University of Southern California Dissertations and Theses

Theory and design of magnetic induction-based wireless underground sensor networks

(USC Thesis Other)

Theory and design of magnetic induction-based wireless underground sensor networks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THEORYANDDESIGNOFMAGNETICINDUCTION-BASED WIRELESSUNDERGROUNDSENSORNETWORKS AThesis Presentedto TheAcademicFaculty by AgneloRochadaSilva InPartialFulfillment oftheRequirementsfortheDegree DoctorofPhilosophyinthe MingHsiehDepartmentofElectricalEngineering, USCViterbiSchoolofEngineering UniversityofSouthernCalifornia August2015 THEORYANDDESIGNOFMAGNETICINDUCTION-BASED WIRELESSUNDERGROUNDSENSORNETWORKS PhDDissertationDefenseCommitteeMembers: ProfessorMahtaMoghaddam,Advisor MingHsiehDepartmentofElectricalEngineering-Electrophysics, USCViterbiSchoolofEngineering UniversityofSouthernCalifornia ProfessorBehnamJafarpour MorkFamilyDepartmentofChemicalEngineeringandMaterialsScience, USCViterbiSchoolofEngineering UniversityofSouthernCalifornia ProfessorHosseinHashemi MingHsiehDepartmentofElectricalEngineering-Electrophysics, USCViterbiSchoolofEngineering UniversityofSouthernCalifornia ProfessorUrbashiMitra MingHsiehDepartmentofElectricalEngineering-Systems, USCViterbiSchoolofEngineering UniversityofSouthernCalifornia DefenseDate: June19,2015 Tomyfamily, fortheircontinuoussupportandencouragement duringalltheseyears. MywifeMonicaSilva: foralwaysbeingthere, cheeringmeupandsupportingme, ingoodandbadtimes. MydaughterRebecaSilva: foryourendurance, forbeingwithusduringallthese difficultyearsofstudies. MymotherMariaLeciAlves andmyfatherArlindoSilva: forgivingbirthtomeatthefirstplace, givingmeloveandsupportinmultipleways throughoutmylife. ii Acknowledgements Iwouldneverhavebeenabletofinishmydissertationwithouttheguidanceofthecommit- teemembers,helpfromlabmates,andlong-termsupportfrommyfamilyandwife. First,I express my deepest gratitude to my advisor, Prof. Mahta Moghaddam, for her exceptional guidance,researchvision,andalsoconfidenceonmypotentialachievements. Shealsopro- videdmeanexcellentatmospherefordoingresearch. I would like to thank the doctorate committee members for helping me in selecting the most important topics to be addressed during my research. Their suggestions also helped me in developing an important background in electrochemistry, a novel research area for me. I also express my gratitute to Prof. Mingyan Liu for her guidance and support as my co-advisorattheUniversityofMichigan,whereIstartedthefirstpartofmydoctoratestud- ies. Iamverygratefultoallmyfellowlabmatesforthestimulatingdiscussionsandappreci- ationwords-whenevereachoneofyoustoppedtohearme,ortomakequestionsaboutmy research, or even to debate about research strategies, I was actually receiving an important contribution regarding my endurance at the PhD program. Special thanks also go to you guyswhohaddirectinteractionsatmyresearch: RuzbehAkbar,RichardChen,PratikShar, JohnStang,andJaneWhitcomb. My research work also had external financial support and I would like to acknowledge thisimportantsupportto: a)NationalAeronauticsandSpaceAdministration,EarthScience Technology Office, Advanced Information Systems Technology Program and b) Brazilian National Research Council (CNPq), under the Brazilian Programme Science without bor- ders. I am also in huge debt with my wife during all those long days and nights helping me withlabandfieldexperiments!! Besidestheemotionalsupport,hercuriosityandquestions alsohelpedmetoachievebalancedsolutions. iii Contents ListofTables viii ListofFigures ix Abstract xv Introduction 1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 TargetApplications&State-of-the-Art . . . . . . . . . . . . . . . . . . . . . . . 7 ResearchObjectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Soil Dielectric Model for Wireless Underground Sensor Networks Operat- ingatSub-MHzRange . . . . . . . . . . . . . . . . . . . . . . . . 17 Method of quantitative identification of the impedance due to the electrode polarizationeffectsonsub-MHz2-electrodemeasurementsystems . 17 Magnetic-Induction(MI)SignalAttenuationModelforWUSNs . . . . . . 18 MI-basedPhysicalLayerforPoint-to-PointCommunicationinSoils . . . . 19 EnergyManagementandNetworkingforMI-basedWUSNs . . . . . . . . 19 ThesisOutline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 I Sub-MHzSoilDielectricModel 21 1 MI-Soil: SoilDielectricModelforSub-MHzWUSNs 22 1.1 IntroductiontotheElectricalPropertiesofSoil . . . . . . . . . . . . . . . 23 1.2 PolarizationEffectsinSoilMixtures . . . . . . . . . . . . . . . . . . . . . 29 1.2.1 CharacteristicsofSoilParticles . . . . . . . . . . . . . . . . . . . 30 1.2.2 DryandWetSoils . . . . . . . . . . . . . . . . . . . . . . . . . . 31 1.2.3 OverviewonBroadbandSpectralResponseofSoils . . . . . . . . . 33 CONTENTS v 1.3 MeasurementsofElectricalPropertiesofSoil . . . . . . . . . . . . . . . . 40 1.4 SoilDispersionatLowFrequencies . . . . . . . . . . . . . . . . . . . . . 46 1.5 SimplifiedSoilDielectricModelforMI-BasedWUSNs . . . . . . . . . . . 56 1.5.1 ElectrodePolarization(EP)Verification . . . . . . . . . . . . . . . 56 1.5.2 EPCorrection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 1.5.3 SimplifiedMI-SoilDielectricModelforMI-WUSNs . . . . . . . . 64 2 TowardanAccurateSub-MHzSoilDiel. Spectroscopy 70 2.1 TheDielectricMeasurementInversionProblem . . . . . . . . . . . . . . . 71 2.2 TheElectrodePolarization(EP)Effects . . . . . . . . . . . . . . . . . . . 75 2.3 MethodstoMitigatetheEPEffects . . . . . . . . . . . . . . . . . . . . . . 85 2.4 NovelEP-IdentificationTechnique: LinearEP-Match . . . . . . . . . . . . 90 2.5 InstrumentationSetup,Experiments,andDiscussion . . . . . . . . . . . . 100 II CommunicationPhysical(PHY)LayerforMI-WUSNs 113 3 MISignalAttenuationModelforWUSNs 114 3.1 FieldsDuetoaMultilayerCoil(TXside) . . . . . . . . . . . . . . . . . . 115 3.2 InducedVoltageinaMultilayerCoil(RXside) . . . . . . . . . . . . . . . 120 3.3 ResonanceTechniquesforMI-BasedWirelessCommunication . . . . . . . 122 3.4 MISignalAttenuationModelforWUSNs . . . . . . . . . . . . . . . . . . 126 3.5 ExperimentsandValidation . . . . . . . . . . . . . . . . . . . . . . . . . . 132 3.5.1 SetofExperimentsA . . . . . . . . . . . . . . . . . . . . . . . . . 133 3.5.2 SetofExperimentsB . . . . . . . . . . . . . . . . . . . . . . . . . 134 3.5.3 SetofExperimentsC . . . . . . . . . . . . . . . . . . . . . . . . . 136 4 OperatingFrequencyRangeforMI-WUSNs 142 4.1 FieldsDuetoaMultilayerCoil(TXside) . . . . . . . . . . . . . . . . . . 142 4.2 InducedVoltageattheRXCoil . . . . . . . . . . . . . . . . . . . . . . . . 144 4.3 ResonanceattheRXSide . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 4.4 MISignalAttenuationModelforWUSNs . . . . . . . . . . . . . . . . . . 145 4.5 OperatingFrequencySelectionStrategy . . . . . . . . . . . . . . . . . . . 146 5 FrequencyAdaptationForExtremeSoilConditions 151 5.1 PracticalMI-SignalAttenuationModel . . . . . . . . . . . . . . . . . . . . 152 5.2 OptimizationProblemStatement . . . . . . . . . . . . . . . . . . . . . . . 157 CONTENTS vi 5.3 PreliminaryResultsandDiscussion . . . . . . . . . . . . . . . . . . . . . 161 5.4 FL-FHFrequency-SwitchingScheme . . . . . . . . . . . . . . . . . . . . 163 6 Dual-WireSchemeAddedtoFrequencyAdaptation 167 6.1 MI-CoilDesignOptimizationProblem . . . . . . . . . . . . . . . . . . . . 168 6.2 PreliminaryEmpiricalEvaluations . . . . . . . . . . . . . . . . . . . . . . 170 6.3 AdaptiveSystem: Analysis&Discussion . . . . . . . . . . . . . . . . . . 171 6.4 MI-WUSNAdaptiveSystems: FinalRemarks . . . . . . . . . . . . . . . . 173 III Energy,Reliability,andNetworkingAspectsofMI-WUSNs 177 7 Energy-ManagementFrameworkforSpecialWSNs 178 7.1 FrameworkForSensorNodesWithConstrainedEnergyProfiles . . . . . . 179 7.2 EnergyManagementinWSNs . . . . . . . . . . . . . . . . . . . . . . . . 182 7.2.1 DesignChallengesandPitfalls . . . . . . . . . . . . . . . . . . . . 182 7.2.2 TheConceptsofEnergyEffortTripodandEnergyControlLoop . . 186 7.3 Energy-AdaptiveFrameworkforWSNArchitectures . . . . . . . . . . . . 191 7.3.1 Foundation1: TheStrategicUseofPrimaryCells . . . . . . . . . . 191 7.3.2 Foundation2: DistributedSystemInsideaSensorNode . . . . . . 194 7.3.3 Foundation3: DualDuty-Cycle(DDC)Operation . . . . . . . . . 198 7.3.4 CharacteristicsoftheLDCMode . . . . . . . . . . . . . . . . . . 201 7.3.5 DDCSystem: ImplementationGuidelines . . . . . . . . . . . . . . 209 7.4 Cross-LayerProtocolforVeryLowDuty-Cycle(LDC)Mode . . . . . . . . 214 7.4.1 RelatedWork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 7.4.2 ProtocolOverview . . . . . . . . . . . . . . . . . . . . . . . . . . 215 7.4.3 Definitions&Terminology . . . . . . . . . . . . . . . . . . . . . . 219 7.4.4 DesignGoals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222 7.4.5 BETS:NormalOperation . . . . . . . . . . . . . . . . . . . . . . 223 7.4.6 BETS:DealingWithErraticScenarios . . . . . . . . . . . . . . . . 226 7.4.7 Energy-EfficiencyoftheCHNode . . . . . . . . . . . . . . . . . . 230 7.5 SimulatedandEmpiricalResults(LDCMode) . . . . . . . . . . . . . . . . 232 7.5.1 ExperimentalSetup . . . . . . . . . . . . . . . . . . . . . . . . . . 232 7.5.2 PerformanceEvaluation . . . . . . . . . . . . . . . . . . . . . . . 234 7.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 CONTENTS vii 8 AchievingHighSystemReliability 243 8.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 8.2 WatchDogTimer(WDT):Initialsteptowardreliablesystems . . . . . . . . 249 8.3 VirtualFiniteStateMachines+WDT . . . . . . . . . . . . . . . . . . . . 254 8.4 Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 9 High-LevelNetworkingAspectsofMI-WUSNs 259 9.1 PreciseSoilIrrigationandEnvironmentalMonitoring . . . . . . . . . . . . 260 9.2 PipelineLeakDetection(PLD) . . . . . . . . . . . . . . . . . . . . . . . . 261 Conclusions&FutureWork 265 Abbreviations 268 AppendixANovelIndoorSub-MHzMI-SoilTestbed 272 References 278 ListofTables 1.1 Characteristicsofthesoilsamplesusedinthisresearch. . . . . . . . . . . . 46 1.2 ElectrodePolarizationImpactonMISignalAttenuation. . . . . . . . . . . 59 2.1 Signalattenuationerrorduetothedielectricconstantestimationerror. . . . 72 2.2 Example of the impact of the EP impedance on dielectric measurements at lowfrequencies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 2.3 Example of the effectiveness of the Pt-Black electrode in reducing the EP effects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 3.1 Asymptoticbehaviorofthemagneticfieldin(3.22). . . . . . . . . . . . . . 120 4.1 Study-case: silty loam soil, targetV c =100μV, and 30cm-diameter air-cored coils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 6.1 Study-case: dual-wirevs. singlecoil. . . . . . . . . . . . . . . . . . . . . . 173 7.1 MainTermsandAcronymsMainlyIntroducedinThisChapter . . . . . . . 181 7.2 Powerprofileusedinthesimulations. . . . . . . . . . . . . . . . . . . . . 202 7.3 TargetApplicationsforLDC-OnlyMode . . . . . . . . . . . . . . . . . . 209 7.4 Defaultparametersforsimulations. . . . . . . . . . . . . . . . . . . . . . . 233 viii ListofFigures 1 Taxonomy of Wireless Communication in Underground or Confined Areas (WCUCA).[1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.1 Soilclassificationaccordingtotheirtexture(dataatthetablesin[2]). . . . 31 1.2 Wetsoilparticle-fluidinteractions(basedonillustrationsin[2]) . . . . . . 32 1.3 Spectraof(a)electronicpolarizationand(b)ionicpolarization. . . . . . . . 33 1.4 Polarizationtypes: electronic,ionic,andorientational(ordipolar). . . . . . 35 1.5 Spectraof(a)orientational,and(b)spatial+doublelayerpolarizations. . . 36 1.6 Cole-Colediagramfor(a)single-timerelaxationand(b)spatialpolarization [2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 1.7 Polarizationtypes: interfacial(orspatial,orM-W),Sternlayer,andDouble layer.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 1.8 Coaxialcapacitorcellusedforthesoildielectricmeasurements. . . . . . . 43 1.9 Empiricalresults: dielectricconstantdispersionfor9differentsoils. . . . . 47 1.10 Abnormallyhighvaluesforthedielectricconstantatlowfrequencies. . . . 48 1.11 Empiricalresults: effectivelosstangentdispersionfor9differentsoils. . . . 49 1.12 Empiricalresults: effectivelosstangentdispersionfor9differentsoils. . . . 50 1.13 Empiricalresults: effectivelosstangentdispersionfor9differentsoils. . . . 51 1.14 EmpiricalresultsforSAN1soil: Dielectricconstantdispersionfordifferent volumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . . . . . . 52 1.15 Empirical results for SAN1 soil: Effective loss tangent dispersion for dif- ferentvolumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . . . 53 1.16 Empirical results for TONZI soil: Dielectric constant dispersion for differ- entvolumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . . . . 54 1.17 Empirical results for TONZI soil: Effective loss tangent dispersion for dif- ferentvolumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . . . 54 ix LISTOFFIGURES x 1.18 Empirical results for NHS-SAT soil: Dielectric constant dispersion for dif- ferentvolumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . . . 55 1.19 Empirical results for NHS-SAT soil: Effective loss tangent dispersion for differentvolumetricwatercontent(VWC)levels. . . . . . . . . . . . . . . 55 1.20 EvidenceoftheEPeffectsonsoilmeasurements. Reproducedfrom[3]. . . 56 1.21 Typical software/mathematical technique used to perform EP correction. Reproducedfrom[3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 1.22 EP-correctionforthedielectricconstantk ′ accordingtoAlgorithm1. . . . . 63 1.23 Fittingforthebcoefficientforthedielectricconstantk ′ ,TONZIclass. . . . 66 1.24 Fittingforthemcoefficientforthedielectricconstantk ′ ,TONZIclass. . . . 66 1.25 Fittingforthebcoefficientforthedielectricconstantk ′′ e ,TONZIclass. . . . 67 1.26 Fittingforthemcoefficientforthedielectricconstantk ′′ e ,TONZIclass. . . 67 1.27 Comparisonbetweenthemeasuredvaluesfork ′ andtheresultedcurvesdue toMI-SOILv6model(TONZIsoilclass). . . . . . . . . . . . . . . . . . . 69 1.28 Comparisonbetweenthemeasuredvaluesfork ′′ e andtheresultedcurvesdue toMI-SOILv6model(TONZIsoilclass). . . . . . . . . . . . . . . . . . . 69 2.1 ConceptualvisionofcomprehensivedualEIS-MISsystemforsub-MHzdi- electricspectroscopywhereaMI-basedwirelesscommunicationsolutionis animportantcomponent. . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.2 EquivalentcircuitmodelsfortheEPimpedance: Warburgmodel(1899). . . 82 2.3 SimulatedresultsfortheWarburgEPmodelandk=0.0002. . . . . . . . . . 82 2.4 EquivalentcircuitmodelsfortheEPimpedance: Frickemodel(1932). . . . 83 2.5 SimulatedresultsfortheFrickeEPmodelandk=0.0002* √ 2π. . . . . . . 83 2.6 InitialinstrumentationsetupusedfortheevaluationoftheLinearEP-Match method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 2.7 Novel use of FFT Analyzer for Total Harmonic Distortion (THD) analysis indielectricmeasurements: thegoalistodeterminewhenthesystementers initsnon-linearregionduetotheEPphenomenon. . . . . . . . . . . . . . 102 2.8 Novel use of FFT Analyzer for Total Harmonic Distortion (THD) analysis indielectricmeasurements: thegoalistodeterminewhenthesystementers initsnon-linearregionduetotheEPphenomenon. . . . . . . . . . . . . . 103 2.9 LinearEP-MatchMethod: definingpotentialpairsofEP-matchpoints. . . . 104 2.10 Linear EP-Match Method: achieving the EP-corrected value for the dielec- tricconstant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 LISTOFFIGURES xi 2.11 Linear EP-Match Method: determining upper-boundary for the true value ofthedielectricconstant. . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 2.12 Linear EP-Match Method: determining the estimation error of the method regardingthedielectricconstant. . . . . . . . . . . . . . . . . . . . . . . . 106 2.13 Linear EP-Match Method: determining the estimation error of the method regardingtheeffectiveconductivity. . . . . . . . . . . . . . . . . . . . . . 107 2.14 MI-Soil dielectric model proposed in Chapter 1 and applied to the NHS- SATsoil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 2.15 Critical scenario to be investigated in this context: saturated soil with very high conductivity. The EP effects are expected to be very strong for such scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 2.16 Recent empirical evaluation using commercial impedance analyzers and two cell constants for the dielectric cell. The EP effects are clearly present atthemeasurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 2.17 Comparison between the preliminary results of the new instrumentation scheme associated to the Linear EP-Match method and the MI-Soil dielec- tricmodelestimations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 3.1 Coordinatesystemandwaveequationduetoatime-varyingsource. . . . . 115 3.2 Horizontalaxesdeployment: strongerRXsignal,constrainednetworktopol- ogy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 3.3 LCresonancecircuitsforMI-WUSNnodes. . . . . . . . . . . . . . . . . . 122 3.4 RXcircuitmodelandexamplesofvoltagegains/levelsfordifferentfrequen- cies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 3.5 Rightside: properLCresonanceconfigurationforRXcircuitofMI-WUSN nodes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 3.6 SetofexperimentsA:soilsettingsnotimpactingtheMIchannel. . . . . . . 134 3.7 SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS- DRY1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 3.8 SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS- DRY2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 3.9 SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS- WET. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 3.10 SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS- SAT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 LISTOFFIGURES xii 3.11 SetofexperimentsC:ErroranalysisoftheMIsoilattenuationmodel. . . . 140 5.1 Scenario investigated in this chapter: practical and optimized performance aspectstargetinglow-powerandmid-range(15..30m)MI-WUSNs. . . . . . 152 5.2 Selectingthebestfrequency/configurationforFL(worstscenario). . . . . . 161 5.3 Energyandbandwidthpenaltieswhenusingafixedfrequency(designedfor theworst-case40%VWC)fordifferentsoilmoisturelevels. . . . . . . . . 164 5.4 Selectingthebestfrequency/configurationforFH (bestscenario). . . . . . 165 6.1 MI-signal attenuation exclusively due to the medium path (referenced to r=1m). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 6.2 Indoor Sub-MHz MI-Soil testbed used for the empirical evaluations in this work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 6.3 RXinducedvoltagelevelasafunctionofdistance,frequency,andsoilcon- ditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 7.1 Effect of sub-zero temperatures on secondary (top) and primary (bottom) cells(AnnArbor, MI,USA). Therechargingprocess ofthe secondarycells is impacted by low temperatures causing node failures (lines in the figure). Primarycellsaremoreresilienttoextremetemperatures. . . . . . . . . . . 185 7.2 Energy effort tripod concept: coordinated efforts involving hardware, net- work algorithms, and application demands lead to an energy-balanced and efficientWSNsolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 7.3 Energy control loop concept: the operation of a WSN node is regulated by its energy state. The decisions are triggered by an energy-management modulethatcanbeimplementedinternallyinthenode,atthenetworklevel, inacentralizeddataserver,orbyacombinationoftheseoptions. . . . . . . 