srsLTE: An Open-Source Platform for LTE

Evolution and Experimentation

Ismael Gomez-Miguelez, Andres Garcia-Saavedra, Paul D. Sutton,
Pablo Serrano, Cristina Cano, Douglas J. Leith

Abstract—Testbeds are essential for experimental evaluation as the number of turbo decoder iterations or the power head-
arXiv:1602.04629v1 [cs.NI] 15 Feb 2016

well as for product development. In the context of LTE networks, room. Such modifications, customizations or instrumentation
existing testbed platforms are limited either in functionality in COTS hardware are extremely difficult, if not impossible,
and/or extensibility or are too complex to modify and customise.
In this work we present srsLTE, an open-source platform for or prohibitively expensive.
LTE experimentation designed for maximum modularity and Software-Defined Radio (SDR) is a popular concept for
code reuse and fully compliant with LTE Release 8. We show implementing radio equipment in software, using low-cost
the potential of the srsLTE library by extending the baseline general purpose computers and radio frontends. In recent
code to allow LTE transmissions in the unlicensed bands and years, it’s gaining popularity as a tool to build close-to-reality
coexistence with WiFi. We also expand previous results on this
emerging research area by showing how different vendor-specific testbeds for experimental research. If the researcher has access
mechanisms in WiFi cards might affect coexistence. to the SDR application code, as it is the case with open source,
the testbed can be easily modified and instrumenting the stack
I. I NTRODUCTION becomes as simple as pulling out the required metrics from
the code. Whilst in terms of performance or capabilities, open
Testbeds are today an essential platform for experimental source SDR testbeds typically stand behind their commercial
research and prototype development. They enable researchers counterparts, this flexibility and openness can be much more
to test, validate and assess the performance of new tech- valuable for many research problems.
nologies for wireless networks. In the context of LTE, a The most popular open source LTE SDR software available
testbed typically includes one or several User Equipments for testbeds today are Eurecom’s OpenAirInterface (OAI) [1]
(UEs), one base station (eNodeB) and an Evolved Packet Core and openLTE [2]. OAI currently provides a standard-compliant
(EPC), although the latter may be minimal. Each of these implementation of a subset of Release 10 LTE for UE, eNB,
components is provided up front by commercial off-the-shelf MME, HSS, SGw and PGw on standard Linux-based comput-
(COTS) hardware solutions. Though usually expensive, the ing equipment (Intel x86 PC architectures). The software can
performance is excellent and the functionality is extensively be used in conjunction with standard RF laboratory equipment
validated. available in many labs (i.e. National Instruments/Ettus USRP
Some research problems, however, require adding new and PXIe platforms). openLTE runs with the Ettus Research
features to the standard or modifying some of its parts. For B2x0 USRP and provides eNB, MME and HSS functionalities.
instance, in the context of 5G research, several groups are openLTE code is well organized, documented and easy to
exploring how new waveforms such as GFDM can fit in customize or modify. However, it is incomplete and many
the current LTE resource grid and physical layer procedures. features are still unstable or under development. Furthermore,
Another example is IoT, where tight power and budget con- it does not provide an UE, limiting the testbed capabilities
straints may require simplified waveforms and protocols to be in terms of instrumentation and measurement. OAI on the
