Telecommunication

GoodforWirelessMulti-hopNetworks?󰀁

MartinKubisch,HolgerKarl,andAdamWolisz

TelecommunicationNetworksGroupTechnischeUniversit¨atBerlinSekr.FT5-2,Einsteinufer2510587Berlin,Germany

{kubisch,karl,wolisz}@ee.tu-berlin.de

Abstract.Ithasbeenshownthatoptimizationofwirelessnetworkoperationscanbeachievedbytransmissionpowercontrolofwirelessnodes.Controllingthetransmissionpowerismostlyperformedasaregulationofthefinalpoweramplifier,whichisthemajorpowersink.Dependingonthedesiredtransmissionpowerlevel,currentlypopularamplifiersworkatdifferentlevelsofefficiency:highestincaseofmaximumemittedpower,lowerincaseofreducedemittedpower.

Inthispaperwepresentanovelapproachforimprovingtheliftimeofmobilede-vicesinmulti-hopwirelessnetworksbyusingnodeswithheterogeneousefficiencycharacteristicsoftheirpoweramplifiers.

Toassessthepotentialofsuchamplifier-heterogeneousnetworks,wehaveperformedmulti-hopnetworksimulationsundertheidealisticassumptionofoptimalroutingbasedonglobalknowledge.Theobtainedresultsdemonstrateanappealingpotentialofareductioninenergyconsumptiontolessthan40%comparedtoclassicalnetworkswithuniformamplifiers.

Keywords:Multi-hopwirelessnetworks,powercontrol,sensornetwork.

1Introduction

Today’spopularpoweramplifiersusedinwirelessnetworkcardsaredesignedtohavethehighestpowerefficiencyatthemaximumoutputpower.Whentheoutputpowerisreduced,thepowerefficiencyoftheamplifierdecreases,i.e.,thepowerconsumedbyanamplifierdoesnotreducewiththesameratioastheoutputpowerdecreases.Asanexample:theRF2155poweramplifier[1],whichisdesignedforcordlessphonesandotherapplicationinthe915MHzISMband,hasfourdifferentoutputpowerlevels.Thepowerefficiencyofthesepowerlevelrangefrom54%efficiencyforthehighestlevelofoutputpowerto1%efficiencyforthelowestlevelofoutputpower.

Poweramplifiersareusuallychosenaccordingtothemaximumrangeonewantstoovercome.AssumingadesiredPERandotherparametersofthereceivercharacteristic,thisleadstoarequirementontheoutputpowernecessarytoovercomethemaximal

󰀁

ThisworkisfundedbytheGermanMinistryofEducationandResearch(BMBF)undertheprojectIBMS2.

M.Contietal.(Eds.):PWC2003,LNCS2775,pp.841–846,2003.cIFIPInternationalFederationforInformationProcessing2003󰀁

842M.Kubisch,H.Karl,andA.Wolisz

distance.Inrealisticsetupshowever,thedistancesbetweencommunicatingnodesarevariable.Thus,theusageofthehighestoutputpowerisrarelynecessary,evenwhendirectcommunicationbetweenthesenderandreceiverispossible.Furtheroptionsforusinglowertransmissionpowerappearwhen(asinIEEE802.11)thecoding/modulationmightbedynamicallyadjusted(slowerdataratesrequirelowertransmissionpowertoachievethesamePER).Inaddition,theusageofmulti-hopschemesinsteadofdirectcommunicationoffersevenmorefreedomforusingloweroutputpower.Itiswellknownthatreducingtheoutputpowerisbeneficialbecauseofbothreductionoftheenergyofthetransmitteraswellasreductionofthepotentialinterferencewithothernodesoperatinginthesamefrequencyband.

