also be considered as a can-
didate for DSA techniques, a trend that is progressively aiming to utilize these sub-bands
for Device-to-Device communication (D2D), as recently advised by 3GPP [25]. D2D can
greatly enhance the network coverage by relaying transmission between eNBs and User
Equipment (UEs) (see Sec. 4.1).
The band 918 - 926 MHz (centered at 922 MHz) is dedicated for Industrial, Scientific
and Medical application (ISM) in addition to RFID. Systems operating over this band should
be designed to be interference resilient. We can notice a very low utilization of this band in
both rural and suburban environment, thus it might be considered as a potential candidate
for DSA applications.
We can notice a null duty cycle in the frequency band 960 - 1700 MHz in suburban
and rural environments against a very low duty cycle in dense urban environment. A large
portion of this band (about 316 MHz) is allocated to aeronautical radionavigation 960 - 1215
MHz, 1300 - 1350 MHz and 1559 - 1610 MHz. These bands are carrying a Safety-of-Life
Service, and they are protected from interference by the International Telecommunication
Union [176], thus these bands are not feasible for DSA. The frequency range of 1400 - 1427
MHz is also protected for radio astronomy since the hydrogen spectral line (1420 MHz) has
very low atmospheric absorption. Another protected region is the band 1,260 - 1,300 MHz
which is dedicated to radio navigation and defence applications, thus it is very difficult to
operate secondary users in these bands, not only because of the technical difficulties, but
also due to the expected legislative issues.
Some of the allocated spectrum for mobile satellite services and satellite broadcasting
lays within the range of this experiment, however it is not possible to capture such signals
without directional antennas, due to the low SNR. Dynamic spectrum access for satellite
bands is being investigated in the literature [177, 178], where initial studies indicate possible
mutual use of such bands between terrestrial services (such as cellular communication), and
satellite services.
Our setup was not able to detect significant transmission above 3 GHz, where the ma-
jority of this band is designed for high gain directive antennas, such as point-to-point links
and satellite-earth transmission. Accordingly we cannot comment on the utilization and the
opportunity to use DSA, while some literature papers suggest the possibility to use underlay
DSA techniques in this band [177].
2. Aerial and D2D Physical Channel 87

2.5 Research Question 1 Conclusion


In this chapter we have addressed research question 1 by studying the path-loss behaviour in
urban environment for mmWave D2D link and demonstrating the feasibility of utilising the
mmWave spectrum for conducting LTE-A D2D communication over ISM bands, however
this can only be achieved by adopting techniques such as beam switching and beam forming,
exploiting the strong reflections and diffraction in the radio path caused by urban structures.
In addition, we have provided an overview of the current DSRC system from the European
standardisation perspective, and illustrated the main characteristics of the vehicular radio
channel. Propagation experiments were conducted at 5.8GHz for three different types of
environments in Melbourne city and the results were presented. The path-loss exponent and
the shadowing standard deviation were estimated from the conducted measurements. The
measurement results can be summarised as follows, firstly the line of sight path-loss expo-
nent is quite depended on surrounding environment ranging from 1.5 to 3.7 and secondly
the conductive rooftop of the car is having a negative impact on the antenna gain reducing
the received power around 2dB to 10dB. Moreover, the shadowing standard deviation was
also computed with a limited amount of time averaging associated with it. We also provided
a statistical generic Air-to-Ground RF propagation model for Low Altitude Platforms, that
can substantially facilitate the planning efforts of airborne wireless services, since the RF
planning can be performed based on merely simple urban parameters, rather than depend-
ing on site-specific 3D-models that are unlikely to be easily available and updated. The
air-to-ground path-loss showed a clear tendency towards two distinct propagation groups
that were thoroughly discussed and analyzed for outdoor receivers. The last contribution of
this chapter provided an initial understanding of the spectrum occupancy in different urban
environments, seen from the perspective of a mobile vehicle. Results indicated a vast dif-
ference in the spectrum utilisation between different urban environments. The chapter also
shed a light on the possibility of utilising dynamic spectrum access in different bands of the
spectrum. We conclude that a potential exploitation of the radio spectrum could be achieved
if dynamic access techniques are utilised, or by exploiting the capabilities of cognitive ra-
dios in D2D and aerial networks, ensuring no harmful interference to the primary user is
caused.
Chapter 3

Coverage and Rate Analysis and


Optimisation

Efficient and effective deployment of wireless infrastructure is essential for meeting end-
user expanding traffic demand by guaranteing a certain level of KPIs (key performance
indicators). A continues cycle of planing-optimization is usually required in operational
radio networks, where radio engineering utilize simulation tools to estimate the coverage
and performance prior the actual deployment of new base stations or new features. These
simulation tools are great tactical response to the dynamic changes in the network, however
it is also required to perform strategic planning for the long-term objectives of the operator.
From this perspective, mathematical and analytic tools are seen to fill the gap in balancing
accuracy and simplicity in estimating the network performance. This chapter addresses re-
search question 2 through three key contributions: (i) cellular coverage and rate estimation
under generic channel fading in Sec. 3.1, (ii) LAP coverage and rate estimation in Sec. 3.2,
and (iii) LAP altitude optimization in Sec. 3.3. We depict in Fig 3.1 a highlight of where
this chapter is focusing, compared to the general layout presented previously in Fig 1.3.
Where In this chapter we provide an analytic framework for computing the expected
downlink coverage probability, and the associated rate of cellular networks, where base
stations are distributed in a random manner. The provided expressions are in computable
integral forms that accommodate generic channel fading conditions. We develop these ex-
pressions by modeling the cellular interference using stochastic geometry analysis, then we
employ them for comparing the coverage resulting from various channel fading conditions
namely Rayleigh and Rician fading, in addition to the fading-less channel. Furthermore,
we expand the work to accommodate the effects of random frequency reuse on the cellular
coverage and rate. Monte-Carlo simulations are conducted to validate the theoretical anal-

88
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 89

Fig. 3.1 Performance estimation of cellular and aerial networks.

ysis, where the results show a very close match. Furthermore, in this chapter we study the
achievable coverage and information rate of an aerial communication link with a low altitude
platform. The aerial link is analyzed by considering random fading and shadowing effects
of the wireless transmissions together by considering a uniform random spatial distribution
of the terrestrial users accessing the aerial base station (AeBS). The building structure and
density of the coverage area are modeled using the ITU-R statistical city model. For a given
urbanization of the environment/region we analytically quantify the achievable coverage
probability and the rate for a given user on the ground at a particular altitude of the AeBS.
Furthermore, we analytically find the optimal altitude of the AeBS to maximize the terres-
trial coverage. Simulations are conducted to verify our analysis and a close match between
the analytical and simulation based results is observed. After that, this chapter provide an
analytical approach to optimize the altitude of such platforms in order to allow the maxi-
mum exploitation of their coverage capabilities over an urban environment. The solution
is presented in a simple mathematical form, which is a function of the maximum allowed
path-loss and of the statistical parameters of the urban environment as defined by the In-
ternational Telecommunication Union. Furthermore, we present a closed form formula for
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 90

predicting the probability of geometrical line of sight between a low altitude platform and a
corresponding ground receiver based on the same environment statistical parameters.
This chapter is based on the following publications contributing towards sections 3.1,
3.2, and 3.3 respectively:

1. Akram Al-Hourani, and S. Kandeepan, “On Modeling Coverage and Rate of


Random Cellular Networks under Generic Channel Fading”, Wireless Networks,
Springer, 2015. DOI: 10.1007/s11276-015-1114-x. (Published) [128] [128]

2. Akram Al-Hourani, S. Kandeepan, Geoff Kaandorp , William Glenn and Abbas


Jamalipour, “Coverage and Rate Analysis of Aerial Base Stations”, IEEE Letters on
Wireless Communication (Submitted)

3. Akram Al-Hourani and S. Kandeepan, S. Lardner "Optimal LAP Altitude for Max-
imum Coverage," IEEE Letters on Wireless Communications, vol.3, no.6, pp.569 -
572, Dec. 2014. DOI: 10.1109/LWC.2014.2342736. (Published) [129]
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 91

3.1 Cellular Networks Performance Analysis Under


Generic Channel Fading
The vast deployment scale of cellular communication has made it as one of the most ubiq-
uitously available piece of infrastructure. The notable expansion rate of cellular networks
is referred to the accelerating demand generated by mobile users, where network operators
are endeavouring to bridge the gap between traffic load and the available network capacity
by deploying additional base stations (BS). It is anticipated that within the next 5 years a
data-traffic growth of around 10 folds will take place in cellular networks alone [6]. The
locations of the deployed base stations, are usually constrained by many factors such as
economical, urban planning codes, and the availability of land/utility etc... these factors, are
very difficult to control and to predict, which leads to an increasing randomness in the BS
locations, where the theoretical hexagonal model is no longer feasible [3, 89]. Due to the
increasing complexity of cellular network, designers and researchers utilize simulation tools
for predicting network coverage and performance. Such approach is widely accepted in the
industry, however it can not give an analytical insight of the influence contributed by the
vast simulation parameters. Rather, it provides a detailed case-specific solution with neither
tractability nor flexibility.
Analytical insight of network dynamics is an essential enabler for strategic planning and
long-term economical modeling [127]. And in order to capture the increasing irregularity
of the network deployment, stochastic geometry models [77, 78, 80, 179, 180] are recently
gaining a paramount interest for studying wireless cellular networks. Stochastic geome-
try allows the analytical understanding of the performance of modern cellular technologies
such as cognitive radios [12], heterogeneous networks, fractional frequency reuse [181] and
device to device communications, in addition to the fundamental coverage and capacity of
the cellular network [3, 82, 88, 89]. The most popular assumption in the literature for the
radio power fading is the Rayleigh channel, where the probability distribution function of
the received power is taking a simple exponential shape. This assumption allows tractability
and enormously simplifies the computation of expressions. However, the desired link might
favour a better performance than Rayleigh model that is usually considered as the worst case
scenario fading [182].
Understanding this gap in the literature, we propose in this section an analytical ap-
proach for studying the coverage and the rate of cellular networks under generic channel
fading conditions, accommodating not only the small scale fading behaviour of the channel
but also the possible effects of shadowing variation. We first model the cellular interference
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 92

in a random cellular network, and then we study the expected performance metrics as spa-
tially averaged over the entire network. In addition we demonstrate the effects on the cov-
erage probability resulting from different channel fading scenarios namely; (i) fading-less
channel, (ii) Rayleigh channel and (iv) Rician channel. We verify our analytical approach
using Monte-Carlo simulations by running repeated random network deployments and ob-
taining the spatial average of the signal-to-noise-plus-interference ratio (SINR). These re-
sults are used to validate the analytical calculations obtained using the integral forms. The
contribution of the section could be summarized in the following points:

• It provides a generic formula (the coverage equation) for calculating the expected
service success probability (or the coverage probability) in random cellular networks,
under general channel fading conditions.

• The coverage equation is flexible to allow different fading models for the serving
signal from one side and the interfering signals from the other side.

• It provides a practical method to compute the expected rate in a random cellular net-
work, without resorting to complicated simulations.

3.1.1 Related Work


Several recent introductory works are available on stochastic geometry in the context of
wireless networks of which we list [80, 102, 183]. However, some of the earliest work on
this regards dates back to the 1970-1990 such as [127, 184]. Since then, several leaps have
taken place, for example the work in [103] draws a mathematical framework for the statis-
tical distribution of the interference generated by random wireless networks, where in our
derivation of cellular interference we follow a similar approach, but taking into considera-
tion the specific properties of cellular networks. Other works related to interference can be
found in [185] addressing slotted ALOHA interference topic assuming a Rayleigh fading
channel, while the authors in [186] address a general fading channel and obtain the opti-
mum transmission probability in slotted ALOHA network. In the context of interference
also, the work in [87] addresses clustered interferes but it is distinguished by addressing the
amplitude and phase of the interference where the interfering signals can interact construc-
tively or distractively. Mentioning that the vast of the literature deals with the aggregated
power of the interferers, that is the algebraic sum of the power of all interfering signals.
In [86] the authors employ stochastic geometry analysis on studying intercell interference
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 93

coordination (ICIC) by muting the transmission from K number of neighbouring stations on


specific resource blocks.
The tractability facilitated by the Poisson point process (PPP) attracts researches to rep-
resent the BS locations according to this process. However, other studies in this field cap-
ture the possible repulsion between base stations, utilizing determinantal point processes
[88, 89]. The accuracy of PPP is proven to increase when a heavy shadowing conditions
affect the network [187], making PPP a valid assumption in most of practical network de-
ployment scenarios.
Applying stochastic geometry for studying cellular communication is an appealing ap-
proach for what it can yield of analytical estimations of the different attributes affecting
such networks. For example the work in [3] addresses the probability of coverage in cel-
lular networks assuming a Rayleigh fading channel affecting the serving signal, while the
work in [85] extends the same approach for multi-tier heterogonous cellular network, where
base stations are implemented with different power and capacity levels. The work in [90]
provides a mathematical framework to compute the expected cellular rate without the need
to obtain the coverage probability, the utilized method depends on the moment generating
function of the interference.
The main difference in our work presented here, is that the coverage and capacity esti-
mations can be obtained for any stochastic fading channel model, with the freedom to select
different stochastic processes for the serving signal and the interfering ones.

3.1.2 Network Model


Achieving high accuracy in estimating network performance requires system level simula-
tion, usually performed for a specific wireless technology that is implemented on a determin-
istic geometrical environment. Such simulations are quite useful in practical deployments of
networks, for example, when wireless operators are deploying a new service, or when they
are upgrading their infrastructure. However, an analytical tractable approach is preferred to
get insight of the different factors contributing to the network performance [3, 88, 89, 186],
these factors include (but not limited to) base stations density, fading models, resource allo-
cation. Although tractable analysis can be achieved using simplified approximation of the
determinist hexagonal models [163], however, these models are very simplistic and might
not reflect the true behaviour of the network, knowing that practical cellular network de-
ployments include vast randomness in the location of base stations. From this perspective,
stochastic geometry is widely used in the research field to model the random location of
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 94

base stations, where for simplicity it is common to assume a homogenous Poisson Point
Process (PPP) to model the BS locations, which emulate a complete randomness in deploy-
ment locations. Accordingly, and in order to preserve the tractability we adopt the PPP
network model assuming homogeneous BS intensity of value λ (BS per unit area). The
point process itself is denoted as Φ = {Xn ∈ R2 }n∈N and is assumed to take place in the
two dimensional Euclidian space R2 . Mobile users are typically associated to the BS of
higher received power, which is characterised with random behaviour (small scale fading
and shadowing), accordingly the cellular boundaries are rather probabilistic. However, tak-
ing aside the fading effects, we can draw the average cellular boundary of each of the PPP
points, simply by taking its Voronoi cell, defined as the region where all users are closer to
the serving BS from any other BS [80]:


V (Xn ) = {u ∈ R2 : ||Xn − u|| ≤ ||Xi − u|| ∀Xi ∈ Φ \ {Xn }}, (3.1)

where V (Xn ) is the Voronoi cell of a base station Xn , and Φ is the set of base stations. The
structure of the cellular system will then be called the Poisson Voronoi Tessellation (PVT)
[179], we depict a sample realization of a PVT layout in Fig. 3.2.
The reason that we can rely on PVT for determining the cell association is that the mean
path-loss is monotonically increasing with respect to distance, so having a closer distance
to a certain BS will result better average received power, than any other BS.
We consider that all base stations are having the same transmit power Po , that is a homo-
geneous network, rather than a multi-tier heterogenous network composed of base stations
of variable power capabilities. No power control is accounted in our model, so that base
stations are assumed to continuously transmit a constant power level. Without loss of gen-
erality, we study a user located at the origin, where its statistical behaviour is typical for all
other users in the network. We call this mobile user, as the typical user, where we estimate
the network performance of this user for all possible spatial realizations of the random net-
work. In other words, we implicitly assume a homogenous distribution of network users, so
that having the performance of the typical user will reflect the spatial average of all users in
the network.

3.1.3 Channel Model


Electromagnetic signals travelling between a BS and a receiver encounter power losses due
to the propagation nature of the surrounding environment, resulting mainly from scattering,
diffraction, reflection and absorption. These power losses are characterized with random
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 95

500
PVT
Typical User
400
Serving BS
Interfering BSs
300

200
Y-Axis [Unit Length]

100

0 Ro

-100

-200

-300

-400

-500
-500 -400 -300 -200 -100 0 100 200 300 400 500
X-Axis [Unit Length]

Fig. 3.2 Cellular Poisson Voronoi tessellation, indicating the typical user located at the ori-
gin.

behaviour and usually categorized into two distinct groups according to the their rate of
change, namely small scale fading and large scale fading or shadowing, where the random
effect of these two categories is independent. Large scale fading results mainly from the
electromagnetic shadowing of obstacles, the random behaviour of the shadowing is mod-
elled here as a random variable denoted as g, where it is widely accepted to be considered
to follow a log normal distribution according to the following:

g = exp (σ N) : N ∼ N (0, 1) (3.2)

and σ = ln(10)
10 σdB represents the standard deviation, usually σdB is provided. The Gaussian
distribution of zero mean and unity standard deviation is denoted as N (0, 1).
On the other hand, the small scale fading is modelled by a generic random variable h,
that can represent any of the common fading channel models, such as Rayleigh, Rician and
m-Nakagami [151].
While the mean loss due to the distance (the path-loss) is modeled in a log-distance
relation [182][62], so that at a location x the mean path-loss between the origin and x is
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 96

given by the following:



l(x) = ||x||−α , (3.3)

where ||.|| represents the Euclidian measure in R2 , i.e. the distance between a source base
station and the mobile station under study. Accordingly the resulting received power at a
certain location will have the following expression:

PRX = Po .g.h.l(x), (3.4)

where Po represents the common power at which all base stations are transmitting. Not-
ing that in this section we represent random variables in bold for convenience and ease of
interpretation.

3.1.4 Modeling Cellular Interference


In our model, we assume that the BSs have a unity reuse factor. That is, for the typical
mobile user, all BSs except the serving one are interfering the downlink signal. Then the
aggregated interference is given by

I= ∑ Pognhnl(x) = ∑ Pnl(Rn), (3.5)


ΦI ΦI

where ΦI = Φ \ {Xo } is the set of interferers, Xo represents the serving BS, and {Pn }n∈N+
is a random variable vector having identical and independently distributed (i.i.d) elements,
so that P = Po gh. The BSs’ distances {Rn }n∈N+ constitute a random vector.
The illustration of a typical receiver located at the origin is shown in Fig. 3.3, where it
is important to note that according to the assumed cellular association, all interfering BSs
should be located outside the ball b(o, Ro ) of radius Ro and centred at the origin, where Ro
is the distance to the serving BS, which is the contact distance to Φ [183].
The statistical distribution of the interference cannot be obtained for a generic case
[103]; however, we can still deduce its Characteristic Function (CF) φI (ω), where the CF for
a random variable X is defined as ϕX (ω) = E eiωX . If the characteristic function is iden-
 

tified, we can utilize the Gil-Pelaez’s inversion theorem [188] to compute the Cumulative
Distribution Function (CDF) of X according to the following:

1 1 1
Z ∞
FX (x) = − Im [ϕX (ω) exp(−iωx)] dω. (3.6)
2 π 0 ω

Proposition 1. The characteristic function of the aggregated interference in a cellular net-


3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 97

Fig. 3.3 All interfering base stations are located outside the ball b(o, Ro ).

work is given by:


ϕI (ω) = exp (−2πλ β ) , where (3.7)

Z ∞
β (ω) = [1 − ϕP (ωl(r))] rdr. (3.8)
Ro
Proof. We start from the definition of the characteristic function of the interference, where
the expectation should be performed over (i) the stochastic processes in P and (ii) over the
geometrical stochastic process of ΦI :

" !#
 iωI 
ϕI (ω) = E e = EΦI EP exp iω ∑ Pn l(Rn )
ΦI
" #
= EΦI EP ∏ exp (iωPnl(Rn))
ΦI
" # " #
(a) (b)
= EΦI ∏ EP [exp (iωPnl(Rn))] = EΦI ∏ ϕP (ωl(Rn)) , (3.9)
ΦI ΦI

where (a) follows from the fact that the combined channel process P = Po gh is independent
of the geometrical process and (b) follows directly from the definition of the CF. Now we
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 98

can apply the probability generating functional of a homogeneous PPP on R2 [80], where
for a function f (x) the following relation is satisfied:
" # Z 
E ∏ f (x) = exp −2πλ [1 − f (x)]rdr . (3.10)
ΦI I

The integration variable r ∈ I is the distance range where the active interferers are located,
that is I = (Ro , ∞). Accordingly, we can write (3.9) as the following:
 Z ∞

ϕI (ω) = exp −2πλ [1 − ϕP (ωl(r))] rdr . (3.11)
Ro

Hence, Proposition 1 is proved.

For a log-distance mean path-loss model, (3.8) will yield:


Z ∞
1 − ϕP ωr−α rdr , noting that
 
β (ω) = (3.12)
Ro

ϕP ωr−α = Eg,h exp iωr−α Po gh .


  
(3.13)

Accordingly, we can rewrite β as the following:


Z ∞

1 − exp iωr−α Po gh

β (ω) = Eg,h rdr
Ro
" #
Ro 2 (−iPo ghω)2/α
   
−2 −α −2
= Eg,h − + Γ , −iPo ghRo ω − Γ , (3.14)
2 α α α

where Γ(.) and Γ(., .) are the Gamma and the incomplete Gamma functions respectively.

3.1.5 Modeling Coverage Probability


Signal to Interference and Noise Ratio (SINR) is an important measure that can determine
the link throughput and the availability of the wireless service. The SINR represents the
strength of the target signal compared to the counterpart interferers’ combined power plus
the thermal noise generated inside the receiver’s electronics. The latter can be represented
S
as an Additive White Gaussian Noise (AWGN). The SINR is expressed as: SINR = I+W ,
where S = Po ho go l(Ro ) is the desired signal which carries the needed information from the
serving BS, I is the aggregate interference power, and W is the AWGN noise power. The
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 99

random variables ho and go model the small scale and large scale fading respectively of the
serving BS channel.

Theorem 1. The probability that a receiver to be covered by a certain level of cellular


wireless service is given by:
Z   
Po ho go
pc = Eho ,go FI −W fRo (r)dr . (3.15)
r>0 Trα

Proof. The outage of wireless service occurs when the SINR level at a receiver falls below
a threshold T . Accordingly, we can express the link-level success probability at a given
distance Ro and a given ho , go as:
 
S S
pL = P[SINR ≥ T |ho , go , Ro ] = P[I ≤ −W ] = FI −W , (3.16)
T T

where FI is the cumulative distribution function (CDF) of the interference I. Thus, the
success probability is found by averaging over Ro , ho , and go as:

pc = ERo ,ho ,go [pL ] = Eho ,go [ERo [pL ]] . (3.17)

Accordingly, we can conclude the result in (3.15) by applying the expectancy rule over the
contact distance Ro , having a probability density function of fRo (r) = 2λ πr exp −λ πr2


in a PPP cellular network. We should note that FI is computed using Gil-Pelaez’s inversion
formula given in (3.6).

We call (3.15) the coverage equation, constituting the main result of this work allowing
the evaluation of the averaged network-level success probability. The various dynamics
affecting the cellular service success probability are visualized in Fig. 3.4, namely (i) the
base station density, (ii) the common base stations’ power, (iii) the path-loss model, (iv)
large scale fading model, (v) small scale fading model, (vi) noise level, and finally (vii) the
target SINR threshold.

3.1.6 Network Performance Analysis


In this section, we first experiment a channel affected by a path-loss only, then we account
for the small scale-fading impairing both the serving BS signal and the interfering signals.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 100

Fig. 3.4 The different dynamics affecting the coverage probability.

3.1.7 Path-Loss Only Scenario


By taking the effect of the path-loss only (i.e. h = 1 and g = 1) we can produce an initial un-
derstanding of the cellular coverage dynamics, representing the upper bound of the channel
performance. According to 3.14, β can be reduced to:

Ro 2 (−iPo ω)2/α
    
−2 −α −2
β (ω) = − + Γ , −iPo Ro ω − Γ . (3.18)
2 α α α

We substitute different values of the path-loss exponent α, assuming here that the cel-
lular network is interference limited, so that the noise can be neglected (i.e., W → 0). The
results are counter-intuitive, since as indicated in Fig. 3.5, a better coverage in a random
cellular network is achieved for higher values of the path-loss exponent, indicating that a
heavier path-loss environment affects the aggregated interference more quickly than affect-
ing the serving signal power. Secondly, we compare the coverage probability for different
intensities of BSs assuming a constant path-loss exponent. The results indicate that the BS
intensity has insignificant effect on the coverage probability when the network is interfer-
ence limited. Note that the same observation was reported in [3] and [88] but for a Rayleigh
fading channel.

3.1.8 Rayleigh Fading Scenario


In this scenario, we assume that both the serving and the interfering signals are impaired
with Rayleigh fading. Namely, an exponential distribution random variable with a unity
mean, where the probability density function PDF of h is given by fh (x) = exp(−x). Ap-
plying this to (3.14), and by neglecting the shadowing variations (i.e. g = 1), the result can
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 101

1
α=2.5
0.9 α=3
α=4
0.8 α=4.5
Coverage Probability (p c) [ ]

α=5
0.7

0.6

0.5

0.4

0.3

0.2

0.1

-1 0 1
10 10 10
Service SNR Threshold (T) [ ]

Fig. 3.5 The coverage probability, when adopting log-distance path-loss channel model
without considering fading effects.

be reduced to the following form:

Ro 2 π Ro 2+α
   
2/α 2π α +2 2 iRo α
β (ω) = − + (−iPo ω) csc +i 2 F1 1, , + 2, − ,
2 α α (α + 2)Po ω α α Po ω
(3.19)

where 2 F1 (., ., ., .) is the hypergeometric function.


We perform numerical integration to calculate the coverage probability as per (3.15) for
5 different path-loss exponent values. We observe that higher path-loss exponent values
have a favorable effect on the service success probability. However, Rayleigh channel gives
lower success probability than fading-less channel, even though both the interference and
the serving signals are affected by the same fading behaviour. We stress the point that the
difference between the framework presented in this section and the one in [3] is the flexibility
provided in choosing the fading model of the serving channel, so it is not limited to Rayleigh
only. This comes at the cost of more complex integral computation. We verify the results of
the coverage equation with the ones obtained in [3], and depict the comparison in Fig. 3.6.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 102

1
α=2.5
0.9 α=3
α=4
0.8 α=4.5
Coverage Probability (p c) [ ]

α=5
0.7 Curves from [3]

0.6

0.5

0.4

0.3

0.2

0.1

-1 0 1
10 10 10
Service SNR Threshold (T) [ ]

Fig. 3.6 The coverage probability, for log-distance path-loss and Rayleigh small scale fad-
ing, showing a comparison with the results of [3].

3.1.9 Rayleigh Interferers with Rician Serving Signal


Rician distributed fading can represent a wireless channel with more flexibility by tuning
the K factor which represents the ratio of the LoS power to the sum of the powers from the
defused multipath components. The probability distribution function describing a Rician
fading channel gain is given as:
p 
−x(K+1)−K
fh (x) = (K + 1)e Io 4xK(K + 1) , (3.20)

where Io (.) is the modified Bessel function of the first kind. In this scenario we assume
that the serving signal follows a Rician distribution, while the interfering signals follow a
Rayleigh distribution, thus β (ω) follows (3.19). The only difference will be in calculating
the expectancy in the coverage equation (3.15), that is over a Rician distributed ho . The
results are illustrated in Fig. 3.7, observing that the coverage probability is more sensitive
to the distribution of the serving signal for lower SINR thresholds, a case which represents
the edge users of the cell.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 103

1
Wider performance gap
0.9 for lower SINR threshold (T)
0.8
Coverage Probability (p c) [ ]

0.7

0.6

0.5

0.4

0.3
α=3, K=0 (Rayleigh)
0.2 α=3, K=10 [dB]
α=4, K=0 (Rayleigh)
0.1 α=4, K=10 [dB]

-1 0 1
10 10 10
Service SNR Threshold (T) [ ]

Fig. 3.7 The coverage probability, for Rayleigh interfering signals and Rician serving signal.

3.1.10 Frequency Reuse


It can be clearly noticed how low is the SINR performance in the previous illustrated sce-
narios, also we note that increasing the density of base stations does not lead to a change in
the coverage performance when the network is interference limited. Accordingly, resource
management techniques should be applied to mitigate the co-channel interference between
cells [86]. The spatial reuse of radio resource has been always the essence of cellular com-
munication. However, there is a trade off between coverage performance and the spatial
spectral efficiency. For example, when applying a frequency reuse scheme, the spectral effi-
ciency will drop since the available spectrum for each cell will reduce, however a frequency
reuse scheme will lessen the interference and boost the SINR performance.
In order to get a better insight of the expected network performance and how it is affected
by radio resource coordination, we adopt the random frequency reuse scheme due to its
simplicity and tractability. In this scheme, base stations can chose from some ∆ available
radio frequencies, so each BS has a probability of a certain frequency assignment equal to
1
∆ . In this case, the co-channel interference will only be received from base stations utilizing
the same frequency. The layout of the a PVT will appear similar to Fig. 3.8, where co-
channel base stations are colored the same. It is obvious that this channel assignment is not
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 104

optimal, since co-channel cells are allowed to be a mutual neighbours. But as mentioned
before, the random frequency reuse greatly facilitates the mathematical analysis.
1000
Frequency 1
Frequency 2
Frequency 3
750
Base Stations

500
Map Y-Axis [Unit Length]

250

-250

-500

-750

-1000
-1000 -750 -500 -250 0 250 500 750 1000
Map X-Axis [Unit Length]

Fig. 3.8 The layout of a random frequency reuse scheme in a Poisson Voronoi tessellation
cellular network.

In Fig. 3.9 we plot the simulation results of the random frequency reuse scheme, show-
ing the probability of the service success pc at an SINR threshold of T = 10dB verses a
range of reuse factor ∆. An interesting observation is that the channel fading effect becomes
more obvious for higher reuse factor, when comparing the three scenarios as explained in the
subsections of Section 3.1.5. Simulation procedure will be explained in details in Section
3.1.12.
For analytically studying the effect of random frequency reuse, we examine the new dis-
tribution of the interferers, noting that it follows a new PPP, since the independent thinning
of a PPP will also yield a PPP [77] with a new intensity equal to λ∆ . In this case the cover-
age equation (3.15) resulted in Theorem 1, still holds the same except that the aggregated
interference follows a different stochastic distribution; the new CDF of the interference is
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 105

0.8
Coverage Probability pc at T=10dB [ ]

0.7

0.6

0.5

0.4

0.3

Path Loss Only (Sim)


0.2 Rayleigh Fading (Sim)
Rician (S), Rayleigh (I) (Sim)
0.1
5 10 15 20 25
Frquency Reuse Factor (∆) [ ]

Fig. 3.9 The effect of frequency reuse on the service success probability at an SINR thresh-
old of T = 10dB. where (S) refers to the serving signal and (I) refers to the interfering
signals.

denoted FbI so that its characteristic function is given by the following:


 
λ
ϕbI (ω) = exp −2π β , (3.21)

resulting a reduced interference effect, thus a better coverage success probability.

3.1.11 Network Rate


A practical system would perform less than the maximum limit set by Shannon capacity
theorem. Accordingly we utilize a practical method
 [146] to estimate the user’s throughput
SINR
(rate) per/Hz following the expression: ρ = ln 1 + SINR o
, where SINRo is a system spe-
cific parameter considered here as a constant, representing the gap between Shannon limit
and the achievable rate of the system. The estimated network rate can be calculated as:
   Z ∞   
γ v
ρc = E[ρ] = E ln 1 + = ln 1 + fγ (v)dv, (3.22)
γo γmin γo
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 106

where the symbol γ is used to represent the SINR, v is merely the integration variable, while
γ min is the effective SINR limit where no useful communication can take place below γ min .
This proposed method can capture the effect of adaptive modulation and coding schemes
in communications systems. This method requires obtaining the PDF of the SINR, i.e., fγ ,
which can be approximated numerically from the points calculated in plotting the service
d c
success probability presented earlier, since fγ = − dv Fγ (v), where Fγc is the complimentary
cumulative distribution function (CCDF) of the SINR as plotted in Figs. 3.5-3.7. By means
of trapezoidal integration, the resulting rate is presented in Fig. 3.10, showing the effect
of γ min on the network performance, where low influence of γmin can be noted at the high
extent of the γmin -axis.

1.2
Rayleigh Fading
Rician (S), Rayleigh (I)
1
Data Rate [nats/Hz]

0.8

0.6

0.4

0.2

0
10-1 100 101
System Minimum SINR (γ min ) [ ]

Fig. 3.10 The expected network rate obtained using trapezoidal integration of equation
(3.22), versus system’s minimum SINR γmin and γo = 1. (S) refers to the serving signal
and (I) refers to the interfering signals.

Alternatively, at the cost of additional computational hurdle we can utilize the following
theorem to calculate the expected rate directly from the system’s parameters:

Theorem 2. The expected rate of a cellular network can be calculated from the following
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 107

expression:
Z   
Po ho go
Z ∞
ρc = Eho ,go FI −W fRo (r)dvdr . (3.23)
r>0 ρmin γo (ev − 1)rα

Proof. We need to obtain the expectancy of the rate over three stochastic processes: (i) the
spatial properties of the point process, (ii) the channel fading distributions of the serving
signal, and (iii) the distribution of the interfering signals. Hence
  
γ
ρc = E ln 1 +
γo
Po ho go Ro −α
   
= Eho ,go ,Ro EI ln 1 +
γo (I +W )
Po ho go R−α
Z ∞     
(a) o
= Eho ,go ,Ro P ln 1 + > v dv
ρmin γo (I +W )
Z ∞   
Po ho go
= Eho ,go ,Ro P I< −W dv , (3.24)
ρmin γo (ev − 1)Rαo
 
where ρmin = ln 1 + γγmino
is the minimum achievable data rates, (a) follows from the fact
that, for a positive random variable X, E[X] = 0∞ P(X > v)dv, a method which was also
R

used in [3]. The final result of the theorem follows from averaging over Ro .

3.1.12 Simulation Results


In order to verify the analytic method, we perform Monte-Carlo simulations for a total
number of users exceeding 500,000. The power received from BSs are combined at every
mobile user taking into consideration the stochastic effects of the radio channel by randomly
generating the individual channel gain. The SINR value is stored for each user. After
that, we obtain its CCDF. Note that the CCDF of the SINR is equivalent to the coverage
probability because: Fγc (T ) = 1 − Fγ (T ) = P(γ > T ) = pc . The three different scenarios
explained in Section 3.1.5 are simulated with α = 4, and the results are depicted in Fig.
3.11. We observe a close matching between the analytical integration and Monte-Carlo
simulations. Also we observe how the performance is bounded by the deterministic channel
conditions (fading-less) as the upper bound and the Rayleigh fading as the lower bond.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 108

1
Path Loss Only (Sim)
0.9
Rayleigh Fading (Sim)
Rician (S), Rayleigh (I) (Sim)
0.8
Coverage Probability (p c) [ ]

Path Loss Only (Analy)


0.7 Rayleigh Fading (Analy)
Rician (S), Rayleigh (I) (Analy)
0.6

0.5

0.4

0.3

0.2

0.1

-1 0 1
10 10 10
Service SNR Threshold (T) [ ]

Fig. 3.11 Coverage probability simulation results of the three proposed scenarios, compared
to the analytical integration obtain using the coverage equation. (S) refers to the serving
signal and (I) refers to the interfering signals.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 109

3.2 Aerial Communication Performance Analysis


Aerial platforms have recently gained a wide popularity as key enablers for rapid deployable
wireless networks where coverage is provided by onboard radio heads. These platforms are
capable of delivering essential wireless communications for public safety agencies during
the aftermath of natural disasters, or to patch surge traffic increases in commercial networks
due to massive public events. An example of the research and development efforts in air-
borne network recovery solutions is the European Commission project ABSOLUTE [4],
focusing on Low Altitude Platforms (LAP) to complement rapidly deployable terrestrial
communication infrastructure.
In this section, we provide an analytic framework to estimate the performance of aerial
coverage, namely we investigate the coverage success probability Pcov and the expected rate
(throughput) ρ of an aerial platform. The estimation is based on the specific nature of the
aerial radio channel that largely differs from legacy terrestrial models commonly used in
cellular communications and broadcasting services. We utilize the model developed by the
EU-project ABSOLUTE and called the ABSOLUTE Model [123] (see Sec. 2.3), taking into
consideration the statistical effects of both the shadowing and small scale fading. In order
to give an insight of the areal coverage performance, we employ the introduced framework
in a realistic urban environment having architectural statistics based on the recommended
parameters of the international telecommunication union (ITU) [94]. These statistics de-
scribes the density and hight distribution of urban structures, having a major effect on the
performance of aerial radio channel. Furthermore, we provide practical simulation results
in addition to the analytic estimations, proving the consistency and validity of the proposed
framework.
Previous works have attempted to study the performance of coverage provided by a low
altitude platform such as the work in [92] which was based on link-level system-specific
simulation of LTE. Similarly, the work in [91] was dedicated to obtain LTE link performance
in a deterministic simulation environment, however taking into account the specific nature of
air-to-ground channel. Optimizing the altitude of aerial platforms was studied in [129] (see
Sec. 3.3) aiming to provide the largest possible coverage spot, the solution was presented
in an analytic easy-to-compute form. The main contribution and novelty of this letter is
the analytic formulation of the coverage and rate based on the ITU-R parameters, allowing
rapid estimation of the performance metrics without relying on time-consuming and system-
specific simulations. Thus, optimizing key parameters of the aerial platform such as the
altitude and transmit power becomes a simple task based on the provided framework.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 110

3.2.1 Air-to-Ground Channel Model


Radio propagation in air-to-ground (AtG) channel differs from the well known terrestrial
propagation models used for cellular, broadcasting and for indoor wireless communication
[189]. Radio waves in AtG channel travel freely without obstacles for large distance before
reaching the urban layer of the man-made structures. The latter layer causes the signal
to scatter, diffract, and refract leading to an excessive amount of losses on top of the free
space path-loss incurred between the aerial platform and a ground receiver. Accordingly, the
signal path loss in an AtG channel can be modelled in two parts [70, 123, 129], firstly is the
free space path loss (FSPL) and secondly is the excessive path-loss denoted as η. The path
loss η is dependent on the elevation angle φ at which the AeBS is seen from the ground
receiver. As per our path loss model (known as the ABSOLUTE model), the excessive
path-loss is also dependent on the urban statistical parameters defined by the international
telecommunication union (ITU) in the recommendation P.1410-2 [94], these parameters are:

• Parameter αo : The ratio of built area to the total area.