189 7.4 Examples of energy-management efforts. (a) At the node level: optimized controlofanenergyharvester,(b)atthenetworklevel: theremainingenergy ofanodeisusedascriteriaforitsselectioninnetworkactivities,and(c)ata centralizedlevel: basedonthesensingdatareceivedfromthenodes,thedata serverdefineswhatnodeswillsenseaccordingtolocation/timescheduling. 190 LISTOFFIGURES xiii 7.5 The Dual Duty-Cycle (DDC) node has multiple microcontrollers (MCUs) and intelligent devices. That is, it encompasses a distributed system inside itself. The main goal is to achieve energy savings at unprecedented levels compared to traditional nodes. The existence of multiple power lines (lines withoutarrows)ratherthanasinglepowerlineisassociatedwiththeuseof the power gating technique [4] and different voltage conditioning schemes fortheinternalmodules. . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 7.6 Dual Duty-Cycle (DDC) Operation: the network switches between LDC and RDC modes. In RDC mode, the network maintains its original charac- teristics. In LDC mode, the BETS protocol becomes active, a planned net- work segmentation occurs, and the network achieves its maximum energy efficiency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 7.7 LifetimeofaWSNnode(seeTable7.2)fordifferentnetworkoverheadsand applicationduty-cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 7.8 DifferenttopologiesforthenodesinLDCmode[5] . . . . . . . . . . . . . 206 7.9 The characteristics of the network while in LDC mode: the BETS protocol isdesignedtoprovidehighenergy-efficiencyforbothEDandCHnodes. . 207 7.10 The cross-layer nature of the BETS protocol: a candidate of choice for the LDCmode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 7.11 BETSfunctionality(EDside): animplementationoftheselfishnodeconcept.216 7.12 An example of regular (no errors) operation of BETS from the CH’s per- spective. AtinactiveMTSs,allnodes(EDsandCH)aresleeping. However, the CH node can still use an inactive MTS for the CH-BS (or CH-Data Server)communication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 7.13 Convergence time assuming that all EDs are turned-on randomly during a 5min-period. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 7.14 Impactofthecommunicationchannelerror(ED-CHlink)ontheBETSnet- workperformance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 7.15 Impact of the communication channel error (ED-CH link) on the energy performance. SimilarscenarioofFig.7.14fordifferentnumberofnodes. . . 238 7.16 CHenergyconsumptionforhomogeneousandheterogeneousschedulings. . 239 7.17 ProposedEnergy-Framework. . . . . . . . . . . . . . . . . . . . . . . . . 240 LISTOFFIGURES xiv 8.1 One of the SoilSCAPE sites: the high degree of network sparsity is clearly visible when virtually transposing the same node topology to the Univer- sity of Southern California campus. These nodes use off-the-shelf 2.4 GHz radio modules (10-18 dBm transmitting power level) and non-rechargeable batteries. No collaboration among nodes is in place and each node directly communicates with the data collector node (LC, marked with a different color). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 8.2 Ripple-2node: theadditionalhardwareandsoftwaremodulesaimingenergy- efficiency can potentially decrease the reliability of the solution. Accord- ingly,aWDT-awaredesignwasadoptedtoallowthenodetoautonomously dealwithproblemswithouthumanintervention. . . . . . . . . . . . . . . . 249 8.3 ExternalWDTschemeusedtoincreasethereliabilityofWSNnodes. . . . 253 8.4 The virtual finite state machine (VFSM) model represents the system be- haviorwithitsexceptionstreatmentassociatedwithWDTactions. . . . . . 255 9.1 MI-WUSN application I: precise soil irrigation. The proposed topology allows the employment of single-axis coils thus reducing the deployment costs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 9.2 MI-WUSN application II: external instrumental (EI) for Pipeline Leak De- tection (PLD) systems. (a) the so-called MI-WUSN-PLD can be employed for existing and new pipeline installations and multiple sensor technology can be employed (b) the combination of special nodes (Coordinator Nodes - CN) and a two-routing technique is proposed as a balanced solution re- garding reliability and energy-efficiency (c) the CN nodes act as sinks thus alleviatingthedataqueuinginlowdata-rateMI-basedsolutions. . . . . . . 261 A1-1Indoor Sub-MHz MI-Soil testbed used for the empirical evaluations in this work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 A1-2InstrumentationandshieldingschemeusedattheSub-MHzMI-Soiltestbed. 275 Abstract Low-power wireless communication in underground settings and confined areas is consid- ered one of the last frontiers in communications. First, there is the energy challenge: be- cause the nodes are embedded in some kind of medium, such as soil, concrete, or debris of a disaster event, many of the traditional energy solutions for communication devices are ruledout. Moreover,typicalradiowave-basedsolutionsforover-the-aircommunicationare significantly impacted in underground settings due to the very strong signal attenuation in lossymedium. To address such challenges, Wireless Underground Sensor Networks (WUSNs) have been proposed. WUSNs potentially enable a wide variety of novel applications mainly in the areas of precision agriculture (PA), concealed electronic fence (border patrol and resi- dencesecurity),anddisastermanagement(landslidedetection,undergroundstructuremoni- toring). Sofar,thehopeisthatthecombinationofcollaborativenetworkingprotocolswitha highnodedensitycanmitigatethementionedissuesofundergroundnodesregardingenergy and signal attenuation. However, the lack of real-world WUSN deployments is a warning thatsomeWUSNaspectsstillneedtobeaddressed. In this research work, the theory and design of a real-world WUSN based on a tech- nique called magnetic induction (MI) are considered. These studies are divided into three main parts. First, a novel soil dielectric model for low frequencies (i.e., 1 kHz to 200 kHz) tailored to MI-based WUSNs is proposed. This preliminary study in than extended to in- clude a novel methodology to identify and separate the electrode polarization (EP) effects fromthedielectricmeasurementsatthesub-MHzrange. Thesecondpartofthisdissertation work is related to physical (PHY) layer of MI-WUSNS: a MI signal attenuation model is developedandimportantdesignstrategiesinvolvingfrequencyandcoiladaptationschemes are proposed. Such solution allows the MI node to dynamically adapt considering energy resources,applicationbandwidth,andsoilconditions. WhilethementionedpartsarerelatedtothePHYlayerofapeer-to-peercommunication system,theconcludingpartofthisworkextendsthesolutionbyincludinguppernetworking layers. Accordingly,across-layerprotocolisproposedasawaytoachieveveryhighenergy- efficiency. In this work, both theoretical and empirical results are considered. Preliminary resultsshowagoodagreementbetweentheempiricalevaluationsandtheproposedmodels. xv Introduction “Inmanywaysthegroundbeneathourfeetisasalienasadistantplanet. Theprocessesoc- curringinthetopfewcentimetersofEarth´ssurfacearethebasisofalllifeondryland,but the opacity of soil has severely limited our understanding of how it functions. As creatures of the aerial world, we have a decidedly distorted view of this nurturing underworld. For ecologists,soilfascinatesandflummoxesinequalmeasure. Thetechniquesandapproaches ofmanybranchesofabovegroundecologydon´ttranslatewelltothesoilenvironment."[6] With the above introductory paragraph, a special issue of the Science mganzine, Soils - the Final Frontier, highlights the challenges in investigating soil ecosystems. Although the article is associated to other areas of Science, those warnings have a remarkable paral- lel regarding underground wireless communication systems. Specifically, current wireless communicationmodelsandtechnologiesmaynotbepromptlyimportedtotheunderground settings without significant penalties in terms of energy, reliability, and costs. In fact, the lackofrobustimplementationsofwirelesscommunicationbetweenlow-powerunderground nodesisanindicationthatundergroundenvironmentcontinuestobeoneofthelastfrontiers incommunications. Thegoalofthisresearchworkistoprovidethetheoreticalfoundations andpracticalguidelinestochangethisscenario. Moreover, while investigating the soil impedance measurements artifacts that occur at sub-MHz frequencies, namely the phenomenon called electrode polarization (EP), it was possible to develop a novel methodology and instrumentation setup based on 2-electrode systems in order to precisely identify EP for diverse frequencies and sample conditions. This achievement potentially has a significant impact on other research areas, such as elec- trochemical,biological,andbiomedicalengineering. 1 INTRODUCTION 2 Background The state-of-the-art approaches in wireless underground communication in general adopt twotechniques: radiowavespropagationandmagneticinduction(MI)[7]. Theformertech- nique is very sensitive to the dynamic characteristics of the environment, in particular the watercontentinsoil[7–10]. Insomecases,theattenuationofthesignalpropagatingthrough soil can reach values higher than 30db/m [10] and, mainly due to this fact, such technique hasbeenconsideredunfeasibleformanyreal-worldapplicationsattheundergrounddomain [11]. If low-power transceivers typically employed in Wireless Sensor Networks (WSNs) are used in underground settings, one can still propose a very high node density, such as 1 nodeforeach1or2m 2 gridarea,asawaytoachieveasolution. Nonetheless,thehighcost and complexity of such solution can still prohibit its practical realization. Therefore, since the beginning of this research project, our objective has been to achievea cost-effective so- lutionforapoint-to-pointcommunicationinvolvingmid-rangedistances(e.g.,15..30m). Wireless Underground Sensor Networks (WUSNs) were introduced as a potential solu- tionforlow-powerundergroundcommunication[7]. Besidestheemphasisoncollaboration among nodes, underground (UG) nodes in WUSNs still need to employ more efficient net- working physical layer (PHY) solutions than the ones adopted in traditional over-the-air (OTA) systems [10, 11]. Accordingly, the majority of the WUSN literature is mainly con- cerned with the PHY-layer. MI-based nodes were introduced as potential candidates for WUSNs because the MI channel has been reported as not strongly impacted by the envi- ronmental parameters and higher distance ranges can be achieved [8, 9, 11, 12]. In this work, we will see that such expectation is true for some cases, but not always. Moreover, many questions regarding MI nodes in WUSNs are still under investigation. One of the main questions is related to the optimum frequency range for MI nodes. Associated to the answer to this question, it is also important to adopt a soil dielectric model for that fre- quencyrange. FortheUHFband(300MHz-3GHz),manysoildielectricmodelshavebeen proposed[13,14]andhavebeenusedinconjunctionwithradiopropagation-basedWUSNs [1, 12, 15]. Unfortunately, soil dielectric models for low frequencies (i.e., sub-MHz range) are more difficult to develop and the few existing studies, such as in [16], do not actually targetwirelessundergroundcommunication. Therefore,thedevelopmentofasub-MHzsoil dielectricmodelisanotheraspectaddressedinthiswork. AswillbediscussedinChapter1,ifweconsidera)mid-rangedistances(i.e.,15..30m), INTRODUCTION 3 b) different soil conditions, and c) energy-efficient MI devices that are expected to operate duringmultipleyearswithoutbatteryexchange,theMIchannelwillbebasicallyconstrained to operational frequencies at the 1 kHz - 200 kHz range (MI systems in soil with distances smaller than 15m may adopt higher frequencies). As a result, the first generation of MI nodes is expected to be tailored to low data-rate applications, at least from the peer-to-peer perspective. Nonetheless,collaborationinMI-WUSNscanstillbeintensivelyusedinorder toincreasethenetworkdatarate,atopictobediscussedinchapter9. Thementionedmid-rangelimitofaround30miscloselyrelatedtothefrequencychoice atlow-frequency(LF)bandsandalsowiththesoilconditions. Nonetheless,higherdistances canstillbeachievedifa)energyconstraintsarerelaxed,orb)biggerdimensionsfortheMI- coils (the antennas, at the MI-WUSN context) are allowed, or c) the soil medium presents exceptional conditions for the MI channel. For the former case, in Chapter 9 the distance of 50m is indicated for pipeline gas/oil leak detection (PLD) systems and such possibility may be achieved with bigger coils with twice the size of the ones mentioned in this work. Inrelationtolatteroption,theadaptationoftheMIsystemaccordingtothesoilconditions, theachievementofanapplication/environmental-awareadaptivesystemisindeedoneofthe maindesigngoalsconsideredinthisdissertationwork. Alternatively, as a way to extend the distance between the MI nodes, the WUSN liter- ature frequently mentions a technique called MI-waveguide which is claimed to extend the communicationlinktohundredofmeters. Inthiscase,passivedevicescalledMI-relaysare employed[11,12]. Nonetheless,MI-waveguidesarenotassumedinthisworkandthejusti- ficationsforthisapproachareprovidedlaterinthischapterandalsoinChapter4. Therefore, inordertoachievethedesignandimplementationofacost-effectiveandrobustMI-WUSN system,thementionedtwohardconstraints-mid-rangedistances(e.g.,30m)andlowdata- rates - are always assumed in this work. The latter constraint will be eventually relaxed in Chapter6bymeansofnoveltechniques. Oneimportantaspectofthisresearchworkisthatnoassumptionisconsideredregarding thesoilconditions(i.e.,soilcompositionandwatercontent). Infact,asignificantnumberof theexperimentsperformedinthisworkarerelatedtosoilswithhighpercentageofclayand waterbecausesuchcombinationisthemostcriticaloneforcommunicationpurposes. Such emphasis on the worst environment conditions is regularly observed in our studies because anexceptionalrobustnesslevelisoneoftheprimarygoalsofMI-WUSNdesigns,inpartic- INTRODUCTION 4 ularduetothehighmaintenancecostsassociatedtoburieddevices. Tothisend,besidesthe PHY-layerbasedontheMItechnology,thelast3chaptersinthisworkarerelatedtodiverse additionalaspectsofMI-WUSNsinvolvingenergy,reliability,andapplication/networking. While an operation at very low frequencies (e.g., 1-10 kHz) can guarantee the com- munication system functionality even in extremes cases (e.g., very wet soil, 30m of dis- tance),thisbandisnotidealintermsofbandwidthandenergy-efficiency,aswillbedemon- strated in future chapters. In order to address this issue and achieve a balance in terms of robustness/energy-efficiency/high-bandwidth, a dynamic frequency-switching is proposed in this work. Two operational frequencies, lower-bound (FL) and upper or higher-bound (FH), are selected according to the distance and circuitry constraints, such as the coil di- mensions and maximum voltage/current levels. The novel FL−FH algorithm is discussed in Chapter 5 where it is shown that this proposed mechanism is paramount to achieve the mentioned balance. For instance, while the soil is very dry, the MI-WUSN can operate at a higher FH frequency favoring higher levels of energy-efficiency and bandwidth. On the other hand, just after a rainfall, the MI-WUSN automatically switches to the lower FL fre- quencyinordertomaintaintheoperationofthenetwork. However, long-term empirical investigation of the variability of soil moisture (SM) in real-world scenarios reveals that the average soil moisture for the mid-range distances con- sidered in this research is relatively stable for long periods of time [17]. Even seasonal changes only drift the average SM value by few percentage points in certain places. Based on this observation, the FL-FH operation scheme may not be the best solution if adopted isolatedly. Specifically,theFLandFH frequenciesareselectedbasedontheworstandbest cases for the soil conditions (e.g., 1% and 40% soil water content, respectively) and not on the typical (time-dominant) soil condition case. For instance, in a scenario where FL and FH are 10 kHz and 100 kHz respectively, the system will not achieve optimum balance if SM is around 15% the majority of the time because the system will unnecessarily operate at 10 kHz potentially using more energy than necessary. Accordingly, it is proposed in this workasolutionforthisscenariocalleddual-wireschemewhichisdiscussednext. While the proposed FL-FH adaptation scheme is potentially essential to achieve effi- cientsolutionsunderavarietyofsoilconditions,thesolutionistypicallyconstrainedbythe selected value for the wire thickness used at the MI coils. This one is an important conclu- sion of our investigations in Chapter 5. Because this design parameter cannot be changed INTRODUCTION 5 once the MI node is deployed, the FL-FH scheme will be typically constrained by very low application bandwidths. Moreover, there is the typical (and non-extreme) soil scenario which was previously discussed that must be considered. To address both challenges, the proposed approach in Chapter 6 is to add the dual-wire scheme to the mentioned FL-FH solution. Thisnewschemeessentiallyreferstotheuseoftwodistinctcoils(oneforFLand otherforFH). However,thechoiceofthewirethicknessforeachcoilisnottrivialandthis discussionisprovidedinChapter6. Thenoveltechniquesdiscussedaboveareonlypracticalapproachesifsignalattenuation expressions due to the soil conditions and frequency are available or they are feasible to be derived. Itiswellknownthattheattenuationofmagneticandelectricfieldsinsoilmedium depend on the complex propagation constant γ of the medium [18] which can be inferred if the complex values of permittivity ε, permeability μ, and conductivity σ are known. Dielectric soil models typically derive expressions for ε and σ considering, at least, the frequency, water content, and soil texture [13, 14, 16]. In general, for non-magnetic soils, μ is assumed to be μ 0 , the permeability of free-space. In this work, we also consider this assumption knowing that magnetic soils are very rare [2] and these few magnetic soils are typically not associated with the target WUSN applications. For sub-MHz range, few soil dielectricworksareavailableandjustonemodel(empirically-determined)isavailable[16]. Therefore, part of the effort in this dissertation work is guided toward to the development ofasub-MHzsoildielectricmodelfordiversesoilconditionswhichisintroducednextand discussedindetailinChapters1and2. Although a sub-MHz soil dielectric model is potentially necessary for practical MI- WUSNs, it is also important to employ a simplified model. Such strategy is typically used for OTA communication when a communication engineer selects the best existing signal attenuationmodelforacertainplacewithoutanalyticallyconsideringthedetailedgeometry and characteristics of the objects nearby the transmitter-receiver pair [19]. Similarly, soil dielectric models that require the knowledge of the soil composition or even the geometric characteristics of the soil particles may not be practical for real-world MI-WUSN imple- mentations. Despitesthelackofaccuracyinrelationtothevalueofγ,itispotentiallymore desirable to adopt a version of a simplified soil model that only have frequency and soil moistureasinputparametersprovidedthattheerrormarginisknownandkeptsmall. Accordingly, 9 soil types with different compositions are studied and 3 classes (or ver- INTRODUCTION 6 sions) of empirically-based soil dielectric models are derived in this work. The procedure is explained in detail in order to allow the repetition of the process for a soil composition