1. Programmable Radios: A Personal Viewpoint Ashu Sabharwal Rice University Houston, TX

2. Congratulations to SoRa Team ! Ashu SabharwalRice University

3. Three Questions Why did I get involved ? Have I learnt anything ? What is on my wishlist ? Ashu SabharwalRice University

4. Why Did I Get Involved ? Major speed innovations occur at PHY/MAC layer –Coding (Convolutional, Viterbi, LDPC, Turbo) –MIMO –Opportunistic scheduling Many more in pipeline –Cooperative coding, network codes, interference alignment… Big departure from traditional networking –Need sanity checks… But do I really need it ? Ashu SabharwalRice University

5. Do I Really It ? Wireless blessed with information/comm/coding theory –Many many success stories Ashu SabharwalRice University

6. Do I Really It ? Wireless blessed with information/comm/coding theory –Many many success stories But hear only success stories –Multiuser detection: 15+ yrs of work, never got deployed –Beamforming: 15+ yrs, maybe finally in WiMax Ashu SabharwalRice University CONCEPT PROOF-OF-CONCEPT Feedback loop too slow & largely broken !

7. What do I Need ? Quick and accurate answers –Physical, MAC layer tests –Full control of all variables Programmable, measurable and deployable Ashu SabharwalRice University

8. Hardware & Design Flows Ashu SabharwalRice University USRPWARP SoRa WARP WARPLab WARPMAC GNURadioSoRa

9. Delivering on Promises ? Many publications for proof-of-concept –GNUradio –WARP –Very soon using SoRa Fundamental shifts ? –Change in models ? –New dominant effects ? –New problem formulations ? Ashu SabharwalRice University

10. Case I: Quantized Beamforming (WiMax/802.11n) Large expected gains from closed loop beam-forming Ashu SabharwalRice University Alamouti (theory) Beamforming (theory)

11. Robustness to Channel Model Large expected gains from closed loop beam-forming Error floor with a small model perturbation Ashu SabharwalRice University Alamouti (actual) Beamforming (actual)

12. Reason for Breakdown Very sensitive to how long the channel remains constant Breaks the equalizer and thus, whole PHY Ashu SabharwalRice University Alamouti (actual) Beamforming (actual)

13. New Model, Simple Fix Re-model, accounting for channel change New packet structure Ashu SabharwalRice University

14. New Model, Simple Fix Beamforming advantage returns Original model did not capture all dominant effects Ashu SabharwalRice University New Beamforming (actual)

15. New Foundations More generally –Feedback errors and delay can cause havoc –Transmitter and receiver get mismatched –Nearly all theory predictions breaks down Better models for physical layer models with fast feedback New fundamental results (Aggarwal & Sabharwal09) –Proof that too many feedback bits not useful –Often more than one feedback bit is a waste ! Ashu SabharwalRice University

16. Case II: Cooperative Coding Physical layer, symbol time-scale cooperation Use both routes simultaneously Pool distributed resources of power/antennas Ashu SabharwalRice University Receiver Relay Source

17. Case II: Cooperative Coding No system demonstration till date Cannot wait 15 years to know its fate Ashu SabharwalRice University Receiver Relay Source

18. Case II: Cooperative Coding on WARP Built with WARPLab Allows fine-grained control of each piece Systematic experiments to understand dominant effects Ashu SabharwalRice University 2x2x2 MIMO Relay

19. Case II: Cooperative Coding, First Results Large gains with optimal processing –6-9 dB over non-relay –3-6 dB over simple No RF or A/D Ashu SabharwalRice University Optimal Simple Wu, Amiri, Duarte, Cavallaro09

20. Case II: Cooperative Coding, First Results With RF –Optimal degrades a lot –Simple is robust Optimal very sensitive to perturbations Why ? –A/D robs important bits –More antennas need more bits Ashu SabharwalRice University Optimal Simple Wu, Amiri, Duarte, Cavallaro09

21. Wish 1: Higher Quality Radios Low-quality signals no post-processing can save the day WARP radios top of the line –But we need better to push the limits ! –Better dynamic range, lower noise floor and bigger A/D,D/A Clean-slate research –Platforms should be an order of magnitude better –Then research can find new sweet spots Ashu SabharwalRice University

22. Case III: Local View in Networks Why current info theory of networks of little use ? Models miss an important component –Nodes only have local network information –Nodes mismatched in their knowledge Ashu SabharwalRice University

23. Theory of Distributed Decisions Two elements (Aggarwal, Liu and Sabharwal09) –A protocol abstraction which quantifies local view –Distributed protocols as channel codes First info theory analysis with hidden nodes –Predicts the losses seen in practice –Losses are unavoidable Ashu SabharwalRice University Local view Full view limit Capacit y

24. Wish 2: Cross-community Fertilization Wireless is many communities –CE + EE + CS –Different languages: VHDL, MATLAB, C –WARP, WARPLab, WARP_MAC Isolation and Integration –Isolated controlled experiments –Integration of concepts Ashu SabharwalRice University

25. Wish 2: Cross-community Fertilization Ashu SabharwalRice University USRP WARP SoRa WARP WARPLab WARPMAC GNURadioSoRa Much remains to be done –Tools remain hard to use –Little coherence across communities

26. Wish 3: Hardware-normalized Results How do you compare results from different testbeds Different hardware –System bandwidth –Speed of processing Some examples –EVM, spectral efficiency –Situation likely to get worse Much remains to be done ! Ashu SabharwalRice University

27. Answers Why did I get involved ? –Problems which are unsolved and relevant Have I learnt anything ? –Yes, more to come ! What is on my wish list ? –Higher quality radios –Cross-community fertilization –Hardware-normalized metrics Ashu SabharwalRice University

28. Ashu SabharwalRice University Exciting times, fun path ahead ! Questions ? WARP: http://warp.rice.edu