• Parameter βo : The mean number of urban structures per unit area, measured in
buildings/km2 .

• Parameter γo : A scale parameter that describes the buildings’ heights distribution


2
according to Rayleigh random variable ∼ γo2 exp −x
2γ 2
.
o

Accordingly, the total AtG channel loss can be expressed as [123]: Lξ = Lo ηξ (αo , βo , γo ),
where Lo represents the free space path-loss between the aerial platform and a terrestrial
user, and ξ refers to the propagation group; the propagation group represents the set of
receivers sharing similar statistical behaviour of the channel, e.g. line of sight (LoS) group
and non-line of sight (NLoS) group. In general, we can consider M propagation groups
ξ = 1 . . . ξ = M, where the probability that a receiver belongs to a certain group depends on
the elevation angle φ [70, 123] and the urban statistical parameters. We call this probability
as the group occurrence probability denoted as pξ which is conditioned on (αo , βo , γo ). In
the event of having only LoS and NLoS propagation groups (which is the most dominant
grouping) represented by {ξ = 1, ξ = 2} respectively the group occurrence probabilities are
then given by P[ξ = 1] = p1 and P[ξ = 2] = p2 .
The excessive path-loss is captured in the random variable η ξ (αo , βo , γo ), where its
statistics is also dependent on the propagation group ξ . For a generic model of the AtG
excessive path loss we can write η ξ (αo , βo , γo ) = ηoξ · gξ · hξ , where for a given ξ we have
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 111

ηoξ represents the mean value that depends on the elevation angle φ , and we have the ran-
dom variables gξ and hξ represent the random shadowing and small scale fading power
gains respectively.

3.2.2 Performance Analysis


For a noise limited communication system, the statistics of the Signal to Noise Ratio (SNR)
determines the performance metrics. The SNR is given by: γ = NS , where N is the noise
power. In order to have a successful communication, the SNR level should exceed a cer-
tain threshold θ determined by the modulation and coding scheme of the wireless system.
Accordingly, the coverage success probability is expressed as:
" #

Pcov = P [γ > θ ] = P gξ > (3.25)
WT Lo ηoξ hξ

if we condition on a certain ground distance R, a propagation group ξ , and a certain small-


scale gain h, then equation (3.25) is merely the Complementary Cumulative Distribution
Function (CCDF) of the random variable g, calculated at the point x = WT LNθo ηo h
. Accord-
ξ ξ
ingly the average coverage probability can be found by obtaining the expectancy over the
conditioned variables R, ξ , and hx i:
" !#

Pcov = E 1 − Fg
WT Lo ηoξ (αo , βo , γo )hξ
" " " !###

= 1 − ER Eξ Ehξ Fgξ . (3.26)
WT Lo ηoξ hξ

Thus by applying the expectancy in (3.26) we can formulate the needed expression to cal-
culate the Pcov presented in (3.29), where Fgξ is the CDF of the random variable gξ . As
well as fhξ and fR are the probability density functions (PDF) of hξ and R respectively, and
m is the total number of the propagation groups. Accordingly, if the statistical model of the
AtG channel is known, we can obtain the expected coverage success probability. Further
utilization of formula (3.29) will be elaborated in the upcoming subsection.
On the other hand, if the SNR statistics are known for a certain link, then the rate can be
estimated in a straightforward manner as the following [146]:
   Z ∞  
γ min [γmax , v]
ρ = E ln 1 + = ln 1 + fγ (v)dv, (3.27)
γo γmin γo
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 112

where fγ is the PDF of the SNR, γ min and γ max are the effective SNR limits where no useful
communication can take place below γ min , and no further rate enhancement can be achieved
above γ max , and γo is an empirical system-specific parameter modelling the gap between the
fundamental channel capacity and the practical performance of the system. Noting that the
resulting rate is measured in natural unit of information nats/s/Hz. fγ can be calculated by
differentiating the CDF of the SNR Fγ , where this CDF is the complement of the coverage
probability calculated in (3.29) against a range of θ , i.e. Fγ (θ ) = 1 − Pcov . Accordingly, by
applying Leibniz integral rule to formula (3.29), fγ (θ ) is written as :

∂  
fγ (θ ) = 1 − Pcov
∂ θ" " " !###
∂ Nθ
= ER Eη Eh Fg , (3.28)
∂θ WT Lo ηoξ (αo , βo , γo )h

where the resulting rate is formulated in (3.30) by substituting (3.28) in (3.27) and applying
the partial differentiation w.r.t θ .

!
m Z Rmax Z ∞

Pcov (αo , βo , γo ) = 1 − ∑ Fgξ pη fhξ (u) fR (r)du dr. (3.29)
ξ =1 0 0 WT Lo ηoξ u
! 
m Z ∞ Z Rmax Z ∞ 
N Nθ min [γmax , v]
ρ(αo , βo , γo ) = ∑ fg ln 1 + pη fhξ (u) fR (r)du dr dv. (3.30)
ξ =1 γmin 0 0 WT Lo ηoξ u ξ WT Lo ηoξ u γo

3.2.3 Model Implementation


To this end, we have formulated a generic framework for calculating the average perfor-
mance of both the coverage and the rate of an aerial platform. This framework accom-
modates both small and large scale fading and is capable of capturing multiple different
propagation groups. In practice, it is sufficient to consider two groups only as suggested
in [123] and [70] corresponding to line-of-sight (LoS) and non-line of sight (NLoS) situa-
tions between the aerial platform and a ground user. The propagation group probability is
defined in [123] equation (12) and (13). Furthermore, we consider a Rician fading impair-
ing the line of site propagation group (ξ = 1) referring to the statistical channel model in
[67–69]. The Rician fading accounts for a dominant multipath component with adjustable
factor represented in its K-factor. On the other hand the NLoS propagation group (ξ = 2)
is considered impaired with Rayleigh fading [36, 190], indicating no single dominant mul-
tipath component in the this propagation group. The PDF of the fading channel gain h2 is
given by fh2 (x) = exp(−x).
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 113

We also take the into account that the channel is impaired with log-normal shadowing
[70, 123] having the following form gξ = exp(σξ N) : N ∼ N (0, 1). σξ = ln(10)10 σξ dB , and
σξ dB is the decibel standard deviation provided by the AtG model where it is a function of
the elevation angle φ , defined in the ABSOLUTE AtG model [123] equation (9). And the
PDF of the log-normal shadowing channel factor is well defined in the literature [182].
We transform the dependency on φ into a dependency on the ground distance R since
simply φ = arctan HR . Thus, for a fixed altitude H, we have φ and R explicitly related. For


convenience, we list all notations and parameters in Table 3.1.


If the receiver is located at a certain certain ground distance R from the projection of
the platform, we can obtain the coverage using the coverage formula (3.29) but without
performing the expectancy over R, thus:
" " !##

PcovR = Eξ Eh 1 − Fgξ , (3.31)
WT Lo ηoξ (αo , βo , γo )h

where by substituting values from Table 3.1, we can plot the graph in Fig 3.12 that de-
picts the coverage probability versus a range of ground distance R assuming a fixed service
threshold T = 10 dB.

Fig. 3.12 The coverage probability PcovR (αo , βo , γo ) versus the users’ ground distance R.
The utilized SNR threshold is T = 10. The desecrate points represent the simulation while
continuous curves are the analytical performance in (3.31).

If terrestrial users are homogenously distrusted within a circular region of radius Rmax ,
then the PDF of the ground distance fR (r) can be written as: fR (r) = R2r
2 : r ∈ [0, Rmax ].
max
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 114

Table 3.1 Notations and Symbols

Symbol Value Explanation


Pcov - The coverage probability
ρ nats/s/Hz The System’s throughput
WT 32 dBm †1 Transmitter’s EIRP
θ Variable Target service SINR threshold
Rmax 1000 m Coverage region
R Variable in m Ground distance
ξ {1,2} Propagation groups (LoS, NLoS)
m 2 Number propagation modes
Freq 2000 MHz Frequency of operation
h {Rician, Rayleigh} Multi-path fading factor
K 10 dB Rician K-Factor
g Log-normal Shadowing factor
H 1000 m Platform altitude
N -98 dBm †2 Thermal noise power
Lo Variable Free space path-loss
γo 1 The performance gap
γmax ∞ Maximum allowable SNR
γmin Variable SNR Threshold by the system

†1 ABSOLUTE Project Deliverable 2.3 [4].


†2 Noise figure of 9dB and bandwidth 5MHz (3GPP TR.814).

Then by substituting in (3.29) we plot Pcov for a varying range of the service threshold θ as
depicted in Fig. 3.13, noticing the effect of the urban environment on the coverage.
In order to avoid additional computational efforts we utilize the points of the curve in
Fig. 3.13 to obtain the sampled probability density function of the SNR by calculating the
differences between adjacent elements of the vector Pcov , of length N, where n represents
the discrete sample index:

fˆγ (n) = {Pcov (T (n + 1)) − Pcov (T (n)) : ∀n ∈ [1, N − 1]}, (3.32)

which is the probability of the SNR to lay between T (n) and T (n + 1), i.e. fˆγ (n) =
P[T (n) < γ < T (n + 1)], where T (n) is the threshold vector sampled at n. The corre-
spondingnSNR sample is calculated asothe mid point between T (n) and T (n + 1), i.e.:
γ̂T (n) = T (n+1)−T
2
(n)
: ∀n ∈ [1, N − 1] , accordingly the vector of the estimated rate is
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 115

Fig. 3.13 The coverage probability Pcov (αo , βo , γo ) versus the SNR threshold. The desecrate
points represent the simulation while continuous curves are the analytical performance in
(3.29).

calculated as:
N−1    
min [γmax , γ̂(n)]
ρ̂(n) = ∑ ln 1 + × fˆγ (n) , (3.33)
arg min[γ̂−γmin ]
γ o
n

a method which represents the discrete numeric equivalent of (3.27). For illustrating the rate,
we utilize γo = 1 representing the Shannon capacity limit, as well as we utilize γmax → ∞,
on the other hand we vary the lower integration limit γmin and observe its effect on the rate
performance. The curves in Fig. 3.14 represents the obtained rate versus γmin . We can notice
from this figure that the average rate holds pretty flat for practical values of γmin indicating
its minor influence.

3.2.4 Performance Optimization


One of the main applications of this analytic framework is the ability to rapidly optimize
the performance of an aerial platform by tuning key design parameters such as its altitude.
It is intuitive to state that higher altitudes will lead to lower coverage in log-distance path-
loss models, however, in an AtG we observe two contradicting effects when increasing the
altitude; firstly, the probability of getting a line of sight is enhancing leading to more users
in the LoS group that have better received power. Secondly, the increasing altitude will lead
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 116

Fig. 3.14 The expected rate of platform versus a range of minimum SNR γmin . The desecrate
points represent the simulation while continuous curves are the analytical performance in
(3.30).

to a higher path-loss. Thus, finding the optimum altitude that maximizes the probability
of coverage is not a trivial approach anymore, as Fig. 3.15 depicts the relation between
Pcov and the altitude of the aerial platform H for a certain SNR threshold T = 20dB and
for different urban environments, indicating the optimum altitude of each. The general
 
approach of finding the optima is formulated as: Hopt = arg maxH Pcov .
It is worthy to note the difference between the altitude optimizing approach here and the
one proposed in [129] and illustrated in Sec 3.3. In the latter the goal is to maximize the
coverage patch of the aerial platform given the maximum allowable path-loss (or the SNR
threshold), while here, the goal is to maximize the probability of coverage within a given
circular patch of radius Rmax where users are expected to be.

3.2.5 System Simulations


We deploy around 31,000 ground receivers within the coverage disk b(0, Rmax ) in a lattice
layout, then we calculate the signal to noise ratio at each of these receivers resulting from the
AtG stochastic channel presented in Sec. 3.2.3. Accordingly the coverage probability can
be obtained by counting the number of receivers having an SNR above a certain threshold
θ divided by the total number of receivers, as the following: P̃cov = N1r ∑i=N r
i=1 1γi >θ , where γi
is the SNR of receiver i, Nr is the total number of receivers, and 1γi >θ is a boolean variable.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 117

Fig. 3.15 Coverage performance versus platform’s altitude, for a fixed coverage area Rmax ,
showing the optimum altitude for a given SNR threshold T = 20dB.

The simulation results are depicted in Fig. 3.12, 3.13 and 3.14 as discrete points. The results
indicates a good match with the analytical curves generated based on formulas (3.31), (3.29)
and (3.30).
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 118

3.3 Aerial Coverage Optimization


Broadband wireless networks are increasingly adopted by users of mission critical commu-
nications, such as public safety agencies and first responders. This adoption is motivated
by the unprecedented development in these networks in terms of capacity and efficiency,
compared to the legacy (second generation) mobile communication systems. However, as
any cellular network, the communication is largely dependent on fixed infrastructure (base
stations) that could be severally disrupted in the case of natural disasters such as floods,
earthquakes or tsunamis. Making the need for finding a rapid and cost-effective temporary
recovery solution an utmost necessity. One of the prospective feasible solutions for realizing
wireless recovery networks is by utilizing airborne base stations. The airborne communi-
cation infrastructure concept has been endorsed by the homeland security bureau in USA
[53]. An example of the recent efforts in airborne network recovery solutions is the ongoing
European Commission project ABSOLUTE [4] focusing on Low Altitude Platforms (LAP).
Due to technical limitations, the number of deployable LAPs could be very limited,
especially during the chaotic aftermath hours of a disaster. This fact mandates a full ex-
ploitation of each of the deployed LAPs by optimizing its altitude in order to provide the
best possible coverage. In this section, we target this issue by providing a mathematical
model capable of predicting the optimum altitude of a LAP based on the statistical param-
eters of the underlaying urban environment. These parameters are described in three folds:
(i) the percentage of build-up area to the total land area, (ii) the number of buildings per
unit area, and (iii) the statistical distribution of buildings heights. An important mediator
parameter in this study is the LAP-to-receiver line of sight probability, which we provide a
closed form for it.

3.3.1 System Model


Low Altitude Platforms (LAP) are quasi-stationary aerial platforms such as quadcopters,
balloons and helicopters, usually characterized with an altitude laying within the tropo-
sphere. In contrary to High Altitude Platforms (HAP) [45, 162] that can reach the upper
layers of the stratosphere. In general, LAPs are much easier to deploy, and are inline with
the broadband cellular concept, since low altitude combines both coverage superiority and
confined cell radius. For the LAP altitude optimization proposed here, we focus on the
lower part of the troposphere, i.e. heights between 200m and 6,000m.
The technology carried by LAPs depends on the end-user’s application, budget and
bandwidth requirements. Applications could be as advanced as LTE-A, Wi-Fi, WiMAX
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 119

Fig. 3.16 Low Altitude Platforms radio propagation in urban environment.

or as legacy as GSM, TETRA or P-25 systems.

3.3.2 RF Propagation Model


Few literature papers are available on characterising the air-to-ground (ATG) propagation
over urban environments, the most comprehensive work in this regards can be found in
[63, 70, 123], where the authors proposed that ATG communication occurs in accordance
to two main propagation groups. These groups are derived statistically in [123], where the
first group correspond to receivers favoring a Line-of-Sight (LoS) condition or near-Line-
of-Sight condition, while the second group generally corresponds to receivers with no LAP
Line-of-Sight but still receiving coverage via strong reflections and diffractions. In [63]
and [70] the propagation groups where similarly classified according to the receivers’ strict
LoS and non-line-of-sight (NLoS) conditions, where each propagation group was studied
independently.
As depicted in Figure 3.16, radio signals emitted by a LAP base station propagate in
free space until reaching the urban environment where they incur shadowing and scattering
caused by the man-made structures, introducing additional loss to the ATG link. We refer to
the additive loss incurred on top of the free space pathloss as the excessive pathloss, which
has a Gaussian distribution [63, 70, 123], however in this study we deal with its mean value
(expectation) rather than with its random behavior, hence η here refers to the mean value of
the excessive pathloss. Another point is that the effect of small-scale fluctuations caused by
the rapid changes in the propagation environment are also not accounted.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 120

Accordingly, The resulting ATG pathloss (expressed in dB) can be modeled as:

PLξ = FSPL + ηξ (3.34)

where FSPL represents the free space pathloss between the LAP and a ground receiver,
and ξ refers to the propagation group. Noticing that, the excessive pathloss η affecting the
ATG link depends largely on the propagation group rather than the elevation angle which is
depicted θ in Figure 3.16.
In order to find the spatial expectation of the pathloss denoted as Λ (measured in dB)
between a LAP and all ground receivers having a common elevation angle θ , we will apply
the following expectation rule:

Λ = ∑ PLξ P(ξ , θ ) (3.35)


ξ

where P(ξ , θ ) represents the probability of occurrence of a certain propagation group which
is strongly dependent on the elevation angle. In our study we are following the assumption
of the two dominant propagation groups that strictly correspond to the LoS condition. Ac-
cordingly ξ ∈ {LoS, NLoS}, and the group probabilities are linked as the following:

P(NLoS, θ ) = 1 − P(LoS, θ ) (3.36)

In the next subsection, we present a closed form expression for obtaining this probability.

3.3.3 Modeling Line of Sight Probability


The International Telecommunication Union (ITU) in its recommendation document [94]
suggests a remarkable method for finding the probability of geometrical LoS between a ter-
restrial transmitter at elevation hTX and a receiver at elevation hRX in an urban environment.
This probability is dependent on three statistical parameters related to the urban environ-
ment:

• Parameter α: Represents the ratio of built-up land area to the total land area (dimen-
sionless).

• Parameter β : Represents the mean number of buildings per unit area (buildings/km2 ).

• Parameter γ: A scale parameter that describes the buildings’ heights


 distribution
 ac-
2
cording to Rayleigh probability density function: f (H) = γH2 exp −H
2γ 2
, where H is
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 121

the building height in meters.

Following the mathematical steps in [94] we can write the resulting LoS probability in
a single equation as:
   2 
(n+ 12 )(hTX −hRX )
m   hTX − m+1 
P(LoS) = ∏ 1 − exp − (3.37)
  
2γ 2 
n=0   

p
where m = floor(r αβ − 1) and r is the ground distance between the transmitter and the
receiver, as depicted in Figure 3.16, while n is merely the product index. It is worthy to
mention that the geometrical LoS is independent of the system frequency, also that equa-
tion (3.37) is generic and can be used for any hTX and hRX heights. A similar geometric
approach was followed in [93] to determine the theoretical likelihood of the LoS in built-up
areas, however the study was not based on the ITU parameters. On the other hand, practical
measurements were presented in [95] for satellite to ground LoS estimations. In the partic-

Fig. 3.17 The calculated line-of-sight probabilities, with their related S-Curve fitting for
different urban environments.

ular case of a LAP we can disregard hRX since it is much lower than the average buildings
heights and the LAP altitude. Also, the ground distance becomes r = h/ tan(θ ), where h
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 122

is the LAP altitude. It is important to note that the resulting plot of the series in (3.37)
will smooth for large values of h, accordingly P(LoS) can be considered as a continuous
function of θ and the environment parameters. Plotting this probability in Figure 3.17 for
four selected urban environments [70] Suburban (0.1, 750, 8), Urban (0.3, 500, 15), Dense
Urban (0.5, 300, 20), and Highrise Urban (0.5, 300, 50) for α, β and γ respectively, we can
notice that the trend can be closely approximated to a simple modified Sigmoid function
(S-Curve) of the following form:

1
P(LoS, θ ) = (3.38)
1 + a exp (−b [θ − a])

where a and b are called here the S-Carve parameters.


This approximation significantly ease the calculation of the LoS probability, and also
it allows the analytical approach presented in Section 3.3.4, because the series in (3.37)
cannot be further reduced. In order to generalize the solution we have linked the S-Curve
parameters a and b directly to the environment variables α, β and γ. This linking was
preformed using two variables surface fitting where (α × β ) is assumed as the first variable,
and (γ) as the second. The surface equation yields a two-variables polynomial having the
following form:
3 3− j
z= ∑ ∑ Ci j (αβ )iγ j (3.39)
j=0 i=0

where z represents the fitting parameter a or b, and Ci j are the polynomial coefficients given
in Table 3.2 and Table 3.3, while the surface fitting is depicted in Figure 3.18.
Table 3.2 Surface polynomial coefficients for a

Ci j i 0 1 2 3
j
0 9.34E-01 2.30E-01 -2.25E-03 1.86E-05
1 1.97E-02 2.44E-03 6.58E-06 -
2 -1.24E-04 -3.34E-06 - -
3 2.73E-07 - - -

3.3.4 Finding the Optimum Altitude


In order to analyse the effect of the LAP’s altitude on the provided service, firstly we define
the service threshold in terms of the maximum allowable pathloss PLmax . When the total
pathloss between the LAP and a receiver exceeds this threshold, the link is deemed as failed.
For ground receivers, this threshold translates into a coverage disk (zone) of radius R, since
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 123

Fig. 3.18 S-Curve parameters 3D-fitting as a relation to the urban environment parameters.
Table 3.3 Surface polynomial coefficients for b

Ci j i 0 1 2 3
j
0 1.17E+00 -7.56E-02 1.98E-03 -1.78E-05
1 -5.79E-03 1.81E-04 -1.65E-06 -
2 1.73E-05 -2.02E-07 - -
3 -2.00E-08 - - -

all receivers within this disk have a pathloss that is less than or equal PLmax , as depicted in
Figure 3.19. Mathematically speaking, the cell radius of the coverage zone can be written
as:
R = r|Λ=PLmax (3.40)

Accordingly, the optimization problem is to find the best altitude that will maximize R. In
order to do so, we deduce a relation between the LAP altitude h and the cell radius R. By
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 124

Fig. 3.19 The coverage zone by a low altitude platform.

rewriting equation (3.34) we have:


 

PLLoS = 20 log d + 20 log f + 20 log + ηLoS
c
 

PLNLoS = 20 log d + 20 log f + 20 log + ηNLoS (3.41)
c
| {z } | {z }
FSPL ηξ

where d is the distance between the LAP and a receiver at a circle of radius r, given by

d = h2 + r2 , while f is the system frequency. The FSPL is according to Friis equation
with the assumption of isotropic transmitter and receiver antennas. Refereing to (3.35):

Λ = P(LoS) × PLLoS + P(NLoS) × PLNLoS (3.42)

According to Figure 3.16, we notice that θ = arctan hr . Now we substitute from equa-


tions (3.36), (3.38), (3.40), (3.41) into equation (3.42), then performing some simple alge-
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 125

braic reductions, we can write:

A
PLmax = h

1 + a exp(−b[arctan R − a])
+ 10 log(h2 + R2 ) + B (3.43)

where A = ηLoS − ηNLoS and B = 20 log f + 20 log 4π



c + ηNLoS . The above equation is
implicit, where neither R nor h can be written as an explicit function of each other. In order
to obtain the optimum point of the LAP altitude hOPT that yields the best coverage, we need
to search for the value of h that satisfies the equation of the critical point:

∂R
=0 (3.44)
∂h

i.e. the point at which the radius-altitude curve in equation (3.43) changes its direction.
The optimum altitude of a LAP is strongly dependent on the specific urban environment
condition. We have plotted different radius-altitude curves according to equation (3.43),
that result from the four selected urban environments in Figure 3.20 maintaining a constant
PLmax = 110 dB, and system frequency f =2,000 MHz, while using the following (ηLoS ,
ηNLoS ) pairs (0.1, 21), (1.0, 20), (1.6, 23), (2.3, 34) corresponding to Suburban, Urban,
Dense Urban, and Highrise Urban respectively [123] (measured in dB).
In order to visualize the effect of varying the maximum allowed pathloss PLmax on the
radius-altitude curve and the optimum altitude solution, we have depicted this relation in
the plot of Figure 3.21, where the cell radius is a function of both, the LAP altitude and
the maximum allowed pathloss PLmax , by maintaining a constant environment parameters
(Urban). We can notice that the resulting line connecting the tips of radius-altitude curves,
indicates a constant ratio between R and hOPT , or in other words, there is a certain elevation
hopt
 a constant ratio of R , we call it here the optimum elevation
angle that always satisfies
angle or θOPT = arctan hOPT R . For obtaining the optimum elevation angle, we first rewrite
the expression in (3.43) in terms of θ and R as the following:

A
PLmax = + 20 log(R sec θ ) + B (3.45)
1 + a exp(−b[θ − a])

the optimum point can then be found by solving the equation ∂∂ θR = 0, which yields the
following:
π abA exp(−b[θOPT − a])
tan(θOPT ) + =0 (3.46)
9 ln(10) [a exp(−b[θOPT − a]) + 1]2
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 126

Fig. 3.20 Cell radius vs. LAP altitude curve for different urban environments.

Fig. 3.21 Cell radius vs. LAP altitude curve for different maximum pathloss, in urban
environment.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 127

the solution of equation (3.46) is clearly independent of the maximum allowed pathloss, and
is also unique for a certain set of parameters (a, b, A). Accordingly, it explains the straight
line in Figure 3.21.

3.3.5 Discussion
It is important to note that there is no explicit solution for equations (3.44), and only numer-
ical solution is possible. However, it is a straightforward task to obtain such solution using a
computer-based tool. Another note is that PLmax depends on the communication technology
of the onboard communication system, the nature of the ground receivers and on the target
quality of service. It has been noticed that for large values of PLmax the optimum altitude
rapidly increases to reach values much higher than earth atmosphere, and in this case the
altitude limitation is set by the mechanical properties of the LAP.
3. Coverage and Rate Estimation and Optimization of Aerial and Cellular Coverage 128

3.4 Research Question 2 Conclusion


This chapter has provided a mathematical framework to analytically compute the coverage
and rate of random cellular networks under generic channel fading conditions. Two main
observations have been made. Firstly, the stochastic process of the radio channel largely
affects the coverage performance when considering lower SINR thresholds (e.g., the perfor-
mance of cell edge users). Secondly, the density of base stations does not affect the cellular
coverage when the network is interference limited, regardless of the stochastic process of
the channel. For obtaining the expected rate of a cellular network two methods have been
illustrated. The first method is based on trapezoidal integration and the second method is
based on a computable integration formula. Furthermore, we have presented a novel ana-
lytical framework to estimate the performance of an aerial platform utilising air-to-ground
channel-specific model. We mapped both coverage and rate to the underlaying urban sta-
tistical parameters as defined by the ITU-R. This analytic framework allows a rapid opti-
misation of the key parameters of an aerial platform such as the altitude and power, given
a certain set of constraints including the quality of service requirements, thus, the provided
framework can substantially facilitate the rapid and optimum deployment of such platforms
by just knowing the underlying urban statistics. Simulation results showed prefect matching
with the analytic curves, indicating the validity of the proposed framework. Finally, we have
provided a mathematical model for obtaining the optimum LAP altitude that maximises the
coverage and the utilisation of a LAP in enabling emergency coverage networks. In addi-
tion, as a mediator parameter, we have showed that the geometrical line of sight between a
LAP and a ground receiver can be expressed in a closed form based on the elevation angle
and the urban statistical parameters.
Chapter 4

Coverage Enhancement with Node


Cooperation and Relaying

One of the promising technologies in future wireless networks is the ability of nodes to
communicate directly, thus expanding cellular coverage by relaying uplink and downlink
traffic towards base stations. This features allows a great resilience against sudden network
damages and traffic congestions. Taking into consideration that public safety agencies are
increasingly depending on such networks for their mission critical communications, thus
node cooperation and direct communication is a vital backup in case of a network infras-
tructure failure or a natural disaster. This chapter addresses research question 3 through
three key contributions: (i) cellular coverage enhancement with node cooperation under dis-
aster circumstances in Sec. 4.1, (ii) node cooperation in LTE-Advanced for disaster relief
in Sec. 4.2, (iii) aerial coverage enhancement with relaying in Sec. 4.3. We depict in Fig
4.1 a highlight of where this chapter is focusing, compared to the general layout presented
previously in Fig 1.3.
Where this chapter is based on the following publications contributing towards sections
4.1, 4.2, and 4.3 respectively:

1. Akram Al-Hourani, S. Kandeepan, and A. Jamalipour, “Stochastic Geometry Study


on Device to Device Communication as a Disaster Relief Solution”, IEEE Trans-
actions of Vehicular Technology, vol.65, no.5, pp.3005 - 3017, May. 2016. DOI:
10.1109/TVT.2015.2450223. (Published) [130]

2. Akram Al-Hourani and S. Kandeepan, "Cognitive Relay Nodes for Airborne LTE
Emergency Networks," 7th International Conference on Signal Processing and Com-
munication Systems (ICSPCS), Gold Coast, Australia, 2013 IEEE, pp. 1-9. DOI:

129
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 130

Fig. 4.1 Coverage enhancement with node cooperation and relaying.

10.1109/ICSPCS.2013.6723940 (Published) [131]

3. Akram Al-Hourani and S. Kandeepan, "Temporary Cognitive Femtocell Network


for Public Safety LTE” 18th International Workshop on Computer Aided Modeling
and Design of Communication Links and Networks (CAMAD), Berlin, Germany,
2013 IEEE, pp. 190-195. DOI: 10.1109/ICSPCS.2013.6723940 (Published) [132]

We start by analytically quantifying the cellular network performance during massive


infrastructure failure, where some terminals can play the role of low-power relay nodes
forming multi-hop communication links to assist farther terminals outside the reach of the
healthy network coverage. Namely, we analytically determine the D2D effect in alleviating
the damage caused by the disaster. The study is based on stochastic geometry analysis, and
presents a novel analytical methodology, applicable to wide scenarios of network conditions
and parameters. This methodology is verified through Monte-Carlo simulations for practi-
cal network parameters based on the latest 3GPP recommendations. Then we address the
deployment of a temporary cognitive secondary LTE Femtocell network in order to supple-
ment and enhance the coverage of a regular LTE network for public safety communications.
We propose a novel approach in deploying such cognitive secondary network by exploiting
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 131

the latest LTE-Advanced HetNet capabilities. We also present two interference mitigation
techniques for mitigating the interference caused by the presence of the secondary cognitive
LTE network. Simulation results are presented to show the enhancement in the coverage
when such a secondary network is deployed together with the proposed interference miti-
gation techniques. Finally we propose a novel concept of Cognitive Relay Node for intel-
ligently improving the radio coverage of an airborne LTE emergency network, considering
the scenarios outlined in the ABSOLUTE research project. The proposed network model
was simulated comparing the different cases of deploying relay nodes to complement the
coverage of an aerial LTE network. Simulation results of the proposed Cognitive Relay
Nodes show significant performance improvement in terms of radio coverage quantified by
the regional outage probability enhancement.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 132

4.1 Cellular Damage Alleviation with Node Cooperation


The recent development in broadband wireless networks has added unprecedented level of
reliability and bandwidth efficiency to cellular communications, leading to the emergence
of new user-applications that would not be possible without such enhancements. In fact,
commercial wireless networks are nowadays an irreplaceable gear in modern economies and
societies, not only used for voice and data communications but also as a carrier for many
mission critical businesses and personal applications such as wireless video monitoring,
transportation signals, logistics tracking and automatic power meters reading.
Prospective users of broadband networks driven by these new advancements are the Pub-
lic Safety (PS) agencies who are actively involved in the development of mission-critical
requirements of the Long Term Evolution-Advanced (LTE-A) communication system. In
fact LTE has been endorsed by major public safety organizations in the US as the technol-
ogy of choice for the 700 MHz band [54], and accordingly a 20 MHz spectrum bandwidth
chunk has been allocated for this purpose [40]. Mission-critical communication for appli-
cations such as digital dispatcher, first responder deployments, fire fighters missions and
police forces operations might sometimes occur in off-network locations or during the af-
termath of natural disasters, where the main cellular network might be partially damaged,
or out of reach. In fact, there is an increasing interest recently in improving communication
technologies for disaster recovery operations [133, 154, 157], a recent example is the Euro-
pean commission project ABSOLUTE, where rapid deployable infrastructure is investigated
[4].
In this section, we analytically study the effect of cellular network damage on the spatial
probability of establishing a certain target wireless service. As well as, we quantify the ef-
fect of introducing multi-hop D2D relays [155] in relieving the endured damage. Numerical
examples are presented leveraging the newly released guidelines by the Third Partnership
Project (3GPP) [25] for implementing and simulating Device to Device (D2D) communica-
tions.
Since D2D communication depends on fully distributed terminals, a better way to ex-
plore this technology is by utilizing stochastic geometry that study the average behavior
of network performance where nodes are positioned according to a certain probabilistic
distribution. The stochastic geometry approach is more effective in analysing modern com-
munication networks, as the conventional hexagonal cellular layout is not applicable in most
of the actual deployments [191]. For instance, Fig. 4.2 shows a cellular network layout that
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 133

Fig. 4.2 Random cellular layout according to a PPP, showing cell dominance boundaries
and base stations’ locations.

could appear when base stations (BS)1 are deployed according to a homogeneous Poisson
Point Process (PPP) with a certain base stations density λ (BS per unit area). Clearly we
can notice that the layout is very far from the neat hexagonal model as usually suggested
by 3GPP as a common evaluation methodology [25]. The locations of the deployed base
stations are usually constrained by many factors such as economical, urban planning codes,
and the availability of land/utility etc... These factors, are very difficult to control, resulting
in an increasing randomness of the BS locations, where the theoretical hexagonal model is
no longer realizable [89]. The plausibility of PPP was also verified by comparing with real
deployments in [3] and [85].
Modeling a cellular network using PPP facilitates the tractability, without resorting to
complex system level simulations. These simulations do not usually lead to a profound
relationship between the different system parameters i.e. the input from one side and output
of the simulation from the other side. However, simulation approach can still be used for
1 We use the notation BS and eNB alternatively in this chapter, where we deal with the BS/eNB as a
standalone entity, so each cell has a single serving BS/eNB with an omni-directional antenna.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 134

verifying the stochastic tractable analysis, as followed in [107][192][193]. Similarly, in


this section we present two approaches. The first, predicts the network performance for
different damage scenarios using tractable stochastic analysis. While the second, focuses on
extensive Monte-Carlo simulation for randomly deployed base stations and mobile users.
The contribution of this section can be summarized in the following folds:

• It provides an analytical method for studying the effect of single and multi-hop D2D
relays in alleviating the incurred damage caused by a natural disaster.

• It introduces a novel expression describing the network-level service success proba-


bility, when chain-relaying is adopted in order to extend the cellular coverage.

• It quantifies the disaster alleviation introduced by D2D technology through the new
concept of the damage alleviation buffer (DAB), measuring the amount of damage
that a D2D-supported network can handle when a disaster occurs.

• It provides a realistic numerical study on the alleviation introduced by D2D technol-


ogy, based on the latest 3GPP network parameters.

• It provides a comparison between two disaster propagation scenarios affecting a cel-


lular network, namely a random BS phase-out scenario and a concentrated phase-out
scenario.

The introduced framework is quite helpful for disaster management and response planning,
since it allows authorities to properly predict the performance of commercial and public
safety networks during the aftermath of natural disasters. Accordingly, they will have a
better insight of how to plan the D2D deployment and how to choose the required number
of relay hops.

4.1.1 Related Work


Device to device communication as an underlay to cellular networks [22] is ongoing stan-
dardization in LTE-A [25]. In addition, studying D2D using stochastic geometry tool is
fairly a new topic. A recent work by the authors in [194] investigate the inter-tier and
intra-tier interferences in D2D underlaying cellular networks, taking into consideration the
specific radio channel properties of D2D. On the other hand, several works related to het-
erogeneous cellular networks are presented in the literature. For example the authors in
[3] derive a mathematical framework for predicting the SINR in a cellular network for the
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 135

downlink direction assuming that base stations are distributed according to a PPP. While
the authors in [88, 89] are utilizing α-Ginibre point process, rather than PPP due to the for-
mer capability of modeling the spatial mutual repulsion of base stations, aiming to resemble
the actual deployment of base stations more accurately. Accordingly, the study derives a
computable coverage probability in multi-tier networks.
Studying multi-tier cellular network using stochastic geometry is gaining an increasing
popularity, with the recent emphasis on heterogeneous base stations deployment in modern
cellular networks. For example, the authors in [85] are also deriving a tractable expression
for the SINR resulting from a multi-tier cellular network. Most of the literature papers
assume a certain coordination between the transmitting nodes (or base stations) in order to
mitigates mutual interference. The authors in [192] are focussing on this issue by studying
the capability of cognitive femtocell in achieving high spectrum reuse gain in a two-tier
cellular network.
On the other hand, the authors in [107] address the two-hop cellular system and obtain its
outage probability. The first hop starts from the cellular base station, while the second one
forwards the traffic via a relay node to farther terminals located in dead spots, they assume
that cellular base stations are deployed in a deterministic square lattice layout, while users
have a PPP distribution. The effect of relays mobility is addressed in [105], where the
authors study the theoretical coverage and capacity extensions using mobile relays. The
focus of the section is the impact of mobility on the probability of route establishment
and the expected availability duration. The section addresses a single cell with idealized
circular coverage area, while here we address the network-level performance of D2D relays
in extending network coverage during disasters.
The work in [103] derives mathematical expressions for the downlink coverage and ca-
pacity of two-hop cellular networks utilizing tools from stochastic geometry, while in our
analysis we deal with both the downlink and the uplink connections in order to estimate the
network-level performance, in addition we analyse the coverage expansion resulting from
n-hop relays. The work in [104] addresses the expansion of cellular network coverage us-
ing relay nodes aiming to find the optimum relay location within a cell, where the utilized
simulation scenario constitute of a single circular cell. Similarly the work in [106] tries
to optimize the deployment locations of relays within a hexagonal cell using brute-force
simulation. While a step further is taken in [195] for providing a methodology to calcu-
late the coverage and spectral efficiency resulting from a multi-hop cellular network using
simulation, where base stations are deployed in a deterministic lattice layout.
To this end, the main differentiator in our work is the provided analytic expressions for
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 136

the network performance covering both single hop and multi-hop scenarios using stochastic
geometry tools. The network performance takes into consideration both uplink and down-
link connections. In addition, we introduce a novel expression that describes the chain-relay
success probability in a series form. To the best of our knowledge, the available litera-
ture does not explicitly quantify the alleviation of the network damage when utilizing D2D
technique.