radically different from the ones evaluated in this study. Nonetheless, as will be discussed inChapter3,theproposedsoildielectricmodelonlyhasitsaccuracysignificantlyimpacted ifaverydifferentkindofsoilisfoundatthetargetdeploymentsiteandhighervaluesofSM are involved. Even for this case, the conservative selection of a lower-bound frequency LF smaller than the one provided by the proposed model may be enough to guarantee the MI operation. While the MI technique is not impacted at the same level of attenuation compared to propagating radio waves in soil medium [11, 12], the magnetic fields may still be signif- icantly impacted in soils with a higher conductivity and for certain combinations of fre- quency and distance, as shown in Chapter 1. To date, few research effort is reported re- gardingLFsoildielectricmodelsinMI-WUSNs. Infact,theexistingMIsignalattenuation modelsarenotdirectlycoupledtoanyspecificsoildielectricmodel. Moreover,theoptimum operational frequency range for MI-WUSNs has not been proposed in the literature. These two aspects may help to explain the lack of real-world MI-WUSNs despite the number of researchworksinthisarea. Whiledevelopingasub-MHzsoildielectricmodel,onemayfacethechallengesrelated to the EP instrumentation effects. In Chapter 1, these effects are explained and a tech- niqueisproposedtomitigate,thatis,tostronglyreducetheEPeffects. SuchEP-correction algorithm is found to have good accuracy regarding its application to a MI-soil signal at- tenuation model. Nonetheless, a more accurate and generic sub-MHz soil dielectric model can be eventually developed if the EP effects are not simply mitigated, but identified and removedfromthedielectricmeasurementsofthebulkymedium(e.g.,thesoilsample). For instance, when a 4-electrode system is adopted, usually the EP effects may be mitigated but they cannot be measured and the EP data is lost. In our study, the simple (and robust) 2-electrodesystemispreservedandcorrectedmeasurementsforsoildielectricconstantand conductivity are achieved. Moreover, the EP data, which is usually seen as an instrumenta- tionartifact thatmustbeeliminated,ispreserved. EP,thenaturalLFphenomenonthatoccursattheinterfacebetweentheelectrodeandthe medium can be indeed exploited. For instance, the EP impedance data may be potentially important for the development of soil dielectric models that are based on physical interpre- INTRODUCTION 7 tation. Usually such models follow the empirical ones, once they are well accepted by the scientific community. Moreover, when the EP data is completely removed from the dielec- tricmeasurements,thesensingphenomenonwhichisbasedondielectricspectroscopymay notprovideadditionalinformationaboutthemedium,suchasthelevelofchemicalcontam- ination of the soil under analysis. Therefore, in this work we started the foundations of the development of an EP-aware soil dielectric model and a specific methodology and instru- mentationsetupisproposedinChapter2. Whiletheimpactofthismoreaccurateimpedance measurement approach is expected to be relatively small regarding underground communi- cations and the associated MI-WUSN design, the underlying EP-related techniques can be important for other research areas that investigate the sub-MHz dielectric spectroscopy of materialsingeneral. Besides a sub-MHz soil dielectric model and the associated MI-soil signal attenuation model,thisresearchworkalsoprovidesthefoundationfortherealizationofMI-WUSNsin relation to aspects beyond the PHY layer. Once a basic and energy-efficient PHY solution for the peer-to-peer communication between two MI nodes is achieved, the next step is to investigate the challenges associated with the channel reliability, modulation, energy, and networkingaspects. Thelattertwoaspectshavebeenintensivelyinvestigatedduringthelast 4yearsregardingenergy-efficiencyinsparseterrestrialWSNs[4,5,17,20–22]. Itturnsout that, in general, WUSNs are sparse WSNs in the sense that a UG node has typically very few one-hop network neighbors. Unfortunately, energy-efficient networking solutions for such cases are much more difficult to be achieved [22]. To this end, we adapt our current successfulnetworkingarchitecturesolutionforterrestrialWSNs(presentedinChapter7)to tworepresentativeMI-WUSNsscenarios,asdiscussedinChapter9. TargetApplications&State-of-the-Art The idea of burying soil probes for some form of soil measurement (such as soil moisture) is common place. However, regarding WUSNs, one can go one step ahead in order to have the entire node buried, including its antenna and battery, thus enabling a wide variety of novel applications [7]. Such scenarios involving the potential use of MI-WUSNs occur for atleastthreemainreasons: • No physical obstruction: Aboveground nodes can easily disrupt the movement of people and machinery. Accordingly, irrigation and soil monitoring systems for crops INTRODUCTION 8 and sports fields are potentially the main WUSN applications. In fact, one of the potential ways to mitigate the global water scarcity problem is by optimizing the ir- rigation systems [23]. This goal can be potentially achieved if a higher soil probe densityisemployed,agoalthatWUSNscanfulfillifthesolutioniscost-effectiveand doesnotimposeanykindofobstructiontotheregularfarmingactivities[10]. • Physicalprotection,concealment: Insomeharshnaturalenvironments,theweather, animals, flooding, fire, etc. can destroy the aboveground environmental monitoring devices and the soil medium can serve as a physical protection if the nodes are fully buried. For instance, the long-term effects of the practice called prescribed burn can be better understood if the monitoring equipment is physically protected before, dur- ing, and after the experimental fire event. This is the case if the devices are placed at the soil subsurface where the fire does not have a strong impact. Considering an- otherscenario,themainthreatcanbepeoplethatcanstealordestroytheaboveground equipments. Inahypotheticalcasementionedin[22],thegoalistocontrolthequality of the sand of a certain beach in a long-term project. If a WUSN is employed in this case, the concealment of the UG nodes can be perceived as an effective way to con- ciliate the project goals with the mitigation of vandalism issues. The same rationale alsoappliestootherscenarios,suchasalandslidemonitoringsysteminstallednearby areaswherepeoplemaystealtheequipments. • Non-optional (or required) medium immersion: In some cases, the nodes are im- mersed in the lossy medium due to the application circumstances and this arrange- mentisnotoptional. ThemostimportantWUSNapplicationinsuchcasesisdisaster management,wherethenodesneedtocommunicateunder-the-debris[24]. Although this scenario is not exactly the one evaluated in this research work, MI-WUSNs for soilscan potentially serveas the foundation for post-disaster sensing/communication solutions in collapsed buildings, mines, and similar structures. In another distinct scenario called infrastructure monitoring, the nodes are deployed inside the concrete blocks due to the needs of the monitoring applications [25]. Again, MI-WUSNs for soilcanserveasthebasisforthecommunicationsolutioninsuchscenarios. Nonethe- less, the design opportunity to be exploited in this specific case involving concrete medium is the relatively stable communication channel compared to the one in soil which can significantly vary according to the water content. On the other hand, the existence of metallic parts inside the concrete blocks nodes pose additional concerns regarding the additional attenuation factor to be considered, this one mainly due to INTRODUCTION 9 Fig. 1 Taxonomy of Wireless Communication in Underground or Confined Areas (WCUCA).[1] theeddycurrentsformedatthemetalpiecesclosertothetransmittingcoil. While the main applicability of this work is on mid-range and low-power MI-based WUSNs deployed in soil medium, it is worth mentioning how the proposed solution fits in a bigger class of wireless systems called Wireless Communication in Underground or Confined Areas (WCUCA) which has the taxonomy shown in Fig.1. For a long time, the mining industry has been looking for effective wireless solutions to help trapped miners [1,26]. SomeoftheseoptionsarecategorizedasThrough-The-Earth(TTE)techniquesthat are point-to-point solutions and typically they are not low-power. Wireless communication techniques for networks located in underground and confined areas have been investigated in different contexts and not always the soil is the actual communication medium. For in- stance, many wireless networks used for mines and tunnels actually employ over-the-air communication, although this scenario has its own set of particular challenges that are not foundinregularOTAandWUSNscenarios. Notethatthedepthofthenodeisanimportant forsometypesofWUSNs,specificallyWUSNsbasedonradiowavepropagation[10]. The effect of having the soil-air interface close to the source and receiving points is critical for radiowavesanddifferentkindsofantennascanbeused[27–29]. However,aswillbeshown inChapter3,theproximityofaTXorRXcoiltothesoil-airinterfacehasnegligibleimpact ontheMIwirelesschanneliflowfrequenciesareemployed. Therefore,thesameMIdevice isexpectedtobetransparentlyemployedinTopsoil,Subsoil,orDeepSoilWUSNs. Surveys on WUSN-related work, including the radio wave propagation technique, are provided in [1, 7, 30]. Moreover, there is an impressive amount of ongoing research re- INTRODUCTION 10 garding Wireless Power Transfer (WPT) which also employs the MI technique for power transfer purposes, not data communication. While part of the information in WPT is com- mon to our studies in MI-WUSNs, it is important to highlight that there are at least two main differences to be considered. First, the resonant circuits at the RX side are distinct: thegoalofWPTistomaximizethepowertransferbymatchingimpedances[31,32],while the goal of MI-WUSN is to maximize the induced voltage for proper data decoding. The detailsofthisdiscussionareprovidedinChapter3. Second,manyWPTsolutionsrequirea certain level of physical proximity between the sender and receiver devices. In such cases, thecircuittheoryofinductively-coupledsystemsconsidersthestrongeffectsoftheRXcir- cuitryimpedanceontheTXcircuitryimpedanceandvice-versa. Infact,themajorityofthe MI-WUSN papers also use the same approach. Nonetheless, for mid-range MI-WUSNs, theTXandRXcircuitsaretypicallyverylooselycoupled andthisfactallowsonetodetach theeffectsoftheRXcircuitonTXsidewhichcansignificantlysimplifytheanalysisofthe problemsinmid-rangeMI-WUSNs. Thesoundnessofthisnovelapproachproposedinthis workisevaluatedinChapter3. In Radio Frequency Identification (RFID) systems [33], both WPT and data transfer aspects are considered at the final free-space solution. However, for RFID systems the involved distances are typically much smaller than the distances for mid-range WUSNs. Another research area which shares some aspects with MI-WUSNs (e.g., low operational frequency, lossy medium) is Through-The-Earth (TTE) technology, as shown in Fig. 1. Nonetheless,besidesbeingapeer-to-peersolution,theenergy-efficiencyisnotsocriticalin TTEsystemsandthedistancesaresignificantlyhigher(e.g.,>200m)comparedtoWUSNs. Therefore,TTEdevicestypicallyhaveaveryhighpowerconsumptioncomparedtoWUSN nodes. Accordingly, in pre-disaster scenarios, energy harvesting is regularly employed in TTEdevices,whichisrarelyanoptionforWUSNnodes[34]. The focus of the remaining part of this section is on the related work in MI technology applied to the wireless communication in lossy environment: soil, water, and fresh (human oranimal). In1997, unidirectionalandhigh-powerMIcommunicationat3kHzwasinves- tigated for military operations in coastal regions [35] and successful results were reported at data rates of up 300bps. This work anticipated what would be highly expected for mid and long-range MI communication in lossy medium: low frequency and small bandwidth. In 2009, the same company behind the work in [35] commercialized a relative low-power device,MIRemoteActivationMunitionSystem[36],whichachievescommunicationrange INTRODUCTION 11 of 200m through soil, rock, vegetation, and water. To compare the distinct MI frequency ranges regarding lossy and lossless media, consider the following case while remember- ing the previous one which operated at 3 kHz. In 2001, the advantages of using MI for wireless close-proximity applications (e.g., <3m), such as those typically employing Blue- tooth technology, were highlighted in [37] and the 11-15MHz carrier is proposed for such MI communication systems. Observe that such higher frequencies for MI are only feasible for low-power devices if they communicate at very short distances, as will be discussed in Chapter3. In2002,theconceptofMIwaveguidesisintroducedin[38,39],althoughnotassociated specifically with data communication solutions. In this approach, resonant coils (passive devices) are included between a pair of transmitting and receiving coils (active devices) in order to reduce the signal path attenuation. In 2006, MI is proposed for the first time as a promising alternative to radio wave propagation technique in WUSNs [7]. Note that from that moment on, the MI-waveguide technique is very often considered at the WUSN liter- ature. In 2007, the MI technology was highlighted as a potential option for underground communications and a prototype involving a 70-inch distance is presented in [40]. At the same year, MI was also considered for communication between implantable devices in hu- manbody[41]. In2009,thefirstcommunicationmodelforMI-WUSNsisproposedin[12,42]andboth the regular MI solution (i.e., a pair of TX and RX coils) and the use of MI-waveguides are consideredundertheassumptionthatthemagneticfieldatthenear-fieldregionisnothighly impacted by the soil conditions, including the soil water content (WC). Simulated results at 300 and 900MHz are presented and it is reported that a regular MI system has similar performance compared to an EM-based solution, being the latter a better or a worse option accordingtothesoilWC.Moreover,withtheuseofMI-waveguides,itisreportedasignifi- cantpathlossreduction(almost80dbfor5.5mat300MHz). Atthesameyear,aMIreceiver solution deployed as an ingestible pill for cattle is reported in [43]. Operating at 125 kHz and using ferrite core, communication ranges of up 20 inches are achieved. An interesting aspect of this work is the comparison of the MI channel in free space and in a lossy saline solution: basically the same signal attenuation levels. We will see in Chapter 3 that such apparentmedium-transparencyofMIchannelsreportedin[12,42,43]typicallyonlyholds foracombinationofsmalldistances,lowfrequencies,andmedium/lowconductivitylevels. INTRODUCTION 12 In 2010, the model in [42] is evaluated at 10MHz in [11]. In this work, it is still as- sumed that soil moisture and composition do not significantly affect the MI channel. Bit ErrorRate(BER)andbandwidthanalysisareperformedconsideringcoildesignaspects. It isconcludedthatwhiletheMIwaveguidetechniquecangreatlyreducethepathloss,itdoes not impact the bandwidth of the system compared to an ordinary MI solution. At the same year, the MI-waveguide model in [42] is theoretically studied for an underground pipeline monitoring application where the sensing points are deployed at the external surface of the underground pipeline in [44]. The frequency of 10 MHz is considered in this study and different design parameters for the MI coil are evaluated. An inter-node distance of 250m isachievedwitharelativelysmallcoilradius(0.15m),0.01ohms/mwire-resistivityforthe coil, and MI-relays placed every 5m (a total of 49 MI-relays). The achieved bandwidth is limitedto1kHzanditcouldincreaseto2kHzifthespacingofrelaysischangedto4m(61 MI-relays). It is concluded that while a higher number of MI-relays is necessary for non- metallicpipelines,whenmetallicpipelinesareemployed,veryfewrelaycoilsarenecessary because the pipeline itself can provide the magnetic core for the MI waveguide. Nonethe- less, the loss due to eddy-currents at the metallic pipelines is not addressed in that work. An important aspect mentioned in the same work, and a critical one for MI-coil designs, is thevalueofthecoilresistance. Ahighervalueisclearlyassociatedwithundesirableohmic lossesatthecoils. However,averylowvalueforsuchresistancecanbetranslatedinavery high Q (merit factor) for the RLC tank thus resulting in very low bandwidth even if larger carrierfrequenciesareemployed. ThisdesignaspectwillbediscussedindetailinChapters 5and6. Also in 2010, a similar WUSN-MI scenario involving steel pipes is discussed in [45], where the solution is expected to be used to control pumps in city heating systems. The pipelines are metallic and a theoretical model for this scenario is developed and simulated results are provided. A low frequency of 3 kHz is used at the design and simulations. The maximum coverage varies from 20 to 40m for a total path loss of 100dB. More recently, in [46], the authors presented the optimal design for transmitter and receiver coil antennas for the system proposed in [45] at 3 kHz, along with a discussion on the advantages and disadvantagesofusingthemetallicpipeascoreforthecoils,afterwhattheauthorsdecided toward air-filled core coils. An interesting aspect to be highlighted in that work is the im- portanceofchoosinglowfrequenciesformid-rangeMI-WUSNs. INTRODUCTION 13 The MI waveguide technique can allow more than one topology in a network. For in- stance, a MI-relay coil can be used by more than one pair of MI transmitter/receiver. With thisobservationinmind,differentdeploymentarrangementsareproposedin[47,48]along with simulated results. In [47], networking aspects of MI-WUSNs are addressed, although heavily based on the MI waveguide feasibility assumption. In [48], an additional MI com- munication model, also using waveguides, is proposed and three excitation methods are investigated. Itisimportanttohighlightthat,althoughmanyWUSNpapersdiscusstheMI- waveguide technique, to date, real-world implementations of MI-WUSNs with MI-relays have not been demonstrated and the topic remains open to further investigation. An excep- tionistherecentindoorMI-WUSNtestbedproposedandimplementedin[49]. In 2012, rescue system for miners (point-to-point communication) based on MI is pre- sented in [50] and this work is potentially one of the first practical implementations of a MI-node. Previously, the same authors had achieved success in implementing underground solutions also involving the MI technology but for localization purposes, not for conven- tional data communication. In that work, the magnitudes of magnetic field of underground nodes with triaxial coils are measured [51, 52]. In [50], a communication range of 35m is empirically achieved in free-space at 2.5 kHz, with a relative low-power device, and em- ploying a novel triaxial antenna for the MI node. At the same work, it is stated that the expectation regarding rock medium is a range of 30m. With such novel antenna, it is pos- sible to model the MI node as an omnidirectional communication device, which is not the casewhenasinglecoilisused,thusfacilitatingnetworkdeploymentstrategies. Itisalsoim- portanttohighlightthatthephysicaldimensionsofthementionedantennaisaround30cm 3 , which is relatively large compared to antennas of traditional WSN nodes. Also in 2012, in [53],thenetworkcapacityofaMI-WUSNemployingMI-relaysisinvestigatedconsidering theimpactsofnodefailure,MI-relaymissinganddisplacement. TheMItechniquehasalsobeenconsideredforcommunicationinlossymediaotherthan soil. In 2012, a theoretical communication model involving MI nodes is proposed