Sensors, Circuits and Instrumentation Systems

《Sensors, Circuits and Instrumentation Systems》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

OlfaKanoun,FaouziDerbelandNabilDerbel(Eds.)Sensor,CircuitsandInstrumentationSystems

1AdvancesinSystems,SignalsandDevices|EditedbyOlfaKanoun,UniversityofChemnitz,GermanyVolume2

2Sensor,CircuitsandInstrumentationSystems|EditedbyOlfaKanoun,FaouziDerbelandNabilDerbel

3EditorsofthisVolumeProf.Dr.-Ing.OlfaKanounProf.Dr.-Eng.NabilDerbelTechnischeUniversitätChemnitzUniversityofSfaxChairofMeasurementandSensorTechnologySfaxNationalEngineeringSchoolReichenhainerStrasse70Control&EnergyManagementLaboratory09126Chemnitz1173BP,3038SFAX,Tunisiaolfa.kanoun@etit.tu-chemnitz.den.derbel@enis.rnu.tnProf.Dr.-Ing.FaouziDerbelLeipzigUniversityofAppliedSciencesChairofSmartDiagnosticandOnlineMonitoringWächterstrasse1304107Leipzig,Germanyfaouzi.derbel@htwk-leipzig.deISBN978-3-11-046819-9e-ISBN(PDF)978-3-11-047044-4e-ISBN(EPUB)978-3-11-046849-6Set-ISBN978-3-11-047045-1ISSN2365-7493e-ISSN2365-7507LibraryofCongressCataloging-in-PublicationDataACIPcatalogrecordforthisbookhasbeenappliedforattheLibraryofCongress.BibliographicinformationpublishedbytheDeutscheNationalbibliothekTheDeutscheNationalbibliothekliststhispublicationintheDeutscheNationalbibliografie;detailedbibliographicdataareavailableontheInternetathttp://dnb.dnb.de.©2017WalterdeGruyterGmbH,Berlin/BostonTypesetting:Konvertus,HaarlemPrintingandbinding:CPIbooksGmbH,Leck♾Printedonacid-freepaperPrintedinGermanywww.degruyter.com

4PrefaceoftheVolumeEditorThesecondvolumeoftheSeries“AdvancesinSystems,SignalsandDevices”(ASSD),containspeerreviewedinternationalscientificarticlesdevotedtothefieldofsensors,circuitsandinstrumentationsystems.Thescopeofthevolumeencompassesallaspectsofresearch,developmentandapplicationsofthescienceandtechnologyinthesefields.Thetopicsincludesensorsandmeasurementsystems,opticalsensors,chemicalsensors,mechanicalsensors,inductivesensors,capacitivesensors,mi-crosensors,thermalsensors,biomedicalandenvironmentalsensors,fexiblesensors,nanosensors,microelectronicsystems,nanosystemsandnanotechnology,sensorsignalprocessing,sensorinterfaces,modeling,dataacquisition,multisensordatafusion,distributedmeasurements,devicecharacterizationandmodeling,customandsemicustomcircuits,analogcircuitdesign,low-voltage,low-powerVLSIdesign,circuittest,packagingandreliability,impedancespectroscopy,wirelesssensors,wirelessinterfaces,wirelesssensornetworks,energyharvesting,circuitsandsystemsdesign.Thesefieldsareaddressedbyaseparatevolumeoftheseries.Allvolumesareeditedbyaspecialeditorialboardmadeupbyrenownedscientistfromallovertheworld.Authorsareencouragedtosubmitnovelcontributionswhichincluderesultsofresearchorexperimentalworkdiscussingnewdevelopmentsinthefieldofsensors,circuitsandinstrumentationsystems.Theseriescanbealsoaddressedforeditingspecialissuesfornoveldevelopmentsinspecificfields.Guesteditorsareencouragedtomakeproposalstotheeditorinchiefofthecorrespondingmainfield.Theaimofthisinternationalseriesistopromotetheinternationalscientificprogressinthefieldsofsystems,signalsanddevices.ItprovidesatthesametimeanopportunitytobeinformedaboutinterestingresultsthatwerereportedduringtheinternationalSSDconferences.Itisabigpleasureofourstoworktogetherwiththeinternationaleditorialboardconsistingofrenownedscientistsinthefieldofsensors,circuitsandinstrumentationsystems.TheEditorsOlfaKanoun,FaouziDerbelandNabilDerbel

5

6AdvancesinSystems,SignalsandDevicesSeriesEditor:Prof.Dr.-Ing.OlfaKanounTechnischeUniversitätChemnitz,Germany.olfa.kanoun@etit.tu-chemnitz.deEditorsinChief:Systems,Automation&ControlProf.Dr.-Eng.NabilDerbelENIS,UniversityofSfax,Tunisian.derbel@enis.rnu.tnPowerSystems&SmartEnergiesProf.Dr.-Ing.FaouziDerbelLeipzigUniv.ofAppliedSciences,Germanyfaouzi.derbel@htwk-leipzig.deCommunication,SignalProcessing&InformationTechnologyProf.Dr.-Ing.FaouziDerbelLeipzigUniv.ofAppliedSciences,Germanyfaouzi.derbel@htwk-leipzig.deSensors,Circuits&InstrumentationSystemsProf.Dr.-Ing.OlfaKanounTechnischeUniversitätChemnitz,Germanyolfa.kanoun@etit.tu-chemnitz.de

7EditorialBoardMembers:Systems,Automation&ControlDumitruBaleanu,ÇankayaUniversity,Ankara,TurkeyRidhaBenAbdennour,EngineeringSchoolofGabès,TunisiaNaceurBenhadj,Braïek,ESSTT,Tunis,TunisiaMohamedBenrejeb,EngineeringSchoolofTunis,TunisiaRiccardoCaponetto,Universita’degliStudidiCatania,ItalyYangQuanChen,UtahStateUniversity,Logan,USAMohamedChtourou,EngineeringSchoolofSfax,TunisiaBoutaïebDahhou,Univ.PaulSabatierToulouse,FranceGérardFavier,UniversitédeNice,FranceFlorinG.Filip,RomanianAcademyBucharestRomaniaDorinIsoc,Tech.Univ.ofClujNapoca,RomaniaPierreMelchior,UniversitédeBordeaux,FranceFaïçalMnif,SultanqabousUniv.Muscat,OmanAhmetB.Özgüler,BilkentUniversity,Bilkent,TurkeyManabuSano,HiroshimaCityUniv.Hiroshima,JapanAbdul-WahidSaif,KingFahdUniversity,SaudiArabiaJoséA.TenreiroMachado,EngineeringInstituteofPorto,PortugalAlexanderPozniak,InstitutoPolitecniko,NationalMexicoHerbertWerner,Univ.ofTechnology,Hamburg,GermanRonaldR.Yager,Mach.IntelligenceInst.IonaCollegeUSABlasM.Vinagre,Univ.ofExtremadura,Badajos,SpainLotfiZadeh,Univ.ofCalifornia,Berkeley,CA,USAPowerSystems&SmartEnergiesSylvainAllano,EcoleNormaleSup.deCachan,FranceIbrahimBadran,PhiladelphiaUniv.,Amman,JordanRonnieBelmans,UniversityofLeuven,BelgiumFrdéricBouillault,UniversityofParisXI,FrancePascalBrochet,EcoleCentraledeLille,FranceMohamedElleuch,TunisEngineeringSchool,TunisiaMohamedB.A.Kamoun,SfaxEngineeringSchool,TunisiaMohamedR.Mékidèche,UniversityofJijel,AlgeriaBernardMulton,EcoleNormaleSup.Cachan,FranceFrancescoParasiliti,UniversityofL’Aquila,ItalyManuelPérez,Donsión,UniversityofVigo,SpainMichelPoloujadoff,UniversityofParisVI,FranceFrancescoProfumo,PolitecnicodiTorino,ItalyAlfredRufer,EcolePolytech.Lausanne,SwitzerlandJunjiTamura,KitamiInstituteofTechnology,Japan

8Communication,SignalProcessing&InformationTechnologyTilAach,AchenUniversity,GermanyKasimAl-Aubidy,PhiladelphiaUniv.,Amman,JordanAdelAlimi,EngineeringSchoolofSfax,TunisiaNajouaBenamara,EngineeringSchoolofSousse,TunisiaRidhaBouallegue,EngineeringSchoolofSousse,TunisiaDominiqueDallet,ENSEIRB,Bordeaux,FranceMohamedDeriche,KingFahdUniversity,SaudiArabiaKhalifaDjemal,Universitéd’Evry,Vald’Essonne,FranceDanielaDragomirescu,LAAS,CNRS,Toulouse,FranceKhalilDrira,LAAS,CNRS,Toulouse,FranceNoureddineEllouze,EngineeringSchoolofTunis,TunisiaFaouziGhorbel,ENSI,Tunis,TunisiaKarlHolger,UniversityofPaderborn,GermanyBertholdLankl,Univ.Bundeswehr,München,GermanyGeorgeMoschytz,ETHZürich,SwitzerlandRaduPopescu-Zeletin,FraunhoferInst.Fokus,Berlin,GermanyBaselSolimane,ENST,Bretagne,FrancePhilippeVanheeghe,EcoleCentraledeLilleFranceSensors,Circuits&InstrumentationSystemsAliBoukabache,Univ.Paul,Sabatier,Toulouse,FranceGeorgBrasseur,GrazUniversityofTechnology,AustriaSergeDemidenko,MonashUniversity,Selangor,MalaysiaGerhardFischerauer,UniversitätBayreuth,GermanyPatrickGarda,Univ.Pierre&MarieCurie,Paris,FranceP.M.B.SilvaGirão,Inst.SuperiorTécnico,Lisboa,PortugalVoicuGroza,UniversityofOttawa,Ottawa,CanadaVolkerHans,UniversityofEssen,GermanyAiméLayEkuakille,UniversitàdegliStudidiLecce,ItalyMouradLoulou,EngineeringSchoolofSfax,TunisiaMohamedMasmoudi,EngineeringSchoolofSfax,TunisiaSubhaMukhopadhyay,MasseyUniversityTuritea,NewZealandFernandoPuenteLeón,TechnicalUniv.ofMünchen,GermanyLeonardReindl,Inst.Mikrosystemtec.,FreiburgGermanyPavelRipka,Tech.Univ.Praha,CzechRepublicAbdulmotalebElSaddik,SITE,Univ.Ottawa,Ontario,CanadaGordonSilverman,ManhattanCollegeRiverdale,NY,USARachedTourki,FacultyofSciences,Monastir,TunisiaBernhardZagar,JohannesKeplerUniv.ofLinz,Austria

9

10ContentsPrefaceoftheVolumeEditor|VS.Kehrberg,P.Wellner,C.GeckelerandJ.MehnerModalAnalysisofMEMSUsingUltrasonicBaseExcitation|1M.Saihi,B.Boussaid,A.ZouinkhiandM.N.AbdelkrimDecentralizedFaultDetectioninWirelessSensorNetworkbasedongaussianfunctionerror|13B.Mezghani,F.TounsiandM.MasmoudiStaticBehaviorAnalyticalandNumericalAnalysisofMicromachinedThermalAccelerometers|27M.HadjSaid,F.Tounsi,P.Gkotsis,M.MasmoudiandL.A.FrancisMEMS-BasedClamped-ClampedBeamResonatorCapacitiveMagnetometer|47G.U.GammandL.M.ReindlRangeExtensionforSingleHopWirelessSensorNetworkswithWake-upReceivers|63A.Ghorbel,M.Jallouli,L.AmouriandN.BenAmorAHW/SWImplementationonFPGAofAbsoluteRobotLocalizationUsingWebcamData|75

11

12S.Kehrberg,P.Wellner,C.GeckelerandJ.MehnerModalAnalysisofMEMSUsingUltrasonicBaseExcitationAbstract:Wedevelopedanewtestfacilitybasedontheprincipleofpiezoelectricbaseexcitationinordertomeettherequirementsformodalanalysisofamicromachinedangularratesensoratveryhighfrequencies.Acombinationofa0.5mmthinpiezoelectricdisc,aseismicmassandadampingelementmountedinavacuumchamberresultsinalinearfrequencyresponseoverafrequencyrangefrom10kHzto200kHz.Furthermore,thetestfacilitycanbeusedforfindingeigenmodesofmicroelectromechanicalsystems(MEMS)upto1MHz.Keywords:MEMS,Modalanalysis,Ultrasonicexcitation,Piezoelectric.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A221IntroductionAtypicalMEMS,likeanangularratesensororaccelerometer,canbedescribedasacoupledspring-masssystem.ModalfrequenciesofMEMScanbemeasuredmorepreciselythangeometryparameters.Therefore,thepreferredwayformodelvalidationandparameterextractionofsuchsystemsistheexperimentalmodalanalysis[1].Micromachinedstructureswithwidthsofafewmicrometerstypicallyhaveeigenfrequenciesabove100kHz.ForMEMStheexcitationisoftendoneelectrostaticallyduetotheeasierimplementationandwidebandwidth[2].Inthiscase,themeasuredfrequencyresponseincludesnonlinearitiesoftheelectrostaticforcesandcrosstalkeffects[3].IftheinterestisinapuremechanicalanalysisofaMEMSdevice,includingmechanicalnonlinearities,vibrationexcitationisneeded.Forthisanalysis,ashakerwithacleanout-of-planemovement,withoutresonancefrequenciesintheinterestedbandwidth,isrequiredtoexcludenonlinearityeffectsfromthebaseexcitation.Standardmeasurementequipments(e.g.electromagneticshakers)formacroscaledevicesarenotentirelysuitableforthistasksincetheylacktheappropriatebandwidth.References[4,5]alreadyshowedthecapabilityofthickerpiezoelectriccuboids/discsasout-of-planeshakersforlowerfrequenciesuptorespectively100kHzand50kHz.S.Kehrberg,P.Wellner,C.Geckeler:AutomotiveElectronics,RobertBoschGmbH,D-72762Reutlin-gen,Germany,email:steven.kehrberg@de.bosch.com.J.Mehner:DepartmentofMicrosystemsandPrecisionEngineering,ChemnitzUniversityofTechnology,D-09107Chemnitz,GermanyDeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.1–11.DOI10.1515/9783110470444-001

132|S.Kehrbergetal.Thedevelopedtestfacilitydescribedinthispaperusesathinnerpiezoelectricdiscandadifferentsubstructuretoallowmeasurementsintherangefrom10upto1000kHz.Thisgivestheopportunitytoanalyzemoremodesandleadstoamorepreciseparameterextraction.AwiderbandwidthisespeciallyimportanttomeetthetrendofsmallerMEMSwithhighereigenfrequencies[6].Fig.1.Angularratesensorgluedonpiezoelectricdisc.2TestFacility2.1PiezoelectricDiscThedevelopedtestfacilityshowninFig.1employsalowcostPRYY+0226piezoelectricdisc(10mmdiameter,0.5mmthickness)madebyPICeramicGmbH.TheactuatordiscismadeofasoftleadzirconatetitanatecalledPIC255.Thewrappedelectrodeoptionisusedforconnectingbothsidesofthepiezoelectricdiscfromthetop.Electricalcon-nectionsaremadeviaenameledcopperwires.Loctite401(cyanoacrylateadhesive)isusedforgluingthemicromachinedangularratesensorontothepiezoelectricdisc.TheelongationofthepiezoelectricdiscandtherebythedisplacementofthewholechipisapproximatelyproportionaltotheappliedvoltageVovertimetduetotheindirectpiezoelectriceffect[7].Therefore,velocityandaccelerationincreaseinanidealcase

14ModalAnalysisofMEMSUsingUltrasonicBaseExcitation|3overfrequencyfasshownin(1)–(3)forasinusoidalvoltage.displacement∝Vsin(2πft)(1)velocity∝2πfVcos(2πft)(2)2acceleration∝−(2πf)Vsin(2πft)(3)2.2SeismicMassThelawofconservationofmomentum(4)connectsthevelocitiesonbothsidesoftheactuator[4].Consequentlyaseismicmass,whichactsascounterweightwithamassmMassandavelocityvMass,mustbeappliedontheothersideofthediscinordertoamplifythevelocityvChipofthechipwiththemassmChip.mChipvChip=−mMassvMass(4)vChipmMass=−(5)vMassmChipIncontrast,thethicknessmoderesonantfrequencyoftheseismicmassmustbeashighaspossibletorealizethedesiredmonotonicbehavior.Wehavechosenasmallsteelplate(25×25×2mm3)ascounterweight.Themassofthesteelplateisfactor400higherthanthemassofthesensorchip.Ascanbeseenin(5),thevelocityofthechipisthereforefactor400higherthanthevelocityoftheseismicmass.Thethicknessmodefrequencyfthoftheseismicmasscanbecalculatedwith(6)wherecisthespeedofsoundandtthethicknessoftheplate[8].Accordingly,thecalculatedthicknessmodefrequencyoftheusedsteelplateisabove1MHz.cfth=(6)2tThepiezoelectricdiscisgluedwithKapton(polyimidefilm)tapeandLoctite401onthesteelplate.Kaptontapecanberemovedandthereforeallowsthereuseofthesteelplate.Sincethetapeis50μmthinandstiff,itaddsnoextradampingtothesystem.Hi-Bond4100C(double-sidedadhesivefoamedacrylictape,1mmthick)isusedforgluingthesteelplateonacylindricalaluminumplate(10mmthick,100mmdiameter).Theviscoelastictapeuncouplesthealuminumbaseplatefromhighfrequencyvibrationsatthesteelplatecausedbythepiezoelectricdisc.Measurementsshowedthatthedouble-sidedtapehaseigenfrequenciesbelow2kHz.Forhigherfrequenciestherubberytapeactsasdamper.Fig.2givesasummaryoverallusedlayers.

154|S.Kehrbergetal.DeviceundertestSupergluePiezoactuatorSuperglueKaptontapeSteelplateDouble-sidedadhesivetapeAluminumplateFig.2.Sketchofthedifferentlayersofthetestfacility.2.3VacuumChamberandMeasurementSystemThewholetestfacilityismountedinavacuumchamber.Anambientpressureof1mbarinsidethevacuumchamberisappliedtoreducedampingandhenceincreasethedisplacementofthemicromachinedangularratesensor.ThevacuumchamberitselfisthenmountedonapositioningtableunderaPolytecMSA500MicroSystemAnalyzer.Thewholemeasurementsystemismountedonanair-cushionedtabletodampenvironmentfrequencies.Weusedtheintegratedscanningvibrometerforout-of-planemeasurementsandthestroboscopicvideomicroscopyforin-planemotiondetection.Thebuilt-infrequencygeneratorconnectedtothepiezoelectricdiscprovidesamaximumpeaktopeakvoltageof20V.3AngularRateSensorAsiliconsurfacemicromachinedBoschangularratesensor[9]isusedtoshowthecapabilityofthedevelopedtestfacility.ThesensorshowninFig.3isacapacitivetypeCoriolisvibratorygyroscope.Thesensingelementconsistsoftwoidenticalspring-massstructuresconnectedbythecouplingspringpresentedinFig.4.ABoschfoundryprocesswithathicknessof∼11μmisusedforthesensor.Astheangularratesensorismadeforautomotivesafetyapplications,suchastheelectronicstabilityprogram,itisveryrobustagainstvibration.Accordingly,thefirstmodecanbeseenatahighfrequencyof∼15kHz.Fabricationtolerancesleadtoaminimalmisalignmentofthesidewallanglesofsurfacemicromachinedstructures,resultinginthespringshavingasymmetricprofileswhichleadtoquadratureeffects[10–12].Quadratureenablesexcitationofin-planemodeswithanout-of-planeforce.Therefore,wecananalyzeallmodesintherangeupto1MHzwithourdevelopedtestfacility.Furthermore,thequadratureallowsdetectingin-planemodesbecausetheywillalsoappearintheout-of-planefrequencyresponsefunction.

16ModalAnalysisofMEMSUsingUltrasonicBaseExcitation|5Fig.3.Boschangularratesensorwithtwoidenticalspring-massstructures.Fig.4.Couplingspringbetweenthetwomasses.4Results4.1BehavioroftheTestFacilityWemeasuredthefrequencyresponsefunctionofthetestfacilityusingaperiodicchirpsignalintherangefrom10kHzto1000kHzwithconstantpeakvoltage.AlaserDopplervibrometerwasfocusedonthesubstrateoftheangularratesensor.Themeasuredfrequencyresponsefunction(Fig.5)showsupto200kHzneitherinaccelerationnorphaseasignificantdeviationfromthedesiredmonotonicbehavior.Therefore,thetestfacilitycanbeusedinthisrangeforanalyzingmechanicalnonlinearitiesofMEMS.

176|S.Kehrbergetal.Furthermore,thetestbenchcanbeutilizedtofindeigenmodesintherangeupto1MHz.Thefrequencyresponsefunctionshowsinthisrangesignificanteigenmodesofthetestfacility.Theseresonancepeakshavelowqualityfactorsoflessthan50at1mbar.Theeigenmodesoftheangularratesensorhaveatthesamepressureandfrequencyatleastoneorderofmagnitudehigherqualityfactors.Consequentlyitiseasilypossibletodistinguishbetweeneigenmodesofthetestfacilityandeigenmodesofthedeviceundertest.1e9Acceleration(a.u.)1e71e51e31e11e–102004006008001000180Phase(°)900–90–18002004006008001000Frequency(kHz)Fig.5.Bodeplotofthemeasuredout-of-planeacceleration.1e3Magnitude(a.u)1e–11e–502004006008001000Frequency(kHz)Fig.6.Bodeplotofthemeasuredout-of-planedisplacementofthecouplingspring.

18ModalAnalysisofMEMSUsingUltrasonicBaseExcitation|74.2AnalysisoftheAngularRateSensorForshowingthecapabilityofthedevelopedfacilitywemeasuredtheBodeplotofthecouplingspringemployingthelaserDopplervibrometerwhichispresentedinFig.6.Weusedaperiodicchirpasinputsignal.Afterthismeasurement,weutilizedasinglefrequencysinusoidalvoltagetostimulateonlyoneeigenmodeatatime.ThelaserDopplervibrometerwasnowusedinscanningmodeover∼100equidistantpointsalongthecouplingspring.Afterwards,thedeflectionofthewholestructurewascalculatedbythePolytecPSVSoftware.Asexamplesofthedifferentmeasuredout-of-planemodesbelow200kHz(measurementswithoutsignificantmodesofthetestfacility),Fig.7presentsasymmetricmodeandFig.8anasymmetricmode.Fig.7.Out-of-planemodeofthecouplingspringat∼115kHz.Fig.8.Out-of-planemodeofthecouplingspringat∼190kHz.

198|S.Kehrbergetal.ThefrequencyresponsefunctionpresentedinFig.6alsoincludesin-planemodesduetothequadratureeffect.Thismakesitsimplertofindin-planemodesbutincreasestherisktoconfusein-planewithout-of-planemodes.Thusweanalyzeddifferentmodesusingstroboscopicvideomicroscopyafterwardstoclearlyidentifyin-planemodes.Fig.9presentsameasuredasymmetricin-planemodeat198kHz.Thecapabilityofmeasuringmodesintherangeupto1MHzcanbeseeninFig.6.Thepresentedfrequencyresponsefunctionincludesalargenumberofresonancepeaksofthecouplingspringwhichcanbeclearlydistinguishedfromtheeigenmodesofthetestfacilitybycomparingqualityfactors.AnexampleofameasuredeigenmodeinthisfrequencyrangecanbeseeninFig.10.Fig.9.In-planemodeofthecouplingspringat∼198kHz.Fig.10.Out-of-planemodeofthecouplingspringat∼704kHz.