4.1.2 System Model


Our study focusses on Frequency Division Duplex (FDD) implementation under LTE-A,
where the downlink and the uplink utilize two separate frequency bands denoted here as fUL
and fDL . It is well known that LTE is based on the Orthogonal Frequency Division Multiple
Access (OFDMA) technology that allows the utilization of a unity cellular frequency reuse
factor. In order to allow this high reuse scheme, several interference mitigation techniques
are implemented within LTE-A, exploiting the ability of OFDMA technology in assigning
adjacent and granular portions of the radio resource to different users. Accordingly, for a
transitional area between two cells (that is the most affected area by interference), base sta-
tions use coordinated frequency-time radio resources assignments for UEs inside these areas
in order to mitigate intercell interference [196]. Several other techniques are introduced in
LTE-Advanced under the name (eICIC) Enhanced Intercell Interference Coordination [97],
that further mitigate the possible interference in heterogeneous networks. In this section we
use the term interference mitigation factor ηb to denote the virtual reduction of the interfer-
ing base stations power due to the interference mitigation techniques.
Another important point in our model is that D2D communication is taking place only in
the fUL for both directions of communication, as it was decided by 3GPP [25]. In this work
we assume that the relaying UE denoted as (UEr) is located within the network coverage
and is operating as a decode-and-forward relay node for helping a farther UE that is located
outside the cellular coverage area, having a Signal to Interference plus Noise Ratio (SINR)
below a certain threshold Θ i.e. (SINR < Θ). Moreover we assume that the latter UE can
also play the role of a relay node, thus even helping more farther UEs and so on, forming a
multi-hop relay chain as depicted in Fig. 4.3.
In a similar manner to the intercell interference coordination, D2D communication can
favor a certain interference mitigation factor ηr , by exploiting multiple-access spectrum
sharing between the neighbouring D2D transmitters that can take place either in a coordi-
nated manner or in uncoordinated manner (when no signaling is required) using spectrum
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 137

Fig. 4.3 The relay chain principle, for extending healthy cells coverage.

Table 4.1 Interference Matrix

Spectral Band Main Signal Interferers

fDL eNB to UE Other eNBs


fUL1 UE to eNB Other UEs
fUL2 Device to Device Other D2D Transmitters

sensing [126] and cognitive radio (CR) techniques [165]. A centralized resource coordina-
tion might be highly vulnerable to failures during the aftermath of a natural disaster, while
the uncoordinated CR techniques, would be more resistant to such disasters. We model the
effect of this self-organized spectrum sharing using the interference mitigation factor ηr ,
assuming that D2D devices can transmit in slotted time periods.
In this study we assume that the uplink communication from the UEs to their serving
eNB can only occur in the a sub-band designate as fUL1 while device to device communi-
cation takes place in another sub-band designated as fUL2 . The reason for this separation
is that the coordinated spectrum sharing between the cellular network and the underlay-
ing D2D might not be possible in all situations, especially during the partial failure of the
cellular network. However the assigned spectrum ratio between fUL1 and fUL2 could be
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 138

adjusted in a semi-static manner whenever the network conditions allow. Fig. 4.4 depicts
the proposed spectrum sharing concept. In fact, implementing this scheme is a straightfor-
ward task in OFDM-based wireless systems, by assigning different number of subcarriers to
D2D connections and to the cellular uplink connections. Although this scheme might result
lower spectrum efficiency than centralized resource management, however, the urgency of
re-establishing network coverage during the aftermath of a natural disaster supersedes the
importance of spectrum efficiency. The possible interference matrix between the different
network nodes is listed in Table 4.1 and depicted in Fig. 4.5.

Fig. 4.4 The proposed semi-static spectrum allocation.

Fig. 4.5 Interference on cellular and D2D links.


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 139

In order to model the effect of a natural disaster on the wireless network we propose
a damage ratio denoted as D ∈ [0, 1] that represents the percentage of the phased out base
stations over the total number of base stations. Mathematically speaking, if the density of
base stations prior a disaster was λb , then in case that a disaster has occurred, a thinning
factor [197] of (1 − D) will affect the original base station density and the resulting new
density will become (1 − D)λb . The implicit assumption of the random BS phase-out pro-
cess is due to the inherited random nature of the effect of natural disasters. While utilizing
stochastic geometry will allow capturing the average behaviour of the network before and
during the natural disaster. This random phase-out scheme allows tractable analysis and for-
mulation of the network performance. While, for comparison purposes, we explore another
phase-out process in Section 4.1.11 assuming that the effect of a disaster is propagating in a
radial manner out of the central point of the disaster, where base stations are phased-out in
non-random manner.
For studying the performance of the network during a natural disaster, we obtain the
network performance behaviour in terms of the service success probability for a range of
damage ratio D. As the thinning of a PPP will yield another PPP [80], the cellular base
stations after the disaster will still follow a PPP when randomly phased-out.

4.1.3 Preliminaries
The Signal to Interference and Noise Ratio (SINR) is an important measure that affects
the link capacity and the availability of the service. SINR represents the strength of the
serving signal compared to the counterpart interferers’ combined power plus the thermal
noise generated inside the receiver’s electronics. The latter is usually modeled as an Additive
White Gaussian Noise (AWGN) that is characterised by a constant power spectral density.
The SINR is given by the following general formula:

S
SINR = , (4.1)
I +W

where S is the serving signal carrying the needed information, I is the combined interfering
power, and W is the AWGN power. Noting that in order to enhance the readability of the
mathematical symbols in this section, we denote random variables in bold font.
The serving signal traveling from a transmitter to a receiver incurs propagation losses
due to power spread, scattering and absorption of electromagnetic waves. These losses
are characterized with a random behaviour due to the rapid changes in the propagation
environment. We model the average value of these losses using log-distance function (see
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 140

Chapter 1), while we model the variations using Rayleigh fading [182]. This kind of fading
occurs when no Line of Sight (LoS) nor quasi LoS is achievable between a transmitter and
a receiver. Accordingly the received power from the serving transmitter has the following
form:
S = PT hr−α , (4.2)

where PT is the signal power as emitted by the transmitter usually represented by the Equiv-
alent Isotropic Radiated Power (EIRP) that includes the effect of antenna gain, α is the
propagation exponent characterizing a certain propagation environment [123], and r repre-
sents a random variable of the distance between the serving transmitter and the receiver.
The factor h represents the fast-fading Rayleigh random variable, having a unity mean and
an exponential Probability Density Function (PDF) form:

fh (x) = exp(−x) (4.3)

On the other hand, the interference I is caused by all co-channel transmitters utilizing
the same spectrum band, and is represented as the following:

I= ∑ PI hn r−α
n , (4.4)
n∈ΦI

where n represents the points of the interferers’ point process ΦI , hn represents the fading
random variable vector that has independent and identically distributed (iid) elements of the
same form as (4.3), PI is the common interferers’ power.
In this study we assume that the interferers set ΦI is distributed according to a homoge-
nous Poisson Point Process (PPP) over a the entire plane R2 . We call a network as Poisson
network if its nodes are distributed according to a PPP. Without loss of generality we con-
sider a receiver located at the origin in our following analysis.

4.1.4 Link-level Success Probability


A stronger SINR level usually leads to a better achievable link throughput (Mb/s) utiliz-
ing higher Coding and Modulation Scheme (MSC). Accordingly a target application, say a
voice-call for example, can only be achieved when the SINR level is above a certain thresh-
old Θ, where any value below this threshold deems the service as failed. So the probability
of getting a successful link serving a certain receiver can be expressed as per the following:
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 141

Ps = P[SINR ≥ Θ]. (4.5)

In order to to find Ps we employ the comprehensive study on interference modeling


[103], that allows the prediction of the combined interference originated from randomly
scattered interferers affecting a certain user. This user is assumed to be located at distance
ro from its serving transmitter. According to [103] Eq. (35), the link success probability in
a Poisson network is given by:

−1
Γ(1 + b1 )  PI 1 #
"

ro 2b ΘW
 πλIC1/b b
Ps = exp − × exp − π ro 2b θ , (4.6)
PT cos( 2b ) PT

where Ca = Γ(2−a)1−acos(πa/2) and Γ(.) is the gamma function, PI is the common power of
the interferers, and λI is the interferers density. According to [103], b is related to the
propagation exponent as b = α2 . Hence, by utilizing simple algebra we can reduce (4.6) to
the following form:
" 2#
πλI Γ(2 − α2 )Γ(1 + α2 ) PI α
  
ro α ΘW α
Ps = exp − × exp − ro Θ , (4.7)
PT 1 − α2 PT

representing a closed form expression for the link-level performance in a Poisson network
with Rayleigh fading. We succinctly express this function as:

Ps (PT , α, PI , λI ,W, Θ, ro ), (4.8)

where by having the parameters PT , α, PI , λI , W , Θ, and ro we can numerically calculate


the corresponding link-level success probability.
The approach for obtaining the link-level performance of a non-fading channel is slightly
different, firstly we start from equation (4.5):

PT
Ps = P[SINR ≥ Θ] = P[I ≤ α −W ]
ro Θ
 
PT
= FI α −W , (4.9)
ro Θ

where FI is the Cumulative Distribution Function (CDF) of the interference. This CDF can-
not be expressed in a closed form expression, while its Characteristic Function is expressed
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 142

as the following [103]:


 h  π i
−1 1/b 1/b
φI (ω) = exp −πC1/b PI λI |ω| 1 − isign(ω) tan . (4.10)
2b
PT
We can then numerically obtain the CDF at a certain value x = roα Θ − W utilizing the Gil-
Pelaez’s inversion theorem [188]:

1 1 1
Z ∞
FI (x) = − Im [φI (ω) exp(−iωx)] dω. (4.11)
2 π 0 ω

In this section we adopt Rayleigh fading in our analysis. However, non-fading scenario is
provided for comparison purpose only, as it will be illustrated later in this section.
It is important to note that the link-level success probability depends on the distance to
the serving transmitter ro , and accordingly it does not represent the overall network-level
service success probability, which we further investigate in the following sections.

4.1.5 Spatial Analysis


Based on the previous mathematical preliminaries, we construct our tractable analysis in
this section. First, we obtain the link-level success probability for the different types of
links assuming that there is only a single serving transmitter that a receiver can attach to.
Then we derive the network-level service success probability, assuming that a receiver can
attach to the nearest serving transmitter where these transmitters are distributed according to
a PPP. We call the network-level success probability as the network performance and denote
it as κ.

4.1.6 Link-level Performance


In order to consider a certain cellular service as a successful service, it is required to have
both a successful uplink service and a successful downlink service. Thus, mathematically
speaking we can represent the link-level cellular service successful probability as:

Pc = PDL PUL , (4.12)

where PDL and PUL are the probabilities of having a successful downlink connection and
a successful uplink connection respectively. According to the interference matrix in Table
4.1, and by referring to (4.8) we can form these probabilities in term of the main signal and
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 143

the interfering signal, the parameters that should be substituted in equation (4.7) are:

Pb Pb
PDL = Ps (PT = , α = αc , PI = ηb , λI = (1 − D)λb ,W = Wm , Θ = Θm , ro ), (4.13)
Loc Loc

where Pb is the common BS power, Loc is the reference path-loss at a distance of 1 meter
from the transmitter in the cellular link environment, λb is the density of base stations, Wm
is the AWGN power affecting the mobile units, and Θm is the downlink SINR threshold.
While the uplink connection success probability can be found using the following pa-
rameters:

Pm Gb Pm Gb
PUL = Ps (PT = , α = αc , PI = ηm , λI = λm ,W = Wb , Θ = Θb , ro ), (4.14)
Loc Loc

where Pm is the common UE power, λm is the UE density, Wb is the AWGN power affecting
the base stations, Θb is the uplink SINR threshold, and Gb is the base station antenna gain.
The reason that Gb is not included in (4.13) is that we consider the Pb to implicitly include
the effect of the base station antenna gain, while the gain of the receiving UE is unity. All
symbols are explained in Table 4.2. Noticing that the applied reduction factor η to the
interference power is the result of the interference mitigation techniques explained below:

• Downlink: Although the frequency reuse factor is unity in LTE, the neighbouring
base stations exchange scheduling information amongst each other in order to avoid
assigning similar downlink radio resources to UEs located in the mutual edge areas of
the cells. We approximate the reduction in the downlink interference due to this ICIC
capability by the factor ηb and assume that it is constant throughout the network.

• Uplink: We assume that the cellular base stations have the capability of assigning the
uplink resources to UEs in an orthogonal manner, thus trying to reduce the uplink
interference coming from other UEs. This mechanism leads to a virtual reduction in
the uplink interference power, approximated in the factor ηm where we assume it is
constant throughout the network.

Alternative mathematical approximation to this interference mitigation can be achieved by


applying a thinning factors to the interferes density as suggested in [80] when the random
frequency reuse scheme is adopted, other interference mitigation schemes applied using
stochastic geometry can be found in [86].
On the other hand a successful D2D link will require two-way successful communica-
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 144

tion, and can be directly written as:

Pm Pm
PD2D = P2s (PT = , α = αd , PI = ηr , λI = γλr ,W = Wm , Θ = Θm , ro ). (4.15)
Lod Lod

where λr is the density of relay nodes, and γ ∈ (0, 1] is a scalar factor which represents the
portion of the D2D devices that are concurrently transmitting. The value γ = 1 represent the
worst case scenario when all D2D pairs are engaged in active transmission simultaneously,
where in our study we focus on this case.
It is important to mention that the presented model does not account for the possible
limitation in traffic handling capability of D2D relays and of the eNBs, but rather it assumes
that network nodes can serve as many UEs as needed. Accordingly, no blockage due to
radio resources limitation is accounted. This assumption is fairly reasonable in a disaster
recovery situation, where coverage rather that capacity is highly important for the search
and rescue operations. However the limitation might not be negligible when the traffic load
increases, accordingly, our model will lose its accuracy in such case.
By applying the numerical values in Table 4.2 to equations (4.12), (4.13), (4.14), and
(4.15) we can plot the link-level success probability as presented in Fig. 4.6. These nu-
merical values are selected according to the network deployment parameters suggested by
3GPP [198], as well as according to the propagation environments suggested by WINNER-
II project [58]. The plot shows both the non-fading and the Rayleigh cases where the draw-
back of fading can be clearly noticed, although that the fading is applied on both the serving
signal and the interfering signal. In addition, we can notice the sharp decay in the success
probability profile of D2D link, that can be referred to the limit imposed by the thermal
noise, while on the other hand, the smooth decay of the cellular service success probability
profile indicates an interference limited network. The latter observation confirms with the
expected performance of modern cellular networks [3], that these networks are interference
limited rather than noise limited.

4.1.7 Network-Level Performance


So far we have focused on the link-level success probability where the distance between the
serving transmitter and the receiver is known. However, in order to evaluate the average
link-level performance, we need to calculate the expectancy of the link-level probability
over the spatial distribution of the distance between a receiver and its serving transmitter.
In an Euclidian space R2 the probability to find at least one member X of a PPP process
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 145

1
Link-Level Success Probability [ ]

Cellular Downlink (Rayleigh)


Cellular Uplink (Rayleigh)
0.8 Cellular Total (Rayleigh)
Cellular Total (non-fading)
D2D (Rayleigh)
0.6 D2D (non-fading)

0.4

0.2

0
0 250 500 750 1000 1250 1500
Distance to Transmitter [m]
Fig. 4.6 Link-level success probability vs. the distance from the serving transmitter, for a
healthy network with D = 0.

Φ, i.e. X ∈ Φ within a search disc of radius R is well defined in the literature as the contact
distance probability, having a CDF of the form [79]:

FΦ (R) = 1 − exp(−λ πR2 ), (4.16)

which represents the probability that a certain UE to find a serving transmitter from a set
of base stations or a set of D2D servers, within a search radius R. Accordingly we can
deduce the PDF of such probability which represents the distribution of the contact distance
between a UE to the serving transmitter as the following:

d
fΦ (r) = FΦ (r) = 2λ πr exp(−λ πr2 ), (4.17)
dr

forming a Rayleigh distribution with a scaling parameter σ = √ 1 . In order to visualize


2λ π
this probability we plot both the CDF in (4.16) and the PDF in (4.17) in Fig. 4.7 according
to the base station density numerical value in Table 4.2, the expected contact distance is
indicated in the same figure.
Our main target is to obtain the average link-level success probability for both the cel-
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 146

×10 -3
1 2
Nearest eNB Distance CDF [ ]

Nearest eNB Distance PDF [ ]


0.75

r CDF
0.5 1
r PDF

E [r]
0.25

0 0
0 250 500 750 1000 1250 1500
Distance [m]
Fig. 4.7 Nearest eNB distance probability distribution.

lular link and for the D2D link. Accordingly, we calculate the mathematical expectancy
(mean) of the service success probability in accordance to the PDF of the contact distance.
Mathematically speaking, we perform the following integration:
Z ∞
κo = E[Pc ] = Pc fc (r)dr, (4.18)
0

where fc is the contact distance PDF between a UE and its nearest serving eNB, given by:

fc (r) = 2(1 − D)λb πr exp −(1 − D)λb πr2 .



(4.19)

Similarly, the average link-level D2D success probability will be:


Z ∞
Ho = E[PD2D ] = PD2D fD2D (r)dr, (4.20)
0

where the D2D contact distance PDF can be expressed as:

fD2D (r) = 2λr πr exp(−λr πr2 ), (4.21)


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 147

representing the probability distribution of the distance between a UE and the nearest D2D
relay.
We define here three scenarios for coverage: (i) the first scenario, is to have a cellular-
only service, without the assistance of D2D relays. (ii) The second scenario is to get the
service directly form BSs or to get the service routed via a single relay hop, we call this
network as single-hop relay network. (iii) The third scenario is similar to the second with the
ability to get the service via multi-hop relays, we call this network as chain-relay network.
In case of scenario (i), the network-level performance will be equal to κo given in (4.18),
while in case of scenario (ii) the network-level performance will be determined by the ability
to establish a successful communication link through either methods of connection: firstly
through a direct connection to the main cellular network, the event is denoted as c where
P(c) = κ0 , or secondly through a relaying D2D that is attached to the main cellular network,
the event is denoted as d1 where P(d1 ) = H1 = κo Ho . Accordingly, the network-level per-
formance in the case of a single hop relay network is the combined probability of the above
two methods given by the following:

κ1 = P(c ∪ d1 ) = P(c) + P(d1 ) − P(c)P(d1 ), (4.22)

that is because the two events c and d1 are mutually independent. Thus, the resulting
network-level performance is given by:

κ1 = κo + H1 − κo H1 . (4.23)

4.1.8 Chain Relay Network Performance


In a similar manner, for an nth relaying chain, the network-level performance κn is related
to the performance of the (n − 1)th relaying chain network-level performance κn−1 and to
the chained D2D chain performance Hn . Thus, it is expressed as the following:

κn = κn−1 + Hn − κn−1 Hn
= κn−1 + κo Hno − κn−1 κo Hno , (4.24)

since the chained D2D chain success probability is given by:

Hn = κo Hno . (4.25)
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 148

Alternatively, by using q-Pochhammer symbol (q-shifted factorial) [199], the resulting


network-level performance can be expressed in terms of κo and Ho only, as per the fol-
lowing:
κo Hn+1
 
o −1
κn = (κo Ho ; Ho )n κo − 1 − , (4.26)
(κo Ho ; Ho )n+1
where the q-Pochhammer symbol is defined as:

 1
 :n=0
(a; q)n = n−1 (4.27)

 ∏ (1 − aqk ) :n>0
k=0

The proof of equation (4.26) is illustrated in the below. This equation provides a simplified
expression for the expected network-level performance for the nth chain relay network as a
function of the D2D link-level mean success rate Ho and the cellular link-level mean success
rate κo . The proof of equation (4.26), is performed through algebraic rearrangement of
terms by substituting different values of n in equation (4.24). The following derivation was
obtained with the aid of Mathematica [200]:

κ1 = κo + κo Ho − κo2 Ho
κo Ho2 − 1
 
= (1 − κo Ho ) κo − 1 −
(1 − κo Ho ) (1 − κo Ho2 )

κ2 = κo + κo Ho + κo Ho2 − κo2 Ho − κo2 Ho2 − κo2 Ho3 + κo3 Ho3


= (1 − κo Ho )(1 − κo Ho .Ho )
κo Ho3 − 1
 
κo − 1 −
(1 − κo Ho ) (1 − κo Ho .Ho ) (1 − κo Ho .Ho2 )

..
.

κn = (1 − κo Ho )(1 − κo Ho .Ho ) . . . (1 − κo Ho .Hon )


[κo − 1 − κo Hon − 1/ [(1 − κo Ho )(1 − κo Ho .Ho )
(1 − κo Ho .Ho2 ) . . . (1 − κo Ho .Hon+1 )

4. Aerial and Cellular Coverage Enhancement with Node Cooperation 149

κo Hon+1 − 1
 
= (κo Ho ; Ho )n κo − 1 −
(κo Ho ; Ho )n+1

If we chained an increasing number of relay nodes, the network performance will even-
tually reach an asymptotic enhancement level. This asymptotic enhancement (denoted as
κL ) is a function of κo and Ho , and can be found in a simplified expression by obtaining the
asymptote of equation (4.26) when n → ∞, according to the following:

κL = lim κn
n→∞
 
0−1
= (κo Ho ; Ho )∞ κo − 1 −
(κo Ho ; Ho )∞
= 1 − (1 − κo ) (κo Ho ; Ho )∞ , (4.28)

that is because:
lim κo Hn+1
o = 0, (4.29)
n→∞

since κo is finite and Ho ∈ (0, 1). Equation (4.28) represents an infinite product extension
of the q-Pochhammer symbol, that can be easily evaluated using any computer based math-
ematical or programming tools. This infinity product extension is given by the following:

(a; q)∞ = ∏ (1 − aqk ), (4.30)
k=0

which follows directly from the definition in (4.27). In fact this product converges very
quickly when a and q ∈ [0, 1].
By examining the network-level performance in equation (4.26), an important trait can
be observed, that it is sufficient to enable few hops for reaching more than 95% of the
asymptotic service level κL , in most of the practical scenarios even 2-3 hops are sufficient
to reach this level. We illustrate this observation in Fig. 4.8 that depicts the number of
hops required to reach 95% of the asymptotic service level denoted as n95 . Mathematically
speaking the plot in this figure represents the following expression:
 
κn
n95 = arg min ≥ 0.95 , (4.31)
n∈N κL

where κn and κL are expressed in (4.26) and (4.28) respectively.


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 150

Fig. 4.8 The number of required hops for reaching 95% of the asymptotic service level κL .

4.1.9 Network Performance During a Disaster


In order to judge the network-level performance during a disaster, we utilize equation (4.26)
against varying values of the damage ratio D. We start with a healthy network D = 0 having
a nominal performance, and then we gradually increase the damage ratio. The results of
these numerical computations are plotted in Fig. 4.9 showing a significant enhancement in
the network-level performance when adopting D2D relaying, we can also notice that further
enhancement due to the extension of hops in the relay chain is gradually becoming marginal
for longer relay chains i.e. for higher n.
We note here the fact that the cellular downlink link-level performance enhances during
the disaster at a given distance ro , that is due to the reduction of interference. However,
when the performance is averaged over all possible values of ro in order to find the network-
level performance, we can clearly notice the performance degradation in Fig. 4.9 for higher
values of D. The physical interpretation is the following; as D approaches one, base stations
becomes more distant to UEs and the average incurred path-loss becomes heavier leading
to a reduction in the serving signal as well as a reduction in the interference, where the
effect of the noise becomes more prominent causing a reduction in the SINR and hence
a lower performance. Some studies such as in [3], proves that for a high-dense cellular
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 151

network, i.e. when the network is interference-limited, the two effects can cancel each
other, but further decrease in the density of base stations, i.e. more damage, will eventually
lead to a lower network-level performance where the network becomes noise-limited. In our
numerical example we start from a medium-density cellular network as per 3GPP simulation
recommendations.

1
Network-Level Success Probability [ ]

0.8

0.6

κ (Cellular Only)
0.4 o
κ 1 (1 Relay)
κ 2 (2 Relays)
0.2 κ 3 (3 Relays)
Device-to-Device E[ P D2D ]

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Damage Ratio [ ]
Fig. 4.9 Network-level success probability against varying values of the damage ratio D.

In order to get a measurable metric for the enhancement introduced by D2D relaying,
we define the Damage Alleviation Buffer (DAB), which is the amount of damage D that a
network can withstand before deteriorating below its original performance level. Firstly, we
define the performance difference ∆Sn as the following:

∆Sn = κn − κo , (4.32)

and then we check at which damage level this performance difference is equal to zero, or
in other words, at which damage level the network is performing in a similar manner to the
original network (prior the disaster). Mathematically speaking:

DABn = D|∆Sn =0 (4.33)


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 152

By applying the numerical parameters in Table 4.2, we plot in Fig. 4.10 the performance
difference ∆Sn against the range of damage ratio D ∈ [0, 0.9]. The figure illustrates the
intersection of ∆Sn curves with the x-axis, indicating the values of DABn . We can notice, for
example, that the DAB could reach as high as 65% for a single hop relay network, meaning
that the network can take damages up to 65% while maintaining the nominal performance, a
significant alleviation of network damage. The figure also depicts the DAB value for higher
numbers of cascaded relays.

0.4
1 Relay
2 Relays
Service Difference ∆S [ ]

3 Relays
0.2 4 Relays

Nominal service
0
DAB (1 Relay) = 65%
DAB (2 Relay) = 78%
DAB (3 Relay) = 84%
DAB (4 Relay) = 87%
-0.2

-0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Damage Ratio [ ]
Fig. 4.10 The damage alleviation buffer due to the introduction of D2D relays.

4.1.10 Monte-Carlo Simulation


Our second approach in analysing the damage effect on cellular networks, is by performing
a larger number of repetitive simulations with a randomized distribution of eNB, UEr and
UE. Each simulation yields a certain outcome value of the performance, and by averaging
all obtained outcomes we can compute the mean network behaviour when affected by a
certain damage ratio D.
In order to deploy base stations or users according to a PPP, we first generate a random
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 153

number of points according to the following Poisson distribution:

λν
PD(ν) = exp(−λ ), (4.34)
ν!

where λ is the density of the deployed points, and ν represents the random variable of the
points count. Then we randomly assign map coordinates (x, y) to these points (eNB, UEr
and UE), using uniform distribution process between (xmin , xmax ) and between (ymin , ymax ),
where max and min represents the upper and lower coordinate extents of the simulation
map. In order to select a base station density λb that is inline with 3GPP recommendations
[25][25], we use the defined Inter-Site Distance (ISD) of 1,750m and convert it to an eNB
density according to the following formula, that can be easily proven using simple geometry:

2 3
λb = , (4.35)
ISD2
we should note here that the “site” according to 3GPP terminology is composed of three
sectors.
For illustration purpose, we calculate the SINR value for each of the pixels of the sim-
ulation patch according to equations (4.1), (4.2) and (4.4). The resulting coverage of the
cellular network is depicted in Fig. 4.11 showing the spatial SINR values in the simulation
patch of 5,000 × 5,000 unit area, in this illustration we didn’t account for the fast fading
effect. We can notice the SINR interaction between neighbouring eNB due to the mutual
increase in interference.
In order to obtain the link-level performance, each link between a UE and its serving
transmitter is compared to the target SINR threshold level Θ, where all tested links above
this threshold are considered as successful. While the tested links below the threshold are
considered as failed. Fig. 4.12 shows one of the simulation runs with the covered UEs
indicated. We can notice from the same figure that the subset of UEs that is directly covered
by the cellular network and the other subset which is covered via a single hop D2D relays,
depicting the coverage extension from the healthy cells.
The simulations of the uplink and the downlink link-success are performed indepen-
dently. Based on these values, the network-level performance measure will be calculated as
the following:
#{DL ∩ UL}
κ̂o = , (4.36)
#{UET }
where DL is the subset of successful UEs in the downlink service, UL is the subset of
successful UEs in the uplink service, UET represents the entire UEs set, and #{.} is merely
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 154

Fig. 4.11 Downlink coverage simulation for a healthy network, i.e. damage ratio D = 0%

the counting measure. The entire scenario is repeated for 2,500 times, where the results are
stored for a each simulated damage ratio and network setup.
On the other hand, for simulating the network level performance when the single hop
relay network is in use, we first identify the D2D relays that are capable of engaging in
network enhancement (denoted as D2D1 ), in other words, the D2D relays that are already
covered by the cellular network through a successful uplink service and a successful down-
link service. Also we obtain the set of UEs that were not able to establish a cellular service,
denoted as UEo , where UEo = {UET \(DL ∩ UL)}, then accordingly we identify the UEs
that are able to be successfully serviced by the D2D1 servers. This successful subset is
denoted as (UEo ↔ H1 ), while the total performance measure will be:

#(UEo ↔ H1 ) + #(DL ∩ UL)


κ̂1 = . (4.37)
#(UET )

In a similar manner, we find the performance of the nth chain relay network as the following:

#(UEn−1 ↔ Hn ) + #(UEn−1 )
κ̂n = , (4.38)
#(UET )
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 155

Fig. 4.12 Serviced UEs, with a network damage ratio = 50%

where the set UEn−1 represent the UEs that are covered by the n − 1 chain relay network.
The analytical curves obtained in Section 4.1.5 are compared to the results of Monte-
Carlo simulation as depicted in Fig. 4.13. It can be noticed form this figure that the theoret-
ical and simulation curves match with a maximum error less than 5%.

4.1.11 Concentrated Damage Model


So far we have adopted the random phase-out of the base stations, where the disaster is
assumed to affect base stations within the afflicted area in a random manner. In this section
we address the phase-out process assuming a concentrated damage that occurs at the center
of the simulation area, then we gradually expand the damage radius according to the severity
of the disaster. Fig. 4.14 depicts the concept of the concentrated damage model showing the
damage area for different damage ratios D.
In order to get an insight on the effect of such model, we compare in Fig. 4.15 the Monte-
Carlo simulation of the network success probability resulting from the random phase-out
and the one resulting from the concentrated phase-out. We can notice, that the performance
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 156

1
Network-Level Success Probability [ ]

0.8

0.6

0.4 κ (Cellular Only) Analytical


o
κ 1 (1 Relay) Analytical
κ 2 (2 Relays) Analytical
0.2
κ 3 (3 Relays) Analytical
Simulation
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Damage Ratio [ ]
Fig. 4.13 Comparison between the analytical approach and Monte-Carlo simulation runs.

of concentrated model is gradually getting lower for more severe damage ratios. This result
can be referred to the fact that in random cellular networks some degree of redundancy is im-
plicitly provided. Mature cellular networks usually focus on providing better capacity, and
thus implementing some redundant base stations. This design approach allows typical cel-
lular networks to withstand random phase-out disasters more effectively. On the other hand,
in the concentrated phase-out the coverage towards the afflicted area can be only originated
from the surrounding base stations in the healthy area. In fact the concentrated damage
model resembles the worst case scenario, when the afflicted area is severely damaged and
all base stations within this area are phased-out.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 157

2.5

1.5

1
Disaster Propagation
Map Y-Axis [km]

0.5

-0.5

-1

-1.5

Healthy BS
-2 Damaged BS
Central Point of Disaster
-2.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Map X-Axis [km]

Fig. 4.14 Concentrated phase-out model, the (+) sign is the central point of the disaster.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 158

Table 4.2 Notations and Symbols

Symbol Simulation Value Explanation

αc 3.52†1 Propagation exponent eNB-UE


αd 3.68†2 Propagation exponent UE-UE
Loc 23.4dB†1 Cellular propagation reference path loss
Lod 26.7dB†2 UE-UE propagation reference path loss
h – Rayleigh fading with unity mean
Θb 5dB Service threshold at the eNB (Uplink)
Θm 10dB Service threshold at the UE (Downlink)
Gb 14dBi[25] eNB antenna directivity
Pb (46+Gb )dBm[25] eNB EIRP power
Pm (23+0)dBm[25] UE EIRP power
ISD 1,750m[25] Inter-site distance

2 3
λb ISD2
eNB initial density
λm 200 km−2 UE density
λr 25%λm D2D relays density
Wb -99dBm†3 eNB thermal noise
Wm -105dBm†4 UE thermal noise
ηb -30dB eNB interference mitigation factor
ηm -60dB UE interference mitigation factor
ηr -60dB D2D-D2D interference mitigation factor
D [0,0.9] Damage ratio

†1 WINNER II scenario C3 (Bad urban macro-cell)[58].


†2 WINNER II scenario A1 (Indoor Office)[58].
†3 Calculated from eNB noise figure 5dB [198], bandwidth 10MHz.
†4 Calculated from UE noise figure 9dB [198], bandwidth 1MHz.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 159

1
Network-Level Success Probability [ ]

0.8
3 Relays
1 Relay
0.6

0.4
Cellular Only
0.2
Random Damage
Concentrated Damage
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Damage Ratio [ ]
Fig. 4.15 Performance comparison between the two models of damage propagation: (i) The
random phase-out, (ii) The concentrated phase-out.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 160

4.2 Temporary Cognitive Femtocell Network


Bandwidth, performance, reliability and cost effectiveness, are all favorite traits in modern
wireless telecommunication systems that encouraged the unmatched acceleration in indus-
try, education, and economy. Also it opened the doors for the new era of online entertain-
ment and socialization. Those same traits in mission critical system has much more crucial
role, they could be decisive in life-or-death situations the public safety personnel repeatedly
encounter. Long Term Evolution (LTE) will enable the firemen and police forces to stream
high quality real-time videos of live operations back to the command and control center. In
addition to that, LTE will allow multimedia and file sharing, facilitating the sensitive time
dependent tasks of the dispatched resources in the field. Policemen could download sus-
pects pictures, firemen can share the layout plan of a building, and ambulance vehicles will
be fitted with live video stream connection to the hospitals, allowing doctors to give lifesav-
ing instructions. In fact LTE has been endorsed by US major public safety organizations as
the technology of choice for the 700 MHz band [54]. To deliver these applications, public
safety LTE network has to feature a wide and reliable RF coverage deployment, however
there are several situations where network coverage is limited or absent, areas could be left
without proper coverage for many reasons such as cost, lack of grid power or technical dif-
ficulties. Those areas include remote rural lands, the inner region of large buildings and
subway tunnels and stations. Another situation is during the unfortunate event of natural
disasters like earthquake, tsunami and forest fires, where public safety network could be-
come completely paralyzed, causing a total failure to the rescue operations. The need for
direct device to device communication, or as it is realized in TETRA system as DMO and
LST [41], is vital for the public safety, since it allows the direct communication between
handheld devices in certain proximity overriding the failed or low coverage network. This
direct connection can bring up voice and low throughput data communication. The simplic-
ity of TDMA in TETRA allowed such feature, while the situation in LTE is quite different
due to the complexity of the air access technology and resource scheduling process. The
realization of direct mode feature in LTE system requires special considerations and en-
hancements for several technical aspects, especially in interference management between
the parent network and devices engaged in the direct mode.
In this section we introduce the concept of Temporary Cognitive Femtocell Network
(TCFN), as a secondary supplement for the main LTE Network (that is referred as the Parent
Network). Cognitive radio networks [165] with intelligently acquired radio environment
maps using sensing and localization is the key enabler in our proposed solution. TCFN
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 161

will provide the needed platform to enable the direct mode (D2D) [155] in LTE that will
allow devices to communicate between each other without the need to route the traffic via
the parent network. Several challenges are associated with TCFN deployment; this section
is discussing, as well as suggesting suitable solutions for each. Those challenges could be
summarized as below:

• TCFN Enabling Mechanism: discussed in section III.

• Cognitive Interference Mitigation: In order to achieve a bandwidth efficient overlay


system that mitigates interference impact on parent network users, exploiting the latest
recommendations in Rel. 10 in regards to HetNet and eICIC. This topic is discussed
in section IV.

• Cognitive Relaying Mechanism: for enabling the communication between the TCFN
and the parent network is discussed in section V.

• TCFN Disabling Mechanism: is discussed in section III.

4.2.1 Network Model


Two main distinct types of networks are proposed:

• The Parent Network (Network A): Based on conventional macro eNBs, providing the
coverage for the intended geographic area.

• The Temporary Cognitive Femtocell Network (Network B): That constitute of several
coverage spots called TCFN Clusters (discussed in sub-section A below.)

