la sp - cornell universitywisl.ece.cornell.edu/workshops/2005_10_11/tong.pdf · lab o rato ry ct a...
TRANSCRIPT
c©2005CornellSensorNetworkWorkshop1
ApplicationSpecificLargeScaleSensorNetworks
Architecture,EnergyEfficientProtocols,andSignalProcessing
LangTong
AdaptiveCommunicationsandSignalProcessingGroup
SchoolofElectricalandComputerEngineering
CornellUniversity,Ithaca,NY14853
http://people.ece.cornell.edu/ltong
ThisworkissupportedinpartbythetheArmyResearchLaboratoryCTAonCommunicationandNetworks,ArmyResearchOffice,OfficeofNavalResearchOffice,andNationalScienceFoundation.
c©2005CornellSensorNetworkWorkshop2
TheInternetandLayeredArchitecture
PSfragreplacements
WAN
LAN
PSfragreplacements WAN
LAN PSfragreplacements
LANPhysical
DataLink
MAC
Network
Transport
Application
Objectives
uMillionsofusers
uThousandsofapplications
uAgrowingvarietyofdevices.
Theneedoflayeredapproaches:
uPartitionnetworkfunctionsintolayers
uDesigneachlayerseparately
c©2005CornellSensorNetworkWorkshop3
SensorNetworksareApplicationSpecific
EnvironmentMilitary Weather
SeismicPSfragreplacements
uNetworkwithpurposes.
u“Unconventional”designmetrics
uHarshdesignconstraints
Whatshouldbetheappropriatedesignparadigm?
c©2005CornellSensorNetworkWorkshop4
LargeScaleSensorNetworks
PSfragreplacements
Internet
¥Low-powerlow-costsensors
¥Randomdeployment
¥Largenumberofnodes.
Applications
2Targetdetectionandtracking.
2Surveillance
2Environmentalmonitoring
2······
DesignChallenges
2Architecture
2PhysicalLayerDesign
2Protocols
2NetworkMaintenance.
2SecurityandTrust.
c©2005CornellSensorNetworkWorkshop5
Outline
Architecture
•SensorNetworkswithMobileAccess(SENMA)
•ExperimentalTestbed.
EnergyEfficientProtocols
•OpportunisticRandomAccess
•EnergyEfficientAdaptiveRouting
CommunicationandSignalProcessing
•Howmanysensorsarealiveandforhowlong?
•Cooperativenetworkwithmisinformednodes
•Nonparametriceventchangedetectionandlocalization.
•TargetdetectionandtrackinginSENMA.
c©2005CornellSensorNetworkWorkshop6
SensorNetworkArchitectures
Multi-hopSensorNetwork
2Stationarysensorsandgatewaynodes.
2Multi-hopto/fromgatewaynodes.
2Sensorstalktosensors.
2Asymmetricandcorrelatedtraffic.
c©2005CornellSensorNetworkWorkshop7
SensorNetworkArchitectures
Multi-hopSensorNetwork
2Stationarysensorsandgatewaynodes.
2Multi-hopto/fromgatewaynodes.
2Sensorstalktosensors.
2Asymmetricandcorrelatedtraffic.
PSfragreplacements
MobileAccessPoints
SensorNetworks
SensorNetworkwithMobileAccess
2MobileAPs+stationarynodes.
2SensorstalkdirectlytoAPs.
2StationarysensorsdrivenbymobileAPs.
2Asymmetricresources.
c©2005CornellSensorNetworkWorkshop8
SENMA:SEnsorNetworkwithMobileAccess
PSfragreplacements
MobileAccessPoints
SensorNetworks
¥MobileaccesspointscanbeUAV,UGV,submarine,doves,dogs...
¥Mobileaccesspointsarenotenergyconstrained.
¥Suitableforclock-drivenandquery-drivenapplications.
L.Tong,Q.Zhao,andS.Adireddy,”SensorNetworkswithMobileAgents(SENMA),”Proc.IEEEMILCOM,Oct.,2003.
c©2005CornellSensorNetworkWorkshop9
AcousticSENMA:SystemConfiguration
Mobile Access Point
Sensor Field
Sound Processing/Network Performance Analysis
c©2005CornellSensorNetworkWorkshop10
Outline
Architecture
•SensorNetworkswithMobileAccess(SENMA)
•ExperimentalTestbed.
EnergyEfficientProtocols
•OpportunisticRandomAccess
•EnergyEfficientAdaptiveRouting
CommunicationandSignalProcessing
•Howmanysensorsarealiveandforhowlong?
•Cooperativenetworkwithmisinformednodes
•Nonparametriceventchangedetectionandlocalization.