20ModalAnalysisofMEMSUsingUltrasonicBaseExcitation|95ConclusionThe0.5mmthinpiezoelectricdisc,incombinationwithaseismicmassandadampinglayer,isappropriateforbaseexcitationintherangebetween10and1000kHz.WeemployedthepresentedtestfacilityforthemodalanalysisofaBoschangularratesensorandshowedthecapabilityofthenewshakersetup.Themainadvantageofthepresentedfacilityis,besidesitswidebandwidth,themonotonicfrequencyresponsefunctionwithoutsignificantdeviationsintherangefrom10kHzto200kHz.Inconsequence,itisqualifiedforanalysisofmechanicalnonlinearitiesofMEMS.Theabilityofperformingmechanicalstimulationsandopticalmeasurementsupto1MHzallowstopreciselycharacterizealmostallkindsofMEMS.Acknowledgment:TheauthorsthankallcolleaguesatRobertBoschAutomotiveElectronics,EngineeringSensorTechnology,fortheirvaluablecontributionstothiswork.Bibliography[1]A.W.PhillipsandR.J.Allemang.AnoverviewofMIMO-FRFexcitation/averaging/processingtechniques.Journalofsoundandvibration,262(3):651–675,2003.[2]J.E.Massad,H.Sumali,D.S.Epp,andC.W.Dyck.Modeling,simulation,andtestingofthemechanicaldynamicsofanRFMEMSswitch.InInt.Conf.onMEMS,NANOandSmartSystems,Proceedings,pp.237–240,2005.[3]N.Dumas,F.Azaïs,F.Mailly,andP.Nouet.StudyofanElectricalSetupforCapacitiveMEMSAccelerometersTestandCalibration.JournalofElectronicTesting,26(1):111–125,2010.[4]D.S.Epp,O.B.Ozdoganlar,P.M.Chaplya,B.D.Hansche,andT.G.Carne.Abaseexcitationtestfacilityfordynamictestingofmicrosystems.In22ndInt.ModalAnalysisConf.(IMAC),Proceddings,2004.[5]J.S.Burdess,A.J.Harris,D.Wood,R.J.Pitcher,andD.Glennie.Asystemforthedynamiccharacterizationofmicrostructures.JournalofMicroelectromechanicalSystems,6(4):322–328,1997.[6]JiriMarek.AutomotiveMEMSsensors–trendsandapplications.2011InternationalSymposiumonVLSITechnology,SystemsandApplications(VLSI-TSA),1–2,2011.[7]H.J.M.T.A.Adriaens,W.L.DeKoning,andR.Banning.Modelingpiezoelectricactuators.IEEE/ASMETransactionsonMechatronics,5(4):331–341,2000.[8]H.ShermanandL.Butler.Transducersasprojectors.TransducersandArraysforUnderwaterSound,76–151,2007.[9]U.-M.Gomez,B.Kuhlmann,J.Classen,W.Bauer,C.Lang,M.Veith,E.Esch,J.Frey,F.Grabmaier,K.Offterdinger,T.Raab,H.-J.Faisst,R.Willig,andR.Neul.Newsurfacemicromachinedangularratesensorforvehiclestabilizingsystemsinautomotiveapplications.TRANSDUCERS’05.The13thInternationalConferenceonSolid-StateSensors,ActuatorsandMicrosystems,DigestofTechnicalPapers,1:184–187,2005.

2110|S.Kehrbergetal.[10]M.S.WeinbergandA.Kourepenis.Errorsourcesinin-planesilicontuning-forkMEMSgyroscopes.JournalofMicroelectromechanicalSystems,15(3):479–491,2006.[11]E.Tatar,S.Alper,andT.Akin.Effectofquadratureerrorontheperformanceofafully-decoupledMEMSgyroscope.IEEE24thInt.Conf.onMechanicalSystems(MEMS),569–572,2011.[12]M.Saukoski,L.Aaltonen,andK.A.Halonen.Zero-rateoutputandquadraturecompensationinvibratoryMEMSgyroscopes.IEEESensorsJournal,7(12):1639–1652,2007.BiographiesStevenKehrbergreceivedhisDipl.-Ing.degreeinelectricalengineeringwithfocusonmicrosystemsandprecisionengineeringattheChemnitzUniversityofTechnology,Germany,in2011.Since2011hehasbeenworkingattheAutomotiveElectronicsdepartmentofRobertBoschGmbHinReutlingen,Germany.Hisresearchinterestsincludedesign,simulationandcharacterizationofMEMSgyroscopes.PatrickWellnerstudiedmaterialscienceattheUniversityofStuttgartandreceivedhisDiplomadegreein1999.From2000to2003hestudiedthemechanicalbehaviorofthinNiAl-filmsattheMax-Planck-InstituteforMetalsResearch(nowMax-Planck-InstituteforIntelligentSystems)andreceivedhisPhDin2003forthiswork.Since2004hehasbeenworkinginthesensordevelopmentunitatRobertBoschGmbH,AutomotiveElectronics.HeworkedasasimulationengineerandteamleaderforaccelerationandangularrateMEMSsensorsforseveralyears.Inhiscurrentpositionheisseniormanagerinthefieldofsensordevelopmentforbothautomotiveandnon-automotivecustomers.CarstenGeckelerreceivedhisDiplomadegreeinphysicsfromtheEberhard-KarlsUniversityinTübingenin2002.From2002to2008heworkedattheuniversityofTübingenandlaterattheuniversityofBonninthefieldofexperimentalatomopticswithBose-Einsteincondensatesinopticaldipoletraps.Since2008hehasbeenworkingattheAutomotiveElectronicsdepartmentofRobertBoschGmbHinReutlingen,GermanyondesignandsimulationofaccelerationandangularrateMEMSsensors.Currentlyhismaininterestsaremulti-axisgyroscopeMEMSforconsumerapplications.

22ModalAnalysisofMEMSUsingUltrasonicBaseExcitation|11JanMehnerreceivedhisDr.-Ing.degreeinelectricalengineeringandinformationtechnologyfromtheChemnitzUniversityofTechnology,Germany,in1994.From1998to1999hewasvisitingscientistattheMassachusettsInstituteofTechnology,Cambridge,wherehewasinvolvedinsoftwaredevelopmentforMEMSdesign.1999hereceivedapost-doctoralqualificationforteachingatuniversities(habilitation).From2004to2007,hewasscientificassistantattheFraunhoferInstituteforReliabiltyandMicrointegrationBerlin.In2007,hebecamefullprofessorformicrosystemsandprecisionengineeringattheChemnitzUniversityofTechnology.Hisresearchinterestsareanalyticalandnumericalmethodsformicrosystemsdesign,coupledfieldanalysis,sensorandactuatorapplicationsaswellasexperimentalcharacterizationofmicrostructures.

23

24M.Saihi,B.Boussaid,A.ZouinkhiandM.N.AbdelkrimDecentralizedFaultDetectioninWirelessSensorNetworkbasedongaussianfunctionerrorAbstract:WirelessSensorNetworks(WSNs)havebecomeaninformationcollectionandmonitoringsolutionforavarietyofapplications.Faultsoccurringtosensornodesarecommonduetothesensordeviceitselfandtheharshenvironmentwherethesensornodesaredeployed.AsoneofthekeytechnologiesinvolvedinWSNs,nodefaultdetectionisindispensableinmostWSNapplications.Inthispaper,wedealwithanewapproachfordetectingdistributedfaultsinwirelesssensornetworkbasedonerrorfunctionforgaussiandistributionproportionaltothedeviationbetweenthemeasurements.Besides,newcoefficientstomeasurethedifferencebetweenmeasurementsandtodothetestonthestatusofthenodearedeterminedinordertoimprovethefaultdetectionalgorithm.Theresultofsimulationsshowanimprovementoftheaccuracyoftheproposedalgorithmcomparedtotheclassicone.Keywords:Wirelesssensornetwork,distributedfaultdetection,gaussiandistribution,errorfunction.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A221IntroductionWirelesssensornetworksareemergingascomputingplatformsformonitoringvariousenvironmentsincludingremotegeographicalregions[1].Theyarecomposedofalargenumberoftinysensornodesequippedwithlimitedcomputingandcommunicationcapabilities[24].FailednodesmaydecreasethequalityofserviceoftheentireWSN[18].ThenodestatusinWSNscanbedividedintotwotypes[2,3]:normalandfaulty.Faultyinturncanbe“permanent”or“static”.Theso-called“permanent”meansfailednodeswillremainfaultyuntiltheyarereplaced,andtheso-called“static”meansnewfaultswillnotbegeneratedduringfaultdetection.WSNnodefaultsareusuallyduetothefollowingcauses:thefailureofmodule(suchascommunicationandsensingmodule)duetofabricationprocessproblems[21],environmentalfactors,batterypowerdepletion[22]andsoon.AsensornetworkmustbecapableofM.Saihi,B.Boussaid,A.ZouinkhiandM.N.Abdelkrim:GabesUniversity,NationalSchoolofEngineersofGabes,Gabes,Tunisia,ResearchUnitofModeling,AnalysisandControlSystems(MACS),emails:saihimarwa@yahoo.fr,dr.boumedyen.boussaid@ieee.org.DeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.13–26.DOI10.1515/9783110470444-002

2514|M.Saihietal.identifyingandreplacingthefaultynodesinordertomakesurethatthenetwork’squality-of-serviceismaintained[19].WeproposedanimprovedDFDschemebasedonerrorfunction.Infact,theoldtestresultusedabinarytest;itcausedanundesiredabruptchangesowehavetriedtocalculatethetestresultdifferentlybyintroducingtheerrorfunctiontoenlargetheaccuracyofthealgorithm.Adistributedfaultdetectionschemeforsensornetworkshasbeenproposed;ituseslocalcomparisonswithamodifiedmajorityvoting,whereeachsensornodemakesadecisionbasedoncomparisonbetweenitsownsensingdataanditsneighborsdata[25],whileconsideringtheconfidencelevelofitsneighbors.Highfaultdetectionaccuracycanbereachedonlywhenitisappliedtothesensornetworkwithmanyneighborsofnodestobediagnosed[20].Inthispaper,weproposeanimprovedDFDschemebydefiningnewdetectioncriterion.Theremainderofthepaperisorganizedasfollows:Insection2,relatedworksintheareaoffaultdetectioninWSN,insection3,faultmodelandDFDalgorithm,theproposedalgorithminsection4andsimulationexampleinthelastsection.2PreviousworkInthissection,webrieflyreviewtherelatedworksintheareaoffaultdetectioninwirelesssensornetworks.Severalworkshaveaddressedtheproblemofhowtodealwithfaultsoccurringinwirelesssensornetworksinordertoachievefaulttolerance[4–6].In[7],awatchdogprocessorisusedforconcurrentsystemlevelerrordetectiontechniques.Awatchdogprocessorisasmallandsimplecoprocessorthatdetectserrorsbymonitoringthebehaviorofasystem.Theyshowedthatalargenum-beroferrorscanbedetectedbymonitoringthecontrolflowandmemory-accessbehavior.Theresearchproposedin[4]makesuseofredundancyandusesatechniquetodecideonwhichnodestokeepactiveandonwhichtoputinasleepmode.Thetechniqueaimstoprovidethesensorfieldwiththebestpossiblecoverage.Inaddition,itmaintainsnetworkconnectivitytorouteinformation.Whenanactivenodefailsitissubstitutedbyoneofthesleepingnodes.However,otherresearchershaveaddressedtheproblemofhavingactivenodesthatprovideincorrectdatawhichresultsinmakinginappropriatedecisions.Theauthorsin[8]proposedageneralapproachtofaultdiagnosisthatiswidelyapplicableandonlyneedsalimitednumberofconnectionsamongunits.Thealgo-rithmusesamajorityvoteamongtheneighborsofaunittodeterminethestatusoftheunit.Theresearchproposedin[5]focusedonsuchissuesandproposedamechanismtodetectanddiagnosedatainconsistencyfailuresinwirelesssensornetworks.

26DecentralizedfaultdetectioninWSN|15Themechanismproposedin[5]usestwodisjointpathstosendthesenseddatatoastaticsink.Afterthesinkreceivesbothcopies,itwillcomparethemtocheckiftheymatch.Ifthetwocopiesmatch,boththedataandthepathsareconsideredtobefaultfreeotherwise,athirddisjointpathwillbeestablished.Thenthesensornodewillsendthreecopiesonthethreedisjointpathstothesink.Thesinkwillcomparethesecopiesanddecidesonthefaultypath.Finally,adiagnosisroutinewillbeexecutedtoidentifythefaultynodewithinthefaultypath.Article[9]proposesanagreement-basedfaultdetectionmechanismfordetectingcluster-headfailuresinclusteredUnderwaterSensorNetworks(UWSNs).Eachclustermemberisallowedtoindependentlydetectthefaultstatusofitsclusterheadandatthesametimeadistributedagreementprotocolisemployedtoreachanagreementonthefaultstatusoftheclusterheadamongmultipleclustermembers.ThedetectionmechanismisbasedonaTDMAMACprotocolusedinthenetworkandrunsconcurrentlywithnormalnetworkoperationbyperiodicallyperformingadistributeddetectionprocessateachclustermember.Itmakesuseofthedataperiodicallysentbyaclusterheadastheheartbeatsforfaultdetection.Theresearchdescribedin[7]proposedaschemebasedonmulti-pathroutingcombinedwithchannelcodingtoachievefaulttolerance.Itusesafuzzylogicbasedal-gorithmthatisenergyandmobilityawaretoselectmultiplepaths.Whenselectingthepaths,thealgorithmtakestheremainingenergy,mobilityandthedistancetothedes-tinationintoaccount.Article[10]showsitisimportanttoprovidefailurerecoveryandavoidunnecessarytrafficwhentheroutingtopologyneedstoberebuilt.Itpresentsaninferenceengine,calledDiffuse,designedtodetectwhentheroutingtopologyneedstoberebuiltbasedondifferentgoals,suchastorecoverfromroutingfailures,improvedataaggregation,andbalancetheenergyconsumption.Diffuseapproachesefficientlyavoidunnecessarytopologyconstructions.Theauthorsuseinformation/datafusiontodetectroutingfailures,whichisadifferentandpromisingapproach.Ageneralframeworktoachievefaulttoleranceinwirelesssensornetworkswasproposedin[9].Theframeworkisbasedonalearningandrefinementmodulewhichprovidesadaptiveandself-configurablesolutions.Theauthorsin[13]proposedandevaluatedalocalizedfaultdetectionschemetoidentifythefaultysensors.Distributedfaultdetection(DFD)methodhassomeshortcomingsasfollows:thefaultdetectionaccuracywilldecreaserapidlyinthecaseofthenumberofneighbornodestobediagnosedisallsmallandthenode’sfailureratioishigh.Highfaultdetectionaccuracycanbereachedonlywhenitisappliedtothesensornetworkwithmanyneighborsofnodestobediagnosed.In[13],theauthorsproposedfaulttolerantalgorithmstodetecttheregionofaneventinwirelesssensornetworks.Also,theyassumethatnodesreportabinarydecisiontoindicatethepresenceofaneventornotandconsideredahazardousbehaviorforthefaultynodes,whichmeansthatthefaultynodeswillbeprovidingarbitraryvalues.Theyproposedarandomizeddecisionschemeandathreshold

2716|M.Saihietal.decisionschemewhichasensornodecanusetodecideonwhichbinarydecisiontosendbycomparingthedecisionithaswiththedecisionsofitsneighbors.Article[14]investigatesusingthespatialcorrelationofsensormeasurementstodetectfaultsinWSNs.Anapproachofweightingtheneighbors’measurementandpresentsamethodtocharacterizethedifferencebetweensensormeasurementsareintroduced.Aweightedmedianfaultdetectionscheme(WMFDS)isproposedandevaluatedforbothbinarydecisionsandrealnumbermeasurements.In[15],afaultmapwasconstructedusingafaultestimationmodel.Inordertobuildthefaultmap,sensornodesarerequiredtosendadditionalinformationthatcanbeusedbythefaultestimationmodel.Furthermore,aclusterbasedalgorithmtoestimatefaultsinwirelesssensornetworkswasproposed.In[16],atargetdetectionmodelforsensornetworkswasproposed.Inaddition,twoalgorithmstofacilitatefaulttolerantdecisionmakingwerepresented.Thefirstalgorithmisbasedoncollectingtheactualreadingsfromtheneighboringnodes.Inthesecondalgorithm,thesensornodeobtainsthedecisionsmadebytheotherneighboringnodestotakeafinaldecision.In[17],thedesignofadistributedfault-tolerantdecisionfusioninthepresenceofsensorfaultswhenthelocalsensorssequentiallysendtheirdecisionstoafusioncenterisaddressed.Acollaborativesensorfaultdetection(CSFD)schemeisproposedtoeliminateunreliablelocaldecisionswhenperformingdistributeddecisionfusion.Basedonthepre-designedfusionrule,assumingidenticallocaldecisionrulesandfault-freeenvironments,anupperboundisestablishedonthefusionerrorprobability.Accordingtothiserrorboundary,acriterionisproposedtosearchthefaultynodes.Oncethefusioncenteridentifiesthefaultynodes,allcorrespondinglocaldecisionsareremovedfromthecomputationofthelikelihoodratiosthatareadoptedtomakethefinaldecision.Theauthorsin[17]applyerrorcorrectingcodestoachievefaulttol-erance.Asaresult,adistributedfaulttolerantclassificationapproachwasproposed.Theapproachproposedisbaseoffaulttolerantfusionrulesthatareusedtoobtainlocaldecisionrulesateverysensor.Inaddition,theauthorsproposedtwoalgorithmsthatcanbeusedtofindgoodcodematricestobeusedbytheclassificationapproach.3NetworkmodelandfaultmodelWeassumesensorsarerandomlydeployedintheinterestedareaandallsensorshaveacommontransmissionrange.Theareaisassumedtobeentirelycoveredbythesensors.AsshowninFig.1,thedarkcirclesrepresentfaultysensorsandthelightgraycirclesaregoodsensors.Therecouldbeafailureoccurringinacertainareaasillustratedinthisfigure.Allsensorsintheareagooutofservice.Sincewearedependingonmajorityvoting,weassumethateachsensorhasatleast3neighboringnodes.Becausealargeamountofsensorsarecastintothe

28DecentralizedfaultdetectioninWSN|17interestedareatoformawirelessnetwork,thisconditioncanbeeasilyobtained.Eachsensornodeisabletolocatetheneighborswithinitstransmissionrangethroughabroadcast/acknowledgeprotocol.Faultsmayoccuratdifferentlevelsofthesensornetwork,suchasphysicallayer,hardware,systemsoftware,andmiddleware[17].Inthispaper,wefocusonhardwarelevelfaultsbyassumingallsystemsoftwareaswellastheapplicationsoftwarearealreadyfaulttolerant.Thefirstofthetwogroupsofcomponentsathardwarelevelconsistsofacomputationengine,storagesubsystemandpowersupplyinfrastructure,whichareveryreliable.Anothergroupofcomponentsaresensorsandactuatorswhicharemostpronetomalfunctioning.Becauseinthefirstgroupofcomponentstheheterogeneousfaulttolerantschemeswillprovidethetargetedleveloffaulttolerance[17],weonlyconsiderthesensorfaultswhichincludethreetypesoffaults:calibrationsystematicerror,randomnoiseerror,andcompletemalfunctioning.Remark:Nodesarestillcapableofreceiving,sending,andprocessingwhentheyarefaulty.FaultynodeWorkingnodeFig.1.WirelessSensorNetworkwithbothfaultyandworkingnodes.4Detectionofthedistributeddefectswithdensityofprobability4.1GaussiandistributionofmeasurementsGenerallythemeasuresstemmingfromsensorsofmeasurehavearandombehaviorbutconcentratedaroundanaveragebytrainingabellasthatofGauss.Consequently,wecanassimilatethatthemeasuresfollowthenormaldistribution.Indeed,the

2918|M.Saihietal.expressionoftheresultofmeasurespellsunderthefollowingshape:x=m+∆xwheremisthebestestimationofthemoderatesizexand∆xtheuncertaintyonthemeasure.Letusbenotafraidofinsistingontheimportanceoftheestimationoftheuncertainty∆x.Infact,wecannotknowwithoutknowingtheuncertaintyifasizeevolved,ifsuchprocessofmeasureleadstothesameresultthatsuchtheotherone,orifthedifferencepossiblyobserved.4.2PresentationofthenormaldistributionThenormaldistributionisatheoreticaldistribution,inthesensethatitisamathe-maticalidealizedimagewhichnevermeetsexactlyinthenature.Butnumerousreallyobserveddistributionsgetclosertoitandhavethisfamousshapeof“bell”(manyindividualsaroundtheaverage,lessandlessinthefuraswegoawayfromit,andthisinasymmetricway).Ontheotherhand,itisveryusedininferentialstatistics:weshallseeinparticularthatanaveragecalculatedonasampleisrandomvariablewhichtendstofollowanormaldistributionwhenthesizeofthesampleincreases,eveniftheinitialpopulationhasquiteadifferentdistribution.4.2.1ThebellcurveLetconsiderthenormaldistributionwiththeparametersmandσnotedN(m,σ)whichisdefinedonRbythedensity:11x−m2f(x)=√exp−(1)σ2π2σLetusnotethat:–therightx=misanaxisofsymmetry,–theinflexionpointsaresituatedatadistanceσofthisaxisofsymmetry.4.2.2Thecentraltheorem-limitTheCTL(CentralTheorem-Limit)willbeverypreciousbecauseheexplainsusthatifwemakethesumofoneverylargenumberofrandom(unpredictable)variablesofanylaw,thissumfollowsapproximatelyanormallaw(infact,withoutgoingintothedetailofthehypotheses,hesaystousthatthevariableX=X1+X2···Xntendstofollowanormaldistributionwhennaimstowardstheinfinity).

30DecentralizedfaultdetectioninWSN|19Ononehand,itallowsustounderstandwhysomanydistributionsobservedintherealityhaveapproximatelythisshapeofbell:theydescribephenomenawhichresultfromtheadditionofalargenumberofindependentcausesoffluctuation.4.2.3ThereducedcenterednormaldistributionTocenterandtoreduceavariable,itistoargueinnumberofstandarddeviationsσwithregardtotheaveragem.AlltherelativeeventsinXcanbeexpressedaswellaccordingtoT.ThereducedcenteredvariableThasforhope0andforstandarddeviation1because:x−m1–E(T)=E=[E(x)−m]=0becauseE(x)=mσσx−m12–V(T)=V=V(x)=1becauseV(x)=(σ)σσ2Thus,thedensityofprobabilityy(t)ofthereducedcenterednormaldistributionN(0,1)spells:1−1t2y(t)=√e2(2)2πThisexpressiongivesinparticulartheprobability,F(x)=Pr(t≤x)thatnomeasureofthesizetissmallerthanthevaluex:1−1x2F(x)=√e2(3)2πRatherthanfandF,wenotegenerallyythedensity,andPthefunctionofdistributionofthedistributionN(0,1).4.2.4TheerrorfunctionWedefinealsothe“errorfunction”erf(t)(seeFig.2)as:x2−t2erf(x)=√edt(4)π0Thisfunctionallowstocalculatethesesameprobabilityinmeansoftheexpression:1x−μPr(t≤x/gaussian,μ,σ)=1+erf√(5)2σ2Wenoticethatthiserrorcanbeexploitedasafunctionoftestreplacingthebinarytest.