However, we also consider a third type of network which is the Emergency Supplemen-
tary network (Network C), in proposing the TCFN concept, which is deployed in the case
of a total failure of the parent network, e.g. disaster situations. Such emergency networks
are rapidly deployed, in afflicted areas, ABSOLUTE is a good example project for such
networks [4]. The concept of TCFN and the three considered networks (A,B, and C) are
depicted in Fig. 4.16. As we see in the figure the TCFN clusters can exist (i) within the par-
ent network coverage area where outage is observed, (ii) at the edge of the parent network
coverage area and (iii) outside the parent network’s coverage area.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 162

The TCFN Clusters

The TCFN has a cluster head UE (chUE) and a set of cluster member UEs (cUE) that are
attached to the chUE (refer to Fig. 4.17) . The cUEs can access the E-UTRAN of the
parent network only through the chUE, moreover the intra cluster communication between
the cUEs is also managed by the chUE in order to minimize the signaling within the cluster.
Note that the intra cluster communication can be performed between the cUEs directly using
the D2D mode, however the chUE is responsible for the local Radio Resource Management
(RRM). Seeing TCFN Cluster from a user perspective, all UEs belonging to a certain TCFN
Cluster are mesh connected using the D2D links, i.e. any device can share, transmit and
receive information to/from any other member of the same TCFN Cluster
The cluster head is an advanced user piece of equipment that has the capability of acting
as a femto cell in addition to its role as a UE, and, it constitutes the heart of the TCFN
coverage cluster. The cluster will occasionally include some associated relays denoted as
rUE that will fill coverage gaps and connect farther users to the TCFN cluster.

Fig. 4.16 Network model with the three types of networks.

TCFN Enabling / Disabling

The decision to commence (Enable) a certain TCFN cluster is taken by the chUE and depend
on several triggers, one of them is the automatic service threshold trigger when the main
wireless service drops below a pre-defined SINR level:
(
0, : SINRUE ≤ SINRth
Service = (4.39)
1, : SINRUE > SINRth ,
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 163

Fig. 4.17 TCFN Cluster Concept.

where SINRth is given by SINRth = SINRmin + ρ, and SINRmin is the SNIR of the minimum
decodable LTE MCS and ρ is a tuning parameter. Other triggers include:

1. Specific local high bandwidth group communication (Automatic), e.g. streaming


video locally to the firemen team leader

2. User initiated TCFN Cluster (Manual)

3. Search & Rescue operation that expects victims outside the parent network service
cell radius (Manual)

Fig. 4.18 shows the various operational stages of a public safety LTE network, Accordingly
TCFN will have the below operational status:

1. Status (1) TCFN could be optionally initiated anytime, however it will be frequently
used in low coverage areas

2. Status (2) TCFN will fill the considerable gaps left by network (A) and (C) and it is
very essential at this point

3. Status (3) TCFN will still fill the coverage gaps left by networks (A) and (C), gradually
becoming as less vital
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 164

Fig. 4.18 Public Safety Network Operational Stages.

4. Status (4) TCFN will go back to normal operation; similar to status (1)

If a UE inside the affected areas detects an established TCFN and opt to join the clus-
ter, it should synchronize with it, and then attachment is done through the normal random
access procedure. Terminating (Disabling) the TCFN cluster could be triggered by several
conditions:

1. Parent coverage is restored and is above the θ level for a certain predefined period of
time (Automatic).

2. chUE is not capable of maintaining the TCFN due to battery drain (Automatic)

3. User decision (Manual)

4. Parent network decision, e.g. severe interference caused by the TCFN (Automatic)

4.2.2 Downlink Interference Mitigation


Thanks to the new Intercell Interference Coordination techniques presented in Release 10,
that overtook the implicit inefficiency in previous releases [98] especially when dealing
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 165

with new elements introduced as part of the heterogeneous network (HetNet) such as re-
lay nodes, Femtocells and remote radio heads. These techniques as seen from the time-
spectral resource space are categorized into two distinct groups, a time domain group and
a frequency domain group, the former concentrates on controlling the transmission of the
sub-frames/Symbols of the overlaid eNBs (such as Femtocell and Picocell), while the latter
focuses mainly on fractional frequency reuse (FFR) methods [96], The commonality be-
tween all techniques is that they lay inside the constrains triangle of bandwidth efficiency,
dynamicity and implementation complexity. For instance, one technique might result in a
high spectrum efficiency but at the same time will require a massive exchange of signaling
information between the different network nodes. In this section we are propose a resilient
solution that will try to utilize both interference coordination groups.

Time Domain Solution

Time domain interference mitigation requires reliable signaling between the parent network
eNB and the TCFN chUE. The proposed interference mitigation approach is OFDM symbol
muting with shift [201], where subframes of the chUE are shifted forward by 5 symbols
compared to the parent eNB subframes (refer to 4.20), shifting will mitigate the interference
caused by the parent eNB’s PDCCH on chUE’s PDCCH, while muting will mitigate the
interference by chUE’s on parent eNB’s cell-specific reference signal CRS that are quite
important for UEs to perform radio link monitoring measurements, in particular, to carry
out the RS-SINR calculations. Low value of RS-SINR will result the UE to falsely consider
the radio link as un-usable and terminate the connection with the eNB [201], another reason
to protect the reference signals is that they play a vital rule in channel estimation and OFDM
demodulation within the UE.
Furthermore, the parent network will report to chUE if any of the cell edge UE has
a severe interference caused by the TCFN; in this case the chUE will configure the inter-
fering subframes as almost blank subframes (ABSFs) [98] in order to further mitigate the
interference on both PDCCH and PDSCH (See Fig. 4.19).

Frequency Domain Solution

Different techniques such as the dynamic power control, dynamic fractional frequency reuse
and formation of groups [96] are implementable in OFDM, however we are discussing the
static Fractional Frequency Reuse (FFR) since it requires less cross-tier coordination (i.e.
less coordination between the parent network and TCFN) and even it can be used in the case
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 166

Fig. 4.19 chUE muted subframes

Fig. 4.20 Symbol muting and shifting concept


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 167

of the total absence of cross-tier signaling that is in contrary to the dynamic FFR methods.
DFFR could be smoothly used in the current LTE commercial deployments since eNBs has
excellent mutual interface X2 allowing ICIC/eICIC information exchange, a privilege might
not exist in the case of chUE-eNB link.
Soft FFR is a static type FFR that has a high spectrum efficiency [96]; its principle
is to divide the available spectrum to N sub-bands, where N is the number of cells in a
frequency reuse cluster. Fig. 4.21 illustrates the proposed reuse pattern, where the available
sub-carriers are divided into three groups (designated as R, G, and B); the parent eNB cell
center is assigned two groups while the cell edge is assigned only one. The chUE will be
free to select from carriers in R or G in the edge region of cell 1, the selection will depend
on the existing level of interference caused by the macro coverage of neighboring cells.
Outer cell areas where service threshold condition is reached (i.e. SINRUE < SINRth ),
the TCFN will have more freedom in the utilization of the spectrum, since the cross-tier
interference probability is much lower. On the other hand, users that manually create TCFN
in the cell center will have to utilize the time domain interference mitigation technique in
addition to adhering FFR pattern as well.

Fig. 4.21 Soft FFR Concept, Underlined bands shall be used only in conjugation with time
domain interference coordination
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 168

4.2.3 Cognitive Relaying Mechanism


TCFN exploits the standardized relay nodes concept [202] in LTE-A Rel. 10, and further
elaborated in Rel. 11 documentations [203]. In addition to that each chUE in TCFN will
be switching intra- TCFN Cluster traffic locally without the need to relay it to the parent
network, however, in the case of a cluster user (cUE) have some data targeted to the parent
network; the chUE will then function in a similar fashion to a Layer-3 relay node, by back-
hauling the cUE traffic towards the parent network donor eNB via the standardized Un link.
While regular relay nodes are operator deployed, and favor directional antennas, chUEs are
meant to be mobile, unplanned, and having smart antennas instead, which will allow better
relaying throughput.
TCFN requires relay nodes cascading for further coverage extension, however this sce-
nario is not standardized in LTE-A, and in order to achieve this important feature, we assume
a layer-2 (Decode and Forward type relay [42]) that will act in a much simplified manner
than the Layer-3 (relay node) RN and without the need for full RRM and user data recogni-
tion which we refer to as the rUE. Reference [43] further explains this concept. Simulation
Model The goal of the presented simulation is to illustrate the need for TCFN, in order to en-
courage further researches and studies concerning this domain, and meant to reproduce the
regular LTE coverage difficulties in indoor environment. The simulation model is partially
derived from guidelines recommended by 3gpp for indoor and home femto cell simulation
scenarios [22]. Results compare the spatial-probability performance of three coverage sce-
narios:

1. First Scenario represents the performance of a conventional standalone network rep-


resented by an eNB placed 250m in the north-east side of the subject building, with
no TCFN capabilities.

2. The second scenario shows the results of introducing a TCFN Cluster inside the build-
ing, with one cluster head and one relay, without the proper interference coordination
with the parent network

3. While the third, demonstrates the effect of using the cross-tier interference mitigation
as suggested in section IV, so the indoor UE has the freedom to attach to either the
Parent network directly or to the TCFN

the building is assumed to have dimensions of 50 m × 50 m with 25 apartments inside


it, 10 m × 10 m each [203], UEs are assumed to be placed randomly inside the building.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 169

Fig. 4.22 illustrates the target building location within the parent network cell, which has an
antenna radiation pattern as the following [203]:
 
φ
G = Go − min 12 , FBR (4.40)
φ3dB

where φ3dB = 70 is the half power beam width , and FBR = 20 dB is front-to-back gain
ratio. The antennas of chUE and rUE are assumed to have 0 dBi gain (on the access side)
that would be the most probable feasible realization in the future.

Fig. 4.22 Simulation model and Target building location

Fig. 4.23 Spatial service probability

The propagation model has been approximated to WINNER II Indoor model, type A1
(Room to Room) [58], assuming thick walls of 12 dB penetration loss each. We define the
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 170

service outage probability of a Receiver (r) for certain scenario (s) by:

P [SINRr,s < SINRth ] (4.41)

where SINRth depends on the target application minimum throughput and on application
requirements. And in our simulation we used SINRth = 10dB [203] which is the LTE
commonly-acceptable value for a minimum-level service.

Table 4.3 Simulation Parameters

Parameter Value

Cellular Layout Hexagonal grid 3 sectors per site


Target Building 250m from the eNB South-West
Simulation Frequency 2000 MHz [198]
eNB antenna max gain 17 dBi [198]
UE Antenna gain 0 dBi [198]
UE Noise Figure 9 dB [198]
Total eNB TX power 46 dBm [198]
Tx power 42 dBm
chUE/rUE Power 24 dBm (250 mW), Class I RN [203]
Indoor Pathloss Model WINNER Case A1 [58]: PL (dB) = 38.44+20 log(d)+ n × Pwall
Outer walls Penetration Loss 18 dB [198]
Inner walls Penetration Loss 12 dB (heavy walls) [198]

Where, (d) in meters, (Pwall) is the penetration loss caused by walls, (n) number of
walls.
Fig. 4.24 depicts the coverage resulting from a conventional stand alone LTE network
(simulation scenario 1) inside the target building illustrated as Reference Signal Received
Power (RSRP) values; white areas indicate a level below the service limits. It is very clear
that the service can barely reach the second raw of apartments, leading to a very high service
blockage as shown in Fig. 4.24.
Fig. 4.25 shows the coverage resulting from implementing simulation scenario 2 (a
TCFN with no interference coordination). From the figure we observe a moderate enhance-
ment in the coverage area due to the deployment of the temporary network TCFN. Note
that having more number of relays may improve the coverage in this sense however the
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 171

Fig. 4.24 RSRP for the Baseline System (Parent Network only)

improvement may only be incremental due to the interferences.

Fig. 4.25 RSRP when uncoordinated TCFN is deployed

On the other hand, Fig. 4.26 shows much better RSRP compared to Fig. 4.25, due to
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 172

the enabling of the interference coordinating between the deployed TCFN and the Parent
Network. Only fractional area is left without proper service, while most of the apartments
are well covered. The enhancement of blockage probability is clear in Fig. 4.23.

Fig. 4.26 RSRP when enabling the coordinated TCFN

Fig. 4.27 Spatial RS-SINR CDF Comparison


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 173

A CDF plot of the Reference Signal-SINR (RS-SINR) in Fig. 4.26 shows a significant
enhancement when deploying the coordinated TCFN, this enhancement will be reflected
as a better D2D throughput as well as enhanced cluster-to-parent overall throughput. The
blockage probability improvement is quite clear in Fig 4.23 since the difference between
scenario 1 and scenario 3 is 46.9%, almost 5.5 times less blockage probability.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 174

4.3 Cognitive Relay Nodes for Airborne Coverage En-


hancement
The unprecedented advancements in Long Term Evolution (LTE) System made it the practi-
cal choice for any -near future- public safety network deployment. By enabling multimedia
contents and file sharing, a whole new dimension will be added to the public safety op-
erations, significantly assisting critical tasks, and facilitating the sensitive time dependent
missions of the dispatched resources in the field. Policemen could download suspects pic-
tures, firemen can share the layout plan of a building, and ambulance vehicles could be
fitted with live video stream connection to hospitals, allowing doctors to give lifesaving
instructions. In fact LTE has been endorsed by US major public safety organizations as
the technology of choice for the 700 MHz band [54]. However, along with the projected in-
creasing dependency on LTE System by public safety agencies, come new challenges facing
network realization, mainly: (i) to provide continuous radio coverage in order to guarantee
a seamless service during nominal operation conditions, and (ii) also it is very important
to have a solid solution for a rapidly deployable network coverage as a failover scenario
in emergency cases, i.e when extreme events occur such as natural disasters. Tackling the
first challenge lays within the LTE-Advanced standardization itself that includes multiple
enhancements over LTE Release 8 allowing cost effective deployment techniques, enabled
by the Heterogeneous Network concept (HetNET), where several types of network elements
can contribute in radio coverage such as Remote Radio Heads, Relay Nodes, Femto Cells
and Pico Cells, thus pushing network infrastructure closer to end users for enhancing spec-
tral efficiency. In fact some of those techniques are nowadays already in use by commercial
network operators.
The serious challenge that still lacks standardization is in providing a rapidly deployable
coverage alternative when a natural disaster occurs, during which, the performance of the
main network is severely affected. In this section we are using the term nominal network
referring to the conventional LTE network that is providing radio coverage during normal
working conditions, while the term Rapid Emergency Network (REN); is used to refer an
assumed rapidly deployable network infrastructure that enables basic telecom services such
as voice and limited data, for a wide geographical afflicted area when the nominal network
service has failed or severely affected by a mean of natural disaster; like flood, bushfire,
severe storm or earthquake. The same REN architecture will be capable of serving sud-
den increases in mobile traffic caused by massive public events where vast population is
expected in a certain area, e.g. sport events. The crucial role of REN in enabling public
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 175

safety operations during the aftermath of disasters, has urged several governmental orga-
nizations to seek reliable deployment scenarios for such networks, for example the Public
Safety and Homeland Security Bureau in the USA has endorsed the use of deployable aerial
communications architecture (DACA) [53] that envisioned the recovery of critical commu-
nications for first responders within 12 to 18 hours. Another example of REN development
efforts is the ongoing European Union funded ABSOLUTE project [4] (Aerial Base Stations
with Opportunistic Links for Unexpected and Temporary Events) focusing on Low Altitude
Platforms (LAP) that are increasingly believed as one of the ultimate answers for RENs.
In ABSOLUTE two tightly interconnected network segments are proposed: an air seg-
ment and a terrestrial segment. The former consists of an innovative helium balloon-kite
structure carrying LTE aerial Base Stations, while the ground segment consists of Portable
Land Rapid Deployment Unit complementing network coverage and providing satellite
backhaul connectivity. It is important to understand that although LAPs are considered one
of the main answers for REN, ground support units are still very essential in complementing
such aerial coverage, and one of those ground units, is the Relay Node as standardized in
LTE-Advanced. In this section propose a novel mechanism to enhance the performance of
standard relay nodes by utilizing cognitive coverage mechanisms, and hence we are calling
those novel relay nodes as Cognitive Relay Nodes (CRN) meant to be rapidly deployable
with full self-configuration capabilities, while at the same time, preserving the compatibility
with Release 10 RN standard.

4.3.1 Relay Nodes in LTE-Advanced


Relaying has been a quite promising trait added to LTE technology, that even before its first
standardization in LTE-Advanced Release 10, researchers, academics and vendors were
endeavoring to exploit the potential power of this technology that will allow a consider-
able enhancement in the cell coverage and throughput [204]. Relays are classified into
several categories according to their protocol stacking [42]: type Layer-1 is the simplest,
that amplifies-and-forward the signal, thus operating only in the physical layer without any
packet error correction mechanism, whoever type Layer-2 performs packets decoding and
hence is called decode-and-forward, while type Layer-3 has most of eNB features, since
it can transmit its own cell identifier, cell specific reference signal (CRS), and primary /
secondary reference signal, which is also standardized in Release 10 and used hereafter as
the base for the enhancements suggested in this section. In particular Release 10 defines a
specific type of Layer-3 called Type-1 which is an inband relay node having the necessary
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 176

Radio Resource Control (RRC) functionalities to support cell handover and mobility man-
agement, also it has its own scheduler to allocate physical resources to User Equipments
(UEs) within its coverage area.

4.3.2 Rapidly Deployabel Cognitive Relay Nodes


The urgent nature of an REN network stipulates a rapid deployable architecture such as
network backhaul, eNBs, ground support units, etc... Similarly, our proposed Cognitive
Relay Node (CRN) shall allow a swift dispatch via the following features:

• Integration with power supply units such as: generator-set, solar panels and power
backup system.

• Mounted on a wheeled trailer.

• Hydraulic telescopic mast holding the antennas (access and backhaul antennas).

• Auto-tracking directional antenna for backhauling.

• 6-Sectors access antenna with Radio Frequency (RF) beam switching capability.

• Automatic configuration when powered up:

– Selection of donor eNB.


– Acquiring Network Parameters.
– Antenna beam control.

This section is focusing on illustrating the enhancement required for the communication
system, while "housekeeping" components such as the power system and mechanical struc-
ture are not discussed here. Our proposed CRNs are specifically tailored to complement the
coverage of LAP Aerial eNBs (AeNB). During emergency deployments, as shown in Fig.
4.28, the coverage of AeNBs can’t be the only source REN service, but rather it should be
complemented by terrestrial elements, this fact is due to the limited weight capabilities of
LAPs causing limited size of RF power amplifier onboard. CRNs like any type-1 Relay
Node will be backhauled via the standard Un interface to a donor eNB, which is in our case
the AeNB). While in its turn AeNB will aggregate all connected CRNs and attached UEs,
then forward the traffic to a ground termination point(s).
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 177

Fig. 4.28 Overview of REN Architecture with CRNs.

4.3.3 Network Model


The model that we have adopted for simulating the performance of our proposed Cognitive
Relay Node, aimed to replicate practical scenarios of a Rapid Emergency Network. And ac-
cordingly we opt to choose the ABSOLUTE network as our baseline system for the network
model.

Propagation Model

The most important part of the wireless network here is the propagation model, this section
consider two models for the propagation as described below:

• Ground-to-Ground RF Propagation (Terrestrial) model.

• Air-to-Ground RF Propagation model.


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 178

In the first model we are adhering 3GPP recommendations for the corrected RN terrestrial
propagation in [205] for predicting CRN coverage, taking into consideration that RNs has
quite different RF propagation behavior than terrestrial eNBs since they are proposed to
be mounted on heights of (5 to 15 m) while models for terrestrial eNBs are typically for
heights of (30-100m) [163]. This corrected model assumes two different propagation con-
ditions: (i) Line of Sight (LoS): for UEs in the near proximity with RN and (ii) Non Line
of Sight (NLoS): for other UEs. The probability of having a LoS between a Relay Node
with a User Equipment of a distance (d in meters) is strongly depending on the average
buildings height and train topography, in our simulation we have chosen to simulate urban
environment in consistence with Air-to-Ground propagation model assumptions, that will
be discussed shortly. LoS probability in Urban environment is given by [205]:
   
1 1 −156 1 −d
PRN−UE
LoS (d) = − min , 5e d + min , 5e 30 , (4.42)
2 2 2

and accordingly the NLoS probability will be given by:

PRN−UE RN−UE
NLoS (d) = 1 − PNLoS (d). (4.43)

In order to combine the two terrestrial propagation scenarios (LoS and NLoS) in a single
formula, logarithmic averaging was used, so the resulting path-loss can be written as:

PRN−UE (d) = PLRN−UE


LoS (d) × PRN−UE
LoS (d) + PLRN−UE RN−UE
NLoS (d) × PNLoS (d) (4.44)

where PLRN−UE
LoS (d) and PLRN−UE
NLoS (d) are the LoS and NLoS mean path loss respectively,
and are statistically obtained from channel measurement campaigns done in Beijing [205],
which resulted the following model for RN height of 5 m and frequency of 2,000 MHz:

PLRN−UE
LoS (d) = 41.1 + 20.9 log(d)
PLRN−UE
NLoS (d) = 32.9 + 37.5 log(d) (4.45)

On the other hand, the radio coverage serviced by the AeNBs will follow an entirely dif-
ferent scheme. In fact airborne communication has a superior radio propagation ability,
compared to conventional terrestrial methods due to the enhanced probability of achiev-
ing LoS condition, however technical difficulties prevent deploying large numbers of aerial
platforms, and in our baseline network model, we are going to use seven (7) AeNBs only.
For simulating Air-to-Ground propagation we have utilized the profound statistical study
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 179

developed in [63] that created a standard model for Air-to-Ground path loss, based on three
types of paths: (i) Line of sight (LoS) for clear and un-foiled paths, (ii) Obstructed line of
sight (OLoS) for partially obstructed paths, and (iii) None Line of Sight (NLoS) for paths
that are totally blocked by buildings or other structures. Similar to ground-to-ground, the
combined path loss that consolidates the probability of all types of paths is given by:

PLAeNB−UE
LoS (θ ) = PLAeNB−UE
LoS (θ ) × PAeNB−UE
LoS (θ )
+ PLAeNB−UE
OLoS (θ ) × PAeNB−UE
OLoS + PLAeNB−UE
NLoS (θ ) × PAeNB−UE
NLoS (θ ), (4.46)

where θ represents the elevation angle between the UE and the serving AeNB and
AeNB−UE
PLoS (θ ), PAeNB−UE
OLoS (θ ), and PAeNB−UE
NLoS (θ ) are path probability of LoS, OLoS and
NLoS respectively, approximated as per the below path loss formulas:

PLAeNB−UE
LoS (θ ) = PLo − 20 log(sin(θ ))
 
90 − θ
PLAeNB−UE
OLoS (θ ) = PLo + 4.77 + 0.35 exp
21.04
 
AeNB−UE 90 − θ
PLNLoS (θ ) = PLo + 15.15 − 12.62 exp , (4.47)
−7.32

where PLo is the free space loss between the AeNB and the ground projection point directly
beneath the LAP:
PLo = 38.47 + 20 log(hLAP ), (4.48)

where hLAP denotes the average LAP altitude (meters).


In ABSOLUTE network each aerial platform is a sectorized eNB carrying three direc-
tional antennas, and in our network model we chose to orient them towards the cells’ centers.
So the down tilt of all antennas will be given by:
 
hLAP
θtilt = arctan , (4.49)
Rcell

where Rcell denotes the design cell radius (meters), which is represented by the cell’s
hexagon radius, as shown in Fig. 4.29.

4.3.4 Cognitive Relay Assisted Emergency Network


As stated previously, the Cognitive Relay Nodes are aimed to assist the coverage served by
the aerial base stations (AeNB), and as an example, Fig. 4.30, shows a target square area
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 180

Fig. 4.29 The proposed AeNB Antenna configuration, adopting the requirements from the
ABSOLUTE project [4].

where REN coverage is needed. The three AeNBs are providing the main service; while
the 9 CRNs are working in a cognitive manner (by steering their RF beams) for serving the
coverage gaps left by the AeNBs.
So the intelligence in our proposed CRN is mainly achieved by adopting three mecha-
nisms (i) the ability to shutdown the transmitting power when no users are spotted in the
CRN’s proximity, which will allow an enormous power saving, (ii) the ability to Calculate
Pathloss with UEs before commencing the relaying TX coverage, (iii) CRNs will be fit with
switched antenna set that will allow directing RF power towards the UEs that are in most
need for the service. These three mechanisms will be further elaborated in the following
sections.
After powering up the CRN, it will acquire its location via a positioning system, such
as GPS/Galileo, while at the same time, it will temporary attach to the strongest available
AeNB, and acquire the list of all LAPs current coordinates / heights (via a high level ap-
plication), then the CRN will mechanically steer its backhaul antenna towards the nearest
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 181

Fig. 4.30 Illustrating the concept of CRN Assisted REN.

and less-congested AeNB and re-attach. Refer to Fig. 4.31 showing the operation sequence.
After that the CRN will start rotating the beam of the access antenna (electronically) several
times per second in order to construct users profile list, upon that the decision on which
sector(s) to lock the access beam will be taken.

Cognitive Relaying Mechanism

For each transceiver the CRN will continuously assess the UEs conditions of neighboring
sectors, and accordingly the UE profile will be maintained for three consecutive sectors
only per transceiver, i.e. the served sector and the two adjacent ones. If there is no more
nominated UEs to be served, the transmitter(s) of the CRN will switch off the TX radio
power and enter into a dormant mode, in which the CRN will keep rotating the access
beam (in receiving mode only) and try to construct the table of any approaching user(s).
CRN will have the capability of steering radio beams towards UE(s) that are in most need
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 182

Fig. 4.31 The process of CRN intelligently attaching to an AeNB.


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 183

of the service. In contrary to conventional Relay Nodes (in nominal networks) that are
operator-deployed and well planned to serve certain cell-edge locations, also designed to
have Omni directional transmission, in REN the ability to perform full RF planning will be
very limited since there is neither enough time nor enough information about the afflicted
area, accordingly the deployment of intelligent self-configure RNs will be the key to REN
success.

RN-UE Path Loss Predication

A key elementary feature in LTE System is the Channel-State Reporting [99] provided
by the User Equipments in order to allow eNBs to perform channel depended schedul-
ing and the Adaptive Modulation and Coding (AMC), one portion of this report is called
the Channel-quality indicator (CQI) representing the highest modulation scheme that the
downlink channel transmission can take place with a block-error rate of at most 10%. CRN
will utilize both CQI feedback and power reports received by the AeNB in order to perform
decisions on where to steer the access antenna RF beam. A Path Loss prediction for the RN-
UE path will be accomplished via the following procedure: CRN will listen to UE (number
n) or UEn transmitting the Physical Uplink Control Channel (PUCCH) that will be received
by the CRN according to the following:

RX,CRN TX
PUE n
= PUEn
− PLUEn −CRN , (4.50)

TX is the power transmitted by the UE , and PL


where PUE n n UEn −CRN is the path-loss between
TX
UEn and CRN. On the other hand PUEn can be predicted by the AeNB since it follows a
well-know LTE specific formula:

TX
PUE n
= min [Pmax , Po,PUCCH + PLDL + ∆format + δ ] (4.51)

where Pmax is maximum allowable transmit power, Po,PUCCH is a cell specific parameter,
∆format is a format dependent offset and δ is a power control offset. What is important
to note is that all above mentioned parameters are well known by the AeNB except the
downlink path loss PLDL , accordingly the formula can be succinctly rewritten as:

TX
PUEn
= f (PLDL ) (4.52)
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 184

On the other hand, the received power at the AeNB will be given by:

RX,eNB TX
PUEn
= PUEn
− PLUL (4.53)

where PLUL is the uplink path loss that could be considered equal to PLDL , i.e. reciprocal
channel. From (4.52) and (4.53) the AeNB can predict the transmitted power by the user
equipment PUETX and then report this power along with CQI to the CRN, that will then use
n
(4.50) to obtain UEn -CRN path loss estimation.

Switched Beam Antenna

As mentioned previously, the access link is equipped with an electronically switched beam
antenna capable of steering the RF power to/from a certain direction (for each transceiver
available in the CRN), a mechanism in which the CRN will be able to serve the UEs that
are in most need of the service. Also, directing CRN beam will mitigate the interference on
AeNB coverage since the frequency reuse factor in LTE networks is unity, and interference
control is a key for spectrum utilization efficiency.

Fig. 4.32 Switched antenna in CRN (with two transceivers).

In order to enable this mechanism 6-directional antennas are connected to RF switches


as illustrated in Fig. 4.32 (showing a CRN with two transceivers configuration example),
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 185

these RF switches are electrically-controlled by the CRN providing steering of RF cover-


age towards the desired direction(s), the only minor negative impact of introducing the RF
switch through the antenna line is its insertion loss (which is in the range of 0.8 dB), however
as it will be shown in the simulation that the overall coverage enhancement will overcome
this minor impact.
Each antenna is assumed to have 65◦ of half power beam width in order to allow suffi-
cient overlapping between sectors. It is important to note that a CRN might have more than
one set of transceiver/antennas and the number of the simultaneous active sectors will be
equal to the available transceiver/antenna sets. The radio power of a certain transceiver will
be directed towards a single direction, and thus having a superior RF performance in this
direction of interest. Fig. 4.33 illustrate this concept, one active sector is shown in red solid
line.

Fig. 4.33 Beam Steering for directing RF power in the network.

Beam Steering Decision

So far, the mechanisms of how to perform UE-RN path loss prediction have been illustrated
and also how to steer the RF beam, while this section will show the method of deciding
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 186

which sector(s) to serve based on the these mechanisms. In fact, during beam rotation
operation period, the CRN is proposed to build a User Equipment Table that has the general
format in Table 4.4.

Table 4.4 UE Database formed by the CRN

Sector Number UE ID UE-CRN Path Loss CQI

Sector 1 UE1 PL1 CQI1


Sector 1 UE2 PL2 CQI2
. . . .
Sector 2 UE1 5 PL1 5 CQI1 5
.. .. .. ..
. . . .
Sector 6 UE4 0 PL4 0 CQI4 0

Thus, each sector a complete list of detectable UEs is constructed, and then the sector is
given a score value according to the following formula:
!
N
1
Si = ∑ W1 × 1
, (4.54)
no PLUEn −CRN +W2 × CQI
n

where W1 and W2 are tuning weights, and CQIn is the average CQI reported by a user-
equipment UEn to the serving AeNB, and N is the total number of detectable UEs within
the sector. Basically, the score of each UE is inversely proportional to its predicted path loss
and to its reported CQI. After completing sectors scoring, the CRN will lock the RF switch
to the sector(s) with the highest score, and commence the cell broadcasting. For example in
the case of two-transceivers the CRN will be able to activate two sectors as a maximum.

Relay Nodes Placement Strategy

During REN deployment, agencies have very limited time for performing radio network
planning and setup, accordingly an assumed planning tool shall nominate the relay nodes
locations based on the number of deployed LAPs and the terrain average parameters. In
our network model we have used a CRN location planning method based on the spatial
blockage ratio; first of all we obtain AeNBs coverage simulation for the target area (without
the effect of any CRN), then we break down obtained coverage simulation map to smaller
adjacent circles of a radius equal to the design cell radius of the CRN, after that we calculate
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 187

the spatial service blockage probability inside each of those circles by counting the blocked
pixels:
Nblocked
BCRNi = , (4.55)
Ntotal
where Nblocked is the number of pixels that are not serviced (blocked), and Ntotal is the total
number of simulated pixels within the study circle, as depicted in Fig. 4.34.

Fig. 4.34 Illustrating the concept of geographical pixels service by AeNB.

A pixel is considered as blocked, when it has an SINR less than the service threshold
required for attaining certain service throughput. Thus a decision whether to place a CRNi
or not in the center of this study circle is given by the following Boolean condition:
(
no, : BCRNi ≤ SBth
Deploy CRNi = (4.56)
yes, : BCRNi > SBth ,

where SBth is the Spatial Blockage decision threshold and is related to the overall target net-
work quality of service (QoS). It is important to notice that this method is entirely different
than the commonly studied RN deployment schemes for terrestrial nominal networks.

4.3.5 Simulation Analysis


The assumed baseline REN Network consists of seven Low Altitude Platforms covering a
square geographical area of 40 km × 40 km, replicating an urban environment of a typical
European city, with an assumed average building height of 11.4m and flat average ground
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 188

level [63], the optimum altitude of the platforms was obtained by searching the maximum
service availability ratio among several simulation runs of different LAP altitudes. The re-
sults in Fig. 4.35 are showing a peak service at around 1,650 m. This altitude is mainly
dependent on the average building height, transmission power and on the target SINR. Gen-
erally, higher buildings demands higher LAP altitude.
Two factors are affecting the produced curve in Fig. 4.35, first the enhancement of
coverage when LAP altitude is rising between 0 m and 1,650 m, after that comes the effect
of path loss factor, starting to be more significant, and negatively affecting the service. More
information about LAP altitude optimization can be found in [129].

Fig. 4.35 Simulation results for obtaining the Optimum LAP Altitude for the assumed urban
environment.

Simulation Scenarios

In order to illustrate network performance impact of introducing the cognitive relay nodes
three different scenarios have been simulated, Scenario (1) represents our baseline, and
consists of the aerial eNodeBs only, without any terrestrial relaying capabilities, while in
Scenario (2) conventional Type-1 Relay Nodes were added in order to assist the coverage
of the AeNBs. It is important to notice here that the proposed conventional RNs has the
same radio transmitting power as of the CRNs, but instead of the switched antennas sys-
tem, standard omni-directional antennas are fitted. In Scenario (3), conventional RNs are
replaced with our Cognitive Relay Nodes for complementing the aerial coverage. Table 4.5
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 189

Fig. 4.36 Comparison between the service blockage probability.

summarizes the main system parameters used in the simulation of our network and scenar-
ios. Users are assumed to be randomly distributed within the service area which is a circle
of radius 20km, having a total number of 2,000 UEs.

Simulation Results

The performed simulation mainly focuses on obtaining the achieved RSRP (Reference Sig-
nal Received Power) for each of the explained scenarios, because RSRP is the main coverage
indicator of how well the target area is being served. RSRP result for the first simulation
scenario is depicted in Figure 9, that also shows the locations of the assumed seven AeNBs
(the red dots) and their hexagon cellular structure. Areas with SINR below the service
threshold level are colored in white indicating service outage. It is noticeable that a consid-
erable portion of the target area is outside the coverage service. While Fig. 4.38 (scenario
2) depicts the RSRP results of the combined coverage of AeNBs and conventional relay
nodes with omni-directional antennas. The situation in scenario (3) is showing a significant
enhancement of the service, with the introduction of CRNs. The placement of CRNs was
done according to the planning method described earlier that yielded 32 CRNs. Fig. 4.39
depicts the resulting combined coverage of AeNBs and CRNs. It can be noticed from the
figure that most of the CRNs activated both sectors, covering the service gaps left by the
aerial network. In order to present more quantitative indication of the gained enhancement,
a cumulative distribution function is plotted comparing the SINR of the three simulated sce-
narios (Fig. 4.40). Finally Fig. 4.36 shows a comparison between the service blockage
probabilities of the target systems calculated based on all assumed UEs in the model.
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 190

Fig. 4.37 RSRP Scenario 1 (AeNBs Only).

Fig. 4.38 RSRP Scenario 2 (AeNBs and Conventional RNs).


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 191

Fig. 4.39 RSRP Scenario 3 (AeNBs and CRNs).

Fig. 4.40 Comparative CDF Plot for the three Scenarios.


