1. Bringing Life to Dead Spots Grace Woo Pouya Kheradpour, Dawei Shen, and Dina Katabi

2. Many APs But Still Poor Coverage Problem increases with mobility and low power devices mit1 mit4 mit5

3. Poor Coverage Is Not No Coverage! 010101011111 X 011101011011 X Can recover a correct packet if we combine the correct bits from these receptions Spatial Diversity  APs are unlikely to have same bit error Loss Currently High Bit Error  Persistent Loss  Dead Spot

4. But Which AP Got the Right Bit? Clearly can’t have per bit checksum Prior work (MRD) tries all block combinations to satisfy checksum Exponential Complexity Works for a few bit errors But not dead spots First bit is “0” First bit is “1”

5. SOFT Recovers a correct packet from its faulty receptions at APs Leverages physical layer hints to identify correct bits SOFT’s delivery rate is up to10x higher than current WLANs and MRD

6. SOFT Architecture APs leverage high-speed Ethernet to combine their receptions Internet AP 1 AP 2 SOFT Combiner AP n Wired Ethernet But which bits are correct? 010001001111 X 010101001011 X 010001001011

7. Physical Layer Knows More! PHY already estimates a confidence in its 0- 1 decision  Soft Value PHY 01011001 Measured Soft Values PDF of per bit soft values Larger absolute soft values  More confidence in bit Soft Value < 0  “0” Soft Value > 0  “1”

8. We Use the Soft Values SOFT changes the PHY interface to expose the soft values to higher layers SOFT combines the soft values of a bit to decode it correctly The combiner forwards the decoded packet if it satisfies the 802.11 checksum AP 1 AP 2 SOFT Combiner Soft packet 010110111

9. How Do We Combine Soft Values? 0.4 - 0.1 - 0.2 How do we decode the bit? Maximum soft value  Bit is “ 1 ” Majority vote  Bit is “0” Average  Bit is “ 1 ” Say for a particular bit, we got Different Combining Methods  Different Answers!

10. SOFT Combining Algorithm Let be the noise variance on the channel to AP i For AWGN and dead spots rule is proven optimal. Let S ij be the soft value of bit j reported by AP i SOFT decision rule: Intuitively, we want to favor less noisy channels

11. But, How Does SOFT Get the Noise Variance? Randomness in soft values is caused by channel noise Estimate from the PDF of the soft values in packet Measured Soft Values PDF of per bit soft values

12. How About Overhead? PHY soft values can be 32-bit float  Excessive Ethernet traffic Solution Invoke SOFT only when associated AP can’t decode Quantize soft values (we used 3 bits)

13. What About the Downlink? Use Time Diversity Combine a packet with its retransmission Use Time Diversity Combine a packet with its retransmission 010001001011 X 010101000011 X

14. Performance

15. SOFT Implementation Software – GNURadio codebase Hardware – USRP frontend GMSK and DBPSK modulations Soft values are inputs to the slicer Poor Coverage: SNR 5 – 12 dB BER about 10 -3

16. Experimental Setup 13 GNURadio nodes Compared –Current 802.11 WLAN (user associates with best AP) –MRD –SOFT Each Experiment –3 random APs –Random source –Transmit 500 packets

17. Does SOFT Help? CDF of 100 experiments Packet Delivery Rate

18. Does SOFT Help? CDF of 100 experiments Packet Delivery Rate

19. Does SOFT Help? CDF of 100 experiments Packet Delivery Rate

20. Does SOFT Help? CDF of 100 experiments Packet Delivery Rate SOFT’s delivery rate can be 10x higher 10x

21. Packet Delivery Rate Bit Error Rate Performance with Increasingly Poor Coverage

22. Packet Delivery Rate Bit Error Rate Performance with Increasingly Poor Coverage Current Approach

23. Packet Delivery Rate Bit Error Rate Current Approach MRD Performance with Increasingly Poor Coverage

24. Packet Delivery Rate Bit Error Rate Current Approach SOFT MRD SOFT Addresses Dead Spots Performance with Increasingly Poor Coverage

25. Current Approach CDF over 50,000 packets Number of Retransmissions Until Correct Packet SOFT on Downlink

26. Current Approach SOFT CDF over 50,000 packets SOFT on Downlink 17 ReTx Number of Retransmissions Until Correct Packet Much Higher Throughput!

27. Combining Method Is Important CDF of 100 experiments Packet Delivery Rate MAX SOFT Majority SOFT Outperforms MAX and MAJORITY

28. Effect of Quantization SOFT Average Delivery Rate 3 Bits 32 Bits 2 Bits All presented results are for 3-bit quantization! Overhead on Wired Ethernet is Acceptable

29. Related Work MRD [Miu05] Soft handoff in cellular networks; Maximum Ratio Combining (MRC) [Brennan55,Yang99] H-ARQ & Chase Combining [ASX03] Partial Packet Recovery [Jam07] We exploit PHY Info We combine across nodes Use quantized soft values Results for WLAN We exploit spatial diversity Quantized soft values Implementation We exploit spatial diversity Combine soft value

30. Related Work Soft and softer handoff in cellular networks Theoretical Maximum Ratio Combining (MRC) [Brennan55,Yang99] H-ARQ & Chase Combining [ASX03] Partial Packet Recovery [Jam07]

31. Conclusion WLAN can have better coverage if the interface to the PHY exposes soft values Delivery rate can be up to 10x higher Ethernet overhead is acceptable The new architecture, SOFT, can co-exist with unmodified 802.11 cards and APs