Combating CrossTechnology Interference Shyamnath Gollakota Fadel Adib Dina

Combating Cross-Technology Interference Shyamnath Gollakota Fadel Adib Dina Katabi Srinivasan Seshan

ISM Band Is Increasingly Crowded Multiple independent studies [Cisco, Ofcom, Jupiter, Farpoint] • Most problems are from cross-technology high-power interferers • Responsible for more than 50% of the customer complaints • Lead to complete loss of connectivity Baby Monitors Microwave Ovens Cordless Phones

Experimental Setup • Two Netgear 802. 11 n devices • Baby monitors, cordless phones and microwave ovens • Wi. Fi devices about 20 feet away from each other • Move interferer 1 -90 feet away from Wi. Fi receiver t e e 0 f 2 Wi. Fi rx Wi. Fi tx

Effect of High-Power Interferers on Wi. Fi 80 60 i Throughput 40 (in Mbps) 20 0 0 1 2 3 4 5 6 7 Interferer Location # 1 foot Line of sight 8 9 10 90 feet Non- Line of sight

Effect of High-Power Interferers on Wi. Fi 80 Without Interferers 60 i Throughput 40 (in Mbps) With Microwave 20 0 With Cordless Phone 0 1 2 With baby Monitor 3 4 5 6 7 Interferer Location # 1 foot Line of sight 8 9 10 90 feet Non- Line of sight

Traditional Solutions to Cross Technology Interference Don’t Work • Avoid interferer frequencies Ø Much wider bandwidth than Wi. Fi Ø Interferer can occupy multiple Wi. Fi channels

Traditional Solutions to Cross Technology Interference Don’t Work • Avoid interferer frequencies Ø Much wider bandwidth than Wi. Fi Ø Interferer can occupy multiple Wi. Fi channels • Treat interferer as noise and use lower rate Ø High power interferers (e. g. , 8 -100 X Wi. Fi power) Ø Can’t get even lowest Wi. Fi rate How can we deal with such high-power interference?

Technology Independent Multiple Output (TIMO) • First Wi. Fi receiver that decodes in presence of high-power cross -technology interferers • Is agnostic to the interferer’s technology • Implemented and evaluated with baby monitors, microwave ovens and cordless phones Ø Convert no-connectivity scenarios to operational networks

Idea: Try to leverage MIMO Client AP Today, streams are of the same technology

Idea: Try to leverage MIMO Client AP If MIMO can work across diverse technologies

Idea: Try to leverage MIMO Client AP If MIMO can work across diverse technologies Challenge: Current MIMO doesn’t work with diverse technologies

MIMO Primer Client AP How do current APs estimate the channels? Client sends a known the twoto antennas If • channels are known, APpreamble can solve on equations decode the • AP correlates with known preamble to estimate channels two streams, S 1 and S 2 • Doesn’t work across technologies

Say, Interferer is One of the Streams Client AP But, AP doesn’t know interferer technology / preamble Can’t compute interferer channels, h 3 and h 4

Fundamental Limitation of Channel Estimation Scenario 1 Channel Interference Scenario 2 Channel Interference Can’t distinguish between the two scenario Impossible to exactly estimate interferer channels

How Does TIMO Work? Client AP AP is not interested in decoding baby monitor • Reduce the number of unknowns to three

How Does TIMO Work? Client AP AP is not interested in decoding baby monitor • Reduce the number of unknowns to three

How Does TIMO Work? Client AP AP is not interested in decoding baby monitor • Reduce the number of unknowns to three • β is the interferer channel ratio

How Does TIMO Work? Client AP AP is not interested in decoding baby monitor • Reduce the number of unknowns to three • β is the interferer channel ratio • Focus on channel ratio instead of channels

Getting Around the Fundamental Limitation Scenario 1 Channel Interference Scenario 2 Channel Interference The scaling factor, c, introduces ambiguity into channels Unlike channels, the channel ratio is not ambiguous

If β Can be Computed, AP Can Decode Wi. Fi Client AP AP can solve the two equations to decode the Wi. Fi client

Question: How do we compute β? Answer: Send known symbol • Wi. Fi client sends known symbol at beginning of its packet

