Chapter 9 Energy Harvesting Aware Power Management

EnergyHarvestingAwarePowerManagement

9.1Introduction

Thetrueautonomyofwirelesssensornetworksdependsontheirreliableoperationforextendedtimeswithouthumanintervention.Energysupplyisacriticalfactorinthisdesign.Wirelessandad-hocdeployment,whichisessentialinsomescenariosandcost-effectiveinothers,precludestheuseofawiredenergyinfrastructure.Thesensornodesarethusforcedtooperateonlimitedbatteryreservesandlowpowerdesignisanimportantdesignconsideration[1,2].

Unlikehumancarrieddevicessuchashand-heldsorcell-phoneswhichcanbereturnedtochargingdocksperiodically,sensornodebatteriesarelimitedinsupply.Theycannotbereplacedinthelargenumbersofnodesastheembeddednatureofdeploymentmakesithardtoaccesseachindividualnode.Alimitedamountofenergysupply,however,isnotsufficienttoensureuninterruptedoperationfortheseveralyearlonglifetimestypicallyexpectedfromanembeddeddeployment.Thesmallnodesizeputsconstraintsonthe

1

maximumbatterysize.Batteriesalreadydominatethenodevolumeinprototypesensornodes.Theenergydensityforcommonbatterytechnologiesvariesintherangeof1200J/cu.cm.(Alkaline)to3780J/cu.cm(Zinc-air).Atsuchenergydensity,assumingasensornodeoperatingat1mW(averageconsumptionafterpowermanagement)andassumingthefullbatterycapacitycanbeutilized,ayear-longoperationrequiresabattery-sizeoftheorderof10cu.cmwhichisratherlarge.Thus,batteriesalonecannotbeexpectedtoreliablysupplyasensornetworkdeploymentforseveralyears.Aviablealternativethenistoendowthesensornodeswithappropriateenergyharvestingtechnologiessuchassolar,vibrational[3],wind/waterflow,thermalgradientscavenging[4 - 6],electromagneticdirectconversion[7]andothers[8 -11].Thesesourcescansupplementorevenentirelyreplacethebatteryenergysupply.Afundamentaldifferencebetweenenvironmentalenergyandbatterysupplyisthattheenvi-ronmentalenergycanbescavengedforaslongasdesiredandifefficientlyutilized,canenableasystemtolasteternally(untilitshardwareisoutdated).However,theintroductionofharvestingcomponentsintosensornodesrequiresdesignchangesspanningthehardwareofthenode,thenodelevelpowermanagement,andnetworkwideenergyscheduling.Considerforinstance,thatthepowerconsumptionatanodeistobematchedtotheenvironmentalenergyavailabletoit.Then,thenodemustbeequippedwithadditionalhardwaretomeasuretheenvironmentalenergyinput,ratherthanjusttheresidualbatterymeasurement.Further,powermanagementdecisionsinanetworkdifferwhenharvestedenergyisavailable.Forinstance,considerasolarenergyharvestingnetwork.Ataparticularinstance,thenetworkmayhavetwoalternativedataroutesavailabletosatisfytheimmediatedatatransferrequirement.However,theharvestedenergyavailableatnodesalongthetworoutesmaybedifferent,sayduetothepresenceofashadowonpartofthenetwork.Insuchacase,thenetworkrequiresaharvestingawarepowermanagementstrategy,alongwiththeenvironmentalenergyinputmeasurement,whichallowsittochoosetheroutepassingthroughnodesoutsidetheshadow.Theremainderofthischapteraddressestheresultingissuesinharvestingawaresystemdesign,includingboththehardwareandpowermanagementsoftware.29.2HarvestingTechnologies

Wedefineaharvestingnodetobeanysensornodewithatleastoneformofenvironmentalenergyharvestingaspartofitspowersupply.Typically,suchadevicewillalsohaveanenergystoragemechanism,suchasabatteryoranultracapacitor,toallowenergyharvestingandconsumptiontooccurwithouttotalsynchrony.However,thestoragedeviceisnotessentialinallscenarios,andharvestingnodessuchasdescribedin[10]usetheenergygeneratedfromthepressofabuttonimmediatelytotransmitapacketandareinactiveotherwise.Anetworkofsuchdeviceswillbereferredtoasaharvestingnetwork.Thedesignofthenetworkinvolvesfurtherconsiderationsthanjusttheindividualnodes,sincethenodesinsuchanetworkmaybeheterogeneousandtheenvironmentalenergyavailableateachnodemaybedifferent.

Weconsidertheharvestingnodefirst.Figure9.1showsthenewmodulesthatarepartofaharvestingnodeinadditiontotheusualsensornodecomponents.Thevariousblocksarediscussedinthesubsequentsections,includinganexampleimplementation.Whilemostoftheblocksshownareimplementedashardwarecircuits,theblocklabeledHarvestingAwarePowerManagementisbestimplementedasasetofalgorithmsonthesensornodeprocessoritself.Adesigncomponentnotshowninthefigurebutwhichinfluencesthehardwareandsoftwaredesignisknowledgeabouttheapplicationbehaviorwithrespecttoenergyconsumptionandthisknowledgeshouldbeexploitedtotheextentavailable.

