Overcoming the Antennas-Per-AP Throughput Limit in MIMO

Shyamnath Gollakota

Samuel David Perli and Dina Katabi

MIMO LANs

Today, MIMO delivers as many concurrent packets as the antennas on the AP

Talk presents a practical technique to double the concurrent packets in MIMO

LANs

MIMO Primer

11h

12h22h

21h1p

2p

APBob

111 ph

112 ph

112

11 phh

222

21 phh

Antenna 1

Antenna 2

hij is the channel from antenna i to antenna j

AP

1p

2p

AP receives the sum of these vectors

MIMO Primer

11h

12h22h

21h

112

11 phh

222

21 phh

Bob

AP

1p

2p

AP projects on a direction orthogonal to interference

p2

p1

How does the AP decode each packet?Current MIMO decodes as many concurrent packets as there are antennas

per AP

MIMO Primer

11h

12h22h

21h

112

11 phh

222

21 phh

Bob

Can We Get More Concurrent Packets?Bob AP

1p

2p

3p

p3

2p

1p p3

No direction is orthogonal to all interference AP can’t decode

All current MIMO LANs are limited by number of antennas-per-AP

1p

2p

Alice

Let the APs Coordinate Over the EthernetNaive solution: Emulate 4-antenna AP by sending every signal sample over Ethernet

Let the APs Coordinate Over the Ethernet

Impractical Overhead,

Naive solution: Emulate 4-antenna AP by sending every signal sample over Ethernet

1p

2p

3p

2p

1p p3

1p

2p

Ethe

rnet

Raw samples

E.g., a 3 or 4-antenna system needs 10’s of Gb/sCan we leverage the Ethernet with minimal overhead?

Ethe

rnet

p1

Bob AP11p

2p 2p

1p

3p

p3

2p

1p p3

• Align P3 with P2 at AP1• AP1 broadcasts P1 on Ethernet • AP2 subtracts/cancels P1 decodes P2, P3

p1

p2 p3

AP2Alice

AP1 decodes P1 to its bits

Interference Alignment and Cancellation (IAC)

• IAC overcomes the antennas-per-AP throughput limit• In IAC, a packet is decoded, then broadcasted once on the Ethernet minimal overhead•

Contributions• First MIMO LAN to overcome the antennas-per-AP

limit• IAC synthesizes interference alignment and

cancellation• Proved that IAC almost doubles MIMO throughput• Implemented IAC in software radios showing

practical throughput gains

How to Change Packet Direction?

How to Change Packet Direction?

101

pH

H h11 h12h21 h22

11hClient AP

12h22h

21h1p

2p

h11h21

p1

h12h22

p2210

pH

10

1p

H21

0p

21 6.4.

5.5.

pp

21 1

05.5.

pp

2211 pvpv

21

0pH

How to Change Packet Direction?

101

pH

Client AP

H

15.5.

pH

26.4.

pH

Sender controls packet direction by multiplying with a vector

11pvH

12 pvH

How Do We Align?Bob AP1 1111 pvH

11H

12H

21H

22H

2211 pvpv

33 pv

3321 pvH

2211 pvH

211321 vHvH

Alice AP2

How Does Alignment Work in Presence of Modulation?

Real

Imaginary

Modulated samples are complex numbers with different phases

Real

Imaginary

Sample in P3

Sample in P2

Alignment is in the antenna domain not the modulation domain

Antenna 1

Antenna 2

Alignment works independent of modulation phases

How Does AP2 Subtract Interference from P1?

Can’t subtract the bits in packet• Need to subtract interference signal as received by

AP2

Solution• AP2 Re-modulate P1’s bits• AP2 estimate and apply the channel P1

traversed to itself on modulated signal– Channel estimation in the presence of

interference as in [ZigZag, SIGCOMM’08]• Subtract!

How Does IAC Generalize to M-Antenna MIMO?

Theorem

In a M- antenna MIMO system, IAC delivers• 2M concurrent packets on uplink• max{2M-2, 3M/2} concurrent packets on downlink

How Does IAC Generalize to M-Antenna MIMO?

E.g., M=2 antennas 4 packets on uplink 3 packets on downlink

Theorem

In a M- antenna MIMO system, IAC delivers• 2M concurrent packets on uplink• max{2M-2, 3M/2} concurrent packets on downlink

How Does IAC Generalize to M-Antenna MIMO?

E.g., M=10 antennas 20 packets on uplink18 packets on downlinkFor a large M,

IAC doubles MIMO throughput

What if There is a Single Client?

ClientAP1

1p

2p

2p

1p 2p

1p

Current MIMO exploits diversity and pick best of two APs

Can’t have more than 2 concurrent packets, but …

IAC can pick the best antenna pair across APs

AP2

•IAC provides higher diversity than Current MIMO•Diversity gain applies to one or more clients

IAC MAC Leverages 802.11 PCF mode

• Clients are simple: APs compute v vectors and send them to clients in the Grant message

• IAC adapts to changing channels because APs get a new channel estimate from each ACK packet

CF- End

Contention-free Contention

Downlink

Uplink

. . . . .

. . . . . ACKs

Gran

t

P4 P5 P6

P1

P2 P3 Time

Performance

Implementation

• GNURadio software• 2-antenna MIMO USRP nodes• Carrier Freq: 2.4GHz

Testbed

• 20-node testbed• All nodes within radio

range of each other• Each run randomly

picks APs and clients

Gain = Client throughput in IACClient throughput in current MIMO

Metric

1 1.5 2 2.5 3 3.50

0.10.20.30.40.50.60.70.80.9

1Uplink Gain

CDF

of R

uns

Per-Client Throughput Gain

Uplink GainCD

F of

Run

s

Per-Client Throughput Gain • On uplink, IAC’s median gain is 2.1x• Gain is partially due to diversity but more to concurrency

Downlink GainCD

F of

Run

s

Per-Client Throughput Gain

On downlink, IAC’s median gain is 1.5x

Gains as a Function of SNR

Gains as a Function of SNR

SNR in dB

Upl

ink

Thro

ughp

ut G

ain

IAC is beneficial across the operational range of SNRs

Related Work• Interference Alignment [AMK’08,JS’08]• Interference Cancellation [GC’80,HWA’08]• MU-MIMO [NJ’06]

IAC provably provides more throughput, and doubles the number of concurrent packets

Conclusion

• First MIMO LAN to overcome the antennas-per-AP limit

• IAC synthesizes interference alignment and cancellation

• Proved that IAC almost doubles MIMO throughput• Implemented IAC in software radios showing that it

works in practice

IAC MAC Leverages 802.11 PCF mode

• APs compute and send v vectors in Grant Clients are oblivious to each other

• APs can track channels, i.e., H, from using ACKs

CF- End

Contention-free Contention

Downlink

Uplink

. . . . .

. . . . . ACKs

Gran

t

P4 P5 P6

P1

P2 P3 Time

Ethe

rnet

1p

2p

Uplink: for M=2 antennas, IAC delivers 2M=4 packets

Clients APs

3p

4p

p1

p2

1p

2p 3p 4p

1p

2p

3p 4p

3p 2p4p

1pp4p3

1pAPs Clients

2p

3p

p1

p2

1p

2p 3p

2p

1p 3p

3p2p 1p

p3

Downlink: - Clients can’t coordinate over Ethernet - For M=2 antennas, IAC delivers 3M/2 = 3

packets

IAC’s concurrency increases capacity bound

C = d log(SNR) + o(log(SNR))

IAC increases degrees of freedom

d is degrees of freedom

Interference cancellation does not increase degrees of freedom but provides a better use of them