3120|M.Saihietal.FunctionError10.80.60.40.2erf0−0.2−0.4−0.6−0.8−1−4−3−2−101234RandomlyproducednumbersFig.2.TheFunctionErrorrepresentation,erf(x).4.3DefinitionsBeforestudyingtheexploitationofthenormaldistributionrulesinDFD,wegivethefollowingdefinitionsthatwillbeusedlater:–n:totalnumberofsensors;–p:probabilityoffailureofasensor;–k:numberofneighborsensors;–S:setofallthesensors;–N(Si):setoftheneighborsofSi;–xi:measurementofSi;–dij(t):measurementdifferencebetweenSiandSjattimet,dij(t)=xi(t)−xj(t)–∆dij(t):measurementdifferencebetweenSiandSjfromtimettot+1,∆dij(t)=dij(t+1)−dij(t)=(xi(t+1)−xj(t+1))−(xi(t)−xj(t))(6)–Cij:testbetweenSiandSj–Ti:tendencyvalueofasensor,Ti∈{LG,LT,GD,FT};–σ1andσ2:thetwoassumedstandarddeviationsassociatedtothenormaldistributionofthemeasurements;

32DecentralizedfaultdetectioninWSN|214.4ApplicationofthenormaldistributioninDFDWeapplythesameimprovedalgorithmandwetrytointroduceresultofmeasureintoafunctionoferrortoobtainamoresuccessfulevaluationofthestatesofknots:dij(t)2−u21erf(dij(t))=√edu=erfij(t)(7)π0∆dij(t)2−u22erf(∆dij(t))=√edu=erfij(t)(8)π0ThenwecalculateCwhichistheproductoferf1anderf2andwerepeatthesameijijijdetailsofcalculationtreatedintheclassicDFDalgorithmwiththeerrorfunction.4.5DescriptionofthealgorithmStep1:ForeverynodeSiandanynodeSjinneighbor(Si)wecalculate:•erf1,•erf2,and•C=(erf1×erf2)ijijijijijStep2:IfCij

3322|M.Saihietal.Tab.1.Asampleoftemperaturemeasurements(part1).Period12345Sensor133.993733.935981.962734.017733.9383Sensor233.797834.180914.081533.969234.0275Sensor333.901833.892024.079933.986834.0601Sensor424.061334.019924.012034.059534.0092Sensor533.994533.847934.057134.104734.1730Sensor633.888133.927634.041333.980233.9391Sensor733.937433.940743.901334.032833.9092Sensor834.025034.040133.076033.976233.8250Sensor933.900734.094233.934334.023034.0910Sensor1034.097534.030033.044034.044034.0867Tab.2.Asampleoftemperaturemeasurements(part2).Period678910Sensor133.992033.885933.871734.024634.0900Sensor234.089833.890733.767134.007033.9700Sensor334.018433.956634.090221.939134.1029Sensor434.029113.983233.816433.877733.9655Sensor534.011333.978134.006734.031744.1013Sensor634.044034.054134.003533.865734.0629Sensor734.010234.038934.222733.896833.9787Sensor834.278730.075133.993134.133128.9134Sensor933.883334.177833.949333.958133.8957Sensor1033.814634.122334.023633.986033.9730WeapplythestandardalgorithmofDFD(withtestbinary)andournewalgorithm(witherrorfunction)onTrueTime1.5WirelessNetworksimulator.ThissimulatorisdevelopedthroughSimulinkMatlabandoffertruetimesimulationforwirelessnetworkenvironmenttoolbox.Aftersimulation,wegetthefollowingresultswhichareregroupedinTab.3andTab.4,respectively,where“1”means“FaultyNode”or“FT”and“0”means“NormalNode”or“GD”.MoretestsaredonewithvariouscasesoffaultsforbothalgorithmwhichprovethatthestandardDFDalgorithmisefficientonlywhenthefaultysensorsarelessthanthehalfofsensors.However,theproposedimprovedalgorithmisefficientuntil6faultsamong10sensorswhichequivalentto60%ofefficiency.ThisresultisveryimportantincaseofhighprobabilityoffaultoccurrenceespeciallywhenthenumberofsensorsinWirelessNetworksisnotveryimportantwhichalwaysthecase.

34DecentralizedfaultdetectioninWSN|23BellCurve0.140.120.10.080.06GaussDistribution0.040.020010203040506070RandomlyproducednumbersFig.3.Thedistributionofmeasurements.Tab.3.StandardDFDalgorithmresult.Period12345678910Sensor10011000000Sensor20011000000Sensor30011000011Sensor41111001100Sensor50000000001Sensor60000000000Sensor70011000000Sensor80000001101Sensor90000000000Sensor100011000000Tab.4.ImprovedDFDalgorithmresult.Period12345678910Sensor10010000000Sensor20010000000Sensor30010000010Sensor41010001000Sensor50000000001Sensor60000000000Sensor70010000000Sensor80000001001Sensor90000000000Sensor100000000000

3524|M.Saihietal.6ConclusionInthiswork,anewapproachtodetectdistributedfaultsinwirelesssensornetworkbasedonerrorfunctionproportionaltothedeviationbetweenthemeasurementsisproposed.Infact,morethedifferenceisgreater,morethepossibilityoffallingdownisgreater.Uponthisremark,wehavedesignedanewalgorithmwhichcalculatesthedifferencebetweenmeasurementsandgeneratesanormalprobabilitybasedonthegaussianfunctionerrorinorderdodeducethestatusofthenode.ThenewtestprocedurereplacesthebinarytestinthepreviousworkinDFDasmentionedabove.Theapplicationofthisalgorithmonasampleof10sensorsshowanefficiencyof60%versus50%intheclassisone.Thisresultisveryimportantifthecaseofasmallnodeneighborhoodinthesmallsmallsensornetworkapplications.Bibliography[1]I.F.Akyildiz,W.Su,Y.SankarasubramaniamandE.Cyirci.Wirelesssensornetworks:asurvey.ComputerNetworks,38(4):393–402,2002.[2]S.ChessaandP.Santi.ComparisonbasedsystemlevelfaultdiagnosisinAdhocnetworks.20thSymp.onReliableDistributedSystems(SRDS),NewOrleans,:257–266,2001.[3]S.ChessaandP.Santi.Crashfaultsidentificationinwirelesssensornetworks.ComputerCommunications,25(14):1273–1282,2002.[4]Y.ZouandK.Chakrabarty.ADistributedCoverageandconnectivityCentricTechniqueforSelectingActiveNodesinWirelessSensorNetworks.IEEETrans.onComputers,54(8):978–991,2005.[5]K.Ssu,C.Chou,H.C.JiauandW.Hu.Detectionanddiagnosisofdatainconsistencyfailuresinwirelesssensornetworks.Int.J.ofComputerandTelecommunicationsNetworking,50(9):1247–1260,2006.[6]G.GuptaandM.Younis.Fault-tolerantclusteringofwirelesssensornetworks.Conf.onWirelessCommunicationsandNetworking(WCNC),:1579–1584,2003.[7]A.MahmoodandE.J.McCluskey.Concurrenterrordetectionusingwatchdogprocessors-asurvey.IEEETrans.onComputers,37(2):160–174,1988.[8]D.Blough,S.Sullivan,andG.Masson.Faultdiagnosisforsparselyinterconnectedmultiporcessorsystems.19thSymp.onFault-TolerantComputing,19:62–69,1989.[9]P.Wang,J.ZhengandC.Li.AnAgreement-BasedFaultDetectionMechanismforUnderWaterSensorNetworks.GlobalTelecommunicationsConf.,Washington,DC,:1195–1200,2007.[10]E.F.Nakamura,C.M.S.Figueiredo,F.G.NakamuraandA.A.F.Loureiro.Diffuse:Atopologybuildingengineforwirelesssensornetworks.SignalProcessing,87(12):2991–3009,2007.[11]J.Staddon,D.BalfanzandG.Durfee.Efficienttracingoffailednodesinsensornetworks.1stACMInt.WorkshoponWirelesssensornetworksandapplications,Atlanta,Georgia,USA,2002.[12]J.R.Chen,S.KherandA.Somani.Distributedfaultdetectionofwirelesssensornetworks.Int.Conf.onMobileComputingandNetworkings,LosAngeles,CA,USA,:65–72,2006.[13]Y.S.Chang,T.Y.Juang,C.J.Lo,M.T.HsuandJ.H.Huang.FaultEstimationandFaultMapConstructiononCluster-basedWirelessSensorNetwork.IEEEInt.Conf.onSensorNetworks,Ubiquitous,andTrustworthyComputing,5–7June2006,:14–19,2006.

36DecentralizedfaultdetectioninWSN|25[14]J.L.Gao,Y.J.XuandX.W.Li.Weighted-medianbaseddistributedfaultdetectionforwirelesssensornetworks.J.ofSoftware,18(5):1208–1217,2007.[15]Y.LaiandH.Chen.Energy-EfficientFault-TolerantMechanismforClusteredWirelessSensorNetworks.16thInt.Conf.onComputerCommunicationsandNetworks,ICCCN,:272–277,Auguste2007.[16]T.Y.Wang,L.Y.Chang,D.R.DunandJ.Y.Wu.Distributedfault-tolerantdetectionviasensorfaultdetectioninsensornetworks.10thIEEEInt.Conf.onInformationFusion,Quebec,Canada,:1–6,2007.[17]F.Koushanfar,M.PotkonjakandA.Sangiovanni-Vincentelli.Fault-ToleranceinSensorNetworks.HandbookofSensorNetworks.I.MahgoubandM.Ilyas(eds.),CRCpress,SectionVIII,no.36,2004.[18]R.Haung,X.QiuandL.Ye.Probability-basedfaultdetectioninwirelesssensornetworks.Int.Conf.onNetworkandServiceMangement(CNSM),Bejing,China,:218–221,2010.[19]A.Akbari,A.Dana,A.KhadmeZadehandN.Beikmahdavi.Faultdetectionandrecoveryinwirelesssensornetworkusingclustering.Int.J.ofWirelessandMobileNetworks,3(1):130–138,2011.[20]A.Mahani,M.S.AnsariandY.S.Kavian.Reliabilityorperformance:Atradeoffinwirelesssensornetworks.8thIEEETCIInt.Symp.onCommunicationSystems,NetworksandDigitalSignalProcessing,:1–5,2012.[21]K.LokMan,C.ChenandD.Hughes.DecentralizedFaultDetectionandManagementforWirelessSensorNetworks.5thInt.Conf.onFutureInformationTechnology(FutureTech),May2010,Busan,Korea[22]J.Hill,R.Szewczyk,A.Woo,S.Hollar,D.CullerandK.Pister.SystemArchitectureDirectionsforNetworkedSensors.ACMSIGPLANnotices,35(11):93–104,2000.[23]http://www.zigbee.org/Specifications/ZigBee/Overview.aspx,consulted:06.10.2013,2013.[24]A.Zouinkhi,E.Bajic,E.RondeauandM.N.Abdelkrim.Simulationandmodelingofactiveproductscooperationforactivesecuritysystemmanagement.Trans.onSystems,SignalsandDevices,5(4):1–23,2011.[25]M.Saihi,B.Boussaid,A.ZouinkhiandM.N.Abdelkrim.NewApproachforDecentralizedFaultDetectioninWirelessSensorNetwork.13thInt.Conf.onSciencesandTechniquesofAutomaticcontrolandcomputerengineering,:1–9,Monastir,Tunisia,2012.BiographiesMarwaSaihibornin1988,inTunis,Tunisia.SheisaPhDstudentattheNationalEngineeringSchoolofGabes(Tunisia)andamemberofModeling,AnalysisandControlSystems(MACS)researchunit.ShereceivedtheNotionalengineeringandMasterDegreesfromtheNotionalEngineeringSchoolofGabes(ENIG),Tunisiain2012inElectricalEngineeringandAutomaticControl.ShereceivedtheBachelorDegreein2007inSciences.Currently,herresearchframeworksfocusesonwirelesssensornetworksandfaultdetectionforreliablecentrelized/decentralizedcontrol.

3726|M.Saihietal.BoumedyenBoussaidbornin1972,isassociateprofessoratNationalSchoolofEngineersofGabes,Tunisia,inElectricalEngineeringandAutomaticControlDepartment.HereceivedhisPhDdegreein2011inAutomaticControlandComputerSciencesfrombothUniversityofNancyandUniversityofGabes.HereceivedhisEngineeringDegreein1997inElectricalengineeringfromNationalSchoolofEngineersofTunis.HeiscurrentlyamemberoftheResearchCentreinAutomationofNancy(CRAN)andtheResearchunitonModeling,AnalysisandControlofSystems(MACS).Hisresearchinterestsfocusonfaulttolerantcontrolandfaultdetectioninwirelesssensornetworkswithapplicationtowindfirmsandrobotsnetworks.AhmedZouinkhiisAssociateProfessorattheNationalEngineeringSchoolofGabes(Tunisia)andamemberofModeling,AnalysisandControlSystems(MACS)laboratory.HereceivedtheNotionalengineeringDegreefromtheNotionalEngineeringSchoolofMonastir(ENIM),Tunisiain1997inindustrialcomputing.HereceivedhisPhDdegreein2011inAutomaticControlfromtheNationalEngineeringSchoolofGabes(Tunisia)andaPhDdegreeinComputerEngineeringfromtheNancyUniversity(France).HisresearchactivitiesarefocusedonDistributedSystems,SmartObjectstheoryandapplications,AmbientIntelligencesystemsandarchitectures,RFIDandWirelessSensorsNetworkConceptsandApplicationsinmanufacturingandsupplychain.MohamedNaceurAbdelkrimwasborninTunisiain1958.HeobtainedaDiplomainTechnicalSciencesin1980,hisMasterDegreeinControlin1981fromtheENSETschoolofTunis(Tunisia),andhisPhDinControlin1985andtheDoctorateinSciencesDegree(Electricalengineering)in2003fromtheENITSchoolofTunis.Since2003heisProfessorattheElectricalEngineeringDepartment(Control)oftheNationalEngineeringSchoolofGabes(Tunisia)andheismanageroftheModeling,AnalysisandControlSystems(MACS)laboratory.

38B.Mezghani,F.TounsiandM.MasmoudiStaticBehaviorAnalyticalandNumericalAnalysisofMicromachinedThermalAccelerometersAbstract:Thispaperpresentsstaticbehavioranalyticalstudyofmicromachinedconvectiveaccelerometers.Thisincludesbothheatconductionandconvectionbe-haviorstudyandmodeling.AmixedmodelingtechniquehasbeenusedtoderivegeneralexpressionsgoverningheatconductionandconvectionofMEMSthermalaccelerometers.ThistechniqueisbasedontheuseofresultsfromFEMsimulationstodevelopananalyticalmodelwhereallderivedexpressionsareasafunctionofbiasingtemperaturesandkeydesigngeometryparameters.Forconductionbehavioranalysis,twovariablesarebeingusedinFEMsimulations:heatertemperatureandmicromachinedcavitydepth.Thelatterparameterhasalargeimpactontheoverallconductivebehaviorofthermalaccelerometerssinceitfixesthevolumewheretheheatbubblecanexpand.Inaddition,heatertemperatureisconsideredtobetheonlyparameterthatfixesheatdistributioninthecavity.Thismodelinghasledtothederivationofexpressionsforbothheaterheattransfercoefficientandcommonmodetemperature.Thesephysically-basedderivedexpressionsgoverntheoverallsensorconductivebehavior.Concerningheatconvectionbehavior,cavitywidthparameterhasbeenaddedasathirdvariable.Usingsimulationdatapoints,fittingtechniquehasbeenusedtodevelopananalyticalexpressionofdifferentialtemperature,proportionaltosensitivity,asafunctionoftheabovedesignandtemperatureparameters.Thisstudyhelpstopredictsensorperformanceatanearlydesignstageandmoreimportantlyfordifferentsensordesigngeometriesandtemperatures.Keywords:Thermalaccelerometer,micromachined,analyticalmodel,FEMsimula-tion,sphericalmodel.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A22B.Mezghani,F.TounsiandM.Masmoudi:UniversityofSfax,NationalEngineeringSchoolofSfax(ENIS)Electronics,Micro-technologyandCommunication(EMC)ResearchGroup,Sfax,Tunisia,e-mails:brahim.mezghani@enis.rnu.tn;tou_fares@yahoo.fr;mohamed.masmoudi@enis.rnu.tn.DeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.27–45.DOI10.1515/9783110470444-003

3928|B.Mezghanietal.1IntroductionSincetheintroductionofminiaturemicromachinedcomponents,manynewsensorsarereportedinliterature[1–7].Oneofthesenewmicroscopicelectromechanicalstructuresistheinertialsensorwhichwasintroducedaftertherecentprogressinthedevelopmentofsemiconductorprocessingtechnology.Inertialsensors,asaccelerom-eters,measureaccelerationsinone,twoorthreeorthogonalaxes.Accelerationmea-surementsaretypicallyusedtocalculatein-planevelocityandposition,inclination,tiltororientationintwoorthreedimensionswithrespecttotheaccelerationofgravity,aswellastomeasurevibrationandimpact.Overthelastfewyears,numeroustypesofmicromachinedaccelerometershavebeendesignedanddeveloped[6–15].Duetotheirintegratednature,theyallhavethekeyfeaturesofminiaturesize,compactdesign,lightweight,sensitivetosmallaccelerations,lowpowerconsumptionandlowcostduetothebatchfabricationintegratedprocess.Thankstotheirnumerousadvantages,MEMSaccelerometershavebeenwidelyusedinbothconsumerandmilitaryfields.Thisincludesautomotiveindustry,consumerapplicationsaslaptops,navigationsystems,militaryindustryasinmissileguidanceandinroboticsandsystemautomation.Asanaturaloutcome,researchcentersareincreasinglyinterestedinthedevelopmentofnew,smaller,moresensitive,fasterandalsomorereliableMEMSaccelerometers.Theseneededdevelopmentsinsensorperformancerequirethedevelopment,inthefirstplace,ofafabricationprocesswhichisobviouslyacrucialfactortoprovidethegoodstabilityandtheneededhighreproducibilityaswellashavingahighyieldoftheintegratedsensor.Usually,micromachinedaccelerometerssuchascapacitiveaccelerometers[8]andpiezoresistiveaccelerometers[9]measuretheappliedaccelerationbymeansofusingaproofmass.Byquantifyingthedisplacementofthisproofmass,thesetypesofaccelerometersareabletomeasuretheappliedacceleration.Wehavetonoteherethatthisoperationmode,whichisbasedontheproofmassdis-placement,constitutesthemaindisadvantageofthesetypesofsensors.Thisismainlyduetothepossibilityofsystembreakdownunderahighsignalgivingalowshocksurvivalrateandalsoduetothecomplexfabricationprocessofsuchsensors.Onedesignthatovercomessuchproblemsisthethermalconvection-basedmicromachinedaccelerometer.Overthelastfewyears,severaldesignswithmultiplesensitiveaxeshavebeenpresentedinliterature[11–15].Thekeyfeatureofthisnewsensorliesinthefactthatitsdesigndoesnothaveasolidproofmassbutitreliesonfreeconvectionofaheatedairbubble.Convectivemicromachinedaccelerometersarebeinglargelyanalyzedinrecentyearsbecauseofthesimplicityofsensorstructure.Theseaccelerometershavebeenwidelystudied,buttheiroptimizationstillneedstobeinvestigatedbecauseofthecomplexeffectsofdesignparameterson

40Analysisofmicromachinedthermalaccelerometers|29bothconductionandconvectionoperation.Inaddition,thenumerousrequirementsforthemonolithicprocesscompatibilityimposestrictconstraintonsensordesignparametersasheatertemperatureandgeometry,cavitysizeandpackagesizeandform[16–23].Amongmanythermalconvection-basedaccelerometers,standardCMOSdesignonesareofparticularinterest.Thisisbecauseitisgivesasolutionwhichwillbecompatiblewithmonolithicintegrationwithallthenecessaryinterfaceandsignalconditioningcircuitryonasinglechipusingwaferlevelpackaging.Thiswillobviouslyleadtoadevicewithlowcost,robustandwithaminiaturediesize.Dynamicresponse,orfrequencybandwidth,istheotherareawherethermalaccelerometersdonotshowgoodperformance.However,formanyconsumerapplicationsasinmobilephones,thisisnotconsideredasanimportantissuebecausethelowerfrequencyresponseisconsideredtobemorethansufficient.Thisconcernhasbeenalreadydealtwithusingcorrectionelectroniccircuitry,asfeedbackscheme,andmaximumfrequencyresponsehasbeenpushedfromabout30Hztoabove100Hz[24].Underspecificconditions,abandwidthgreaterthan300Hzhasbeenreportedwithoutanyelectronicsignaltreatment[25].Thisenhancedfrequencyresponsehasbeenattainedusingvariousgasnaturemediumandpressureinsidethecavity.Usingbothanalyticalandnumericaltools,modelingandsimulationofthermalaccelerometersstillneedtobelargelyinvestigated.Thisisduetothenumerousnewdesigngeometriesofferingsingle,dualandthreeaxesconfigurations.Asacontrasttosingleaxisthermalaccelerometers,wherecylindricalmodelisgener-allybeingused,dualandthreeaxiscounterpartsarenormallymoreadaptedforspherical3Dmodel.Severaltheoreticalanalysesusingthismodelaredetailedin[19–23,26,27].However,thermalaccelerometersspecificdesignparameterscanbequitedifferentfromonesensordesigntoanother.Therefore,tobeabletoproperlyusesphericalmodelexpressions,FEMsimulationscouldbeusedtofittheseparametersintodifferentmodelingequations.Thederivedresultwillbecompletegeneralanalyticalmodelexpressionswhicharefunctionofdifferentdesignparameters.Basedonanewlydeveloped3Dgeometricalmodelofthermalaccelerometers,FEMsimulations,usingtheAnsys©software,aredoneonthesensor.Theobtainedspecificvaluesandtheirrespectivefittingcurveexpressionsareusedindetailedanalyticalmodelingtoderivedifferentmodelingequationsofthesensorundertestgeneraldesign.Wewillfirsttreattheconductivebehaviorwhichincludestheexpressionsderivationoftheheattransfercoefficientandcommonmodetemperature.Second,wewillmodeltheconvectivebehaviorwhichcomesdowntothederivationofthedifferentialtemperatureexpressionasafunctionofkeygeometricdesignparametersandbiasingtemperatures.

4130|B.Mezghanietal.2ThermalaccelerometerandFEMmodelpresentationAllthreeknowntypesofMEMSconvectiveaccelerometershavesimilaroperatingmode,thoroughlyexplainedinliterature,concerningstaticbehavior:commonmodeandconvectionphenomena.Thisincludessingle,dualandthreeaxismicromachinedthermalaccelerometers.Differentconvectiveaccelerometerconfigurationsandde-signswerefabricatedthencharacterizedandpublished:–In[12],Garraudetal.presentedadual-axisCMOSMEMSthermalaccelerometer(Fig.1)withasquareshapedhotplatecontainingametallicmeanderformingtheheater.Sensingdirectionsrdet2.rin2.routtAlMicromachinedSiO2cavityPolysiliconFig.1.Schematicofadualaxisaccelerometershowingmaingeometryparametersandsensingdirections[12].Temperaturedetectorsareplacedorthogonallyintheheaterplane.Apost-processetchingstepisaddedtotheCMOSchiptoreleasethestructures.Detectorsaremadewithasetofthermocouples,whicharebasedontheSeebeckeffectandprovideapotentialdifferenceproportionaltothetemperaturedifferencebetweentheirhotandcoldjunctions.Thisgivesabetterperformanceofthebiasshiftwithrespecttoanyroomtemperaturevariation.