Thedesiredandbeneficialreductionoftheoutputpowerdoes,unfortunatelyenough,notleadtoaproportionalreductionofpowerusedtodrivetheamplifier,astheamplifierismovingintoalessefficientoperationrange.But,infact,thereisnophysicalrulethatmandatesthatpoweramplifiershavethehighestefficiencyatthehighestoutputpower.ApracticalexampleforadifferentamplifierdesignsisCripps[2],whichpresentsamplifierswhicharedevelopedwithefficiencyenhancementtechniquesinmind.Thepresented“Doherty”amplifierisanexampleofanadaptationofpowerefficiency.Ithasthehighestpowerefficiencyatapowerlevel6dBlessthanthemaximumone.Obviously,usingsuchamplifierswithshiftedpowerefficiencywouldpayoffif—mostly—loweroutputpowerisapplied,evenifun-proportionallyhighpowerconsumptionforthehighestpowerlevelwouldbeused.

Asthiscouldbeextendedwiththeuseofdynamicallyadjustablecharacteristicsofpoweramplifierstoresultinanoptimalselectionforaspecificnetworktopology,weconstrainourselvesonlytonodeswithtwotypesofamplifiers,furtherreferredtoasshort-orlong-rangespecialists(notethateventheshort-rangespecialistsinourmodelcantransmitatthehighestpowerifnecessary,butthecostsarecomparablyhigh).Intuitively,suchamixtureofnodesmightimproveenergyefficiency.Inthispaper,wewanttoassesswhetherthisintuitionactuallyholds:westudyavaryingpercentageoflong-andshort-rangespecialistnodesatvariousnodedensitiesandstudytheresultingenergyefficiencyinawirelessnetwork.

Ourconsiderationsarestructuredasfollows:firstwegiveashortoverviewofre-latedworkprovingthatadifferentiationofoutputpowerlevelisanattractivewaytooptimizewirelessnetworks.Furtherwedescribethemethodologyofthisstudy,detailsofoursimulationsetupandsomeencouragingresults.Asthispaperpertainstoworkinprogress,wecompleteourconsiderationswithcommentsonongoinginvestigations.

2RelatedWork

Thereexistnumerouspossibilitiesforreductionofthecommunicationenergyinmulti-hopwirelessnetworks.Wewillnotdiscussthepaperswhichputtheirnodesintosleep,evenifthisisaninterestingfunctionalityofsomeMACdesigns(includingIEEE802.11);itcanbecombinedwiththeapproachdiscussedinthispaper.

Relevanttothispaperareapproachespertainingtotheselectionofthetransmissionpowerlevel.Thismightbegloballyunifiedforthewholenetwork,e.g.withrespecttodesiredconnectivity[3][4]orindividuallyforeachpairofnodes[5][6][7][8][9].[10]

AreClassesofNodeswithDifferentPowerAmplifiersGood843

Asshownin[5]and[3],theuseoflowerpowerlevel,whichshortenthedistanceandrequiremoreintermediatehops,canincreasetheenergyefficiencyaswellasthecapacityofanetwork.Especiallytheconsiderationoftheremainingbatterycapacitycanleadtoanextensioninnetworklifetime.AnotherapproachistheCOMPOWprotocol[4]whichexchangeslifemessagesonseparate,discretepowerlevelstofindacommonpowerlevelthroughoutthenetwork.UnlikethesenetworksbasedonIEEE802.11,isthesearchfortheminimumpowerlevelofanode[6]amoregeneralapproach,whichcanbeappliedtolowdataratenetworksaswell,e.g.,sensornetworks.Theydeterminethepowerlevelwhilerelyingonmeasurementsofthereceivedsignalpowerstrength.Additionallytothispowerlevelcalculation,[7]usesapathrecalculationineverynodetodetermineifitcouldprovideamoreenergy-efficientpath.

Aprovenwayoffindingaminimumtransmissionpowerisdonein[9]and[10].Whilethesealgorithmsrelyeitheronlocationinformationoronangleofarrival,thealgorithmsin[8]donot.There,locallyavailableinformationisusedtoadaptthepowerlevelinansensornetworkanditisshownthatthisincreasesthenetworklifetime.