in [54] for underwater communication networks (UWCNs). One of the differences in relation to WUSNs is the need of considering a 3D scenario rather than typical 1D/2D WUSN cases. Because the nodes are expected to be deployed at sea water, the model considers the losses duetotheeddycurrentsnearbythetransmittingcoilthatareduetothehighconductivityof the medium. That work concludes that wireless communication using MI nodes is possible in this scenario for operating frequencies between few and tens of kHz. Later, in chapter 3, INTRODUCTION 14 it is shownthat such frequencyrange is also appropriate for the lower-boundfrequency FL in mid-range MI-WUSNs considering the worst soil scenario (wet soil with relatively high conductivity). AnotherareaofpotentialapplicationoftheMItechniqueisBodyAreaNetworks(BANs), where the MI nodes are deployed close to the body or inside it, depending on the BAN ap- plication. In2013,theNearFieldMagneticInductionCommunication(NFMIC)isreported as the appropriate PHY layer for BANsin [55]. Once MI systems typically havelow band- width,thisworkproposescooperativecommunicationschemesinordertoachieveahigher networkcapacityinBANs. Also in 2013, important conceptual adjustments at the WUSN literature are reported in [56]. Potentially, the most important one related to the initial assumption that the MI chan- nelisnothighlyimpactedbythesoilproperties. Accordingly,anewMImodelisproposed which takes into account the losses due to the electrical properties of the soil medium. A second concept that was reviewed is that MI-waveguides indeed impact the bandwidth of the solution. A third highlighted aspect in that work is the impact of the resonant capacitor onthecapacityperformanceofthesystem. Morespecifically,itisshownthatastheselected carrier frequency increases, the value of the resonant capacitor approaches the parasitic ca- pacitance of the coil. Similar approach will be analyzed in Chapter 3 with the addition of otherfactors,suchastheimpactoftheskindepthresistanceandpotentialinefficienciesdue totheproximityeffectsinmultilayercoils. In 2014, the addition of the channel estimation feature in MI-WUSNs is proposed in [57]. In this case, explicit feedback from the receiver node, related to the channel con- ditions, is not necessary. Such technique can be used, for instance, to properly trigger a frequency-switching solution for WUSNs according to the soil conditions. Such kinds of mechanismsarealsoconsideredinthisdissertationwork. InChapters5and6,itisproposed an alternative scheme where both frequency and coil-switching schemes are used. More- over, if soil moisture sensors at different depths are employed at the MI node, it is possible for this node to forecast the channel conditions. Such feature is based on the observation thatchangesatthesoilmoisturetypicallyoccurinarelativelyslowrate[17]. Whenoneanalyzestheprevioussurveyinvolvingmorethanadecadeofresearchworks inMI-WUSNs,itbecomesapparentthecurrentlackofempiricalworkinthisarea. Besides INTRODUCTION 15 their validation role, preliminary real-world experiments are also an important source of valuable information regarding hidden challenges in WUSN deployments. Therefore, the lackofsuchexperimentscanpotentiallyimpacttherapidproliferationofMI-WUSNs. One exception, previously commented [50], provides some initial feedbacks of typical charac- teristics of mid-range MI-WUSNs, at least for the first generation of MI devices: low fre- quencies, large coils, and tiny bandwidths. Another missing aspect observed at the WUSN literature is related to the operational frequency range for MI-WUSNs. The recent work in [57] reports realistic upper frequency boundaries potentially smaller than 1MHz. The sameworkalsohighlightsthepossibilityofadoptingafrequency-switchingschemeforMI- WUSNs. Nonetheless,thesimulationsatthepreviousmentionedworkstypicallyarebasedontwo extreme cases for soil conditions (dry/wet) with specific values for the soil electrical prop- ertiesreportedforradioandmicrowavebands. Therefore,animportantopenresearchtopic is the development of a specific soil dielectric model for MI-WUSNs. Such model must be tailored for the expected low frequencies (LF) in MI-WUSNs. Unfortunately, a proper LF soil model is not available mainly due to the instrumentation challenges at this frequency band. Once such model is available, potentially a more accurate answer in relation to the exact frequency range for MI-WUSNs can be achieved. In this dissertation work, we fol- lowedthesesteps. Another open research topic at the WUSN area is related to energy aspects and eco- nomical ways to deal with this challenge. Energy harvesting in WUSNs is an option but it is strongly constrained in many ways in comparison to traditional methods for indoors and outdoors WSN nodes [34] and it is not expected prompt solutions in a short period of time. Althoughtheadoptionofwirelesspowertransfer(WPT)technologyisfrequentlyproposed for BANs in order to recharge the energy of nodes installed inside the human/animal body, this method if adopted in WUSNs may not be used very frequently due to the human costs involvedin physicallyvisiting eachnode location. Another approach towardthe mitigation oftheenergyissueinWUSNsistoexploitthefactthattheUGnodeshavestaticandsparse topologyandradicallychangethenetworkoperationincomparisonwithtraditionalWSNs. A first step toward this goal is our work in [5] and its adaptation to mid-range MI-WUSNs isdiscussedinChapter7. INTRODUCTION 16 ResearchObjectives The main objective of this thesis is to analyze the key-factors behind the current non- proliferation of WUSNs and to propose the design of a low-cost, reliable, and energy- efficient WUSN solution for mid-range distances (e.g., 15 to 30m), particularly robust to changesatsoilsettings,andwithamultiple-yearlifetime. The second objective of this thesis is to analyze the factors that mainly influence the electrode polarization (EP) effects on sub-MHz dielectric measurements in general (and in soil samples, in particular) and to provide the initial methodology and guidelines for instrumentation apparatus in order to identify, rather than to eliminate, the EP effects on 2-electrodemeasurementsystems. Regarding the first objective, the theory and design of a real-world WUSN based on the MI technique are discussed and divided into three parts: MI-Soil dielectric model (Part I), PHY layer for MI-WUSNs which includes the MI-Soil signal attenuation model (Part II), and Energy and other aspects for MI-WUSNs (Part III). At Part I, a novel soil dielec- tric model for low frequencies (i.e., 1 kHz to 200 kHz) tailored to MI-based WUSNs is proposed. At Part II (Chapters 3 to 6), a simplified MI signal attenuation model is devel- oped considering the mentioned soil model and also circuitry aspects of MI systems. Next, the proper frequency range for the operation of MI-WUSNs is studied and the algorithms to implement an adaptive frequency and coil-switching operation are provided, as well the guidelines for the coil design. Finally, at Part III (Chapters 7 to 9), the networking aspects, beyondthePHY-layer,arealsoaddressedinordertoachievearobustandpracticalenergy- efficientsolution. Regardingthesecondobjectiveofthisresearchwork,theEPeffectsareinitiallyconsid- ered in Chapter 1 and studied in detail in Chapter 2. E. Warburg (1899) is usually recog- nized as the first investigator to identify the existence of an electrode-electrolyte interface impedance in [58]. Such apparent instrumentation artifact distorts the dielectric analysis of samples at sub-MHz frequencies. In 1937, H. Fricke and H. Curtis developed in [59] a mathematically accurate method to remove the so called electrode polarization effects from the instrumentation measurements by means of what was called later by H. Schwan (1963) as the electrode distance variation technique in [60]. Unfortunately, this technique was not widely adopted because it requires that the EP impedance to be the same for two INTRODUCTION 17 measurements involving different distances between the electrodes and this is the open re- search problem which is addressed in this part of the dissertation work. Various techniques toeliminatetheEPeffectsarediscussedandanovelwaytoidentifyandisolatetheseeffects fromthebulkdielectricmeasurementsisprovidedinChapter2. Thepreliminaryresultsare analyzedwithparticularattentiontoacriticalkindofsoil,namelyNHS-SAT.Thissoilsam- ple was removed very close to a lake and it has a very high conductivity level, even higher thantypicalfertilizedsoilsincropareas. Therefore,thissoilispotentiallyagoodrepresen- tative for the worst scenario in MI-WUSNs, in particular considering this soil in saturation condition. These results initially validate the proposed EP-identification methodology and allowthecontinuationoftheworkinordertoachieveamoreaccuratesub-MHzsoildielec- tricmodelthatcanbeemployedforbothcommunicationandsoilsensingpurposes. Soil Dielectric Model for Wireless Underground Sensor Networks Op- eratingatSub-MHzRange Thisresearchworkstartswiththeinvestigationofpossibleexistingsub-MHzsoildielectric models. Sofar,suchmodelforMI-WUSNsisnotavailableandtheexistingsignalattenua- tion models for these networks do not capture soil medium electrical parameters or require the specific knowledge of the values for them, such as permittivity (ε) or conductivity (σ), which is a challenge in the design of practical, real-world, solutions. The model provided in this work is an empirical one and it is tailored to 3 representative classes of soils based on an investigation of 9 distinct soil samples. The input parameters of the model are soil class, frequency, and volumetric water content. The procedure is explained in detail and it adoptsrelativelow-costinstrumentation(2-electrodeimpedanceanalyzer)inordertofacil- itatetheeventualextensionofthemodelinrelationtosoilcompositionsnotcoveredbythe 3mentionedclassesofsoils. Method of quantitative identification of the impedance due to the elec- trodepolarizationeffectsonsub-MHz2-electrodemeasurementsystems Chronologically,thisisthelastpartofthisdissertationresearchwork. Afterconcludingthe initial sub-MHz soil dielectric model discussed above, it became apparent that such model could still be enhanced in order to increase its accuracy in particular for frequencies below 10 kHz. Nonetheless, the main challenge was to find a way to eliminate the EP effects. INTRODUCTION 18 To this end, around 1,200 experiments are performed with different materials and shapes for the electrodes. During these tests, the capabilities of the EP effects as an natural and embedded sensor revealed to be a relevant surprise. For instance, even a very small trace of soil (e.g., 0.1% of the total volume) close to one of the electrodes in a dielectric cell containing deionized water can be easily detected by means of an abrupt increase of the EP impedance. However, if typical EP-elimination techniques are employed, such as 4- electrode systems and the adoption of Platinum-Black electrodes, such detection capability is essentially voided. Moreover, while these EP-elimination techniques are popular regard- ingfluids,thereareevidentissuesintheirapplicationtosoilsamples,inparticularregarding the repeatability of experiments, as will be discussed in Chapter 2. These two aspects are themainreasonsbehindthedevelopmentofatechniquetoidentify,ratherthantoeliminate, the EP impedance in dielectric measurements at sub-MHz range. Besides the investigation of historical attempts in reducing the EP effects, a particular attention wasgivento the pre- cise method of distances proposed by Fricke. Apparently, few works were successful in employing this method and we tried to identify what are the necessary conditions to apply the method. Besides the Fricke’s works, three additional studies are fundamental for this investigation: in1965,H.SchwanandJ.Maczukpublishedannovelinvestigationaboutthe linearitylimitsoftheEPimpedance[61]. In1974,C.GabrielliandM.Keddaminvestigated thegenerationofharmonicswhenahighersinusoidalvoltagelevelisemployedinasystem where EP is present [62]. In 1984, B. Onaral, H. Sun, and H. Schwan published a detailed study about the linear and non-linear electrical properties of bioelectrodes [63]. Nonethe- less, to the best of our knowledge, no attempt was made regarding correlating the 1937 Fricke’s work with the generation of harmonics due to non-linear EP effects. This correla- tion is actually the foundation of the solution achieved in this research work and presented inChapter2. Magnetic-Induction(MI)SignalAttenuationModelforWUSNs It is proposed a novel MI signal attenuation model for mid-range MI-WUSNs which is based on the mentioned soil model. It is a simplified model in the sense that the transmis- sion (TX) circuitry of one node is separated, for analysis purpose, from the reception (RX) circuitry at another node. Note that it is assumed that the same coil can be used for TX and RX roles. Such novel approach of detaching TX and RX circuits is only acceptable at the Circuit Theory for very loose-inductive couplings, where the mutual inductance M (alternatively, the coupling factor k) has a very small value, which is the case for the inter- INTRODUCTION 19 node distances considered in this work. The error analysis due to this approach is provided and it is demonstrated in Chapter 3 that it has negligible impact on the signal attenuation. The main advantage of this simplified signal model is the fact that the input parameters are directly associated with the circuitry design ones. Moreover, the model facilitates the separation between soil path losses aspects from circuitry capabilities or inefficiencies. In particular, the analysis of the resonant frequency deviation impact is strongly facilitated in thismodelcomparedtotheuseofexistingMI-WUSNmodels. Theproposedsignalattenu- ationandthesoildielectricmodelsareconvenientlyembeddedinagraphicalusersoftware interfaceinMATLAB(TheMathWorks,Inc.). MI-basedPhysicalLayerforPoint-to-PointCommunicationinSoils At this part of the research, the mentioned signal attenuation model is intensively used to answer the question about the best frequency range for the operation of mid-range MI- WUSNs. It is shown in Chapter 4 that the answer to this question depends on both soil and circuitry aspects. The lower-bound frequency, FL, is mainly governed by the worst soil scenario (relative high conductivity and wet soil). On the other hand, the upper-bound frequency,FH,considersthebestsoilscenarioanditisshownthatthefrequencyboundary in this case is mainly constrained by circuitry aspects, such as the AC resistance of the coil wiring,whichistypicallyneglectedattheWUSNliterature. OnceFLandFH arefoundby theproposedalgorithminChapter5,itisshownthatsuchsolutionisstillmainlyconstrained by the choice of the wire thickness adopted for the coil. Such choice will be essentially a decision between energy and application bandwidth performances. However, when the soil is not at extreme dry/wet conditions, the penalties due to the choice of the wire thickness willbepermanent,despitestheadvantagesoftheadaptivefrequencyscheme. Therefore,in Chapter 6, an additional mechanism is employed to solve the issue: the dual-wire scheme. This approach is basically the adoption of 2 distinct coils, one designed for FL operation and the other one for FH. As a result, the MI node can properly decide in operating with one of the 4 modes that result from combining 2 frequencies and 2 wire thicknesses for the coil. The solution is demonstrated to be an efficient way to balance energy-efficiency, applicationbandwidthneeds,andenvironmentalconditions. EnergyManagementandNetworkingforMI-basedWUSNs It is proposed a cross-layer networking solution called Best-Effort Time-Slot allocation (BETS) protocol for sparse and very energy-efficient terrestrial WSNs in order to achieve INTRODUCTION 20 a long-term, multiple-year, lifetime for the solution. Fortunately, this already time-tested architecture [64] can be adapted to the underground scenario with few modifications. In Chapter 7, it is shown that typical WSN networking protocols with more than 5% of net- work overhead causes a dramatic impact on the battery lifetime of nodes if low duty-cycle applications are used, which is typically the case for MI-WUSNs. Moreover, while homo- geneous energy consumption among the nodes is a very desirable feature for WUSNs, it is stillanopentopicinWSNs. AnadaptedversionofBETSforMI-WUSNscanprovidesuch feature and also a very small overhead for WUSNs at the cost of hardware and software modifications,someofthemdetailedinthiswork. Withtheadoptionofsuchmodifications, theriskofhavingalessreliablesolutionincreasesduetoitshighercomplexityandthissce- nario is particularly critical for buried MI nodes. Therefore, in Chapter 8, it is shown how toincreasethereliabilityofsuchsolutionsbymeans ofhardwareandsoftwareengineering techniques. Finally,toillustratehowthementionedmodificationsoftheBETSprotocolfor different MI-WUSNs can be achieved, two applications are investigated in Chapter 9: a) precise crop irrigation and environmental monitoring and b) pipeline gas/oil leak detection (PLD). ThesisOutline This research work is organized as follows: in Chapter 1, a soil dielectric model for low- frequencies and tailored to MI-WUSNs is investigated. A deeper investigation of the elec- trode polarization (EP) effects on dielectric measurements and a methodology to identify these effects is proposed in Chapter 2. Next, based on the proposed soil model of Chapter 1, a signal attenuation model for mid-range MI-WUSNs is developed in Chapter 3 and it is validated with preliminary outdoors experiments. With this signal (or MI channel) model, itispossibletoinvestigatethelower-boundfrequency(FL)ofaMIsystemconsideringthe expectedworstsoilscenario(andalsothebestcasewithFH frequency)andthisdiscussion is presented in Chapters 4 and 5. In order to guarantee a balanced solution with any soil condition,thedual-wireschemeisalsoproposedforMI-WUSNdesignsanditispresented inChapter6. Atthethirdandlastpartofthisresearchwork,thePHYlayerproposedforMI-WUSNs evolves from a robust peer-to-peer, low-power, and mid-range MI communication solution for soils to a networked one in Chapters 7, 8, and 9, where energy, reliability, and higher networkingaspectsareconsidered. PartI Sub-MHzSoilDielectricModel 21 Chapter1 MI-Soil: SoilDielectricModelfor Sub-MHzWUSNs At the studies of the electromagnetic wave (EM) propagation in soil, including the mid- range MI-WUSN applications, three properties of the medium are important: dielectric permittivity,electricalconductivity,andmagneticpermeability[2,18]. Byassuminganon- magnetic soil (which is usually the case, according to [2], pp. 365), a soil dielectric model typically refers to the first two of the mentioned soil properties. A key characteristic of the MI-based communication in comparison with the radiation-based one is the smaller sensi- tivitytothechangesatthesoilsettings[11,12,50]oftheformer. Forinstance,theexistence of big and high-dense rocks between two MI nodes may not have a huge impact for the MI communicationperformancewhichisnotthecaseforradiowavepropagation. Nonetheless, dependingontheparametersofthescenario(frequency,distance,soilsettings),particularly the water content, the MI communication can still be forbidden or severely impacted. Ac- cordingly, the main goal of this chapter is to provide a proper soil dielectric model that can be used to estimate the soil path attenuation for the magnetic field in MI-based WUSNs consideringdifferentsoilsettingscenarios,althoughconstrainedtolowerfrequencies(LF), e.g.,1kHz-200kHz. Although a significant number of soil dielectric models have been proposed targeting relatively higher frequencies (i.e., >1MHz), apparently no research has been done in rela- tion to a LF soil dielectric model associated to wireless underground communication. An exceptionistheempiricalmodelprovidedin[16]whichmaystillpresentdistortionsdueto thepotentialEPimpedanceeffectsatthosesoilmeasurements. Therefore,thelossesdueto conductivity and polarization effects of soil mixtures at sub-MHz range are investigated in 22 CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 23 thischapterandasimplifiedsoildielectricmodelforMIcommunicationisproposed. 