42Analysisofmicromachinedthermalaccelerometers|31–In[13],Parketal.introducedadual-axisMEMSconvectiveaccelerometerwithdiamond-shapedheatergeometry.Sincetemperaturegradient,proportionaltosensitivityislargerforthisheatershapethenitisbetterthanthesquareshape.Toachieveagreaterconcentrationofgeneratedheat,anewimplementationofheatercomplexpattern,whichisthinnerandlongerthanconventionalheaters,isused(Fig.2).Fig.2.SEMimageofadiamond-shapedheaterprototypeshowingnewheaterandthermopilesdesignimplementation[13].–In[14],Nguyenetal.publishedaprototypeofamonolithic3-axisthermalaccelerometer.ThisCMOSMEMSsenorisclaimedtobethefirsttobedesignedandfabricatedusingstandardplanartechnology,availablethroughtheCMPserviceinGrenoble,France.Theusedprocessisa0.35μmCMOSprocessfromAMS,followedbyapost-processusedtoformthebottomcavityandreleasethestructures.AsshowninFig.3,theheaterisdesignedonthecentralsquareplateandthefourthermaldetectorsareplacedonmediansofthecavity.Forxandyaxissensingdirections,allfourresistivesensorsareused.Forthethirdz-axisdetection,onlytwosensingresistancesareusedandtworeferenceresistancesareaddedonsiliconsubstrate.Eachz-sensingresistanceisthensplitintoapairtomeasuretheaveragecommonmodetemperatureonthefoursensinglocationsonxandymedians.Intheliteratureandalsoinpatents,itcanbeseenthatvariousresearchgroupsdesigned,fabricatedandtestedconvectiveaccelerometers.Mostshownresultsfocusonthechosendesigngeometryandfabricationsteps.However,bothanalyticalandnumericalanalyseswhichareinsomecasesbeingshownweredevelopedtofitonlyontheirspecificdesign.Thisleadstosaythatthespecificmodelingdevelopmentcannotbeusedforadifferentdesignorbiasingtemperatures.Moreover,onlyfewstudiesfocusedontheuseofanalyticalmodelinginpredictingsensorperformance

4332|B.Mezghanietal.asafunctionofkeygeometricparameters.Thisisobviouslyduetothecomplicatednatureofthiskindofstudyinsphericalmodelcoordinate.Fig.3.3Drepresentationofthemicro-heaterandthe4sensingresistances[14].Inthisstudy,wetrytouseanewlydeveloped3DgeometrymodelinFEMsimulationstoadapttheavailablesphericalmodelequationsonageneralthermalaccelerometerdesign.The3Dmodel,atfirstpresentedin[19,20]forasingleaxisthermalaccelerom-eter,isnearlythesamefordualandthreeaxisones.Moreover,theexactsameFEMtechniqueisusedtoevaluateaveragecommonmodeanddifferentialtemperaturevaluesofthesensor.Inthisactualstudy,numericalsimulationsareperformedforheattransferofalaminarandincompressibleflowofathermal-bubbleinsideanenclosedchamber.Incomputations,theair(fluidused)thermalpropertiesareassumedtobeconstantintheoperatingtemperaturerange.Boundaryconditionsforbothmicromachinedcavityandpackageareregardedastheisothermalwallsandaresetto300K.Inaddition,ameshingrefinementisusedsothatwehavethefinestmeshsizeneartheheater,temperaturedetectionlocation,andlargestattheboundaries.Thisusedmeshingmodelhasbeenvalidatedbyprovingthatsimulationresultsfordifferentdesignsareindependentofthemeshminimumsize.Inaddition,wehaveusedthermalpropertiesofcompressibleairasafunctionoftemperatureandsteadylaminarnaturalconvectionwithcontinuity,mass,energyandmomentumconservation.Wewillfirsttreattheconductivebehaviorwhichincludestheexpressionsderiva-tionoftheheattransfercoefficientandcommonmodetemperature.Second,wewillmodeltheconvectivebehaviorwhichcomesdowntothederivationofthedifferentialtemperatureexpression.Inthefollowingsections,thesensorparametersandtheir

44Analysisofmicromachinedthermalaccelerometers|33nominalvalues,usedinbothFEMsimulationsandequationsevaluation,aregiveninTab.1.Tab.1.ListofaccelerometerparametersandnominalvaluesSymbolDescriptionValueUnitrdDistancefromheatercentertodetector120μmrHHeaterhalfwidth30μmroCavityhalfwidth300μmh1Bottomcavitydepth300μmh2Topcoverheight1000μmeSuspendedstructuresthickness7μmd2Distancefrombottomcavityedgetocover600μmwTopcoverwidthw=2(ro+d2)1800μmAsshownin[21,22],valueschosenforrd,h2,h1andd2givemaximumsensitivityread-ingsfortherovalueof300μm.Inaddition,thespecifiedrHvalueoffersmaximumeffi-ciency.Thesekeygeometricparameterswhichwillbeusedinbothnumericalandan-alyticalmodelingareexplainedinthesimplifiedcross-sectionalviewshowninFig.4.Theanalyticalmodelingwhichwillbeusedconcernssphericalmodelwhichisthemostadaptedfordualandthreeaxisaccelerometerdesigngeometries.Thismodelisbasedonfreeconvectionheattransferbetweentwoconcentricspheres(Fig.5):fromaninnersmallspheretoanouterlargersphere.Relatingthismodeltoourthermalaccelerometer,wecanstatethat:–theinnersmallsphere,withradiusriandtemperatureTi,modelsthesuspendedcentralplatewheretheheaterisbeingimplemented,–theouterlargersphere,withradiusroandtemperatureTo,modelstheisothermalwallswhichareinourcasethesubstrateandpackage.SiPackagehHeaterBonding2BondingDetectorrordd22rHBottomh1cavitySisubstrateFig.4.Simplifiedthermalaccelerometercross-sectionalviewwithdifferentmodelingparameters.

4534|B.Mezghanietal.rTiTooriFig.5.Innerandouterspheres,modelingtheheaterandsubstrate.3ExpressionsdescribingtheconductionbehaviorHarryC.Hardeestudiedindetailsbothconductionandfreeconvectionheattransferinaclosedchamber[26,27].HeexpressedbothcommonmodeanddifferentialtemperaturesasafunctionofparametersshowninFig.5.Aswaspreviouslyshownin[21,22],thermalaccelerometersinnerandouterproducedheatisothermshapesshowconsiderabledeformationwhencomparedtoaperfectsphereshapeasthoseshowninFig.5.Specifically,theouterisothermradiusandformwerefoundtobecloselyrelatedtotheoverallsensorgeometryandthereforefixesbothconductiveanddifferentialtemperaturereadings.Amongmostthermalaccelerometerparameters,bottomcavitydepth,h1,istheonlydistancewhichisdefinitelyfixedonlyattheendofthepost-processmicromachiningstep.Thisisbecauseitcanbeeasilyaffectedbyetchingtimeandetchingsolutionmovements,compositionandalsotemperature.Asaresult,thefinalcavitydepthvaluecanbeeasilymodifiedifidealconditionsarenotmetduringthemicromachiningpost-process.Basedontheseimportantfabricationdetails,oneofourmainmodelingparameterwillthereforebeh1.Inconvectiveaccelerometers,heatconductionbehaviorcanbeanalyzedbythespecificationoftwomainparameters:–Heattransfercoefficientwhichisrelatedprimarilytotheheatertemperaturethroughheatconductioninthecavity,andiscloselyrelatedtocavitydepth,–Commonmodetemperaturesetfromheatertemperaturevalueandwhichwillsetdetectorsinitialtemperature.3.1Heater-HeattransfercoefficientIn[23],thestepbystepderivationoftheheattransfercoefficienthH,expressionwasdetailed.Theevaluationofananalyticvalueofthiscoefficientcanbedoneusingthefollowingexpressions:Roλ0Cλ(T)hH=(1)Ri(Ro−Ri)

46Analysisofmicromachinedthermalaccelerometers|35whereCλ(T)isgivenby:T2−T2T3−T3T4−T4δ1ioδ2ioδ3ioCλ(T)=1+×+×+×(2)2Ti−To3Ti−To4Ti−Towhereλ=−3.93×10−4Wm−1K−1istheairconductivityextrapolatedatT=0K(the0negativesignisduetothefactthatweextrapolateat0Kanexpressionthatisonlyvalidfrom100Kto600K).Theotherthreeparametersaregivenby:δ=−0.259K−1,1δ=1.23×10−4K−2andδ=−3.87×10−8K−3,whichrepresentthecoefficientsof23thermalconductivityvariationforair.TheequivalentinnersphereequivalentradiusvalueisRi=28μm.Thischosenvalueisconsideredtobeatrade-offbetweenaradiusobtainedfromthecentralplateandthatoftheheatermeander.TheouterisothermequivalentradiusRocanbefoundusingthefollowingexpression:roh1Ro=×(3)2r4440h1+3Asitisquiteclear,theanalyticalevaluatedvalueoftheheattransfercoefficientwillbedependentonbothdesigngeometryparametersandheater/substratetemperatures.Therefore,wecanpredictthisvaluewellbeforefabricationwhichispreferablyattheearlydesignlevel.3.2Commonmode-FluidconductionDetectors’initialtemperatureissetbyheatconductionintheairfillingthecavity.ThisparameterisdenotedcommonmodetemperatureTCM.TheanalyticalgeneralexpressionderivationofTCMasafunctionofdesigngeometryparametersandbiasingtemperatureswasalsodetailedbyMezghanietal.in[23].Thisexpressionisgivenby:δ122δ233δ344(TCM−To)+2(TCM−To)+3(TCM−To)+4(TCM−To)(4)RδδδRio(Ro−d)122233344=Rd(R(Ti−To)+2(Ti−To)+3(Ti−To)+4(Ti−To)oo−Ri)whereC(T)isgivenby:λT2−T2T3−T3T4−T4δ1CMoδ2CMoδ3CMoCλ(T)=1+×+×+×(5)2TCM−To3TCM−To4TCM−ToTheequivalentouterradiusRderivedanalyticalexpressionisgivenby:oro+dh1ro−dRo(h1)=2×+2(6)44ro4h1+2

4736|B.Mezghanietal.Again,itisquiteclearthattheanalyticalevaluatedvalueofthecommonmodetem-peraturewillbedependentonbothdesigngeometryparametersandheater/substratetemperatures.Therefore,wecanpredictthisvaluewellbeforefabricationwhichispreferablyattheearlydesignlevel.4ExpressionsderivationdescribingconvectionbehaviorThesecondmostimportantparameterdescribingsensorperformanceisknownassensitivity,S.Whennoaccelerationisapplied,temperaturedetectorsarelo-catedonidenticaltemperatureisotherms,obtainedforsymmetryreasons.Therefore,common-modetemperature(TCM)willbeindicatedbybothdetectors.Detectortemperaturereadingismadefeasibleusingtwodifferenttechniques.Oneusesthehighresistivitydependenceofpolysilicontotemperature(TemperatureCoefficientofResistance,TCR=9×10−4/rC).Thesecondusesasetofthermocouples,whichareˇbasedontheSeebeckeffectandprovideapotentialdifferenceproportionaltothetemperaturedifferencebetweentheirhotandcoldjunctions.Underaccelerationalongaspecificsensitiveaxis,cavitytemperaturedistributiondeformsduetofreeconvectionandeachdetectorwillthereforeindicateanewtemperature,TDi.Then,thetemperaturevariation∆TDiofeachdetectorcanbeevaluatedbytakingthedifferencebetweenthetwopreviousreadings(∆TDi=TDi−TCM).Thedifferentialtemperature,proportionaltothesensorsensitivity,isthenobtainedfrombothdetectorsandgivenbyδTD=∆TD1−∆TD2.Finally,thethermalsignalisconvertedintoanoutputvoltagebymeansofanintegratedWheatstonebridge.Thisvoltageisthenamplifiedbyanon-chipinstrumentationamplifier.Sensorsensitivity(SexpressedinK/g)isthereforeproportionaltothemeasureddifferentialtemperatureδTD:δTDδTD=∆TD1−∆TD2andS=(7)ΓConsequently,whenanacceleration(Γing,1g=9.81ms−2)of1gisappliedalongonesensitiveaxis,thedifferentialtemperature(δTD)isobtainedfromtemperaturevariations∆TD1and∆TD2,whicharesymmetricallymeasuredbybothdetectors.Fromtheabovestudy,weclearlyseethattheanalysisofsensitivityexpressioncanbedonethroughthestudyofthedifferentialtemperaturebehaviorandmeasuredvaluesusingunityacceleration.Inthefollowing,wewillreportthisstudyandwillexpressthesensordifferentialtemperatureexpressionasafunctionofkeygeometricdesignparametersandbiasingtemperaturevalues.

48Analysisofmicromachinedthermalaccelerometers|374.1DifferentialtemperaturedependenceoncavitydepthOneofthemostimportantgeometricparameterisobviouslythedepthofthebottommicromachinedcavity.Thisisclearlyexplainedatthebeginningofsectionthree.Therefore,wewillstartourdifferentialtemperaturemodelingfromthisspecificparameter.Alongwiththislattervariable,theheatertemperaturevalueisalsoconsideredextremelyimportantsinceitistheonlyparameterthatfixestheinitialtemperaturesettingsinthecavitywhichobviouslywilldeterminethisthermalsensoroverallresponse.Inordertobeabletofindthisdependence,FEMsimulationsweredone,usingournew3Dmodel,toevaluatedifferentialtemperaturevaluesforvarioustechnologicallypossiblebottomcavitydepthsandfordifferentheatertemperatures.Next,theseFEMevaluatedvaluesareplottedandafittingisdoneonthedatapoints.BothplottedvaluesandrespectivefittingcurvesareshowninFig.6.0.06Ti=600K0.05T=550Ki0.04T=500KiδT(K)0.030.02Ti=440K0.01T=390KiT=350Ki0050100150200250300350400Bottomcavitydepth,h1(μm)Fig.6.Differentialtemperaturevs.bottomcavitydepthfordifferentheatertemperatures(solidline:fittingcurves,dots:FEMdata).FromFig.6,wecanclearlyseethatdifferentialtemperaturevaluesareproportionaltoheatertemperatures.Thisphenomenonisonlyduetothevariationofheatisothermssurroundingtheheaterasitstemperatureisbeingvaried.Itisthereforeconcludedthat:–forhighh1values,differentialtemperaturevaluesareconstantforconstantheatertemperatures.Thiscanbeexplainedbythefactthatsiliconbodywallsofthemicromachinedcavitywillbethelimitingfactoroftheheatbubbles.

4938|B.Mezghanietal.–forlowh1values,differentialtemperaturevaluesincreaseforconstantheatertemperatures.Asbottomsubstratesilicongetsclosertotheheater,thismicro-machinedcavitydepthwillbethelimitingfactoroftheheatbubbleswhichinturnareconsideredtobetheonlymechanismthatfixesthemeasureddifferentialtemperature.Thiscanbeexplainedbytheshapeofthehotbubblecreatedbytheheater,asillustratedinFig.7.Inthecasethatcavitydepthishighenough;thefourlateralsiliconcavitysidewallswilllimitthehotbubblesize.Therefore,hotbubblesizeisnotaffectedbycavitydepthvalueandthedifferentialtemperaturejustdependsonheatertemperatureandcavitylateraldimensions.Ontheotherhand,thehotbubblesizereducesconsiderablyforlowcavitydepthvalues,whichbecomesthemainlimitingfactoroftheproducedisothermsandthereforeofthedifferentialtemperaturemeasuredvalues.(a)h1=300μm(b)h1=75μmFig.7.Temperatureprofilecrosssectioninsidecavityfortwocavitydepths.FromFig.6,wecanseethatthetransitionbetweenthesetwobehaviorsoccursaroundh1=150μm.Itisveryimportanttonotethatthistransitionvalueisindependentofheatertemperature.Thefittingofonesinglecurvecanbewrittenusingthefollowingexpression:h1δTh1=4(8)44−6h1+150×10Inthisequation,thedenominatorrootorderwaschosentogivethebesttransitionfitbetweenlinearandsaturationregionsshowninFig.6.Tomakethisexpressionageneraloneandspecificallydependantonthekeygeometricparameterwhichfixesthecavitysidewallsdistance(cavityhalfwidth,ro),weshouldincludethislatterparameterinthefittingexpression.Here,weexpressthedifferentialtemperatureexpressiontoreflectalinearrelationshipforlowvaluesofh1andasaturationata

50Analysisofmicromachinedthermalaccelerometers|39valuearoundro=150μmforhighvaluesofh1.Aswecannotice,thiscorrespondstoadistanceof(ro/2)fromtheheatercentertothecavityborder.Therefore,thegeneralfittingexpressionforasinglecurveinFig.6canbewrittenusingthefollowingformula:h1δTh1=4(9)44roh1+2Itshouldbenotedthattheeffectofroismodeledinthisexpressiononlyforlowvaluesofh1.However,theeffectofroisnottakenintoaccountforbothtransitionandsaturationeffects.Thiswillbestudiedindetailinthenextsectionsbyanalyzingthecompleterelationofcavityhalfwidthvariationondifferentialtemperaturereadings.4.2DifferentialtemperaturedependenceonheatertemperatureInordertobeabletospecifydifferentialtemperatureexpressionforallheatertemperaturesshowninFig.6,wehavetoquantifyinthefirstplacethedifferentialtemperaturevariationasafunctionofheatertemperaturevalue.Thisbiasingtemper-aturerelationcanbeeasilyfoundfromcurvefittingofmaximumδTDvaluesplottedasafunctionofheatertemperature.Therefore,toexpressthedependenceofsensorsensitivityonheatertemperature,weproposetofindtherelationbetweenthemaximumdifferentialtemperaturereadings,evaluatedusingFEMsimulations,andheatertemperature.TherelationcanbeeasilyfoundthroughcurvefittingofFEMdatapointswhichwillthengivetheanalyticalexpressiondescribingthisimportantdependence.ThefittingexpressionoftheFEMdatapointscanbewrittenas:−61.7δTTi=3.6×10(Ti−To)(10)Therefore,usingunityacceleration,sensorsensitivitydependenceonbothheatertemperatureandmicromachinedcavitydepthcanbeexpressedas:−61.7h1δTh1,Ti=3.6×10(Ti−To)×4(11)44roh1+2Again,itshouldbenotedherethatthisexpressionmodelstheeffectofroonlyforlowvaluesofh1.Forsensitivitytransitionandsaturationregions,thiseffectisnottakenintoaccountinthisexpression.Asitwasclearlyexplainedearlier,sincethe

5140|B.Mezghanietal.volumewhereconvectionphenomenaoccurswillbelargelyaffectedwhenrovaries,therefore,thiswillhaveadirecteffectondifferentialtemperaturemeasuredbybothdetectors.Accordingly,transitionandsaturationsensitivityvalueswillbelargelyaffected.Thisgivesthehalfwidthcavitydesignparameteranaddedimportanceandshouldthereforebeincludedinthemodelinggeneralequation.Thiswillhelpinpredictingtheeffectofthisparametervariationonthethermalaccelerometersensitivityvalueandthereforeontheoverallsensorperformance.4.3DifferentialtemperaturedependenceoncavitywidthTostudythedifferentialtemperaturevaluedependenceonrofordifferentsensordesigngeometries,weusethe3DmodelinFEMsimulationswhereallotherdesignparametersaresettotheirmaximumvalues.Therefore,weuseasquarecoverwithh2=w=3mm,heatertemperatureTHvariable,cavitydepthh1variableandcavityhalfwidthrovariable[200μm–800μm].Heatertemperatureandcavitydepthvariationsaredoneinpairsgivenby:[(TH,h1)=(390K,0.25ro),(440K,0.5ro),(500K,0.75ro),(550K,ro),(600K,1.25ro)].Thisisdonesothatallpossiblevariationvalueswillbestudiedandderivedexpressionscanthereforebeasgeneralaspossible.Thismeansthatthesederivedexpressionscanbeusedtoestimatedifferentialtemperaturereadingsforanytechnologicallypossibledesigngeometryandalsoatanearlydesignstage.Usingtheabovesetofdata,variousFEMsimulationsweredonetocomputeandplotsensitivity(differentialtemperature)valuesforafixedcavityhalfwidthofro=300μmasafunctionofvariousrovaluesfrom200μmto800μm.EvaluateddatapointsareplottedinFig.8,wherethedashedlinesareincludedtoshowthatthereisalinearrelationshipbetweendifferentialtemperatureandmicromachinedcavityhalfwidth.Thislinearrelationshipbetweenevaluateddifferentialtemperaturevalues(proportionaltosensitivity)andcavityhalfwidthvaluesobviouslysaysthatwhenvaryingro,sensitivitywillbemultipliedbyafactor.Thissocalledmultiplicationfactorcanbecomputedbyplottingitandthenusingdatafittingonthesepoints.Thecurvefittingexpressionwillbetheexactmultiplicationfactorwhichcanbeusedasageneralmodelingexpressiontopredictthesensitivityvaluewhencavitywidthisvaried.BothofthesedatapointsandrespectivefittingcurveareshowninFig.9.FromFEMdatafitting,wededucetheanalyticexpressiondescribingtheeffectofrovariationsondifferentialtemperaturereadings.Thismultiplicationfactor,Cro,isgivenby:62Cro=2.63×10ro+5119ro(12)

52Analysisofmicromachinedthermalaccelerometers|41450r=200μ400or=300μo350ro=400μr=500μo300r=600μor=700μo250r=800μo=200−800μm(mK/g)200o150100Sensitivityforr5000102030405060708090Sensitivityforro=300μm(mK/g)Fig.8.Sensitivityvaluesforafixedcavitywidthvssensitivityvaluesusingvariouscavitywidthvalues(dashedlinestoshowthelinearrelationship).65o432Multipicationfactor,CrFEMdata1Fittingcurve0200300400500600700800Cavityhalfwidth,ro(μm)Fig.9.Multiplicationfactorforvariouscavityhalfwidthvalues(solidline:fittingcurve,dots:FEMdatapoints).Itshouldbenotedherethattherovalueisinμm.Furthermore,thisfittingexpressioncannotbeusedbyitselfandshouldbeaddedtotheexpressionofthedifferentialtemperaturepreviouslyderived.Thiswillgiveageneralmodelingequationwhichwill

5342|B.Mezghanietal.predictthevalueofthedifferentialtemperatureforafixedh1,Tiandrovalues.Thisexpressioncanthereforebewrittenunderthefollowingform:−61.7h1δTh1,Ti,ro=3.6×10(Ti−To)×444roh1+262×2.63×10ro+5119ro(13)Anotherveryimportantparameteristheoptimalpositionofdetectorsinthecavity.Thisalsocanbeseenashowfaristhedetectorfromtheheater.Thisisacrucialconcernsinceitcandramaticallyaffectdifferentialtemperaturereadingsifdetectorsarenotplacedintheiroptimallocation.Tostudytheeffectofdetectorslocationondifferentialtemperaturereadings,FEMdatapointsaretakenalongthesensitivex-axisforvarioushalfcavitywidthvalues.Whenplotted(Fig.10),thesevaluesclearlyshowthatdetectorslocationcaninfactaffectdifferentialtemperaturereadings.250200150100500δT(mK)−50r=300μm−100oro=400μm−150ro=500μmr=600μmo−200r=700μmo−250−600−400−2000200400600xpositioninthecavity(μm)Fig.10.Differentialtemperaturevaluesevolutionalongsensitivex-axisforvariouscavityhalfwidths.Thisprovesthatdetectorlocationinthecavityisacrucialfactorandshouldbeincludedinthisspecificmodelingofdifferentialtemperatureasafunctionofcavitywidth.Thisanalysisandequationsderivationshouldobviouslybecontinuedsothatitincludesfurtherdesignparameterstofinallygiveageneralmodelingexpressionthatcontainsallgeometryfactorsandbiasingtemperatureofthermalaccelerometers.The

54Analysisofmicromachinedthermalaccelerometers|43otherdesignparameterswhichcanbeincludedinthestudyconcerncoverheightandwidthandalsoheaterwidthandthickness.5ConclusionInthispaper,weusedanewlydeveloped3DmodelinFEMsimulationswithpreviouslyderivedanalyticalmodeltostudyconductionandconvectionbehaviorofCMOSMEMSthermalaccelerometers.Heatconductiondistributioninthecavityisgovernedbyheatertemperature,setbyitsheattransfercoefficient.Thiswillfixcavitycommonmodetemperatureandspecificallyalongeachsensitiveaxis.Moreover,thevolumewheretheconductionphenomenonistotakeplaceisfixedbybottomcavitydepth.ThesedominatingparametersareanalyticallymodeledandvalueswereverifiedusingFEMsimulations.Adetailedmodelingofdifferentialtemperatureasafunctionofheatertemperaturealongwithmicromachinedcavitydepthandwidthhasbeendone.Aderivedexpressionwasgivenwhichincludestheseeffectsandthereforepredictsasensitivitylevelatanearlydesignstage.Thisanalyticalstudycouldobviouslybeappliedfordifferentsensordesignparametersandbiasingtemperatures.Acknowledgments:AuthorsareindebtedtotheircolleaguesattheLIRMMinFranceforprovidingvaluablehelpduringthefirststepsinunderstandingconvectiveac-celerometermodeling.SpecialthanksareduetoFredericMaillyandPascalNouetnotonlyforproviding2DAnsysscript,butalsoforofferingtheuseofavailablesimulationtools.Bibliography[1]S.SemancikandR.Cavicchi.KineticallyControlledChemicalSensingUsingMicromachinedStructures.AccountsofChemicalResearch,31(5):279–287,1998.[2]A.Padmanabhan,M.Sheplak,K.S.BreuerandM.A.Schmidt.Micromachinedsensorsforstaticanddynamicshear-stressmeasurementsinaerodynamicflows.Int.Conf.onSolidStateSensorsandActuators,TRANSDUCERS’97,1(16–19):137–140,1997.[3]H.LakdawalaandG.K.Fedder.TemperaturecontrolofCMOSmicromachinedsensors.5thIEEEInt.Conf.onMicroElectroMechanicalSystems,:324–327,2002.[4]A.M.Shkel,A.A.Trusov,I.P.PrikhodkoandS.A.Zotov.Three-dimensionalwafer-scalebatch-micromachinedsensorandmethodoffabricationforthesame.UnitedStatesPatent8,567,247,October29,2013.[5]J.Hermann.Micromachinedmeasuringcellwitharmsupportedsensor.UnitedStatesPatent5,551,294,September3,1996.[6]M.L.Lee,C.H.Je,S.S.Lee,S.H.Jung,C.A.ChoiandG.Hwang.Micromachinedsensorformeasuringvibration.UnitedStatesPatent7,975,550,July12,2011.[7]B.E.BoserandR.T.Howe.Surfacemicromachinedaccelerometers.IEEEJ.ofSolid-StateCircuits,31(3):366–375,1996.