4. Aerial and Cellular Coverage Enhancement with Node Cooperation 192

Table 4.5 Simulation Parameters

Parameter Value

Layout of AeNBs Hexagonal grid 3 sectors per site 0◦ 120◦ 240◦


Layout of CRNs 6 Sector per CRN 0◦ 60◦ 120◦ 180◦ 240◦ 300◦
Propagation Environment Urban (City)
Simulation Frequency 2,000 MHz
System Bandwidth 5 MHz
AeNB Antenna Max Gain 8 dBi [ABSOLUTE Project]
AeNB Antenna H-Beam Width 60◦ degrees [ABSOLUTE Project]
AeNB Antenna V-Beam Width 60◦ degrees [ABSOLUTE Project]
AeNB Antenna Port power 23 dBm [ABSOLUTE Project]
LAP Height (AeNB) 1,650 m
CRN Antenna Max Gain 18 dBi
CRN Antenna H-Beam Width 65◦
CRN Antenna V-Beam Width Effect Not Simulated
CRN Antenna Height 5m
CRN Antenna Port power 37.7 dBm
Conventional RN Antenna Max Gain 6 dBi (Omni Directional)
Conventional RN Antenna H-Beam Width 360◦
Conventional RN Antenna V-Beam Width Effect Not Simulated
Conventional RN Antenna Height 5m
Conventional RN Antenna Port power 38.5 dBm
UE Antenna gain 0 dBi
UE Noise Figure 9 dB
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 193

4.4 Research Question 3 Conclusion


In this chapter, we presented a novel analytical method for studying the effect of a disas-
ter on cellular network performance, and showed how D2D technology can alleviate such
harm by extending the coverage originated from healthy cells towards the damaged cells.
A novel expression has been presented that describes the expected performance of chain
relay network based on q-Pochhammer symbol. We represented the alleviation introduced
by D2D as a buffer of how much damage the network can absorb while maintaining its
nominal service level. By applying practical network parameters we showed that a buffer
as high as 50-70% can be easily reached when D2D technology is embraced even with a
single hop relay. This fact is depicted in Fig. 4.41, illustrating the resulting damage al-
leviation buffer against a range of n number of cascaded chain relays. We have further

Fig. 4.41 Damage alleviation buffer against the number of cascaded relays.

verified our analytical approach by performing extensive Monte-Carlo simulations and then
comparing with the analytical network-level performance. The comparison showed very
close and homogenous statistics, indicating the validity of the presented analytical method.
Furthermore, we have presented the concept of deploying a temporary secondary cognitive
LTE based femto-network for supplementing and enhancing the network coverage, also we
have showed that significant enhancement was achieved by implementing a supplementary
network to the conventional LTE network. We also presented two feasible mechanisms for
cross-tier downlink interference management to further improve the signal to interference
ratio by mitigating the interference.
In this chapter we also introduced a novel practical approach in realizing cognitive re-
4. Aerial and Cellular Coverage Enhancement with Node Cooperation 194

lay nodes for serving rapidly deployable emergency networks. The obtained results show
significant performance enhancement over conventional relay nodes in terms of service and
coverage. Although the simulation was conducted for a specific REN architecture, CRN
concept can be adopted in some special scenarios related to commercial networks.
Chapter 5

Energy Efficiency Enhancement with


Node Cooperation and Relaying

The increasing energy consumption in expanding wireless networks poses a great challenge
for both user equipments manufacturers and mobile operators; where manufacturers are en-
deavouring to exploit better battery technologies and solutions for energy storage, and mo-
bile operators are trying to control their ever soaring energy bill. Moreover, reducing energy
consumption in public safety mobile equipments and nodes is of an utmost important dur-
ing critical operations where grid power is limited or unavailable, a factor that will directly
affect the duration of the active life time of the network and its nodes. One of the design so-
lutions is to utilise node cooperation and relays that can break down lengthy communication
distances into shorter segments, thus reducing the required energy for delivering the end-to-
end message. Where the total energy spent by all participating nodes should be lower than
the energy spent without using node cooperation. This chapter addresses research ques-
tion 4 through three key contributions in: (i) energy efficiency analysis for assisted D2D
under cellular networks in Sec. 5.1, (ii) energy efficiency with node cooperation for aerial
backhauled networks Sec. 5.2, and (iii) energy efficiency with complementary Mérn point
process Sec. 5.3. Where this chapter is based on the following publications contributing
towards sections 5.1, 5.2, and 5.3 respectively:

1. Akram Al-Hourani, Sithamparanathan Kandeepan, and Ekram Hossain, “Relay-


Assisted Device-to-Device Communication: A Stochastic Analysis of Energy Sav-
ing”, IEEE Transactions on Mobile Computing. (Accepted) [134]

2. Akram Al-Hourani, Sathyanarayanan Chandrasekharan, Abbas Jamalipour, Lau-


rent Reynaud and Sithamparanathan Kandeepan, “Optimal Cluster Head Spacing for

195
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 196

Energy-Efficient Communication in Aerial-Backhauled Networks”, in Global Com-


munications Conference (GLOBECOM), 2015 IEEE. (Accepted).

3. Akram Al-Hourani, Bill Moran and Sithamparanathan Kandeepan, “Regularizing


random points: Complimentary Matérn Point Process “, IEEE Transactions on Wire-
less Communications (Submitted)

In this chapter, we lay a mathematical framework for estimating the energy saving of a
relay assisting a pair of wireless devices. We derive closed-form expressions for describing
the geometrical zone where relaying is energy-efficient. In addition, we obtain the proba-
bilistic distribution of the energy saving introduced by relays that are randomly distributed
according to a spatial Poisson point process. Furthermore, we present a comparison method-
ology for fairly evaluating the energy consumption of conventional cellular network from
one side and relay-assisted device-to-device communication from another side. Results sug-
gest that a significant energy saving can be achieved when relay-assisted device-to-device
communication is adopted for distances below a certain threshold. In addition, we intro-
duce a novel analytic approach in clustering wireless nodes on ground fields through the
assistance of an aerial base station, forming an aerial backhauled network. On the ground,
some of the deployed nodes act as cluster heads, and aggregate and relay the terrestrial data
towards the aerial platform. We present a novel cluster-head selection algorithm based on
Matérn hard-core point process, and then we obtain the analytical formulation to optimise
the spacing between cluster heads to minimise the overall energy consumption. The formu-
lated problem utilises stochastic geometry to capture the random nature of the locations of
the deployed nodes, leading to a tractable analysis of the expected network metrics, where
we compare the performance of the proposed approach with other clustering algorithms,
and show the improvement in the energy-efficiency. Furthermore, we present a tractable ap-
proach for regularising randomly placed points, by splitting them into two subsets: the first
is generated by means of the Matérn hard-core point process, while the remaining points
constitute the complementary Matérn hard-core point process (CMHC). We study the char-
acteristics of this new process, deriving its pair-correlation functions, and the distribution
of the distance to the nearest neighbour. The results have several applications in wireless
communications, including the modeling of wireless sensor networks, where we investigate
an example of regularising such networks and illustrate its advantage in reducing the energy
consumption of wireless nodes.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 197

5.1 Energy Efficiency for Cellular Networks


The ever-expanding wireless communication infrastructure is withdrawing higher energy
than ever, raising the need for finding more efficient systems. The challenge of developing
efficient wireless systems can be addressed on several levels, starting from device electron-
ics, up to the network-level architecture and protocols [36, 111]. The anticipated gains
of achieving such efficiency is the essence of extending mobile devices’ battery life and
reducing environmental and economic impacts of wireless communication infrastructure.
Efficient wireless communication is not only important for commercial cellular networks,
but it is also vital in situations where mobile devices are forming an infrastructure-less ad-
hoc networks that rely on scarce energy sources. Examples of such networks are sensors
networks and public safety networks [122, 155]. Device-to-Device (D2D) communication
technology is one of the many prospective network-level technologies aiming for achieving
energy efficiency in cellular networks.
D2D technology allows peers to exchange data directly, without the need for routing
their traffic via a typical cellular base station (BS). Thus, for short distances, a consider-
able energy saving can be achieved by transmitting at lower power. Nevertheless, when the
distance between a communicating D2D pair is large, attaining such saving will no longer
be achievable. Determining the maximum distance for energy-efficient D2D communica-
tion is one of the currently open research areas [206]. Furthermore, in order to extend the
efficient communication distance between D2D pairs a mediating relay can be utilized to
breakdown the distant path into two shorter segments (see chapter 4), forming a two-hop
link. A significant energy saving might be achieved if the relaying device resides inside a
certain geometric region. We call this region here the energy saving zone, and we introduce
a new geometrical method for calculating the boarders of this zone based on the propagation
conditions, specifically, the path-loss exponent.
Assuming that the relaying devices are scattered in a random manner following a spatial
Poisson point process, we derive a mathematical framework to quantify the energy saving
resulting from the use of such relays to assist D2D pairs. Then we employ this framework
in a cellular communication system to compare the achievable energy saving resulting from
the utilization of the assisted D2D technology1 . In summary, the contributions of this work
can be stated as follows:

• We mathematically describe the energy saving zone between a communicating pair,


1 We refer here to a D2D pair with a mediating Relaying device as assisted D2D, and the assisting relay as

helper.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 198

Inside this zone, a relay would be energy efficient if utilized by the communicating
pair.

• We obtain the stochastic distribution of the energy saving introduced by relays inside
this region. In this case, the communicating devices have the choice to select the relay
candidate that provides the best energy saving.

• We present a novel approach to compare cellular energy consumption from one side
with the consumption of D2D and assisted D2D from another side.

• We obtain a closed-form expression for the maximum theoretical limit of energy effi-
ciency enhancement that could be achieved by assisted D2D technology.

• We find an expression of the distance threshold, beyond which no energy-efficient


D2D communication can take place. Also, we study the effect of mediating relays in
extending this distance threshold.

5.1.1 Related Work and Novelty of our work


Relay placement and selection is an important issue in most of wireless networks, having ap-
plications in cellular systems, ad-hoc and in sensor networks. Relay selection criteria varies
according to the imposed constraints on the network performance. Mainly, the selection
optimization aims to enhance the following network metrics: (i) energy-saving [115, 116]
(ii) capacity-enhancement [117–119], and (iii) QoS/connectivity of the nodes [120, 121].
Optimizing relay selection based on multiple criteria is also an open research topic. The
general approaches in investigating relay selection mechanisms and performance enhance-
ment ranges from studying the fundamental performance limits to studying the detailed
application-specific protocols, where the concept of having a relaying region is investigated
in the literature as part of either exploring the theoretical limits or to reduce the complexity
of relay selection protocols [115].
Several works aim to determine the optimum relaying region that is defined as the geo-
metrical region where relays can satisfy a target performance metric such energy, capacity
or QoS. In this study we focus on the energy-efficiency metric as the key parameter of
performance in energy-limited networks. An example study on this kind of networks is in-
vestigated in [36] where a proper relay selection technique is showed to be highly useful
for energy-constrained deployments such as temporary aerial-to-terrestrial communication
systems, in addition, it demonstrates that this efficiency can be achieved without affecting
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 199

Table 5.1 Comparison Among Different Relaying Region Formulation

Work Summary Ref. Network Approach Objective Relay Region


type
Provides an elliptical closed-form This Thesis Cellular / Stochastic ge- Energy- Elliptical
expression for the relaying region. D2D ometry efficiency
Quantifies the the expected en-
ergy saving using stochastic ge-
ometry. Quantifies the distance
threshold, beyond which no energy-
efficient D2D communication can
take place.
Proposes a protocol for locally opti- [115] Ad-Hoc Protocol- Energy- Bounded by
mizing the next hop relaying region. based efficiency several curves
Investigates optimal relay selection [116] Sensor net- Protocol- Energy- Near circular
and power allocation in cooperative works based efficiency
sensor networks.
Protocol-based relay selection for [33] Cellular Protocol- Energy- Simulation
energy-constraint networks to min- based efficiency near-elliptical
imise total-energy consumption per
information bit.
Tries to determine the probabil- [120] Two-hop Geometry- Connectivity Intersection of
ity that a relay exists between two based two circles
wireless derives.
Proposes a forwarding technique [121] Ad-Hoc / Geometry- Connectivity Intersection
with random selection of relays. Multihop based with a circle
Investigates how a certain relay can [117] Source- Geometry- QoS Intersection
be chosen within a relaying region. Sink based with a circle
Investigates random relay selection [119] Cellular Stochastic ge- QoS Intersection of
within a relaying region and then ometry two circles
compares the performance with
best relay selection.
Investigates two selection strategy: [118] Two-hop Protocol- Throughput Not specified
(i) random selection and (ii) greedy based
selection within a subset of relay
nodes.
Optimize the power of cellular and [101] Cellular / Stochastic ge- Throughput Circular
D2D systems and investigates opti- D2D ometry
mum relay selection range.

the target quality-of-service (QoS). The study in [207] also shows that if proper relay se-
lection strategy is adopted, the total expended energy can be minimized in a cooperative
network. The incentives for relays to facilitate the communication of other network nodes
can be modelled using game theoretic tools [12].
Determining the relaying region has been investigated in the literature using various ap-
proaches, we list a wide range of studies on relay selection in Table 5.1. These approaches
can be roughly grouped into two main categories (i) geometry-based (including stochas-
tic geometry) and (ii) protocol-based. In the geometry-based approaches, the relaying re-
gion is mapped into a deterministic geographic region, implicitly requiring the nodes to be
location-aware. The advantage of this method is the possible closed-form expression that
can accurately describe the relaying region. While the protocol-based approaches on the
other hand, heavily relies on signalling between nodes in order to learn the best cooperation
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 200

strategy. These type of approaches are more suitable for practical realizations.
The main novelty and differentiator in this work can be seen in three folds: (i) the ellip-
tical closed-form approximation of the relaying region when energy-efficiency is adopted as
the performance metric. (ii) Quantifying the the expected energy saving of randomly located
relay nodes. (iii) Developing a benchmarking methodology that fairly compares the energy
performance of cellular communication with D2D and assisted D2D communications, by
obtaining the distance threshold at which no efficient D2D communication can take place
beyond.

5.1.2 System Model and Assumptions


Energy Consumption Model

In our model the received signal power is given by [57]:

Pt −α
Pr = d h, (5.1)
Lo

where Pt is the transmit power, Lo is a constant that depends on the transmission frequency
and the antenna gains, d is the distance between the transmitter and the receiver, h is the
random variable representing the channel fading, and α is the path-loss exponent. A cer-
tain desired level of the signal-to-interference-plus-noise ratio (SINR) can be achieved by
adjusting the transmit power Pt , where the SINR is given by:

Pr
SINR = , (5.2)
PI + PN

the symbol PI is a random variable representing the interference power summing at the
receiver, and PN is the receiver noise which we assume to be the same for all the receivers
for simplicity.
In order to achieve a desired rate we set the SINR value to SINRo , and accordingly, by
using (5.1) and (5.2), the total energy consumption for the transmission and reception of the
signal over a session interval Tm is given by:

ε = Pt Tm + ERX
(PI + PN )
= SINRo Lo Tm d α + ERX = µd α + ERX , (5.3)
h

where ERX is the energy required to receive and decode the signal at the destination. The
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 201

physical interpretation of the variable µ = (PI +P


h
N)
SINRo Lo Tm is the amount of energy re-
quired to deliver a standard message to a receiver located at a reference distance, with the
aim of achieving a desirable SINRo at the receiver, we call µ here as the reference energy,
and we call ε the energy cost. The essence of the utilized model is very generic and ap-
plies to most of practical mobile devices. Note that the energy exerted for communication
between a transmitter and a receiver consists of two main parts: (i) a constant part rep-
resents the energy required to receive and decode the signal at the destination ERX , and
(ii) a distance-dependent part represented in µd α . The energy consumption model at the
signal/transmission (PHY) level is only relevant to our work here, while modelling of en-
ergy consumption at the MAC level signalling on the other hand is irrelevant since we only
consider signals at the transmission level at the PHY layer but not at the packet level.

Network Model

Fig. 5.1 The three different scenarios considered for the benchmarking.

The main aim of our work is to model the theoretical average energy saving that yields
from exploiting D2D and assisted D2D technologies, when compared to conventional cellu-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 202

lar communications. Accordingly, we define three distinct scenarios as depicted in Fig. 5.1.
The first scenario (i) represents the conventional cellular communication, where a source
can communicate to its destination only via the cellular network. In the second scenario
(ii) a source can transmit a message directly to its destination. The third scenario (iii) is
similar to the second scenario, except that a mediating relay can assist the communicating
pair in delivering the message, we call this scenario the assisted D2D scenario. The medi-
ating relay or helper is selected so that the total cost of delivering a message is minimized.
The communicating pairs in scenarios (ii) and (iii) can be located within the same cell or
in different cells, assuming that D2D communication links can be formed independently of
the cellular layout.
The cellular BSs are assumed to be randomly distributed following a homogeneous Pois-
son Point Process (PPP) with density λc BS per unit area [3], where the PPP model is quite
plausible in most of the practical cellular deployment cases [187]. The distribution of relays
follows another independent PPP with a density λr (relays per unit area).
Relays are assumed to have a decode-and-forward capability, i.e., if a D2D message is
routed via a relay, a negligible additional bit error rate (BER) is added to the end-to-end
link. Accordingly, relays can follow the same power control scheme mentioned previously
in order to hold the destination received power at a constant level.
The path-loss exponent is assumed to be the same for all scenarios, resulting from a
homogenous propagation environment. It is important to note that the model does not ac-
count for traffic congestions that might arise at a certain relay or BS, which is a reasonable
assumption if the capacities of these nodes are large enough. The mode selection, media
access control (MAC), or resource allocation for D2D communication are not the focus of
this thesis. The D2D devices could utilize spectrum underlay [22] or overlay methods, while
the former results in a higher spectral efficiency, the latter can reduce coordination between
the nodes. One main assumption here is that the utilized MAC should guarantee an average
homogenous interference temperature across the network.

5.1.3 Energy Saving by Relaying


In this section, we analysis the possible energy saving of relay-assisted D2D communica-
tion. We first determine the geometrical region where an assisting relay can be utilized
efficiently, i.e., can save energy when used. Secondly, we look into the probability of get-
ting a certain energy saving gain if a single relay is placed randomly inside this saving zone.
Finally, we extend the one-relay model to any arbitrary number of relays within the sav-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 203

ing zone, and then we deduce the expectancy of energy saving gain if relays are randomly
deployed according to a homogeneous PPP.

5.1.4 Energy Saving Zone


Without loss of generality, we assume a communicating D2D pair ∈ R2 where a sender S
is located at the cartesian coordinates (− d2 , 0) and a destination D at (+ d2 , 0). Furthermore,
we assume a relay R is located at a general point location (x, y).

Erx
Fig. 5.2 The energy saving zone for a α = 3, µ1 = 0.4, and different values of β , plotted as
per inequality (5.4).

Lemma 1. In order that the route through a relay R to cost less total-energy than the direct
path between S → D, the relay R should be located within an energy saving region Rα ⊂ R2
described in the following inequality:
" #α " #α
d 2
 2 2 2
d Erx
Rα : x+ + y2 + β x− + y2 < dα − (5.4)
2 2 µ1

Proof. Following equation (5.3), the energy cost of the direct path S → D is given by:

εD = µ 1 ||SD||α + Erx = µ 1 d α + Erx , (5.5)


5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 204

Fig. 5.3 The energy saving zone for a path-loss exponent α = 3, β = 1, and different values
of Eµrx1 , plotted as per inequality (5.4).

Fig. 5.4 Different realizations of energy saving zones in (5.4) corresponding to various val-
ues of the path-loss exponent α, notice the consistency with the elliptical approximation in
(5.10).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 205

where ||.|| is the Euclidian distance measure. The total-energy cost of the route via the relay
R is the summation of the energy costs between the S → R and R → D. That is,

εR = µ 1 ||SR||α + Erx + µ 2 ||RD||α + Erx


" 2 #α " 2 #α
2 2
d 2 d 2
= µ1 +x +y + µ2 x − +y + 2Erx . (5.6)
2 2

The condition of saving energy via a relaying route is to have εR < εD . We define β =
µ2
µ1 as the ration of the reference energies. Accordingly, the result in (5.4) will follow by
substituting from (5.5) and (5.6).

We can notice from Fig. 5.2 that the energy saving regions tend to shift towards the
µ
receiver in case of larger β = µ 2 . Also, we notice from Fig. 5.3 that these regions tend to
1
shrink for higher values of Eµrx , that is when the cost of receiving and decoding the message
1
becomes comparable to the cost of transmitting the message. All relays laying outside the
energy saving region are considered energy inefficient, that is, utilizing any of them by the
communicating pair will incur additional total-energy cost than the direct D2D link. We
study the average boundaries of the energy saving zone by taking the mean value of µ 1 , and
in order to keep the tractability of our work we assume that µ 1 d α ≫ Erx , thus d α − Eµrx ≈ d α .
1
This approximation is reasonable when the communicating distance is sufficiently large
[208], in contrary to multi-hop networks, where the accumulating effect of Erx cannot be
neglected. We also assume that µ is an i.i.d random variable in the network, resulting from
a homogenous channel fading processes, and a homogenous interference temperature [209],
where the media access protocol is capable of keeping the interference level of the network
as homogeneous and constant as possible. Thus β = 1, i.e., µ 1 ≈ µ 2 ≈ µ.
Accordingly, the saving zone described by the inequality in (5.4) depends only on the
propagation exponent α for normalized units with respect to distance d. Fig. 5.4 illustrates
this dependency, and also indicates the interesting case of α = 2, i.e., for free space path-loss
(FSPL) scenario where the energy saving region becomes a perfect circular disk.

Definition 1. In order to quantify the energy saving gain introduced when using a relay R,
we define the energy saving gain denoted by η as the normalized energy cost difference:

△ εD − εR
η= . (5.7)
εD

We can notice that the best energy saving gain obtainable by a mediating relaying device
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 206

is given by the following equation:

ηbest = 1 − 21−α . (5.8)

It is straightforward to see from equation (5.6) that the total-energy of the relaying route
εR increases when either x or y increases. Accordingly, the lowest εR can be achieved
when x = 0 and y = 0, i.e., when a relay is located at the origin. In this case the relaying

route energy cost is given by εR = 2 d2 , by substituting the latter and (5.5) into (5.7)
we can obtain (5.8). We notice that the best energy saving gain is only dependent on the
path-loss exponent α. When α = 2 (i.e., FSPL case), the best energy saving gain becomes
ηbest = 50%, i.e., routing a message via a relay placed at the origin can save 50% of the
energy cost.

Observation 1. The energy saving zone can be approximated by an elliptical region.

This observation can be easily made from Fig. 5.4. Approximating the energy saving
region by an ellipse leads to a much simpler analysis of the energy saving performance. We
start by defining the exact contour Cα surrounding the saving region Rα , where the energy
cost through the relay route of any relay R ∈ Cα is equal to the energy cost over the direct
path, i.e., εR = εD . Accordingly, we can write
" # α " #α
d 2
 2 2 2
2 d 2
Cα : x+ +y + x− +y = dα . (5.9)
2 2

Then we select the following equation for an ellipse to approximate the contour of the saving
region:
x 2
   2
y
Ĉα : + = 1, (5.10)
d/2 kd/2
where k is a stretching parameter for controlling the stretch of the ellipse over the y-axis. If
k is properly selected the four vertices of Ĉα can perfectly match with Cα vertices, also the
relative error of the surrounded area will be less than 1.5% over a practical range of α. Fig.
5.5 indicates the relative errors of the surrounded area for different values of α.
The stretching parameter k can be found by forcing the two contours Cα and Ĉα to match
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 207

Fig. 5.5 The relative error in the area of the energy saving zone when approximated by a
perfect ellipse.

at the point (0, k d2 ) as indicated in Fig. 5.6-a. Accordingly, we can write the following:

εr = εd
µ||SR||α + µ||RD||α = µ||SD||α
"    # α
d 2 d 2
2

2µ + k = µd α . (5.11)
2 2

The solution will give the value of k that satisfies the contour approximation:
q
2
k = 22− α − 1 : α ≥ 1. (5.12)

We can clearly notice that k is independent of the distance d between the source and the
destination.

5.1.5 Spatial Distribution of the Energy Saving Gain


In this section, we explore the spatial distribution of the energy saving gain η inside the
saving region Rα . We first define an equi-energy elliptical contour Cα (γ).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 208

Fig. 5.6 Approximating to an ellipse by matching the vertices.

Definition 2. For an equi-energy elliptical contour Cα (γ), all relays located at Cα (γ) have
the same amount of energy cost and can save the same amount of energy, defined as follows:
" # α " #α
d 2
 2 2 2
2 d 2
Cα (γ) : x+ +y + x− +y = γd α , (5.13)
2 2
where γ is the ratio between the energy cost via the relay and the direct link energy cost,
and is defined as follows:
△ εR
γ= = 1 − η. (5.14)
εD
Observation 2. The equi-energy contours can be approximated by elliptical contours.

This approximation can be clearly noticed from the plot in Fig. 5.7. The approximated
contours are selected to have the following form:
 2  2
x y
Ĉα (γ) : + = 1, (5.15)
S1 d/2 S2 d/2

where S1 and S2 are scaling parameters, in a similar approach to the approximation per-
formed in Observation 2.
In order to find the scaling parameters of Ĉα (γ) that are basically stretching the semi-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 209

Fig. 5.7 Equi-energy contours (5.13) for α = 3, also showing the scaled approximated el-
lipses (5.15).

major axis and the semi-minor axis, we force the two contours to match at two known points,
such as P1(S1 d2 , 0) and P2(0, S2 d2 ) as illustrated in Fig. 5.6-b. For point P1 in this figure,
we can write:

εr = γεd
µ||SP1 ||α + µ||P1 D||α = γ µ||SD||α
 α  α
d d d d
µ + S1 + µ − S1 = γ µd α . (5.16)
2 2 2 2

Similarly, for point P2 we can write


"   2 # α2
d 2 d
2µ + S2 = γ µd α . (5.17)
2 2

Solving equations (5.16) and (5.17) yields the following:

S1 = Solution of [(1 − S1 )α + (1 + S1 )α = 2α γ]
q
2/α
S2 = (2α−1 γ) − 1 : α ≥ 1, γ > 2α−1 . (5.18)
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 210

We notice that there is no generic closed-form expression for S1 . However, numerical solu-
tions are straightforward to obtain, also some closed-forms are obtainable for integer values
of α. We list some of these solutions in Table 5.2.

Table 5.2 Solutions of S1

Path-loss (α) S1 Condition


√ 1
2.0 2γ − 1 γ> 2
q
4γ−1
√ 1
3.0 3
γ> 4
q √
√ 1
4.0 2 2 γ +1−3 γ > 8

The set of equations describing the equi-energy contours in (5.13) indicates that higher
values of γ yield higher cost in delivering the traffic via the relay route. Accordingly, the
following conditions are always satisfied:

ε[C(γ1 )] > ε[C(γ2 )] ⇐⇒ γ1 > γ2 , (5.19)

where ε[.] is the energy cost of the equi-energy contour. We also denote the region enclosed
by the contour C(γ) as RC(γ) , where all the points within this area have less energy cost than
the contour points.

Lemma 2. If a single relay is randomly placed inside the saving region R, the cumulative
distribution function (CDF) of η is given by

S1 S2
Fη (u) = 1 − . (5.20)
k

Proof. By definition, the CDF of η is Fη (u) = P(η ≤ x). By looking at (5.14), we notice
that for larger values of γ we will have less energy saving. Mathematically,

Fη (x) = P(η ≤ u) = P(1 − γ ≤ u)


= 1 − P(γ ≤ v) = 1 − Fγ (v), (5.21)

where Fγ (v) is the CDF of γ. We note that the event of having γ ≤ v is equivalent of
having the relay point R(x, y) belonging to the region RC(γ) . Since the contours are non-
overlapping according to (5.19), the probability that R(x, y) ∈ RC(γ) will depend on the
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 211

spatial probabilistic distribution of R(x, y). Accordingly

(a) |RC(γ) |
P(χ ≤ γ) = P(R ∈ RC(γ) ) =
|RC(1) |
(b) π d2 S1 × d2 S2 S1 S2
= = , (5.22)
π d2 × d
2k
k

where (a) follows only when the spatial distribution of R inside the energy saving zone is
homogeneous and that there is only one relay inside the energy saving zone. The measure |.|
is the Lebesgue measure on R2 which is merely the area measure. RC(1) = Rα is the entire
energy saving region. The step (b) follows from Observation 2 where C(γ) is approximated
by an ellipse, knowing that the area of an ellipse is the product of the semi-major and semi-
minor axes and π. Using (5.21) and (5.22), the CDF of η can be obtained as given in the
Lemma.

In order to preserve the tractability in our study we focus on the case when α = 2.
This assumption allows us to derive closed-form expressions giving a clear insight into the
behaviour of the system. Later in Section 5.1.10, we illustrate that for higher values of α,
similar performance trends are observed.
In the particular case of α = 2, the CDF of energy saving gain η can be easily derived
from Lemma 2 as:
Fη (u) = 2u, u ∈ (0, 0.5], (5.23)

where η = 0 corresponds to the situation when the relay ∈ Cα=2 , and η = 0.5 corresponds
to the situation when the relay is located at the origin.

5.1.6 Energy Saving Gain With Multiple Relays


Now, we have obtained the CDF of the energy saving, resulting from a single relay existing
inside the energy saving zone. We look further for the general situation where several relays
exist inside the energy saving zone. In such a case, the message will have the choice to be
routed via the most energy efficient route, i.e., the communicating pair will select the relay
that will cost the least energy amongst the available set of candidate relays inside the energy
saving zone.

Theorem 3. In case of α = 2, the average energy saving obtainable from randomly de-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 212

ployed relays between two communicating devices is given by

N + e−N − 1
η̄ = , (5.24)
2N

d 2

where N = λr π 2 is the mean number of relays inside the energy saving zone.

Proof. If we have a number of n relays inside the energy saving zone, the overall energy
saving will be given by:
η(n) = max(η1 , η2 , . . . , ηn ), (5.25)

where {η1 , η2 , . . . , ηn } corresponds to the energy saving gain random variable vector result-
ing from each relay respectively. Therefore we can write:

Fη(n) (u) = P(η(n) ≤ u) = P(max(η1 , η2 , . . . , ηn ) ≤ u)


a
= Fη1 (u) × Fη2 (u) · · · × Fηn (u) = (Fη )n , (5.26)

where step (a) follows from the fact that η1 , η2 , . . . , ηn are independent and identically dis-
tributed random variables (i.i.d). For the particular case of α = 2, the CDF of η is charac-
terized in (5.23). Therefore, we can write

Fη(n) (u) = (2u)n . (5.27)

Accordingly, the PDF can be found be differentiation, as the following:

fη(n) (u) = 2n(2u)n−1 . (5.28)

We further obtain the average energy saving gain Eη [η(n)], when the number of relays is
known, as follows:
Z 1
2
η(n) = E[η(n)] = u fη(n) (u)du
0
 
1 n
= , (5.29)
2 n+1

expressing the energy saving gain expectancy over all possible realizations of a fixed number
of relays n inside the energy saving zone. It is clearly noticeable that the average value will
tend to 0.5 as the number of relays grows. Fig. 5.8 depicts this trend.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 213

Fig. 5.8 The expected value of the energy saving gain (η) for a fixed number of relays within
the saving zone, as per equation (5.29).

Since the distribution of the relays follows a PPP over the two dimensional euclidian
space R2 , and that the study region Rα ⊂ R2 , the number of relays within Rα is subject to
a Poisson distribution of the following form [77]:

(λr |Rα |)n


P(Φ(Rα ) = n) = exp(−λr |Rα |), (5.30)
n!
2
where Φ(z) is the counting measure of points in a region z. We note that |Rα | = π d2 .
In the light of equation (5.30) we can find the global expected value of H(n) using the
expectancy rule w.r.t. the discrete random variable n.

η̄ = En [H(n)]
h
d
2 in h
d 2
 i

1

n
 λr π 2 exp −λr π 2
=∑
n=0 2 n + 1 n!
∞   n
1 n N exp(−N)
=∑ . (5.31)
n=0 2 n + 1 n!
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 214

Furthermore, in order to have a practical expression of the global expected value η̄ without
resorting to the computation of the infinite sum in (5.31), we further simplify (5.31) to the
form indicated in (5.24), where the steps of simplification are shown below:

N n exp(−N)
∞  
1 n
η̄ = ∑
n=0 2 n+1 n!
1 ∞
Nn
= e−N ∑ n
2 (n + 1)!
"n=0   ∞  n+1 #
e−N ∞ (n + 1)N n+1 N
= ∑ −∑
2N n=0 (n + 1)! n=0 (n + 1)!
"   ∞  k #
−N ∞
kN k
(a) e N
= ∑ −∑
2N k=1 k! k=1 k!
e−N  N
(b)
Ne − eN + 1

=
2N
N + e−N − 1
= ,
2N

where (a) follows from the replacement of variables such that n + 1 → k, and (b) follows
from Taylor’s expansion of the exponential function.

We may notice from equation (5.24) that the upper limit of energy saving is:

N + e−N − 1 1
η̄best = lim = , (5.32)
N→∞ 2N 2

which again confirms our previous observation about the maximum possible energy saving
in equation (5.8) when utilizing a relay placed at the origin.

5.1.7 Energy Consumption Analysis


In this section, we derive the formulas required for describing the energy consumption in
each of our model scenarios as explained in Section 5.1.2, namely, the (i) conventional
cellular communication, (ii) D2D communication, and (iii) assisted D2D communication.

Cellular Communication Scenario

As depicted in Fig. 5.1, in a conventional cellular call the message is transmitted wirelessly
by a sender (source) to its serving BS, then travels through a fixed network (usually a wired
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 215

network) towards the far-end BS, and then finally it travels wirelessly again to the destina-
tion device. In order to understand how much energy is spent for communicating wirelessly
via a cellular network we should first analyze the distribution of the distance between a com-
municating UE and its serving BS, and then derive the expectancy of the energy associated
with the wireless transmission.
Since we have assumed previously that the cellular BSs are distributed stochastically
according to a homogeneous PPP denoted Φc , then we can apply the first contact distance
distribution [210] according to the following:

FRo (r) = 1 − exp(−λc πr2 )


fRo (r) = 2λc πr exp(−λc πr2 ), (5.33)

where FRo (r) and fRo (r) are the CDF and PDF of the first contact distance Ro , respectively,
representing the distribution of the distance between devices and their serving BSs. We
assume that a device will attach to its nearest neighbouring BS, that usually provides the
highest average SINR.
In addition, we know that the energy cost to transmit a message towards the serving BS
is ε = µ c rα , where µ c is the average cellular reference energy. Accordingly the expected
value of ε over Ro is given by:

Z ∞
ε̄ = ERo [ε] = ε fRo (r)dr
Z ∞ 0

= µ c rα × 2λc πr exp(−λc πr2 )dr


0
− α2 1 α
= µ c λc √ α Γ(1 + ). (5.34)
π 2

This indicates that in conventional cellular communication scenario the average consumed
energy in the wireless part of the network, only depends on the density of the BSs, and the
path-loss exponent.
The total cellular average energy is twice the average energy exerted in one side, since
we need the same average amount on the other end between the BS and the destination user
equipment (UE). Thus, the total cellular average energy for delivering end-to-end message
is:
−α 1 α
εc = 2 × µ c λc 2 √ α Γ(1 + ). (5.35)
π 2
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 216

Equation (5.35) is used here as a baseline for benchmarking the two other scenarios.

Device-to-Device Scenario

In contrary to conventional cellular communication, D2D capability allows more efficient


energy utilization when the distance between the communicating devices is relatively short.
According to our energy model, the consumed energy to deliver a message between two
D2D devices is merely related to the separating distance d, and is given by the following:

εd = µ d d α , (5.36)

where µ d is the average D2D reference energy.


The statistical distribution of the D2D distance is application specific. For instance, in
public safety networks, mobile users might be distributed in a clustered manner, allowing
a higher energy efficiency for D2D links, while in commercial cellular applications voice
calls usually take place between devices located at largely separated distances. Accordingly,
in order to highlight the possible merit of D2D we quantify the distance at which if D2D
communication is used, it can perform better than the average cellular communication.

Relay-Assisted Device-to-Device Scenario

In the assisted D2D scenario, the communicating devices have the choice to select the best
relay available in the energy saving zone. In this case, the introduced energy saving gain in
Theorem 1 will apply to equation (5.36). Hence, the resulting energy consumption is given
by:
εr = (1 − η̄)µ d d α , (5.37)

where η̄ is the average energy saving gain, that takes the form of equation (5.24) only when
α = 2.

5.1.8 Performance Benchmarking


Measuring the three scenarios (i), (ii), and (iii) in a single fair scale is a tricky approach.
Numerical examples might be case-specific and might not reflect the true behaviour of the
proposed model. Accordingly, we normalize the variables in order to avoid numerical com-
parison.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 217

Normalized Energy Cost

We normalize the D2D distance d with respect to average cellular contact distance do , i.e.,
our x-axis variable becomes x = ddo . Following (5.33), the expected value of the contact
distance is given by:

1
Z ∞
do = E[r] = r · 2λc πr exp(−λc πr2 )dr = √ . (5.38)
0 2 λc

On the other hand, we represent the density of relays as a multiple of the density of cellular
BSs, with a scaling factor κ, i.e., λr = κλc . The physical interpretation of κ is the average
number of relays underlaying a BS. This parameter can be easily dimensioned by network
planners.
α
ελc2
The normalized energy cost (y-axis) is selected to take the form , which gives a
µ
common scale to measure the energy consumption of all scenarios, assuming that µ c ≈ µ d ,
that is a homogenous interference and radio environment. The resulting normalized energy
consumption will follow from equations (5.35), (5.36), and (5.37) as given below:

2 α
yc = √ α Γ(1 + )
π 2
1
yd = α xα
2
1
yr = α xα [1 − η̄(x)], (5.39)
2

where yc , yd , and yh are the normalized energy consumptions for (i) cellular, (ii) pure D2D,
and (iii) assisted D2D scenarios, respectively. In case of α = 2, the average energy gain as
function of x is expressed as follows:

N + e−N − 1 π
η̄(x) = , N = κ x2 . (5.40)
2N 16

Fig. 5.9 shows the resulting plots based on equations (5.39) for the case α = 2. The
figure shows that there is a significant reduction in the energy consumption when utilizing
D2D technology (in scenario-ii) for short-range communications.

Normalized Distance Threshold

We denote the normalized distance threshold at which D2D communication becomes inef-
ficient as θo , beyond this threshold, D2D communications can not introduce energy saving.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 218

We call this threshold to be the saving distance threshold. From Fig. 5.9 we can notice
that adopting the relay-assisted D2D technique will further enhance the energy saving by
pushing the saving distance threshold θ farther.

Fig. 5.9 Normalized analytical benchmarking between the three scenarios.