1.1 IntroductiontotheElectricalPropertiesofSoil Consideratime-harmonicelectromagneticwavewhichpropagatesinalossymedium. From firstprinciplesinelectromagnetics,thefollowingMaxwell-Ampereequationinitsdifferen- tialformisthenusedtostudytheproblem[18]: ∇×H=J i +J c + jωε ∗ E (1.1) whereJ i istheimpressedcurrentdensity,J c istheeffectiveelectricconductioncurrentden- sityandtheimaginarytermistheeffectivedisplacementelectriccurrentdensity. Considering the soil as a source-free medium, that is, J i = 0, expanding the complex permittivityε ∗ ,andapplyingtherelationJ c =σ s E,whereσ s isthestaticfieldconductivity ofthemedium,wehave: ∇×H=σ s E+ jω(ε ′ − jε ′′ )E (1.2) ∇×H=σ s E+ jωε ′ E+ωε ′′ E (1.3) ∇×H=(σ s E+ωε ′′ )E+ jωε ′ E (1.4) wherethetermωε ′′ correspondstothealternatingfieldconductivity. Wewillseelaterinthischapterthattypicaldielectricmeasurementinstrumentationper- ceiveslossofthematerialundertestasawhole,notdifferentiatingohmiclosses(duetothe conduction currents) from pure dielectric or polarization losses. Therefore, it is convenient tointroducetheeffectiveconductivityσ e as[18]: σ e ,σ s +ωε ′′ (1.5) Applying(1.5)into(1.4),wehave: ∇×H=σ e E+ jωε ′ E (1.6) Assuming a non-magnetic soil (μ =μ o , where μ o is the permeabillity of vacuum), ho- mogeneous, and unbounded medium, we can use (1.6) to solve the traditional wave equa- tions[18]andintroducethecomplexpropagationconstantγ: CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 24 ∇ 2 E= jωμ o σ e E−ω 2 μ o εE=γ 2 E (1.7) ∇ 2 H= jωμ o σ e H−ω 2 μ o εH=γ 2 H (1.8) γ 2 = jωμ o σ e −ω 2 μ o ε ′ (1.9) γ 2 = jωμ o (σ e + jωε ′ ) (1.10) γ = p jωμ o (σ e + jωε ′ ) (1.11) Letγ =α+ jβ andusingitin1.11: α =ω p μ o ε ′ 1/2 r 1+( σ e ωε ′ ) 2 −1 1/2 (1.12) β =ω p μ o ε ′ 1/2 r 1+( σ e ωε ′ ) 2 +1 1/2 (1.13) where α, given in Np/m, is the attenuation constant and β, given in rad/m, is the phase constant. The attenuation constant α is zero for lossless medium, which is not the case of soil. Because α is related to the attenuation of the electric field E or the magnetic field H, it is possibletoconvertNp/mindB/masfollows[18]: |α(Np/m)|= 1 8.68 |α(dB/m)| (1.14) Typically, at the electrical specifications of materials, a value for the effective loss tan- gent tanδ e , also called Dissipation Factor (DF), is provided. DF or tanδ e represents the degree of losses of the material. When tanδ e ≫ 1, the material is considered lossy and when tanδ e ≪1, the material is considered lossless. The definitions forδ e are given below [18]: CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 25 tanδ e , σ e ωε ′ (1.15) tanδ e =tanδ s +tanδ a (1.16) tanδ e = σ s ωε ′ + ωε ′′ ωε ′ (1.17) tanδ s = σ s ωε ′ (1.18) tanδ a = ωε ′′ ωε ′ = ε ′′ ε ′ (1.19) where tanδ s (dimensionless) is the static electric loss tangent and tanδ a is the alternating electriclosstangent. It was already commented that, in general, the dielectric instrumentation cannot differ- entiatebetweenohmicanddielectriclosses. Similarly,itcannotdifferentiatebetweentanδ s and tanδ a . Therefore, the effective loss tangent tanδ e is the parameter which we usually havetodealand,forconvenience,ausefuldefinitionfortanδ e isprovided: tanδ e , ε ′′ e ε ′ (1.20) TheimportanceofthesenotationsforourMI-WUSNresearchisexplainedasfollows: in ordertostudyasoildielectricmodelatlowfrequencies,aninstrumentsuchasanimpedance analyzer is employed. In the case of soil samples, typically the measured parameters are C (capacitance) and DF (dissipation loss, which is the same as loss tangent). Given the phys- ical parameters of the soil sample holder, the value of ε ′ is calculated. Alternatively, the dielectric constant k ′ = ε ′ ε o (ε o is the permittivity of the free-space) can be also calculated. Withthisdata,itispossibletocalculateσ e orε ′′ e ,butnottheactualvaluesofσ s andε ′′ e . This isalimitationoftheinstrumentationthatwemustbeawarebeforehand. However,thereare techniques that can be used to calculate σ s andε ′′ e from the dielectric measurements if the frequencyisswept(spectroscopy)tobediscussedlaterinthischapter. Therefore,infuturepartsofthisdocument,wheneverwewanttorefertothewell-known parametersε ′′ orσ, we will potentially need to refer toε ′′ e orσ e (also represented asσ ef f in this work), respectively, in particular at the context where dielectric instrumentation is used. ForthepurposeofusingasoildielectricmodelforaMIsignalattenuationmodel,the mentioned limitation is not an issue. However, from the perspective of physical intuition, CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 26 whenever the value ofε ′′ e becomes abnormally high at low frequencies, we need to remem- ber that this phenomenon may be mainly affected by the conductivity. On the other hand, whenever the value of σ e becomes very high at higher frequencies, it is potentially due to the dominance of ε ′′ . Next, we will derive mathematical expressions that help us in such kindof analysis. Westart with the followingformulas that can be directly derivedfrom the previousequations: ε ′′ e =ε ′ tanδ e (1.21) ε ′′ e = σ e ω (1.22) σ e =ωε ′ tanδ e (1.23) Because there is no universal convention regarding symbols related to permittivity, the following one is adopted in this work, whereε o =8.8542x10 −12 F/m, is the permittivity of free-space: • Complexrelativepermittivity: symbolk ∗ andk ∗ =k ′ − jk ′′ • Relativepermittivity(realpart),alsocalleddielectricconstant: symbolk ′ orε r • Relativepermittivity(imaginarypart),alsocalleddielectricloss: symbolk ′′ • Complexpermittivity: symbolε ∗ andε ∗ =ε ′ − jε ′′ • Realpermittivity: symbolε ′ andk ′ =ε ′ /ε o • Imaginarypermittivity: symbolε ′′ andk ′′ =ε ′′ /ε o With these initial definitions, it is possible to derive one typical expression given in the dielectricliteraturewhichisusefultounderstandthatwhile k ′ isrelatedtothecapabilityof themediumtostoreenergy(nolossinvolved),k ′′ reflectsacombinedtotallossesduetothe polarizationprocessandalsoduetoconduction. Tothisend,westartbyapplying(1.5)into (1.22)andusethepreviousdefinitionsfork ′′ : CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 27 ε ′′ e = σ e ω (1.24) ε ′′ e = (σ s +ωε ′′ ) ω (1.25) ε ′′ e = σ s ω +ε ′′ (1.26) k ′′ e = ε ′′ e ε o (1.27) k ′′ e = σ s ω +ε ′′ ε o (1.28) k ′′ e = σ s ε o ω +k ′′ (1.29) where k ′′ e is the effective dielectric loss (dimensionless), k ′′ is the actual relative imaginary permittivity of the medium (dimensionless), and σ s is the static field conductivity of the medium,typicallycalledσ DC ,andgiveninS/m. Similarly, it is possible to derive another well-known expression in the dielectric litera- tureforthepropagationconstantγ. Tothisend,westartbydeveloping(1.11)andapplying (1.5)andtherelationc o =(μ o ε o ) −1 : γ = p jωμ o (σ e + jωε ′ ) (1.30) γ = r j ω 2 ω μ o σ e −ω 2 μ o ε ′ (1.31) γ =ω r j μ o σ e ω −μ o ε ′ (1.32) γ =ω r j μ o ε o σ e ωε o −μ o (k ′ ε o ) (1.33) γ =ω √ μ o ε o r j σ e ε o ω −k ′ (1.34) γ =ω √ μ o ε o s j (σ s +ωε ′′ ) ε o ω −k ′ (1.35) γ = ω c o s j( σ s ε o ω + ε ′′ ε o )−k ′ (1.36) γ = ω c o r −k ′ + j( σ s ε o ω +k ′′ ) (1.37) CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 28 whereγ is the complex propagation constant,ω is the angular frequency (ω =2πf) given in rad/s, c o is the speed of light in vacuum (c o = 299,792,458 m/s), e o is permittivity of freespace(ε o =8.8542x10 −12 F/m),k ′ andk ′′ arethedielectricconstantanddielectricloss of the medium (dimensionless), respectively, and σ s is the static field conductivity of the medium,typicallycalledσ DC ,andgiveninS/m. The complex propagation constant γ has a real part α and an imaginary part β. The attenuation factorα is related to the energy losses (polarization and conduction) within the material [18]. The phase factor β, although not directly associated with losses, is related to the geometric spreading or spatial attenuation of the amplitude of the signal attenuation. This aspect, so far not properly highlighted in the WUSN literature, is extremely important for MI-WUSNs and the topic will be exploited in section 1.5. Besides these two kind of signalattenuation,typicallyothertwoaspectsareconsideredinthestudyofwavepropaga- tion: thescatteringofthewaveduetoobstacles(suchasrocks,rootsofplants,etc.) andthe interfaceeffects,inthiscase,thesoil-airinterface. Preliminarytestsareperformedinordertoinvestigatetheneedofconsideringthesetwo latter aspects in a LF soil model for MI-WUSNs. The field experiments are performed in harshenvironment(stateprotectedlandsatJackson,CA)andtheyhavethefollowingsetup: 10kHz,TXpowerlevel<500mW,30-cmdiametercoils(65turnsontheaverage),receiver sensitivity around 50μV, inter-node distances of 17.5m and 20m, and vertical deployment (surface of TX/RX coils facing each other and not the soil interface). The results indi- cate no perceptible variation of the received signal in the presence or not of non-metallic objects, such as human body, trees, and rocks close to the transmitting and receiving de- vices. Additional test was performed varying the relative depth of one coil (TX) while the other maintained fixed depth (RX). The maximum variation was 1m, until the point where the transmitting coil was completely above the soil surface. The maximum received signal variation was 3.7% (the higher received signal occurring when the TX coil is above the ground), insidethe normal varianceof thesignal innormal circumstances andwith thetwo coils properly aligned. Therefore, based on these results, we decided to neglect the effects ofnon-metallicobstaclesandthesoil-airinterface. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 29 1.2 PolarizationEffectsinSoilMixtures ConsideranexternalelectricfieldEwhichisbeingimposedonasoilmixture. Theinternal atoms, or molecules, or ions of the mixture will tend to minimize the electrical potential inside the material by deforming, rotating, or moving charges. This process is called polar- ization because the external force caused a displacement of charges from their equilibrium position. This process stores energy, similar to what occurs with a capacitor [65]. Because real-world dielectric materials including soil mixtures are not ideal, some aspects must be considered regarding the overall polarization process [2, 65]. First, the polarization does not occur instantaneously and there is a phase lag between the moment when the field is imposed and the complete polarization response of the material. To capture this physical behavior,acomplexpermittivityε ∗ (ε ∗ =ε ′ − jε ′′ )isadopted. Second,becauseanalternat- ing field is being considered, the imposed polarization-equilibrium-..-polarization cyclical process represents a loss, called dielectric loss. Third, if the frequency is sufficiently high, thepolarizationmaynotoccur(orbeverysmall)andthedielectriclossesdecreasemarking whatiscalleda polarization relaxation. Therefore,thegeneralspectralresponseofmateri- als is a non-linear descending k ′ curve as the frequency increases. Fourth, if the material is relativelylossy,conductioncurrentswillalsooccurinparallelwiththepolarizationprocess similar to a lossy capacitor (capacitance C and resistance R in parallel) and such additional lossesarecalledohmicorconductancelosses. Bothlosses,dielectricandohmiclosses, are capturedtogetherwhentanδ e ,σ e ,andε ′′ e areused,aspreviouslydiscussed. Thepolarizationeffectsatsoilmixturesatlowfrequenciesareverycomplex,aswewill verify shortly. Due to the longer time between alternating cycles, there is enough time for many kinds of charge reorganizations to take place, at both microscopic and macroscopic levels. In general, the sub-MHz frequency band is avoided for purposes of soil measure- ments, with few exceptions in geophysics and static conductivity. Therefore, there are not many soil dielectric works at low frequencies available, in particular a simple one which canbeusedforaMIsignalattenuationmodel(thegoalofthispartoftheresearch)without the needs of capturing all the details regarding the diverse electrochemical polarizations of the soil mixture. To this end, the next sections of this chapter are organized as follows: in Section 1.2.1, the characteristics of soil particles regarding this research are discussed. In Section 1.2.2, the effects of the water penetrating a dry soil are investigated. In Section 1.2.3,anoverviewofdiversekindsofpolarizationsinasoilmixturearebrieflydiscussed. In Section1.3,theselectionandsetupoftheinstrumentationusedforthisstudyarepresented. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 30 In Section 1.4, the empirical results regarding 9 kinds of soils are discussed. Finally, at Section 1.5, a novel method to mitigate the electrode polarization effects on the dielectric measurements is presented and the technique is applied to the mentioned empirical results resultinginaLFsoildielectricmodeltailoredspecificallytoMI-WUSNs. 1.2.1 CharacteristicsofSoilParticles The soils are formed from parent rocks, igneous, metamorphic, and sedimentary ones due to three factors: weathering (natural breaking down, not to be confused with erosion), tem- perature, and pressure. Besides the chemical composition, the soil particles have different sizes, sphericity, and angularity [2]. For the studies of the dielectric properties of soils, we are primary interested on the size of the particles because it is directly associated with the capability of the soil to hold water. This characteristic is associated to the specific surface area of the particles and can be understood by a simple analogy. Assume that you have a boxwithacertainvolumeanditmustbefilledwithpaintedballs: exclusivelysmallerballs, orlargerones,butnotboth. Noneoftheballswasalreadypainted. Ifyouselectthesmaller balls option, naturally the box will contain more balls than if you had selected bigger balls and you will end up having to use more ink to paint the smaller balls. This occurs because the necessary set of smaller balls have a higher surface area (S a ) compared to the set or larger balls. Usually for soil particles, S a is given in m 2 /g. Similarly, the tiny clay particles have higher S a than the bigger sand grains and consequently a higher volume of water is necessarytocoverthesurfaceoftheclayparticlesincomparisonwithsandparticles[2]. In short,thewaterretentionishigheratclayeyandsandysoils. Thewatermoleculesstronglylinkedtothesoilparticlesarecalledboundwater andthe water molecules at the voids between the soil particles are called free water. Bound and free water have distinct behaviors at the presence of an electric field. Free water molecules rotate according to the electric field orientation and have a dielectric permittivity close to 80. However,fortheboundwatercase,therotationofthemoleculesisimpactedbytheex- istingboundtothemineralsurfaceswhichdecreasesthedielectricvalueoftheboundwater. Therefore, the size of the grains is an important parameter in the studies of the dielectric properties of soils and typically some form of soil texture classification is used, as shown in Fig. 1.1. At this figure, with the data provided in [2], pp.31, it is possible to verify the inverse relation between particles’ size and their surface area. Also, a soil with texture P is CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 31 percent SAND 0 100 0 90 100 0 10 20 30 40 50 60 70 80 90 100 ¨Clay¨ Montmorillonite 1E-9 - 5E-7 ¨Clay¨ Kaolinite 4E-6 - 5E-6 Silt 2E-6 - 6E-5 Sand 6E-5 - 2E-3 2 Surface area (m /g) ¨Clay¨ Montmorillonite 400-800 ¨Clay¨ Kaolinite 10-20 Silt 0.04-1.1 Sand 0.001-0.04 Average length or diam. (m) Fig.1.1Soilclassificationaccordingtotheirtexture(dataatthetablesin[2]). marked in the figure and it is classified as clay loam with 30% of clay, 30% of sand, and 40% of silt [15, 29]. Besides the grain size (and its associated surface area), the sphericity and angularity of the particles, the mass density, porosity, and the chemical composition aspectsarealsoimportantforsoildielectricmodelsandacomprehensivestudyofthesesoil characteristicscanbefoundin[2]. 1.2.2 DryandWetSoils Typically, when the soil is dry, manyionic substances are tightly attached to the bigger soil particles (i.e., clay, silt, or sand particles). An ion is formed when an electrically neutral atomgains/losesoneormoreelectrons[2]. Whenanelectricfieldisappliedtothisdrysoil, the formation of conduction currents is very difficult to occur because typically the ionic formations are separated and the flow of a current may not be feasible. Moreover, because ingeneralsinglematerials,suchasisolatedsandorclayparticles,havelowpermittivity,the lossesduetopolarizationandconductionatdrysoilaresmall,bothinhighandlowfrequen- cies. However, this scenario dramatically changes when water is added and that is why it istypicaltohaveattenuationstudiesofelectromagneticwavesinsoilshowingdistinctplots fordryandwetsoils. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 32 A C A Stern layer Host (soil particle) _ _ A Free water Hydrated cation Hydrated anion Double layer Bulk electrolyte Fig.1.2Wetsoilparticle-fluidinteractions(basedonillustrationsin[2]) The hydrogen atoms in a water molecule are oriented in such a way to form an angle of 109 o . At distances larger than 3 molecules, the water molecule can already be properly modeled as an electric dipole [2] which is easily oriented (rotation) by an external electric field, which is the case of free water in soils. This explains why water is highly polarizable (permittivity around 78-80) in comparison with other materials. When water is added to drysoils,aquaeouselectrolytesareformedwhicharecomprisedoffreewaterandhydrated ions. When an electric field is applied to this wet soil, the movement of hydrated ions oc- curs,thusconductionlossesarehigher. Moreover,thepolarizabilityoffreewaterdecreases duetotheincreaseofconductioncurrents. The previous discussion is relatively simplistic. Actually, the physical interpretation of what occurs at wet soil mixtures exposed to electric fields requires the addition of layers of interactions. Duetotheirhighersize,thesoilparticles(sand,silt,clay)areconsideredhosts wheresuchinteractionswilloccuratdifferentdistancesorlayers. Intheliterature,typically threelayers(onlythefirstoneappliestodrysoils)arementioned,aslistedbelowandshown inFig. 1.2. Atthenextsection,theeffectsoftheselayersonthedielectricpropertiesofwet soilswillbediscussed. • Sternlayer: watermoleculesandionsaretightlyattachedtothehostsurface • Double layer: intermediate region where the ionic formations and water molecules arepartiallyinfluencedbythehostpresence • Bulkelectrolytes: minimalornoinfluenceduetothehostpresence CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 33 due to Electronic polarization 10 15 Frequency Real permittivity resonance due to Ionic polarization Frequency Real permittivity 10 12 resonance Range of Electronic polarization (up to PHz) => ¨always present¨ Range of Electronic polarization (up to THz) => ¨always present¨ (a) (b) Fig.1.3Spectraof(a)electronicpolarizationand(b)ionicpolarization. 