5544|B.Mezghanietal.[8]T.MinetaS.,Kobayashi,Y.Watanabe,S.Kanauchi,I.Nakagawa,E.WuganumaandM.Esashi.Three-axiscapacitiveaccelerometerwithuniformaxialsensitivities.J.ofMicromechanicsandMicroengineering,6(4):431–435,1996.[9]A.Partridge,J.K.Reynolds,B.W.Chui,E.M.Chow,A.M.Fitzgerald,L.Zhang,N.I.Maluf,T.W.Kenny.Ahigh-performanceplanarpiezoresistiveaccelerometer.J.MicroelectromechanicalSystems,9:58–66,2000.[10]L.C.SpanglerandC.J.Kemp.ISAAC:integratesiliconautomotiveaccelerometer.SensorsandActuatorsA,54:523–529,1996.[11]S.J.ChenandC.H.Shen.Anoveltwo-axisCMOSaccelerometerbasedonthermalconvection.IEEETrans.onInstrumentationandMeasurement,57(8),2008.[12]A.Garraud,A.Giani,P.Combette,B.CharlotandM.Richard.AdualaxisCMOSmicromachinedconvectivethermalaccelerometer.SensorsandActuatorsA,170:44–50,2011.[13]P.Usung,P.Byoungkyoo,M.Il-Kwon,K.DongsikandK.Joonwon.Developmentofadual-axismicromachinedconvectiveaccelerometerwithaneffectiveheatergeometry.MicroelectronicEngineering88(3):276–281,2011.[14]H.-B.Nguyen,F.Mailly,L.LatorreandP.Nouet.Designofamonolithic3-axisThermalConvectiveAccelerometer.Symp.onDesign,Test,IntegrationandPackagingofMEMS/MOEMS(DTIP),Barcelona,Spain,2013.[15]Y.ZhaoandY.Cai.Z-axisthermalaccelerometer.UnitedStatesPatent7,392,703,July,2008.[16]A.A.Rekik,B.Mezghani,F.Azaïz,N.Dumas,M.Masmoudi,F.MaillyandP.Nouet.InvestigationontheeffectofgeometricaldimensionsontheconductivebehaviourofMEMSconvectiveaccelerometers.Symp.onDesign,Test,IntegrationandPackagingofMEMS/MOEMS(DTIP),Aix-en-Provence,France,2011.[17]X.B.Luo,Z.X.Li,Z.Y.GuoandY.J.Yang.Thermaloptimizationonmicromachinedconvectiveaccelerometer.Int.J.ofHeatandMassTransfer,38(7–8):705–712,2002.[18]J.Courteaud,P.CombetteandN.Crespy.Thermalsimulationandexperimentalresultsofamicromachinedthermalinclinometer.SensorsandActuatorsA,141:307–313,2008.[19]B.Mezghani,A.Brahim,F.Tounsi,A.A.Rekik,M.MasmoudiandP.Nouet.From2Dto3DFEMsimulationsofaCMOSMEMSconvectiveaccelerometer.Int.Conf.onMicroelectronics(ICM),Tunisia,2011.[20]B.Mezghani,F.Tounsi,A.A.Rekik,F.Mailly,M.MasmoudiandP.Nouet.SensitivityandpowermodelingofCMOSMEMSsingleaxisconvectiveaccelerometers.MicroelectronicsJ.,July2013.[21]B.Mezghani,F.TounsiandM.Masmoudi.SensitivityModelingofdual-axisCMOSMEMSConvectiveAccelerometersusingFEMandSphericalModel.Symp.onDesign,Test,IntegrationandPackagingofMEMS/MOEMS(DTIP),Barcelona,Spain,2013.[22]B.Mezghani,F.TounsiandM.MasmoudiM.ConvectionbehavioranalysisofCMOSMEMSthermalaccelerometersusingFEMandHardee’smodel.AnalogIntegratedCircuitsandSignalProcessing(ALOG),Springer,2013.[23]B.Mezghani,F.Tounsi,H.YaichandM.Masmoudi.ConductiveBehaviorModelingofDual-axisCMOSMEMSConvectiveAccelerometersUsing3DFEMandSphericalModel.Int.Multi-Conf.onSystems,Signals&Devices,(IEEE-SSD’13),Tunisia,2013.[24]http://www.memsic.com/index.cfm.[25]A.Garraud,P.Combette,F.Pichot,J.Courteaud,B.CharlotandA.Giani.Frequencyresponseanalysisofanaccelerometerbasedonthermalconvection,J.ofMicromechanicsandMicroengineering21(3),2011.[26]H.C.Hardee.NaturalconvectionbetweenconcentricspheresatlowRayleighnumbers.Ph.D.Dissertation,UniversityofTexas,Austin,Texas,1966.[27]L.R.MackandH.C.Hardee.NaturalconvectionbetweenconcentricspheresatlowRayleighnumbers.Int.J.ofHeatandMassTransfer,11(3):387–396,1968.

56Analysisofmicromachinedthermalaccelerometers|45BiographiesBrahimMezghanireceivedtheBScandMScdegreesfromtheUniversityofMinnesota,U.S.A.HereceivedthePhDdegreefromNationalEngineeringSchoolofSfax(ENIS)inTunisia,whereheiscurrentlyholdinganassistantprofessorposition.Heiscurrentlyworkingonthedesign,modelingandsimulationofnewmicromachinedsensors.Thisincludesbothmechanicalandelectronicconditioningparts.Hisrecentinterestsincludethedevelopmentofnanomaterialsforsensorapplications.FaresTounsireceivedtheBSc’01andMSc’03degreesfromNationalEngineeringSchoolofSfax(ENIS)inTunisia,andthePhD’10inMicroandNano-electronicsfromGrenobleInstituteofTechnology(INPG),France.HeiscurrentlyanassistantprofessorinInstitutSupérieurd’InformatiqueetdeMathématiquesdeMonastir(ISIMM),Tunisia.Heiscurrentlyworkingonthedesign,modelingandsimulationofnewCMOScompatiblemicromachinedsensors.Specifically,heisinterestedinnoveldesignsofmicrophones,accelerometersandRFswitches.Inaddition,heisnowfocusedonthefieldofnanotransducers,evaluationofnewmaterials/structuresforMEMSandadvancedmicrosystems.MohamedMasmoudiwasborninSfax,Tunisiain1961.HereceivedtheBSc’85fromNationalEngineeringSchoolofSfax(ENIS)andthePhD’89inMicroelectronicsfromtheLaboratoryofComputerSciences,RoboticsandMicroelectronicsofMontpellier,France.From1989to1994,hewasholdinganAssociateProfessorpositionatNationalEngineeringSchoolofMonastir,Tunisia.Since1995,hehasbeenaProfessoratENISwherehehasbeenengagedindevelopingMicroelectronicsintheengineeringprogramoftheuniversity,andwhereheisalsotheheadoftheresearchgroupElectronics,MicrotechnologyandCommunication(EMC).Heistheauthorandco-authorofseveralpapersintheMicroelectronicfield.Hehasbeenareviewerforseveraljournals.Prof.MasmoudiorganizedseveralinternationalConferencesandhasservedonseveraltechnicalprogramcommittees.

57

58M.HadjSaid,F.Tounsi,P.Gkotsis,M.MasmoudiandL.A.FrancisMEMS-BasedClamped-ClampedBeamResonatorCapacitiveMagnetometerAbstract:Inthispaper,aMEMSresonantmagnetometerbasedonLorentzforceispresentedmodeledandcharacterized.AMEMSmagnetometersensorisgenerallyusedtodetectthevariationofanexternalmagneticfieldusingaresonantstructuresuspended.Intheproposeddesign,thedeformationofthisstructure,excitedatitsfirstresonancefrequencythroughtheestablishmentofaLorentzforce,willbemeasuredbyacapacitivesensingmethod.TheLorentzforceisgeneratedbytheinter-actionofanalternativecurrent(AC)whichexcitesthesuspendedclamped-clampedconductingbeam,obtainedthroughabulkmicro-machining,anda150mTexternalstaticmagneticfield.Theout-of-planestructuredisplacementisconvertedintoacapacitancechangebyaseriesofinterdigitatedcombs.ThefrequencyoftheACsupplycurrentischosentobeequaltotheresonantfrequencyofthesuspendedbeam,leadingtotheincreaseofitsdeflectionmagnitudeandtheoverallsensitivity.Through-outthispaper,thefirst-orderresonancefrequencyandthecapacityexpressionsaredeveloped,simulatedwithComsolandthereaftercomparedtothemeasuredvalues.Theresonancefrequencysimulatedandmeasuredisinthevicinityof15kHz.Ontheotherhand,thequality-factor,electricmodelandsensitivityofthesensor,whichisfoundtobearound10fF/mA,aredeterminedfromthecharacterizationofthefabricatedsensorprototype.Keywords:Capacitivesensing,Lorentzforce,Magneticfieldsensor,MEMSMagne-tometer,ResonantSensors.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A221IntroductionMAGNETIC-fieldsensorsaremostlybasedintheiroperationprincipleontheHalleffect,magneto-resistive(amorphousandgiantmagneto-resistors),magneto-transistor,magneto-diode,fluxgate,orothersemiconductoreffectsinordertomeasurethemagneticfields[1–5].MagnetometersensorsoffernewcapabilitiesM.HadjSaid,P.Gkotsis,L.A.Francis:Sensors,MicrosystemsandActuatorsLabora-toryofLouvain-la-Neuve(SMALL),UniversitécatholiquedeLouvain,Belgium,email:Laurent.Francis@uclouvain.be.M.HadjSaid,F.Tounsi,M.Masmoudi:UniversityofSfax,NationalEngineeringSchoolofSfax,Electronics,MicrotechnologyandCommunication(EMC),Sfax,Tunisia,email:mohamed.masmoudi@enis.rnu.tn.DeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.47–62.DOI10.1515/9783110470444-004

5948|HadjSaidetal.thatcanbeusedindifferentdomainsofengineering,industry,militaryortelecommunications[6].Thelistofmagneticsensorapplicationsincludesposition-sensing,noncontactswitching[7],vehicledetection[8],navigation[9],mineralprospecting[10],brainfunctionmapping[11],etc...Insomebiomedicalapplications,withthehighrequirementsinsensitivityandaccuracy,magnetometersshouldalsobesmallenoughandhavelowpowerconsumption,whereastheperformancesofmostpresentsensorsarenotsatisfactory.MEMS-basedmagneticfieldsensorprovidesanopportunitytosolvethisproblembyofferingsmall-sizesolutionformagneticfieldsensing.Inaddition,smallerdevicecanbeplacedclosertothemeasurementspotsandtherebyachievinghigherspatialresolution.ForMEMStechnologies,manyresonantmagneticfieldsensorsexploittheLorentzforceprinciple,wherethestructuredisplacement,whichvibratesattheresonance,ismeasuredeitherbyoptical,piezoresistive,orcapacitivesensingtechniques[6].Thefirstsectionofthisarticlewillintroducetheoperationalprincipleoftheresonantsensor.Thesecondsectionwillprovideacomparisonbetweenthetheoret-icalandthemeasuredresonancefrequencyvaluesforbothtypesofmanufacturedprototypesfollowedbytheinterdigitatedcombscapacitanceevaluationofthesensor.Then,afterthedeterminationofthesensorqualityfactor,thecompleteelectricmodelwillbededucedandanalyzed.Finally,thesensorsensitivity,whichisafunctionofthecapacitancevariation,wasdetermined.2OperationalprincipleSuspendedstructureshaveaninfinitenumberofresonancevibrationmodes(oreigenmodes).Thestructuredeflectionmagnitudeattheresonancefrequencyisthemaximumthatcanbeachieved,thustheresonantsensormostlyusesstructuresoperatingatoneoftheireigenmodes,generallythefirstvibrationmode.Basedonthisnotion,thedesignedsensorconsistsofasuspendedclamped-clampedbeamwithcombelectrodesdefinedalongitsentirelength.Theseelectrodesforminterdigitatedpairswithcombsthatareanchoredtothesubstrate.Whenthesus-pendedclamped-clampedbeamvibratesintheout-of-planedirection(z-axis),theinterdigitatedelectrodescapacitanceschange(seeFig.1).Ifanalternativecurrentix,whichpassesthroughthesuspendedbeam(alongthex-axis),iscoupledtoamagneticfieldBy(alongthey-axis),aLorentzforcewillbegeneratedalongthez-directionandsubsequentlythebeamwillmoveoutofplane.TheLorentzforceexpressionisgivenasfollowing:FL=ixByL(1)

60MEMS-basedclamped-clampedbeamresonatorcapacitive|49whereListhebeamlength.Thebeamshouldbelongenoughtopermitarelativelyhighnumberofcombfingers,andeventuallyasufficientvariationinthecapacitanceformedbytheinterdigitatedcombs.ThechoiceofLisdefinedbythemaximumareaofferedbythefabricationtechnologyandthatcanbeetchedonthesubstrate.Moreover,thecombwidthshouldbeminimizedinordertoreducethedampingeffectsduetoair,whileitsthicknessisalsodefinedbythefabricationprocess.TwoprototypesofthisMEMSmagnetometerhavebeenfabricatedthroughtheSOIMUMPsfoundryprocess(MEMSCAPInc.,USA).Thefirstprototypehasasus-pendedconductingbeamthicknessequalto10μmwhilethesecondprototypeisequalto25μm.ThestructuregeometricparametersthatwereusedforthefabricationofthemagneticfieldsensorareshowninTab.1.Inthemanufacturedlayout,threecontactpadspermittheelectricalaccessesformeasurement;padsAandBallowaccesstothetwoendsoftheclamped-clampedbeamandpadCisconnectedtotheelectrodeswhicharefixedonthesubstrateandisusedforthecapacitancemeasurements.800MovableFinger-600FixedFinger-ElectrodesElectrodes400yzx2000AB–200CCurrent–400BeammagneticfieldB–400–200020040060080010001200140016001800200022002400Fig.1.TopviewoftheMEMS-basedmagnetic-fieldsensor.3DeterminationoftheresonancefrequencyTheclamped-clampedbeamusedhasarectangularsectionshapefixedatitsbothends.Duringthesensoroperation,itwillbesubjectedtobendingduetotheLorentzforce.Tomaximizethedeformationvalue,thebeamwillbeexcitedwithitsresonancefrequency.TheresonantfrequencyofsuspendedstructuresisusuallydeterminedbytheRayleighmethod[6].Thismethoddeterminestheeigenmodesbytheratiobetweenthemaximumpotentialenergyandthemaximumkineticenergyofthestructure[6].Fromthisdefinition,then-eigenmodesfrequencyformulaofaclamped-clamped

6150|HadjSaidetal.Tab.1.Useddimensionsintheresonantmagnetometersensordesign.PartSensorparametersValueLength(L)1.969μmBeamWidth(W)62.5μmHeight(h)10μmor25μmLength(l)100μmWidth(a)3μmFingersThickness(t)10μmor25μmFingersnumber(n)312Gapbetweenfingers(d)3μmbeamisgivenbythefollowingequation:R2EInfn=(2)2L2ρAwhereρisthedensityofthestructuralmaterial,Aisthebeamsectionarea,EistheYoung’smodulus,Rnisaconstantthatdependsontheboundaryconditionsandthenaturalmodeofthebeam,itsvaluesaregiveninTab.2,andfinallyIistheareamomentofinertiagivenbythefollowingequation:wh3I=(3)12Tab.2.Rn-constantvaluesforthefirstfiveeigenmodes.ParameterR1R2R3R4R5Value4.7387.85310.99514.13117.278ThusaftersubstitutingtheR1-value,theexpressionforthefirstresonantfrequencyoftheclamped-clampedbeamcanbegivenby[12]:4.7382EIf1=(4)2L2ρAWithinthesameobjective,amodalanalysisoftheclamped-clampedbeamstructurewasperformedusingComsolMultiphysicsinordertodetermineitsresonancefrequenciesandcompareittotheanalyticalvalue.Figure2showsthefirstmodal

62MEMS-basedclamped-clampedbeamresonatorcapacitive|51shapeofthebeam,andTab.3summarizesthefirstresonantfrequencyofthe10μmand25μm-thickbeamstructuresobtainedbysimulations.Forsymmetryreasonsandtosavetimeduringthesimulation,wemodeledonlythequarterofthebeam(seeFig.2).Thetheoreticalandsimulationresultsarequitesimilar,andaresummarizedinTab.3forthebothsensordesignedstructures.05004020Eigenfrequency=15148.737851Hz0–500Fig.2.Thebeamdisplacementwithinthefirstvibrationmode(forthe10μm-thickbeam).Experimentally,theresonantfrequenciesareobtainedbythelaserDopplervelocime-try(PolytecMSA-500),wherethebeamisactuatedbyaperiodicchirpsignaloffixedamplitudewhichisappliedbetweenpadsAandB(seeFig.1).Indeed,thedeviceisplacedina150mTstaticmagneticfieldgeneratedbyasetoftwoneodymiumpermanentmagnetsplacedinfrontofeachothersinordertoobtainafieldashomogeneousaspossible.TheresonancefrequenciesresultsexposedinFig.3showalargediscrepancywiththosefoundintheoreticalandsimulation.ThedifferenceiscausedbytheJouleheatingeffectduetothermsvalueofthedrivingcurrent,andtheincreaseofthethermalenergyofthebeam.Becausethebeamisclampedatbothends,thisincreaseleadstoanexpansionofthebeamandeventuallytoitsbuckling.ThelocalizedthermalstressgeneratedbyJouleheatingplaysamajorroleleadingtotheresonantfrequencymodification.Ingeneral,whencompressivestressesaregenerated,thestructurebecomessofter;thereforethereisadecreaseintheresonantfrequency.Otherwise,ifthestressesaretensilestressesthenthestructurebecomesmorerigid,andtheresonantfrequencyincrease.Astaticanalysisusingthe“JouleheatingandtheThermalexpansion”modulesinComsolMultiphysicswasperformedforthetwothicknessesofthebeam(namely10μmand25μm).Figure4showsthetemperatureprofileinthebeamwhenacurrentof50mAisused.Duringthesimu-lation,thetemperatureisfixedatthebeamendstobeequal295K(∼22°C,ambient

6352|HadjSaidetal.20.2Structure10mm2019.819.619.419.2Frequency[kHz]1918.820406080100120(a)Current[mArms]40.9Structure25mm40.840.740.640.540.440.340.2Frequency[kHz]40.140253035404550(b)Current[mArms]Fig.3.Measuredresonantfrequencyvariationasafunctionofthesupplycurrentforthe(a)10μm-thickbeam(b)25μm-thickbeam.temperature)andincreasingto325K(∼52°C)inthemiddleforthe10μm-thickbeamandto313K(∼40°C)forthe25μmthickbeam.Theconvectiveheattransfercoefficientwasassumedtobeequalto100W/m2/K.Inthiscase,thebeamstructurebecomesmorecompliantcausingtheresonantfrequencyreduction[13].Figure3showsthecurveoftheresonantfrequencyshiftingasafunctionofthesupplycurrent.Twophenomenacanbedenoted,forthe25μm-thickbeamwecanseethereducingoftheresonantfrequencybecauseoftheinherentcompressivestress.However,forthe10μm-thickbeamwecanseethattheresonantfrequencyincreaseswhenthecurrentisincreasing.Itisbelievedthatinthiscase,thenon-linearvibrationeffectsbecomedominantastheLorentzforceincreaseswiththesupplycurrentinthebeam.Ontheotherhand,theSOIMUMPsfabricationprocessusesafinegoldlayeraddedontopofthesuspendedbeam.Duetothehighthermalexpansioncoefficientofgold,wecannoteagreaterstressinthelatterthaninthesiliconlayer.Indeed,thegoldlayertendstoexpandmorebutitscohesionwiththesiliconpreventsthat.Thisshould

