Professional Documents
Culture Documents
Performance Analysis and Comparison of GPP-based SDR Systems
Performance Analysis and Comparison of GPP-based SDR Systems
Performance Analysis and Comparison of GPP-based SDR Systems
net/publication/322409448
CITATIONS READS
14 1,285
5 authors, including:
Kan Zheng
Beijing University of Posts and Telecommunications
359 PUBLICATIONS 9,564 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Kan Zheng on 03 December 2018.
Abstract—With the development of wireless communications, based on a unified platform. Software Defined Radio (SDR)
varies of wireless protocols have come into being. For quickly technology therefore emerges at the right moment which holds
verifying the new communication technology, the most effective the promise of universal and fully programmable wireless
method is through software. Simple software simulation has been communication systems. It gets rid of the traditional radio de-
unable to meet the needs of people for technical verification, thus sign ideas that are completely dependent on fixed hard ware for
Software Defined Radio (SDR) emerges at the right moment. In a special purpose, trying to implement various radio functions
this paper, we first make a brief overview of general purpose through software based on a modular universal hardware plat-
processor based SDR (GPP SDR), then review the current state form instead. Wireless communication technology has reached
of development and future trends in this area. Moreover, we in-
a new level taking advantage of software which is programma-
troduce two famous GPP SDR platforms, i.e. OpenAirInterface
(OAI) and srsLTE. OAI has been well known by many research-
ble, easy to modify and low cost. Thus SDR is recognized as
ers and widely used in scientific research. However, to the best of the third revolution following the fixed-to-move, analog-to-
my knowledge, there is no detailed introduction to srsLTE. We digital in the communications field.
set up OAI and srsLTE in laboratory and give a specific analysis The SDR platform can be constructed by combining a pro-
to the key source codes of physical layer (PHY) downlink (DL) on grammable radio frequency (RF) front-end with a baseband
the platforms. Furthermore, the PHY DL performance is evalu- processing platform via an appropriate interface. The leading
ated and compared. Our result demonstrates that OAI and example RF front-end is Universal Software Radio Peripheral
srsLTE are both feasible and stable SDR platforms, which proves
(USRP) [2] provided by Ettus Research. General purpose pro-
that GPP is powerful enough to process wideband signal like
LTE. GPP SDR has become the mainstream in the design of SDR.
cessor (GPP) is selected by more SDR designers as baseband
We wish to introduce these two platforms to open source com- processing platform to complete functions like wireless trans-
munity and our results could provide a reference for further re- ceiver, signal generation, modulation/demodulation, encod-
search or deployment of GPP SDR. ing/decoding etc. There have been many SDR prototypes based
on GPP. One of the well-known GPP-based SDR (GPP SDR)
Keywords—SDR; GPP; OAI; srsLTE. projects is Sora [3] developed by Microsoft Research Asia. It
makes extensive use of modern GPP multi-core CPU and
I. INTRODUCTION SIMD extensions (SSE). GNU Radio [4] is a famous free open-
source SDR framework based on off-the-shelf GPP. The com-
With the development of wireless communications, new bination of GNU Radio and USRP is one of the most popular
technologies and services are emerging in an endless stream. SDR platforms for radio designers and researchers. Amarisoft
As a typical example, cellular phones have evolved from ana- [5] LTE100 is a commercial LTE soft base station based on
log wireless communication system which supports single- GPP, which performs better in terms of functionality, availabil-
standard and voice-only into high-bandwidth multi-standard ity and ease of use compared to open source products.
communication based on long term evolution (LTE). The 5G
standard [1] is also in a heated discussion at present. However, In this paper, we will focus on GPP SDR and analyze the
the traditional development in communication system has al- characteristics of it. Furthermore, we conduct the code analysis
ways been in a static paradigm providing non-adaptable radio and performance test of physical layer (PHY) downlink (DL)
systems. The wireless infrastructure radio designs have been on two famous open source GPP SDR platforms i.e. OpenAir-
implemented for a specific purpose using a combination of Interface (OAI) [6] and srsLTE [7]. We wish our results could
Application Specific ICs (ASICs), Digital Signal Processors provide a reference for further research and deployment of
(DSPs) and Field Programmable Gate Array (FPGA) devices. GPP SDR.