1.2.3 OverviewonBroadbandSpectralResponseofSoils Polarizationmechanismsindielectricmaterialscanpresenttwocharacteristicsaccordingto the frequency: resonance and relaxation. In resonance response, both real and imaginary parts of the permittivity display some form of peak in their values in a certain frequency region and these values drop to smaller ones at frequencies sufficiently smaller or greater than the resonance frequency. One can model such behavior with a RLC lumped circuit bothbyaseriesoraparallelconfiguration[66]. Theelectronicpolarizationpresentsareso- nancebehaviorasshowninFig.1.3a. Similarly,theionicpolarizationalsodisplaysresonant behavior,asshowninFig.1.3b. Whenthedispersivebehaviorofthepermittivityofadielectricmaterialisanalyzed,that is, the variation of its permittivity according to the frequency, it is important to analyze the spectrafromtheveryhighfrequencyand,step-by-steppassingbyeachkindofpolarization toward the very low frequency. This right-to-left broadband analysis is recommended be- cause all the polarizations effects accumulate toward lower frequencies. Or, alternatively, we can say that once a sufficient high frequency is achieved, certain form of polarization does not hold anymore. It was already commented that for the completion of the polar- ization mechanism it is necessary a finite amount of time (displacement time) and this one must be smaller than the time of half-cycle of the signal, before the polarization reverts to thecontrarydirection. Ifthefrequencyissufficientlyhigher,thepolarizationtimecannotbe achievedandthematerialdoesnotpresentthatpolarizationeffectforthatspecificfrequency and higher ones. For instance, in the case of the electronic polarization, we can see that its effectsceaseatfrequencieshigherthan10 15 Hz. Naturally,forourLFstudiesofsoils,both effectsofelectronicandionicpolarizationwillbepresentalltime. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 34 Itispossibletoplotsuchcurveswithresonantbehavior,althoughsuchanalysiswillnot be necessary for our goals regarding MI-WUSNs. The expression for the complex relative permittivityofamaterialfortheresonancecaseis[2]: k ∗ =k ′ − jk ′′ (1.38) k ∗ =k ′ ∞ + 1 2 k ′ o −k ′ ∞ (1− ω ω res )+ j β ω res (1.39) where k ′ o and k ′ ∞ are the dielectric constant values at frequencies much lower and much higher, respectively, than the angular resonance frequencyω res andβ is an adjustment fac- tor,inthiscase,acoefficientassociatedwiththelossbyradiation[2]. The physical interpretation of both electronic and ionic polarizations is illustrated in Fig. 1.4. For the electronic polarization, the electron clouds are displaced eccentrically ac- cording to the direction of the external electric fieldE [67]. When E changes its direction, according to the sinusoidal cycles, the cloud of electrons reverts its orientation. Moving up inthescaletowardthemacroscopiclevelofatomsorgroupsofatoms,wecanthenconsider the ionic polarization. As shown in Fig. 1.4, at equilibrium the atoms present some form of symmetry for the charge distribution which is broken when E is imposed [18]. Because themassoftheatomsishigherthanelectrons,thedisplacementtimeoftheatomsaremuch higherandthisfactisshownbyasmallerresonancefrequency(around10 12 Hz)compared totheonefortheelectronicpolarization,asshowninFig. 1.3. Movingfurthertowardlowerfrequencies,theremainingtypesofpolarizationidentified for soil mixtures present relaxation spectrum, that is, the polarization mechanism does not have associated restoring forces. Therefore, for the real part of the permittivity, a step-like behavior is observed as illustrated in Figs. 1.5a and b, for both orientational (or dipole), spatial (or interfacial), and double layer polarizations [2, 65, 67]. As illustrated in Fig. 1.4, the orientational polarization is similar to the ionic polarization, but associated with bigger (i.e.,heavier)players,inthiscasedipolarmolecules. Forsoilmixtures,thewateristhemain component affected by this kind of polarization which is evidently strong considering the fact the water by itself has dielectric constant value around 78 which is value much higher than the dielectric constant of the other soil components, considered each one separately (i.e., single phase analysis). Free water presents stronger orientation polarization, while bound water is partially affected although it is associated with other kinds of very strong polarizations. Therefore, from the physical interpretation standpoint, the higher is the de- CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 35 Electronic Polarization + _ E _ Unpolarized state + _ _ _ _ _ _ _ _ Polarized state Ionic Polarization C A C A C A C A C A C A C A C A C A C A C A C A C A C A C A C A net dipole moment = 0 net dipole moment > 0 Player Orientational Polarization Atoms Ions (e.g., sodium chloride [NaCl]) Dipolar Molecules (e.g., water) Polarization Type _ _ _ _ _ _ _ _ _ _ Mass increases: displacement time increases + Fig.1.4Polarizationtypes: electronic,ionic,andorientational(ordipolar). gree of porosity of the soil, potentially the higher is the amount of free water that can be subjectedtotheorientationpolarization. Nonetheless, the effort of fully characterizing the soil dispersion due to the orientation andotherpolarizationsismuchmorecomplexthanthispreliminarydiscussionanditisstill an open research topic, in particular for sub-MHz range. This is the case because there are manyotherfactorstobeconsideredbesidesthesoilporosity. Forinstance,thehigheristhe surfaceareaoftheparticles(e.g.,clayeysoil),thehigheristheamountofboundwater,thus increasing the net effect of the orientation polarization despites the smaller amount of free water available. Another aspect is the impact of the shape and angularity of the existing clay particles in soil. In fact, the high diversity of variables associated with this problem is reflected by the big frequency range (10 6 −10 9 Hz) where different soil settings present theirrelaxationfrequencies,asshowninFig.1.5a. Ingeneral,themanufacturersofsoilmea- surement devices, in particular soil moisture sensors, usually avoid to employ operational frequencies strongly impacted by the relaxation processes and it is still very common the needofdevicecalibrationsfordistinctkindsofsoils. Theorientationpolarizationtypically isthephenomenonwhichisbehindthiscalibrationeffort. NotethatinFig. 1.5a,evenatthe flat regions of the curve, a certain level of variation for the values of dielectric constant is stillobserved. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 36 10 9 Frequency Real permittivity relaxation Frequency Real permittivity 10 6 relaxation Range of Orientational polarization for free/bound water (GHz / MHz) => varies with temperature and type of molecule Range of Spatial (sub-GHz) => very complex to model (mixtures) due to Orientational polarization due to Spatial and Double layer polarizations 10 6 (a) (b) 10 8 Fig.1.5Spectraof(a)orientational,and(b)spatial+doublelayerpolarizations. The relaxation phenomenon cal also be modeled by a lumped circuit. A basic series or parallel RC configuration can be used [66], but due the existence of multiple polarizations playersatsoilmixture,amorecomplicatedconfigurationwithdifferentvaluesofresistance and capacitors in multiple-order circuits may be necessary. Another approach that give us curves for the complex permittivity of the soil mixture, similar to the highlighted parts in Fig. 1.5,istheuseofthefollowingexpressionknownasDebye’sequation[2,68]: k ∗ =k ′ − jk ′′ (1.40) k ∗ =k ′ ∞ + k ′ o −k ′ ∞ 1+ j ω ω rel (1.41) k ∗ =k ′ ∞ + k ′ o −k ′ ∞ 1+ jωτ (1.42) k ′ =k ′ ∞ + k ′ o −k ′ ∞ 1+ω 2 τ 2 (1.43) k ′′ =(k ′ o −k ′ ∞ ) ωτ 1+ω 2 τ 2 (1.44) where k ′ o and k ′ ∞ are the dielectric constant values at frequencies much lower and much higher, respectively, than the angular relaxation frequencyω rel = 1 τ , andτ is the relaxation time[2],suchthattheoperatingfrequency f = 1 2πτ . For a relaxation dominated by the dipole polarization, the Debye’s equation (1.42) is sufficient to represent the polarization mechanism. In fact, this equation holds for any phe- nomenon characterized by a single-time relaxation [68]. However, it is important to high- light that for soil mixtures the remaining kinds of polarization other than the dipole lead to certainapparentanomalies(i.e.,giganticvaluesfork’)thatarenotcapturedby(1.42)when applied to lower frequencies. At such frequencies (e.g., sub-MHz), in particular for wet CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 37 rel =1/ (a) (b) e Fig.1.6Cole-Colediagramfor(a)single-timerelaxationand(b)spatialpolarization[2]. soils,modificationstotheDebye’sequationareusuallyrequiredforaproperrepresentation of the phenomenon. The following technique can be used to identify this case. We start by narrowing our frequency analyzes around the frequency where the orientational polariza- tion is expected to dominate. Next, we can use (1.42) to plot the complex permittivity in a differentway,formingasemi-circle,bymeansofadiagramcalledCole-Coleplot[2]. This procedure is very helpful to identify anomalous dispersion or the existence of other lower frequencies polarization effects besides the orientational polarization. The Cole-Cole plot issimplytheplotofthedielectricconstantk ′ (atthex-axis)andthedielectriclossk ′′ (atthe y-axis). Note that the axes are in linear-axis mode, while typically we use log-log mode to plotfrequencyvs. k ′ (ork ′′ ). A Cole-Cole diagram for a single-time relaxation process, such as the orientational po- larization, is shown in Fig.1.6a. In real-world experiments, usually the extreme values for k ′ , that is, k ′ o and k ′ ∞ may not be so well defined as in the figure. However, typically its is possible to use the Cole-Cole diagram to infer these values. This procedure is used in this research as we will see soon. Moreover, typically for soil mixtures more than one relaxation-time are involved and different plot outputs are expected, as illustrated in Fig. 1.6b which also represents a spatial polarization. As mentioned before, each one-time or multiple-time relaxation processes can be represented by a lumped circuit involving R and Ccomponentsanditispossibletolookforintheliteratureaboutthedetailsofhowtoderive the proper representative lumped circuit, as in [69]. At that work, it is also to observe how to empirically-derive a Cole-Cole diagram for soil mixtures in frequencies higher than the expectedonesformid-rangeMI-WUSNs. Similardielectric-relatedworkstargetingbiolog- icalareascanbefoundandtheyaresourceofimportanttechniques[68]. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 38 + _ E _ Unpolarized state Polarized state Players * Water-soil particle * Air-soil particle Polarization Type Mass increases: displacement time increases Multi-phase (mixtures) Interfacial Polarization, also called: * Spatial Polarization * M-W Polarization [Maxwell-Wagner] (no relaxation) _ + + + + + + + + + + + + + + + _ _ _ _ _ _ _ _ _ _ _ _ * Water-soil particle with low specific surface (e.g., sand) * Air-soil particle with low specific surface _ _ _ _ _ _ _ _ _ _ _ _ + + + + + + + + + + _ _ _ _ _ _ _ _ _ _ _ _ + + + + + + + + + + * Water-soil particle with high specific surface (e.g., clay) Stern layer and Double layer Polarizations Stern layer Double layer Double layer with different arrangement Diffuse layer Fig.1.7Polarizationtypes: interfacial(orspatial,orM-W),Sternlayer,andDoublelayer. An equivalent way to show the mathematical function behind the behavior of lumped circuits derived for multiple-time relaxation process is the use of modifications to Debye’s equation [2] and one of such modified equations that frequently appears in microwave soil dielectric models is the Cole and Cole equation which adds a spread in relaxation times. In other words, rather than a perfect semicircle at the Cole-Cole diagram, sometimes the phe- nomenonshowsupasacompressedsemi-circleandthisisthescenariowherethefollowing ColeandColeequationcanproperlyfit: k ∗ =k ′ ∞ + k ′ o −k ′ ∞ 1+(jωτ) 1−α (1.45) where the parameters are the same as in (1.42) with the addition of α, which is a polariz- abilityfittingparameternormallyempirically-determined. Itisimportanttorememberthatgoingloweratthefrequencyspectrum,thematerialac- cumulates the mentioned polarizations such that the effective polarizability of the material CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 39 isthesumofthementionedelectronic,ionic,andorientationalpolarizationseffectswiththe remainingLFpolarizationmechanisms. As we move even further at lower frequencies (e.g., sub-MHz), heavier players such asmulti-phasemixturesinvolvingsoilparticles,water,andairhaveenoughtimetodevelop theirpolarizationprocess,despitestheirhighdisplacementtimes. OneofsuchLFrelaxation phenomenaisthespatialpolarization(alsocalledMaxwell-Wagner,M-W,orinterfacialpo- larization), as shown in Fig. 1.5b. For this polarization, the physical form and orientation of the soil particles are significantly important [2]. As shown at the first case in Fig. 1.7, if the soil particles are plate-like and are not randomly oriented, the spatial relaxation may not occur. In general, sand particles are more spheroid-like due to the higher size and the spatial polarization effects are potentially observed, as illustrated at the second case in Fig. 1.7. This kind of polarization occurs due to the electrical differences between the players, suchasconductivityandpolarizability,causingaccumulationofchargesattheinterfacesof thehostparticles. Suchhostparticles,suchassandandsiltones,arelargeenoughtobehave as micro-capacitors resulting in a relatively high polarizability. However, observe that the polarizationeffectisdirectlyrelatedtothepresenceofwater. The mentioned micro capacitive structures involved in the spatial polarization, that is sand and silt particles, do not form layers and the polarization effect is not drastically high. The scenario changes when high specific surface inclusions are involved, such as tiny clay particles, as illustrated in the third case in Fig. 1.7. The inclusions themselves do not be- havelikemicro-capacitorsbuttheirpresenceinanelectrolytesolutionallowstheformation ofdifferentkindsoflayers,asalreadydiscussedandillustratedinFig. 1.2,basedonfigures in [2]. These polarizations are called Stern layer and Double layer and they are topics of increasing research, not only related to soil mixtures. Many models have been proposed for these kinds of polarizations. For soil mixtures, we will see at the next section that the interfacial polarization has a lower impact (at least, it has an almost static contribution) to the MI signal attenuation considering diverse soil settings. However, the Stern and Double layer polarizations have a strong impact that clearly differentiate the MI signal attenuation for a sandy soil compared to a clayey soil. Therefore, we can conclude that the real per- mittivity of soil mixtures potentially increases with decreasing frequencies. This fact has strongimplicationsinsub-MHzMI-WUSNdesigns. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 40 1.3 MeasurementsofElectricalPropertiesofSoil Laboratory measurements of DC electrical conductivity and magnetic permeability of soil are straightforward and many commercial instruments are available. However, for the per- mittivity measurements of soil samples, the task is more complex and it may require ex- pensiveinstrumentationinordertoachieveaccurateresults. Inparticular,thedielectriccell (i.e., the soil sample holder) has a typical cost many folds higher than the main electronic device used at the instrumentation. On the other hand, we will show in this chapter that, for the development of mid-range MI communication for WUSNs, it is possible to employ a cheaper dielectric instrumentation which also has the potential of being an intrinsic part of MI nodes. With this goal in mind, we develop a low-cost apparatus for laboratory and in-situsoilmeasurementswhichhassatisfactoryaccuracyforourapplication. An impedance analyzer (also called LCR meter) is usually a proper instrument for di- electric measurements of material in general for frequencies below 10MHz [70]. The fol- lowing explanation applies for the majority of the cases involving impedance analyzer and alsomoreexpensiveinstrumentationinvolvingfrequencyanalyzers. Nomatterifthedevice is used to measure inductor or capacitor electronic components, or to measure dielectric properties of blood or soil samples, in fact the instrument directly measures only the com- plex impedance of the material: the real resistance (R) and the angleθ related to the phase differencebetweencurrentandvoltage. Typically,suchmeasurementsspanacrossacertain frequency range, e.g., 10Hz-1MHz with linear- or logarithmic-spaced sampling points (the latterisusuallythepreferenceforsoilmeasurements). Themagnitudeoftheimpedance|Z|isdirectlymeasuredbydividingtheappliedvoltage across the specimen under analysis to the measured current passing through this material. θ is achieved by measuring the phase shift between the source voltage (reference) and the phase of the current. From|Z| and θ, the values of R and C (or L) are calculated. The instrumentdecreasesitsaccuracyifθ istoolow/high(e.g.,verycloseto0 o and90 o ). These measured and calculated values are then converted to other impedance-related parameters, if desired [70]. Note that the below R andC values are related to a series-RC configuration representingthesetsoilsample+dielectriccell+associatedinstrumentwiring: CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 41 |Z|= R cosθ (1.46) Z =R+ jX =|Z| θ (1.47) X =X L +X C X =ωL− 1 ωC (1.48) where ω is the angular frequency and X is the apparent reactance measured by the instru- ment. Note that, for soil measurements and depending on the frequency, L accounts for inductive effects of the wiring and also of the dielectric cell. Similarly,C accounts for ca- pacitive effects of the wiring, the dielectric cell (e.g., fringe effects), and as expected, also forthecapacitanceofthebulkmaterialunderanalysis. CommercialimpedanceanalyzersandLCRscanalsoprovidethevalueofDF (losstan- gent) which is the inverse of the quality factor of the material (DF =1/Q). DF typically indicates how lossy is the dielectric material for a certain frequency. A DF value which is very close to 0 indicates a material with lossless characteristics. We will see in this thesis that, particularly for soils, DF has a very wide dynamic range, mainly depending on the frequency and wetness condition of the soil sample. In some cases, the value of DF is so high that the traditional consideration of the soil sample as a dielectric material loses its physicalmeaning. Inthiscase,apracticalwaytohaveaproperphysicalinterpretationoftheproblemisto consider the soil sample as a lossy capacitor (i.e., a parallel-RC configuration; not be con- fused with the apparent series-RC measured by the equipments). Note that the expression for DF is different than in (1.52) which is shown below: it is actually its inverse which is (ω·R ˙ C) −1 . Therefore, for an almost lossless dielectric material (R high), DF is small (i.e., DF is close to 0), that is the conduction current passing through R of the parallel-RC schemeispotentiallymanyorderssmallerthanthedisplacementcurrent. Ontheotherhand, for a very lossy dielectric material (R small), DF is very high, typically going to hundreds for very lossy soils and the conduction current dominates the apparent measured current passing through the system where the displacement current may be negligible compared to the conduction current. The problem may be difficult to analyze because it involves a product (i.e., R·C). Returning to the