Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control...

42
Andrea Goldsmith Stanford University Stanford Networking Seminar May 28, 2009

Transcript of Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control...

Page 1: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Andrea Goldsmith

Stanford University

Stanford Networking gSeminar

May 28, 2009

Page 2: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Future Wireless NetworksFuture Wireless NetworksUbiquitous Communication Among People and Devices

Next‐generation CellularWireless Internet AccessWireless MultimediaSensor Networks Smart Homes/SpacesAutomated HighwaysIn‐Body NetworksAll this and more …

Page 3: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

F t C ll PhFuture Cell PhonesSan Francisco

Everything wireless in one deviceBurden for this performance is on the backbone network

BSBS

San Francisco

BSBS

Internet

PhoneSystem

New YorkNth-GenCellular

Nth-GenCellular

BS

Much better performance and reliability than today‐ Gbps rates, low latency, 99% coverage indoors and out 

Page 4: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Future Wifi:Future Wifi:Multimedia Everywhere, Without WiresPerformance burden also on the (mesh) network

802.11n++

• Streaming video• Gbps data rates

Wireless HDTVand Gaming

• Gbps data rates• High reliability• Coverage in every room

Page 5: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

D i Ch llDevice ChallengesBT

Size and CostCellular

GPS

FM/XM

Multiband AntennasMultiradio Coexistance Apps WLAN

DVB-H

Multiradio CoexistanceIntegration

ppProcessor

MediaProcessor

WLAN

Wimax

Page 6: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Software Defined Radio:Software‐Defined Radio: Is this the solution to the device challenges?

Cellular

BT

GPS

FM/XM A/D

A/D

AppsProcessor

M di

WLAN

DVB-HA/D

DSPA/D

Wideband antennas and A/Ds span BW of desired signals

MediaProcessor Wimax A/D

Wideband antennas and A/Ds span BW of desired signalsDSP is programmed to process the desired signal based on carrier frequency, signal shape, channel characteristics, etc.Avoids specialized hardware

Today, this is not cost, size, or power efficient

Page 7: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

D i S l tiDevice Solutions Silicon evolution will reduce size and powerp

130nm→95nm→65nm→45nm→ 32nm→…

Circuit design not anticipatedCircuit design  not anticipatedCMOS PA efficiency and power will improveA/D technology will improveWideband antenna design will improveWideband antenna design will improveTools for digital design will improve

Room for innovation at the RF/baseband interfaceRoom for innovation at the RF/baseband interface

Dedicated silicon will remain faster, cheaper, and lower power than processor‐based designs

But less flexible and with more costly development

Page 8: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

S t Ch llSystem Challenges

Managing interferenceManaging interferenceReliabilityHigh‐bandwidth applicationsHigh‐bandwidth applicationsScarce spectrumReal‐time constraintsReal time constraintsUbiquitous coverage indoors and out

Page 9: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

System SolutionsSystem SolutionsBetter link layer design

Low‐complexity OFDM and MIMO (PHY wars are over)High‐performance modulation and coding Adaptive techniques (in time space and frequency)Adaptive techniques (in time, space, and frequency)

Better MAC and networking techniquesMore efficient use of wireless spectrumMore efficient use of wireless spectrum

RelayingPicocells and FemtocellsPicocells and FemtocellsCooperation and Cognition

Cross‐Layer DesignCross Layer Design

Much room for improvement and innovation

Page 10: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

MIMO in Wireless NetworksMIMO in Wireless Networks

How should MIMO be fully exploited?How should MIMO be fully exploited?At a base station or Wifi access point

MIMO Broadcasting and Multiple AccessMIMO Broadcasting and Multiple Access

Network MIMO: Form virtual antenna arraysDownlink is a MIMO BC, uplink is a MIMO MACCan treat “interference” as a known signal or noiseCan cluster cells and cooperate between clusters

Page 11: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Multiplexing/diversity/interferenceMultiplexing/diversity/interference cancellation tradeoffs in MIMO networks

Stream 2

Interference

Stream 1

Spatial multiplexing provides for multiple data streamsSpatial multiplexing provides for multiple data streamsTX beamforming and RX diversity provide robustness to fadingTX beamforming and RX nulling cancel interferenceg g

Optimal use of antennas in wireless networks unknown

Page 12: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

MIMO Receiver ComplexityReceiver Complexity is a problem

It affects design time, size, cost, battery life, etc. Complexity Exponential in Constellation Size/Antenna No.