We can obtain the saving distance threshold of D2D scenario, i.e., θo , by solving the
following set of equations from (5.39):

1 2
yd |θo = yc , thus θo2 = , (5.41)
4 π

and taking α = 2. This yields the solution θo = 2√π2 . In a similar manner, when utilizing
the assisted D2D technique, we can obtain the generic saving distance threshold by solving
the following equations:

1 2
yr |θ = yc thus θ 2 [1 − η̄(θ )] = , (5.42)
4 π

where the solution of (5.42) yields the value of the θ as a function of the relative density of
the helpers (i.e., relays) κ > 0 as follows:
r
W [e1−κ ] + κ − 1
θ =4 , (5.43)
πκ
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 219

in which W [z] is the Lambert W function [211], defined as the inverse of t → tet = z. Thus,
W [z]eW [z] = z. The proof of the solution is van be achieved by starting from equations (5.42)
and multiplying both sides by κ π4 as follows:

1 2 2
θ [1 − η̄(θ )] =
4 π
π N + e−N − 1 κ
κ θ 2 [1 − ]= .
16 2N 2
π 2
Since N = κ 16 θ as per (5.40),

e−N = N − κ + 1,

which is a transcendental function. It can be solved by multiplying both sides with eN−k+1 :

e1−κ = (N − κ + 1)eN−κ+1
(a)
W [e1−κ ] = N − κ + 1
r
(b) W [e1−κ ] + κ − 1
θ =4 ,
πκ

where (a) follows directly from the definition of Lambert function [211], and (b) from sub-
π 2
stituting N = κ 16 θ and rearranging the terms.
The relation between the saving distance threshold and the relative density of relays
is plotted in Fig. 5.10, showing that with the increase of the density of the helpers, the
saving threshold increases rapidly, then becomes marginal, and eventually reaches a certain
asymptotic limit denoted as θmax when the theoretical density of relays becomes infinite.
Noting that W [0] = 0, this limit can be mathematically expressed as:
r
W [e1−κ ] + κ − 1 4
θmax = lim 4 =√ . (5.44)
κ→∞ πκ π

Thus, in case

of α = 2, the upper and lower bounds of the saving distance threshold are given
2√ 2 √4
by θ = [ π , π ), where the lower bound corresponds to no relays in the network, i.e., the
D2D alone scenario, and the upper bound corresponds to an infinite density of relays.
It is interesting to look at the rate of change of the threshold distance θ with respect to
the density of relays κ, that is in order to visualize the effectiveness of adding more relays
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 220

Fig. 5.10 Saving distance threshold with respect to the relative density of relays, plotted as
equation (5.43).

to the network. Mathematically,


  
1−κ 2 + κW e1−κ − 1
  
dθ 2 W e
= −√ p . (5.45)
dκ πκ 3/2 (W [e1−κ ] + 1) κ +W [e1−κ ] − 1

The interpretation of (5.45) can be visualized through the plots in Fig. 5.11 showing a
sharply decaying trend. This trend can be seen not only for the case when α = 2, but also
for the higher values of α, as obtained by simulation.

5.1.9 Asymptotic Behaviour of the Distance Threshold


The results on the asymptotic behaviour of the distance threshold shown in (5.44) were
based on α = 2. In this section, we study the asymptotic behaviour of distance threshold
with varying α.

Lemma 3. The asymptotic value of the saving distance threshold when κ → ∞ is given by

4 h α i α1
θmax = √ Γ(1 + ) . (5.46)
π 2
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 221

Fig. 5.11 The rate of change in the saving distance threshold with respect to the relative
density of the assisting relays. Obtained by simulation and compared to equation (5.45).

Proof. Starting from equations (5.39) and writing yr |θ = yc , we have:

1 α 2 α
θ [1 − η̄(θ )] = √ α Γ(1 + ). (5.47)
2 α
π 2

Considering the fact that when κ → ∞ the energy saving gain η will asymptotically reach
the value as per (5.8), then equation (5.47) can be easily reduced to the form in (5.46).

5.1.10 Monte-Carlo Simulations and Discussion


Simulations

In order to validate the analytical approach presented in Sections 5.1.7 and Section 5.1.8,
we perform extensive Monte-Carlo simulations over 100,000 runs. In each run, we generate
a random number of BSs based on Poisson distribution and adhering to a certain density2 . It
is important to note that our analytical model is independent of the BS density, but rather it
accounts for the relative density of relays λr with respect to the density of BSs λc . We have
2 The actual simulation parameters are: λc = 200 × 10−6 BS/ul2 , number of users= 5, 000, and the map
size = 1, 000 × 1, 000 ul2
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 222

previously denoted this ratio as κ = λλcr . After that, BSs are deployed in a random manner
within the simulation map, their locations constitute the seeds of a Voronoi tessellation that
represents the dominance regions of each BS. The dominance region of a base station X or
its Voronoi cell is mathematically defined as [80]:


V (X) = {u ∈ R2 : ||X − u|| ≤ ||Z − u|| ∀Z ∈ Φc \ {X}}, (5.48)

where Φc is the realized set of BSs. It is important to note that the above definition is for
an infinite plane, while in our case the users are located within a confined simulation region
u ∈ R. In order to mitigate inaccuracy issue that might arise from the finite dimensions
of the simulation map, we limit the stretch of the deployment region of users so it does not
cover the entire simulation map, by this we guarantee that a user is having sufficient freedom
to attach to the nearest BS. Fig. 5.12 depicts one of the realizations of the cellular layouts,
with a reduced density, indicating the active calls between BSs and their associated users.

Fig. 5.12 Cellular layout indicating simulated calls via the cellular base stations.

For each of the simulation runs, we deploy a random density of relay nodes κ = 0 →
100. Also, we generate random calls between the deployed users, then for each call the
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 223

total-energy consumption using the three scenarios is registered. The total number of calls
in the simulation is 2.5 × 108 . The reference energy µ is calculate at users locations an
then averaged, assuming that the interference is originated from all base stations except the
closest one.

Fig. 5.13 Monte-Carlo simulation comparison with the theoretical equation plot of the dis-
tance threshold (5.43).

Results

After performing all the runs of Monte-Carlo simulation, we average the results obtained
for each value of κ as indicated in Fig. 5.13. The simulation results show high level of
consistency between the analytical approach and Monte-Carlo simulations. We also confirm
the independency of the BS densities, by repeating the simulation for different values of λc .
Extending the simulation to accommodate higher values of the path-loss exponent α
yields an expected result in that it positively affects the saving distance threshold. In other
words, the assisted-D2D technology enhances the energy performance further when the
radio propagation environment is impaired with heavier path-loss. Fig. 5.14 depicts this
relationship for four different values of α.
Also, we confirm the validity of the benchmarking of the normalized energy cost equa-
tions in (5.39) by comparing the averaged simulation runs with the theoretical plots as indi-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 224

Fig. 5.14 Normalized saving distance threshold (θ ) obtained for different values of α using
Monte-Carlo simulation.

cated in Fig. 5.15, showing high level of consistency and matching.


In Fig. 5.16, we show the simulated distance threshold when the density of relays κ = 12
as compared with the analytical asymptote obtained from equation (5.46). The figure also
suggests a very interesting result, that for practical destinies of helpers, we can achieve gains
ranging from 35% to 51% in extending the saving distance threshold of D2D communica-
tion. From the same figure, it is interesting to note that the asymptote is nearly proportional
to the path-loss exponent in a linear approximation.

Discussion

By observing the rate at which the saving distance threshold reaches its upper bound (refer
to Fig. 5.10), we can deduce that even for small values of κ the assisted D2D technique can
perform near to its theoretical limit, i.e., θmax . In fact, a value of κ ≈ 12 can reach 95%
of the maximum possible enhancement for low-loss environment. The interpretation of this
fact, loosely speaking, is that within a cell, it is sufficient to have an average of 12 relays to
obtain a saving distance closed to (2.20-2.40) times the average cell contact distance. It is
important to note that in practical deployments the actual performance will be also affected
by the traffic handling capacity of the relays. For example, if a relay is capable of routing
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 225

Fig. 5.15 Energy cost obtained using Monte-Carlo simulation as compared to the analytical
plots of equations (5.39).

Fig. 5.16 Achievable gains in the saving distance threshold for an example practical value
of κ = 12, as compared to the theoretical asymptote expressed in equation (5.46).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 226

one call only, thus, the overall virtual density of relays will be related to the mobile users
density. However, as indicated in the model assumptions, in this study we do not account
for the capacity limitations of the system.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 227

5.2 Energy Efficiency for Aerial Backhauled Networks


Based on the recent development in electronics and wireless systems, the concept of broad-
band networks facilitated by aerial platforms is becoming more feasible. In this type of net-
works, an airborne base station acts as a communication hub for terrestrial devices, enabling
a rapid and cost effective deployment of wireless coverage. Aerial-backhauled networks are
suitable for many envisioned applications, such as the disaster recovery of mobile commu-
nications (see section 4.2 and section 4.3 in chapter 4), where the concept of aerial disaster
recovery is gaining an increasing popularity among government agencies exemplified by the
EU-funded ABSOLUTE project [4]. In this project a major role is played by Low Altitude
Platforms (LAPs) that are used for providing airborne mobile coverage for first responders
and public safety agencies. An additional application area of aerial-backhauling is in the
field of Wireless Sensor Networks (WSN) where the deployed sensors can leverage good
communication conditions with the aerial platform, allowing a direct sensor-to-sink com-
munication scheme [212]. Aerial platforms [45] vary widely in terms of their operational
altitude, load capacity and mobility. In this section, we consider wireless networks back-
hauled by stationary or quasi-stationary platforms.
Depending on the nature of the target application, the terrestrial nodes could have plenty
of duplicated information that are transmitted towards the aerial platform. To deliver ter-
restrial traffic to an airborne platform, the conventional method is that for each node to
transmit directly to the aerial platform. Alternatively, in clustered networks an aggregating
node, called the cluster head (CH), can gather the information from several other neighbour-
ing nodes, called cluster members (CM), and then upload the information in a one extended
session towards the aerial platform. This mechanism can introduce significant energy sav-
ing when the information is compressible [109], or when the individual sessions have less
efficient payload-to-overhead ratio. Example applications that utilize this type of aggregate-
and-upload techniques ranges from rescue teams, telemetry/operation, seismic sensing to
Supervisory Control and Data Acquisition (SCADA) telemetry and environmental sensing,
where it is vital to extend the operational lifetime of terrestrial terminals that are battery-
powered and hence energy-constrained [155].
In order to facilitate the clustering scheme, an important challenge that needs to be
addressed is to find the best strategy for deploying CHs in order to minimize the overall en-
ergy consumption of the nodes. Some of the previous works such as LEACH [213] assume
a randomized selection of CHs which results in a spatially-irregular deployment of CHs.
Thus, leading to an increased energy consumption due to the widened distance distribution
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 228

between CMs and their serving CHs. Other related works such as HEED [110] and the
algorithm presented in [5] acknowledge the effect of having geometrical regularity in the
deployment of CHs, however they do not provide any tractable solution to achieve it. In this
section, we propose a CH selection algorithm that guarantees a minimum spacing between
CHs in order to increase their geometrical regularity, this algorithm is based on a hard-core
point process called Matérn hard-core point process type-II [80] that facilitates a tractable
mathematical analysis. Furthermore, by using tools from stochastic geometry, we present
an analytically tractable approach to optimize the minimum spacing between CHs, leading
to higher energy efficiency in the network. The contribution of this section resides in the
following folds:

• It provides a practical realization to utilize Matérn hard-core process in order to deploy


CHs in a regularized manner.

• It provides a novel mathematical model for representing the energy consumption for
both Matérn and randomly clustered networks (such as LEACH) underlying aerial
platforms.

• It provides an analytically tractable approach for optimizing the minimum spacing


between CHs, in order to maximize the energy saving.

• It provides a tight approximation of the contact distance distribution in Matérn hard-


core process.

• It provides Monte-Carlo simulation to verify the presented analytical approach.

5.2.1 Network Model


We consider an aerial-backhauled network, where an aerial base station is covering a certain
geographical area. Terrestrial nodes are deployed in a random and homogeneous manner
within this geographical area. We assume that all traffic should be either routed onboard
or forwarded to further destination(s) via the aerial platform. This assumption is applicable
when the majority of traffic is either server-client type or an off-site destination type. We
also assume that the terrestrial nodes are capable of forming direct links with each other,
where some nodes are selected to play the role of a decode-and-forward relay station, thus
serving the remaining nodes, by receiving and aggregating their information-payload, and
then uploading it towards the aerial platform. The whole scenario is depicted in Fig. 5.17
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 229

showing the layout of the proposed aerial-backhauled network. Each CH is responsible of


serving a certain area called a cluster.

[!t]

Fig. 5.17 The layout of an aerial-backhauled network, with Matérn clustering.

As stated previously the terrestrial nodes are assumed to be randomly and homoge-
neously distributed, where we study a temporal snapshot of the network assuming nodes to
be stationary, thus following a stationary Poisson Point Process (PPP). The intensity of this
process is denoted as λ , and accordingly the intensities of the CHs and the CMs are denoted
as λc and λm respectively, where λ = λc + λm .
CMs will associate to the closest CH, that is because the the closest CH provides the
highest average signal power. This scheme inherently divides the plane into randomly
shaped clusters called Voronoi tessellation, with the cluster heads acting as seed points in
this tessellation. We illustrate the concept of Voronoi clustering in Fig. 5.18 showing the
boundaries of the cluster regions and the associated cluster members. Noting that the mini-
mum allowable spacing between the cluster heads (denoted as δ ), is yet to be optimized.

Cluster Head Selection Algorithm

Our main aim is to increase the geometrical regularity in the CHs selection by defining
a minimum spacing δ between any two given CHs. We start by allocating a uniformly
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 230

[!t]
Fig. 5.18 Voronoi tessellations representing the service area of the cluster head.

distributed random mark t ∼ U (0, 1) to all nodes in the network. After that, every node
broadcasts its mark to other neighboring nodes in the circular zone having a radius of at
least δ . It is important to note that the parameter δ is assigned by a supervisory mechanism
(e.g. by the aerial platform), while the formula for calculating the best δ , which is one of
the major contributions of this work, is yet to be illustrated in Section 5.2.4.
After conducting the broadcasting phase, every node compares its own mark to all other
detectable nodes. If a node has the lowest mark, then it is flagged as a CH. When all nodes
perform this kind of comparison, all non-CH nodes will become CMs. The set of CHs is
denoted by Φc , while the set of cluster members is denoted by Φm . We illustrate the CH
selection algorithm in Algorithm 1, detailing the actual process in a wireless network, where
the operator ||x − y|| represents the Euclidean distance between points x and y. This method
of assigning the role of CHs is equivalent to Matérn Hard-Core Process of type II (MHC),
and it is described as [80]:


Φc = {x ∈ Φ : M (x) < M (y)∀y ∈ Φ ∩ b(x, δ ) \ {x}}, (5.49)
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 231

where M (x) represents the mark of point x, b(x, δ ) is a ball centered at x having a radius
of δ , and Φ is the parent PPP representing all nodes in the network. The above clustering
scheme yields two point processes, the first process Φc includes the CHs described in (5.65).
While the second process Φm includes the CMs, that can be expressed as Φm = Φ \ Φc . The
density of the MHC process is defined as [80]:

1 − exp(−λ πδ 2 )
λc = . (5.50)
πδ 2

Algorithm 1 Cluster Head Selection Algorithm


1: for all X ∈ Φ do
2: Mark the node X as M (X) = t ∼ U (0, 1)
3: Broadcast the mark M (X)
4: end for
5: for all X ∈ Φ do
6: X ∈ Φc “Assign X to the cluster heads set”
7: Build the list of detectable nodes Φ
b
8: for all Y ∈ Φb do
9: if ||X −Y || < δ and M (X) > M (Y ) then
10: X ∈ Φm “Assign X to the cluster members set
11: and remove it from the cluster heads set”
12: break
13: end if
14: end for
15: end for

Noticing that we implicitly assume that the distances to the detectable nodes are known.
This kind of awareness is not trivial to acquire when we aim high accuracy localization
[214], however a rough estimation of the distance can be easily achieved when relying on
the received power level.

Contact Distance Distribution

Each member of the CMs’ set Φm associates to its nerest CH, thus the serving region of a
CH is represented by a Voronoi cell as illustrated previously. It is to be noted that in this
work we concentrate on maximizing energy efficiency in a single clustering cycle, where
the presented work here can be further utilized to develop full clustering algorithms that
take into consideration the total life time of the network. To obtain the average energy
consumption in the terrestrial nodes, we need to understand the contact distance distribution
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 232

which is the distance distribution between the CMs and their respective CHs. We denote the
contact distance as R. In order to obtain the Probability Density Function (PDF) of R we
perform Monte-Carlo simulations for around 40,000 points using varying values of δ . After
that, we fit the results into the best empirical surface equation fˆR (r, δ ). Resulting from many
trials and errors we found a simple and accurate formula to best fit the Monte-Carlo samples
as the following:    
b √
ˆfR (r, δ ) = a r exp − r : λ δ ≥ 1, (5.51)
δ2 δ
where a and b are fitting parameters given by a=2.23 and b=3.32, resulting a root mean

square error of around 1% over the range λ δ ∈ [1, 6]. We illustrate the surface fitting in
Fig. 5.19, showing the simulation samples as circular points and the corresponding best fit
surface. We can notice from this figure the extent of fitting goodness.
[!t]

Fig. 5.19 Fitting the probability density function of the contact distance R between cluster
members and their serving cluster heads, for standard λ = 1.

5.2.2 Energy Consumption Model


The Signal to Interference and Noise Ratio (SINR) is an important measure that determines
the link quality and the achievable throughput. Thus, for a particular data payload say a
video streaming application, the required minimum throughput can only be achieved if the
SINR is above a certain threshold. We denote this threshold as SINRo . In general the SINR
Pr
is given by SINR = PI +P N
, where Pr is the received power from the serving transmitter, PI
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 233

is the total interference power summing at the receiver, this interference is originated from
all other transmitters that are active at the same radio resource interval, and PN is the noise
power.
In order to achieve an SINR level of SINR = SINRo at the receiver, the transmitter
should adjust its transmit power to counter the effects of the radio path-loss, the interference
and the noise. In this case, the SINR at the receiver will be of the form

Po
SINRo = , (5.52)
PI + PN

where Po = PLt is the received power, L is the average path loss incurred between the trans-
mitter and the receiver and Pt is the controlled transmit power.
The energy spent in the process of delivering a message between a communicating pair
transmitted over a period Tm can be modeled as [134]:

ε = Pt Tm + Erx
= SINRo (PI + PN )L × Tm + Erx , (5.53)

where Erx is the energy required to receive and decode the signal at the destination. We call
ε the energy cost of delivering a message between a communicating pair. Our interest is to
obtain the optimum performance of the average energy, that represents the average system
behavior, accordingly, we consider the average interference power PI in (5.53), assuming
that the media access protocol is capable of keeping the level of the interference temperature
[209] as homogeneous as possible.
In case of the terrestrial nodes i.e. cluster heads and cluster members, the exerted energy
to deliver a message from a cluster member towards its serving cluster head located at a
distance R, can be deduced starting from (5.53) as the following:

(a)
εTerr (R) = SINRo (PITerr + PNnode )Lo Rα × Tm + Erxnode
(b)
= µTerr Rα + Erxnode , (5.54)

where step (a) follows from assuming a log-distance channel model with average path-loss
L = Lo Rα , where α is the path-loss exponent, and Lo is the reference path-loss [1]. Step
(b) follows from defining the constant µTerr = SINRo (PITerr + PNnode )Lo Tm as the amount
of energy required to deliver a message with length Tm to a receiver located at a unity
distance, with the aim of achieving a desirable SINRo at the receiver. The symbol PITerr
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 234

denotes the average terrestrial interference power, and PNnode is the nodes’ internal noise
power. It is noteworthy that in our approach, the overhead required for CH selection is only
the broadcasting of the individual nodes’ scores and therefore is minimal.
On the other hand, air-to-ground (AtG) communication channel follows a specific path-
loss model which largely deviates from the log-distance approximation (see chapter 1),
namely the path-loss is related to the angle at which the platform is seen from the terrestrial
node (called the elevation angle). In our study we assume that the radius of the deployment
area is much smaller than the altitude of the platform, accordingly the elevation angle is ap-
proximated to hold the same. Thus, all terrestrial nodes share the same fixed path-loss LT2A
towards the aerial platform. In this case, the energy consumption involved in the process of
delivering a message towards an aerial platform, can be derived starting from (5.53) as the
following:

εT2A = SINRo (PIT 2A + PNa )LT2A Tm


= PoT2A Tm , (5.55)

where PIT 2A is the average aerial interference power, PNa is the platform’s noise power, and
PoT2A is the fixed transmit power required to deliver the message. The reason we have
disregarded the aerial platform’s receiver energy consumption is that in this work we are
only interested in the energy consumption of the terrestrial nodes.

5.2.3 Energy Consumption Analysis


Our focus in this work is on minimizing the total energy consumption involved in uploading
data from terrestrial terminals, the case where we assume that these terminals have limited
energy storage capacity. This scenario is quite common in sensor networks, where the
deployed sensors are battery-powered. We also assume that the system is not constrained
by the possible limitation in the aerial platform’s energy, and accordingly we do not account
its energy consumption in the optimization analysis.
The total energy consumption in terrestrial-terrestrial communication can be seen as
composed of two parts: (i) firstly, the total transmit energy from all CMs towards their CHs,
(ii) secondly, the total reception and processing energy exerted by all cluster heads. Thus,
the expected energy consumption can be expressed as the following:

εTotalTerr = E[ ∑ (µ| Terr Rα + E )]


x∈Φm
{z x} | rx{znode}
in CMs in CHs
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 235

(a)
= nm E [µTerr Rα ] + nm Erxnode
Z ∞
(b)
≈ nm µTerr rα fˆR (r, δ ) + nm Erxnode
0
   
a α 2+α
= A (λ − λc ) × δ Γ µTerr + Erxnode , (5.56)
b b

where E[.] is the expectancy operator, Rx is the distance between the cluster member x and
its serving cluster head. Step (a) follows from the fact that the distances between CMs and
their serving CHs are independent from each other. This independence is inherited from
the properties of the parent PPP. On the other hand, nm = E[Φm (A)] is the average number
of cluster members laying inside the target area. And since the deployment of the nodes is
assumed to be homogeneous, then nm is simply given by nm = Aλm , where A is the area of the
deployed nodes. The approximation in step (b) is the result of assuming that the simulated
contact distance distribution closely represents the actual distribution i.e. fˆR (r, δ ) ≈ fR (r, δ )
given in (5.51).
We can notice from equation (5.56) that terrestrial energy vanishes when δ → 0, that is
when all nodes are cluster heads, and no terrestrial traffic is taking place. In this case, the
traffic is uploaded directly to the aerial platform.
The total messages being relayed towards the platform are originated from both the CMs
as well as the CHs, since we assume that CHs still perform the function of normal nodes
(e.g. sensing) in addition to their function as aggregation nodes. By extending (5.55), we
can express the terrestrial-to-aerial energy cost as the following:

εTotalT2A = (nc + nm )PoT2A ξ Tm


= λ APoT2A ξ Tm = λ AµT2A ξ , (5.57)

where nc is the average number of CHs in the deployment area i.e. nc = Aλc , and µT2A =
PoT2A Tm . The factor ξ ∈ (0, 1] is the clustering energy-reduction factor that is achieved
through two main mechanisms:

• Aggregating and compressing terrestrial data, by removing redundant information ob-


tained from CMs.

• Overhead reduction, where the aggregated packets at the CHs are stripped and re-
encapsulated in a jumbo packet.

The compression gain profile depends on the application and the type of the data being re-
layed towards the aerial platform. For example, the compression in data acquisition sensors
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 236

(temperature, pressure, acoustic, location, etc...) will depend on the correlation between the
data being aggregated, and the details required by the sink (the base station). The aggre-
gation process itself is an energy-consuming process, where this extra energy consumption
can also be embedded within the energy-reduction factor ξ . Noticing the situation when no
clustering is used then ξ = 1. Our energy model provides flexibility in embedding the com-
pression and aggregation effect, however in the next section we provide an example formula
for ξ in order to get a better insight of the model dynamics.

5.2.4 Performance Benchmarking


In order to judge the performance of the clustered network against the original performance
of the unclustered network, we define the normalized energy difference, or the clustering
energy-efficiency, as the following:

εunclustered − εclustered
η= , (5.58)
εunclustered

where the energy consumption of the unclustered network is simply given by equation (5.57)
with ξ = 1, accordingly εunclustered = λ AµT2A . While the energy consumption of the clus-
tered network is the sum of the energy exerted in communicating terrestrially plus the energy
exerted in uploading the packets towards the aerial platform, accordingly:

εclustered = εTotalTerr + εTotalT2A . (5.59)

After substituting from (5.56) and (5.57), and performing some algebraic reductions, we
can write:
η = 1 − (β + ξ ), where (5.60)

    
λ − λc a α 2 + α µTerr Erxnode
β= δ Γ + . (5.61)
λ b b µT2A µT2A
Similarly we can find the normalized energy difference for LEACH [213] analytically,
that is when the locations of the CHs are uncorrelated. In this case CHs will follow a
PPP process because the random thinning of the parent PPP will yield another PPP [80],
and will have the well known closed form contact distance distribution given by fR (r) =
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 237

2πλc r exp(−πλc r2 ). Following the same steps in (5.56), we can obtain the related β as:

−α/2
  !
λ − λc λc h α i µTerr Erxnode
βLEACH = × √ α Γ 1+ + . (5.62)
λ ( π) 2 µT2A µT2A

In order to have an efficient clustering the normalized energy difference should be posi-
tive, namely η ∈ (0, 1]. The minimum energy consumption will be reached when the cluster
spacing δ maximizes η. Accordingly, the optimal minimum CH spacing can be obtained
as:
δopt = arg max (η) . (5.63)
δ

In order to get a deeper insight of the system performance dynamics, we utilize a simple
linear aggregation model where the number of the output message packets of the aggregator
is given by [215] y = κx + γ where κ is the compression factor, while γ is the overhead and
x is the number of input packets. Accordingly:

εTotalT2A = nc PoT2A yTm


λm
= nc µT2A [κ( 1 + ) + γ]
λ
| {z } | {zc }
CH message CMs messages
 
λc
= λ AµT2A κ + γ , (5.64)
λ

By comparing (5.64) and (5.57) we can deduce that ξ = κ + λλc γ, however by setting the
condition of ξ = 1 at λλc = 1 (i.e. no energy-reduction when all nodes are CHs), then we get
ξ = (1 − κ) λλc + κ. An illustration of the energy-reduction factor is shown in Fig. 5.20 for
different values of the compression factor κ, noting that there is no obtainable optimum for
the case κ = 1.
We illustrate the energy performance by substituting numerical values from Table 5.3
into equation (5.60) and then we plot in Fig. 5.21 the resulting curves for different values
of the compression factor κ. Also, we show the results of Monte-Carlo simulations, that
exhibit a trend very close to the analytical results, indicating the validity of our proposed
analysis.
We also compare the performance of the proposed approach to HEED [110] and the
algorithm presented in [5] by performing simulations. The results are shown in Fig. 5.21.
In HEED, the secondary parameter used to select the CHs is called AMRP and for both
the algorithms, simulations were performed for different cluster radii. In order to unify the
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 238

[!t]

Fig. 5.20 The energy-reduction factor ξ for different values of the compression factor κ.

Table 5.3 Notations and Symbols

Symbol Numerical Value Explanation

SNIRo 10 dB Target Service SINR


λ 1/1000 m−2 Terminals density
α 3.68†1 Terrestrial propagation exponent
Lo 26 dB†1 Terrestrial reference path-loss
LT2A 120 dB†2 Terrestrial to Air Path-loss
PITerr †3 0 Terrestrial interference
PIT2A †3 0 Aerial interference
PNa = PNnode -99 dBm †4 Noise power
Erxnode 100 mJ [36] Nodes receiving energy

†1 Winner II scenario A1 [58].


†2 Free space path-loss of aerial altitude 12 km and frequency 2 GHz.
†3 Interference is ignored for simplicity.
†4 Calculated from noise figure of 5 dB and bandwidth 10 MHz.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 239

x-axis with the proposed approach in this work, we translate the resulting λc from other
algorithms into δ using (5.50).

[!t]

Fig. 5.21 The normalized energy difference η for different values of the compression factor
κ, showing Matérn hard-core case, LEACH, HEED and the algorithm presented in [5].

We can note from Fig. 5.21 that MHC clustering approach, proposed in this work, out-
performs random clustering (LEACH), HEED and the algorithm presented in [5] especially
for lower densities of CH (i.e. higher spacing δ ). We can also observe from the same figure
that clustering is rendered inefficient when there is no compression i.e. κ = 1.
We note here that the scores given to the nodes in Alg. 1 could be altered to include
parameters such as the node’s residual energy. However, it is out of the scope of this work
and will be considered for future work.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 240

5.3 Energy Efficiency with Matérn Hard-Core Point Pro-


cess
The relative tractability of the Matérn hard-core Process (MHC) has earned it attention in
several practical fields. An additional advantage for MHC is the fact that points are banned
from being closer to each other than a pre-defined distance called the hard-core distance.
This should be compared to a Poisson point process (PPP) where the interaction between
pair of points is absolutely null.
Practical random point sets are usually featured with a kind of “correlation” between
their points. As an example, consider trees in a forest represented as a point process. Bi-
ological constraints reject the possibility of two trees growing at a very close proximity
distance R → 0. Similarly, while the deployment of cellular wireless base stations has a
certain degree of randomness because of the uncertainties related to the availability of land,
utility, planning permits and other socioeconomic factors; however, a strong correlation ex-
ists for close-by base stations. It is common to use PPP model to represent the distributions
of base stations [80, 85, 89], aiming to characterize the worst-case deployment scenario,
and to exploit the tractability facilitated by PPP. However, in practical scenarios, two base
stations that use the same frequency (or radio resources) cannot co-exist in a very close
proximity because of the strong induced mutual interference [88].
Wireless sensor networks (WSN), where sensors are usually deployed randomly, consti-
tute a rich field for applying stochastic geometry, and specifically hard-core point processes.
These processes can be applied to model some media access techniques such as the Carrier
Sense Multiple Access (CSMA) [83], and to model the cluster head spacing [84]; cluster-
ing randomly places sensors so that cluster heads are chosen to perform some designated
tasks, such as the aggregation and compression of data from other sensors (called cluster
members). The concept of clustered sensors is depicted in Fig. 5.22, where originally all
points (squares and circles) follow a Poisson point process, assuming a full spacial random-
ness in the deployment of the sensors. However, the sensors self-organize in clusters, where
each cluster is served by a cluster head (squares) and has an almost surely finite number
of associated cluster members (circles). In such a clustered deployment, the regularity of
cluster heads is an important aspect to minimize the variations in distances between cluster
members and their serving cluster heads. This is particulary important in WSN network
applications, otherwise some sensors will spend very high transmission power to reach their
cluster heads. Instead it is preferred to have a homogenous energy consumption [112] so as
to extend the life span of the energy-limited sensors (e.g. when battery-powered).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 241

10
Cluster Heads (MHC)
Cluster Members (CMHC)
8 b(x,δ)

4
Map Y-Axis [Unit of length]

-2

-4

-6

-8

-10
-10 -8 -6 -4 -2 0 2 4 6 8 10
Map X-Axis [Unit of length]

Fig. 5.22 Clustered wireless sensor network using Matérn hard-core point process, showing
the hard core region i.e. b(x, δ ), where no two cluster heads are allowed within. The leftover
points constitute the complementary process CMHC (cluster member). λ = 1, δ = 2.

We present an extension of the well-known Matérn hard-core (MHC) type-II point pro-
cess [80, 216]. A new point process, the Complementarity Matérn Hard-Core point process
(CMHC), is defined and studied. In particular, we study its pair-correlation function and its
two-point Palm probability, aiming to shed some light on the regularity of the points of the
CMHC. We prove that the CMHC process behaves asymptotically like a Poisson point pro-
cess. We also study the statistical distribution of the nearest neighbour distance between the
points of CMHC and the MHC processes, presenting this distribution in an empirical closed
form approximation. As an example application, we employ this empirical form to illustrate
the reduction in the energy consumption of WSN nodes due to the increased regularity of
the MHC process.

5.3.1 Complementary Matérn Hard-Core Point Process


In this study we focus on MHC type-II since it has a higher density than MHC type I [179],
thus yielding a more efficient child point process out of the generating parent PPP process.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 242

To construct an MHC type II process, we start with some real number δ > 0 and a parent
stationary Poisson point process (PPP) denoted by Φo , with a homogeneous intensity of λo
(points per unit area), and then for each point x ∈ Φo a uniformly distributed random mark
M (x) = t ∼ U (0, 1) is assigned. A point x ∈ Φo will be flagged-for-removal if its mark
is the least compared to other points within a ball of radius δ , b(x, δ ), around it. Only after
this test is done for all points in Φo , the points that are flagged-for-removal are removed
together. The remaining points, constituting the resulting MHC process are described by
[80]:

ΦMHC = {x ∈ Φo : M (x) < M (y) ∀y ∈ Φo ∩ b(x, δ ) \ {x}}, (5.65)

where δ is the the hard-core distance. It be can easily shown that the resulting MHC has an
intensity of [80]:
1 − exp(−λo πδ 2 )
λMHC = . (5.66)
πδ 2
One of the interesting features of the MHC is that the outcome of its thinning process is
independent of the order in which the points are considered for removal, because the actual
removal of points (the thinning) is only performed after all points in Φo have been checked
against their neighbours.
In this way, we extract a child MHC process ΦMHC out of the parent PPP Φo . As a
side product, another child process denoted by ΦCMHC constituting the leftover points, i.e.
ΦCMHC = Φo \ ΦMHC is produced. This child process is called the Complementary Matérn
Hard-Core point process (CMHC). The resulting MHC and CMHC of the same parent are
called sibling processes. We formally define it in Definition 3.

Definition 3. The Complementary Matérn Hard-Core point process (CMHC) is defined by:


ΦCMHC = {x ∈ Φo : M (x) > M (y) ∃y ∈ Φo ∩ b(x, δ ) \ {x}}, (5.67)

where Φo is the generating parent PPP.

Since the parent process is exactly divided into two child processes, then the intensity of
CMHC can be calculated as follows:

λCMHC = λo − λMHC
1 − exp(−λo πδ 2 )
= λo − . (5.68)
πδ 2

Figure 5.22 is based on a realization of the MHC point process and its complement
CMHC; also we indicate the hard-core region of the MHC b(x, δ ). This realisation is applied
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 243

to a wireless sensor network as an example, where cluster heads are modelled using the
MHC and cluster members are modelled using the left over points CMHC.

5.3.2 Points Interaction in CMHC


Points interaction behaviour in a given point process can range from repulsive to attractive,
while PPP is distinct with no interaction among its points. A fully regular point process
(with repulsive behaviour) forms a lattice e.g. square, triangular, where point locations
become deterministic, while on the other hand an attractive behaviour between points leads
to geometrical-clustering, that is an inhomogeneous intensity where some points in close
proximity are grouped together in clusters. It is important to note that geometrical-clustering
is a different concept than protocol-clustering. The latter is commonly referred in wireless
sensor network literature as the process of selecting cluster heads in a network, and thus
dividing the network into logical clusters (or subsets) where each cluster is served by a
cluster head.
For the CMHC process we should expect a sort of interaction between its point as it is
the result of a non-random thinning of a PPP. In order to quantify this interaction we start
with the following Lemma.

Lemma 4. For a point x1 ∈ ΦCMHC , the probability of another point x2 located at a distance
r to migrate from the parent PPP to ΦMHC is given by:

β (r, δ ) = P(x2 ∈ ΦMHC |x1 ∈ ΦCMHC )


ρMHC (δ ) − kMHC (r, δ )
= , (5.69)
1 − ρMHC (δ )

where kMHC (r) is the MHC two-points Palm probability, ρMHC (δ ) is the retention probabil-
ity of a single point when the hard-core distance is δ .

Proof. Assume that two points x1 and x2 , originating in the parent PPP Φo , are separated by
distance r. After MHC-thinning, we face four distinct possibilities: (i) both points migrate
to ΦMHC , (ii) both points migrate to ΦCMHC , (iii) x1 migrates to ΦMHC and x2 to ΦCMHC ,
and (iv) vise-versa. The first event is well defined in the literature as the MHC two-point
Palm probability: the probability of two points separated by distance r to be both retained
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 244

in (migrate to) the resulting MHC process. It has the form [81]:

 0
 r<δ
2

kMHC (r) = 1−e−πδ λo −λo ν(δ ,r)
(5.70)
2 + e ν(δ ,r) −1
πδ 2
r≥δ


λo2 (ν(δ ,r)−πδ 2 )

where ν(δ , r) is the area of the union of two disks of radius δ with centres separated by a
distance r given by:
 q
r 2
2πδ 2 − 2 arccos δ2 + δ 2 − r4 r r ≤ 2δ
 
ν(δ , r) = 2δ , (5.71)
 2πδ 2 r > 2δ

and illustrated in Figure 5.23.

Fig. 5.23 The intersection of the balls b(x1 , δ ) and b(x2 , δ ).

We also calculate the probability of a point x to individually migrate to the child CMHC
(also called the retention probability). This event is the complement of the event of having
the point selected as member of the MHC child process, and the probability of the latter is
[179]:
Z 1 Z 1
exp −πδ 2tλo dt

ρMHC = V dt =
0 0
1 − exp −πδ 2 λo

= , (5.72)
πδ 2 λo

where the integrand


V = exp −πδ 2tλo ,

(5.73)

is the void probability in a Poisson point process of intensity tλo , since in order for a point
x to be retained in MHC, the mark of x is less than that of all other points in the ball b(x, δ ).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 245

The parent marked PPP, can be seen as a t-thinning process when considering the points
that are marked with M (x) ≤ t, yielding an of intensity λot. That is because the random
thinning of a PPP yields another PPP [79].
Accordingly, the retention probability of the CMHC process is:

1 − exp −πδ 2 λo

ρCMHC = 1 − ρMHC = 1 − . (5.74)
πδ 2 λo

On the other hand, applying Bayes’ rule we can write:

P(x1 ∈ Φ ∩ x2 ∈ Φ) = P(x1 ∈ Φ) × P(x2 ∈ Φ|x1 ∈ Φ), (5.75)

where we simplify notation: Φ ≡ ΦMHC and Φ ≡ ΦCMHC for convenience. Similarly we


write:
P(x1 ∈ Φ ∩ x2 ∈ Φ) = P(x1 ∈ Φ) × P(x2 ∈ Φ|x1 ∈ Φ). (5.76)

A further application of Bayes’ rule gives

P(x1 ∈ Φ ∩ x2 ∈ Φ)
P(x2 ∈ Φ|x1 ∈ Φ) = . (5.77)
P(x1 ∈ Φ)

This probability is the complement of the probability P(x2 ∈ Φ|x1 ∈ Φ). Taking this fact
and substituting (5.77) into (5.75), we obtain
 
P(x1 ∈ Φ ∩ x2 ∈ Φ)
P(x2 ∈ Φ ∩ x1 ∈ Φ) = P(x1 ∈ Φ) × 1 − . (5.78)
P(x1 ∈ Φ)

Since the probabilities P(x2 ∈ Φ ∩ x1 ∈ Φ) and P(x1 ∈ Φ ∩ x2 ∈ Φ) are equal, we can use
(5.78) and (5.76) to deduce that

P(x1 ∈ Φ) − P(x2 ∈ Φ ∩ x1 ∈ Φ)
P(x2 ∈ Φ|x1 ∈ Φ) = . (5.79)
P(x1 ∈ Φ)

Using the facts that P(x2 ∈ Φ ∩ x1 ∈ Φ) is the MHC two points Palm probability kMHC (r),
that P(x1 ∈ Φ) is the MHC retention probability ρMHC , and that P(x1 ∈ Φ) = 1 − P(x1 ∈
Φ) = 1 − ρMHC , we deduce the result of the Lemma.