Figure9.1:Blockdiagramofaharvestingnode.

Thekeydistinguishingcharacteristicoftheharvestingnodeappearsatthetopleftintheblockdiagram-theharvestingdevice.Itisbasedononeoftheharvestingtechnologiesrelevanttothedeploymentenvi-ronmentandoutputsenergy.Thekeyharvestingtechnologiesofinteresttoembeddedsystemsarediscussedbelow.

3

9.2.1SolarEnergy

Solarorotherlightenergycanbeconvertedtoelectricpowerusingsolarcells.Themagnitudeofenergygen-eratedvariesfromapproximately15mW/sq.cminnoon-timesunlightto10µW/sq.cminindoorincandescentlighting.Theenergyoutputdependsonthematerialused.Crystallinematerialssuchassiliconandgalliumarsenidehavemoderateabsorptionefficiencyandhighconversionefficiency(15-30%)whilethinfilmma-terialssuchascadmiumtelluridehavehighabsorptionefficiencyandlowerconversionefficiency(≤10%).Thechoiceofmaterialalsodependsonitsspectralresponseandthelightsourceofinterest.Forthepurposesofcircuitdesign,thesolarcellmaybemodelledasavoltagesourcewithaninternalresistance.Theoutputvoltageisfairlyconstantintheusefuloperatingrangeandthesupplycurrentvarieswithlightintensity.Asinglesolarcelloutputis0.6Vbutpanelswithseriesofsuchcellscangenerateanyrequiredvoltageforthecircuit.9.2.2VibrationEnergy

Vibrationsareavailableinmanyenvironmentsofinterestincludingcommercialbuildings,parkingstructures,aircrafts,trains,industrialfacilitiesandevenresidentialbuildings.Preliminaryanalysisandexperimentspre-sentedin[3]showthat300µW/cm3isavailableinsuchenvironments.Thesourcesofvibrationswhichmaybeheavymachinery,homeappliances,HVACvents,movementofpeopleorvehicles,andothermovementsvaryagreatdealintheiraccelerationcharacteristicsandthefrequencyspectra[3].Methodstoconvertthisenergytoelectricitycanbeclassifiedintoelectromagnetic,electrostaticandpiezoelectric[12 -16].Electromagneticconversionusesvibrationtomoveaconductorinamagneticfield.Existingprototypes[15,12]generateverylowvoltageoutputtobeusable.Electrostaticconversionusesvibrationenergytomovetheconductorsofachargedcapacitor.Thedisadvantageofthisapproachisthataseparatevoltagesourceisrequiredtochargethecapacitor.Anadvantagehoweveristhattheoutputvoltageisintheusablerangeoftwotoseveralvolts.Piezoelectricconversionusesmaterialswhichwhenmechanicallydeformedgenerateanelectricpotential.Thepiezoelectricmethodcombinestheadvantagesofelectromagneticand4electrostaticconversionbutaredifficulttoimplementatmicro-scale.Withthecurrenttechnology,theyhavethegreatestavailableenergydensityamongthethreemethods.

9.2.3Othersources

Windorwaterflowcanbeconvertedtoenergy.Whilemacro-scalegeneratorsbasedontheseflowsarewidelyused,compacttechnologiestoextractsuchenergyarelacking.Asensornetworkingapplicationofwindenergyisalsoforlocomotion,suchasusedinNASA’sTumbleweeds[17].Theseareinflatablesphereswhichcanrollalongthedeploymentsurfaceusingwindenergy,andareaimedatMartianandpolarexploration.ThermoelectricgenerationusingSeebeckeffect(flowofcurrentinaloopmadefromtwowiresofcer-tainmetalswhenatemperaturedifferenceisappliedtothewirejunctions)andothermethodshavebeendemonstratedtoyield10µW/cm2to40µW/cm2usinga5-10degreeCelciustemperaturegradient[18,6,4].Pressurevariations,suchasthepressureofafluidorgasinanenclosedspacechangingwithtimeofday,canalsobeusedtogenerateenergy.IntheAtmosclockforexample,inventedin1928byJean-LeonReutter,amixtureofgasandliquidenclosedinasealedcapsuleexpandsasthetemperaturerisesandcontractsasitfalls,movingthecapsulebackandforthprovidingsufficientmotiontoruntheclock.Methodstoconvertsuchlimitedpressurevariationormotionintoelectricityarenotreadilyavailable,however.

Theexactchoiceoftheharvestingtechnologydependsseveralfactors,includingontheachievedenergydensityfromaparticulartechnology,sensornodeformfactor,andmostimportantly,theavailabilityinthedeploymentscenario.Forinstance,asensornetworkdeployedforenvironmentalmonitoringorprecisionagricultureinoutdoorsettingsmayhavereadyaccesstosunlightandcouldhenceusesolarenergy,whileasensornetworkoperatingindoorsforindustrialapplicationssuchasmachinehealthmonitoringmayhaveplentyofvibrationsavailableforharvestingenergy.

