AirShare Omid Abari Hariharan Rahul, Dina Katabi and Mondira Pant Distributed Coherent Transmission Made Seamless.

AirShare

Omid Abari Hariharan Rahul, Dina Katabi and Mondira Pant

Distributed Coherent Transmission Made Seamless

Wireless networks are getting denser and denser

Smart Homes

Factories

Wi-Fi connections

But, wireless spectrum is limited

Distributed cooperative protocols

Distributed MIMO

Distributed Modulation

Distributed lattice coding

Noisy network coding

Distributed compressed

sensing

higher throughput & higher efficiency

Protocols assume: Wireless nodes transmit at exactly the same frequency

Reality: Nodes have small offsets in their frequencies

Carrier Frequency Offset (CFO)

I

Q

Transmissions from different nodes rotate relative to each other

Independent wireless nodes have slightly different carrier frequencies

Ideal

Reality

Multiple wireless nodes transmit concurrently

I

Q

Reference Signal

Wireless Node

Clock

Root Cause of CFO is Clock

Each node uses its own clock as a reference

Reference Signal

Radio

Wireless Node

Clock

Carrier Signal2.4 GHz

10 MHz

Radio

Wireless Node

ClockRadio

Wireless Node

Clock

Root Cause of CFO is Clock

Each node uses its own clock as a reference

Carrier Signal2.4 GHz+240Hz

10 MHz +1Hz

Carrier Signal2.4 GHz-720Hz

10 MHz -3Hz

Crystals have slightly different frequencies

Different nodes have offset in their carrier frequency (CFO)which varies over time

How can we eliminate CFO?

Naïve Solution

Connect all nodes to a shared reference clock

Defeats the notion of a wireless network

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

ClockCarrier Signal

Our IdeaTransmit a reference over-the-air

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

Radio

Wireless Node

Clock

AirSharetransmits the reference clock over the air

&eliminates CFO

Protocol independent

Cheap and Low-Power

Supports mobility

Clock Emitter

AirShare ArchitectureEmitter

Recipient

Clock Emitter

ChallengesEmitter

Recipient

Challenges

RecipientHow can we build a cheap and

low-power recipient?

Clock Emitter

Emitter

How can emitter transmit a clock?

Clock Emitter

Emitter

How can emitter transmit a clock?

Problem: Reference clocks are typically 10-40 MHz

- FCC forbids transmitting such a low-frequency signal

- Requires large antennas

Transmits two signals separated by the clock frequency

f1 f2

(s∈(2𝜋 𝑓 1𝑡 )+ s∈(2𝜋 𝑓 2 𝑡 ) )

Transmit a Differential-reference

Instead of transmitting a signal at the clock frequency (10 MHz)

fref = 10 MHz

Clock Emitter

Emittersin (2𝜋 𝑓 𝑟𝑒𝑓 𝑡 )

fref

10 MHz

f2 , f1 = any frequency

Recipient f1 f2 fref

Transmits two signals separated by the clock frequency

f1 f2

(s∈(2𝜋 𝑓 1𝑡 )+ s∈(2𝜋 𝑓 2 𝑡 ) )

Transmit a Differential-reference

Instead of transmitting a signal at the clock frequency (10 MHz)

fref = 10 MHz

Clock Emitter

Emittersin (2𝜋 𝑓 𝑟𝑒𝑓 𝑡 )

fref

10 MHz

f2 , f1 = any frequency

Recipient f1 f2 fref

?

Recipient f1 f2 fref

?

Using trigonometric identities:

(s∈(2𝜋 𝑓 1𝑡 )+ s∈(2𝜋 𝑓 2 𝑡 ) )× (sin (2𝜋 𝑓 1𝑡 )+sin (2𝜋 𝑓 2 𝑡 ) )f1 f2

f1 f2 f1 f2

receives the signal and multiplies the signal by itself

DC 2f1 2f2f2-f1 f1+f2f2-f1 f1+f2= 10 MHz

AirShare transmits the reference clock without violating FCC regulations

Clock Emitter

ChallengesEmitter

RecipientHow can we build a cheap and

low-power recipient?

How can emitter transmit a clock?Transmit a Differential-reference

Use Passive Architecture

BandPassFilter Reference

Signal

LNA

Mixer

Wireless Node

Simple, passive, off-the-shelf components Cheap and Low-power

Passive Passive

f1 f2f1 f2f1 f2 f2-f1DC 2f1 f1+f2 2f2f2-f1DC 2f1 f1+f2 2f2

• Low power consumption: • < 10% for wireless sensors• < 0.1% for Wi-Fi APs

• Low cost:• Off-the-shelf components• Costs only a few dollars

Our AirShare Prototype

Antenna

We built a prototype of recipient in a custom designed PCB

Evaluation

• Implemented AirShare using off-the-shelf components

• Evaluated AirShare in an indoor testbed using USRPs

• Evaluated two applications:– Distributed Rate Adaptation– Distributed MIMO

CFO(Hz)

CD

F

0.01 0.1 1 10 100 1000 100000

0.2

0.4

0.6

0.8

1

Today Nodes

CFO(Hz)

CD

F

0.01 0.1 1 10 100 1000 100000

0.2

0.4

0.6

0.8

1

CFO(Hz)

CD

F

0.01 0.1 1 10 100 1000 100000

0.2

0.4

0.6

0.8

1AirClock

Today Nodes

Synchronization Accuracy

AirShare reduces the CFO by multiple orders of magnitude

Measured CFO between nodes at 2.4 GHz– 500 Experiments– Different nodes and locations

2-3 orders of magnitude

Ideal Zone[sigcomm’12]

Solution: Distributed Rate Adaptation

multiple sensors transmit together higher throughput

Application 1: Distributed Rate Adaptation

Ideally: Better channel quality Higher throughput

I

Q2 31 N

Problem: Sensors support only single low data rate

Throughput Gain

• Data throughput for 6 sensors

Throughput gains of 1.6-3× over today sensors for 6 sensors

20-2516-2412-185-120

40

80

SNR (dB)

Th

rou

gh

pu

t (K

bp

s)TDMA

120

20-2516-2412-185-120

40

80

SNR (dB)

Th

rou

gh

pu

t (K

bp

s)

120

20-2516-2412-185-120

40

80

SNR (dB)

Th

rou

gh

pu

t (K

bp

s)

This WorkTDMA

120

Application 2: Distributed MIMO

Multiple APs transmit to multiple clients concurrently

Network throughput scales with the number of APs

1

1 2 N

2 N

Ethernet

Throughput GainDistributed MIMO network including 5 clients and 5 APs

Throughput gain of 4.4× over traditional 802.11 for 5 transmitters

2 3 4 51

2

3

4

5

Number of Receivers

Th

rou

gh

pu

t G

ain

2 3 4 51

2

3

4

5

Number of Receivers

Th

rou

gh

pu

t G

ain

Related Work• Using wires or power-lines to distribute a shared clock

[SenSys’09, SIGCOMM’14]

• Designing algorithms to estimate and correct for CFO [SIGCOMM’12, ToN’2013]

• Equip each node with a GPS disciplined oscillator [Trimble, Jackson Labs]

Cheap, Low-PowerProtocol independent

Supports mobility

AirShare

Conclusion

• Described AirShare, a simple method to eliminate CFO in wireless nodes

• Provides large throughput gains

• Enables many new applications such as distributed MIMO, distributed modulation, etc.