No Time to Countdown: Backing Off in Frequency Domain Souvik Sen, Romit Roy Choudhury, Srihari Nelakuditi

Current WiFi Channel Contention Random Backoff = 15 Random Backoff = 25 AP1 AP2 R1 R2 B1=15 AP1 DIFS AP2 Time B2=25 2

Current WiFi Channel Contention Random Backoff = 0 Random Backoff = 10 AP1 AP2 R1 R2 B1=15 B1=0 AP1 DIFS AP2 Time B2=25 B2=10 3

Current WiFi Channel Contention Carrier Busy Transmit AP1 AP2 R1 R2 B1=15 B1=0 AP1 Data and ACK DIFS AP2 Wait Time B2=25 B2=10 4

Current WiFi Channel Contention Random Backoff = 18 Random Backoff = 10 AP1 AP2 R1 R2 B1=15 B1=0 B1=18 AP1 Data and ACK DIFS B2=10 AP2 Wait Time B2=25 B2=10 5

Current WiFi Channel Contention Random Backoff = 8 Random Backoff = 0 AP1 AP2 R1 R2 B1=15 B1=0 B1=18 B1=8 AP1 Data and ACK DIFS B2=10 B2=0 AP2 Wait Time B2=25 B2=10 6

Current WiFi Channel Contention Carrier Busy Transmit AP1 AP2 R1 R2 B1=15 B1=0 B1=18 B1=8 AP1 Data and ACK Wait DIFS B2=10 B2=0 AP2 Wait Data and ACK Time B2=25 B2=10 7

Current WiFi Channel Contention Carrier AP1 Busy AP2 Transmit R1 R2 B1=15 B1=0 B1=18 B1=8 High channel wastage due to backoff 35% overhead at 54Mbps AP1 Data and ACK Wait DIFS B2=10 B2=0 AP2 Wait Data and ACK Time

Current WiFi Channel Contention  Backoff is not fundamentally a time domain operation  Its implementation is in time domain Time Domain Frequency Domain Can we implement backoff in frequency domain? Are there any benefits in doing so?

Frequency domain contention resolution 802.11 a/g/n PHY adopts OFDM Wideband channel divided into 48 narrowband subcarriers Copes better with fast, frequency selective fading Purely a PHY motivation Subcarriers: 1 2 3 4 … 48 Frequency We propose Back2F MAC Opportunity: Pretend OFDM subcarriers as integers Emulate randomized backoff

 Replace temporal with subcarrier transmission Back2F: Main Idea  Replace temporal with subcarrier transmission 6 18 47 47 AP1 AP2 Backoff = 6 Backoff = 18 R1 R2

 Replace temporal with subcarrier transmission Back2F: Main Idea  Replace temporal with subcarrier transmission Other’s Backoff = 18 Other’s Backoff = 6 6 18 47 47 Listen Antenna Listen Antenna AP1 AP2 Backoff = 6 Backoff = 18 R1 R2 Both APs learn AP1 is the winner

Back2F: Scheduled Transmission Active subcarriers imply backoff chosen by other APs Each AP knows its rank in the sequence Enables back to back TDMA like transmission Self Backoff Other’s Backoff 12 47 Rank in TDMA: 3 AP1

frequency domain backoff? Is there a benefit with frequency domain backoff? 1500 bytes at 54Mbps ~ 250 micro sec. Avg. temporal backoff ~ 100 micro sec. Big overhead Frequency backoff = 1 OFDM symbol = 4 micro sec 4 microsecond overhead  dramatic performance enhancement relative to 100 microsecond overhead

Will APs Collide During Contention?  Introduce a second round of contention  Winners of first go to second Subcarrier 1 2 3 First Round 4 5

Will APs Collide During Contention?  Introduce a second round of contention  Winners of first go to second Subcarrier 1 2 3 First Round 4 5 Subcarrier 1 2 3 Second Round 4 5

Only a Few APs in Second Round? TDMA will not be effective

 Instead of only winners, a few more APs to second round Optimize for TDMA  Instead of only winners, a few more APs to second round Subcarrier 1 2 3 First Round 4 5 Subcarrier 1 2 3 Second Round 4 5

 Instead of only winners, a few more APs to second round Optimize for TDMA  Instead of only winners, a few more APs to second round 1 2 3 First Round 4 5 0 2 4 0 2 4 0 2 4 Enabling TDMA Rank 1 Rank 2 Rank 3 Subcarrier 1 2 3 Second Round 4 5

Improved Channel Utilization 0 2 4 0 2 4 0 2 4 Enabling TDMA Rank 1 Rank 2 Rank 3 Subcarrier 1 2 3 4 5

Improved Channel Utilization 0 2 4 0 2 4 0 2 4 Enabling TDMA Rank 1 Rank 2 Rank 3 Subcarrier 1 2 3 4 5 Data/ACK Data/ACK Data/ACK .......... Time 802.11: Contention per packet Frequency Backoff TDMA Data/ACK .......... Time Back2F: Contention per TDMA Schedule

Multiple Collision Domains  Does Back2F work with real-world scattered APs?