introduced in the standard and the performance assessed in a other hand is comparatively very complete and provides very
real scenario. Other research problems require instrumentation, good performance. However, the code structure is complex
often of the entire network stack (PHY to IP) to measure and difficult for a user external to the project to modify or
the impact these metrics might have on the user applica- customize.
tion. Such metrics include for example, the channel Doppler In this work, we present an open source LTE library
spread, the interference, the number of HARQ retransmissions, (srsLTE) and a complete software radio LTE UE (srsUE) [3].
We describe the architecture, evaluate the computational effi-
I. Gomez-Miguelez and P. D. Sutton are with Software Radio Systems ciency and discuss the suitability to research on future LTE
Limited, Ireland. enhancements. As a case study we also present a modification
A. Garcia-Saavedra is with NEC Labs Europe, Germany.
P. Serrano is with Dept. of Telematics Engineering, Universidad Carlos III to srsLTE that implements a duty cycle-based access mecha-
de Madrid, Spain. nism for unlicensed LTE [4] and evaluate the results obtained
C. Cano and D. J. Leith are with School of Computer Science and Statistics, in a coexistence scenario with WiFi. Given the attention that
Trinity College Dublin, Ireland.
This work was supported by Science Foundation Ireland grants 13/IF/I2781 the use of the unlicensed spectrum by LTE is gathering,
and 11/PI/11771. with concerns being raised by the Federal Communications
Commission (FCC) [5] as well as by the WiFi alliance [6], The modular library approach allows researchers to easily
experimentation of emerging coexistence mechanisms in real customize, improve or completely replace components without
testbeds becomes relevant in order to ensure that coexistence affecting the rest of the code. Modules are organized hierar-
to WiFi is guaranteed in real deployments. Since the Medium chically in the following categories, as illustrated in Figure
Access Control (MAC) layer of LTE needs to be modified 1a:
to implement changes in the way LTE accesses the channel, • Core: The core modules are the main building blocks
the srsLTE platform is a perfect candidate for experimentation. used within the PHY layer. In this category we find the
The results presented in this work extend those reported in [7], turbo and convolutional coders and decoders, modulator
that were obtained using a PHY-only implementation of LTE- and demodulator, synchronization, channel estimation
A, by showing the impact of: i) the LTE periodically leaving and reference signal generation, OFDM and SC-FDMA
the channel empty to WiFi and ii) vendor-specific nuances of processing and so forth.
different WiFi cards, which may affect WiFi performance in • Physical Channels: There is one module for each uplink
initially unforeseen ways. and downlink channel (e.g. PDSCH, PUSCH, PDCCH,
The rest of this article is organised as follows. In Section PUCCH, etc). Each module uses the core building blocks
II we describe the srsLTE library and analyse its computa- to implement the signal processing required to convert
tional efficiency, while the UE implementation is described in bits into samples ready to send to the digital converter
Section III. Then, in Section IV, we present the LTE/WiFi co- and vice versa. Some physical channels share some
existence scenario we use for evaluation and discuss the results functionality, which is implemented in common auxiliary
obtained. Finally, we conclude with some final remarks. modules, e.g., PUSCH and PDSCH share many process-
ing functions which are defined in the module SCH.