5

9.3DescriptionsoftheComponentsofaHarvestingNode

Figure9.1alludedtovariousnewcomponentsthatarerequiredwhenasensornodeisconvertedintoahar-vestingnode.Thissectiondiscussesthereasonswhytheseblocksareneededandtheirdesignrequirements.Therechargingcircuitreceivespartoftheenergyoutputoftheharvestingdeviceandstoresit.Thisblockisneededsincethecurrentandvoltagecharacteristicsoftheharvestingdeviceoutputmaynotbedirectlysuitableforchargingthestoragedevice.Forexample,arechargeablebatterymustbechargedavoltageequaltoorhigherthanitsoutput,andiftheharvestingdeviceoutputscurrentatalowervoltagethanthis,avoltageconversioncircuitmayhavetobeused.

Therechargingcircuitalsodependsontheenergystoragemechanism,andincaseofbatteries,onthebatterychemistryused.Forabatterychargedfromasolarcell,thesimplestimplementationistodirectlyconnectasolarpaneltothebatterythroughadiodetopreventthebatterybeingdrainedthroughthesolarcell.However,thiscircuitwouldyieldlowlongevityforthebatteryasthebatterycouldberepeatedlydamagedfromovercharge(continuedchargingofbatteryevenwhenitischargedtoitsfullcapacity)andundercharge(continueddischargingafterthebatteryvoltagehasdroppedbelowabatterychemistryspecificthreshold).Ingeneral,toensurebatterylongevity,arechargingcircuitisrequiredtoproduceanacceptablechargingprofileandtodisconnectthebatteryatitsunder-chargelimit.

Theconsumptionarbiterusesacombinationoftheenergyfromtheharvestingsourceandthestoragedevicetosupplythepowerrequirementsoftheharvestingnode.Therearethreereasonswhichnecessitatethisarbiter.

First,theharvestedenergymaynotalwaysbesufficienttopowertheloadandhenceamechanismisneededtosharetheloadbetweentheharvestedandstoredenergy.Storageefficiencyisalwayslessthan1.Thearbitermustensurethattheenvironmentalenergyissuppliedtotheloaddirectly,andonlythedeficitisdrawnfromthestorage.Whentheenergyavailablefromtheharvestingdeviceexceedstheconsumption,theexcessenergyisroutedtotherechargingcircuit,whichmaystoreitifthestorageisnotalreadyfull.

6

Second,foragivenenvironmentalstimulus,theenergyoutputoftheharvestingdevicedependsonitsowninternalresistanceandthecharacteristicsoftheloadpresentedtoit.Theconsumptionarbitermayalsomodifytheloadcharacteristicsappearingattheoutputoftheharvestingdevicetomaximizethescavengedenergy.

Third,thestoragemechanismusedmaynotsupplytheenergyatthevoltagerequiredbythesensornodeandthearbiterregulatesthesupplyvoltage.

Thearbitermayalsobedesignedtoprioritizestorageofenergyincertaincasessuchaswhenthebatteryhasdrainedtoaverylowlevel.Insuchacase,iftheharvestedenergylevelisjustsufficienttopowerthesensornodebutnoexcessisavailabletochargethebattery,thenthearbitermayshutoffthenodeandchargethebatteryfirst,inordertoensuresystemavailabilityinemergencysituations.

Theenergytrackermonitorstheenergyavailablefromtheharvestingdeviceandalsothecurrentstateoftheenergystore.Suchdatamaybeusedbytheharvestingawarepowermanagementalgorithmsforlearningtheenergyenvironment.Theinstantaneousbatterystatusandenvironmentalenergyavailabilityalongwiththelongtermavailabilitybehaviorisusedbyseveralnetwork-wideschedulingalgorithmstodistributework-loadasdiscussedlaterinsectionsonpowermanagement.

Thesub-modulepowerswitchingblockistypicaltoanydevicewhichprovidesforshuttingdownpartsofitscircuitforpowermanagement.Thisisverycrucialinsensornetworkswheresensorsandminimaldirectmemoryaccessperipheralsmaybekeptactiveformuchlongerdurationsthantheenergyintensivecommunication,processingandlargermemorymodules.Inaharvestingnode,thisblockisusefulforturningonoroffvariouscomponentsastheamountofenvironmentalavailabilityvaries.Thisswitchingfunctionalityiscontrolledbytheharvestingawarepowermanagementalgorithms.

Aprototypeharvestingnode,implementedattheNetworkedandEmbeddedSystemsLaboratory,Uni-versityofCaliforniaLosAngeles,isshowninFigure9.2.Thisnodeusesapairofsolarcellsconnectedinparallelastheharvestingdevice.ThestorageisapairofAAsizedNiMHre-chargeablebatteries.TherechargingcircuitprovidesthechargingprofilerequiredforNiMHbatteries.Thearbitersuppliesthesolar

7

celloutputenergytotheloadwhiletheexcessordeficitflowsthroughthebattery.TheheliomotearbiteralsoregulatesthevoltageoutputoftheNiMHbatterypackwhichvariesbetween2.2and2.8Vtoastable3Vrequiredbythesensornode.Theenergytrackermodulemonitorsthetotalenergychargedintothebattery,theinstantaneousbatterycurrentandvoltage.Themeasureddataisprovidedtothesensornodeprocessorviaaone-wireinterface.Thedetailedcircuitschematicsareavailablein[19].

