Interference Alignment By Motion

Swarun Kumar

Fadel Adib, Omid Aryan, Shyamnath Gollakota and Dina Katabi

Major Advances in MIMO

E.g. Interference Alignment

Significant gains in throughput

Single-Antenna Devices

Single Antennas, due to limits on power and sizeLargely left out of these MIMO benefits

Bring MIMO Benefits to Single Antenna Devices

“Interference Alignment”

Goal

Interference Alignment

antenna 1

C1

C2 C3

antenna 2

AP 1

C1

interfere interfere

2-antenna node can decode only 2 signals

1 2

C2 C3

Interference Alignment

antenna 1

C1

C2 C3

antenna 2

AP 1

C1

interfere interfere

1 2

C2 C3

2-antenna node can decode only 2 signals

Interference Alignment

antenna 1

C1

C2 C3

antenna 2

AP 1

C1

interfere interfere

1 2

C2 C3

“align”

2-antenna node can decode only 2 signals

Interference Alignment

antenna 1

C1

C2

C3

antenna 2

AP 1

C1

interfere interfere

1 2

C2 C3

“align”

2-antenna node can decode only 2 signals

Interference Alignment

antenna 1

C1 antenna 2

C1

interfere interfere

1

C2 C3

one unwanted interferer

2

AP 1

“align”

2-antenna node can decode only 2 signals

Single-Antenna Devices

C1

interfere interfere

1

C2 C3

antenna 1

C1

C2 C3

antenna 2

Can we still perform interference alignment?

Signals from all clients will change

2

AP 1

Single-Antenna Devices

C1

interfere interfere

1

C2 C3

antenna 1

C1

C2 C3

antenna 2Signals from all clients will change

2

AP 1

Perform Interference Alignment purely at the AP• Eliminates feedback/cooperation with clients• Brings benefits of alignment to new devices

Can we still perform interference alignment?

MoMIMO

• Moves the AP’s antenna to positions that achieve interference alignment

• Needs to only displace antenna by up to 2 inches

• Achieves 1.98x gain in throughput over 802.11n

1. How do we “find” positions of alignment?

2. How does it impact general wireless networks?

Feasibility of “Alignment by Motion”

Record antenna displacement for interference to drop below noise

AP 12 inch radius

21

interferedesired

C1 C2 C3

Feasibility of “Alignment by Motion”

Feasibility of “Alignment by Motion”

Why is the required displacement small?

Median: 0.3 inch

90th Percentile: 1 inch

A Simple Example

antenna 1

C1

ante

nna

2align ReferenceReference

C1

1 2AP 1

A Simple Example

antenna 10align ReferenceReference

C1

1 2AP 1

Goal: Minimize signal from C1 to antenna 2

C1 ante

nna

2

• Paths combine constructively or destructively based on phase

Indoor Environments Rich in MultipathHigh signal @2

(poor alignment)

C1

1 2AP 1

Indoor Environments Rich in Multipath

Paths differ by extra 2”

• Paths combine constructively or destructively based on phase

• For Wi-Fi, 2” ≈ λ/2

λ0° 360°C1

1 2AP 1

Indoor Environments Rich in Multipath

• Paths combine constructively or destructively based on phase

• For Wi-Fi, 2” ≈ λ/2

• In-phase paths now out-of-phase!

Low Signal @2 (good alignment)

0° 180°λ2C1

1 2AP 1

Paths differ by extra 2”

• Small displacement suffices for alignment• Generalizes to many reflectors, any alignment

{{

How Can We Find Good Alignment?We must quantify goodness of alignment

antenna 1

C1

C2

antenna 2

interference

C1

C2 interference

Poor

GoodC1

C2 interference ≈ 0

Goal: Find antenna location that minimizes interference

Naïve solution: Random walk

• Simulated the spatial profile of interference

• Ten reflectors placed in randomly chosen locations

• Applied standard multipath models

Does not work!