II. SRS LTE: AN O PEN -S OURCE LTE L IBRARY FOR SDR
• UE Processes: The UE processes implement the physical
In this section we describe the srsLTE library and we channel procedures for uplink and downlink making use
analyse its computational efficiency proving its suitability of the physical channel modules.
for experimentation with future enhancements of the LTE • Example Applications: On top of the hierarchy we
standard. find a number of examples showing how to use the
A. Description library through the UE processes modules. Among others,
these examples include a PDSCH transmitter and receiver
srsLTE is an open source library for the PHY layer of LTE application and cell search examples.
Release 8. It is designed for maximum modularity and code
reuse with minimal inter-module or external dependencies. B. Computational Efficiency
The code is written in ANSI C and makes extensive use Computational efficiency is the most challenging aspect
of Single Instruction Multiple Data (SIMD) operations, when of an SDR application. As in many other systems, the LTE
available, for maximum performance. In terms of hardware, receiver is much more complex than the transmitter. Though it
the library deals with buffers of samples in system memory can be ported to embedded processors, srsLTE initially targets
thus being able to work with any RF front-end. It currently general purpose processors (GPP), which is suitable for most
provides interfaces to the Universal Hardware Driver (UHD), research problems. Since memory is very cheap in GPPs,
giving support to the Ettus USRP family of devices. The efficient code will make extensive use of look-up tables (LUT)
aim of the library is providing the tools to build LTE-based and pre-generate as many signals as possible where memory
applications such as a complete eNodeB or UE, an LTE sniffer access is more efficient than computation. For instance, all
or a network performance analyser. scrambling sequences, reference signals and some PUCCH
The current features provided by the library are: signals can be generated for each subframe and input data
• LTE Release 8 compliant in FDD configuration; combination. The CRC and encoder can use a LUT, and every
• Supported bandwidths: 1.4, 3, 5, 10 and 20 MHz; interleaver can be pre-computed.
• Transmission mode 1 (single antenna) and 2 (transmit Clock speeds for individual processing cores in GPPs are
diversity); beginning to reach their practical limits. For this reason,
• Cell search and synchronization procedure for the UE; further performance improvements in modern CPUs must
• All DL channels/signals are supported for UE and eN- exploit data and instruction parallelism. These techniques are
odeB side: PSS, SSS, PBCH, PCFICH, PHICH, PDCCH, used for the most expensive parts of the receiver, namely
PDSCH; the turbo decoder, the channel estimator and equalizer, the
• All UL channels/signals are supported for UE side: demodulator and the Viterbi decoder. In order to provide
PRACH, PUSCH, PUCCH, SRS; maximum portability, the VOLK library [8] has been used as
• Highly optimized turbo decoder available in Intel much as possible. VOLK is a library that contains kernels of
SSE4.1/AVX (+100 Mbps) and standard C (+25 Mbps); hand-written SIMD code for different architectures, including
and a general one written in C. At runtime, VOLK will select the
• MATLAB and OCTAVE MEX library generation for correct kernel for maximum performance. When not possible,
many components. compiler intrinsics are used to provide two alternative versions
 (a) Module diagram for the srsLTE library. (b) Threading architecture in srsUE. Boxes with coloured borders are threads.
Fig. 1. srsLTE library and srsUE threading architectures.