Figure9.2:Heliomote.

9.4HarvestingAwarePowerManagement

Powermanagementstrategiesdiffersignificantlywhenthepowersourcechangesfromafixedbatterysupplytoaharvestingdevice.Therearetworeasonswhichleadtothesedifferences.First,theenvironmentalenergysourceishighlyvariable.Unlikethestoredsupplywhichissimplycharacterizedbytheamountofresidualenergyofthebatteryandisreliablyavailable,theenvironmentalenergyrequiresamoresophisticatedcharacterization.Second,theenvironmentalenergyisnotalimitedresourceandhasthepotentialtobeusedeternally.Below,wepresentamodeltocharacterizetheproductionandconsumptionofenvironmentalenergywhichcanbeusedtodesignpowermanagementalgorithmsforharvestingnodes.

9.4.1HarvestingTheory

Sincetheharvestingsourceishighlyvariable,animportantdesignconsiderationistoprovideareliablesystemperformance.Anotherobjectiveistodecideiftheenvironmentalenergyaloneissufficientforthesystemtooperateatthedesiredlevelofperformance.Ifso,thesystemcanoperateindefinitely.Otherwise,thedesignermaywanttodeterminetherequiredharvestingadditions.Thegoalofdevelopingaharvestingtheoryistoenableanalysisofawidevarietyofharvestingtechnologieswithrespecttosystemperformance.Webeginwithasourcecharacterization.

8

Definition1.(ρ,σ1,σ2)−source:SupposeE(t)isacontinuousandboundedfunctionofacontinu-ouslyvaryingparametert.E(t)issaidtobea(ρ,σ1,σ2)−sourceifandonlyifforanyfiniterealnumberT,itsatisfies:

󰀂

E(t)dt≥ρT−σ1E(t)dt≤ρT+σ2

(9.1)(9.2)

󰀂T

T

Thisdefinitionusesonlythreeparameterskeepingitanalyticallytractablebutstillallowingittomodelawidevarietyofvariationsinenergysources.E(t)modelsthepoweroutputattimet.Themodelcapturestheasymptoticrateofavailability,whichisthemaximumpoweratwhichthesystemcanoperate.Sincewearemodellingphysicalenergysources,therestrictionsplacedonthefunctionE(t)arejustified.Itmaybenotedthattheunitofρispower,e.g.Wattsandtheunitofσ1andσ2isenergy,e.g.Joules.

Theconsumptionitselfmaynotbeconstantandthefollowingdefinitioncancharacterizemostconsump-tionprofiles.

Definition2.(ρ′,σ)−consumer:Adeviceissaidtobea(ρ′,σ)consumerifitspowerconsumption,Ec(t),satisfiestheconstraint

󰀂

Ec(t)dt≤ρ′T+σ

(9.3)

T

foranyvalueofT.

Withthisdefinition,thefollowingtheoremspecifiestheminimumperformancethatcanbeguaranteed.Theachievableperformancemayinfactbehigher,suchasifanapplicationrequiresnodeoperationonlywhentheenvironmentalenergyisavailable.

Theorem1.SustainablePerformanceatEternity(VariableConsumptionProfile):Ifa(ρ′,σ)-consumerdeviceispoweredbya(ρ,σ1,σ2)-source,hasanenergystoragecapacityofσ+σ1+σ2,andρ′<ρ,thenthedevicecanoperateforever.

Theproofisavailablein[20].Thefollowingexampledemonstratesanimmediateapplicationoftheabovetheorem.

9

Table9.1:Solarcellparametersinexperimentalenvironment

Parameter

ρσ1σ2

Value23.61.4639×1061.8566×106

UnitsmWJJ

Example1.Figure9.3showsthepowerflowingintothebatteryobservedbyatestheliomote,whenplacedinsunlight,inthemonthofJanuaryintheNorthernhemisphere.Negativevaluesrepresentbatterydrain.

Figure9.3:Solarenergybasedchargingpowerrecordedfor9days.

Measuringthebatterycurrentinsteadofthecurrentoutofthesolarcellensuresthatonlytheactualsolarpoweravailable,andnotanypowerlostduetocircuitinefficiencies,isconsidered.Sincethenegativeportionsofthewaveformdominate,thebatteriessufferanetdischarge.ForthewaveformplottedinFigure9.3,thesourcecharacterizationparametervaluesaregiveninTable9.1.

Withthesevaluesthesolarcellisa(ρ,σ1,σ2)-sourceinthetestenvironment.Wenowneeda(ρ′,σ)classificationoftheconsumernode.ThesensornodeusedintheheliomotehasasleepmodepowerdrawnPsleep≤3mWandthemaximumcurrentdrawnPmax=100mW.Thus,(ρ′=100mW,σ=0)isavalidclassification.However,toachieveρ′=ρ,wecansetthenodetosleepfor78.7%ofthetime.Iftheminimumwakeupdurationis2s,thisleadstoσ=153J.Theorem2impliesthatarechargeablebatterywithcapacity=σ+σ1+σ2=3.32×106J,or922.43mAh,isrequiredforsustainingthenodeindefinitelyfromsolarenergy.ThishelpsthedesignertochooseaAAAsizedNiMHbattery.Usingalargerbatterydoesnothelpimprovelongtermperformance.

