Fine-grained Channel Access in Wireless LAN

1. Fine-grained Channel Access in Wireless LAN SIGCOMM 2010 Kun Tan, Ji Fang, Yuanyang Zhang,Shouyuan Chen, Lixin Shi, Jiansong Zhang, Yongguang Zhang

2. Trends in 802.11 WLANs • PHY data rate increases • 802.11n up to 600Mbps • 802.11ac/ad up to >1Gbps • Data throughput efficiency degrades with PHY data rate

3. Reasons for Low Throughput Efficiency • Contention resolution overhead due to CSMA • Coarse-grained channel allocation • Whole channel allocated to a single station

4. Possible solutions • Reduce overhead • Infeasible, physical laws/technology • Increase useful channel time – frame aggregation • OK, used in 802.11n but • Practical limitations: 80% efficiency at 300Mbps requires frame size of 23KB!

5. An Alternative ApproachFine-Grained channel Access • Divide channel into smaller subchannels • Multiple users contend for and use subchannels simultaneously • Based on traffic demands • Amortize MAC coordination, increase channel efficiency

6. Challenges • Need to avoid interference between neighbor subchannels • Traditional approach: guard bands • High overhead • OFDM – Orthogonal Frequency Division Multiplexing • “Eliminates” need for guard bands • Requires tight synchronization (100s of nsec)

7. OFDM – High Level Overview • Divides spectrum into many small, partially overlapping subcarriers • Subcarrier frequencies “orthogonal” to each other • OFDM system with FFT size N • N subcarriers, each with bandwidth B/N

8. OFDM as multi-access technology • Different stations assigned different subcarriers in the same channel • WiMAX, LTE • Symbol timing alignment is critical • Requires tight synch with cellular BS • Use of guard times, CP (cyclic prefic) • 802.11: CP-to-symbol length ratio 1:4 (0.8μs to 3.2μs)

9. OFDM-based Channel Access in WLANs • Challenge 1: Coordinate random access among multiple stations • Cannot use cellular-type synchronization • Need a new OFDM architecure for distributed coordination • Challenge 2: Longer symbol length to maintain 1:4 CP-to-symbol length ratio • Makes backoff mechanism inefficient • Need new MAC contention mechanism, new backoff scheme

10. Paper Contributions • Design and implementation of FICA • Cross-layer architecture based on OFDM • Enables fine-grained subchannel random access in WLANs • Two key techniques • New PHY architecture based on OFDM • Novel frequency domain contention method

11. FICA Overview • Uplink transmission • Downlink transmission similar

12. Symbol Time Misalignement • Using carrier sensing • Using reference broadcast

13. PHY Architecture • Each 802.11 channel (20Mhz) divided into 1.33Mhz subchannels • 14 + guardband • Each subchannel divided into 17 subcarriers • 16 + pilot • Data is transmitted over all 16 subcarriers

14. Frequency Domain Contention • Allocate K subcarriers per subchannel • Contention band • Each node contending for a subchannel picks randomly a subcarrier and sends a ‘1’ in M-RTS • AP arbitrates contention and sends winning subcarriers in M-CTS

15. Issues in Frequency Domain Contention • What if 2 nodes choose the same subcarrier? • Collision • No transmission • How large should K be? • K=16 (initial backoff value in 802.11) • Who is returning M-CTS? • Only potential receivers • Allocate 40 subcarriers, hash receiver’s ID into 0..39, set appropriate subcarrier

16. M-RTS, M-CTS

17. Frequency Domain Backoff • How many subchannels can a node contend for? • n=min(Cmax, lqueue)

18. Downlink Transmission • AP can transmit simultaneously to many clients • Different subchannels per client, has to contend for each subchannel • Two-way traffic • FICA uses no backoff, AP and station can send M-RTS simultaneously • Solution: use different DIFS to prioritize transmissions • Fixed DIFS to all stations, 2 DIFS to AP • If AP uses short DIFS, use long DIFS next time • If AP receives M-RTS, use short DIFS next time • Fair interleaving of uplink-downlink, not among all stations!

19. Multiple Domains – Hidden Terminals • Hidden terminals • Collisions may cause M-RTS/M-CTS loss • Random backoff after M-CTS loss • Multiple domains • Nodes may receive inconsistent M-CTS from different nodes • Node only allowed to transmit if wins contention in all domains it participates.

20. Evaluation • Simulation • Implementation

21. Simulation Setup • Event-based simulator • Only uplink traffic • Packet loss only due to collisions • Compare against 802.11n • No aggregation • Full aggregation • Mixed traffic

22. Simulation Results No Aggregation

23. Simulation Results Full Aggregation • All nodes saturated, frame size 18KB!

24. Simulation Results Mixed Traffic

25. Implementation • Sora platform [NSDI ‘09] • Fully programmable software radio platform • Implementation cannot run in real time • Takes too long to transfer PHY frames from CPU to RCB (Even though Sora is the fastest platform available) • Have to prestore all PHY frames in RCB

26. Evaluation – Time Misalignment With Broadcasting With Carrier Sensing

27. Reliability of PHY Signaling

28. Demodulation Performance

29. Conclusion • Trend in 802.11 WLANs • Throughput efficiency decreases as data rate increases • Fundamental reason • Entire wide-band channel allocated to one node • FICA • Cross-layer design to enable fine-grained subchannel random access • New PHY arhitecture based on OFDM • New frequency domain backoff scheme