TABLE I
of some modules (a generic and a SIMD one) and the best one T OTAL PDSCH RECEIVER PROCESSING TIME AND BREAK - DOWN OF THE
is chosen at compile time. CPU UTILIZATION FOR 20 M HZ BANDWIDTH CONFIGURATION .
We have measured the processing time of the UE PDSCH
receiver for different bitrates (modulation and coding combina- Percentage of CPU
tions) and analysed the per-module CPU usage in an Intel Core 75 Mbps 30 Mbps 3.62 Mbps
Module Name
i7-3540M 3 GHz CPU. Table I shows the obtained results. 64QAM 16QAM QPSK
Turbo decoder (1 iteration) 78.14 % 64.21 % 20.89 %
The PDSCH includes all symbol and bit processing, including
OFDM receive processing 6.08 % 11.70 % 33.33 %
OFDM demodulation, channel estimation, symbol detection,
Resource Element de-mapping 4.92 % 9.31 % 25.26 %
rate recovery, resetting the soft buffer, turbo decoding and Rate recovery 4.49 % 5.64 % 8.34 %
CRC check. Since the interval at which data arrives in LTE is CRC checksum 2.92 % 2.23 % 0.72 %
1 ms, 1 processing core is able to process the entire PDSCH Soft demodulation 1.76 % 2.11 % 3.38 %
chain in real-time. In practice, more cores are needed because Equalization 0.16 % 1.84 % 4.98 %
the CPU needs to perform other tasks, such as synchronization, Others 1.53 % 2.96 % 55.12 %
uplink signal generation, physical layer procedures, etc. The Total Execution Time 954 µs 488 µs 170 µs
per-module usage results indicate that the turbo decoder is
by far the most demanding part and its relative weight in
the system increases as the rate increases. The SIMD turbo • MAC, RLC, PDCP, RRC, NAS and GW layers;
decoder implementation in srsLTE is based on the max-log- • Soft USIM supporting Milenage and XOR authentication;
MAP algorithm for SIMD introduced in [9] with a minor • Detailed log system with per-layer log levels and hex
modification to compute the horizontal maximum using the dumps;
Intel’s SSE4.1 phminposuw instruction. • MAC layer Wireshark packet capture;
III. SRS UE: A COMPLETE UE SDR I MPLEMENTATION • Command-line trace metrics;
srsUE is a software radio LTE UE covering all layers of • Detailed input configuration file; and
the network stack from PHY to IP. It is written in C++ • virtual network interface (i.e. tun/tap device) created upon
and builds upon the srsLTE library which provides the PHY network attach.
layer processing. For some security functions and RRC/NAS A configuration file is provided to set parameters such as
message parsing, it uses some functions from the openLTE the downlink carrier frequency and log or packet capture
project. Running on an Intel Core i7-4790, srsUE achieves options, making the software easy to use. The UE starts setting
more than 60 Mbps downlink with a 20 MHz bandwidth the USRP sampling rate to 1.96 MHz, in order to capture
SISO configuration, when tested against an Amarisoft LTE 100 the synchronization PBCH signals. Once synchronization and
eNodeB. Apart from the features listed above for the srsLTE MIB decoding is successful, it reconfigures the sampling rate
library, srsUE provides the following additional features: to the appropriate sampling rate for the LTE signal bandwidth.
Next, the UE attempts an attachment by sending a PRACH has 3 ms to decode the PDSCH, generate the PUSCH with
sequence and if the correct response is received, it continues the ACK/NACK information in it and send the samples to the
the connection setup procedure. Following successful network converter buffer.
attachment, a new virtual network interface is created in the Dividing the processing into uplink and downlink threads
system and the user can then establish IP sessions with the is ineffective, because the uplink processing needs to wait the
eNodeB network. downlink processing to finish. A more efficient approach is to
This instrumentation and the Wireshark capture capabilities parallelise all the processing associated with every subframe
make srsUE an ideal tool for many applications, including into a pipeline avoiding any delays (for thread synchronization
cross-layer performance analysis, education and prototype or data transfer) within the tasks to be done in the 3 ms
development. srsUE opens the insights of the LTE stack to deadline. This architecture is efficient for dual- or quad-core
the user: the real-time traces, logs and packet captures can be CPUs. If more than 4 cores are available, increasing the
used to correlate user experience in video streaming or web number of PHY threads above 3 does not add any benefit,
surfing with the signal quality or any other metric inside the because the maximum tolerable latency is 3 ms (3 pipeline