Thekeyutilityoftheabovetheoryisthatitenablescharacterizingtheenergyavailabilityusingasmallnumberofparameterswhichcanbeusedforpowermanagementasdiscussednext.

10

9.4.2SchedulingAlgorithms

Inbatterypoweredsystems,thepowermanagementstrategiesaredesignedmerelytominimizetheenergyconsumption.Severalpowerscalingstrategiestominimizebatteryconsumptionhavebeenstudied[21 -24].Manyofthesecanbeusedtominimizebatterydraininharvestingnodesalso.However,thesestrategiesarenotsufficientforefficientlyutilizingtheharvestingsource.Inharvestingsystems,theproblemspaceexpands.Theaimisnotalwaystominimizeconsumptionbutadapttheconsumptiontotheavailableenvironmentalenergywithintheperformanceconstraints.Forinstance,thetasksmayberescheduledwhenpossibletooperatedirectlyofftheenvironmentalenergy.Thisdoesnotreducetheenergyconsumedinexecutingthetask,butreducesthewastageduetobatteryinefficiency.Researchhasrecentlybegunintheareaofharvestingawarepowermanagement[25,26].Variouspowerscalingmechanismsaretypicallyavailabletocontrolthepowerconsumptioninembeddedsystems,suchasdynamicvoltagescaling(DVS),transmitpowercontrol,anddutycyclingthenode.Foreachmechanism,reducingpoweraffectstheperformance.ForDVSanddutycyclingforinstance,usingalowerenergymodeincreasesthelatencyofexecutingatask.Thealgorithmsbelowareillustratedusingdutycyclingasanexamplepowerscalingmechanismsinceitisavailableinmostprocessors.Theorem1givestheasymptoticrateofenergyconsumptionavailabletoaharvestingnode.Asmen-tioned,afundamentaldifferenceinharvestingsystemscomparedtobatterypoweredsystemsisthatifthesystemperformanceisadaptedtotheenvironmentalavailability,thesystemcanlasteternally.Toadapttheperformanceinthismanner,thesystemonlyneedstoestimatetheparametersρ,σ,σ1,andσ2foritsdeploy-mentscenario.Thiscanbeachievediftheenvironmentalbehavioroveralearningphaseisassumedtoberepresentativeofthelongtermbehavior,whichisvalidformanysources.Forinstance,solarenergyfollowsadiurnalandannualcycle.Windshaveknownrepetitivepatterns.Thesustainableperformancedependsdirectlyonρandhencewewishtoestimateρtowithinanerrormargin∆.Thepowerparameterρcanbeestimatedbyaveragingtheenergyobtainedfromtheenergysourceovertime.Lettheaverageattimet,ρ(t)11becalculatedas:

ρ(t)=

󰀁t

0

E(t)dtt

(9.4)

SincethedevicewillsampleE(t)atdiscretetimes,theaboveintegralwillbeevaluatedasadiscretesum-mation.Apartfromtherunningestimateofaverage,alsostorethemostrecentlocalminimaandmaximaobservedinρ(t).Whenthedifferencebetweenthemaximaandtheminimareaches∆weassumethatestimationofρiscomplete.TheE(t)waveformcanalsobeusedtoestimateσ1andσ2[20].

Oncethesourcehasbeencharacterized,thedevicecanadjustitsperformancetooperateattheavailablerateρ.SupposeactivemodepowerisPmaxandsleepmodepowerisPsleep.Thenthedutycyclexsatisfies:

ρ=xPmax+(1−x)Psleep

(9.5)

neglectingtheenergyconsumptionofswitchingbetweenmodes.Thevalueofxdeterminedfromthisequa-tioncanbeusedtodecidethesleepdurationiftheminimumtimespentinactivemode,Tminisknown.Thus,performancecanbeadjustedatasinglenodetomatchtheenvironmentalavailability.

9.5DistributedHarvesting

Inadistributedsystemwithseveralharvestingnodes,theharvestingawarepowermanagementproblemwouldbetodistributetheworkloadamongthenodesinsuchawaythattheoverallperformanceofthesystemismaximized.Findingtheoptimalsolutionrequirescompleteknowledgeofwhatenergyresourcesareavailableateverynodeandthecompletesetoftaskstobeperformedbythesystem.Manyofthesetasks,suchasroutingadatapacketfromonenodetoanotherinvolveenergyconsumptionatseveralintermediatenodes,makingtheschedulingdecisionscoupledamongnodes.Also,inadistributedsystem,scalabilityconcernsdictatethatalltheinformationateverynodenotbecommunicatedtoacentralnodeforschedulingdecisions.Moreover,inmanydistributedsystemssuchassensornetworks,communicationitselfisthemajorenergyconsumeranddistributeddecisionsaretheassumednorm.Distributedpowermanagementsolutionsarehenceused,eventhoughtheymaynotbegloballyoptimal.