•TargetdetectionandtrackinginSENMA.
c©2005CornellSensorNetworkWorkshop11
OpportunisticCommunicationsandNetworking
PSfragreplacementsγ2 γ1
r
MobileAP
Sensors
OpportunisticCommunications
2MobilityofAPinducesfading.
2Fadingleadstochannelfluctuations.
PSfragreplacements
t
γ1(t)γ2(t)
Onlyuser1Onlyuser2
DesignRationale
2Waitforchances.Savepower
2Actwhenopportunitiesarise.
c©2005CornellSensorNetworkWorkshop12
OpportunisticCommunicationsandNetworking
PSfragreplacementsγ2 γ1
r
MobileAP
Sensors
OpportunisticCommunications
2MobilityofAPinducesfading.
2Fadingleadstochannelfluctuations.
PSfragreplacements
t
γ1(t)γ2(t)
Onlyuser1Onlyuser2
DesignRationale
2Waitforchances.Savepower
2Actwhenopportunitiesarise.
c©2005CornellSensorNetworkWorkshop13
OpportunisticMediumAccessControl
PSfragreplacements
γ∼F(γ)
PSfragreplacements Slott
Beacon
Data
DistributedOpportunisticMAC
2ChannelAcquisition
Estimateγfrombeacon.
2OpportunisticALOHA
Transmitwithprobabilitys(γ).
2OpportunisticCSMA
2Sensecarrierwithbackoffτ(γ).
Advantages
2Simplicity.2Scalability.
2EnergyEfficiency.
P.Venkitasubramaniam,S.AdireddyandL.Tong”SensorNetworkswithMobileAccess:OptimalRandomAccessandCoding,”
ToappearinIEEEJSAC:SpecialIssueonSensorNetworks
Q.ZhaoandL.Tong,”DistributedOpportunisticInformationRetrievalinSensorNetworks:CSI-BasedCarrierSensing,”
ICASSP’04,AlsosubmittedtoIEEETrans.WirelessCommunications.
c©2005CornellSensorNetworkWorkshop14
OpportunisticMediumAccessControl
PSfragreplacements
γ∼F(γ)
PSfragreplacements Slott
Beacon
Data
DistributedOpportunisticMAC
2ChannelAcquisition
Estimateγfrombeacon.
2OpportunisticALOHA
Transmitwithprobabilitys(γ).
2OpportunisticCSMA
2Sensecarrierwithbackoffτ(γ).
Advantages
2Simplicity.2Scalability.
2EnergyEfficiency.
P.Venkitasubramaniam,S.AdireddyandL.Tong”SensorNetworkswithMobileAccess:OptimalRandomAccessandCoding,”
ToappearinIEEEJSAC:SpecialIssueonSensorNetworks
Q.ZhaoandL.Tong,”DistributedOpportunisticInformationRetrievalinSensorNetworks:CSI-BasedCarrierSensing,”
ICASSP’04,alsosubmittedtoIEEETrans.WirelessCommunications.
c©2005CornellSensorNetworkWorkshop15
EnergyEfficientAdaptiveRouting������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������������ ������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������ ������
������ ������ ������ ������ ������ ������ ������ ������ ������ ������
PSfragreplacements
S
D
R
r10
−210
−1
5
10
15
20
25
30
35
40
45
λ
Total energy consumption
R=1000,N=30k,Bm=500,τ=300,po=0.01
proactivereactivehybrid
PSfragreplacements
S
D
R
r
Q.ZhaoandL.Tong,”EnergyEfficiencyofAdHocNetworks:AnAnalysisofProactive,Reactive,andHybridNetworkingStrategies,”
submittedtoIEEEINFOCOM,July,2004.
c©2005CornellSensorNetworkWorkshop16
Outline
Architecture
•SensorNetworkswithMobileAccess(SENMA)
•ExperimentalTestbed.
EnergyEfficientProtocols
•OpportunisticRandomAccess
•EnergyEfficientAdaptiveRouting
CommunicationandSignalProcessing
•Howmanysensorsarealiveandforhowlong?
•Cooperativenetworkwithmisinformednodes.
•Nonparametriceventchangedetectionandlocalization.
•TargetdetectionandtrackinginSENMA.
c©2005CornellSensorNetworkWorkshop17
OperatingSensorsandEnergyDistribution
−0.200.20.40.60.811.20
5
10
15
Power level [%]
Num
ber of nodes [%]
Power level distribution
deadhiddenobserved
Problem:Estimatethenumberofoperatingsensorsandthehistogramofthe
energydistributionusingtheinformationinthereceivedpackets.
c©2005CornellSensorNetworkWorkshop18
TwoApproachesinMonitoringWirelessSensorNetworks
CountingviaScheduledInterrogation
Askeachsensortransmitandcountthenumberofsuccesses
•Centralschedulingisneeded.MobileAPmustaddresssensors
individually.