Outofthispartiallistonecanseethatsettingdifferenttransmissionpowerlevelsisbeneficialfortheenergyconsumptionandotheroptimizationcriteria.Whenoptimizingthepowersettingitisimportanttounderstandthatmostlytheemittedpower(assuringacertainSNRatthereceiver,definedbyadesiredPER)isusedastheoptimizationcriterion.

Inallofthesepapersithasbeenassumedthattheparametersoftheamplifierofallnodesareidentical,andonlyindividualsettingsarepossible.Toourknowledge,thisisthefirstpaperconsideringamixtureofnodeswithdifferenttypesofamplifiercharacteristics.

3EvaluationBasedonGlobalKnowledge

Thesystemscenariounderconsiderationisawirelessmulti-hopnetwork,e.g.,amobileadhocnetwork.Nodeswanttocommunicatewithothernodesoverlongerdistancesthanispossibleevenwiththehighesttransmissionpower,necessitatingmulti-hopcommu-nication.Forsuchnetworksweneedanadhocroutingprotocolwhichchoosesenergy-efficientroutesandthusthenodescanusetransmissionpowerlevelsmallenoughtosafelycommunicatewiththeirnexthopneighbor.

Theadvantageofsuchenergy-efficientroutingschemesisthatthereductionofdistance,usefulformulti-hopcommunication,impliesareductionofnecessarytrans-missionpower,butastheefficiencyofthefinalamplifierdoesdecreaseaswell,thegainisnotashighasitcouldbe.Ourhypothesisisthatthisshortcomingcanbeovercomebyusingdifferentnodesinthenetwork,equippedwithdifferentpoweramplifiers,opti-mizedfordifferentdistances.Weclaimthatsuch“specialists”forlong-andshort-rangecommunicationwillincreasetheenergyefficiency.

Ideally,anadhocroutingprotocolshouldbemodifiedtoexploitthedifferentcharac-teristicsofsuchnodes,routingshort-rangecommunicationovernodeswithamplifiersthatoperateefficientlyatlowtransmissionpowersandbridginglongdistanceswithothernodes.Developingsuchanextensiontoanadhocprotocolisactuallyanon-trivial

844M.Kubisch,H.Karl,andA.Wolisz

activity.Hence,tofirstobtainanunderstandingoftheprinciplefeasibilityandbenefitsofthisapproach,weusedasimplifiedapproachandexamineit.

Inourapproachwehaveafixednumberofuniformrandomlydistributednodes,whereeachnodehasonlytwooutputpowerlevels(theoutputpowerlevelsarethesameforshort-andlong-rangespecialists),andwechangethepercentageofnodeswhichareshort-rangespecialists.Foreachnodeandeverypowerlevelinsuchanetworkthemaximumtransmissionrangeiscalculated.Usingtheglobalknowledgeofallnode’slocation,wedeterminewhichnodecanbereachedandannotatetheedgesbetweenthenodeswiththeenergyconsumedperpacketasthecostfunction.

ThenweuseDijkstra’salgorithmtocalculatetheoptimalpathbetweenanytwonodesinthenetwork,resultinginaforwardingtableforeverynodewhichalsocontainsthenecessarytransmissionpowerlevel.Havingthesepaths,weusedthefollowingtrafficpatterntodeterminethetotalenergyused:Everynoderandomlyselectsonedestinationnodeinthenetworkandtransfersonepackettoit.Attheend,thetotalconsumedenergyofallnodesservesasafigureofmeritfordifferentnetworkdensitiesandratiooflong-/short-rangespecialists.