12

Distributedpowermanagementisanascentareaofresearch.Methodshavebeenproposedfordistributedbatterypoweredsystems[27 -31].Thesemethodsattempttoreduceandbalancetheenergyusageacrossmultiplenodestoensuresystem-widesustainabilityratherthanminimizingthetotalenergyconsump-tionwhichcouldcausecertainnodestodepletefasterandmaketheoverallsystemuseless.Inharvestingnetworksontheotherhand,insteadofbalancingtheenergyconsumptionamongnodes,theobjectiveistosharetheworkloadinproportiontotheharvestingopportunityavailabletoeachnode.Thedistributionofworkloadmaybehighlyapplication-dependentandsuchmethodstypicallyrequireanunderstandingoftheapplicationcharacteristics.Onesuchmethodwhichaddressesatypicalusagescenarioisdiscussedbelow.Oneoftheapplicationsforsensornetworksistomonitoradeploymentsceneforspecificeventsandreporttheoccurrenceofsucheventswithlowlatency.Thesensingtransducersconsumeminimalenergyandarekeptactive,whilethepowerintensivemodules-processorandradio,arewokenupwhenaneventistriggeredbyatransducer.Theeventmustnowbecommunicatedwithlowlatencytoacentralbasestation.Thepowermanagementproblemistoprovideanenergyefficientcommunicationtopologyforthenetwork,whichadaptstheenergyconsumptionatdifferentnodestotheirindividualharvestingopportunities.Theschemebelowachievesthisinacompletelydistributedfashion.Also,itcanreporttheexpectedperformancetothebasewithminimalcommunicationoverhead.9.5.1CommunicationProtocol

ThecommunicationprotocolfollowedisdepictedinFigure9.4.Whenanodehasdatatotransmittoitsnexthopneighboralongthedatapath,thenodetransmitsaBEACONpacketandlistensforresponseforaperiodTack.ItrepeatsthisprocessuntilanACKisreceived.TackisthetimerequiredbyanactivenodetosendanACKafterreceivingaBEACONpacket.SupposethetimerequiredtotransmitaBEACONpacketisTbeacon.EverynodeinthenetworkwakesupforadurationTmin=2Tbeacon+TacktolistenforanyBEACONmessagesfromnodesattemptingtosenddatatoit.Aftereveryawakeperiod,itsleepsfora13duration

Tsleep=

1−x

Tminx

(9.6)

wherexisthesustainabledutycycleatthisnode,determinedfrom(5).WheneveranodereceivesaBEACONpacket,ittransmitsanACKandstaysawakeuntilithasreceivedtherelevantdataandforwardedittothenexthop.Withthisarrangement,theworstcasedelayinreceivinganACKwhenrepeatedlysendingaBEACONisTsleep+Tmin.Theenergyspentintransmittingtheeventdataisnegligiblecomparedtotheenergyspentinperiodicallyenteringactivemodetolistenforpotentialdatawheneventsareinfrequent.

Figure9.4:Communicationwithsleepcycle.

Theroutediscoveryfromallnodestothebaseisbasedontheformationofadatagatheringtree,adoptedfrom[32].ThebasenodetransmitsanINITpacket.Allnodeswhichreceivethispacket,treatthesenderoftheINITpacketastheirparentandsendanACKwithrandomback-offdelay.Theynowknowtheroutetothebase,whichisonehopaway.Thesenodesassignthemselvesdepth=1.TheyretransmittheINITmessagewiththeirownID.Allnodeswhichhavenotalreadyassignedthemselvesadepthacknowledgethispacket.Thesenodesnowknowtheirnexthopontheroutetothebaseandassignthemselvesdepth=2.Theprocesscontinues.WhenanodetransmitsanINITmessagebutdoesnotreceiveanyresponsesforatimeoutduration,itassumesthattherearenonodesdeeperthanitselfalongthepaththatpassesthroughit.Suchnodesaredenotedleafnodes.

9.5.2DistributedPowerManagement

Thenetworkstartsatanarbitraryperformancelevel.Oncetheestimationofρ,σ1andσ2parametersiscompleted,usingthesamemethodsasdiscussedintheprevioussection,thenodeswillbegintransitiontothesustainableperformancelevel.Eachnodesetsitsresponselatency,L,equaltoTmin+Tsleep.Ifanodeisnotreceivinganyenergyfromtheenvironment,itmayhavetoselectapresetvalueofLforitself.WhenaleafnodehasestimateditsL,itsendsaLATENCYmessagetoitsparentcontainingitsestimateofL.A

14

nodewhichisnotaleaf,waitsfortheLATENCYpacketfromallitschildren.LetLidenotethelatencyvalueheardfromi-thchildandletithaveNcchildren.WhenithascollectedallLATENCYpacketsfromitschildrenandhasestimateditsownL,itsets

L′=

max

{Li}+L

(9.7)

i∈[1,...,Nc]

ItsendsthiscumulativevalueofL′toitsparent.L′representsthetotalworstcasedatatransferlatencyalongthepaththroughthisnodefromtheworstleaf.Theprocessrepeats.ThebasethenchoosesLmaxequaltothelargestL′amongtheonesitreceivesfromitschildren.ThisLmaxisthemaximumlatencyofdatatransferinthenetworkalongtheworstpath.EachnodeswitchestothedutycyclebasedonitsestimatedρafterhavingsenttheLATENCYmessage.Thusthenodesgraduallytransitiontotheirindividualperformancelevels,withnocentralizedcoordination.