All those hardware chips promise a good performance and reli- The rest of the paper is organized as follows: Section Ⅱ
ability, but the tight coupling of product and protocols makes it provides a brief overview of state-of-the-art of GPP SDR. Sec-
difficult to operate between various standards. A specific radio
tion Ⅲ describes OAI and srsLTE respectively. We focus on
station can only be supported in a specific system. Users and
operators want to build a multi-band, multi-mode and multi- the main procedures and algorithms of PHY DL. Real-time
functional compatible system with features of plug and play
experimental results on these two platforms are presented in GPP SDR uses popular high level programming languages
Section Ⅳ. We draw conclusions in Section Ⅴ. such as C/C++, Python as development tools. So it can be port-
ed between processors with minimal rewriting required. And a
II. OVERVIEW OF GPP SDR great number of experienced programmers and well-written
libraries [10] are available. Besides, GPP has a great advantage
There are many silicon solutions for implementing base-
in human-computer interaction making it easier for developers
band process. On the basis of the physical medium, SDR plat-
to program in a familiar architecture and environment, which
form is mainly divided into three categories, i.e. FPGA-based
SDR, DSP-based SDR, GPP-based SDR. FPGA is flexible to opens the door to create communities of software radio devel-
some extent because they are capable of reprogramming and opers.
reconfiguration. The execution time of DSP can be fully de- Powerful computation.
termined. However, it is very difficult to program communica- Technology evolution in areas such as multi-core, SIMD,
tion modules on these two platforms. GPP is a flexible alterna- large on-chip caches, low latency on-chip system memory and
tive to fixed-function silicon like FPGA and DSP, and it is the high-speed I/O is making GPP a real alternative to process high
most common form of SDR currently. speed digital signals both for current and next generation of
A few of the latest SDR platforms are based on GPP. Alt- wireless systems. Other branches of computer science like
hough different platforms may have subtle differences among Cloud computing can also contribute to the improvement of
each other, they generally follow similar designs and share performance in GPP SDR by distributing multiple cores geo-
many common algorithms. In this section, we describe the ar- graphically concentrated into a virtual baseband pool.
chitecture of the GPP SDR system, then summarize some Better flexibility and scability.
common characteristics of it.
We can use multi-threads and multi-cores to support time
A. System Description crucial computational tasks in wireless communication systems
enhancing the real-time processing capability, which is a flexi-
GPP SDR platform is designed to utilize GPP to perform
ble design approach offered by GPP. It is also easy to integrate
the time crucial signal processing functions as well as other
protocol entities. As shown in Fig. 1, GPP SDR System is con- with other applications by blurring the boundaries of radio and
structed by a RF front-end connected with a General PC other applications, improving end to end effectiveness.
through the interface such as Peripheral Component Intercon- Relatively cheap.
nect Express (PCIe) bus / Universal Serial Bus (USB) 3.0. The Compared with other SDR platforms for a special purpose
RF front-end is a programmable hardware implementing only like FPGA and DSP, GPP has a larger market size and more
the RF and signal conversion functions as mentioned above; application scenarios. Besides, the development and debug
PHY and upper layer functions like medium access control cycle of SDR is shorter by means of GPP. Therefore it has rela-
(MAC) [8] are performed in software in a GPP, then upper
tively lower development costs.
layer interface distributes the signals to different service appli-
cations [9]. III. OAI AND SRSLTE: GPP SDR PLATFORMS
General PC OAI is the first and the most complete open source SDR
LTE platform implementing the full protocol stack of 3GPP
The Real Time OS over Intel X86 CPU Interface between RF standard in both Evolved Universal Terrestrial Radio Access
front-end and computer
Higher Layer PHY Processing (E-UTRAN) and Evolved Packet Core (EPC) currently.