4ModelAssumptions

Wehaveconsideredvariousmodelassumptionshere.Inallscenarios,100nodesareuniformrandomlydistributedinanareabetween5km∗5kmand18.25km∗18.25km.Thisisequaltonetworkdensitiesbetween4(allnodescancommunicatedirectly)and0.3nodesperkm2(lowervaluesresultinpartitionednetworks).TheefficiencymodelforthetransmissionpowerisbasedonvaluesfortheRF2155poweramplifier:Eachnodehastwotransmissionpowerlevelsof70mWand447mW.Thelong-rangespecialistshaveanefficiencyof54%(consumedpowerof826mW)forthehighpowerleveland20%(consumedpowerof337.5mW)forthelowpowerlevel.Theshort-rangespecialistshaveanefficiencyof20%(consumedpowerof2235mW)forthehighpowerleveland54%(consumedpowerof129.6mW)forthelowpowerlevel.Theothervaluesof-85dBmreceiversensitivity(implyingaPERof1%),200mWreceptionpowerand200mWcomputationpowerwhiletransmittingarebasedonthe“SieMoS50037BluetoothModule”[11].Theamplifiercharacteristicusedinaparticularnodeinaparticularscenariodependsonthepercentagenecessaryforthisscenario,e.g.,outofthis100nodes10areselectedtobeshort-rangespecialiststheothersarelong-range.Forthetrafficweusedapacketsizeof1500byteaswellasaimmediateacknowledgementof30bytesatadatarateof1Mbit/s—takingintoaccounttheacknowledgmentaswellisimportantbecauseoftheheterogeneousenergycostsofdifferentdevices!

5SimulationResults

Figure1(a)displaysthetotalenergyconsumed,eachaveragedover40differentrandomplacementsofnodes.Apathlosscoefficientof3wasused.Onthex-axis,thepercentageofshort-rangespecialistsisdisplayedandonthey-axistheenergynecessaryfortheusedtrafficpatternwhichresultsinatransferof100packets.Thelowerlineistheenergy

AreClassesofNodeswithDifferentPowerAmplifiersGood

2.5Energy needed for communication [J]Energy needed for communication [J] 7 6 5 4 3 2 1 0

0.6 nodes/(km*km)

0.3 nodes/(km*km)0.4 nodes/(km*km)0.5 nodes/(km*km)845

2

1.5

0.75 nodes/(km*km)1 node/(km*km)2 nodes/(km*km) 1

0.5

4 nodes/(km*km)

0

0 20 40 60 80 100Percentage of nodes with shifted power consumption [%]

0

20 40 60 80 100

Percentage of nodes with shifted power consumption [%]

(a)Highdensity(b)Lowdensity

Fig.1.Totalconsumedenergyoverpercentageofshortrangespecialists

averageforadensityof4nodesperkm2,theupperonefor0.75nodesperkm2.Thelinesbetweenareintermediatedensities.

Forthehigh-densitynetworks,usingonlyshort-rangespecialistsisbeneficialandtheenergyneededislessthen38.4%comparedtoanetworkhavingonlylong-rangeones.

Figure1(b)alsodisplaysthetotalenergyconsumedoverthepercentageofshort-rangespecialists,butthenetworksarelessdense.Thelowerlineistheenergyaverageforadensityof0.6nodesperkm2,theupperonefor0.3nodesperkm2.

Thethreelowerlinesaretheenergycurveswherethenetworkdensityissparseandusingonlyshort-rangespecialistsisnotmostbeneficial.Instead,thereisanoptimalpointdependingonthedensity.Foradensityof0.6nodesperkm2thisratiois80%,for0.5nodesperkm2theratioisaround70%andfor0.4nodesperkm2itisaround30%.Whenthedensityis0.3nodesperkm2orsmaller,itisnotbeneficialtouseshort-rangespecialistsatall.Additionally,furtherreductionindensityleadstoanotherproblem:theconnectivity.Whenamuchlowerdensityisused,theprobabilityofhavingadisconnectednetworkbecomeshigher,thusfurthercurvesareleftout.

6ConclusionsandFurtherWork

Astheresultsshow,heterogeneityofnodeswithdifferentlyoptimizedpoweramplifiersarebeneficialforenergyefficiency.Thebestratiobetweenlong-andshort-rangespe-cialistsdependsonthedensityofthenetwork,thenodelayout(onecouldimagineanetworkwhereparticularnodesareofbetterusewhentheyadapttheiramplifierchar-acteristic)andalso,implicitly,onthecharacteristicsofthedifferentamplifiers.Thus,itisacomplexproblem,butweintendtotackleitstepbystep.