Question: How do we compute β? Answer: Send known symbol Known • Wi. Fi client sends known symbol at beginning of its packet • Solve equations to get β • Once β is known, it can be used to decode subsequent symbols

Question: How do we compute β? Answer: Send known symbol Time Known symbol But, Use what if interferer is concentrated in time β to decode subsequent symbols

Question: How do we compute β? Answer: Send known symbol Time Known symbol if interferer is concentrated in time We. But, havewhat a solution to compute β without known symbols

Intuition: Exploit the Wi. Fi Symbol Structure • BPSK – ‘ 1’ bit sent as +1 and ‘ 0’ bit sent as -1 Imaginary -1 +1 Real

Intuition: Exploit the Wi. Fi Symbol Structure • BPSK – ‘ 1’ bit sent as +1 and ‘ 0’ bit sent as -1 • If no interference, received symbols are close to expected symbols Imaginary -1 +1 Real

Intuition: Exploit the Wi. Fi Symbol Structure • BPSK – ‘ 1’ bit sent as +1 and ‘ 0’ bit sent as -1 • If no interference, received symbols are close to expected symbols • If interference, received symbols are far from expected symbols Imaginary correct Error -1 +1 Real Correct estimate Average error is small

Intuition: Exploit the Wi. Fi Symbol Structure • BPSK – ‘ 1’ bit sent as +1 and ‘ 0’ bit sent as -1 • If no interference, received symbols are close to expected symbols • If interference, received symbols are far from expected symbols Imaginary correct +1 -1 Error Real guess 1 Bad estimate Average error is big

Intuition: Exploit the Wi. Fi Symbol Structure • BPSK – ‘ 1’ bit sent as +1 and ‘ 0’ bit sent as -1 • If no interference, received symbols are close to expected symbols • If interference, received symbols are far from expected symbols Imaginary correct -1 Error • • +1 Real guess 2 guess 1 Better Estimate Average error reduce Design gradient descent style algorithm to iteratively converge to actual channel ratio Paper described algorithm that works across modulations

Performance

Implementation • Implement using USRP 2 s • Wi. Fi modulations and coding rates • OFDM over 10 MHz • Bits rates between 3 -27 Mbps • No carrier sense

Testbed • Place USRP prototype for 802. 11 at blue locations • Change the location of interferer over red locations Tx Rx

Throughput Performance with Baby Monitor 25 20 15 11 Throughput (in Mbps)10 Wi. Fi 5 0 1 2 1 foot Line of sight 3 4 5 6 7 8 Interferer Location # Non- Line of sight 9 10 90 feet

Throughput Performance with Baby Monitor 25 20 15 11 Throughput (in Mbps)10 60 feet away USRP Wi. Fi 5 0 Wi. Fi 1 2 3 4 5 6 7 8 9 10 Interferer Location # foot 90 feet Despite 1 disabling carrier sense, complete loss of connectivity in more than half the. Line location Line of sight Nonof sight

Throughput Performance with Baby Monitor Without interference 25 20 USRP Wi. Fi with TIMO 15 11 Throughput (in Mbps)10 USRP Wi. Fi 5 0 Wi. Fi 1 2 1 foot Line of sight 3 4 5 6 7 8 Interferer Location # Non- Line of sight 9 10 90 feet

Throughput Performance Cordless Phones 25 Microwave Ovens 25 with TIMO 20 20 15 15 oughput (in Mbps) 10 with TIMO 802. 11 Throughput (in Mbps) w/o TIMO 10 5 5 0 0 w/o TIMO 1 3 5 7 9 Interferer Location scenarios # Interferer Location TIMO transforms with a complete loss of # connectivity to operational networks

Related Work • Decoding Interference [IC, SAM, Beamforming, …] - Don’t work with cross-technology interference • Cognitive Communication [Samplewidth, Jello, Swift, …] - Don’t operate on the same frequency First system to decode in the presence of cross-technology interference on same band

Conclusions • First Wi. Fi receiver that decodes in presence of high-power cross-technology interferers • Enable MIMO to work across technologies • Implemented and evaluated with baby monitors, microwave ovens and cordless phones Ø Convert no-connectivity scenarios to operational networks