Multiple Collision Domains  Does Back2F work with real-world scattered APs? BO = 5 BO = 2 BO = 3 BO = 1 Blue waits for Purple, but Purple waits for Green But Blue and Green should transmit simultaneously  Lost transmission opportunity  However 802.11 does not suffer from this problem Blue will wait for DIFS, continue counting down and eventually transmit

Multiple Collision Domains  Does Back2F work with real-world scattered APs? BO = 5 BO = 2 BO = 3 BO = 1  Blue waits for Purple, but Purple waits for Green  But Blue and Green should transmit simultaneously  Lost transmission opportunity Back2F Solution: Emulate 802.11  However 802.11 does not suffer from this problem Blue will wait for DIFS, continue counting down and eventually transmit

Multiple Collision Domains BO = 5 BO = 2 BO = 3 BO = 1 Frequency Backoff G -> 1 R -> 5 4 P -> 2 DIFS Time B -> 3 2 DIFS Reduce BO by winner’s BO Channel idle > DIFS Frequency Backoff, Wait for turn Frequency Backoff My turn Transmit

Back2F: Performance Evaluation Three important questions: Can Back2F detect subcarriers reliably? What is Back2F’s collision probability? How much throughput gain over 802.11?

Back2F: Performance Evaluation  Three important questions:  Can Back2F detect subcarriers reliably?  Evaluated on USRP/Gnuradio What is Back2F’s collision probability? How much throughput gain over 802.11?  Evaluated using traces at 65 locations

Back2F: Performance Evaluation  Three important questions:  Can Back2F detect subcarriers reliably?  Evaluated on USRP/Gnuradio Practical Challenge: High Self Signal Transmit Antenna Listen Antenna

Self Signal Overflows into Adjacent Subcarrier 60 50 40 B n 30 NR 20 10 60 64 pt. FFT 64 pt. FFT 50 Self Signal 40 30 20 10 S SNR in dB d i 10 10 20 30 40 50 60 Subcarrier Number 20 30 40 Subcarrier Number 50 60

Solution: Use a Higher Point FFT 60 60 50 40 B n 30 NR 20 10 128 pt. FFT 128 pt. FFT 50 Self Signal 40 30 20 10 S SNR in dB d i 10 10 20 30 40 50 60 Subcarrier Number 20 30 40 Subcarrier Number 50 60

Solution: Use a Higher Point FFT 60 60 50 40 B n 30 NR 20 10 256 pt. FFT 256 pt. FFT 50 Self Signal 40 30 20 10 S SNR in dB d i 10 10 20 30 40 50 60 Subcarrier Number 20 30 40 Subcarrier Number 50 60

Subcarrier Detection Performance 1 0.8 0.6 0.4 0.2 8dB SNR 10dB SNR 12dB SNR 14dB SNR Detection Accuracy 0 1 2 3 4 5 6 7 8 Distance in Subcarriers Robust subcarrier detection at 14dB Reliable subcarrier detection at 14dB

Back2F: Performance Evaluation Collect traces to answer: What is Back2F’s collision probability? How much throughput gain over 802.11 AP Client 20 AP locations 45 client locations

Back2F: Performance Evaluation Collect traces to answer: What is Back2F’s collision probability? How much throughput gain over 802.11 RSSI Per Subcarrier SNR Optimal Bitrate Traffic pattern Collision Probability Emulate 802.11, Back2F for various topologies

Back2F: Collision Probability 0.7 Back2F: 2 rounds Back2F: single round 802.11 Collision Probability(%) 0.6 0.5 Benefit of second round 0.4 0.3 0.2 0.1 5 10 15 20 25 30 35 40 45 50 Number of APs Small collision probability in dense networks

Throughput Evaluation 1 0.8 0.6 0.4 0.2 HD streaming Skype traffic Web browsing CDF 0.1 0.2 0.3 0.4 0.5 0.6 Throughput gain over 802.11 Higher throughput gain for real time traffic 36

Throughput Evaluation Throughput gain (%) over 802.11 0.6 0.5 0.4 0.3 0.2 0.1 5 10 15 20 25 30 35 40 45 50 Number of clients 6 Mbps 8 Mbps 4 Mbps /o batch 1 3 5 54 Mbps w Throughput Gain of Upto 50%

Limitation and Discussion Robustness of subcarrier based backoff Back2F more sensitive to channel fluctuation Need for additional antenna Back2F is complementary to MIMO Gain over packet aggregation Aggregation may not be possible for real time traffic Back2F provides gain with aggregation at higher rates Interoperability with 802.11 May interoperate but will cause unfairness

Summary  Using OFDM subcarriers Randomization is an effective method for contention resolution 802.11 time domain backoff requires channel to remain idle Observation: randomization possible in frequency domain  Using OFDM subcarriers Back2F: practical system realizing frequency domain contention Prevents collisions, provides upto 50% improvement in throughput

Duke SyNRG Research Group Questions, comments? Thank you Duke SyNRG Research Group http://synrg.ee.duke.edu