Process RF font-end
srsLTE is a high-performance LTE library for SDR develop-
Buffer Control
Symbol Level
Bit Level
TX / RX
ment (UE). The library is highly modular with minimal inter-
Layer2 ADC / DAC module or external dependencies under a free open source
Process
software AGPL v2 license.
a) Synchronization
OAI UE will not continue to decode PSS and SSS after Pilot Channel
synchronizing with eNB. While srsLTE UE continues to per- Insertion Estimation
form synchronization maintenance after synchronization.
b) Frequency Offset Estimation
IFFT FFT
OAI uses frequency tracking synchronization algorithm
based on pilot signal to estimate fractional frequency offset in
the fourth OFDM symbol of each slot. In srsLTE, Maxi-
Transmitter Receiver
mumLikelihood (ML) algorithm based on Cyclic Prefix (CP) is
adopted to estimate fractional frequency offset, and integer
frequency offset estimation is carried out taking advantage of Fig. 3. Physical procedures.
orthogonality of PSS sequences in frequency domain.
The main functions for processing PDSCH corresponding
c) Time Offset Estimation to Fig. 3 in OAI and srsLTE are summarized in Table Ⅰ.
OAI uses channel impulse response to estimate time offset;
while srsLTE estimates the timing offset using PSS and SSS TABLE I FUNCTION NAME OF PHY DL IN OAI AND SRSLTE.
sequences.
PHY
d) Channel Estimation DL Functions
OAI srsLTE
For the sake of simplicity, UE of srsLTE and OAI both Encoding dlsch_encoding srslte_dlsch_encode
adopts Least-Squares (LS) algorithm and linear interpolation to Scrambling dlsch_scrambling srslte_scrambling_bytes
achieve channel estimation. srslte_mod_modulate_by
T Modulation dlsch_modulation
tes
X
However, there is a significant difference in the time inter- Pilot
generate_pilots_slot
srslte_refsignal_cs_put_s
polation of channel estimation between OAI and srsLTE. For a Insertion f
subframe in OAI, the channel estimation of OFDM symbol 12, IFFT do_OFDM_mod srslte_ofdm_tx_sf
FFT slot_fep srslte_ofdm_rx_sf
13 in current subframe is determined by the symbol 11 of cur-
Channel lte_dl_channel_estimat srslte_ue_dl_decode_esti
rent subframe and symbol 0 of the next subframe as shown in Estimation ion mate
Fig. 2. While in srsLTE, decoding the current subframe does R
srslte_demod_soft_demo
not require the next subframe to help. X Demodulation rx_pdsch
dulate s
Descrambling dlsch_unscrambling srslte_sequence_pdsch
Decoding dlsch_decoding srslte_dlsch_decode
a) Transmitter
As shown in Fig. 3, for each subframe, a transport block eNB OAI PHY DL UE
firstly will be encoded to a codeword. The codeword is then USRP B210
USRP B210
scrambled and modulated using one of the modulation schemes
of QPSK, 16QAM, 64QAM resulting in a block of complex-
valued modulation symbols and then mapped to resource ele-
ments. After pilot insertion, a codeword is transformed to
OFDM symbols. Now the data are ready to be transmitted in
RF front-end.
b) Receiver
On receiving link, FFT is firstly performed on each OFDM
symbol of received subframe. Then the UE of srsLTE and OAI
achieves channel estimation using LS as mentioned above.
Both srsLTE and OAI demodulate symbols using soft decision
decoding method based on log likelihood ratio (LLR) algo- srsLTE PHY DL
rithm. After demodulation, the data are descrambled and de- eNB UE
USRP B210 USRP B210
coded. In order to improve the execution speed of the program,
the SSE instruction set is adopted to complete the turbo decod-
ing function in OAI and srsLTE.