•RequiringatleastO(N)andmoreifchannelhasfadingandinterferences.
CountingviaRandomCollection
Collectpacketsrandomlyandcountthenumberofsuccesses.
•Distributedtransmissionwithsimplemediumaccess.
•Forsuchamethodtobeeffective,itrequiresNlogN+O(N)successfully
receivedtransmissions.
•Example:N=1000,ontheorderof10,000successfulreceptionsare
needed.
c©2005CornellSensorNetworkWorkshop19
TwoApproachesinMonitoringWirelessSensorNetworks
TheEstimationApproach
•Innetworkmonitoring,theexactvaluesofthenetworkparametersare
notrequired.
•Estimatethenumberofoperatingsensorsusingthestatisticalproperties
ofthereceivedsamples.
Question:
CanweestimateaccuratelyusinglessthanNlogN+O(N)samples?
•C.Budianu,S.Ben-DavidandL.Tong,“EstimationoftheNumberofOperatingSensorsinLargeScaleSensorNetworks”,
ICASSP2004,alsosubmittedtoIEEETrans.SignalProc.,June,2004.
c©2005CornellSensorNetworkWorkshop20
CooperativeSENMAwithMisinformedSensors
�������� ����
PSfragreplacements
q(y|x)
!!
""##$$%%
PSfragreplacements
q(y|x)
•Incooperativesensornetworks,sensorsmaynotagreewitheachother.
•Whatisthemaximumrateofinformationretrievalwiththepresenceof
misinformedsensors?
•Z.YangandL.Tong,”CooprativeSensorNetworkswithMisinformedSensors,”Proc.ofthe38thAnnualConferenceon
InformationSciencesandSystems(CISS’04),Princeton,NJ,March2004.TobesubmittedtoIEEETrans.InformationTheory,
July,2004.
•Z.YangandL.Tong,“Theeffectoffadingontheachievablerateofcooperativesensornetworkswithmisinfomredsensors,”to
appearinIEEE2004GLOBECOM,
c©2005CornellSensorNetworkWorkshop21
NonparametricEventChangeDetectionandLocalization
PSfragreplacements
FirstdatacollectionSeconddatacollection
•Isthereachangeinthedistributionsof“red”and“green”?where?
•Canweguaranteedetectionprecision?Whatisthecomputation
complexity?
•S.Ben-David,T.HeandL.Tong,”Non-ParametricApproachtoChangeDetectionandEstimationinLargeScaleSensor
Networks,”Proc.ofthe38thAnnualConferenceonInformationSciencesandSystems(CISS’04),Princeton,NJ,March2004.
c©2005CornellSensorNetworkWorkshop22
DetectionofSteppingStoneAttack
SteppingStoneAttack
Theattackerusesachainofcompro-
misedhoststoconstructasequence
ofinteractiveattacks
QuickestAttackDetection
Detectpossiblestepping-stoneattack
pairsbasedonstatisticalpropertiesof
incomingandoutgoingtraffic.
•T.HeandL.Tong,”DetectionofEncryptedInteractiveSteppingStoneAttack:AlgorithmsandPerformance,”tobesubmitted
toIEEETrans.SignalProcessing.
c©2005CornellSensorNetworkWorkshop23
TargetDetectionandTracking
ControlCenterPSfragreplacements
MobileAccessPoints
SensorNetworks
•SensorsperformlowqualityDoAestimationandtargetclassification.
•Thefusioncenterestimatesthefieldparameters.
•Developoptimaldetection,estimation,andfusionalgorithms.
c©2005CornellSensorNetworkWorkshop24
RelatedProjects
NationalScienceFoundation
•SignalProcessingforRandomAccess:ACrossLayerApproach.
•SENSORS:TheReachbackChannelinWirelessSensorNetworks.
(Servetto(PI))
•NET-NOSS:UltraLow-PowerSelf-ConfiguringWireless.(Wicker(PI))
•TRUST:TeamforResearchinUbiquitousSecureTechnology(Wicker
(PI))
OfficeofNavalResearch
ChannelandQoSAdaptiveMultimediaWirelessAd-HocNetworks(withHaas,
Hemami,Manohar,Wicker,Proakis,Nahrstedt).
DARPAConnectionlessNetworking.
ArmyResearchLaboratory
SignalProcessingforSensorNetworks.CrossLayerDesignforWireless
Networks.DirectionofArrivalEstimationwithNetworkedSensors.