stack, for example. stages). The performance may then be improved by creating
srsUE classes are organized by layers, one class per stack multiple threads associated to each subframe and split the
layer. Each class provides a separate clean interface to any multiple codeblocks in a transport block among them, a feature
other class that make use of it, which is used for message currently under development.
passing between layers of the stack.
A set of threads are created for performance and priority IV. E XPERIMENTING WITH AN U NLICENSED LTE & W I F I
management reasons. Figure 1b illustrates the different layers COEXISTENCE SCENARIO
and the threads within them. Red dashed arrows indicate data As mentioned earlier, a potential use of the srsLTE platform
paths whereas dark arrows indicate interaction between threads is to analyse the coexistence of unlicensed LTE/WiFi. The
or classes. Each thread performs the following tasks: great advantage of using real hardware rather than simulations
• PHY DSP: These threads perform all the processing is that it allows us to evaluate the impact of the complex
associated with a subframe, including: OFDM demodula- wireless propagation effects encountered indoors (where such
tion, PDCCH search, PDSCH decoding, PUSCH/PUCCH small cells are most likely to be deployed) and also explore
encoding, uplink signal generation and transmission to issues such as capture and carrier sense which are often
the digital converter. After all processing is done, the difficult to model adequately.
thread returns to idle;
• PHY SYNC: Receives 1 subframe from the converter, A. LTE/WiFi Testbed
performs time and frequency synchronization, copies the Our interest is in a WiFi link and an LTE link operating in
aligned frame into an idle PHY DSP thread and triggers the same band. The LTE stations consist of an USRP B210
its execution; board from Ettus connected via an USB 3.0 interface to a
• MAC PROCEDURES: Manages MAC procedures, in- standard PC (Intel Core i7) running Linux Ubuntu Trusty,
cluding random access, scheduling request and up- with the uhd_driver and version 1.0.0 of srsLTE. The
link/downlink HARQ; distance from the LTE BS to the WiFi transmitter is 34 cm,
• MAC TIMER: Provides timer services to upper layers. and 35 cm to the WiFi receiver. The WiFi link is 94 cm long.
All timers in LTE have a resolution of 1 ms, thus are We use a VERT2450 Dual Band (2.4 to 2.48 GHz and 4.9
implemented with simple counters synchronised to the to 5.9 GHz) omni-directional vertical antenna with 3dBi Gain
subframe reception instead of using system clocks. and, unless otherwise noted, LTE is configured to use 100
• MAC DL READER: Reads downlink transport blocks physical resource block (PRBs), i.e., to use a 20 MHz channel
from a buffer, processes the PDU and pushes the packet bandwidth (the same as that used by WiFi).
up the stack until the GW; and The WiFi nodes are Soekris net6501-70 devices, which are
• GW UL READER: Reads from the virtual network inter- low-power single-board computers equipped with a 1.6 GHz
face and stores it into the PDCP buffer. Intel Atom E6xx series CPU, 2 Mini-PCI sockets, 2048 Mbyte
The threading architecture in the PHY is motivated by the DDR2-SDRAM and 8 Gbyte usb-based storage. These run
stringent latency requirements. In LTE, the required response Linux Ubuntu (kernel 3.13). In all of our experiments, in
time is 4 ms (4 subframes), which is imposed, in the UE, order to detect vendor-specific performance issues that might
by (i) the transmission of an ACK/NACK HARQ indication be present in the 802.11 hardware (see e.g. [10]), we repeat all
after decoding a PDSCH transport block, (ii) the transmission measurements using two different wireless NICs, namely, an
of PUSCH after the reception of an uplink grant or (iii) the Atheros AR9390-based 802.11a/b/g/n card and a Broadcom
re-transmission of PUSCH after the reception of a NACK BCM4321 card. We configure the Atheros cards to operate
HARQ indication. The worst case scenario happens by the in the 5 GHz band, which measurements using a spectrum
superposition of case (i) and (ii) or (i) and (iii): in the same analyser confirmed were free from other transmissions, and
subframe the UE decodes the PDSCH and encodes a PUSCH. configure the Broadcom cards to operate in the more populated
Considering the time needed to buffer 1 subframe, the UE 2.4 GHz band (since they do not support 5 GHz operation).
 WiFi’s MCS (Mb/s) 6 24 54 WiFi’s MCS (Mb/s) 6 24 54