For a full MAP RX

C∝NI ⋅ NT⋅2log2(M)xNNI : No. RX IterationsNT: No. OFDM TonesM: Constellation Size

Reduced complexity receiver options:(Iterative) MMSE Spherical‐decoders M‐Algorithm etc

N: No. Antennas

(Iterative) MMSE, Spherical‐decoders, M‐Algorithm, etc.Performance/complexity tradeoffs depend on N and M

For 64QAM, 211tones, 6 antennas: C NI×211×236

Key area for innovation

Page 13: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Coverage Indoors and Out:Coverage Indoors and Out:Cellular versus Mesh

FemtocellOutdoors Indoors

Femtocell

Wifi MeshCellular has good coverage outdoors

Relaying increases reliability and range 

Wifi mesh has a nichemarket o tdoors

Cellular cannot provide reliable indoor coverageWifi t k l dWifi mesh has a nichemarket outdoors

Hotspots/picocells enhance coverage, reliability, and data rates. 

Wifi networks already ubiquitous in the homeAlternative is a consumer‐installed Femtocell

Multiple frequencies can be leveraged to avoid interference

installed FemtocellWinning solution will depend on many factors

Page 14: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

S Wi l S tScarce Wireless Spectrum

$$$$$$

and Expensive

Page 15: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

S t l RSpectral ReuseDue to its scarcity, spectrum is reusedy, p

In licensed bands and unlicensed bands

BS

Cellular, Wimax Wifi, BT, UWB,…

Page 16: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

I t f F i d F ?Interference: Friend or Foe?If treated as noise: FoeIf treated as noise: Foe

PSNR =IN

SNR+

Increases BER, reduces capacity

If decodable: Neither friend nor foe

Increases BER, reduces capacity

If decodable: Neither friend nor foe

Multiuser detection can completely remove interference

Page 17: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Ideal Multiuser DetectionSignal 1

Signal 1

- =Signal 1

Demod

IterativeMultiuserDetection

Signal 2Demod

Signal 2

- =

Why Not Ubiquitous Today?  Power and A/D Precision 

Page 18: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

I t f F i d F ?Interference: Friend or Foe?

If exploited via cooperation and cognitioncooperation and cognition

Friend

Especially in a network settingEspecially in a network setting

Page 19: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Cooperation in Wireless NetworksCooperation in Wireless Networks

Many possible cooperation strategies:Virtual MIMO , generalized relaying, interferenceVirtual MIMO , generalized relaying, interference forwarding, and one‐shot/iterative conferencing

Many theoretical and practice issues:Many theoretical and practice issues:Overhead, forming groups, dynamics, synch, …

Page 20: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

C i i h C i MIMOCapacity with Cooperative MIMO 

2 2 MIMO B dTX1

G

RX12x2 MIMO Bound

2x1 Broadcast BoundTX Coop Only

GG

TX2 RX2

TX Coop OnlyTX/RX Coop

RX Coop OnlyTime

Division

TX cooperation needs large cooperativeTX cooperation needs large cooperative channel gain to approach broadcast bound2x2 MIMO bound unapproachable

Page 21: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

General Relay StrategiesTX1 RX1TX1

relay

X1

Y3=X1+X2+Z3

Y4=X1+X2+X3+Z4

X3= f(Y3)

TX2 RX2X2

Y5=X1+X2+X3+Z5

Can forward message and/or interferenceR l f d ll f h

RX2

Relay can forward all or part of the messages Much room for innovation

Relay can forward interferenceRelay can forward interference To help subtract it out

Page 22: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

B fi i l t f d b thBeneficial to forward bothinterference and message

Page 23: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Intelligence beyond Cooperation:Intelligence beyond Cooperation: CognitionC i i di i l iCognitive radios can support new wireless users in existing crowded spectrum