Now the conditional probability given by Lemma 4, provides the following interesting
corollary.

Corollary 1. If a point x ∈ PPP migrates to the CMHC process, then the probability of its
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 246

nth nearest neighbour to migrate to the sibling MHC is given by:


Z ∞
Pn (δ ) = β (r, δ ) × fn (r)dr, (5.80)
0

where fn (r) is the contact distance probability density function to the nth nearest neigh-
bour in the parent PPP process, given by [80]:
n
−λ πr2
2
fn (r) = 2 exp(−λ πr ) , (5.81)
rΓ(n)

where Γ(n) is the gamma function.

Proof. This is a straightforward consequence of the fact that the expectation of the proba-
bility function β (r, δ ) over r, that is Pn (δ ) = Er [β (r, δ )], where r is considered as a random
variable.

We visualize in Figure 5.24 the conditional probability Pn (δ ) for a range of hard-core


distances δ assuming a unity generating intensity. Notice that this probability is rapidly
flattening for higher values of δ indicating a rapid decrease in the interaction between the
point x ∈ CMHC and its MHC neighbours. The conditional probability Pn (δ ) approaches
the corresponding MHC retention probability ρMHC , as β (r, δ ) becomes less dependent on
r for larger distances.
Based on Lemma 4, the following theorem characterizes the pair correlation function in
a CMHC process. The pair correlation function captures the regularity of the points in a
point process; when it is is equal to unity, the points have no correlation in their locations
(indicating complete randomness, as is the case of PPP), while values less than unity indi-
cates a repulsive behaviour between points e.g gas molecules, and values greater than unity
indicate an attractive behaviour [80] e.g. matter distribution in the universe.

Theorem 4. The pair correlation function of the complementary Matérn hard-core point
process is given by the following expression:

1 − 2ρMHC + kMHC (r)


gCMHC (r) = 2
. (5.82)
1 − 2ρMHC + ρMHC

Proof. We use the concept of the second moment density (or the second order product
density) [179], that describes the probability of having two points separated by a distance r
to exist in infinitesimal balls of radii dr. When the generating point process is homogeneous
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 247

0.5
δ=1
δ=1.5
δ=2
0.4
Conditional Probability P n [ ]

δ=3

0.3 ρMHC =0.3

0.2

ρMHC =0.14
0.1
ρMHC =0.08

ρMHC =0.035
0
5 10 15 20 25 30
Neighbour Order (n) [ ]
Fig. 5.24 The conditional probability of the nth neighbour of a CMHC point to migrate to
the MHC process.

and stationary, this density is described in terms of the two-point Palm probability k(r) as
follows [179]:
ρ (2) (r) = λo2 k(r). (5.83)

Accordingly we can calculate the corresponding pair correlation function [80, 84], from

ρ (2) (r)
g(r) = , (5.84)
λ2

where λ is the intensity of the point process under examination, while λo is the intensity of
the parent point process. The pair correlation function captures the regularity degree of a
point process.
The two-point Palm probability of the CMHC process kCMHC (r) is the probability that
both points x1 and x2 migrate to the CMHC from the generating PPP, so that

△ 
kCMHC (r) = P x1 ∈ Φ ∩ x2 ∈ Φ . (5.85)
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 248

By applying Bayes’ rule we have


 
kCMHC (r) = P x1 ∈ Φ × P x1 ∈ Φ|x2 ∈ Φ
  
= P x1 ∈ Φ × 1 − P x1 ∈ Φ|x2 ∈ Φ
a
= 1 − 2ρMHC + kMHC (r), (5.86)

where the last step (a) follows from Lemma 4, and the fact that P x1 ∈ Φ = 1 − ρMHC . By
substituting into (5.83) and then into (5.84) we deduce the conclusion of Theorem 4, taking
into consideration that the density of a CMHC process is λCMHC = λo ρCMHC .

The pair correlation functions of both the CMHC and MHC processes are plotted in
Figure 5.25 for several different values of δ , using a unity parent intensity. The graphs
indicate that the points in CMHC rapidly lose their interaction for higher values of δ , and
that there is no correlation taking place beyond r > 2δ . On the other hand, the graph of MHC
indicates that the points are banned from co-existence closer than r < δ , since kCMHC (r) = 0,
and then exhibit a rapid attractive behaviour between r = δ and r = 2δ , since kCMHC (r) > 1.

Theorem 5. The complementary Matérn hard-core point process converges weakly to a


Poisson point process as δ → ∞.

Proof. We invoke the definition of weak convergence of point processes [217] and require-
ments to prove such convergence. Consider, for this, a family of bounded and disjoint Borel
subsets of R2 denoted by A1 , . . . , Am ∀m ∈ N+ . Convergence of the CMC process to the PPP
is then equivalent to:

d
{NCMHC (A1 ), . . . , NCMHC (Am )} → {NPPP (A1 ), . . . , NPPP (Am )} , (5.87)

where NCMHC and NPPP are the count measures on the point processes CMHC and its parent
PPP respectively. The letter d refers to convergence in distribution. Invoking Definition 1,
we see that
lim P(NCMHC (A) = k) = P(NPPP (A) = k) (5.88)
δ →∞

is satisfied for any bounded Borel set A, because a ball of radius δ → ∞ will eventually cover
the Borel set A regardless of the location of the centre of the ball. This ball prevents any
point x ∈ A from being thinned, and accordingly all points will remain in A almost surely.
Now (5.87) follows.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 249

1.2
Pair Correlation Function g CMHC(r) [ ]

0.8
Hard-core distance
0.6 (δ) = 1, 1.5, 2, 3

0.4
δ=1
δ=1.5
δ=2
0.2
δ=3
PPP
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Two-Points Distance (r) [Unit of Length]

1.2
Pair Correlation Function g MHC (r) [ ]

0.8

0.6 Hard-core distance


(δ) = 1, 1.5, 2, 3
0.4
δ=1
δ=1.5
δ=2
0.2
δ=3
PPP
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Two-Points Distance (r) [Unit of Length]

Fig. 5.25 Pair correlation function of the MHC process and its complement CMHC,
gCMHC (r) and gMHC (r) respectively.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 250

It is also interesting to note that asymptotic value of the pair correlation approaches unity
when δ → ∞ because:

lim ρMHC = 0, and lim kMHC = 0. (5.89)


δ →∞ δ →∞

Referring to the result of Theorem 4, we see that

lim g(r)CMHC = 1, (5.90)


δ →∞

indicating that no interactions between points takes place as δ → ∞, as is the case for the
PPP. To visualize the rate at which the pair correlation function of the CMHC approaches
unity, we investigate the relation between the gCMHC (r) at r < δ from one side versus δ on
the other side; that is, we plot the following function:

1 − 2ρMHC
gCMHC (r < δ ) = 2
1 − 2ρMHC + ρMHC
2
ρMHC
= 1− 2
, (5.91)
(1 − ρMHC )2

where kMHC (r < δ ) = 0 according to (5.70). The asymptotic behaviour is depicted in Fig-
ure 5.26, showing a rapid approach to unity for increasing values of δ .

5.3.3 Nearest Neighbour Distribution


The nearest neighbour distribution [80, 218] is an important property of a point process,
where the nearest neighbour to a point x ∈ ΦCMHC in ΦMHC is given by :


NNMHC (x ∈ ΦCMHC ) = arg min{∥y − x∥}, (5.92)
y∈MHC

where ∥y − x∥ is the Euclidian distance measure between x and y. The distance to this
nearest neighbour is a random variable, given by


R(x ∈ ΦCMHC ) = ∥x − NNMHC (x ∈ ΦCMHC )∥, (5.93)

To approximate the Probability Density Function (PDF) of R, we fit the empirical histograms
generated by performing a large number of Monte Carlo simulations. This empirical distri-
bution is denoted by fˆR (r, δ ), where the selected form is chosen to balance between simplic-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 251

1
Pair Correlation Function g CMHC(r<δ) [ ]

0.8

0.6 Intensity (λ) = 0.5, 1, 2, 4

0.4

δ=0.5
δ=1
0.2 δ=2
δ=4
PPP
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Hard-Core Distance (δ) [Unit of Length]

Fig. 5.26 The asymptotic behaviour of the CMHC pair correlation function gCMHC (r < δ )
formulated in (5.91).

ity and accuracy. As a result of many trials and errors we found a simple and quite accurate
formula that provides a good fit to the Monte Carlo sample histogram is

r
   
r b √
fˆR (r, δ ) = a 2 exp − : λ δ ≥ 1, (5.94)
δ δ

where a and b are fitting parameters, chosen to be a=2.24 and b=3.32. This results in a root

mean square error of around 1% over the range λ δ ∈ [1, 6]. It is worthy mentioning that a
and b are dependent since 0∞ fˆR (r, δ ) = 1, so that a = Γ b2 . We illustrate the empirical fitting
R
[b]
in Figure 5.27, showing the simulation samples as discrete points and the corresponding
fitted curves.

5.3.4 Energy Consumption Analysis


To provide an insight to the benefits resulting from introducing regularization in a clustered
wireless sensor networks using CMHC, we analysis the average energy consumption in-
volved in transmitting a wireless packet from a cluster member to its serving cluster head.
We compare three scenarios: (i) in the first scenario the cluster heads are assumed to be
distributed according to a PPP randomly thinned from the parent PPP consisting of all the
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 252

1.4
δ=1
Probability Density Function (PDF) [ ]

δ=1.5
1.2
δ=2
δ=3
Fitting
1

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Contact Distance (r) [Unit of Length]
Fig. 5.27 The simulation of the contact distance distribution (normalized histogram), com-
pared with the empirical formula in (5.94).

Cluster Heads
Cluster Members

PPP MHC Triangular Lattice

Fig. 5.28 The Voronoi tessellation of the three scenarios having equal nodes density (equal
cluster heads density and equal cluster members density), indicating an increase of regular-
ity from left (PPP) to right (triangular lattice).
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 253

nodes; (ii) the second scenario assumes an MHC model for the cluster heads and CMHC for
the cluster members as illustrated in Figure 5.22; (iii) in the last scenario we assume a fully
regular cluster heads distributed according to a triangular lattice, while cluster members
follow a PPP.
The received signal power at a cluster head originating from a cluster member can be
modeled as S = Phr−α , where P is the transmit signal power, h is the channel power gain
random variable having mean 1, r is the distance between a cluster member and its nearest
cluster head, and α is the path-loss exponent. The received Signal-to-Noise Ratio (SNR) is
given by SNR = S/N, where N is the noise power. Accordingly, a desired SNR = γo can
be achieved by adjusting the transmit power to counter the effect of the distance loss and
channel loss. As a result the power P can be expressed in terms of the desired SNR by


P = Nγo . (5.95)
h

Thus the exerted radio energy is given by:


ε = Tm Nγo , (5.96)
h

where Tm is the message transmission period. According to (5.96), the average transmit
radio energy by a cluster member to deliver a message to its cluster head becomes
 

Z ∞
α
E[ε] = E Tm Nγo = µE [r ] = µ rα f (r)dr, (5.97)
h 0

where h is independent of the distance r, f (r) is the contact distance PDF, and µ =
Tm Nγo E 1h is a constant, called the reference energy. Given that the density of the cluster
 

heads is λh and using (5.97), we are now able to compare the energy consumption of the
three deployment scenarios.

Scenario 1 PPP

In this scenario, cluster heads are randomly thinned from the parent PPP having a resulting
intensity of λh . Thus cluster members also constitute another PPP. The PDF of the contact
distance between a randomly placed point and the cluster heads is given by [130]:

fPPP (r) = 2πλh r exp −πλh r2 ,



(5.98)
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 254

so that the average energy in (5.97) becomes

1 h αi
εPPP = µ p α Γ 1 + , (5.99)
πλh 2

Scenario 2 MHC

In this scenario, cluster heads are thinned from the parent PPP using MHC-thinning, while
cluster members constitute the sibling CMHC process. The contact distance from a CMHC
process towards the sibling MHC process is approximated by the empirical formula (5.94).
As a result the average energy becomes
 
a α 2+α
εMHC ≈ µ δ Γ , (5.100)
b b

Scenario 3 Triangular Lattice

8
ǫ PPP
7 ǫ MHC
Normalized Average Energy [ ]

ǫ TRI
6

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
-1
Cluster Heads Density [Unit Area ]

Fig. 5.29 The average energy spent by a cluster member to contact its cluster head, compar-
ing the three geometrical scenarios with the same cluster heads density, and for a path-loss
exponent α = 3.

In this scenario, the cluster heads follow a triangular lattice, while cluster members fol-
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 255

low PPP. The triangular lattice constitute the most dense deterministic planner arrangement
[80]. It is formed when circles are closely packed together, so that the resulting centres form
equilateral triangular lattice. The Voronoi tessellation of the lattice points forms a hexagonal
tiles as indicated in Fig. 5.28 showing a comparison between the resulting Voronoi tessel-
lation of the three scenarios. The Cumulative Distribution Function (CDF) of the contact
distance from a randomly placed point to an equilateral triangular lattice is given by [219]:

2πr2 s
 √
3s2
r≤ 2
FTRI (r) = q √ , (5.101)
2 12r2 4 3r2
2πr
cos−1 s s √s ,

 √
3s2
+ s2
−3− s2 2r 2 <r< 3

where s is the triangular lattice side, representing the inter cluster heads distance. The
PDF is then deduces by differentiating with respect to r, and the resulting average energy
is calculated according to (5.97) and formulated in (5.102), where 2 F1 (., .; .; .) is the hyper
geometric function [220].

α
21−α 3− 2 sα
εTRI =µ  (5.102)
(α + 1)(α + 2)2 Γ − α2 − 1
  
α+1
 α  1 1 1−α 3
× 2 (α + 2)Γ − − 1 2 F1 , (−α − 1); ;
2 2 2 2 4
√ α+1
 
α 1
− π3 2 (α + 1)Γ − − .
2 2

The average energy consumption of the three scenarios is compared using a common
cluster head density λh = λo − λm where λo is the total nodes density and λm is the cluster
members density. The relation between the average energy and the cluster heads density
is explicitly described for the PPP in (5.99); however, for the MHC scenario we first map
the density to a choice of δ using (5.66) and then substitute into (5.100).
q For the triangular
2
lattice scenario the density is mapped to the side length s by s = √3λ . We also assume
h
a unity reference energy µ = 1 for comparison purposes. This comparison is depicted in
Figure 5.29 showing a clear advantage of regularizing cluster heads in reducing the average
energy cost of delivering a message between a cluster member towards its cluster head.
Noticed that the MHC process performs much closer to the ideal packed lattice (i.e. the
triangular lattice) than to the Poisson point process.
5. Aerial and Cellular Energy Efficiency Enhancement Using node Cooperation 256

5.4 Research Question 4 Conclusion


In this chapter, we have presented an analytical approach to determine the geometrical re-
gion where having a relaying device would be energy-efficient. Using this concept, we have
developed a tractable method to estimate the energy saving gain introduced by relay-assisted
D2D communication. The analysis is based on stochastic geometry exploiting the stochastic
properties of Poisson point process. In order to compare the communication energy-cost of
the cellular communication from one side and the D2D/assisted D2D from another side, we
have developed a normalisation technique for which the energy and the distance are nor-
malised. The results suggest that relay-assisted D2D technology can introduce substantial
energy saving up to a certain distance referred to as the saving distance threshold, where we
have quantified the possible extension of this distance when using assisted D2D communi-
cation. The asymptotic behaviour of the saving distance threshold has also been obtained as
a closed-form expression for general values of the path-loss exponent, indicating almost a
linear increasing relation with respect to the path-loss exponent. Using Monte-Carlo simu-
lations, we have showed that the extension of the saving distance is significant when using
only a quite moderate number of relays, after which, further increase of the density of re-
lays will only have a marginal effect on the performance. In addition, we provided a novel
cluster head selection approach for networks underlaying an aerial platform. This clustering
scheme helps in reducing the energy-cost of delivering the terrestrial data towards the aerial
platform. We optimised the proposed clustering scheme in order to maximise the network
energy efficiency, providing a novel analytical approach for calculating the optimal cluster
heads spacing. For better exploring the concept of the complementary Matérn hard-core
point process (CMHC), we studied its regularity by obtaining its pair correlation function.
We have also proved that the CMHC process converges to a PPP for large δ . Since the
CMHC process constitute the remaining points of an MHC-thinning, it can be applied in
several fields where the regularisation of a portion of randomly deployed nodes is required.
We have also presented an empirical formula describing the PDF of the contact distance be-
tween the CMHC process towards the regularised points in the sibling MHC process. The
idea of regularising random points has been applied in a wireless sensor network to illus-
trate the energy saving in delivering wireless messages between cluster members and their
cluster heads by using this approach.
Chapter 6

Conclusion and Future Research

6.1 Conclusion
In this thesis we explored the two complementary technologies of node cooperation and
aerial platforms, we started with our first objective of obtaining the physical layer char-
acteristics of the radio channel characterising both air-to-ground link and device-to-device
link, where we have accomplished this objective by obtaining a statistical channel model for
air-to-ground channel by characterising the path-loss and the associated shadowing based
on the ITU urban statistical parameters. In addition to that, we determined the path-loss ex-
ponent in different urban environments for device-to-device communication channel within
the millimeter-wave ISM band, concluding that there is a great opportunity in exploiting this
band for future D2D communication. We investigated node cooperation channel for vehic-
ular environment in the 5.8 GHz band by conducting field measurements for the path-loss
exponent and the shadowing standard deviation in different urban environments. Finally,
within the same objective we investigated the level of utilisation of the radio spectrum in
Melbourne metropolitan area in order to determine the possible spectral holes that might be
used opportunistically for D2D or emergency aerial networks, this utilisation was quantified
in correlation with the spatial population density.
We continued with our second objective of determining the coverage and throughput of
both cellular networks and arial platforms based on the design parameters and using analyti-
cal tools. We accomplished this objecting by deriving a novel generic expression (we called
it the coverage equation) that can analytically determine the service success probability of
cellular networks under generic channel fading conditions, this novel expression was de-
rived using stochastic geometry that captures the randomness in base stations locations, we
also presented a practical method to calculate the expected throughput of a cellular network

257
6. Conclusion 258

based on the coverage equation. We also derived a similar set of equations for determining
the coverage and throughput of an aerial platform given the ITU statistical parameters of
the underlaying urban environment. Using novel analytical approach, we presented a math-
ematical frame work for optimising the altitude of an aerial platform based on its design
parameters, a method that can greatly facilitate the efficient deployment of aerial networks.
The third object was to determine the possible enhancements in the network coverage
when utilising node cooperation and relaying technologies to assist both cellular base sta-
tions and aerial platforms to reach farther users outside the reliable coverage zones. We
have accomplished this objective by quantifying the alleviation of the possible cellular net-
work damage when D2D technology is utilised using a single hop D2D and multi-hop D2D
links, this performance-enhancement analysis was achieved using tools from stochastic ge-
ometry capturing the randomness in the location of base stations and the location D2D-
enabled users by calculating their mutual interference, we also compared the performance
of two damage propagation schemes, the first is the centralised damage scheme and the
second is the random damage scheme. We further introduced the concept cognitive fem-
tocell networks where mobile nodes can initiate local coverage to complement a damaged
or poor macro-cellular service, these femtocells utilise LTE-specific interference mitigation
techniques to operate in a self-organised manner underlaying the macro network. For ac-
complishing the third objective from the aerial networks prospective, we introduced a novel
algorithm to enable ground cognitive relay nodes to assist aerial base stations in expanding
their coverage and reaching more ground users, this is achieved by intelligently switching
the coverage beam towards the users with lower QoS level.
Finally our fourth objective was to quantify the energy-efficiency introduced by node
cooperation to both cellular networks and aerial platforms. We have accomplished this
objective through three contributions, firstly, we determined the geometrical region where
relaying is energy efficient, then we applied tools from stochastic geometry to determine the
statistical behaviour of the energy-efficiency for a randomly and homogeneously distributed
relays in a cellular network, we benchmarked the performance of three scenarios: (i) con-
ventional cellular networks, (ii) D2D supported networks and (iii) assisted-D2D supported
network, where we found that D2D is only efficient up to a certain distance of which we
quantified and analysed its asymptotic behaviour, we found that this threshold distance can
be enhanced when using assist-D2D technique. In the second contribution under this objec-
tive, we applied Matérn hard core point process to analytically quantify the energy saving of
clustered nodes underlaying an aerial platform, then we optimised the cluster heads spacing
to maximise this energy saving. In the last contribution under the energy-efficiency objec-
6. Conclusion 259

tive, we a analysed a novel point process we called it the complimentary Matérn hard core
point process (CMHC), where we determined its points interaction behaviour by obtaining
a closed form for its pair correlation function, and showed that regularising random points
into clusters using the CMHC process is quite useful for energy-efficiency applications.
Throughout this research we relied on verifying the analytic results and formulas against
computer simulations using Monte-Carlo results. We also presented practical numerical
examples to reflect the usefulness of the presented methodologies. The vast majority of the
work presented in this dissertation was published in-part or as a whole in peer-reviewed
journals, conference proceedings, book chapters, or otherwise currently undergoing a peer
review process. These publications has been highlighted and identified in each chapter.
6. Conclusion 260

6.2 Future Research Paradigms


Based on the work accomplished in this thesis we draw some future research paradigms
that can possibility extend this work for further analysing on node cooperation and aerial
networks. Starting with research question 1 on D2D and AtG channel characterisation,
a possible future work will include the physical verification of the air-to-ground model by
launching an actual aerial platform transmitter along with ground drive test setup, allowing
the measurement of the received power in various locations, under several urban conditions
and considering mobility, in addition to a comprehensive study of Air-to-Indoor penetration.
Under the same research question, future work can also focus on the experimental analyses
of the D2D link in mmWave spectrum within different urban environments, and on the ex-
perimental verification of beam switching technique that include the physical realisation of
the antenna arrays. For the spectrum survey, future work can include zoomed-in frequency
measurements on some selected bands that are more potential to be utilised for D2D and
Aerial Networks.
Under research question 2, and based on the developed mathematical frame work, fu-
ture research can include the modelling of different cellular interference mitigation schemes
aiming to enhance the spectral efficiency of cellular networks while keeping the target qual-
ity of service. On the aerial network coverage analysis, future work can include the study
of aerial platforms performance for MIMO systems, where multipath propagation in dense
urban environment might increase the channel capacity. On the optimisation of aerial plat-
forms altitude, future work can include the analysis of the random behaviours of ATG radio
channel including the large-scale variations as well as the small scale fading effect.
Under research question 3, future work might expand the analysis of the cellular dam-
age alleviation taking into consideration various radio propagation environments, account-
ing for the effects of the large scale fading processes, in addition, future work can study the
effects of the limited traffic handling capability of the relaying devices, taking the possibility
of traffic congestion in the relaying nodes and base stations, Thus, modelling more practical
network scenario.
Finally, for research question 4, an extension of the single hope assisted D2D is the
study of multi-hop relaying impact on energy saving, which is a considerable expansion
requiring tools from graph theory. In addition, more factors can be taken into account for
better accuracy in estimating the energy-efficiency, such as link outage probability and the
stochastic processes of the interference and the wireless communication channel. For the
complementary Matérn point process, a future work can can address further implementation
6. Conclusion 261

of this point process in wireless sensor networks to investigate different behaviour aspects,
including cooperation of nodes and network connectivity, in addition to the analytic inves-
tigation of the contact distance between the CMHC/MHC process pair.
References

[1] S. Chandrasekharan, A. Al-Hourani, K. Magowe, L. Reynaud, and S. Kandeepan,


“Propagation Measurements for D2D in Rural Areas,” in IEEE International Confer-
ence on Communication Workshop (ICCW), June 2015, pp. 639–645.
[2] Australian Bureau of Statistics, “Australian Population Grid,” 2011. [Online].
Available: http://www.abs.gov.au/
[3] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A Tractable Approach to Coverage and
Rate in Cellular Networks,” IEEE Transactions on Communications, vol. 59, no. 11,
pp. 3122–3134, 2011.
[4] (2013) EU-FP7 ICT IP Project ABSOLUTE. [Online]. Available: http://www.
absolute-project.eu/reports/publications
[5] S. Chandrasekharan, S. Kandeepan, R. Evans, A. Munari, R. Hermenier, M.-A. Mar-
chitti, and K. Gomez, “Clustering Approach for Aerial Base-Station Access with
Terrestrial Cooperation,” in IEEE Globecom Workshops (GC Wkshps), Dec 2013, pp.
1397–1402.
[6] Ericsson, “Ericsson Mobility Report: On the Pulse of the Networked Society,”
Report, June 2014. [Online]. Available: http://www.ericsson.com/mobility-report
[7] DOCOMO, “DOCOMO 5G White Paper, 5G Radio Access: Re-
quirements, Concept and Technologies ,” Report, July 2014. [Online].
Available: https://www.nttdocomo.co.jp/english/binary/pdf/corporate/technology/
whitepaper_5g/DOCOMO_5G_White_Paper.pdf
[8] NOKIA, “Ten key rules of 5G deployment Enabling 1 Tbit/s/km2 in 2030,”
Report, March 2015. [Online]. Available: http://networks.nokia.com/sites/default/
files/document/nokia_5g_deployment_white_paper.pdf
[9] E. Hossain, D. Niyato, and Z. Han, Dynamic Spectrum Access and Management in
Cognitive Radio Networks. Cambridge university press, 2009.
[10] Z. Qing and B. M. Sadler, “A Survey of Dynamic Spectrum Access,” IEEE Signal
Processing Magazine, vol. 24, no. 3, pp. 79–89, 2007.
[11] M. Song, C. Xin, Y. Zhao, and X. Cheng, “Dynamic spectrum access: from Cognitive
Radio to Network Radio,” IEEE Wireless Communications, vol. 19, no. 1, pp. 23–29,
2012.

262
6. Conclusion 263

[12] S. Kandeepan and A. Giorgetti, Cognitive Radio Techniques: Spectrum Sensing, In-
terference Mitigation, and Localization. Artech House, 2012.
[13] C. Jiang, Y. Chen, K. Liu, and Y. Ren, “Renewal-Theoretical Dynamic Spectrum Ac-
cess in Cognitive Radio Network with Unknown Primary Behavior,” Selected Areas
in Communications, IEEE Journal on, vol. 31, no. 3, pp. 406–416, March 2013.
[14] J. Zhang, I. Collings, C. S. Chen, L. Roullet, L. Luo, S.-W. Ho, and J. Yuan, “Evolv-
ing Small-Cell Communications towards Mobile-over-FTTx Networks,” Communi-
cations Magazine, IEEE, vol. 51, no. 12, pp. 92–101, December 2013.
[15] A. Anpalagan, M. Bennis, and R. Vannithamby, Eds., Design and Deploy-
ment of Small Cell Networks. Cambridge: Cambridge University Press, 2015.
[Online]. Available: http://www.cambridge.org/au/academic/subjects/engineering/
wireless-communications/design-and-deployment-small-cell-networks
[16] M. Bennis, S. Perlaza, P. Blasco, Z. Han, and H. Poor, “Self-Organization in Small
Cell Networks: A Reinforcement Learning Approach,” IEEE Transactions on Wire-
less Communications, vol. 12, no. 7, pp. 3202–3212, July 2013.
[17] L. Lu, G. Li, A. Swindlehurst, A. Ashikhmin, and R. Zhang, “An Overview of Mas-
sive MIMO: Benefits and Challenges,” IEEE Journal of Selected Topics in Signal
Processing, vol. 8, no. 5, pp. 742–758, Oct 2014.
[18] E. Larsson, O. Edfors, F. Tufvesson, and T. Marzetta, “Massive MIMO for next gen-
eration wireless systems,” IEEE Communications Magazine, vol. 52, no. 2, pp. 186–
195, February 2014.
[19] T. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. Wong, J. Schulz,
M. Samimi, and F. Gutierrez, “Millimeter Wave Mobile Communications for 5G
Cellular: It Will Work!” Access, IEEE, vol. 1, pp. 335–349, 2013.
[20] P. Zhouyue and K. Farooq, “A Millimeter-Wave Massive MIMO System for Next
Generation Mobile Broadband,” in ASILOMAR, 2012 Conference, 2012, pp. 693–
698.
[21] A. L. Swindlehurst, E. Ayanoglu, P. Heydari, and F. Capolino, “Millimeter-wave
Massive MIMO: The Next Wireless Revolution?” IEEE Communications Magazine,
vol. 52, no. 9, pp. 56–62, 2014.
[22] P. Phunchongharn, E. Hossain, and D. I. Kim, “Resource Allocation for Device-
to-Device Communications Underlaying LTE-Advanced Networks,” IEEE Wireless
Communications, vol. 20, no. 4, pp. 91–100, 2013.
[23] X. Lin, J. Andrews, A. Ghosh, and R. Ratasuk, “An overview of 3gpp device-to-
device proximity services,” IEEE Communications Magazine, vol. 52, no. 4, pp. 40–
48, April 2014.
[24] K. Doppler, M. Rinne, C. Wijting, C. B. Ribeiro, and K. Hugl, “Device-to-Device
Communication as an Underlay to LTE-Advanced Networks,” IEE Communications
Magazine, vol. 47, no. 12, pp. 42–49, 2009.
6. Conclusion 264

[25] 3GPP, “TR 36.843 Study on LTE Device to Device Proximity services (Release 12),”
Report, 2014.
[26] L. Wei, R. Hu, Y. Qian, and G. Wu, “Enable Device-to-Device Communications
Underlaying Cellular Networks: Challenges and Research Aspects,” IEEE Commu-
nications Magazine, vol. 52, no. 6, pp. 90–96, June 2014.
[27] X. Zhang, Z. Zheng, Q. Shen, J. Liu, X. Shen, and L.-L. Xie, “Optimizing Network
Sustainability and Efficiency in Green Cellular Networks,” IEEE Transactions on
Wireless Communications, vol. 13, no. 2, pp. 1129–1139, February 2014.
[28] H. Min, J. Lee, S. Park, and D. Hong, “Capacity Enhancement Using an Interference
Limited Area for Device-to-Device Uplink Underlaying Cellular Networks,” IEEE
Transactions on IEEE Transactions on Wireless Communications, vol. 10, no. 12, pp.
3995–4000, December 2011.
[29] M. Erturk, S. Mukherjee, H. Ishii, and H. Arslan, “Distributions of Transmit Power
and SINR in Device-to-Device Networks,” IEEE Communications Letters, vol. 17,
no. 2, pp. 273–276, February 2013.
[30] A. Asadi, Q. Wang, and V. Mancuso, “A Survey on Device-to-Device Communica-
tion in Cellular Networks,” IEEE Communications Surveys Tutorials, vol. 16, no. 4,
pp. 1801–1819, Fourthquarter 2014.
[31] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklos, and Z. Turanyi,
“Design Aspects of Network Assisted Device-to-Device Communications,” IEEE
Communications Magazine, vol. 50, no. 3, pp. 170–177, March 2012.
[32] M. Tehrani, M. Uysal, and H. Yanikomeroglu, “Device-to-device communication in
5g cellular networks: Challenges, solutions, and future directions,” Communications
Magazine, IEEE, vol. 52, no. 5, pp. 86–92, May 2014.
[33] W. Yang, L.-H. Li, W.-L. Sun, and Y. Wang, “Energy-Efficient Relay Selection and
Optimal Relay Location in Cooperative Cellular Networks With Asymmetric Traffic,”
The Journal of China Universities of Posts and Telecommunications, vol. 17, no. 6,
pp. 80–88, 2010.
[34] H. Nishiyama, M. Ito, and N. Kato, “Relay-by-Smartphone: Realizing Multihop
Device-to-Device Communications,” IEE Communications Magazine, vol. 52, no. 4,
pp. 56–65, 2014.
[35] L. Goratti, K. Gomez, R. Fedrizzi, and T. Rasheed, “A Novel Device-to-Device Com-
munication Protocol for Public Safety Applications,” in Globecom Workshops (GC
Wkshps), 2013 IEEE, Dec 2013, pp. 629–634.
[36] S. Kandeepan, K. Gomez, L. Reynaud, and T. Rasheed, “Aerial-Terrestrial Commu-
nications: Terrestrial Cooperation and Energy-Efficient Transmissions to Aerial Base
Stations,” IEEE Transactions on Aerospace and Electronic Systems, vol. 50, no. 4,
pp. 2715–2735, October 2014.
6. Conclusion 265

[37] Y. Shibata, N. Uchida, and N. Shiratori, “Analysis of and Proposal for a Disaster
Information Network from Experience of the Great East Japan Earthquake,” IEEE
Communications Magazine, vol. 52, no. 3, pp. 44–50, March 2014.
[38] K. Morrison, “Rapidly Recovering from The Catastrophic Loss of a Major Telecom-
munications Office,” IEEE Communications Magazine, vol. 49, no. 1, pp. 28–35,
January 2011.
[39] A. Ghosh, R. Ratasuk, B. Mondal, N. Mangalvedhe, and T. Thomas, “Lte-advanced:
next-generation wireless broadband technology [invited paper],” IEEE Wireless Com-
munications, vol. 17, no. 3, pp. 10–22, June 2010.
[40] N. P. S. T. Council, “Public Safety Broadband Push-to-Talk over Long Term Evolu-
tion Requirements,” NPSTG Communications Report, 2013.
[41] ETSI, “Technical requirements for Direct Mode Operation (DMO), 300 396-1
V1.2.0,” TETRA, vol. 1, pp. 396–1, 2011.
[42] M. Iwamura, H. Takahashi, and S. Nagata, “Relay technology in LTE-Advanced,”
NTT DoCoMo Technical Journal, vol. 12, no. 2, pp. 29–36, 2010.
[43] K. Vanganuru, S. Ferrante, and G. Sternberg, “System Capacity and Coverage of a
Cellular Network with D2D Mobile Relays,” in Military Communications Confer-
ence, Oct 2012, pp. 1–6.
[44] S. Karapantazis and F. Pavlidou, “Broadband communications via high-altitude plat-
forms: A survey,” IEEE Communications Surveys & Tutorials, vol. 7, no. 1, pp. 2–31,
2005.
[45] D. Grace and M. Mohorcic, Broadband Communications via High-Altitude Plat-
forms. John Wiley & Sons, 2011.
[46] B. El-Jabu and R. Steele, “Cellular Communications Using Aerial Platforms,” IEEE
Transactions on Vehicular Technology, vol. 50, no. 3, pp. 686–700, May 2001.
[47] K. Gomez, K. Sithamparanathan, M. M. Vidal, and T. Javornik, “Aerial Base Stations
with Opportunistic Links for Next Generation Emergency Communications,” IEEE
Communications Magazine, 2016.
[48] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Unmanned Aerial Vehicle with
Underlaid Device-to-Device Communications: Performance and Tradeoffs,” IEEE
Transactions on Wireless Communications, vol. PP, no. 99, pp. 1–1, 2016.
[49] ——, “Drone Small Cells in the Clouds: Design, Deployment and Performance Anal-
ysis,” in 2015 IEEE Global Communications Conference (GLOBECOM), Dec 2015,
pp. 1–6.
[50] V. Sharma, M. Bennis, and R. Nagdev, “UAV-Assisted Heterogeneous Networks for
Capacity Enhancement,” IEEE Communications Letters, vol. PP, no. 99, pp. 1–1,
2016.
6. Conclusion 266

[51] K. Gomez, L. Goratti, T. Rasheed, and L. Reynaud, “Enabling Disaster-Resilient