Thein-networkprocessingofthetransmittedLvaluesateachparentreducesthedatasenttotheparentnodesandtheamountofdatasenttothebasenodeisthusproportionaltothenumberofdepth1nodesandnottothetotalnumberofnodesinthenetwork.Thisensuresscalabilitywhileatthesametimereturninganetworkwideperformancemetrictothebase.Iftheachievedperformanceisnotacceptable,additionalresourcesmayhavetobeadded.

Theabovemethodprovidedalgorithmstoadaptthepowerconsumptiononanetworkwidebasistothelocalenergyavailabilityatthevariousnetworkcomponents.Furtherresearchisunderwaytoexploitmoredetailedinformationabouttheenergyavailabilityprofile,includingitstemporalvariationsandspatialpatternstooptimizeperformanceingivenenergyenvironments.

9.6Conclusions

Energyharvestingprovidesaviablealternativetocreatelonglastingwirelesssensornetworks,usinghar-vestedenergytosupplementorreplacestoredbatteryreserves.Harvestingisnotonlyusefulbutessentialinsomeusagescenariossuchasspaceexplorationandextra-terrestrialsensornetworkdeploymentswhere

15

theoperatinglifetimerequiredisverylongandbatteryreplacementsarenexttoimpossible.TheunmannedvehiclesexploringthesurfaceofMarsusesolarenergyharvestingforalltheirenergyneeds.Differenttypesofharvestingopportunitiesexistinmostdeploymentscenariosandthisformofenergyshouldbeactivelyex-ploitedforenhancedsystemperformance.Theuseofharvestedenergyleadstoseveralresearchissueswhicharedifferentfromconventionalpowermanagementbasedononlystoredbatteryenergy.Notonlyisthedesignofthehardwaredifferenttoaccountfortheharvestingsource,butworkloadschedulingnowdependsonthenatureoftheenvironmentalsource.Theschedulingisalsodifferentincaseofdistributedsystemswithmultipleharvestingnodes.Harvestingtheoretictechniquescanbeusedtodeterminethesustainableperformancelevelsandadditionalresourcesrequiredforachievingrequisiteperformance.Thesetechniquesalsofacilitatethedesignofdistributedschedulingmethods.Anidealsystemwouldutilizeitsstartingbatteryresourcesandharvestedenergytoachievethemaximumpossibleapplicationthroughputwiththeavailableresources.Suchpowermanagementforharvestingnetworksisanopenproblemwitharichsetofresearchchallenges.

