GPU-accelerated SDR Implementation of Multi-User Detector for Satellite Return Links

Transcription

1 DLR.de Chart 1 GPU-accelerated SDR Implementation of Multi-User Detector for Satellite Return Links Chen Tang chen.tang@dlr.de Institute of Communication and Navigation German Aerospace Center

2 DLR.de Chart 2 Preamble German Aerospace Center National aeronautics and space research center of Germany

3 DLR.de Chart 3 Preamble German Aerospace Center National aeronautics and space research center of Germany Wide range of R&D projects in national and international partnerships DLR & NASA operate the flying infrared telescope SOFIA DLR operates/coordinate the Columbus (European lab module on ISS) Galileo satellite navigation system

4 DLR.de Chart 4 Preamble The work presented here has been developed in the scope of NEXT (Network Coding Satellite Experiment) project funded by German Space Agency that paved the way to the GEO research communication satellite H2Sat (2017) H2Sat: explore and test new broadband (high data rate) satellite communication

5 DLR.de Chart 5 Overview What Problems? Introduction and Motivation How to Solve? Multi-User Detection (MUD) System Design GPU-accelerated SDR Implementation of MUD Result and Outlook

6 DLR.de Chart 6 Overview What Problems? Introduction and Motivation How to Solve? Multi-User Detection (MUD) System Design GPU-accelerated SDR Implementation of MUD Result and Outlook

7 DLR.de Chart 7 Introduction and Motivation Unidirectional satellite broadcast service

8 DLR.de Chart 8 Introduction and Motivation Bidirectional satellite communication Forward link Return link e.g. internet over satellite; interactive satellite TV services Multi-user access issue

9 DLR.de Chart 9 Introduction and Motivation Bidirectional satellite communication Forward link Return link e.g. internet over satellite; interactive satellite TV services Multi-user access issue

10 DLR.de Chart 10 Introduction and Motivation Bidirectional satellite communication Forward link Return link e.g. internet over satellite; interactive satellite TV services TDMA Multi-user access issue Multi-access schemes: Time Division Multiple Access f t

11 DLR.de Chart 11 Introduction and Motivation Bidirectional satellite communication Forward link Return link e.g. internet over satellite; interactive satellite TV services FDMA Multi-user access issue Multi-access schemes: Frequency Division Multiple Access f t

12 DLR.de Chart 12 Introduction and Motivation Bidirectional satellite communication Forward link Return link e.g. internet over satellite; interactive satellite TV services MF-TDMA (e.g. DVB-RCS) Multi-user access issue Scarcity and high cost of satellite frequency spectrum (millions of dollars) How to improve spectrum efficiency? Multi-User Detection (MUD) f t

13 DLR.de Chart 13 Overview What Problems? Introduction and Motivation How to Solve? Multi-User Detection (MUD) System Design GPU-accelerated SDR Implementation of MUD Result and Outlook

14 DLR.de Chart 14 Multi-User Detection (MUD) System Multiple users transmit at the same frequency and time A transparent satellite return link Multiuser Detection (MUD) Increase spectrum efficiency Few practical MUD implementations for satellite systems High complexity Sensitive to synchronization and channel estimation errors Main objectives: Develop a MUD receiver Increase decoding throughput real-time processing

15 DLR.de Chart 15 MUD System Design p user 1 user 2 transmit user 2 for free f Successive Interference Cancellation (SIC) Sequentially decode users & cancel interference Linear complexity on number of users Straightforward extension to support more users

16 DLR.de Chart 16 Overview What Problems? Introduction and Motivation How to Solve? Multi-User Detection (MUD) System Design GPU-accelerated SDR Implementation of MUD Result and Outlook

17 DLR.de Chart 17 MUD System Design SDR = Software Defined Radio Components (e.g. filter, amplifier, modulator etc.) in a communication system are implemented via software Benefits vs hardware-based devices: Flexible to change Lower cost Shorter development time Drawback vs hardware-based devices: Low processing power

18 DLR.de Chart 18 SDR Programmable radio devices DSP (Digital Signal Processor) FPGA (Field Programmable Gate Arrays) SoC (Programmable System on Chip) GPGPU (General-Purpose GPU)

19 DLR.de Chart 19 GPU-based SDR Restriction of FPGA-based SDR Long development time and complexity No standardized protocols, interfaces or architectures less portable Nvidia CUDA GPU-based SDR High performance Nvidia Tesla c2070: 448 cores; 515 GFLOPs of double-precision peak performance

20 DLR.de Chart 20 GPU-based SDR Restriction of FPGA-based SDR Long development time and complexity No standardized protocols, interfaces or architectures less portable Nvidia CUDA GPU-based SDR High performance Less effort to develop Unified architecture more portable GPU: Nvidia GTX285 HC1: 5 x Virtex-5 FPGA Ref: GPU vs FPGA for high productivity computing, 2010 (David H. Jones, A. Powell, C. Bouganis, Peter Y.K. Cheung)

21 DLR.de Chart 21 MUD System Design Real-time implementation of MUD is challenging T ddd T fffff Processing bottlenecks: LDPC channel decoding EM channel estimation Resampling and interference cancellation V 1 V 2 V 3 V 4 V n... V i C j U1: n = 4800 k = 3200 LDPC... C 1 C 2 C 3 C n - k C j V i U2: n = 4800 k = 2400

22 DLR.de Chart 22 GPU-based MUD Processing bottlenecks LDPC Channel Decoding EM Channel Estimation Interference Cancellation To be accelerated by GPU 4800 nodes to be processed iteratively Thousands-points FFT iteratively Resampling, thousands-points FFT

23 DLR.de Chart 23 MUD receiver on GPU Processing bottlenecks: LDPC channel decoding EM channel estimation Resampling and interference cancellation Data transfer between host and device memory (144GB/s of Nvidia Tesla vs. 8GB/s of PCIe*16) GPU CPU GPU CPU GPU CPU GPU CPU All parts of each single user receiver and interference cancellation on GPU Minimize the latency of intermediate data transfer between host and device memory

24 DLR.de Chart 24 Overview What Problems? Introduction and Motivation How to Solve? Multi-User Detection (MUD) System Design GPU-accelerated SDR Implementation of MUD Result and Outlook

25 DLR.de Chart 25 Simulation Setup GPU Nvidia Tesla c2070 (1.15GHz, CUDA compatibility: 2.0) Comparison benchmark: Intel Xeon CPU E5620 (2.4GHz) Channel coding: LDPC Irregular Repeat Accumulate Blocklength: 4800 bits U1 coderate: 2/3, U2 coderate: 1/2 Baud-rate: symbols/second real-time decoding threshold: ca. 85ms (66 kbps)

26 DLR.de Chart 26 Simulation Result Comparison of total processing time of MUD between CPU and GPU

27 DLR.de Chart 27 Simulation Result Comparison of total processing time of MUD between CPU and GPU

28 DLR.de Chart 28 Simulation Result Real-time threshold

29 DLR.de Chart 29 Summary SDR implementation of MUD receiver High flexibility and low cost Extension to support more users GPU acceleration 1.8x ~ 3.8x faster than the real-time decoding threshold Still space to improve New GPU better performance GPU CUDA is very promising for powerful parallel computing Low learning curve Heterogeneous: mixed serial-parallel programming Scalable Days/weeks of simulation hours

30 DLR.de Chart 30 Thank you very much! Q&A