30

20

10

0

-10 Inte

rfere

nce

(dB)

Naïve solution: Random walk

High interference

x (in)

y (in

)

3

1

2

0

-1

-2

-332

1

-1-2-3 0

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10 Inte

rfere

nce

(dB)

Naïve solution: Random walk

Low interference

Goal: Find blue spotsy

(in)

x (in)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Naïve solution: Random walk

x (in)

y (in

)

Blue spots of low interference are small Hard to stumble upon in a random walk

Goal: Find blue spots

Inte

rfere

nce

(dB)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Key Observation: Interference is smooth

• Wireless channels are continuous and smooth functions over space

x (in)

y (in

)

Inte

rfere

nce

(dB)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Solution: A Hill Climbing Algorithm• Move in random direction and track interference

x (in)

y (in

)

Inte

rfere

nce

(dB)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Solution: A Hill Climbing Algorithm• Move in random direction and track interference– If interference : continue in that direction

x (in)

y (in

)

Inte

rfere

nce

(dB)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Solution: A Hill Climbing Algorithm• Move in random direction and track interference – If interference : continue in that direction

x (in)

y (in

)

Inte

rfere

nce

(dB)

3

1

2

0

-1

-2

-332

1

0-1-2-3

30

20

10

0

-10

Solution: A Hill Climbing Algorithm• Move in random direction and track interference– If interference : continue in that direction– If interference : continue in opposite direction

x (in)

y (in

)

Inte

rfere

nce

(dB)

Algorithm converges to spot of minimum interferenceGuides antenna to find positions of alignment

1. How do we “find” positions of alignment?

2. How does it impact general wireless networks?

Interference Alignment

AP 1

C1 C2 C3

AP 2 AP 3

AlignC2 and C3

Interference Alignment

AP 1

C1 C2 C3

AP 3

AlignC1 and C3

AP 2

Interference Alignment

AP 1

C1 C2 C3

AlignC1 and C2

AP 2 AP 3

Interference Alignment

AP 1

C1 C2 C3

AP 2 AP 3

• 3 concurrent streams Gain in throughput!N antenna APs enable N+1 concurrent uplink streams

What about downlink traffic?

AP 1

C1 C2 C3

AP 2 AP 3

What about downlink traffic?

AP 1

C2 C3

AP 1 has 2 antennas

null ??

2 antenna node can null interference at up to 1 antenna

Nothing!

C2 & C3 alignedat AP 1

AP 1

C2 C3

AP 1 has 2 antennas

null

2 antenna node can null interference at up to 1 antennaC2 & C3 aligned

at AP 1AP 1

C2C3

null for free!

Uplink Wireless Channels

AP 1

C2 C3

h1

h2 h3 h4 antenna 1an

tenn

a 2 (h1, h2)

(h3, h4)

h1

h2

h3

h4=

Downlink Wireless Channels

AP 1

C2 C3

h1

h2 h3 h4

Channel Reciprocityx

h1x + h2αx

null

αx

Downlink Wireless Channels

AP 1

C2 C3

h1

h2 h3 h4

Channel Reciprocityx αx

h1x + h2αx = 0

null

Downlink Wireless Channels

AP 1

C2 C3

h1

h2 h3 h4

Channel Reciprocityx αx

α =

null

-h1

h2

α = -h3

h4

h1

h2

h3

h4=

null

Alignment on the uplink enables nulling on the downlink, with no extra movement

Downlink Traffic

AP 1

C1 C2 C3

AP 2 AP 3

Downlink Traffic

AP 1

C1 C2 C3

AP 2 AP 3

Downlink Traffic

AP 1

C1 C2 C3

AP 2 AP 3

Downlink Traffic

AP 1

C1 C2 C3

AP 2 AP 3

• 3 concurrent streams on the downlinkMoMIMO provides gains to uplink & downlink traffic

Experimental Results

MoMIMO Implementation

• Implemented on USRP N210

• Mounted antenna on Roomba to emulate sliding antennas

• Compare MoMIMO with 802.11n, n+

TestbedRandomly assign nodes to red locations

Class RoomOffice Space

Can Alignment Reduce Interference?CD

F

Interference (dB)

Can Alignment Reduce Interference?CD

F

802.11nMoMIMODownlink

Median: -2.5dB

Interference (dB)

ThroughputHeterogeneous mix of 1 & 2-antenna nodes

CDF

Network Throughput (Mbps)

ThroughputHeterogeneous mix of 1 & 2-antenna nodes

1.98x

CDF

Network Throughput (Mbps)

802.11nMoMIMO

ThroughputHeterogeneous mix of 1 & 2-antenna nodes

1.31xCDF

Network Throughput (Mbps)

802.11n

n+MoMIMO

Conclusion

• Performs Interference Alignment purely by moving an antenna of the AP

• Displaces antenna by up to 2 inches

• New applications at intersection of networkingand robotics