LTE Tx Power=−16dBm LTE Tx Power=−1dBm LTE Tx Power=12dBm WiFi Tx Power=0dBm WiFi Tx Power=8dBm WiFi Tx Power=17dBm
1.00 1.00

0.75 0.75

Atheros (5 GHz)

Atheros (5 GHz)
WiFi’s Throughput (Relative)

WiFi’s Throughput (Relative)

0.50 0.50

0.25 0.25

0.00 0.00
1.00 1.00

Broadcomm (2.4 GHz)

Broadcomm (2.4 GHz)

0.75 0.75

0.50 0.50

0.25 0.25

0.00 0.00
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 −10 0 10 −10 0 10 −10 0 10
Duty cycle (%) LTE Tx Power (dBm)

Fig. 2. Impact of LTE duty cycle on performance. Fig. 3. WiFi throughput vs. LTE transmission power

B. Impact of duty cycle to the WiFi transmissions (i.e. the LTE silent period leaves
We start our experiments by investigating the impact of enough idle time to accommodate transmission of several WiFi
LTE transmissions on a WiFi link. To this end, we con- packets). When the 5 GHz band is used the results show
figured a WiFi station to send 1500 Bytes UDP traffic. little dispersion and the use of a lower MCS increases the
The WiFi station is constantly backlogged (always has a robustness of the WiFi transmissions. However, this effect is
packet to send) and is configured to use a transmit power minor when the LTE transmission power is very low (so the
of 17 dBm. We vary the activity on the LTE link by LTE interference with WiFi is then negligible) and also when
modifying the srsLTE code (namely, the example program the LTE transmission power is high (when all choices of WiFi
src/examples/pdsch_enodeb.c) that is executed in MCS are equally affected by the interference from LTE). When
user space. In more detail, we implement a periodic duty- the 2.4 GHz band is used the results are qualitatively similar,
cycle channel access scheme similar to the proposed CSAT although there is more dispersion, with the 54 Mbps MCS in
coexistence mechanism [11] where we fix an active interval particular being extremely sensitive to LTE interference.
during which the LTE base station transmits data to the UE To sum up, these measurements confirm that scheduling the
followed by a randomized silent period.1 The mean period LTE transmissions using a randomised duty cycle, with relative
(active plus silent) is equal to 150 ms. By changing the mean long on and off periods, has the expected qualitative impact for
duration of the silent period we vary the “duty cycle” of both of the WiFi wireless cards considered. We next analyse in
the LTE link. The LTE base station does not perform carrier more detail the impact on performance of the choice of MCS,
sensing and sends traffic at the start of the frame boundary. transmission power and band used.
LTE control plane signals are transmitted during the on periods C. Impact of transmission power
but the channel is completely freed during the silent periods.
The foregoing results confirm that adjusting the LTE duty
We vary the LTE duty cycle between 0% (no LTE trans-
cycle allows the impact of LTE interference on WiFi to be
missions) and 100% (no LTE silent periods), and measure the
controlled. We now fix the LTE duty cycle to be 50% and vary
resulting WiFi throughput for different settings of the LTE
the transmission power of both the WiFi and LTE transmitters,
transmit power and of the Modulation and Coding Scheme
with the aim of understanding the sensitivity and robustness
(MCS) used by the WiFi link. The results are shown on
of WiFi performance to these parameters. Measurements are
Figure 2, where for each configuration the average throughput
shown in Figure 3, for the two WiFi wireless cards studied
is measured over a 10 s experiment, which is repeated 5 times,
and for different choices of WiFi MCS.
and we use box-and-whisker plots to represent the measured
It can be seen that the WiFi transmission power has little, if
median, 25th and 75th percentiles, and the 1.5× interquartile any, impact on performance. This confirms that, in our small
range. Note that we plot the normalised WiFi throughput,
testbed, the signal quality of the WiFi link is good for all the
i.e., the ratio of the measured throughput to the maximum
configurations of this parameter. The measurements also show
achievable throughput (with 0% duty cycle), for ease of that, as in the previous experiments, the use of a lower MCS
comparison across different choices of MCS.
results in an increased WiFi normalised throughput, which
In all of these experiments the WiFi throughput is inversely
suggests that a large number of WiFi transmissions collide
proportional to the LTE duty cycle, as might be expected as
with LTE transmissions (i.e., the reduction in throughput
the LTE active and silent periods are relatively long compared
performance is not only caused by channel deferral), and
1 Randomising the LTE silent period alleviates potential quantisation ef- therefore the use of more robust transmission schemes favors
fects [12]. the appearance of the capture effect. In fact, for the case of the
 WiFi’s MCS (Mb/s) 6 24 54 WiFi’s MCS (Mb/s) 6 24 54

LTE Tx Power=−16dBm LTE Tx Power=−1dBm LTE Tx Power=12dBm LTE Tx Power=−16dBm LTE Tx Power=−1dBm LTE Tx Power=12dBm
1.00 1.00

0.75 0.75

Atheros (5 GHz)

Atheros (5 GHz)
WiFi’s Throughput (Relative)

WiFi’s Throughput (Relative)

0.50 0.50

0.25 0.25

0.00 0.00
1.00 1.00

Broadcomm (2.4 GHz)

Broadcomm (2.4 GHz)

