1. 11 1 The Next Generation Challenge for Software Defined Radio Mark Woh 1, Sangwon Seo 1, Hyunseok Lee 1, Yuan Lin 1, Scott Mahlke 1, Trevor Mudge 1, Chaitali Chakrabarti 2, Krisztian Flautner 3 1 Advanced Computer Architecture Lab, University of Michigan 2 Department of Electrical Engineering, Arizona State University 3 ARM, Ltd.

2. 22 2 University of Michigan -SAMOS 2007 3G Wireless Large Coverage Outdoor - High Mobility Up to 14Mbps

3. 33 3 University of Michigan -SAMOS 2007 Expected Wireless Growth  The growth of wireless will require more bandwidth

4. 44 4 University of Michigan -SAMOS 2007 4G Wireless  What we need  Adaptive high performance transmission system  Great candidate for SDR Large Coverage – 100Mbps Coverage Outdoor - High Mobility Macro Cells Pico Cells Isolated HotSpots – 1Gbps Coverage Indoor – Very Low Mobility

5. 55 5 University of Michigan -SAMOS 2007 Next Generation Wireless – 4G  3 Major Components to 4G  Modulation/Demodulation  Multiple-Input Multiple-Out (MIMO)  Channel Decoder/Encoders

6. 66 6 University of Michigan -SAMOS 2007 Modulation - OFDM  Can be implemented with IFFT/FFT

7. 77 7 University of Michigan -SAMOS 2007 Major Component of Modulation – FFT/IFFT  Very wide data level parallelism  Requires complex operations

8. 88 8 University of Michigan -SAMOS 2007 MIMO (Multiple Input – Multiple Out)  Previously we used single antenna systems  Now we use multiple antennas to increase the channel capacity  Diversity - High Reliability  Space Time Block Codes (STBC)  Multiplexing – High Throughput  Vertical-BLAST (V-BLAST)

9. 99 9 University of Michigan -SAMOS 2007 Space Time Block Codes (STBC)

10. 10 University of Michigan -SAMOS 2007 STBC  Requires complex operations  Low Data Movement  Highly parallelizable

11. 11 University of Michigan -SAMOS 2007 Vertical-BLAST (V-BLAST)

12. 12 University of Michigan -SAMOS 2007 V-BLAST  Implementation Based on Square Root Method for V-BLAST  Original requires repeated pseudo-inverse calculation for finding the strongest signal  This algorithm has reduces complexity  Complexity  Requires matrix operations on complex numbers  Many Matrix Transformations

13. 13 University of Michigan -SAMOS 2007 Channel Decoding  3G Technologies in 4G  Viterbi  Turbo Decoder  New to 4G  LDPC  Better performance characteristics compared to Turbo and Viterbi

14. 14 University of Michigan -SAMOS 2007 LDPC

15. 15 University of Michigan -SAMOS 2007 LDPC  Min-Sum Decoding Used  Regular LDPC code  Can get benefit from Wide SIMD  Can do the Bit Node and Check Node  Alignment of Check and Bit nodes is a problem

16. 16 University of Michigan -SAMOS 2007 SODA PE Architecture  SIMD – 32 Wide, 16-bit datapath, Predicate Execution

17. 17 University of Michigan -SAMOS 2007 Key 4G algorithms 100 Mbps1 Gbps MCycle/s FFT2x3604x360 IFFT2x3604x360 STBC240- V-BLAST-1900 LDPC77004x18500 4G Workload on SODA  100 Mbps 4G requires 8Ghz SODA PE  1 Gbps 4G requires 20Ghz SODA PE

18. 18 University of Michigan -SAMOS 2007 SODA With Technology Scaling 180nm130nm90nm65nm45nm32nm22nm Vdd (V)1.81.31.1 10.90.8

19. 19 University of Michigan -SAMOS 2007  We can’t do any of 4G with technology scaling on one core  Would 8GHz cores even be an energy efficient solution?  What about 1Gbps?  Are we ever going to get a 20GHz core?  Cannot rely on technology scaling to give us 4G for free  4G SDR will require algorithmic and architectural innovations SDR Challenges In 4G

20. 20 University of Michigan -SAMOS 2007 4G Algorithm-Architectural Co-design  Architectural improvements (SODA II)  Specialized functional units  CISC-like complex arithmetic operations  Specialized data movement hardware  Less strain on the memory system  Wider SIMD  How wide can we go?  More PEs  What does the interconnect look like?  Algorithmic optimization through parallelization  Reduce intra-kernel communication  Reduce memory accesses  Arithmetic is much cheaper than data movement

21. 21 University of Michigan -SAMOS 2007 Thanks  Questions?

22. 22 University of Michigan -SAMOS 2007 Successive Cancelling for V-BLAST  V-BLAST successive interference cancelling (SIC)  The ith ZF-nulling vector w i is defined as the unique minimum-norm vector satisfying  Orthogonal to the subspace spanned by the contributions to y i due to the symbols not yet estimated and cancelled and is given by the ith row of

23. 23 University of Michigan -SAMOS 2007 Alamouti Scheme  Assumption: the channel remains unchanged over two consecutive symbols  Rate = 1  Diversity order = 2  Simple decoding

24. 24 University of Michigan -SAMOS 2007 Advantages of Software Defined Radio  Multi-mode operations  Lower costs  Faster time to market  Prototyping and bug fixes  Chip volumes  Longevity of platforms  Enables future wireless communication innovations  Cognitive radio