64MEMS-basedclamped-clampedbeamresonatorcapacitive|53325.6Surface:Temperature(K)3253203153105003053000295Fig.4.Temperaturevariationinthe10μm-thickbeamduetotheJouleheatingeffect.explainalsowhythestressinthestructureof10μm-thickbeamishigherthanintheotherone.Duringtheexperimentforthe10μm-thickbeams,wenotethatwhenthecurrentincreasesthecurvatureofthebeam,eveninstaticstandbymode,largelyincreasesreaching3μm,knowingthatthethicknessofthegoldlayerisequalto520nm.Inthiscasethenonlinearvibrationispresentedduetothegoldlayerandmoredominantthanthethermalstresseffect.Converselyfor25μm-thickbeamsstructures,thiseffectisnotoutstandingsincethebeamisthickerandthetotaldisplacementwasinorderofnm.Toavoidthisproblem,the10μm-thickbeamswasactuatedusingasweepssignalbetween17kHzand22kHz,and22kHzand17kHzrespectivelyshowninFig.5.Ahysteresisisshownbetweenthetwospectrathatcanbeexplainedbythenonlineareffectofthebeam.Infact,Zooketal.[14]foundthatwhenthedrivevoltagesincrease,theresonantfrequencyshiftstoahighervalue,leadingtohysteresisdependingonthedirectionoffrequencyscan.Theoreticallyandduetothegoldlayer,somecorrectionadjustmentcanbeintroducedtoequation(4)resulting:wE2h2+E2h2+EhEh+4h2+6hh+4h2112211221122EI=(5)12(E1h1+E2h2)andρA=ρ1A1+ρ2A2(6)

6554|HadjSaidetal.351V17–22kHz1V22–17kHz30X:2.013e+004Y:2.604e–00825X:1.929e+00420Y:1.759e–00815Displacement[nm]10500510152025303540Frequency[kHz]Fig.5.Spectrumofthebeamwithsweepactuation.whereE1andh1arerespectivelytheYoung’smodulusandthethicknessofthesiliconlayer,andE2andh2aretheYoung’smodulusandtheheightofthegoldlayer,respectively.While,ρ1andA1arerespectivelythedensityandsectionofsilicon,ρ2andA2arethedensityandthesectionofthegoldlayer,respectively.Basedonthiscorrection,theresonantfrequencyresultsareshowninTab.3.Wecanconcludethesameresult,isthatthegoldlayerincreasesthedensityandinducedaresonantfrequencyhigher.4InterdigitatedcombscapacitanceevaluationAsperdesignthebeamismovingoutofplane,themagnetometerissensitivealongthez-axisdirection.Thetheoreticalcapacitanceexpressioninthisconfigurationisgivenby:l(t|z|)nεif|z|

66MEMS-basedclamped-clampedbeamresonatorcapacitive|5535Surface:Electricpotential(V)25umthickbeam10umthickbeam1301250.8200.6150.4Capacitance[fF]100.205–1.4288´10–170(a)(b)–40–30–20–10010203040Displacement[um]Fig.6.(a):TheelectrostaticpotentialdistributionbetweentwofingerssimulatedbyFEM.(b):Capacitancechangeasafunctionofthedisplacementbetweentwofingers.fingersbythetotalnumberoffingerspairsthroughouttheclamped-clampedbeam.Table3showsthedifferentresultsfromthissimulation.Tab.3.Resonancefrequencyandstatictotalcapacitancevalues,forthetwodifferentstructuresthick.10μm-thickstructure25μm-thickstructureAnalyticresultFEMresultAnalyticresultFEMresultFirstmodeFrequency(kHz)17.515.14842.9839.465Capacitance(pF)–2.57–5.315QualityfactorThequalityfactor,Q,isdefinedastheratiobetweentheenergystoredinthesystemandtheenergylostpercycle[15].Ifthestructurehasahighqualityfactorthenalowenergywillbeneededtokeeptheresonanceatconstantamplitudeandtheelectroniccircuitrywillhaveaminimumeffectontheresonantfrequency[16].Figure7presentsthedisplacementoftheclamped-clampedbeamactuatedat50mArmsasafunctionofthefrequencyobtainedbylaserDopplervibrometry.TheQ-factorcanbeeasily

6756|HadjSaidetal.–340–360–380–400–420–440Displacement[dB]–460–4800510152025303540Frequency[KHz]Fig.7.Displacementoftheclamped-clampedbeam(indB)vs.thefrequency.Tab.4.Valueofthequalityfactorderivedfromthedisplacementspectrumwithanexcitationcurrentof50mA.ResonanceWidebandwidthQualityfrequency(kHz)for−3dB(kHz)factor50mA19.2700.3850.71calculatedfromthiscurveusingthefollowing:f0Q=(8)∆fwheref0istheresonantfrequencyofthebeamleadingtothemaximumvibrationamplitude,and∆fisthedifferencebetweenf1andf2frequencies.Thesefrequenciesf1andf2correspondtovibrationamplitudesof−3dBlowerthanthemaximum(thatoff0).FromFig.7,weextractthevaluesneededtodeterminethequalityfactor;theseparametersareshownintheTab.4.ThemeasuredQfactorisaround50foroursensor(forthe10μm-thickstructure).6MEMSmagnetometerequivalentelectricmodelTheparametersoftheelectricmodelofthesensorsarededucedfrommeasurementsoftheinputandoutputimpedance.BoththesequantitiesaremeasuredusingtheAgilent4284ALCRmeteronafrequencyrangebetween20Hzand1MHz.Theinputimpedanceofthesensorisequivalenttoaresistanceformedbythemetallayerthatisdepositedontopofthebeam.Experimentally,thisimpedanceismeasuredbetween

68MEMS-basedclamped-clampedbeamresonatorcapacitive|57padsAandB(seeFig.1).InFig.8theinputimpedancein(Ω)andthecorrespondingphasein(°)asafunctionoffrequencyareplottedforthe10μm-thickstructure.Fromthesecurve,theresistanceoftheAulayerwasdeducedequalto5.62Ω.Theoutputimpedanceofourstructureisalsoimportantbecauseitisdefinedbythetotalcapacitanceofthefingerelectrodes.Experimentally,thisimpedanceismeasuredbetweenpadsAandC(seeFig.1).FromtheschematicinFig.9(modulusandphaseofoutputimpedance),wecandeducearesistor-capacitorserialcircuit,sowemeasuredirectlythesevalues(resistorandcapacitorserial)usingtheLCRmeter.ThevaluesthatwereobtainedfromthesemeasurementsatdifferentfrequenciesaresummarizedinTab.5.ThemagnetometerelectricmodelcanberepresentedasshowninFig.10;withRactrepresentingtheinputimpedance,andtheimpedanceCeq,ReqaretheequivalentoftheRC-dipolewhichwasderivedfromtheoutputimpedance.ThebeamisactuatedbetweentheAandBpadsandthecapacitivemeasurementiseffectuatedbetweentheAandCpadsorBandCpads.Finally,thetheoreticalsensitivityofoursensorscanbedefinedbythefollowingequation∆C∆C∆zS==×B(9)∆ix∆z∆ixItmaybenotedthatthesensitivityisdirectlyproportionaltothecurrentpassingthroughthebeam,howeverincreasingthiscurrentcausesashiftoftheresonantfrequency,whichpresentadrawbackforourresonantsensors.So,wecansolvethisproblembyusingatuningsystem(electrostatics,thermal,etc.)fortheresonantfrequencyorusingothermetallayerinsteadtothegoldiftheprocesspermitsit.Tocalculatethesensitivity,wecanextractitstwoexpressionconstituentpartsfromgraphs.Infact,thefirstpartcanbedeterminedfromFig.6.bandthesecondpartcanbeextractedfromFig.11,whichpresentsthedisplacementasafunctionoftheactuatingrmscurrent.FromFig.12,wecancalculatethetotalsensorspracticalsensitivitythatisaround10fF/mA,forthe10μm-thickstructure,and1fF/mAfortheotherone.ThesensorsensitivitydependsmainlyonthechosengeometricalparametersintheSOIMUMPsfabricationprocessandthequalityfactor.Thesensorperformancecanbeimprovedbyusingavacuumpackage.Indeed,invacuumthequalityfactorincreaseandtheresonantbeamwillgetlargeramplitudesbyreducingthedampingeffect.

6958|HadjSaidetal.Tab.5.Capacityandresistancevaluesobtainedfromtheoutputimpedancemeasurementsatdifferentfrequenciesforthe10μm-thickstructure.Frequencies(kHz)0.511001000Capacity(pF)6.806.766.726.61Resistance(MΩ)267.55223.3132.01596.97040302010Impedance[Ω]0123456101010101010Frequency[KHz]806040Phase[deg]200123456101010101010Frequency[KHz]Fig.8.Experimentalmeasurementoftheinputimpedance.40302010Impedance[Ω]0123456101010101010Frequency[KHz]806040Phase[deg]200123456101010101010Frequency[KHz]Fig.9.Experimentalmeasurementoftheoutputimpedance.

70MEMS-basedclamped-clampedbeamresonatorcapacitive|59R_act/2R_act/2C_eqABR_eqCFig.10.Electricmodelofthesensor.4803.57032.5602501.5displacement[um]1displacement[um]400.503001020304050253035404550(a)current[mArms](b)current[mArms]Fig.11.Measureddisplacementofthebeamvariationasafunctionofthesupplycurrentfor:(a)10μm-thickbeam,and(b)25μm-thickbeam.2.62.552.52.452.42.352.3Capacitance[pF]2.252.22.152.101020304050Current[mArms]Fig.12.simulatedcapacitanceasfunctionofactuationcurrentforthe10μm-thickbeam.7ConclusionInthispaperaMEMS-basedmagnetometerwithcapacitivedetectionwaspresentedfoundedonanout-of-planeLorentzforceexcitation.Wehavesimulatedthefirst

7160|HadjSaidetal.vibrationmodeoftheclamped-clampedbeamandithasbeencomparedwiththemeasuredvalue.Thesensorresonancefrequencywasfoundedaround15.146kHzwithComsolforthe10μm-thickbeamand39.465KHzforthe25μm-thickone.Wenotedadifferencebetweenthesimulatedandmeasuredvaluewhenthebeamwasactuatedbyanalternativecurrent.ThisdifferenceisaresultoftheJouleheatingeffectforthe25μm-thickbeam,asitgeneratesacompressivestress,andthenon-lineareffect,mostlycausedbythepresenceofthegoldlayer,forthe10μm-thickbeam.Wenoticedalsoadifferencebetweenthesimulatedandthemeasuredcapacitance,whichwasabout2.57pF(themeasuredwasabout6.7pF)forthe10μm-thickstructure.Thisdifferenceisduetotheeffectoftheelectrodesandpadeffect(seeFig.1)whichhasnotbeenconsideredinthesimulatedmodelsincewemodeledjustonepairoffingers,andwemultiplythecapacitancefoundbythetotalnumberoffingerspairsthroughouttheclamped-clampedbeam.Themeasuredqualityfactorwasabout50,andtheelectricequivalentmodelofthe10μm-thicksensorisdeducedfromthemeasurementoftheinputandoutputimpedanceusingtheLCRmeter.Itwasfoundthattheinputimpedanceisasimpleresistanceequalto5.62ΩandtheoutputresistanceisformedbytheRCserialcircuitswithRwasabout223.313MΩandCwasabout6.7pF.Finally,thesensorsensitivityisfoundtobeabout10fF/mA(forthe10μm-thick)and1fF/mA(forthe25μm-thick).Thesensitivityissignificantlyaffectedbythecurrentpassingthroughthebeamthatcausesresonantfrequencyshifts.ThisproblemcanbesolvedbyusingtuningelectrostaticssystemsbyapplyingaDCvoltagebetweenfingerstoadjusttheresonantfrequency.Workstoovercomethecharacterizationproblemsencounteredduringourfirstdesign;wenotetheshiftingoftheresonancefrequencyandthequalityfactorreductionareongoing,andfurtheroptimizedstructuresarealsounderstudyasaddingspringontheendsofthebeam.Acknowledgements:TheauthorswouldliketothankMr.PierreCourtoisforhavingdesignedandfabricatedthesensorintheSOIMUMPsprocessflow.Bibliography[1]S.Kordic.Integratedsiliconmagnetic-fieldsensors.SensorsandActuators,10:347–378,1986.[2]R.S.Popovic,J.A.FlanaganandP.A.Besse.Thefutureofmagneticsensors.SensorsActuatorsA,56:39–55,1996.[3]R.Gottfried-Gottfried,W.Budde,R.Jahne,B.Sauer,S.UlbichtandU.Wende.Aminiaturizedmagnetic-fieldsensorsystemconsistingofaplanarfluxgatesensorandaCMOSreadoutcircuitry.SensorsandActuators,54:443–447,1996.[4]M.J.Caruso.Applicationsofmagnetoresistivesensorsinnavigationsystems.SensorsandActuators,SAESP-1220:15–2,1997.[5]P.Losantos,C.Cané,D.FlandreandJ.-P.Eggermont.Magnetic-fieldsensorbasedonathin-filmSOItransistor.SensorsandActuatorsA,67:96–101,1998.[6]A.L.Herrera-May,L.A.Aguilera-Cortés,P.J.García-RamírezandE.Manjarrez.ResonantmagneticfieldsensorsbasedonMEMStechnology.J.ofSensors,39:7785–7813,2009.

72MEMS-basedclamped-clampedbeamresonatorcapacitive|61[7]A.ArfanandD.K.Potter.AnewcontactlesstrackballdesignusingHalleffectsensors.SensorsandActuatorsA,147:110–114,April2008.[8]W.Hernandez.ImprovingtheResponseofaRolloverSensorPlacedinaCarunderPerformanceTestsbyUsingaRLSLatticeAlgorithm.SensorsJournal,MolecularDiversityPreservation,Int.(MDPI),5:613–632,2005.[9]F.AyaziandK.Najafi.DesignandFabricationofAHighPerformancePolysiliconVibratingRingGyroscope.11thIEEE/ASMEInt.WorkshoponMicroElectroMechanicalSystems,Heidelberg,Germany,January,1998.[10]S.D.Senturia.PerspectivesonMEMS,PastandFuture:TheTortuousPathwayFromBrightIdeastoReal.12thIEEEInt.Conf.onSolidStateSensors,ActuatorsandMicrosystems,Boston,June,2003.[11]B.Ziaie,T.W.Wu,N.Kocaman,K.NajafiandD.J.Anderson.AnImplantablePressureSensorCuffForTonometricBloodPressureMeasurement.TechnicalDigest,Solid-State,SensorandActuatorWorkshop,June,1998.[12]W.Weaver,S.P.TimoshenkoandD.H.Young.VibrationProblemsinEngineering.5thEdition,WileyPublishers,1990.[13]T.RemtemaandL.Lin.Activefrequencytuningformicroresonatorbylocalizedthermalstressingeffects.SensorsandActuators,91:326–332,2001.[14]J.D.Zook,D.W.Burns,H.Guckel,J.J.Sniegowski,R.L.EngelstadandZ.Feng.Resonantmicrobeamstraintransducers.6thInt.Conf.Solid-StateSensorsActuators(Transducers’91),:529–532,SanFrancisco,CA,June1991.[15]F.Ahmad,J.OjurDennis,N.H.Hamid,M.HarisandM.Khir.DesignandSimulationofMechanicalBehaviorofMEMS-basedResonantMagneticFieldSensorwithPiezoresistiveoutput.Int.Conf.onMechanicalandElectricalTechnology,September,2010.[16]M.ElwenspoekandR.J.Wiegerink.MechanicalMicrosensors.Springer-Verlag:Berlin,Heidelberg,Germany,2001.BiographiesMohamedHadjSaidwasborninMonastir,Tunisiain1987.HereceivedtheB.Sc.’10andMSc’12degreesbothfromtheHigherInstituteoftheinformaticsandmathematicsofMonastir(ISIMM,Tunisia).HisMasterisdoneincollaborationwiththelaboratoryofSensors,MicrosystemsandActuatorsofLouvain-la-Neuve(SMALL),Belgium.HeiscurrentlyaPhDstudent,intheNationalEngineeringSchoolofSfax(ENIS),workingonthenoveldesign,modelingandsimulationofaMEMSmicrophonebasedinmagneticdetectioninCMOSTechnologyincollaborationwithTIMALaboratoryFrance.Furthermore,heinterestsinmagneticsensors,surfaceacousticwavemodeling.FaresTounsireceivedtheBSc’01andMSc’03degreesfromtheNationalEngineeringSchoolofSfax(ENIS)inTunisia,andthePhD’10inMicroandNano-electronicsfromGrenobleInstituteofTechnology(INPG),France.HeiscurrentlyanassistantprofessorintheInstitutSupérieurd’InformatiqueetdeMathématiquesinMonastir(ISIMM),Tunisia.Heiscurrentlyworkingonthedesign,modelingandsimulationofnewCMOScompatiblemicrosystemsmicromachinedsensors.Specifically,heisinterestedinnoveldesignofmicrophone,accelerometersandRFswitches.Inaddition,heisnowfocusedonthefieldofnanotransducers,evaluationofnewmaterials/structuresforMEMSandadvancedmicrosystems.

7362|HadjSaidetal.PetrosGkotsiswasborninAthens,GreeceandreceivedhisdegreeinPhysicsfromtheNationalandKapodistrianUniversityofAthensin2002.In2010afterthePhDdegreeawardfromCranfieldUniversity,EnglandhejoinedUniversitécatholiqueLouvainatBelgiumasapost-doctoralresearchassistantinvestigatingthereliabilityofMEMSinharshenvironments.Hiscurrentresearchinterestsincludethestudyoffunctionalmaterials,thin-filmcharacterization,designandmodellingofmicrosystems,effectsofradiationandextremetemperaturesonthereliabilityofmicroelectromechanicalsystemsandthedevelopmentofprocessestoincorporatenewmaterialsinmicrofabrication.MohamedMasmoudiwasborninSfax,Tunisiain1961.HereceivedtheB.Sc.’85fromNationalEngineeringSchoolofSfax(ENIS)andthePh.D.’89inMicroelectronicsfromtheLaboratoryofComputerSciences,RoboticsandMicroelectronicsofMontpellier,France.From1989to1994,hewasholdinganAssociateProfessorpositionatNationalEngineeringSchoolofMonastir,Tunisia.Since1995,hehasbeenaProfessoratENISwherehehasbeenengagedindevelopingMicroelectronicsintheengineeringprogramoftheuniversity,andwhereheisalsotheheadoftheresearchgroupElectronics,MicrotechnologyandCommunication(EMC).HeistheauthorandcoauthorofseveralpapersintheMicroelectronicfield.Hehasbeenareviewerforseveraljournals.Prof.MasmoudiorganizedseveralinternationalConferencesandhasservedonseveraltechnicalprogramcommittees.LaurentA.Francis(MEng’01,PhD’06)holdstheMicrosystemsChairpositionattheUniversitécatholiquedeLouvain(UCL,Belgium)asAssociateProfessorandistheleadingmemberoftheSensors,MicrosystemsandActuatorsLaboratoryofLouvain(SMALL).Hisscientificinterestsarerelatedtoultra-lowpowermicrosensorsforbiomedicalapplicationsoroperationinharshenvironments,onbio-inspiredapproachesandonmicro-andnano-fabricationtechnologiesincludinginparticularatomiclayerdeposition.HisPhDthesiswasrelatedtoacoustic/opticalbiosensorsatIMEC,theInteruniversityMicroElectronicsCenterlocatedinLeuven,Belgium.Between2000and2007hewaswithimecasresearcher,successivelyintheBiosensorsandRF-MEMSgroups.In2011,hewasavisitingprofessorattheUniversitédeSherbrooke,Canada.HeisregularmemberoftheIEEE.Hehasauthoredorco-authoredmorethansixtypublicationsininternationalscientificjournals,conferences,andbookchaptersandholdsonepatent.

74G.U.GammandL.M.ReindlRangeExtensionforSingleHopWirelessSensorNetworkswithWake-upReceiversAbstract:WirelessWake-upreceiversareusedwheneveralonglifetimeandaperma-nentaccessibilityisrequired.Themaindisadvantagetheyhaveistheirshortrangeofcommunicationofafewmeters.Wethereforepresentinthisworkaninfrastructureforasinglehopnetworkthatmakesuseofapowerfulmainnodethatcansendoutwake-upsignalswithupto+20dBm.Theendpointnodeshaveanintegratedwake-upreceiverandconsume3.5μAofcurrent.Thewake-uprangewasmeasuredupto90metersinanopenairfield.Keywords:Wake-Up,WSN,LowPower.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A221IntroductionAwirelessinfrastructurewhereahostnodecommunicateswithanumberofclientnodescanbefoundindifferentscenarioslikeasmartmeteringapplicationorthesurveillanceofsoilparametersinfarming.Oftentheinformationgatheredbytheclientnodesisneededonlyonceadayorevenonlyonceaweek.Ifthepointoftimeisnotknownwhenthehostdemandsthedatafromtheclients,theclientsneedtohavearadiolisteningconstantlytoachannel.Anormallowpowerreceiverconsumesabout15mAofcurrentinreceivemode.WhentheclientispoweredbyastandardCR2032coincellwith230mAhthebatterywouldbedepletedinlessthanoneday.Inthisworkweuseclientnodesthathaveanintegratedwake-upreceiverandconsumeonly3.5μAofcurrent.WithanormalCR2032coincellatheoreticalstandbytimeofmorethan8yearsisachieved.Weachievethelowpowerconsumptioninreceivemodebyusinga125kHzreceiver.ToavoidlargeantennasthatcannotbeintegratedonasmallPCBwemodulatethe125kHzwake-upsignalinthesenderona868MHzcarrierusingOOK(OnOffKeying).InthereceivingclientswedemodulatetheincomingsignalusingaSchottkyDiodefollowedbyalowpassfilter.Onlytheenvelopesignalisthenpassedtothe125kHzreceiver.Thiswaywecombinethelowcurrentconsumptionofreceiversworkingatlowfrequencieswiththesmallantennasizeofreceiversworkingathigherfrequencies.Inmostnetworktopologiesthehostnodecomeswithawiredpowersourceoratleasthasaneasychangeablebattery.WethereforeincorporatedinthehostnodeanRF-frontendthatenablesanoutputpowerofupto+20dBm.ThisG.U.GammandL.M.Reindl:LaboratoryforElectricalInstrumentation,DepartmentofMicrosystemsEngineering-IMTEK,Freiburg,Germany,email:gerd.ulrich.gamm@imtek.uni-freiburg.de.DeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.63–74.DOI10.1515/9783110470444-005

7564|G.U.GammandL.M.Reindlwaythewake-upsignalcanpenetratewallsandcanwake-upsleepingclientsinsidebuildingswhichisespeciallyimportantforsmartmeteringapplications.Theremainderofthispaperisstructuredasfollows.Section2givesanoverviewonrelatedwork.Section3presentstherealizednetworktopology.Section4introducestheclientnodeandshowssomemeasurements.Section5introducesthehostnodeandgivessomemeasurements.Thefinalconclusionandafurtheroutlookwillbegiveninsection6.2RelatedWorkIn[1]Ansarietal.presentaTelosBsensornodeoperatinginthe868MHzISMbandwithincludedaddressdecoding.Decodingisdonebyaseparatemicrocontroller.Theachievedwake-uprangeisupto10meters.In[2]LunCuifenetal.presentaZigbeebasedsystemforreadingoutmeterswireless.TheelectronicattachedtothemetersconsistsofaCC2430SystemonChipwithintegratesZigBeefunctionality.Thesystemisespeciallydesignedforruralareas.Theauthorsconnecttheirdevicestothestandardpowernetwork.Thereforeanadditionalpowersupplyisneededforthedevicewhichincreasesthecosts.AnotherZigBeebasedsystemispresentedin[3].BaodingZhangandJialeiLiupresentasystemwhereasocalleddataconcentratorcollectsdataviaZigbeeandpassthemviaGPRStoagateway.Thepaperstatesthatthepowerconsumptionisoneofthemainproblems,butthereisnosolutiongiven.In[4]Gammetal.giveamoredetaileddescriptionofthewake-upreceivercircuitaswellasadditionalmeasurements.Otherexamplesofwake-upreceivercircuitscanbefoundbyGuetal.in[5]andbyVanderDoornetal.in[6].In[7]ThomasWendtandLeonhardReindldescribedifferentmethodstowake-upasleepingsensornode.OneofthepresentedmethodsisaFrequencyDiversityWake-UpschemeinwhichalsoanAtmelwake-upreceiveroperatingat125kHzisused.Wendtcalculatesthelifespanofasensornodewithsuchawake-upreceiverupto4–5years.3NetworkTopologyWeproposeastarnetworktopology.Thecentralnodeofthenetworkisthehostnodewhichcansendoutcodedwake-upsignalstotriggerselectivenodesfromtheirsleeptoactivemode.Everyclientnodethatwakes-upautomaticallytransmitsitscollectedandstoreddatatothehostnode.Incaseofasmartmeteringapplicationtheclientnodes