0.75 0.75

0.50 0.50

0.25 0.25

0.00 0.00
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100 −20−15−10 −5 0 5 10 15 20 −20−15−10 −5 0 5 10 15 20 −20−15−10 −5 0 5 10 15 20
LTE Channel Width (PRBs) LTE central frequency offset (MHz)

Fig. 4. WiFi throughput vs. LTE bandwidth Fig. 5. WiFi throughput vs. LTE central frequency

Atheros card, there is a small but consistent improvement in between LTE and WiFi, this may result in vendor-dependent
performance as the transmit power is increased from 8 dBm performance. One potential consequence is that this may intro-
to 17 dBm when the MCS is 54 Mbps and the LTE trans- duce fairness issues in a multi-vendor environment. Therefore,
mission power is -6 dBm, which lends further support to this we argue that more experimental research, such as the enabled
observation. by our srsLTE-based platform, is required prior to the rollout
Given these results, we conclude that careful tuning of the of LTE in unlicensed bands.
transmission power could potentially be used to adjust the
E. Impact of the central frequency
share of resources between WiFi and LTE. However, given the
steepness of the variation in throughput vs. LTE transmission Finally, we explore performance when the amount of spec-
power and the only minor impact of the WiFi configuration, trum overlap between LTE and WiFI is varied. We again
this might not be a practical approach. configure LTE to use 100 PRBs (corresponding to a channel
bandwidth of 20MHz) and now vary the central frequency used
D. Impact of bandwidth used for LTE transmissions from -20 MHz to +20 MHz in steps of
We next investigate the impact of varying the number 5 MHz. Similarly to the previous experiments, we perform the
of subcarriers used by LTE (i.e., its bandwidth). To this tests for a range of LTE transmission powers and WiFi MCS.
end, we perform a sweep on the number of PRBs used by The resulting measurements are shown in Figure 5.
LTE, and meausre the resulting WiFi throughput for different On the one hand, when the LTE transmission power is
choices of the LTE transmission power and WiFi MCS. The highest (12 dBm) then the WiFi throughput performance is
measurements obtained are shown in Figure 4. similar for both wireless cards: when there is large overlap
In contrast to the previous experiments, here there are in the spectrum used by LTE and WiFi the interference is
significant differences in the WiFi throughput depending on large and only for the edge cases where the centre frequency
the WiFi card used. While when the LTE transmit power is is ± 20 MHz is it reduced (it is also worth noting that the
12 dBm the number of PRBs does not affect WiFi throughput, interference pattern is practically symmetrical). On the other
when the LTE transmit power is -16 dBm the WiFi throughput hand, when the LTE transmission power is smallest (-16 dBm),
varies considerably as the number of PRBs changes. It can also the behaviour changes depending on the WiFi card used. For
be seen that the Atheros and Broadcom WiFi NICs exhibit the the case of the Broadcom card the results are as might be
opposite behaviour: for the latter, when the MCS is 54 Mbps expected: the smaller the frequency offset the lower the WiFi
the throughput reduces as interference increases (an arguably throughput for an MCS of 54 Mbps, while for more robust
intuitive result, as LTE occupies more spectrum), while for the MCS the performance does not change significantly with the
former throughput increases with the number of PRBs used. frequency offset. For the case of the Atheros cards instead
Furthermore, this behaviour is qualitatively consistent for the of a U-shaped figure we see that, for all choices of MCS,
case of -1 dBm with the Atheros cards, while the Broadcom throughput is maximised either when the LTE transmissions
cards and an MCS of 6 Mbps now show a different pattern are placed 20 MHz away from the center frequency or when
as compared to when the LTE transmit power is -16 dBm. LTE uses the exact same frequency as WiFi transmissions. For