4G Mobile Communication Networks,” IEEE Communications Magazine, vol. 52,
no. 12, pp. 66–73, December 2014.
[52] I. Bucaille, S. Hethuin, T. Rasheed, A. Munari, R. Hermenier, and S. Allsopp,
“Rapidly Deployable Network for Tactical Applications: Aerial Base Station with
Opportunistic Links for Unattended and Temporary Events ABSOLUTE Example,”
in IEEE Military Communications Conference, MILCOM 2013, Nov 2013, pp. 1116–
1120.
[53] FCC, “The Role of Deployable Aerial Communications Architecture in Emergency
Communications and Recommended Next Steps,” White Paper, 2009. [Online].
Available: http://hraunfoss.fcc.gov/edocs_public/attachmatch/DOC-309742A1.pdf
[54] A. Lucent, “How-to Guide: for LTE in Public Safety,” Government Technology Mag-
azine, 2010.
[55] (2015) Google Loon Project. [Online]. Available: http://www.google.com/loon/
[56] G. De la Roche, A. Alayón-Glazunov, and B. Allen, LTE-advanced and next genera-
tion wireless networks: Channel modelling and propagation. John Wiley & Sons,
2012.
[57] ITU-R, “Guidelines for Evaluation of Radio Interface Technologies for IMT-
Advanced,” International Telecommunication Union (ITU), M Series, Radiowave
propagation ITU-R M.2135-1, 2009.
[58] W. I. Project, “WINNER II Channel Models,” European Commission, Deliverable
IST-WINNER D, vol. 1, 2007.
[59] S. Geng, J. Kivinen, X. Zhao, and P. Vainikainen, “Millimeter-Wave Propagation
Channel Characterization for Short-Range Wireless Communications,” IEEE Trans-
actions on Vehicular Technology, vol. 58, no. 1, pp. 3–13, Jan 2009.
[60] P. Zhouyue and F. Khan, “An Introduction to Millimeter-Wave Mobile Broadband
Dystems,” Communications Magazine, IEEE, vol. 49, no. 6, pp. 101–107, 2011.
[61] ITU-R, “Rec. ITU-R P.676-9 Attenuation by Atmospheric Gases,” P Series, Ra-
diowave propagation, 2013.
[62] International Telecommunication Union-Radio, “Guidelines for Evaluation of Radio
Interface Technologies for IMT-Advanced,” International Telecommunication Union
(ITU), M Series, Radiowave Propagation ITU-R M.2135-1, 2009.
[63] Q. Feng, J. McGeehan, E. Tameh, and A. Nix, “Path Loss Models for Air-to-Ground
Radio Channels in Urban Environments,” in Vehicular Technology Conference. VTC
2006-Spring. IEEE 63rd, vol. 6, May 2006, pp. 2901–2905.
[64] M. Simunek, F. Perez Fontain, and P. Pechac, “The UAV Low Elevation Propaga-
tion Channel in Urban Areas: Statistical Analysis and Time-Series Generator,” IEEE
Transactions on Antennas and Propagation, vol. 61, no. 7, pp. 3850–3858, July 2013.
6. Conclusion 267

[65] P. Chini, G. Giambene, and S. Kota, “A Survey on Mobile Satellite Systems,” Inter-
national Journal of Satellite Communications, p. 28–57, 2010.
[66] A. Jamalipour, Low Earth Orbital Satellites for Personal Communication Networks.
Artech House, Inc„ 1997.
[67] C. Loo, “A statistical model for a land mobile satellite link,” IEEE Transactions on
Vehicular Technology, vol. 34, no. 3, pp. 122–127, Aug 1985.
[68] Y. Matsumoto, R. Suzuki, K. Kondo, and M. Khan, “Land mobile satellite propa-
gation experiments in Kyoto City,” IEEE Transactions on Aerospace and Electronic
Systems, vol. 28, no. 3, pp. 718–727, Jul 1992.
[69] K. D’Souza, M. Khan, and Y. Matsumoto, “Further results on LMSS experiment in
Kyoto City,” IEEE Transactions on Aerospace and Electronic Systems, vol. 29, no. 4,
pp. 1237–1243, Oct 1993.
[70] J. Holis and P. Pechac, “Elevation Dependent Shadowing Model for Mobile Com-
munications via High Altitude Platforms in Built-Up Areas,” IEEE Transactions on
Antennas and Propagation, vol. 56, no. 4, pp. 1078–1084, 2008.
[71] R. Kouyoumjian and P. Pathak, “A Uniform Geometrical Theory of Diffraction for
an Edge in a Perfectly Conducting Surface,” Proceedings of the IEEE, vol. 62, no. 11,
pp. 1448–1461, Nov 1974.
[72] M. Wellens, J. Riihijarvi, and P. Mahonen, “Empirical Time and Frequency Domain
Models of Spectrum Use,” Physical Communication, vol. 2, no. 1, pp. 10–32, 2009.
[73] M. H. Islam, C. L. Koh, O. Ser Wah, Q. Xianming, Y. Y. Lai, W. Cavin, L. Ying-
Chang, B. E. Toh, F. Chin, G. L. Tan, and W. Toh, “Spectrum Survey in Singapore:
Occupancy Measurements and Analyses,” in 3rd International Conference on Cog-
nitive Radio Oriented Wireless Networks and Communications., 2008, Conference
Proceedings, pp. 1–7.
[74] M. Lopez-Benitez, A. Umbert, and F. Casadevall, “Evaluation of Spectrum Occu-
pancy in Spain for Cognitive Radio Applications,” in IEEE 69th Vehicular Technol-
ogy Conference. IEEE, 2009, pp. 1–5.
[75] R. I. Chiang, G. B. Rowe, and K. W. Sowerby, “A Quantitative Analysis of Spectral
Occupancy Measurements for Cognitive Radio,” in IEEE 65th Vehicular Technology
Conference. IEEE, 2007, pp. 3016–3020.
[76] Y. Liang, S.-x. YIN, W. Shuai, E.-q. ZHANG, W.-j. HONG, and S.-f. LI, “Quantita-
tive Spectrum Occupancy Evaluation in China: Based on a Large Scale Concurrent
Spectrum Measurement,” The Journal of China Universities of Posts and Telecom-
munications, vol. 19, no. 3, pp. 122–128, 2012.
[77] F. Baccelli and B. Błaszczyszyn, Stochastic Geometry and Wireless Networks: Vol-
ume I Theory, ser. Foundations and Trends in Networking, 2010, vol. 3.
[78] ——, Stochastic Geometry and Wireless Networks: Volume II Applications, ser.
Foundations and Trends in Networking, 2010, vol. 4.
6. Conclusion 268

[79] A. Baddeley, I. Bárány, R. Schneider, and W. Weil, Stochastic Geometry. Springer,


2007.
[80] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge University
Press, 2012.
[81] ——, “Mean Interference in Hard-Core Wireless Networks,” IEEE Communications
Letters, vol. 15, no. 8, pp. 792–794, August 2011.
[82] V. Chandrasekhar and J. G. Andrews, “Uplink Capacity and Interference Avoidance
for Two-Tier Femtocell Networks,” IEEE Transactions on Wireless Communications,
vol. 8, no. 7, pp. 3498–3509, 2009.
[83] H. Nguyen, F. Baccelli, and D. Kofman, “A Stochastic Geometry Analysis of Dense
IEEE 802.11 Networks,” in IEEE INFOCOM 2007. 26th IEEE International Confer-
ence on Computer Communications., May 2007, pp. 1199–1207.
[84] R. K. Ganti and M. Haenggi, “Regularity in Sensor Networks,” in International
Zurich Seminar on Communications. IEEE, 2006, Conference Proceedings, pp.
186–189.
[85] H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, “Modeling and Analysis
of K-Tier Downlink Heterogeneous Cellular Networks,” IEEE Journal on Selected
Areas in Communications, vol. 30, no. 3, pp. 550–560, 2012.
[86] X. Zhang and M. Haenggi, “A Stochastic Geometry Analysis of Inter-cell Interfer-
ence Coordination and Intra-cell Diversity,” IEEE Transactions on Wireless Commu-
nications, vol. PP, no. 99, pp. 1–1, 2014.
[87] K. Gulati, B. L. Evans, J. G. Andrews, and K. R. Tinsley, “Statistics of Co-channel
Interference in a Field of Poisson and Poisson-Poisson Clustered Interferers,” IEEE
Transactions on Signal Processing, vol. 58, no. 12, pp. 6207–6222, 2010.
[88] N. Miyoshi and T. Shirai, “A Cellular Network Model with Ginibre Configurated
Base Stations,” Advances in Applied Probability, vol. 46, no. 3, pp. 832–845, 2014.
[89] I. Nakata and N. Miyoshi, “Spatial Stochastic Models for Analysis of Heterogeneous
Cellular Networks with Repulsively Deployed Base Stations,” Performance Evalua-
tion, vol. 78, pp. 7–17, 2014.
[90] M. Di Renzo, A. Guidotti, and G. E. Corazza, “Average Rate of Downlink Heteroge-
neous Cellular Networks over Generalized Fading Channels: A Stochastic Geometry
Approach,” IEEE Transactions on Communications, vol. 61, no. 7, pp. 3050–3071,
2013.
[91] K. M. Gomez, A. Al-Hourani, L. Goratti, R. Riggio, S. Kandeepan, and I. Bucaille,
“Capacity Evaluation of Aerial LTE Base-Stations for Public Safety Communica-
tions,” in European Conference on Networks and Communications, Jun 2015, pp.
138–143.
6. Conclusion 269

[92] K. Gomez, T. Rasheed, L. Reynaud, and S. Kandeepan, “On The Performance of


Aerial LTE Base-Stations for Public Safety and Emergency Recovery,” in IEEE
Globecom Workshops, Dec 2013, pp. 1391–1396.
[93] E. Ogawa and A. Satoh, “Propagation Path Visibility Estimation for Radio Local
Distribution Systems in Built-Up Areas,” IEEE Transactions on Communications,
vol. 34, no. 7, pp. 721–724, 1986.
[94] ITU-R, “Rec. P.1410-2 Propagation Data and Prediction Methods for The Design
of Terrestrial Broadband Millimetric Aadio Access Systems,” P Series, Radiowave
propagation, 2003.
[95] E. Lutz, D. Cygan, M. Dippold, F. Dolainsky, and W. Papke, “The Land Mobile
Satellite Communication Channel-Recording, Statistics, and Channel Model,” IEEE
Transactions on Vehicular Technology, vol. 40, no. 2, pp. 375–386, 1991.
[96] N. Saquib, E. Hossain, and K. Dong In, “Fractional Frequency Reuse for Interference
Management in LTE-Advanced HetNets,” IEEE Wireless Communications, vol. 20,
no. 2, pp. 113–122, 2013.
[97] D. Lopez-Perez, I. Guvenc, G. de la Roche, M. Kountouris, T. Quek, and J. Zhang,
“Enhanced Intercell Interference Coordination Challenges in Heterogeneous Net-
works,” IEEE Wireless Communications, vol. 18, no. 3, pp. 22–30, 2011.
[98] ——, “Enhanced Intercell Interference Coordination Challenges in Heterogeneous
Networks,” IEEE Wireless Communications, vol. 18, no. 3, pp. 22–30, June 2011.
[99] E. Dahlman, P. S., and S. J., LTE/LTE-Advanced for Mobile Broadband. Academic
Press, 2013.
[100] N. Lee, X. Lin, J. Andrews, and R. Heath, “Power Control for D2D Underlaid Cellu-
lar Networks: Modeling, Algorithms, and Analysis,” IEEE Journal on Selected Areas
in Communications, vol. 33, no. 1, pp. 1–13, Jan 2015.
[101] L. Zhesheng, L. Yuancao, W. Si, G. Yuehong, Z. Xin, and Y. Dacheng, “Stochastic
Geometry Analysis of Achievable Transmission Capacity for Relay-assisted Device-
to-Device Networks,” in IEEE International Conference on Communications, 2014,
Conference Proceedings, pp. 2251–2256.
[102] M. Haenggi, J. Andrews, F. Baccelli, O. Dousse, and M. Franceschetti, “Stochastic
Geometry and Random Graphs for The Analysis and Design of Wireless Networks,”
IEEE Journal on Selected Areas in Communications, vol. 27, no. 7, pp. 1029–1046,
2009.
[103] M. Z. Win, P. C. Pinto, and L. A. Shepp, “A Mathematical Theory of Network Inter-
ference and Its Applications,” Proceedings of the IEEE, vol. 97, no. 2, pp. 205–230,
2009.
[104] A. K. Sadek, Z. Han, and K. R. Liu, “Distributed Relay-Assignment Protocols for
Coverage Expansion in Cooperative Wireless Networks,” IEEE Transactions on Mo-
bile Computing, vol. 9, no. 4, pp. 505–515, 2010.
6. Conclusion 270

[105] X. Lei, T. E. Fuja, and D. J. Costello, “Mobile Relaying: Coverage Extension and
Throughput Enhancement,” IEEE Transactions on Communication, vol. 58, no. 9,
pp. 2709–2717, 2010.
[106] W.-H. Sheen, S.-J. Lin, and C.-C. Huang, “Downlink Optimization and Performance
of Relay-Assisted Cellular Networks in Multicell Environments,” IEEE Transactions
on Vehicular Technology, vol. 59, no. 5, pp. 2529–2542, 2010.
[107] R. K. Ganti and M. Haenggi, “Spatial Analysis of Opportunistic Downlink Relaying
in a Two-Hop Cellular System,” IEEE Transactions on Communications, vol. 60,
no. 5, pp. 1443–1450, 2012.
[108] H. Mustafa, M. Shakir, M. Imran, A. Imran, and R. Tafazolli, “Coverage Gain and
Device-to-Device User Density: Stochastic Geometry Modeling and Analysis,” IEEE
Communications Letters, vol. 19, no. 10, pp. 1742–1745, Oct 2015.
[109] D. Wei, Y. Jin, S. Vural, K. Moessner, and R. Tafazolli, “An Energy-Efficient Clus-
tering Solution for Wireless Sensor Networks,” IEEE Transactions on Wireless Com-
munications, vol. 10, no. 11, pp. 3973–3983, 2011.
[110] O. Younis and S. Fahmy, “HEED: a Hybrid, Energy-Efficient, Distributed Clustering
Approach for Ad Hoc Sensor Networks,” IEEE Transactions on Mobile Computing,
vol. 3, no. 4, pp. 366–379, 2004.
[111] S. Kandeepan, S. K. Jayaweera, and R. Fedrizzi, “Power-Trading in Wireless Com-
munications: A Cooperative Networking Business Model,” IEEE Transactions on
Wireless Communications, vol. 11, no. 5, pp. 1872–1880, 2012.
[112] B. Seung Jun and G. de Veciana, “Spatial Energy Balancing Through Proactive Multi-
path Routing in Wireless Multihop Networks,” Networking, IEEE/ACM Transactions
on, vol. 15, no. 1, pp. 93–104, 2007.
[113] J.-H. Chang and L. Tassiulas, “Energy Conserving Routing in Wireless ad-hoc Net-
works,” in IEEE Nineteenth Annual Joint Conference of the IEEE Computer and
Communications Societies, vol. 1, 2000, pp. 22–31 vol.1.
[114] Y. Chia-Hao, O. Tirkkonen, K. Doppler, and C. Ribeiro, “Power Optimization of
Device-to-Device Communication Underlaying Cellular Communication,” in IEEE
International Conference on Communications, 2009, pp. 1–5.
[115] V. Rodoplu and T. H. Meng, “Minimum Energy Mobile Wireless Networks,” IEEE
Journal on Selected Areas in Communications, vol. 17, no. 8, pp. 1333–1344, 1999.
[116] L. Simic, S. M. Berber, and K. W. Sowerby, “Partner Choice and Power Allocation for
Energy Efficient Cooperation in Wireless Sensor Networks,” in IEEE International
Conference on Communications., 2008, pp. 4255–4260.
[117] Z. Lin, E. Erkip, and A. Stefanov, “Cooperative Regions and Partner Shoice in Coded
Cooperative Systems,” IEEE Transactions on Communications, vol. 54, no. 7, pp.
1323–1334, 2006.
6. Conclusion 271

[118] C. K. Lo, S. Vishwanath, and R. W. Heath, “Relay Subset Selection in Wireless


Networks Using Partial Decode-and-Forward Transmission,” IEEE Transactions on
Vehicular Technology, vol. 58, no. 2, pp. 692–704, 2009.
[119] S. Cho, W. Choi, and K. Huang, “QoS Provisioning Relay Selection in Random
Relay Networks,” IEEE Transactions on Vehicular Technology, vol. 60, no. 6, pp.
2680–2689, 2011.
[120] L. M. Feeney, D. Hollos, M. Kubisch, S. Mengesha, and H. Karl, “A Geometric
Derivation of The Probability of Finding a Relay in Multi-Rate Networks,” in Net-
working. Springer, 2004, pp. 1312–1317.
[121] M. Zorzi and R. Rao, “Geographic Random Forwarding (GeRaF) for Ad Hoc and
Sensor Networks: Multihop Performance,” IEEE Transactions on Mobile Comput-
ing, vol. 2, no. 4, pp. 337–348, 2003.
[122] S. Kandeepan, T. Rasheed, and S. Reisenfeld, “Energy Efficient Cooperative
HAP-Terrestrial Communication Systems,” in Springer, Personal Satellite Services.
Springer, 2011, pp. 151–164.
[123] A. Al-Hourani and A. Kandeepan, Sithamparanathanand Jamalipour, “Modeling Air-
to-Ground Path Loss for Low Altitude Platforms in Urban Environments,” in IEEE
Global Communications Conference (GLOBECOM), Dec 2014, pp. 2898–2904.
[124] A. Al-Hourani, S. Chandrasekharan, and S. Kandeepan, “Path Loss Study for Mil-
limeter Wave Device-to-Device Communications in Urban Environment,” in IEEE
International Conference on Communications Workshops (ICC), June 2014, pp. 102–
107.
[125] A. Al-Hourani, S. Chandrasekharan, G. Baldini, and S. Kandeepan, “Propagation
measurements in 5.8GHz and pathloss study for CEN-DSRC,” in International Con-
ference on Connected Vehicles and Expo (ICCVE), Nov 2014, pp. 1086–1091.
[126] A. Al-Hourani, V. Trajkovic, S. Chandrasekharan, and S. Kandeepan, “Spectrum oc-
cupancy measurements for different urban environments,” in European Conference
on Networks and Communications (EuCNC), June 2015, pp. 97–102.
[127] F. Baccelli, M. Klein, M. Lebourges, and S. Zuyev, “Stochastic Geometry and Archi-
tecture of Communication Networks,” Telecommunication Systems, vol. 7, no. 1-3,
pp. 209–227, 1997.
[128] A. Al-Hourani and S. Kandeepan, “On Modeling Coverage and Rate of Random
Cellular Networks under Generic Channel Fading,” Wireless Networks, pp. 1–13,
2015. [Online]. Available: http://dx.doi.org/10.1007/s11276-015-1114-x
[129] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP Altitude for Maximum
Coverage,” IEEE Wireless Communications Letters, vol. 3, no. 6, pp. 569–572, Dec
2014.
[130] A. Al-Hourani, S. Kandeepan, and A. Jamalipour, “Stochastic Geometry Study on
Device-to-Device Communication as a Disaster Relief Solution,” IEEE Transactions
on Vehicular Technology, vol. 65, no. 5, pp. 3005–3017, May 2016.
6. Conclusion 272

[131] A. Al-Hourani and S. Kandeepan, “Cognitive Relay Nodes for airborne LTE emer-
gency networks,” in 7th International Conference on Signal Processing and Commu-
nication Systems (ICSPCS), Gold Coast, Australia, Dec 2013, pp. 1–9.
[132] ——, “Temporary Cognitive Femtocell Network for public safety LTE,” in IEEE 18th
International Workshop on Computer Aided Modeling and Design of Communication
Links and Networks (CAMAD), Berlin, Germany, Sept 2013, pp. 190–195.
[133] A. Al-Hourani, S. Kandeepan, and S. Arunthavanathan, Design and Deployment
of Small Cell Networks. Cambridge: Cambridge University Press, 2015, ch.
Temporary Infrastructure with Cognitive Radios for Small Cell Networks.
[Online]. Available: http://www.cambridge.org/au/academic/subjects/engineering/
wireless-communications/design-and-deployment-small-cell-networks
[134] A. Al-Hourani, S. Kandeepan, and E. Hossain, “Relay-Assisted Device-to-Device
Communication: A Stochastic Analysis of Energy Saving,” IEEE Transactions on
Mobile Computing, 2015.
[135] I. Belski, A. Al-Hourani, A. Valentine, and A. Belski, “Can Simple Ideation Tech-
niques Enhance Idea Generation?” in AAEE 2014: Engineering the Knowledge Econ-
omy: Collaboration, Engagement and Employability, 2014, pp. 1–9.
[136] C. Tang, S. Kandeepan, A. Hourani, A. Munari, and M. Berioli, “Spectrum sensing
for cognitive maritime VHF networks,” in OCEANS 2014 - TAIPEI, April 2014, pp.
1–7.
[137] ICC, “ICT Facts and Figures,” International Telecommunications Union,
2013. [Online]. Available: http://www.itu.int/en/ITU-D/Statistics/Documents/facts/
ICTFactsFigures2013.pdf
[138] Cisco Visual Networking Index, “Forecast and Methodology, 2012–2017,” White Pa-
per, 2013.
[139] L. Ippolito, “Millimeter Wave Propagation: Spectrum Management Implications,”
FCC Bulletin, vol. 70, 1997.
[140] Radio Regulations, “International Telecommunication Union Radiocommunication
Sector. ITU-R. Geneva,” 2012.
[141] Australian Media and Communication Autority, “60 GHz Band Milimeter Wave
Tchnology,” Radiofrequency Planning Group, Australian Communications Author-
ity, 2004.
[142] A. Nesic et al., “Millimeter Wave Printed Antenna Array with High Side Lobe Sup-
pression,” in IEEE APSISy 2006, 2006, pp. 3051–3054.
[143] C. Balanis and P. Ioannides, “Introduction to Smart Antennas,” Synthesis Lectures on
Antennas, vol. 2, no. 1, pp. 1–175, 2007.
[144] R. Elliott, “Beamwidth and Directivity of Large Scanning Arrays,” Microwave Jour-
nal, vol. 7, pp. 74–82, 1964.
6. Conclusion 273

[145] 3GPP, “Parameters for FDD HeNB RF Requirements,” 3GPP TSG-RAN WG4 R4-
092042, 2009.
[146] P. Mogensen et al., “LTE Capacity Compared to The Shannon Bound,” in IEEE 65th
VTC2007-Spring., 2007, pp. 1234–1238.
[147] J. Kenney, “Dedicated Short-Range Communications (DSRC) Standards in the
United States,” Proceedings of the IEEE, vol. 99, no. 7, pp. 1162–1182, July 2011.
[148] C. F. Mecklenbrauker, A. F. Molisch, J. Karedal, F. Tufvesson, A. Paier, L. Bernado,
T. Zemen, O. Klemp, and N. Czink, “Vehicular Channel Characterization and Its Im-
plications for Wireless System Design and Performance,” Proceedings of the IEEE,
vol. 99, no. 7, pp. 1189–1212, 2011.
[149] F. Bai, T. Elbatt, G. Hollan, H. Krishnan, and V. Sadekar, “Towards characterizing
and classifying communication-based automotive applications from a wireless net-
working perspective,” in Proceedings of IEEE Workshop on Automotive Networking
and Applications (AutoNet), 2006, pp. 1–25.
[150] K. Yun Hee, S. Iickho, K. Hong Gil, C. Taejoo, and K. Hyung-Myung, “Performance
Analysis of a Coded OFDM System in Time-Varying Multipath Rayleigh Fading
Channels,” IEEE Transactions on Vehicular Technology, vol. 48, no. 5, pp. 1610–
1615, 1999.
[151] A. Saakian, Radio Wave Propagation Fundamentals. Artech House, 2011.
[152] M. Russell and G. L. Stuber, “Interchannel Interference Analysis of OFDM in a Mo-
bile Environment,” in Vehicular Technology Conference, 1995 IEEE 45th, vol. 2,
1995, Conference Proceedings, pp. 820–824 vol.2.
[153] T. Rappaport, Wireless Communications: Principles and Practice. Prentice Hall
PTR New Jersey, 1996, vol. 2.
[154] K. Gomez, T. Rasheed, L. Reynaud, and S. Kandeepan, “On The Performance of
Aerial LTE Base-Stations for Public Safety and Emergency Recovery,” in IEEE
Globecom Workshops (GC Wkshps), Dec 2013, pp. 1391–1396.
[155] S. Kandeepan, K. Gomez, T. Rasheed, and L. Reynaud, “Energy Efficient Coop-
erative Strategies in Hybrid Aerial-Terrestrial Networks for Emergencies,” in IEEE
22nd International Symposium on Personal Indoor and Mobile Radio Communica-
tions (PIMRC), Sept 2011, pp. 294–299.
[156] K. Gomez, S. Kandeepan, L. Reynaud, and T. Rasheed, “Adaptive Energy Efficient
Communications for Rapidly Deployable Aerial-Terrestrial Networks,” in IEEE In-
ternational Conference on Communications Workshops (ICC), June 2013, pp. 452–
457.
[157] L. Reynaud, T. Rasheed, and S. Kandeepan, “An Integrated Aerial Telecommunica-
tions Network that Supports Emergency Traffic,” in 14th International Symposium
on Wireless Personal Multimedia Communications (WPMC), Oct 2011, pp. 1–5.
6. Conclusion 274

[158] Q. Feng, E. Tameh, A. Nix, and J. McGeehan, “Modelling the Likelihood of Line-
of-Sight for Air-to-Ground Radio Propagation in Urban Environments,” in IEEE
GLOBECOM ’06, 2006.
[159] M. Weiner, “Use of the Longley-Rice and Johnson-Gierhart Tropospheric Radio
Propagation Programs:0.02-20 GHz,” IEEE Journal on Selected Areas in Commu-
nications, vol. 4, no. 2, pp. 297–307, 1986.
[160] S. Rohde and C. Wietfeld, “Interference Aware Positioning of Aerial Relays for
Cell Overload and Outage Compensation,” in IEEE Vehicular Technology Confer-
ence (VTC Fall), 2012, pp. 1–5.
[161] A. Valcarce, T. Rasheed, K. Gomez, S. Kandeepan, L. Reynaud, R. Hermenier,
A. Munari, M. Mohorcic, M. Smolnikar, and I. Bucaille, “Airborne Base Stations
for Emergency and Temporary Events,” in Personal Satellite Services. Springer,
2013, pp. 13–25.
[162] T. Tozer and D. Grace, “High-altitude platforms for wireless communications,” Elec-
tronics & Communication Engineering Journal, vol. 13, no. 3, pp. 127–137, 2001.
[163] G. Stuber, Principles of Mobil Communication, Chapter 2. Springer, 2011.
[164] V. Erceg, et al, “Comparisons of a Computer-Based Propagation Prediction Tool with
Experimental Data Collected in Urban Microcellular Environments,” IEEE Journal
on Selected Areas in Communications, vol. 15, no. 4, pp. 677–684, 1997.
[165] K. Sithamparanathan and A. Giorgetti, Cognitive Radio Techniques: Spectrum Sens-
ing, Interference Mitigation, and Localization. Artech House, 2012.
[166] S. Filin, H. Harada, H. Murakami, and K. Ishizu, “International Standardization of
Cognitive Radio Systems,” Communications Magazine, IEEE, vol. 49, no. 3, pp. 82–
89, 2011.
[167] J. Andrews, S. Buzzi, C. Wan, S. V. Hanly, A. Lozano, A. C. K. Soong, and J. C.
Zhang, “What Will 5G Be?” IEEE Journal on Selected Areas in Communications,
vol. 32, no. 6, pp. 1065–1082, 2014.
[168] S. Pagadarai, A. M. Wyglinski, and R. Vuyyuru, “Characterization of Vacant UHF TV
Channels for Vehicular Dynamic Spectrum Access,” in IEEE Vehicular Networking
Conference (VNC). IEEE, 2009, pp. 1–8.
[169] S. Chen, A. M. Wyglinski, S. Pagadarai, R. Vuyyuru, and O. Altintas, “Feasibility
Analysis of Vehicular Fynamic Spectrum Access via Queueing Theory Model,” IEEE
Communications Magazine, vol. 49, no. 11, pp. 156–163, 2011.
[170] S. Kandeepan, A. Sierra, J. Campos, and I. Chlamtac, “Periodic Sensing in Cognitive
Radios for Detecting UMTS/HSDPA Based on Experimental Spectral Occupancy
Statistics,” in IEEE Wireless Communications and Networking Conference (WCNC).
IEEE, 2010, Conference Proceedings, pp. 1–6.
[171] C. A. Balanis, Antenna Theory: Analysis and Design. John Wiley & Sons, 2012.
6. Conclusion 275

[172] T. Yucek and H. Arslan, “A Survey of Spectrum Sensing Algorithms for Cognitive
Radio Applications,” IEEE Communications Surveys & Tutorials, vol. 11, no. 1, pp.
116–130, 2009.
[173] Australian Communication and Media Authority, “Australian Radiofrequency
Spectrum Plan 2013,” 2013. [Online]. Available: http://www.acma.gov.au/
[174] Australian Radiation Protection and Nuclear Safety Agency, “Base Station
Frequency Bands,” 2013. [Online]. Available: http://www.arpansa.gov.au/
[175] Australian Communications and Media Authority, “Digital Dividend
Auction Results,” 2013. [Online]. Available: http://www.acma.gov.au/
Industry/Spectrum/Digital-Dividend-700MHz-and-25Gz-Auction/Reallocation/
digital-dividend-auction-results
[176] International Telecommunication Union-Radio, “The Protection of Safety Services
from Unwanted Emissions, Recommendation ITU-R SM.1535, Question ITU-R
211/1,” International Telecommunication Union (ITU), Report, 2012.
[177] S. Kandeepan, L. De Nardis, M. Di Benedetto, A. Guidotti, and G. Corazza,
“Cognitive Satellite Terrestrial Radios,” in Global Telecommunications Conference
(GLOBECOM 2010), 2010 IEEE, Dec 2010, pp. 1–6.
[178] M. Hoyhtya, J. Kyrolainen, A. Hulkkonen, J. Ylitalo, and A. Roivainen, “Application
of Cognitive Radio Techniques to Satellite Communication,” in IEEE International
Symposium on Dynamic Spectrum Access Networks (DYSPAN), Oct 2012, pp. 540–
551.
[179] D. Stoyan, W. Kendall, and J. Mecke, Stochastic Geometry and Its Applications.
Jhon Wiley, 1987.
[180] J. F. C. Kingman, Poisson Processes. Oxford University Press, 1992, vol. 3.
[181] T. D. Novlan, R. K. Ganti, A. Ghosh, and J. G. Andrews, “Analytical Evaluation of
Fractional Frequency Reuse for OFDMA Cellular Networks,” IEEE Transactions on
Wireless Communications, vol. 10, no. 12, pp. 4294–4305, 2011.
[182] A. F. Molisch, Wireless communications. John Wiley & Sons, 2010, vol. 15.
[183] H. ElSawy, E. Hossain, and M. Haenggi, “Stochastic Geometry for Modeling, Anal-
ysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks: A Survey,”
IEEE Communications Surveys & Tutorials, vol. 15, no. 3, pp. 996–1019, 2013.
[184] E. S. Sousa and J. A. Silvester, “Optimum Transmission Ranges in a Direct-Sequence
Spread-Spectrum Multihop Packet Radio Network,” IEEE Journal on Selected Areas
in Communications, vol. 8, no. 5, pp. 762–771, 1990.
[185] J. Venkataraman, M. Haenggi, and O. Collins, “Shot Noise Models for Outage and
Throughput Analyses in Wireless Ad Hoc Networks,” in IEEE MILCOM 2006, Mili-
tary Communications Conference, 2006, Conference Proceedings, pp. 1–7.
6. Conclusion 276

[186] F. Baccelli, B. Blaszczyszyn, and P. Muhlethaler, “Stochastic Snalysis of Spatial and


Opportunistic Aloha,” IEEE Journal on Selected Areas in Communications, vol. 27,
no. 7, pp. 1105–1119, 2009.
[187] B. Blaszczyszyn, M. K. Karray, and H. P. Keeler, “Using Poisson Processes to Model
Lattice Cellular Networks,” in Proceedings IEEE INFOCOM, 2013, Conference Pro-
ceedings, pp. 773–781.
[188] J. Gil-Pelaez, “Note on The Inversion Theorem,” Biometrika, vol. 38, no. 3-4, pp.
481–482, 1951.
[189] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels.
John Wiley & Sons, 2005, vol. 95.
[190] A. Zajić, Mobile-to-mobile Wireless Channels. Artech House, 2012.
[191] C.-H. Lee, C.-Y. Shih, and Y.-S. Chen, “Stochastic Geometry Based Models for Mod-
eling Cellular Networks in Urban Areas,” Wireless networks, vol. 19, no. 6, pp. 1063–
1072, 2013.
[192] C. Shin-Ming, A. Weng-Chon, T. Fan-Min, and C. Kwang-Cheng, “Design and Anal-
ysis of Downlink Spectrum Sharing in Two-Tier Cognitive Femto Networks,” IEEE
Transactions on Vehicular Technology, vol. 61, no. 5, pp. 2194–2207, 2012.
[193] T. YungLan and H. ChingYao, “Analysis of Femto Base Station Network Deploy-
ment,” IEEE Transactions on Vehicular Technology, vol. 61, no. 2, pp. 748–757,
2012.
[194] P. Mugen, L. Yuan, T. Q. S. Quek, and W. Chonggang, “Device-to-Device Underlaid
Cellular Networks under Rician Fading Channels,” IEEE Transactions on Wireless
Communications, vol. 13, no. 8, pp. 4247–4259, 2014.
[195] K. R. Jacobson and W. A. Krzymien, “System Design and Throughput Analysis for
Multihop Relaying in Cellular Systems,” IEEE Transactions on Vehicular Technol-
ogy, vol. 58, no. 8, pp. 4514–4528, 2009.
[196] L. Daewon, G. Y. Li, and T. Suwen, “Intercell Interference Coordination for LTE
Systems,” IEEE Transactions on Vehicular Technology, vol. 62, no. 9, pp. 4408–
4420, 2013.
[197] A. Baddeley, I. Bárány, R. Schneider, and W. Weil, Stochastic Geometry, Chapter 1.
Springer, 2007.
[198] 3GPP, “TR 25.814 Physical Layer Aspects for Evolved Universal Terrestrial Radio
Access (UTRA) (Release 7),” 2009. [Online]. Available: http://www.3gpp.org/ftp/
specs/archive/25_series/25.814/
[199] G. Gasper and M. Rahman, Basic Hypergeometric Series. Cambridge university
press, 2004, vol. 96.
[200] Wolfram Research, Inc., “Mathematica,” 2012.
6. Conclusion 277

[201] 3GPP, “R1-104661, Comparison of Time-domain eICIC Solutions,” Report, 2010.


[202] ——, “ETSI TS 136 216, (E-UTRA) Physical Layer for Relaying Operation, 3GPP
TR 36.215 (Rel 10),” Report, 2011.
[203] ——, “TR 36.826, "(E-UTRA) Relay radio transmission and reception" (Rel 11),”
Report, 2012.
[204] Y. Yuan, LTE-Advanced Relay Technology and Standardization, Chapter1. Springer,
2013.
[205] 3GPP, “TR1-091566 Relay to UE channel model for LTE-Advanced,” Report, 2009.
[206] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklós, and Z. Turányi,
“Design Aspects of Network Assisted Device-to-Device Communications,” IEEE
Communications Magazine, vol. 50, no. 3, pp. 170–177, 2012.
[207] R. Madan, N. B. Mehta, A. F. Molisch, and Z. Jin, “Energy-Efficient Cooperative
Relaying over Fading Channels with Simple Relay Selection,” IEEE Transactions on
Wireless Communications, vol. 7, no. 8, pp. 3013–3025, 2008.
[208] M. Lauridsen, P. Mogensen, and L. Noel, “Empirical LTE Smartphone Power Model
with DRX Operation for System Level Simulations,” in IEEE 78th Vehicular Tech-
nology Conference, 2013, Conference Proceedings, pp. 1–6.
[209] X. Yiping, C. N. Mathur, M. A. Haleem, R. Chandramouli, and K. P. Subbalakshmi,
“Dynamic Spectrum Access with QoS and Interference Temperature Constraints,”
IEEE Transactions on Mobile Computing, vol. 6, no. 4, pp. 423–433, 2007.
[210] D. Moltchanov, “Distance Distributions in Random Networks,” Ad Hoc Networks,
vol. 10, no. 6, pp. 1146–1166, 2012.
[211] R. M. Corless, G. H. Gonnet, D. E. Hare, D. J. Jeffrey, and D. E. Knuth, “On the
LambertW Function,” Advances in Computational Mathematics, vol. 5, no. 1, pp.
329–359, 1996.
[212] M. Lotfinezhad, B. Liang, and E. S. Sousa, “Adaptive Cluster-Based Data Collection
in Sensor Networks with Direct Sink Access,” IEEE Transactions on Mobile Com-
puting, vol. 7, no. 7, pp. 884–897, 2008.
[213] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “An Application-
Specific Protocol Architecture for Wireless Microsensor Networks,” IEEE Transac-
tions on Wireless Communications, vol. 1, no. 4, pp. 660–670, 2002.
[214] K. Magowe and S. Kandeepan, “Cooperative Blind Localization of Primary User in a
Cognitive Radio Environment,” in 8th International Conference on Signal Processing
and Communication Systems (ICSPCS), Dec 2014, pp. 1–8.
[215] V. Mhatre and C. Rosenberg, “Design Guidelines for Wireless Sensor Networks:
Communication, Clustering and Aggregation,” Ad Hoc Networks, vol. 2, no. 1, pp.
45–63, 2004.
6. Conclusion 278

[216] B. Matérn, “Spatial Variation,” Meddelanden Fran Statens Skogsforskningsinstitut,


vol. 49, no. 5, 1960.
[217] T. Mikosch, Non-life Insurance Mathematics, Chapter 9: Weak Convergence of Point
Processes. Springer Science & Business Media, 2009.
[218] D. Stoyan, “On Estimators of The Nearest Neighbour Distance Distribution Function
for Stationary Point Processes,” Springer Metrika, vol. 64, no. 2, pp. 139–150, 2006.
[219] P. Holgate, “The Distance From a Random Point to The Nearest Point of a Closely
Packed Lattice,” Biometrika, pp. 261–263, 1965.
[220] F. W. Olver, NIST Handbook of Mathematical Functions. Cambridge University
Press, 2010.

You might also like