AlthoughthisideawasdemonstratedforWLANnetworks,webelieveitcouldalsobeappliedtootherwirelessnetworks,e.g.,sensornetworks.Inthesenetworkstheproblemisquitesimilarapartfromthefactthatotherpacketsizesandtrafficpatternsareused.Asanextstepweintendtointegratethecharacteristicsofsuchdifferentshort-andlong-rangespecialistsintoadhocroutingprotocols.Theparticularchallengeistohandlethedifferentenergycostsforbothdirectionsinasimpleandefficientmanner.Moreover,

846M.Kubisch,H.Karl,andA.Wolisz

theapplicationofsuchaheterogeneousnodesapproachshouldbeparticularlyusefulinwirelesssensornetworks,wherethedensityofthenetworkcanbeestimatedbeforehand.

Insuchanetwork,alsotheuseofcluster-basedroutingprotocolsisverypopular,anditmightbeattractivetousethelong-rangespecialistsasclusterheads.

Acknowledgement.WethankProf.G.B¨ockandW.Chenfortheirdiscussionregardingpoweramplifiers.

References

1.RFMicroDevices:RF21553VProgrammableGainPowerAmplifier.Designer’sHandbook(2003)2–173–2–180

2.Cripps,S.C.:RFPowerAmplifiersforWirelessCommunications–EfficiencyEnhancementTechniques.ArtechHousePublishers,Boston,MA,USA(2000)

3.Monks,J.P.,Ebert,J.P.,Wolisz,A.,Hwu,W.:Astudyoftheenergysavingandcapacityimprovementpotentialofpowercontrolinmulti-hopwirelessnetworks.In:WorkshoponWirelessLocalNetworks,Tampa,Florida,USA,alsoConf.ofLocalComputerNetworks(LCN).(2001)

4.Narayanaswamy,S.,Kawadia,V.,Sreenivas,R.S.,Kumar,P.R.:Powercontrolinad-hocnetworks:Theory,architecture,algorithmandimplementationoftheCOMPOWprotocol.In:EuropeanWirelessConference,Florence,Italy,EUREL,VDE(2002)156–162

5.Kubisch,M.,Mengesha,S.,Hollos,D.,Karl,H.,Wolisz,A.:Applyingad-hocrelayingtoimprovecapacity,energyefficiency,andimmissionininfrastructure-basedWLANs.InIrmscher,K.,ed.:KommunikationinVerteiltenSystemen(KiVS),13.ITG/GI-Fachtagung,Leipzig,Germany,InformatikAktuell,Springer(2003)

6.Bergamo,P.,Maniezzo,D.,Giovanardi,A.,Mazzini,G.,Zorzi,M.:Distributedpowercon-trolforpower-awareenergy-efficientroutinginadhocnetworks.In:EuropeanWirelessConference,Florence,Italy,EUREL,VDE(2002)237–243

7.Doshi,S.,Bhandare,S.,Brown,T.X.:Anon-demandminimumenergyroutingprotocolforawirelessadhocnetwork.MobileComputingandCommunicationsReview6(2002)50–668.Kubisch,M.,Karl,H.,Wolisz,A.,Zhong,L.C.,Rabaey,J.:Distributedalgorithmsfortrans-missionpowercontrolinwirelesssensornetworks.In:WirelessCommunicationsandNet-workingConference(WCNC),NewOrleans,LA,IEEE(2003)

9.Rodoplu,V.,Meng,T.H.Y.:Minimumenergymobilewirelessnetworks.IEEEJournalonSelectedAreasinCommunications17(1999)1333–1344

10.Wattenhofer,R.,Li,L.,Bahl,P.,Wang,Y.M.:Distributedtopologycontrolforpowereffi-cientoperationinmultihopwirelessadhocnetworks.In:INFOCOM–TheConferenceonComputerCommunications,Anchorage,Ak,IEEE(2001)

11.SiemensAGAustria:SiemensS50037BluetoothModule(2001)