Migrating Backoff to the Frequency Domain No time to countdown: Migrating Backoff to the Frequency Domain Souvik Sen Romit Roy Choudhury, Srihari Nelakuditi MobiCom 2011

Introduction

Current WiFi Channel Contention Conventional WiFi networks >> perform channel contention in time domain more than 30% reduction in throughput due to backing off Solution >> migrate the backoff operation to the frequency domain

Back2F 802.11 a/g/n PHY adopts OFDM Pretend OFDM subcarriers as integers The process is similar to current WiFi Picks a random value from specified range [0,CW-1] ex) a Once the channel becomes idle, it begins the backoff operation Instead of counting down, it transmits a symbol on the a_th sbcarrier It has two antenna One for transmit, the other listens to determine which of the subcarriers are active

Back2F Challenges >> Accuracy >> Collision >> Ranking Contributions >> Time domain -> Frequency domain >> Design an OFDM based system >> Make prototype on the USRP/GNU Radio platform Challenges 1. Active subcarriers need to be detected accurately 2. Collision among nodes need to be mitigated successfully 3. Since nodes are located in different contention neighborhoods, the node ranking do not obey any global order Contributions We identify an opportunity to migrate protocol operations from the time to the frequency domain. Although we instantiate our ideas through a WiFi based MAC, they may be generalized to other arbitration strategies. We design an OFDM based system where random backoff is realized by selectively transmitting on a subcarrier. A logical order among senders is enforced in a decentralized manner, for improved channel usage. We address the challenges behind such a scheme, and prototype it on the USRP/GNU Radio platform. Promising results, in terms of throughput, fairness, and scalability, give us confidence to build a larger system

802.11 and OFDM

802.11 channel access Chose CW (contention window) Count down only if the channel is idle (DIFS) If the node senses a busy channel, frozen Revived only after the channel is idle Node completes the countdown first begins transmission DIFS >> DCF Interframe Space SIFS >> Short Interframe Space PIFS >> PCF Interframe Space

Channel under-utilization due to 802.11’s backoff

OFDM Benefit >> Its ability to cope with channel adversities; narrowband interference and frequency-selective fading Back2F >> Back2F node picks a random number and transmits a short signal only on the that subcarrier

Architecture and Design

Backoff within single collision domain Basic operation of Back2F

Backoff within single collision domain 2nd Round situation

Backoff within single collision domain How about single winner? >> Must go 2nd round Is it enough 2nd round? >> Two rounds of contention suffice for up to 50 contenders There is no way to reliably tell between a sole winner and collisions Because a Back2F node can detect which subcarriers are active, but cannot tell how many nodes transmitted on a given subcarrier

Backoff over multiple collision domain 1st domain: AP2 is winner 2nd domain: AP3 is winner

Backoff over multiple collision domain

Coping with Misdetection due to Fading False positive >> No one has transmitted on that subcarrier, but a node detects a subcarrier False negative >> A node fails to detect a subcarrier False positive does not affect Back2F >> cause the channel to be wasted for a DIFS duration Let assume that self subcarrier i fails to detect a legitimate subcarrier f False negative has no effect when number i is less than f However, if f is smaller than i in the second round, then it causing a certain collision It defeats the purpose of backing off

Coping with Misdetection due to Fading To alleviate the impact of false negative A node can transmit on subcarrier i and ((F/2 + i) mod F) F is the total number of subcarriers

Adjusting backoff values Pseudo code for Back2F with batched transmissions 2nd round의 가장 큰 backoff 값을 1st round 의 나머지 loser들에서 빼준 값이 새로운 loser node 들의 값이다 Batched transmissions over multiple collision domains >> Back2F incurs a penalty of DIFS whenever the batch order is not followed

Points of Discussion

Active subcarrier detection 10 consecutive subcarriers (out of 52) FFT: 고속 푸리에 변환 DFT와 그 역변환을 빠르게 수행하는 효율적인 알고리즘 N-point FFT N: the length of your sequence

Active subcarrier detection Higher point FFT reduces leakage When high point FFT, even in presence of a high self subcarrier, transmissions on adjacent and nearby subcarriers are reliably discerned

Active subcarrier detection 259 pt. FFt, the subcarrier adjacent to the self-subcarrier experiences less than 10dB leakage

Collision Probability Collision probability remains less than 2%

Implementation and Evaluation

USRP/GNURadio Prototype Transmitter >> 52 subcarriers, converted into a time domain signal using 64pt. IFFT Listening antenna >> execute 64, 128, 216 pt. FFTs GNU Radio is a free software toolkit for learning about, building, and deploying software-defined radio systems. Universal Software Radio Peripheral (USRP) The USRP product is intended to be a comparatively inexpensive hardware device for software radio.

USRP/GNURadio Prototype Evaluated using traces at 65 locations

Subcarrier detection 64pt. FFT 128pt. FFT

Subcarrier detection 256pt. FFT Subcarriers above 14dB can be detected reliably 14dB 또는 그 이상의 SNR에서 97%의 정확도로 adjacent subcarrier들을 검출하였다.

False positive/negative False positive는 앞서 말했듯이 Back2F에 그다지 큰 영향을 끼치지 못하고, False negative의 경우에는 영향을 미치지만, 14dB 정도에서 확률이 작다. 14dB는 이전의 subcarrier detection에서 보면 14dB 이상이면 97%확률로 검출함

Throughput gain with batch

Throughput gain with batch Batching contributes 5% of the gain The gains with batching is small because basic Back2F has already reduced the backoff duration substantially

Traffic type Gains are better with Skype traffic due to smaller packet sizes This is because backoff overheads are fixed, making it proportionally larger for short packets

Performance in dense networks They simulated HD traffic in a single collision domain under varying densities and different bitrates Back2F provides gains are in the range of 15% to 30%

Limitations and Discussion

Limitation and Discussion Robustness of subcarrier based backoff Collisions due to hidden terminals Gain over packet aggregation Interoperability with 802.11 Their results are derived from lab experiments, without node/environment mobility Back2F more sensitive to channel fluctuation They observe that Back2F collisions are solely due to hidden terminals perhaps turn on RTS/CTS (Request-to-send/Clear-to-send) Aggregation may not be popssible for real time traffic Back2F provides gain with aggregation at higher rates Back2F wait for longer than DIFS May interoperate but will cause unfairness

Conclusion

Summary Randomization is an effective method for contention resolution 802.11 time domain backoff requires channel to remain idle Randomization possible in frequency domain by using OFDM subcarriers Back2F: practical system realizing frequency domain contention