76Singlehopwirelesssensornetworkswithwake-upreceivers|65wouldbeinstalledclosetothemeteringdevice.Thehostnodecouldbeahandhelddevicethatisusedbyapersontocollectthemeteringdatawhiledrivingbyinacar.Fig.1showsadiagramoftherealizednetworktopology.Wake-upClient-NodeDataWake-upHost-NodeDataClient-NodeFig.1.Diagramoftherealizednetworktopologie.ImpedanceRectifierLowpassMatchingAntennaControlSwitchWake-UpMainMicro-Wake-upTransceivercontrollerReceiverFig.2.Blockdiagramofthesensornode.Insleepmodeallincomingsignalsareroutedtotheupperpath.AfterdemodulationandlowpassfilteringtheAS3932chipdetectsavalidwake-upsignalandswitchestheclienttoactivemode.4ClientNode4.1ClientHardwareDesignThemaincomponentoftheclientnodeistheCC430F5137[8]SystemonChipfromTexasInstruments.ItincludesalowpowermicrocontrollercoretogetherwithanintegratedradiotransceiveroftypeCC1101inonesingledie.ComingfromtheSMAjacktowardsthecenteroftheboardfollowstheADG918[9]antennaswitchthatisusedforselectingthewake-uppathortheconnectiontotheradiooftheCC430controller.Ontopoftheantennaswitchfollowsthedemodulationcircuitconsistingofimpedance

7766|G.U.GammandL.M.Reindlmatching,SchottkyHSMS285Cdiodeandlowpassfilter.TheAS3932[10]isawake-upreceiverworkingat125kHzwithincludedaddresscorrelation.Whenreceivingavalidwake-upsignalittriggersthemicrocontrollerfromsleeptoactivemode.Fig.2showsablockdiagramoftheclientnode.FortheclientnodeafourlayerdesignonFR4substratewascreated.Additionalcomponentsarepushbuttons,JTAGprogramminginterface,LEDsanda26MHzcrystal.TheimpedancematchingofthemainradiowasdoneusingachipbalunfromJohansonTechnologies.ThemainpowersupplyisaCR2032coincell.Thistypeofbatteryissmallandveryeconomic.Thebacksideoftheboardaccommodatestheblockingcapacitors.AphotooftheclientnodecanbeseeninFig.3.CC430F5137AS3932ADG918DemodulationFig.3.PhotoofaclientnodemanufacturedonafourlayerFR4substrate.4.2ClientSoftwareInnormaloperatingmodethenodewillbeinthesleepmode.Allperipheralsandcomponentsareswitchedofftosaveenergyexceptthe125kHzwake-upreceiver.Theantennaswitchisinsuchposition,thatincomingsignalwillberoutedtothedemodu-lationandlowpasscircuit.Ifthewake-upreceiverchipdetectsavalidwake-upsignalitwillinterruptthemicrocontrollerfromitssleepmode.Themicrocontrollerthenreadsoutthemeteringdata,turnsonthemaintransceiverandtogglestheantennaswitchsothatthemaintransceiverisconnectedwiththeantenna.Thenthemetering

78Singlehopwirelesssensornetworkswithwake-upreceivers|67dataissentoutusingGFSKmodulationschemewithincludedCRCcheck.TheflowchartofthenodesoperatingsystemcanbeseeninFig.4.4.3ClientMeasurementsCurrentconsumptionmeasurementofthecompletenodeinsleepingmodewasdoneusingaKeithley6514Systemelectrometer.Thecurrentwasmeasuredto3.5μAofwhich2.8μAconsumesthewake-upreceiverchipandabout0.7μAtheCC430controllerinlowpowermode4.Whenusinga230mAhcoincellwecancalculatethemaximumlifetimeoftheclientnodeinsleepmodebytheequation1.StartTurnonmainradiopermanentSenddataIncomingSignal?NoYesToggleAntennaSwitchInterruptμCμCtosleepmodeToggleAntennaSwitchEndFig.4.Flowchartofthenodesoperatingsystem.230mAhTmaxidle==65714h(1)3.5μAThisequalsto7.5yearsofstandbytimeinwhichthenodecanalwaysbewokenupbyreceivingavalidwake-upsignal.WecandefineanintervalTintinwhichtheclientnodewakes-uponceandsendsbackthestoreddata.ThelengthofthisintervalcanbeseeninFig.5onthex-axis.Onthey-axistheresultinglifespanofthenodeisplotted.Theoveralllifetimedoesnotchangesignificantlyduetothecurrentneededtosend

7968|G.U.GammandL.M.Reindlbackthestoreddata.Inalongtermscenariothemajorityofthepowerisconsumedbythecurrentinsleepmode.876CR-2032(230mAh)5(years)node43LifespanT21001002003004005006007008009001000TimeIntervalTint(s)Fig.5.Onthex-axistheintervalTintisplotted.Itrepresentsthespanoftimeinwhichthesleepingnodeiswoken-uponceandtransmitsitsstoreddatabacktothehost.Theresultinglifespanforeachintervalisplottedonthey-axis.5HostNode5.1HostHardwareDesignThehostnodeiscontrolledbyanCC430F5137SystemonChipthatcombinesmicro-controllerandtransceiverICinonepackage.ForimpedancematchingachipbalunfromJohansonTechnologiesisused.TosuppresssporadicemissionsatunwantedfrequenciesaB3716SAWfromEpcoshasbeeninsertedintheantennapath.Sincethewake-upsignalhastopassconcretewallstoreachthesleepingclientnodesahighsendingpowerisofadvantage.WethereforeincludedtheCC1190RFfrontendfromTexasInstruments[11]onthehostnode.Itincreasestheoutputpowerupto+20dBm.FortransferringthereceiveddatatothePCaFT232UARTtoUSBchipisused.ThehostnodecanbepoweredusingfourAAbatteriesorusingtheUSBconnection.Ajumperselectsthedifferentpowersupply.ToguaranteeastableoperationaMIC5318dropoutregulatorisused.Furthercomponentsareapushbutton,threeLEDs,anOn/OffSwitchandaJTAGprogramminginterface.InFig.6ablockdiagramofthehostnodecanbeseen.Fig.7showsaphotooftheassembledhostnode.

80Singlehopwirelesssensornetworkswithwake-upreceivers|69QuartzUSB-to-UARTMicrocontroller+RFChipBalunSAW-FilterTransceiverFrontendVoltageRegulatorLEDsFig.6.Blockdiagramofthehostnode.FT232R-USBCC430F5137MIC5318B3716SAWCC1190Fig.7.Fotooftheassembledhostnode.5.2HostsoftwareThesoftwareforthehostnodestartswithaninitializationroutine.Afterconfiguringthetransceiveramainstatemachineisstarted.WhenreceivingastartcommandviaUARTorwhenthepushbuttonispressedthenodesendsoutacodedwake-upsignalforclientnode1.Afterwardsthetransceiverisswitchedtoreceivemodetolistenfortheincomingdataofnode1.Thenawake-uppacketfornode2isbuiltandtheprocess

8170|G.U.GammandL.M.Reindlisrepeated.ThereceiveddataispassedviaUARTtoaconnectedcomputerwhereitcanbestoredandvisualized.AflowchartofthesoftwarecanbeseeninFig.85.2.1Wake-UpsignalbuildupThewholewake-upsignalisbuiltbymodulatinga125kHzsquarewavesignalonan868MHzcarrierusingOOK(OnOffKeying).Anadditionaladdressinformationforselectivewake-upofnodescanbemodulatedonthe125kHzsignalbyagainusingOOK.ThemodulationstepsareshowninFig.9.TheOOKmodulationofthe125kHzsquaresignalonthe868MHzcarrierintheCC430F5317isdonebysettingthedatarateto250kbit/sandbysendingoutconstantlythedatabyte0xAA.Thisresultsinasubsequentlyturnedonandoffcarriersignal.Theresultingenvelopethenrepresentsthe125kHzsquaresignal.Figure10showsthedatabyte0xAAandtheresultingcarriersignalwhensentoutwith250kbit/s.StartWakeNode2IdleListenforData2••Interrupt•NoYesSendDataviaUARTWakeNode1IdleListenforData1EndFig.8.Flowchartofthehostsoperatingsystem.

82Singlehopwirelesssensornetworkswithwake-upreceivers|71125kHzPeriod868MHzCarrierModulationDataModulation1010101101001Fig.9.Modulationstepsofthewake-upsignal.0xAA10101010868MHzCarrier4ms=>250kBit/s8ms=>125kBit/sFig.10.Sendinga0xAAbytewithadatarateof250kbit/sresultsinthemodulationofa125kHzsquarewavesignalonthe868MHzcarrier.5.3HostMeasurementsTohaveanideaoftheamountoffrequenciesthatareinvolvedinmodulatingthe125kHzwake-upsignalonthe868MHzcarrierameasurementwithanRhodeandSchwarzZVLSpectrumAnalyzerwasdone.Thehostnodewasprogrammedtosendthewake-upsignalinaninfiniteloopwithoutanydelayinbetween.Theantennasignalwasfeddirectlytothespectrumanalyzer.Forsafetyreasonsa10dBmattenuatorwasinserted.TheresultingspectrumcanbeseeninFig.11.Duetotheattenuatortherealamplitudevaluesare10dBmhigherthanshowninthefigure.

8372|G.U.GammandL.M.ReindlAmpindBm100–10–20–30finMHz840860880900Fig.11.Spectrumofthewake-upsignal.Theamplitudevaluesseenonthey-axisaremeasuredwiththeattenuatorinline.Therealvaluesare10dBmhigher.5.4Wake-upDistanceMeasurementTodeterminethemaximumrangethathostandclientcanbeseparatedadistancemeasurementhasbeendone.Thehostnodewasprogrammedtosendawake-upsignalwithincludedaddressinformationinaninfiniteloop.Thesleepingclientnodewasprogrammedtowake-upatthesameaddress.Bothhostandclientnodesweremountedtopolesofonemeterheight.Thehostnodehadafixedpositionwhereastheclientnodewasmovedsuccessivelyaway.SwitchingfromsleeptoactivemodewasindicatedintheclientnodebyturningonandoffanLED.Themaximumrangeweachievedbyacceptingapacketerrorrate(PER)of5%andasendingpowerofRangeinmFriis–Formula80604020PoutindBm0–1001020Fig.12.Wake-uprangeasafunctionofoutputpowerusingFriistransmissionequation.

84Singlehopwirelesssensornetworkswithwake-upreceivers|73+20dBminthehostwhere90metersinanopenfreefield.Thetheoreticalpossiblewake-uprangecanbecalclulatedusingFriistransmissionequation.Solvedforritlooksasfollows:cPSGSGEr=(2)4πfPEFigure12showsaplotoftheoutputpowerofthesenderonthex-axisandtheresultingwake-uprangeonthey-axis.Themeasurementpointisrepresentedbyabluedotandisinthetheoreticalexpectedrange.6ConclusionandOutlookInthisworkwepresentedawirelessinfrastructureconsistingofonehostnodeandthreeclientnodeswithincludedwake-upreceivers.Whileinsleepmodetheclientsconsume3.5μAofcurrentbutcanbewokenupbyacorrectwake-upsignal.Thewake-upsignalitselfconsistsofan125kHzsignalmodulatedonan868MHzcarrierthatissendoutbythehostnodewithamaximumoutputpowerof+20dBm.Whenwokenup,theclientnodesswitchtotheirmainradiotransceiverandsendouttheircollecteddata.Thehostnodecanbepoweredandcontrolledbyanconnectednotebookwhichmakesthesystemusableformobileapplications.Furtherworkcanbedonetoincreasethesensitivityofthesleepingnodessincethemainradiosrangeisstillsignificanthigherthanthewake-uprange.Acknowledgment:ThisworkhaspartlybeensupportedbytheGermanResearchFoundation(DFG)withintheResearchTrainingGroup1103(EmbeddedMicrosys-tems).Bibliography[1]J.Ansari,D.PankinandP.Mahonen.Radio-triggeredwake-upswithaddressingcapabilitiesforextremelylowpowersensornetworkapplications.IEEEInt.Symp.onPersonal,IndoorandMobileRadioCommunications(PIMRC),:1–5,September2008.[2]L.Cuifen,Z.Xiaoqin,L.YanpingandL.Ce.Theelectricmeterreadingsysteminruralareasbasedonwirelessmicro-computer.Int.ComputerDesignandApplications(ICCDA)Conf,vol.1,2010.[3]B.ZhangandJ.Liu.Akindofdesignschemaofwirelesssmartwatermeterreadingsystembasedonzigbeetechnology.Int.E-ProductE-ServiceandE-Entertainment(ICEEE)Conf,:1–4,2010.[4]G.U.Gamm,M.Sippel,M.KosticandL.M.Reindl.Lowpowerwake-upreceiverforwirelesssensornodes.6thInt.IntelligentSensors,SensorNetworksandInformationProcessing(ISSNIP)Conf.,:121–126,2010.

8574|G.U.GammandL.M.Reindl[5]L.GuandJ.Stankovic.Radio-triggeredwake-upforwirelesssensornetworks.RealTimeSystems,vol.29,no.2,pp.157–182,2005.[6]B.vanderDoorn,W.KavelaarsandK.Langendoen.Aprototypelow-costwakeupradioforthe868mhzband.Int.J.SensorNetworks,5:22–32,2009.[7]T.WendtandL.Reindl.Wake-upmethodstoextendbatterylifetimeofwirelesssensornodes.Conf.onInstrumentationandMeasurementTechnology(IMTC),:1407–1412,2008.[8]CC430F513xMSP430SoCwithRFCore,TexasInstruments,sLAS554E,:43,2009.[9]ADG918/ADG919,AnalogDevices,2008.[10]AS39323DLowFrequencyWakeupReceiver,austriamicrosystemsAG,revision1.2,2009.[11]CC1190850–950MHzRFFrontEnd,TexasInstruments,sWRS089A,:2,2009.BiographiesG.U.GammdidhisDiplomainElectricalEngineeringandInformationTechnologyattheKarlsruheInstitutofTechnology(KIT),Karlsruhe,Germany.CurrentlyheisdoinghisPhDinMicrosystemsEngineeringattheInstitutofMicrosystemsTechnology(IMTEK)inFreiburg,Germany.HeismemberofthePhDprogram“EmbeddedMicrosystems”andhisworkisconcernedwithwake-upreceiversforembeddedmicrosystems.L.M.ReindlisheadoftheLaboratoryforElectricalInstrumentationattheDepartmentofMicrosystemsEngineering,oftheUniversityofFreiburgsinceMay2003.HereceivedaDipl.Phys.degreefromtheTechnicalUniversityofMunich,Germanyin1985andaDr.Sc.Techn.degreefromtheUniversityofTechnologyVienna,Austriain1997.From1999to2003,hewasauniversitylecturerforcommunicationandmicrowavetechniquesattheInstituteofElectricalInformationTechnology,ClausthalUniversityofTechnology.Hisresearchinterestsincludewirelesssensorandidentificationsystems,andsurfaceacousticwavedevices.

86A.Ghorbel,M.Jallouli,L.AmouriandN.BenAmorAHW/SWImplementationonFPGAofAbsoluteRobotLocalizationUsingWebcamDataAbstract:ThispaperpresentsanimplementationofabsoluterobotlocalizationalgorithmusingFPGAtechnology.Theadoptedlocalizationmethoduseswebcamtrackingimages.Thistechniquehasbeendevelopedandimplementedforthemotionoftherobotfromaninitialpositiontowardsanotherdesiredposition.Firstly,wehavevalidatedtheproposedapproachonPCplatformusingClanguageandOpenCVlibrary.Secondly,tofacilitaterobotautonomousnavigation,amixedHW/SWimplementationwasbeendevelopedusingahighperformanceversionoftheNIOSprocessorcoupledwithacustomhardwareaccelerator.ExperimentaltestsonAlteraCycloneIIIFPGAStarterKitshowanimprovementof85%overpurelysoftwareexecutionwhichprovestheeffectivenessofthisproposedarchitecture.Keywords:HW/SWimplementation,Autonomousnavigation,Robotlocalization,AlteraFPGA,Webcamdata,Imageprocessing.MathematicsSubjectClassification2010:65C05,62M20,93E11,62F15,86A221IntroductionRobotnavigationisaquicklydevelopingareainthescienceofrobotics.Therapidtechnologyprogressofsensorandimagegivesnewopportunitiesforautonomousrobotnavigation.Mobilerobotsareparticularlybeingusedasasubstituteforhumansortodosimpleworkthatiseitherinclosedenvironmentoroutside.Itisbecomingincreasinglyimportanttobeabletoaccuratelydeterminethepositionofarobotinitsenvironment,aswellastomanagealltherelatedelectronic,mechanicalandsoftwareissues.Tomaketrulyautonomousrobots,oneoftheissuestoresolveisrobotlocalization.Forselflocalization,arobothasaccesstoabsoluteandrelativemeasurementsgivingitfeedbackaboutitsdrivingactionsandthesituationoftheenvironmentaroundit.A.Ghorbel,M.Jallouli,N.BenAmor:UniversityofSfax,NationalEngineeringSchoolofSfax,ResearchLaboratoryonComputerandEmbeddedSystems(CES),Sfax,Tunisia,emails:ghorbel.agnes@gmail.com,mohamed.jallouli@enis.rnu.tn,nader.benamor@enis.rnu.tn.L.Amouri:UniversityofSfax,NationalEngineeringSchoolofSfax,ResearchLaboratoryonControlandEnergyManagement(CEM),Sfax,Tunisia,emails:lobnaamourijmail@yahoo.fr,lobna.amouri@isecs.rnu.tn.DeGruyterOldenbourg,ASSD–AdvancesinSystems,SignalsandDevices,Volume2,2017,pp.75–92.DOI10.1515/9783110470444-006

8776|A.Ghorbeletal.2StateoftheartRelativelocalizationconsistsinevaluatingthepositionandorientationusingonlyinertialsensordata.Thesedatacanbedisplacementinformation(odometer),speedinformation(velocity)oraccelerationinformation(accelerometer).Nevertheless,itcannotbeusedespeciallyforlongandwindingtrajectories[1],sinceitsuffersfromseveraldrawbacks.Thoughthetechniqueissimple,itispronetoerrorduetoimprecisioninmodeling,noise,driftandslip[2].Sincethepositionestimationisbasedonearlierpositions,theerrorintheestimatesincreasesovertime.Absolutelocalizationprovidespositionmeasurementsbasedonobservationsmadefromtheenvironment.Thispositioninformationisindependentofpreviouspositionestimates[3].Thelocationisnotderivedfromintegratingasequenceofsuccessivemeasurements,butdirectlyfromonemeasurement.Unlikerelativeposition,theerrorinthepositiondoesnotgrowunbounded[3].Themajordisad-vantageofabsolutemeasurementsistheirdependenceonthecharacteristicsoftheenvironment.Inordertocompensatethedrawbacksofthetwotechniques,substantialim-provementisprovidedbyapplyingKalmanFilteringtechniques[4].Thesefilterscanestimatestatesofnoisysystemsinnoisyenvironment.Anotherapproach,presentedin[5,6],adaptsthepositionandtheorientationofamobilerobotthroughaweightedExtendedKalmanFilter(EKF).Thesemethodsrequiretooheavycomputingcapacityforamobilerobottoperformatask.Otherdisadvantagesareeithertheshortrangeofusedsensorslikeinfraredsensorsorthenecessitytoknowtheinitialpositionoftherobot.Othersolution,presentedin[7],usesamethodwhichpermitsthevehicletocorrectitsdriftbydirectobservationusingauniqueembeddedCCDcameraonamobilerobot.In[8],anotherlocalizationtechniqueispresented.Itusesthecorrespondencebetweenacurrentlocalmapandtheglobalmappreviouslystoreinmemory.In[9],amethodcalculatesthepositionoftherobotinordertointerceptamovingtargetthroughvisualfeedback.Themostsignificantdisadvantageofthesemethodsisthenecessitytoknowthestartingpositionofthemobile.Inrecentyears,themostadvanceddigitaltechnologieswereintroducedintoroboticcontrolapplications:DSPandFPGA.In[10],amulti-DSPplatform,basedonTIsDSPsfromtheC2000/C5000familiesisusedformotioncontrolanddataprocessingonthemobilerobotF.A.A.K.Thistechnologyisparticularlyefficientforimplementationofcomplexlocalizationtechnique.TheDSPinternalparallelstructureissuitableforimageprocessingalgorithm.TheemergenceofcomplexandlowpowerFPGAoffersaflexibleandrealisticapproachtodesignhighperformanceinafastertimeandatlowercosts.Itallowstodefineuserprogrammablehardwaresubsystemwhichcanbeeasilyupdated.Thus,theintegrationofFPGAcancarryoutalotofcomputing-intensiveandtime-criticaltaskssuchasinformationacquisitionanddataprocessioninrobotcontrolapplications[11].

88FPGAImplementationofAbsoluteRobotLocalization|77AzharandDimondin[12],useFPGAsfortheimplementationofcontrolandsensorfusionalgorithmsintheinertialnavigationsystemofaMobileInvertedPendulum(MIP)robot.TheFPGAtechnology,despiteitslittleuseinrobotics,hassignificantadvantagesespeciallyimportantcapacitythatallowstouseasinglePFGAtocontrolarobotespeciallyforparallelapplications.AhardwareimplementationofartificialneuralnetworkusedforvisualtrajectorycontrolofamobilerobotusingFPGAispresentedin[13].Anotherapplication,in[14],consistsoninterfacingallthemodulesusedbytherobottodetectobstaclesandcontroltherobotspeed.Thesedifferentmodulesareactuators,sensors,wirelesstransmissionscircuits.AfuzzylogicandanhybridfuzzylogiccontrollerforroboticnavigationhavebeencarriedoutusingFPGAin[15,16].TheresearcheshavedemonstratedthattheFPGAisaninterest-ingsolutionthatcombinessatisfactoryperformanceandcostdesignforroboticapplications.Theproposedtechniquepresentedinthispaperconsistsoncombiningsensorsmeasurementswithexternalabsolutedatacomingfromatrackingcamera,toreducetheencoderpositionerrorsandprovidethebestestimateoftherobotposition.Since,therearefewworkonFPGAinrobotlocalizationarea,wedecidetoimplementthisabsolutelocalizationalgorithm(basedonwebcamdata)onFPGAembeddedsystem.Thecompletetechniquewillbepresentedonongoingpublications.Thepaperisorganizedasfollows.Insection3,wedescribetheproposedapproachforabsolutelocalization.Then,wepresentthedifferentalgorithmsusedtodetermineabsolutepositionoftherobotinthetestplatformbyexposingprogressivelyvalidationresults.Insection4,wedetailtheHW/SWimplementationontheAlteraFPGAembeddedsystem.3DescriptionoftheabsolutelocalizationalgorithmTheproposedabsolutelocalizationalgorithmisbasedonwebcamdata.Theprovidedimagesaretreatedwithvariousimageprocessingalgorithmstoobtainthepositionofthemobilerobotonamapformedbytwoaxiscartesiancoordinatesystem.AsweshowninFig.1,theexternalabsolutedataareobtainedfromacameramountedontheceilingofthetestenvironmentandensuredtheabsolutelocalization.Thewebcamprovidesacolorimagewhichwillsubsequentlybehandledbyanimageprocessingprogramtodeterminefirstly,thereferencesystemandsecondlytherobotposition.TherobotusedinthisworkistheMiniKheperaII(Fig.2).Itisequippedwitha68331Motorolaprocessorwith25MHzoffrequency,8InfraredproximityandambientlightsensorsandaserialportprovidingcommunicationwiththePC(sendingandreceivingcommands).ThisapproachisbasedonthreestepsasshowninFig.3.