Without degrading performance of existing usersWithout degrading performance of existing users

Utilize advanced communication and signal processing techniquestechniques

Coupled with novel spectrum allocation policies

Technology could Revolutionize the way spectrum is allocated worldwide 

Provide sufficient bandwidth to support higher quality and higher data rate products and services

Page 24: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Cognitive Radio ParadigmsUnderlay

Cognitive radios constrained to cause minimal interference to noncognitive radiosinterference to noncognitive radios 

InterweaveCognitive radios find and exploit spectral holes toCognitive radios find and exploit spectral holes to avoid interfering with noncognitive radios

OverlayOverlayCognitive radios overhear and enhance noncognitive radio transmissions

Knowledgeand

Complexity

Page 25: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Underlay SystemsCognitive radios determine the interference theirCognitive radios determine the interference their transmission causes to noncognitive nodes

Transmit if interference below a given threshold

IPNCR

The interference constraint may be met

NCRNCR

CR CR

The interference constraint may be metVia wideband signalling to maintain interference below the noise floor (spread spectrum or UWB)the noise floor (spread spectrum or UWB)

Via multiple antennas and beamforming

Page 26: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

I t S tInterweave SystemsMeasurements indicate that even crowded spectrum is 

d ll i d f inot used across all time, space, and frequenciesOriginal motivation for “cognitive” radios (Mitola’00)

These holes can be used for communicationInterweave CRs periodically monitor spectrum for holesHole location must be agreed upon between TX and RXHole location must be agreed upon between TX and RXHole is then used for opportunistic communication with minimal interference to noncognitive users

Page 27: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Overlay Cognitive Systemsh k l d f hCognitive user has knowledge of other 

user’s message and/or encoding strategyUsed to help noncognitive transmission

Used to presubtract noncognitiveUsed to presubtract noncognitiveinterference

RX1CRCR

RX2NCR

Page 28: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

P f G iPerformance Gains from Cognitive Encoding

outer bound

our scheme

prior schemes

Only the CRtransmits

Page 29: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Crosslayer Protocol DesignCrosslayer Protocol Design

Application

NetworkNetwork

Access

Link

dHardware

Substantial gains in throughput efficiency andSubstantial gains in throughput, efficiency, and end-to-end performance from cross-layer design

Page 30: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Di it /M lti l i i MIMODiversity/Multiplexing in MIMO

Use antennas for multiplexing:Use antennas for multiplexing:

ST C d

f d

High-RateQuantizer

ST CodeHigh Rate Decoder

Error Prone

Use antennas for diversity 

Low-RateQuantizer

ST CodeHigh

DiversityDecoder

Low Pe

How should antennas be used? Depends on end-to-end metric.

Page 31: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

DMT i MIMOWi l N kDMT in MIMO Wireless Networks

3.5

4DMT of (4,1,3) half-duplex relay channel

d4,1(r)

2

2.5

3

y ga

in, d

(r)

,

d1,3

(r)

1

1.5

Div

ersi

ty

dDDF(r)

dvDF(r)

dfDF

(r), a=0.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.5

Multiplexing gain, (r)

vDF( )

Page 32: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Wh b I f C ll i ?What about Interference Cancellation?

b d f l l d• Antennas can be used for multiplexing, diversity, or interference cancellation

•Cancel M‐1 interferers with M antennas

• What metric should be optimized?p

Cross-Layer Design

Page 33: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Delay/Throughput/RobustnessDelay/Throughput/Robustness across Multiple Protocol Layers

BA

B

Multiple routes through the network can be used for l l d d d l /lmultiplexing or reduced delay/lossSpatial dimension of MIMO adds new degree of freedom

A li ti i l d i ti lti lApplication can use single‐description or multiple description codes

Can optimize optimal operating point for theseCan optimize optimal operating point for these tradeoffs to minimize distortion

Page 34: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Network Fundamental Limits

Network Metrics

Capacity Delay

Network Fundamental Limits

Fundamental Limitsof Wireless Systems

(DARPA Challenge Program)