Our hypothesis for this behaviour is that these WiFi cards intermediate LTE transmission powers (-1 dBm), although in
make use of different (proprietary) clear channel assessment all cases WiFi throughput improves when using a more robust
mechanisms for carrier sensing. MCS, the results vary qualitatively depending on the hardware
These results indicate that while changing the number of considered: for the case of Atheros, the offset has little impact
PRBs used by LTE might be useful to manage coexistence on thoughput apart from when it is 20 MHz; in contrast,
for the Broadcom cards and an MCS of 6 Mbps there is a
remarkable drop in throughput for an offset of 15 MHz. These
measurements confirm the presence of a dependency on the
WiFi hardware used when considering LTE/WiFi coexistence.
These results further support the two conclusions in the pre-
vious experiments: firstly, an adequate tuning of the spectrum
used by LTE may improve coexistence with WiFi; secondly,
more real-life experimentation is required to better understand
the reasons for the observed behaviour and to detect vendor-
specific issues that might preclude fair coexistence.
V. C ONCLUSIONS
In this work we have presented an open-source, modular and
fully compliant with LTE Release 8 platform that allows for
LTE extension and experimentation. We have described the
architecture of the srsLTE library as well as the srsUE and
evaluated its computational efficiency proving its suitability
to current LTE testing. As a case study we have shown the
potential of the library for extension of LTE to work in
the unlicensed bands and coexist with WiFi networks. Our
results in this regard motivate further experimental validation
of emerging coexistence mechanisms as particularities of off-
the-shelf wireless cards might affect fair coexistence in ways
difficult to predict via analysis or simulations.
R EFERENCES
[1] “OpenAirInterface.org,” 2015. [Online]. Available:
http://www.openairinterface.org/
[2] “openLTE,” 2015. [Online]. Available: http://openlte.sourceforge.net/
[3] “srsLTE,” 2015. [Online]. Available: https://github.com/srsLTE/
[4] D. Flore, “Chairman Summary,” 3GPP workshop on LTE in unlicensed
spectrum, June, 2014.
[5] Federal Communications Commission, “Office of engineering
and technology and wireless telecommunications bureau
seek information on current trends in LTE-U and
LAA technology,” May 2015. [Online]. Available:
https://www.fcc.gov/document/oet-and-wtb-seek-information-trends-lte-u-and-laa-technology
[6] WiFi Alliance, “Wi-Fi Alliance statement on License-
Assisted Access (LAA),” February 2015. [Online]. Available:
http://www.wi-fi.org/news-events/newsroom/wi-fi-alliance-statement-on-license-assisted-access-laa
[7] Y. Jian, C.-F. Shih, B. Krishnaswamy, and R. Sivakumar, “Coexistence of
Wi-Fi and LAA-LTE: Experimental evaluation, analysis and insights,” in
IEEE International Conference on Communication Workshop (ICCW),
2015, 2015.
[8] T. O. T. Rondeau, N. McCarthy, “SIMD programming in GNURadio:
Maintainable and user-friendly algorithm optimization with VOLK,”
in Proc. Conference on Communications Technologies and Software
Defined Radio (SDR12). Brussels, Belgium: Wireless Innovation Forum
Europe, 2012.
[9] K. Loo, T. Alukaidey, and S. Jimaa, “High performance parallelised
3GPP turbo decoder,” in Personal Mobile Communications Conference,
2003. 5th European (Conf. Publ. No. 492), April 2003, pp. 337–342.
[10] G. Bianchi, A. Di Stefano, C. Giaconia, L. Scalia, G. Terrazzino, and
I. Tinnirello, “Experimental Assessment of the Backoff Behavior of
Commercial IEEE 802.11b Network Cards,” in INFOCOM 2007. 26th
IEEE International Conference on Computer Communications. IEEE,
May 2007, pp. 1181–1189.
[11] A. K. Sadek, T. Kadous, K. Tang, H. Lee, and M. Fan, “Extending
LTE to unlicensed band-Merit and coexistence,” in IEEE International
Conference on Communication Workshop (ICCW), 2015, 2015.
[12] C. Cano and D. J. Leith, “Unlicensed LTE/WiFi Coexistence: Is LBT
Inherently Fairer Than CSAT?” arXiv preprint arXiv:1511.06244, 2015.