8978|A.Ghorbeletal.Fig.1.Theexperimentalenvironment.Fig.2.ThemobilerobotKheperaII.RobotinitialInputHoughpositionGrayscalingSobelFilterThresholdingErosionImageTransformCurrentrobotStep2Creationofthecroppedpositionrectangle+ReferenceHough–GrayscalingSobelFilterThresholdingΣImageTransformlandmarkStep1kStep3RobotrealpositionFig.3.Theadoptedapproach.

90FPGAImplementationofAbsoluteRobotLocalization|79Thestep1consistsonthelocalizationofthefourlandmarksanddefinitionoftherobotcartesiancoordinatesystemsetupagainstwhichthepositionoftherobotwillbecalculated.Inthestep2,wecalculatethecurrentpositionofourrobotinasequenceofimages.Instep3andlaststep,weestimatethepositionoftherobotontheplatform.ThistechniquehasbeenimplementedandtestedonPCasafirststep,underVisualStudiowithClanguageusingOpenCVLibrary.OpenCVisanopensourceandfreecomputervisionlibrary.TheonlyOpenCVfeaturesusedinapplicationareloadinganddisplayingimagesingraphicalwindows(TheimagesarecapturedfromacameraandsavedintheharddiskofPC).Nopredefinedimageprocessingfunctionwasused.Thisisinordertobeabletoembedquicklytheapplicationandfacilitatethehardwareaccelerator’sdesign.Itisimportanttomentionthatwhenthistechniquewasestablished,OpenCVcannotbecrosscompiledontheNios-IIprocessor.Thethreestepsshownpreviouslywillbedetailedinthissection.3.1TherobotCartesianCoordinateSystemTheaimofthispartistoidentifythelabelsusedforthecartesiancoordinatessystemsetupagainstwhichthepositionoftherobotwillbecalculated.Weextractthepixelcoordinatesofthereferencesystemfromareferenceimage,whichiscapturedandstoredinadvance.Thisimage,asshowninFig.4,describestheworkspaceofthemobilerobotlimitedbyfourlandmarksplacedonthecorners.Afterextractingcoordinatesofthelandmarks,wecandefineourreferencesystemshowninFig.5.Fig.4.Thereferenceimage.Fig.5.TheCartesianCoordinatesSystem.

9180|A.Ghorbeletal.Fig.6.Croppedrectangle:70×70pixels.3.2DeterminationofrobotpositionWedeterminatetherobotpositioninasequenceofimages:weextractfromeachsequence’sframetherobot’sposition.Inordertodecreasetheprocessingtime,weapplyaminimumboundingbox(MBB),embracingtherobot,tothebinaryinputimageinordertoreducethenumberofthetreatedpixels.Theproposedalgorithmispresentedasfollows:–Inthefirststep,wehavethefirstcolorimagewiththedefaultwebcamimagesize640×480pixels.Onthisimage,wecalculatetheinitialrobotpositioninpixelscoordinates.–Inthesecondstep,wedefineacroprectangle(Fig.6)thatdependsonpreviousrobotcoordinates.3.3RobotandlandmarklocalizationtechniqueThedifferentalgorithmsusedarethegrayscaling,theSobelfilter,thethresholds,erosionandHoughCircleTransform.AsshowninFig.7,thefirststepconsistsonconvertingthecolorimagetoagraylevel(Fig.7b)inwhichthered,greenandbluecomponentshaveequalintensityinRGBspace.Then,thesecondstepconsistsontransforming,usingSobelfilter,thegrayscaleimageintoablackimageunlessatthepointswhereacontourisdetectedthatismarkedinwhite(Fig.7c).Afterthat,wemustreduceinFig.7calargequantityofinformations(thewhitelines)whileconservinginFig.7dnearlyallpertinentinformationstoseparateobjectsfrombackground.Toisolatetherobotfromthefourmarkers(Fig.7e),weapplytheerosiononthebinaryimage.Finally,thelaststepistolocatethefourlandmarksandtherobotusingtheHoughCircleTransformalgorithm.Afterapplyingthistreatment,wehavesucceedtolocatesimultaneously,fourlandmarks(definingthereferencesystemoftherobot)(seeFig.7g)andidentifytherobot(seeFig.7f).

92FPGAImplementationofAbsoluteRobotLocalization|81(a)Inputcolorimage(b)Grayscaleimage(c)Edgedimage(d)Binaryimage(e)Erodedimage(f)Centerofrobot(g)CenteroflandmarksFig.7.Resultsofimageprocessingalgorithms.

9382|A.Ghorbeletal.Theabsolutelocalizationsystemusingcameradeterminesthecoordinatesinarefer-encelinkedtothecameraimage(A,j,i).Thesecoordinateshavetobetransformedintherobotcoordinatessystem(O,x,y).3.4TransformationfromimagecoordinatestorealcoordinatesBasedonboththerealandimagelandmarks’coordinateswiththegravitycenteroftherobot,wecancalculatetherobotrealposition.Hence,wedefinetheleftbottomlandmarks(j1,i1)astheoriginofoursystemasshowninFig.8.Yi(j3,i3)(j4,i4)yrobotrixOriginO(j1,i1)xrobot(j2,i2)AjjrFig.8.Thepositioningarchitectureadopted.Weobtaintherealimagecoordinatesinthecoordinatesystem(O,x,y)usingequation(1):⎧⎨xrobot=jr−j1(1)⎩yrobot=ir−i1wherejrandiraretheposition(j,i)oftherobotontheimageandj1andi1aretheposition(j,i)oftheorigin.Theresultingcoordinatesaremeasuredusingpixelnumberasunit.Totransformthemincentimeters,amultiplicationbyconstantcoefficientskxandkyisused.Thesecoefficientsarecalculatedonthebasisofboththerealandthepixelsdistancebetweenthelandmarksalongthex-axisandthey-axis.

94FPGAImplementationofAbsoluteRobotLocalization|83Thecoefficientskxandkyaregivenby:⎧⎪⎪Dx⎨Kx=j2−j1(2)⎪⎪Dy⎩Ky=i4−i1whereDxandDyarerespectivelytherealdistanceincentimetersbetweenthelandmarksalongthex-axisandthey-axis.ByapplyingthepreviouslydescribedlocalizationtechniqueonaPC(Core2duo,2GHz,3GbRAM),theaverageprocessingtimeisabout0.009secondswhichissuffisanttoprovidecontinuousrobotnavigationwithmaximumvelocity.InordertoensureautonomousnavigationwithoutanencombrantPC,wedecidetoimplementthistechniqueoflocalizationonanALTERAFPGAembeddedsystemlikeshowninFig.9.CameraFPGARobotFig.9.Thehardwareimplementation.4ImplementationonAlteraFPGAplatformInthispaper,wegiveproofofconceptofusingFPGAtechnologyinourlocal-izationtechnique.WeadoptanAlteraCycloneIIIFPGAStarterKit.Thus,inthissection,wepresentthestepsneededtoimplementthistechniqueontheNIOS-IIprocessor.Thenwedetailthevariousstagestomeasuretheprocessingtime.Finally,weexposedifferenttechniquesusedtooptimizeandreducetheoverallexecutiontime.

9584|A.Ghorbeletal.4.1PurelySoftwareImplementationThefirststepwehavetodoistheadaptationofthecodewhichhasbeenexecutedunderVisualStudiousingOpenCVlibrary.Twopossiblesolutionsexist.ThefirstconsistsonbuildingtheembeddedLinuxoperationsystemonNios-IIandcrosscompiletheOpenCVlibrary.ThissolutionissuitablebutimpossibletorealizewiththepreviousNiosversion.Also,thismethodincreasesthecomputationaltime(alreadyhugeasdiscussedafter).So,theadoptedsolutionistodeclareimagesinheaderfile(.h)directlyinthesourcecodeasdescribedinFig.10.1typedefstruct{2unsignedcharB;3unsignedcharG;4unsignedcharR;5}Pixel;6Pixelimage[307200]={{0,0,3},{1,2,6},{5,9,10},{9,13,14},{11,1Fig.10.Exampleofheaderfile.Afterimporting,configuringandexecutingtheprogramontheFPGA,weneedtoverifytheresult.SinceourFPGAisnotequippedwithaVGAport,wehavedisplayedintheNIOSconsoletheRGBcomponentsvalues,weputeachinamatrixandweconcatenatetheminordertoobtaina3channelsmatrix,usingMATLABcommand:MyRGB=cat(3,Red,Green,Blue)(3)TheimagesobtainedareidenticaltothoseshowninFig.7onthePCversion.ThisvalidatesthepurelysoftwareCversiondevelopedfortheNIOS-II.4.2TimeprocessingmeasureTheimplementationoftheabsolutelocalizationalgorithmontheFPGA,under100MHzoffrequency,leadstoanassemblyoftimeexecutionwhichisresumedinTab.1andTab.2.Wecannoticethatweobtain118.106secondsastimeprocessingforthefirstimage(640×480pixels)usedtoidentifythestartingrobotposition.However,therestofimages(70×70pixels),usedforrobottracking,needonly12.316seconds.Toensurefluidrobotnavigation,theprocessingtimemustbelessthan0.48seconds.Then,wedecidetoaccelerateoursystem.

96FPGAImplementationofAbsoluteRobotLocalization|85Tab.1.Measuringresultforthefirstimage.Section%oftotaltimeTime(sec)Time(clocks)GrisandSobel83.198.134039813403474Binarisation0.2570.3030530305130Erosion0.7880.9311593115463HoughTransform15.918.737431873743278TotalTime:118.106secondsTab.2.Measuringresultfortheremainingimages.Section%oftotaltimeTime(sec)Time(clocks)GrisandSobel192.33668233668191Binarisation0.2970.036583657597Erosion1.430.1761617616146HoughTransform79.39.76661976661326TotalTime:12.3169seconds4.3TechniquesforacceleratingtimeprocessingInordertoacceleratethesystemandreducetheexecutiontime,weusedifferentmethods:4.3.1ToolsavailableontheNiosIIprocessor–TheuseofthefastversionoftheNIOS(NIOS-II/fcore)thatprovides133MHzasfrequency,aperformanceover300MIPSandsix-stagespipelinetoachievemaximumMIPSperMHz.–Theuseofthefloating-point(FP)custominstruction:likecustomperipherals,custominstructionsallowustoincreasesystemperformancebyaugmentingtheprocessorwithcustomhardwaredirectlyaddedtotheNiosUAL(Fig.12).Thefloating-pointcustominstructions,optionallyavailableontheNiosIIprocessor,implementsingle-precision,floating-pointarithmeticoperations.Whenthefloating-pointcustominstructionispresentinthedesign,thecodewillbebuildusingthecustominstructions(addition,subtraction,multiplicationanddivision)likeshowninFig.11.

9786|A.Ghorbeletal.MultiplicationAdditionNios-IIprocessorSubtractionDivisionAvalonBusPerformancecounterMemory(slave)(slave)Fig.11.ArithmeticoperationsconnectstotheNiosII.CustomInstructionNiosIIAALUResultBFig.12.CustomInstructionconnectstotheNiosIIALU.ThisFPunithasbeenusedtocomputethegrayscaleequation:Gray=0.299×Red+0.587×Green+0.114×Blue(4)TheaddedhardwarefeaturesleadtoanassemblyoftimeexecutionwhichisresumedinTab.3andTab.4.

98FPGAImplementationofAbsoluteRobotLocalization|87Tab.3.Measuringresultforthefirstimage.Section%oftotaltimeTime(sec)Time(clocks)Sobel56.59.89234989234461Binarisation0.3010.052655265333Erosion2.180.3811238111921HoughTransform417.17303745904065TotalTime:17.4994secondsTab.4.Measuringresultfortheremainingimages.Section%oftotaltimeTime(sec)Time(clocks)Sobel7.480.2967229671767Binarisation0.1510.00599598544Erosion0.9360.037123712309HoughTransform91.43.62636362636394TotalTime:3.9664secondsWecannoticethat,withthesecustomhardwarelogicblocks,weobtain17.499sec-ondsastimeprocessingforthefirstimagewhichcorrespondsto85%ofimprovement.Therestofimages(70×70pixels)needonly3.966secondswhichnearlycorrespondsto70%ofimprovement.4.3.2CreationofhardwareacceleratorAccordingtoTab.3,theSobelfunctionmonopolizes56.5%oftotaltimesinceitcontainscomplexoperationslikemagnitudeequation(5):|Mag|=(Eh)2+(Ev)2(5)whereEhandEvrepresentrespectivelythegradientintensityinimagealonghorizon-talandverticaldirections.So,wedesignahardwareacceleratorthatcomputesequation(5)toHWinordertoaccelerateexecutiontimeforthefirstimage.ThisacceleratorcommunicateswiththeNios-IIprocessorthroughAvalonbus(Fig.13).TheschematicblockoftheSobelHWacceleratorisshowninFig.14.

9988|A.Ghorbeletal.NiosIIMemory(master)(slave)AvalonBusSobelAcceleratorFig.13.TheSobelhardwareaccelerator.MultiEh[15..0]2XInputAddSqrtMag[15..0]A+RQOutputBrOutputMultiEʋ[15..0]2XInputFig.14.TheschematicblockofSobelaccelerator.ThevisualizingimagesafteraddingtheSobelacceleratorareidenticaltothoseshowninFig.7.TheseresultsvalidatethedesignofFig.13.Withalltheseacceleratingtools(NIOS-II/f,FloatingPointUnitandtheSobelaccelerator),wereach11.052secondsofexecutiontimeforthefirstimage(Tab.5)whichcorrespondsto35%oftimedecreaseagainstprevioustime.Fortherestoftheimages,wereach3.788seconds(Tab.6).Tab.5.Measuringresultforthefirstimage.Section%oftotaltimeTime(sec)Time(clocks)Sobel29.33.23552323552263Binarisation0.4760.052635262942Erosion3.440.3807238071911HoughTransform66.87.38325738324576TotalTime:11.0523seconds

100FPGAImplementationofAbsoluteRobotLocalization|89Tab.6.Measuringresultfortheremainingimages.Section%oftotaltimeTime(sec)Time(clocks)Sobel4.930.1868718686747Binarisation0.0240.0009191047Erosion0.2280.00864863854HoughTransform94.83.59213359212649TotalTime:3.78876seconds5ConclusionInthiswork,wehaverealizedanapplicationthatensurestheabsolutelocalizationoftherobotKheperaIIinitsworkspaceusingwebcamdata.WedevelopaCversionofthelocalizationalgorithmusingOpenCVlibrary.ThenweimplementitonanALTERAFPGAembeddedsystemusingtheNIOSprocessor.Weusespecificcomputationhardware(floatingpointcoprocessor)andacustomhardwareaccelerator(Sobel)toreducetheoverallexecutiontime.TheobtainedresultsprovedtheeffectivenessoftheproposedHW/SWarchitectureinacceleratingtheprocessingtime.However,itstillenoughtoensurecontinuousandrealtimerobotnavigation.Thefurtherimprovedtheobtainedvalues,anim-plementationofaparallelarchitecturewithamulti-processorFPGAsystemwillbeheldinfuturework.DuetothelimitedcapacityoftheusedFPGA(CycloneIIIFPGAStarterkit),otheroptimizationstechniquescannotbeusedespeciallymultiprocessortechniques.Thesetechniquesareparticularlyefficientespeciallyforapplicationwithhighinherentparallelismlikeournavigationapplication.DirectnavigationoftherobotusingtheCycloneIIIFPGAStarterkitisnotpossibleduetothelackofDVIinputnorVGAoutput.WeplaninongoingworktorealizealltheapplicationusingtheML507platformwithPowerPCprocessorwhichismoreperformantthanNIOSandintegratingthecameraforimagesacquisition.Furthermore,theuseofthecameraisrestrictedtoaninternalenvironment(indoor)tobeabletofixit,whileinanexternalenvironment,itwillbehardertodoit.So,toovercomethisproblem,amulti-sensors(GPS,camerasorDGPS)fusionapproachisawaytoimproveenvironmentperceptioninareal-lifescenariowouldinvolveamuchlargerarea.Bibliography[1]M.Jallouli,L.AmouriandN.Derbel.Aneffectivelocalizationmethodforrobotnavigationthroughcombinedencoderspositioningandretimingvisualcontrol.InJ.ofAutomation,MobileRoboticsandIntelligentSystems,3(2),2009.[2]S.I.Roumeliotis,G.S.SukhatmeandG.A.Bekey.Smootherbased3DAttitudeEstimationforMobileRobotLocalization.InIEEEInt.Conf.onRoboticsandAutomation,May1999.

10190|A.Ghorbeletal.[3]R.Negenborn.RobotLocalizationandKalmanFiltersOnfindingyourpositioninanoisyworld.ThesisforthedegreeofMasterofScience,InstituteofInformationandComputingSciences,UtrechtUniversity.September1,2003.[4]M.S.GrewalandA.P.Andrews.KalmanFiltering,TheoryandPractice.PrenticeHall,1993.[5]J.Z.SasiadekandP.Hartana.SensorDataFusionUsingKalmanFilter.Int.Conf.oftheInt.SocietyofInformationFusion,19–25,2000.[6]F.Kobayashi,F.Arai,T.Fukuda,K.Shimojima,M.OnodaandN.Marui.SensorFusionSystemusingrecurrentfuzzyinference.J.ofIntelligentandRoboticSystems,23:201–216,1998.[7]K.ACHOURandA.O.DJEKOUNE.Localizationandguidancewithanembarkedcameraonamobilerobot.AdvancedRobotics,16(1):87–102,2002.[8]A.Kak,K.Andress,C.Lopez-Abadia,M.CarrollandJ.Lewis.HierarchicalEvidenceAccumulationinthePseikiSystemandExperimentsinModel-DrivenMobileRobotNavigation.ComputingResearchRepositoryJ.,abs/1304.1513,2013.[9]L.FredaandG.Oriolo.Vision-basedinterceptionofamovingtargetwithanon-holonomicmobilerobot.emRoboticsandAutonomousSystemsJ.,55:419–432,February2007.[10]I.MasárandM.Gerke.DSP-BasedControlofMobileRobots.DSPEducationandResearchSymp.,Birmingham,U.K,November2004.[11]P.He,M.H.Jin,L.Yang,R.Wei,Y.W.Liu,H.G.Cai,H.Liu,N.Seitz,J.ButterfassandG.Hirzinger.HighPerformanceDSP/FPGAControllerforimplementationofHIT/DLRDexterousRobotHand.IEEEInt.Conf.onRoboticsandAutomation,NewOrleans,April2004.[12]M.A.H.B.AzharandK.R.Dimond.FPGA-basedDesignofanEvolutionaryControllerforCollision-freeRobot.ACM/SIGDA11thInt.Symp.onFieldProgrammableGateArrays,2003.[13]S.T.BrassaiandL.Bako.VisualtrajectorycontrolofamobilerobotusingFPGAimplementedneuralnetwork.PollackPeriodica,4(3):129–142,DOI:10.1556/Pollack.4.2009.3.12,December2009.[14]S.KaleandS.S.Shriramwa.FPGA-basedControllerforamobilerobot.InInt.J.ofComputerScienceandInformationSecurity,3(1):1–5,July2009.[15]J.M.Ramos-Arreguin,E.Guillen-Garcia,S.Canchola-Magdaleno,J.Pedraza-Ortega,E.Gorrostieta-Hurtado,M.A.Aceves-FernndezandC.A.Ramos-Arreguin.FuzzyLogicHardwareImplementationforPneumaticControlofOneDOFPneumaticRobot.AdvancesinSoftComputingbook.SpringerBerlinHeidelberg,:500–511,2010.[16]N.Zhang,R.Kamdem,E.Ososanya,W.MahmoudandL.Wenxin.VHDLimplementationofthehybridfuzzylogiccontrollerswithFPGA.Int.Conf.onIntelligentControlandInformationProcessing(ICICIP),:5–10,August13–15,2010.BiographiesAgnèsGhorbelborninApril1987.SinceDecember2012,sheisaPh.DatComputerandEmbeddedSystemLaboratory(CES_lab)intheNationalEngineeringSchoolofSfax,Tunisia.ShereceivedherMasterdegree(withhonors)inNewTechnologiesofDedicatedComputerSystemsdisciplinein2012.Herresearchinterestsareembeddeddevices,hardware/softwareco-design,roboticapplicationsbasedonvision.ShehasworkedextensivelyonmappingofimageprocessingalgorithmsontoFPGAs.

102FPGAImplementationofAbsoluteRobotLocalization|91MohamedJallouliwasreceivedhisDEAinAutomaticsfromUniversityofValenciennes,France,in1986andPhDinRoboticsEngineeringfromUniversityParisXII,France,in1991.In1987,hethenjoinedFrenchUniversityineducationactivitiesforhispostdoctoralperiod.InApril1991,hejoinedtheTunisianUniversitywherehehelddifferentpositionsinvolvedinbotheducationandresearchactivities.HeiscurrentlyanAssociateProfessoratHigherInstituteofIndustrialSystemsofGabesandaComputer&EmbeddedSystemlaboratory’smember.Hiscurrentinterestsincludetheimplementationofintelligentmethods(neuralnetwork,fuzzylogicandgeneticalgorithm)inroboticandvisionsystemaswellasinmulti-sensorydatafusionmobilebases.LobnaAmourireceivedthePh.DDiplomainElectricalEngineeringfrombothUniversityofSfax,TunisiaandUniversityofOrleans,France,in2012.InSeptember2009shejoinedTunisianUniversityineducationandresearchactivities.SheiscurrentlyanassistantprofessorofSoftComputingatENET’COMinTunisia.SheisamemberofthelaboratoryCEMLab(Control,EnergyandManagementLaboratory),UniversityofSfax,EngineeringSchool(ENIS).AssistantProfessorAmouriiscurrentlyworkingonEmbeddedSystemsControlinvolvingbothadaptivecontrol,imageandsignalprocessing.NaderBenAmorNaderBenAmorisanAssistantProfessorattheNationalEngineeringSchoolofSfax,Tunisia.HereceivedhisPhDinElectricalEngineeringfrombothSfaxUniversity(Tunisia)andBretagneSudUniversity(France)in2005.HisresearchinterestsareHardware-SoftwareSystemonChip,co-designmethodology,selfadaptivesystems,embeddedrealtimesystem,realtimeimageprocessingonFPGAsystems,robotics.

103