Outage

(DARPA Challenge Program)

BC

Cross‐layer Design andE d t d P f

A

DCapacity

Delay

End‐to‐end Performance

(C*,D*,O*)

Research Areas‐ Fundamental performance limits      and tradeoffs

D

Robustness

and tradeoffs ‐ Node cooperation and cognition‐ Adaptive techniques‐ Layering and Cross‐layer design

Application Metrics

‐ Network/application interface‐ End‐to‐end performance optimization and guarantees

Page 35: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Wireless Sensor Networks

• Smart homes/buildingsSmart homes/buildings• Smart structures• Search and rescue• Homeland securityy• Event detection• Battlefield surveillance

Energy (transmit and processing) is the driving constraintData flows to centralized location (joint compression)Data flows to centralized location (joint compression)Low per‐node rates but tens to thousands of nodesIntelligence is in the network rather than in the devices

Page 36: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Cross Layer TradeoffsCross‐Layer Tradeoffs under Energy Constraints

HardwareAll nodes have transmit, sleep, and transient modesEach node can only send a finite number of bitsy

LinkHigh‐level modulation costs transmit energy but saves circuit energy (shorter transmission time)Coding costs circuit energy but saves transmit energy

AccessAccessPower control impacts connectivity and interferenceAdaptive modulation adds another degree of freedom

Routing:Circuit energy costs can preclude multihop routing

Page 37: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Mi i E R iMinimum Energy Routing

0 085

0.1 Red: hub nodeGreen: relay/source

4 3 2 10.1150.185

0.085 Green: relay/source

ppsRppsR

8060

2

1

==

(0,0) (5,0) (10,0) (15,0)

0.515 ppsRppsR

2080

3

2

=

• Optimal routing uses single and multiple hops

• Link adaptation yields additional 70% energy savings• Link adaptation yields additional 70% energy savings

Page 38: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Cooperative Compression

Source data correlated in space and time

Nodes should cooperate in compression as well as communication and routing

Joint source/channel/network coding

What is optimal: virtual MIMO vs. relaying

Page 39: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Distributed Control over WirelessDistributed Control over Wireless Automated Vehicles‐ Cars‐ Airplanes/UAVs‐ Insect flyers

Interdisciplinary design approach• Control requires fast, accurate, and reliable feedback.• Wireless networks introduce delay and loss• Wireless networks introduce delay and loss• Need reliable networks and  robust controllers • Mostly open problems: Many design challenges

Page 40: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Wireless Biomedical SystemsWireless Biomedical SystemsWireless 

TelemedicineTelemedicine

WirelessNetwork

fIn‐ Body Wireless Devices‐Sensors/monitoring devices 

Recovery fromNerve Damage

‐Drug delivery systems‐Medical robots‐Neural implants  Challenges being defined 

Page 41: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

Tech Transfer ChallengesTech Transfer Challenges

• Communication and network theory can be implemented in a real system in 3‐12 months 

f / h h l fl

with sufficent $$$

• Information/Communication Theory heavily influence next‐gen. wireless systems (mainly at the PHY & MAC layers)

• Idealized assumptions have been liberating• Idealized assumptions have been liberating

• Above PHY/MAC, there is little fundamental theory, which has prevented real breakthroughs

Industry people read our papers and implement our ideas

• Launching a startup is the best way to do tech transferLaunching a startup is the best way to do tech transfer

• We need more/better ways to exploit academic innovation

Page 42: Stanford Universitynetseminar.stanford.edu/past_seminars/seminars/Net... · Distributed Control over Wireless Automated Vehicles ‐Cars ‐Airplanes/UAVs ‐Insect flyers Interdisciplinary

SummaryTh t i i l t h l iThe next wave in wireless technology is upon us

Thi t h l ill bl li ti th t illThis technology will enable new applications that will change people’s lives worldwide

Design innovation will be needed to meet the requirements of these next‐generation systemsrequirements of these next generation systems

A systems view and interdisciplinary design approach y p y g ppholds the key to these innovations