16

Bibliography[1]Raghunathan,V.,C.Schurgers,S.Park,andM.Srivastava,“Energyawarewirelessmicrosensornet-works,”IEEESignalProcessingMagazine,Vol.19,No.2,2002,pp.40–50.[2]Min, R.,M.Bhardwaj,S.Cho,A.Sinha,E.Shih,A.Wang,andA.Chandrakasan,“Anarchitectureforapower-awaredistributedmicrosensornode,”inIEEEWorkshoponSignalProcessingSystems(SiPS’00),October2000.[3]Roundy,S.,P.K.Wright,andJ.M.Rabaey,EnergyScavengingforWirelessSensorNetworkswithSpecialFocusonVibrations.KluwerAcademic,2004.[4]Stordeur,M.,andI.Stark,“Lowpowerthermoelectricgenerator-selfsufficientenergysupplyformi-crosystems,”in16thInternationalConferenceonThermoelectrics,pp.575–577,1997.[5]Weber,W.,“Ambientintelligence:industrialresearchonavisionaryconcept,”inProceedingsofthe2003internationalsymposiumonLowpowerelectronicsanddesign,pp.247–251,ACMPress,2003.[6]Pescovitz,D.,“Thepowerofsmalltech..”Smalltimes,Vol.2,No.1,2002.[7]“Darpaenergyharvestingprojects.”http://www.darpa.mil/dso/trans/energy/projects.html.[8]Kymisis,J.,C.Kendall,J.Paradiso,andN.Gershenfeld,“Parasiticpowerharvestinginshoes,”inSecondIEEEInternationalConferenceonWearableComputing,(ISWC),pp.132–139,IEEEComputerSocietyPress,October1998.17[9]Shenck,N.S.,andJ.A.Paradiso,“Energyscavengingwithshoe-mountedpiezoelectrics,”IEEEMicro,Vol.21,May-June2001,pp.30–42.[10]Paradiso, J.A.,andM.Feldmeier,“Acompact,wireless,self-poweredpushbuttoncontroller,”inACMUbicomp,(Atlanta,GA,USA),pp.299–304,Springer-VerlagBerlinHeidelberg,September2001.[11]Starner,T.,“Human-poweredwearablecomputing,”IBMSystemsJournal,Vol.35,No.3-4,1996.[12]Williams,C.B.,C.Shearwood,M.A.Harradine,P.H.Mellor,T.S.Birch,andR.B.Yates,“Devlopmentofanelectromagneticmicro-genrator,”IEEEProceedings-Circuits,DevicesandSystems,Vol.148,December2001,pp.337–342.[13]Ottoman,G.K.,H.F.Hofmann,andG.A.Lesieutre,“Optimizedpiezoelectricenergyharvestingcircuitusingstep-downconverterindiscontinuousconductionmode,”IEEETransactionsonPowerElectron-ics,Vol.18,No.2,2003,pp.696–703.[14]Glynn-Jones,P.,S.P.Bebby,E.P.James,andN.M.White,“Themodelingofapiezoelectricvibrationpoweredgeneratorformicrosystems,”inTransducers01/EurosensorsXV,June2001.[15]Amitharajah,R.,andA.P.Chandrakasan,“Selfpoweredsignalprocessingusingvibration-basedpowergeneration,”IEEEJournalofSolidStateCircuits,Vol.33,No.5,1998,pp.687–695.[16]Shearwoo, C., andR.B.Yates,“Developmentofanelectromagneticmicrogenerator,”IEEElectronicsLetters,Vol.33,October1997,pp.1883–4.[17]“Exploringmars:Blowinginthewind.”http://www.jpl.nasa.gov/technology/features/tumbleweed.html,2001.[18]Whalen,S.,M.Thompson,D.Bahr,C.Richards,andR.Richards,“Design,fabricationandtestingofthep3microheatengine,”SensorsandActuators,Vol.104,No.3,pp.200–208.[19]“HeliomoteR.”http://cvs.nesl.ucla.edu/cvs/viewcvs.cgi/HeliomoteR/.CVSRepository.18[20]Kansal,A.,D.Potter,andM.Srivastava,“Performanceawaretaskingforenvironmentallypoweredsensornetworks,”inACMJointInternationalConferenceonMeasurementandModelingofComputerSystems(SIGMETRICS),June2004.[21]Simunic,T.,H.Vikalo,P.Glynn,andG.D.Micheli,“Energyefficientdesignofportablewirelesssys-tems,”inACM/IEEEInternationalSymposiumonLowPowerElectronicDesign,2000.[22]Min,R.,T.Furrer,andA.Chandrakasan,“Dynamicvoltagescalingtechniquesfordistributedmicrosen-sornetworks,”inIEEEComputerSocietyWorkshoponVLSI,pp.43–46,April2000.[23]Schuette, M.A.,andJ.R.Barr,“Embeddedsystemsdesignforlowenergyconsumption,”inProceed-ingsofthe1994IEEE/ACMinternationalconferenceonComputer-aideddesign,pp.534–540,IEEEComputerSocietyPress,1994.[24]Simunic,T.,L.Benini,G.D.Micheli,andM.Hans,“Sourcecodeoptimizationandprofilingofenergyconsumptioninembeddedsystems,”inIEEEInternationalSymposiumonSystemSynthesis(ISSS),2000.[25]Kansal, A.,andM.B.Srivastava,“Anenvironmentalenergyharvestingframeworkforsensornetworks,”inInternationalsymposiumonLowpowerelectronicsanddesign,pp.481–486,ACMPress,2003.[26]Voigt,T.,H.Ritter,andJ.Schiller,“Utilizingsolarpowerinwirelesssensornetworks,”inThe28thAn-nualIEEEConferenceonLocalComputerNetworks(LCN),(Bonn/Konigswinter,Germany),October2003.[27]Rong, P.,andM.Pedram,“Extendingthelifetimeofanetorkofbatterypoweredmobiledevicesbyremoteprocessing:Amarkoviandecisionbasedapproach,”inACMDesignAutomationConference(DAC),(Anaheim,CA,USA),June2003.19[28]Chang, J.H.,andL.Tasiulas,“Maximumlifetimeroutinginwirelesssensornetworks,”inAdvancedTelecommunicationsandInformationDistributionResearchProgram(ATIRP),(CollegePark,MD,USA),March2000.[29]Li,Q.,J.Aslam,andD.Rus,“Onlinepowerawareroutinginwirelessad-hocnetworks,”inACMSIGMOBILE,(Rome,Itay),July2001.[30]Xu,Y.,J.Heidemann,andD.Estrin,“Geography-informedenergyconservationforadhocrouting,”inProc.ACMMobicom,pp.70–84,ACMPress,2001.[31]Schurgers,C.,V.Tsiatsis,S.Ganeriwal,andM.Srivastava,“Optimizingsensornetworksintheenergy-density-latencydesignspace,”IEEETransactionsonMobileComputing(TMC),January-March2002,pp.70–80.[32]Intanagonwiwat,C.,R.Govindan,andD.Estrin,“Directeddiffusion:Ascalableandrobustcommuni-cationparadigmforsensornetworks,”inACMMobicom,August2000.20