instrumentation analysis, depending if the material has predominant inductive or capacitive characteristics, DF is calculated as follows (for a CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 42 series-RCrepresentation): ifθ>0: X =X L −0=X L =|Z|sinθ (1.49) L app = X L ω = X L 2πf DF =tan(90 o −θ)= R X L = R ωL app (1.50) ifθ<0: X =0−X C =X C =|Z|sinθ (1.51) C app = −1 ωX C = −1 2πfX C DF =tan(90 o −θ)=− R X C =−ωRC app (1.52) whereC app and L app are the apparent values of the capacitor or inductor-like material, re- spectively. For the context of this work, only C-related values are measured and analyzed (i.e.,θ <0). Forconvenience,typicallythevaluesofX C inthisworkarepresentedaspositivevalues. However, it is important to always keep track of the sign of X C when mathematically eval- uating the expressions. While a negative X C will be associated to a positive value for the dielectric constant k ′ , a positive X C will lead to a negative value of k ′ which does not have physical meaning for the materials under analysis here. Therefore, a mathematical evalua- tionofapositivevalueofX C forsoilsmeansthatthemeasuredθ ispositiveandthesystem has a predominant inductive characteristic. Such case can occur when one moves to higher frequenciesandX L in(1.48)becomeslargercomparedtoX C . Suchscenariocanpotentially occur when the impact of the instrumentation wiring inductive behavior becomes stronger compared to the capacitance value of the bulk material. This examples illustrates the need of calibrating the system before directly usingC app as theC value for the series-RC repre- sentingthebulkmaterial. As previously mentioned, a soil sample can be modeled as a lossy capacitor, that is, C in parallel with R. Note that R is the effective resistance of the material and it is associated to both conductive and dielectric losses. This represents a challenge in some cases, such as to derive σ DC of the soil. Fortunately, the effective values σ e and k ′ are sufficient for our purposesrelatedtoMIsoilattenuationmodels. Thenextstepatthesoildielectricproperties determinationistoconvertthevalueofC andDF (alternatively,|Z|andθ)intoσ e andk ′ . So far, no assumption has been made in relation to the kind of container which is being usedtoholdthesoilsample. Themostusedanddocumentedspecimencontainer(ordielec- triccell)isthe parallel-plate capacitor cell. Thiscontainerisrelativelysimpletobuildand CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 43 Fig.1.8Coaxialcapacitorcellusedforthesoildielectricmeasurements. many soil-related works successfully adopted such solution [3, 16, 71–74]. In this research work, we preliminarily adopted a customized parallel-plate cell. However, mainly due to differentfringeeffectsaccordingtothesoilwetness,wefacedrepeatabilityissuesinvolving parallel-plate dielectric cells and soil samples. On the other hand, the other option for the dielectric cell was much more robust during our empirical investigation: the coaxial capac- itorcell[71,75,76]. Besidessmallerfringe(orstray)effects,thecoaxialcapacitorcellalsohastheadvantage ofpotentiallybeingabettercontainerforfluidsandwet/saturatedsoils,asshowninFig. 1.8, where our customized coaxial cell is holding deionized water during a calibration process. Thedetailsrelatedtothecalibration,includingshort/opentestsandsampleholderstrayca- pacitancecorrection,canbefoundat[70,75–77]. InFig. 1.8,oneoftheinstrumentsusedin this work, a Quadtech 1910 impedance analyzer, is also shown. This equipment if used for dielectricmeasurementsmustbeperceivedasatwo-electrode(2E)device,althoughitactu- allyemploysfourBNCconnectors. Thepurposeofthesefourportsistomitigatetheeffects of resistance and parasitics inductance of the dielectric cell cable. This scheme must not be confused with a four-electrode (sometimes called 4-terminal, a terminology we avoid) (4E) device which will be mentioned shortly as is used to reduce/eliminate the electrode polarizationseffect. Once theC and DF values are provided by the instrumentation, the next step is to cal- culate the values of dielectric parameters of interest, as shown below for parallel-plate and coaxialcapacitorcellcases[70,75]. Itisimportanttohighlightagainthattherearetwoways tomodeltheimpedanceofamaterial: asaseriesconfiguration(R± jX)orasaparallelcon- CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 44 figuration (G± jB), where G and B are the conductance and susceptance, respectively, of the specimen [70]. In general for soil measurements, the parallel mode is the best one for a prompt derivation of the dielectric constant and conductivity value for a lossy capacitor model of the sample. If the instrument is configured for RC-series scheme, the values of R and C must be converted to the parallel ones (consult the instrument manual or check the detailsatthenextchapter)beforeapplyingsuchvaluesatthefollowingequations: foraparallel-platecapacitorcell: C = ε ′ A d (1.53) ε ′ = Cd A k ′ (d.constant)= Cd ε o A ε ′′ e =ε ′ (DF) σ e =2πfε ′′ e (1.54) foracoaxialcapacitorcell: C = 2πhε ′ ln b a (1.55) ε ′ = C·ln b a 2πh ε ′′ e =ε ′ (DF) σ e =2πfε ′′ e (1.56) where A is the disk area, given in m 2 , d is the distance between the plates of a parallel ca- pacitor,giveninm. Also,histhelengthofthecoaxialcapacitor,giveninmanda,barethe radius of the internal metallic pin and the radius of the capacitor, respectively, given in m, asshowninFig. 1.8. A potential issue to be aware while employing or designing such kind of instrumenta- tioniscalledelectrodepolarization(EP),whichoccursatverylowfrequencies(e.g.,<100 kHz) [2, 78–82]. EP can create uncertainties in relation to the apparent high measured val- ues for the dielectric constant: if they really reflect the inherent physical properties of the soil sample, or if they are due to the artifacts of the instrumentation errors (i.e., due to EP), or a combination of both cases. Accordingly, the EP problem is analyzed and guidelines areprovidedtomitigatetheissueatthenextsectionandwithagreateramountofdetailsin Chapter2. The challenge of dealing with the EP effect (e.g., abnormal high values for real per- mittivity) associated with the dielectric measurements of wet soils is pretty similar to the well-known problems in measuring the dielectric constant of electrolytic solutions studied at the fields of physical chemistry [83] and biosystems. Accordingly, the EP-mitigation CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 45 techniqueproposedhereisbasedonsimilareffortsattheseresearchareas. The currents flowing in dielectric measurement instrumentation system involve elec- trons, while ions are involved in wet soil mixtures. Due to this incompatibility, an accumu- lation of ions at the instrumentation electrodes occurs while a DC current passes through a high-conductivity sample (i.e., electrolytic solution, soil mixture). The ions accumulate because they cannot chemically combine with the electrodes. At very low AC frequencies [2,16,83],theEPphenomenonalsooccurs. Athigherfrequencies,thereisnotenoughtime for the ions to migrate to the electrodes and the EP effect is minimum or non-existent. The accumulationofionsattheelectrode-sampleinterfacecanbeperceivedasahighimpedance layer(doublelayercapacitance)thatcausesanapparentveryhighpermittivityvalueforthe sampleunderanalysis[2,16,81]. ThemitigationoftheEPproblemisnottrivialforwetsoil mixtures because, disregarding the EP phenomenon, the true value of the soil permittivity may also dramatically increase at low frequencies due to the natural soil dispersion behav- ior. Therefore, for our signal attenuation model purposes, the challenge is to remove the artificial EP-related part from the abnormally high measured values of permittivity without distortingtheactualvaluesofpermittivity. DrysoilssamplesarenotsignificantlyaffectedandmethodstomitigateorcorrecttheEP effectonthemeasurementsofdrysoilsareingeneralnotnecessary. However,fordielectric measurementsofwetsoilsatLF,theEPphenomenoncannotbeneglected[81]. Oneoption istheadoptionofspecial4E instrumentationwith4-electrodes(again,notnottobeconfused with LCRs or impedance analyzers that have 4 ports [e.g., IL, PL, PH, IH]) to mitigate the EPissue[2,16,78,82]. However,ithasbeenreportedthatevenwitha4Einstrumentation, additionalissuescanoccuranditmaynotbethedefinitesolutionfortheEPeffects[78,79]. Another option to solve the EP issue is called reversible electrode systems and refers to the adoption of special coating at the electrode interfaces in order to allow the ions to combine with the electrode material (redox reaction) voiding the EP effect while not intro- ducing significant loss to the system [16]. Unfortunately, this technique and similar ones have been reported as ineffective or unreliable [2]. Moreover, the mentioned EP solutions are not practical for a soil sample holder which is permanently below the ground and such arrangement is one of possibilities envisioned in this work. A more detailed treatment of theEPsubject,containingasurveyoftheEP-correctiontechniques,canbefoundin[84]. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 46 Table1.1Characteristicsofthesoilsamplesusedinthisresearch. Label Location Clay Sand Silt Textureclassification NHS NewHoganLake(siteA)-ValleySprings,CA 25% 27% 48% MediumLoam LOA1 Urbanresidence’sgarden-Pasadena,CA 15% 52% 33% MediumLoam LOA2 Universitycampus’sgarden-LosAngeles,CA 19% 56% 25% SandyLoam SAN1 Commercialplaygroundsand 2% 91% 7% Sand TONZI Particularcattleranch-Jackson,CA 23% 25% 52% MediumLoam NHS-DRY1 NewHoganLake(siteB)-ValleySprings,CA 19% 35% 46% MediumLoam NHS-SAT NewHoganLake(siteC)-ValleySprings,CA 23% 17% 60% SiltyLoam NHS-WET NewHoganLake(siteC’)-ValleySprings,CA 23% 27% 50% MediumLoam NHS-DRY2 NewHoganLake(siteC")-ValleySprings,CA 27% 23% 50% MediumLoam It is important to highlight that the EP issue significantly may impact only a relative small range of frequencies of the MI-communication system. Moreover, even for this af- fectedfrequencyrange,onlywetsoilsarestronglyimpacted. Therefore,weinitiallydecided to adopt a relatively simple software-based solution in order to address the EP issue while maintaining a conventional, low-cost dielectric 2E instrumentation. This approach proved tobesuccessfulforthedevelopmentofasub-MHzsoildielectricmodelforMI-basedcom- munication. However, to achieve more accurate results for soil sensing purposes, a more elaboratedsolutioniseventuallydevelopedwhichisdiscussedatthenextchapter. 1.4 SoilDispersionatLowFrequencies Preliminaryexperimentswiththeinstrumentationdescribedatthelastsectionareconducted involving 9 types of soils, as shown in Table 1.1. Different frequencies, ranging from 1 kHz to 200 kHz are employed and initially only two extreme soil volumetric water content (VWC) are considered: naturally dry and saturation conditions. The main goal of these ex- periments is to identify classes of soils with similar dielectric characteristics under dry/wet conditions. A secondary goal is to identify the electrode polarization (EP) issues for a sig- nificant number of soil samples. As already mentioned, the EP effects are not intrinsic polarization characteristics of the soil sample being tested, but it can be perceived as an instrumentationartifactthatcancauseseriousdistortionsonthedielectricmeasurementsof materialsatlowfrequencies,inparticulariftheyhavearelativelyhighconductivity. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 47 The empirical results related to the dielectric constant dispersion are shown in Fig. 1.9. The soil samples labeled as NHS- are from the New Hogan Lake (Valley Springs, CA) and the samples were collected from an area which is not open to people. Moreover, differ- ent distances (couple of meters) from the boundary of the lake are considered (i.e., natural variation of the soil moisture). Accordingly, we have samples that were visually very dry whencollected(NHS-DRY1andNHS-DRY2),somerelativelywet(NHS-WET),andsome in saturation condition (NHS-SAT). All samples are collected and evaluated in oven-dry andsaturatedconditionsafterthesoilpreparationprocess. Forthisreason,wehaveacurve labeled such as NHS-WET-dry: in this case, the rightmost part of the label denotes the soil moistureconditionatthelaboratorytest. Fig.1.9Empiricalresults: dielectricconstantdispersionfor9differentsoils. Whenoperatingatlowfrequencies,abnormallyhighdielectricconstantvalues,asshown in Fig. 1.9, may occur and similar results have been reported in many LF experiments with soil mixtures, such as in [71]. There are two main reasons explaining such high values for k ′ . First, at the kHz range, a significant number of polarization mechanisms take place, as discussedattheprevioussection,suchastheinterfacialandtheDoublelayerpolarizations. In particular, the latter polarization is associated to what is called gigantic double layer ef- fect, when values of k ′ becomes higher than hundreds or thousands. A second reason for highvaluesofk ′ isthealreadydiscussedinstrumentationfactorcalledelectrodepolarization (EP), as shown in Fig. 1.10. EP is typically associated with 2E impedance analyzers that CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 48 10 4 Frequency Real permittivity 1) Gigantic double layer effect (at low frequencies) 2) Electrode polarization: * instrumentation error cause by double layer effect at the electrode-specimen interface * not characteristic of the material due to Macrospace polarizations 10 5 Fig.1.10Abnormallyhighvaluesforthedielectricconstantatlowfrequencies. may be exacerbating or distorting the results [2, 73], which is typically the case when we getvaluesofk ′ higherthan10 4 ,assumingthatthemeasurementsareforfrequencieshigher than 1 kHz. Many investigators have proposed different ways to solve the EP issue, but all techniquesareconsideredinefficientorsourceofadditionalproblemsaccordingto[2],with the exception of the adoption of a 4E impedance analyzer. It is important to understand what were the primary reasons for not adopting 4E instrumentation for soil measurements inthisresearch. It is well known that 4E instrumentation can significantly mitigate the EP effects [2, 3], however2importanttrade-offsmustbeconsidered: a)atrue4Eimpedanceanalyzeritisnot easily available as off-the-shelf equipment and this fact can impact the repeatability of the experiments by the research community and b) it requires special procedures regarding the soilsampleanditsholder[3]. Inrelationtothelatterconstraint,atthecontextofthecontin- uation of this research work related to the feasibility of in-situ soil moisture measurements atlowfrequenciesbyMI-WUSNnodes,anyprocedurethatmustberegularlyperformedat laboratory is a serious constraint. Moreover, as will be discussed in Chapter 2, soil sam- ples and the 4E technique may not be compatible in terms of repeatability of experiments. Therefore, we decided to maintain the 2E instrumentation and develop a method to correct theEPdistortionandthisonewillbepresentedatthenextsection. AnalyzingthedielectricresultsinFig. 1.9,someconclusionscanbereached: • Forfrequenciesaround100kHzorhigher,thevaluesofk ′ arenotextremelyhighand theyareclosetoexperimentsstartingat1MHz[3,73,74]. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 49 • In general, there is a significant separation between the values of k ′ for dry and wet soil cases (e.g., one order of magnitude). With one exception, these dry/wet cases are clearly identified as the lower (−− trace) or upper (solid trace) group of lines, respectively. • The mentioned exception is the soil NHS-WET which has very high values of k ′ for bothdryandwetcases. • For frequencies around 10kHz or smaller, the values of k ′ are significantly high (i.e., >10 3 )forallcasesofwetsoils. Fig.1.11Empiricalresults: effectivelosstangentdispersionfor9differentsoils. To better understand these results, one may plot the Cole-Cole diagram. The value of effective dielectric loss k ′′ e can be calculated according to (1.21) and using the relation k ′′ e = ε ′′ e ε o . Oneofthecases,Tonzi,isselectedasagoodrepresentativecaseforthisdiscussion and the associated plots are shown in Fig. 1.11 and 1.12 for dry and wet (saturation) cases, respectively. Fewconclusionscanbereachedbyanalyzingbothplots: • The effective loss tangent is defined as the ratio of k ′′ e with k ′ . Therefore, for lossless dielectric materials, values of the y-axis are at the same order as the values of x-axis. In both plots, this is not the case and values of loss tangent significantly higher than then unity can be observed. Therefore, we are dealing with lossy soils for both dry andwetcases. • Any of the plots looks like a semicircle which indicates the existence of multiple polarizationseffectsandnotasingle-timerelaxation. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 50 • Ifverysmallfrequenciesweremeasured,goingdowntomHz,potentiallytheplotfor the wet case would be more similar to the one in Fig. 1.6b. Anyway, the wet soil condition is presenting significant losses due to interfacial, spatial, and double-layer effects. Nonetheless,thesignificantlossesalsoforthedrycasecanbeawarningthat theelectrodepolarizationisdistortingtheresults. Fig.1.12Empiricalresults: effectivelosstangentdispersionfor9differentsoils. Because the analysis with the Cole-Cole diagram only provides partial conclusions, we can plot the effective loss tangent dispersion for all 9 soil cases in order to get some con- clusions and the results are presented in Fig. 1.13. One can verify that this plot provides a betterinsightofwhatishappeningwiththesoilmeasurements,asdiscussedbelow: • For all wet soil cases, it is possible to clearly identify a relaxation frequency. This frequency is relatively distinct for each soil type, varying from 22kHz and 138kHz. There is no clear association between a type of soil and the relaxation frequency. For instance, although NHS and TONZI have very similar soil texture, at saturation condition,onehasarelaxationfrequencyaround110kHz,whiletheotherafrequency of 22kHz. Moreover, with a completely different soil composition (91% sand), the SAN1samplepresentsrelaxationfrequencyaround41kHzatsaturationcondition. • Looking at the dry soil cases, it is not possible to identify a relaxation frequency, but an increasing total loss as the frequency decreases. Because the wet soil presents a relaxation frequency, it means that another form of polarization is dominating the regular polarizations process already discussed. We can raise an hypothesis that the CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 51 Fig.1.13Empiricalresults: effectivelosstangentdispersionfor9differentsoils. electrodepolarization(EP)isthecauseofthedistortionbecauseatfrequenciessmaller than10kHztheEPeffectsarestronger[2,3,73]. • To confirm the above hypothesis related to EP, we return our attention to Fig. 1.9, regarding the dry soil cases. Many of the values are abnormally high for dry soils withwatercontentsmallerthan5%,inparticulartheNHSandNHS-SAT.Thisresult reinforces the EP hypothesis, once it is reported that the EP effect also occurs in dry soilswithrelativelyhighconductivity. Basedonthispreliminaryanalysis,thenextstepattheresearchwastorepeattheexper- iments with diverse values for water content (VWC) levels and with extra-care to maintain thesoildensityrelativelythesameinallVWCcasesforthesoilsampleunderinvestigation. Three of the soil samples are selected as representative cases for 3 classes of soil: SAN1, TONZI, and NHS-SAT. This selection is justified as follows: SAN1 has 91% of sand and only 2% of clay; therefore, polarizations particularly strong in clayey soils will not impact this class of soil. TONZI is an average-case soil, medium loam texture, with 23% of clay and25%ofsand. Finally,NHS-SATisaspecialcaseofhigh-conductivitysoilwhichcanbe verifiedbyinspectionontheprecedingplotsforbothdryandwetcases. Theintuitionbehind this short-list selection is to reduce the universe of variables and to concentrate the model efforts in just 3 soil cases that are initially associated with the concepts of best, medium, andworsecaseforcommunicationpurposesinMI-WUSNs. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 52 Eachofthefollowinglaboratoryexperimentsareperformedtwiceinordertoverifythe existence of outliers caused by the instrumentation, which did not occur. The variance of the measurements between these two set of measurements were smaller than the expected average error given by the manufacturer of about 1%. Moreover, the impedance analyzer was programmed to take 10 measurements for each test-scenario and to send the average number as the final measurement value. The results of the experiments for the first case, SAN1soil,areshowninFigs. 1.14and1.15. Fig.1.14EmpiricalresultsforSAN1soil: Dielectricconstantdispersionfordifferentvolu- metricwatercontent(VWC)levels. AnalyzingtheresultsinFigs. 1.14and1.15,someconclusionscanbereached: • In general, for sandy soils, the relaxation frequency increases with increasing VWC (12−42kHzfor5%-saturation). • However, for very high water content level (i.e.,≥35%), the above relation does not hold. • Nonetheless,theamplitudeofthelosstangentalwaysincreaseswithincreasingVWC. • It is possible to identify a frequency range (i.e., 300Hz-3kHz) where the variation of the dielectric constant k ′ l follows monotonically the VWC variation. Based on CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 53 this observation, it is potentially feasible to design a soil moisture probe operating at low frequencies which only needs two measurements (each one in one distinct frequency)toaccuratelydeterminetheVWCvalue,providedthatasinglecalibration wasperformed. Fig. 1.15 Empirical results for SAN1 soil: Effective loss tangent dispersion for different volumetricwatercontent(VWC)levels. The same experiments are repeated again for TONZI soil and the results of the exper- iments are shown in Figs. 1.16 and 1.17. Analyzing the results in these figures, some conclusionscanbereached: • In general, for soils with a significant percentage of clay, the relaxation frequency (varying from 6-12kHz) does not necessarily increases with increasing VWC. The initial explanation for these is that multiple polarizations are involved, including the EPeffectsandtheimpactofsomeofthemforclayeysoilisdifferentaccordingtothe VWClevel. • Nonetheless,theamplitudeofthelosstangentincreaseswithincreasingVWC. • Again,similarlytosandysoils,itispossibletoidentifyafrequencyrange(i.e.,300Hz- 3kHz)wheresoilmoisturemeasurementscanbeaccuratelyperformed. For the last experiment in this section, NHS-SAT soil is evaluated and the results of the experiments are shown in Figs. 1.18 and 1.19. Analyzing the results in these figures, some conclusionscanbereached: CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 54 Fig.1.16 Empirical results forTONZIsoil: Dielectric constant dispersion fordifferentvol- umetricwatercontent(VWC)levels. Fig. 1.17 Empirical results for TONZI soil: Effective loss tangent dispersion for different volumetricwatercontent(VWC)levels. • Thisclassofsoilistheonethatpresentedthehighestvaluesofthedielectricconstant k ′ ,varyingfromaround4x10 4 (drycase)tomorethan4x10 8 (wetcase). • In general, for soils with a significant percentage of clay, the relaxation frequency (varyingfrom12-47kHz)doesnotnecessarilyincreaseswithincreasingVWC. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 55 • Nonetheless,theamplitudeofthelosstangentincreaseswithincreasingVWC. • Again, similarly to the other soil classes, it is possible to identify a frequency range (i.e.,300Hz-3kHz)wheresoilmoisturemeasurementscanbeaccuratelyperformed. Fig. 1.18 Empirical results for NHS-SAT soil: Dielectric constant dispersion for different volumetricwatercontent(VWC)levels. Fig.1.19EmpiricalresultsforNHS-SATsoil: Effectivelosstangentdispersionfordifferent volumetricwatercontent(VWC)levels. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 56 1.5 SimplifiedSoilDielectricModelforMI-BasedWUSNs A preliminary conclusion from the last section is that the EP effects are potentially exacer- batingthevaluesofk ′ attheempiricalmeasurementsofthesoilsamples. Ingeneral,theEP effects do not significantly impact the effective loss tangent [2, 73, 74]. Therefore, one can use (1.15) to calculate σ e = tanδ e ωε ′ . If that preliminary hypothesis about EP is correct, thenthemeasuredvaluesε ′ aremuchhigherthantheactualvalues. TheinitialgoalofthissectionoftheresearchistoverifyiftheEPeffectreallyoccurred at the soil measurements that were just analyzed and, if positive, to provide a way to math- ematically correct the EP distortion, assuming that the same dielectric instrumentation is maintained. 1.5.1 ElectrodePolarization(EP)Verification Fig.1.20EvidenceoftheEPeffectsonsoilmeasurements. Reproducedfrom[3]. The hypothesis of the EP effects at our LF soil measurements is justified by three rea- sons: a) the comparison of measurements with 2E and 4E instrumentation, b) the existing literature reports, and c) empirical evidence based on the actual measured wireless signal attenuation in soils. Regarding the first reason, soil dielectric measurements are performed CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 57 at 5Hz to 1.3GHz band for different soil settings in [3] by means of both 2E/2T and 4E/4T instrumentations. One of the figures in that work is shown here, by means of Fig. 1.20, to illustrate the EP impact which is associated with 2E instrumentation. Besides k ′ plots, the paper also presents the same behavior occurring to the effective conductivity σ e , as previ- ously discussed. This specific plot considers the case of Bentonite clay. Observe that, for bothmediumandsaturationlevelsofsoilwetness,thereisasignificantincreaseofk ′ atthe left side of the plot, that is, for frequencies smaller than 100kHz. The arrows in that plot indicate the limits of using one or other instrumentation. Note that 4E devices are typically notbroadbandinstrumentsandtheymaybeusedatfrequencyregionsthatEPeffectsoccur. Alsoobservethatthevalueofk ′ at10andhighsoilmoistureisaround10,000accordingto the4Einstrumentationand100,000ifmeasuredwitha2Einstrumentation. Fig. 1.21 Typical software/mathematical technique used to perform EP correction. Repro- ducedfrom[3]. The EP issue is also a very well documented topic, although the majority of papers come from the biochemistry-related areas. A detailed survey regarding the problem and the techniques that have been used to address the EP issue are provided in [84] for further investigation. Fromthatwork,theFig. 1.21isselectedforouranalysis. Observethatasoft- ware/mathematicaltechniqueisusedbasedontheslopeofthecurveforhigherfrequencies before the curve deformation occurs. This technique must be used with care because such approximation is only valid if it follows a physical analysis of the real phenomenon. For CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 58 instance,forsoilmixtures,themacrostructurepolarizations,includinginterfacial,Sternand Doublelayer,arestronglyenhanceatfrequenciesbelow10or1kHz,asshowninFig. 1.20. Therefore, although the technique show in Fig. 1.21 has been properly used in many areas, such as blood analysis, an additional investigation is necessary before applying the tech- nique for dielectric soil measurements. In fact, we adapted this technique in our research specificallyforMIsignalattenuationconsideringdiverseaspects,tobediscussedshortly. The third and last evidence we will consider related to the EP effects comes from field experiments: we want to measure the variation of the wireless UG signal strength by com- paring very dry and saturated scenarios. For this goal, we introduce one of few works that provides an empirical soil dielectric model for low frequencies [16]. The work is based on regression analysis and a curve fitting (method of least squares) related to diverse experi- mentsinvolving5distinctsoiltypes. Ourevaluationhereisbasedonthesecondandfourth equationsofTable4ofthatwork. Atthesecondequationpresentedin[16],thereisasignal error which was corrected later by the author in [85]. Although the authors of the original paper gave careful attention to the EP problem, the knowledge about EP effects was rela- tively limited at that time (1967) and the employed measurement methods (silver paint and non-polarizing electrodes) were later considered as having limited efficiency in mitigating theEPeffects[2]. The following experiment was performed in a residence garden (soil LOA1) and in- volved12-cmmultilayercoils(air-cored)buriedat25cm-depth,10kHz,a0.8Wtransmitter, and inter-node distance of 5m. In one experiment, the soil was naturally very dry (1.6% VWC) and, for the other, the soil was saturated by continuous irrigation of the place. Our goalistouseabasicsignalattenuationmodel(detailsatthenextchapter)andforecasthow the induced voltage at the RX side will be negatively impacted when the soil changes from dry to saturated condition. The results are shown in Table 1.2, where the values due to the MI-Soilmodelwhichwillbepresentedsoonisalsopresentedforcomparison. The values of k ′ andσ e in this table for LOA1 come from the data in the Figs. 1.9 and 1.13, with the help of equation (1.15). Therefore, we can consider that the EP effects are embedded inthisLOA1data. ThesymbolV ind atthetablemeansinducedvoltageattheRX side and this value also depends on the circuitry parameters. V ind is included at the table only for completeness; however, theΔ ind , the variation ofV ind for dry and saturation cases, is independent of the circuit aspects and it is more appropriate for comparison regarding CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 59 Table1.2ElectrodePolarizationImpactonMISignalAttenuation. Model k ′ (dry) k ′ (wet) σ e (dry) σ e (wet) V ind (dry) V ind (wet) Δ ind (μV) LOA1 7.5 9500 1.5E-6 5.3E-2 621.2 485.6 -21.8% Scott’s[16] 12.6 855.9 2.4E-4 4.6E-3 612.1 491.4 -19.7% MI-SOILv.6 81.0 869.4 1.5E-4 4.2E-3 615.0 583.3 -5.2% MI-SOIlv.7 65.0 284.4 2.2E-4 2.1E-3 613.0 594.5 -3.0% (Empirical) - - - - 557.0 551.4 -1.0% exclusively the soil dielectric properties. Accordingly, we are interested in verifying the error in estimating Δ ind . This novel technique based on MI is an effective way to perform initial evaluation of soil dielectric model because the EP effects do not occur in this MI- communicationscenario: thecoilshavegalvanic-isolationregardingthesoilmedium. SomecommentsandconclusionscanbereachedbyanalyzingTable1.2: • We acknowledge that such experiment is not very rigorous, in particular if we con- sider the homogeneity of the water distribution while it penetrated the soil until the saturationconditionwasreached. Althoughweconfirmedsaturationpointsataround 30-cm depth at each TX/RX coil side and also at the middle path between them, clearly the scenario is a limited and small version of what actually occurs in soil af- ter significant natural precipitation. Anyway, the experiment was designed to give an additional feedback regarding the EP effects. The preliminary validation of the pro- posedmodelwillbegivenlaterinthisworkbymeansofadditionalandmorerigorous experiments. • TheMI-SOILv6andv7models,tobediscussedshortly,weredevelopedfewmonths afterthisexperimentbuttheyareincloseagreementwiththeempiricalresults. • ThementionedScott’smodel,althoughwithafairagreementfordrysoils,isevidently pessimistic in relation to the signal attenuation in wet soils. While the RX signal actually only dropped 1% at the experiments, the Scott’s model has the prevision of 19.7%ininducedvoltagedrop. • The worst case was the empirical model based on the dielectric measurements of this soil LOA1 (first line at the table). Both values related to the dry and saturation conditionsdeviatesignificantlyfromtheothermodels. Moreover,theLOA1empirical model’s prevision of 21.8% in induced voltage drop was the most pessimistic of all models evaluated and potentially the EP impedance is influencing the measurements andthesoilattenuationmodelisactuallyusingwrongvaluesfork ′ andσ e . CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 60 • One possibility in relation to the difference between Scott’s and LOA1 models is that, while Scott’s experiments partially mitigate the EP issues, the dielectric mea- surementsforLOA1aremoreimpactedbytheEPeffectsconsideringthefactthatno techniquewasemployedforthispurpose. • For this low frequency (10kHz) and inter-node distance of 5m, it becomes evident thatsignificantvariationsofk ′ arenotnecessarilyassociatedwithsignificantchanges of V ind . This is a novel finding, at least for the MI-WUSN context. For instance, if one compares the results for the pair MI-SOIL v6 and v7, he will see that they are relatively close, while each model has very different values for k ′ . Similarly for the case involving the pair Scott’s and LOA1, the results are close, but the k ′ difference reaches almost one order of magnitude. The same conclusion does not hold for σ e which seems to be the dominating factor for V ind values. At the next section, these conclusionswillbeproperlyexploitedatthecontextofMIsignalattenuationmodels. We can anticipate that the mentioned conclusions in this bullet are only valid for the combinationoflowfrequenciesandrelativelysmalldistances. 1.5.2 EPCorrection Inspired by the conclusions at the last section, the next step in the research is to inves- tigate how an EP-correction technique can be used for 3 soil classes, SAN1, TONZI, and NHS-SAT.Theideaistousetheoriginaldatasetforthesessoilclassesanduseamathemat- ical/software technique similar to the one discussed in [3, 84] and shown in Fig. 1.21. The novelalgorithm,tobedescribednext,isbasedonfouressentialsassumptions/characteristics regardingdielectricsoilmeasurements: (a) Soilmeasurements for frequencies higher than 100kHz, although yet under the influ- enceofEP,arenotstronglyaffectedbytheEPeffects[2,71,73]. (b) EP-correction techniques, in particular the one proposed here, is expected to fail for frequencies smaller than 1kHz [3]. Therefore, its validity is for the 1kHz to the upper-frequency point (UFP), where UFP is the highest frequency available in a 2- port impedance analyzer (IA) used for the soil measurements, provided that UFP is higherorequalthan200kHz. Infact,this200kHzfrequencyistypicalUFPifwecon- sider relatively cheap commercial IA devices. Other UFP options are 1 and 2MHz. (c) EP-correction techniques will potentially fail for the rare cases involving saturated soils with extreme high-conductivity, where the soil dispersion (not related to EP) is CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 61 very significant and may be associated with values for k ′ close to values of apparent dielectricconstantofthesamplebutonlyrelatedtotheEPeffects. (d) TheEP-correctiontechniquebeingproposedherehasbeinginvestigatedinrelationto its accuracy based on MI signal attenuation results only, not on actual deviations in termsofk ′ orσ e . Therefore,theuseofthetechniqueislimitedtothefinalMI-WUSN signal attenuation model. As a result, the proposed soil dielectric model cannot be accuratelyusedasagenericLFsoildielectricmodel. TheEP-correctiontechniquewhichisbeingproposedhere,specificallytobeusedinLF MI soil attenuation models, is based on the linear interpolation between two points at the originalk ′ curve(inalog-log,k ′ vs. frequencyplot). Thisk ′ curverefertothedatasetafter the instrumentation calibration and stray capacitance removal procedures. We assume here thattheparametersforthedatasetarecomprisedof k ′ and DF (dissipationfactor,whichis thesameastanδ e )valuesfordifferentfrequenciesandwatercontent(VWC)levels. Forthe firstalgorithm(EP-correction),onlyk ′ -setisnecessary. Thementionedfrequencypointsfor thelinearinterpolationmustbeinaregionrelativelyfreeofsignificantEPeffects,butatthe same time such area must be closer to the beginning of the frequency region where drastic EPeffectsareobserved. Consideringourinstrumentation,weselectedthefirstpoint,called lower-frequency point (LFP) as 100kHz and the second point, called UFP, as 200kHz. It is important that many measurements points between LFP and UFP are availablefor accurate interpolation. In our research, we performed dielectric measurements using LOG-SWEEP in order to have a sufficient number of points at the mentioned LFP-UFP section of the k ′ curve. Anlinearextrapolationforthek ′ curve,consideringacertainkindofsoilclassandVWC level, is performed for the frequency segment between 1kHz and LFP (e.g., 100kHz). Be- cause we are dealing with for log-log plots, a logarithmic conversion is necessary for such linear techniques, as shown at the algorithm below, to be presented shortly. This process is repeated for different VWC values, maintaining the same soil class. In our investigations, we selected the 5% step for the VWC measures as a balanced decision between accuracy andprocedurecomplexity. TheresultofthisEP-correctionprocedure,detailedatAlgorithm 1,isasetofk ′ curveswiththeEP-correction. However,eachcurveinthisclassofsoilisstill associated with a specific value of VWC, such as 1%, 5%, 10%, and so on. For a MI soil attenuationmodel,anotherprocedureisnecessarytoallowtheestimationoftheattenuation foranotherVWCvalues,suchas3.7%,andthisisthetopicofthenextsection. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 62 Algorithm1ElectrodePolarization(EP)CorrectionProcedurefork ′ Require: k ′ -set: dielectric constantvaluesfora specific volumetricwatercontent(VWC) leveland foraspanoffrequenciesbetween1kHzandanupper-frequencypoint(UFP),suchas200kHz Require: LFP:lower-frequencypoint(UFP),e.g.,100kHz Require: UFP:upper-frequencypoint(UFP),e.g.,200kHz Ensure: Outputisk ′ setwithnewEP-correctedvaluesforthefrequencyregion1kHz-UFP f1←lower-frequencypoint(LFP),e.g.,100kHz f2←upper-frequencypoint(UFP),e.g.,200kHz k1←valueofk ′ associatedwithfrequency f1 k2←valueofk ′ associatedwithfrequency f2 m← logk1−logk2 log f1−logf2 b←logk1−mlog f1 index←0 whilenotatendofthe (k ′ , f)datasetdo index←index+1 k ′ index ←10 b (f index ) m endwhile The results of the EP-correction for k ′ is illustrated in Fig. 1.22, where we changed the Algorithm 1 only to extend the correction down to 100Hz, than 1kHz. Although the case is specific for TONZI soil, 20%VWC, all the other curve cases have the same pattern and areomittedforconvenience. Consideringthephysicalmeaningbehindsuchapproximation, it is important to highlight again that the accuracy of the process is strongly impacted at the region around 1kHz and smaller frequencies. Therefore, a different approach would be necessary to address such very low frequency range. Fortunately, for MI-WUSNs, 1kHz is already a very small frequency in terms of associated application bandwidth and it is not expectedthatthesub-kHzfrequencyrangewillbeusedonedayforthisspecificscenario. The previous procedure must be repeated multiple times, according to different VWC levels available at the dielectric measurements data set. The naive approach we had just after concluding Algorithm 1 was to try to find a function with parameters frequency and VWC that could capture the dielectric behavior under unknown (not measured) VWC val- ues. ManydifferentmathematicalfittingapproacheswereusedbymeansofMATLABtool. Unfortunately, when we tried to apply such preliminary soil model to the soil attenuation model,wefounderrorshigherthan90%fortheRXinducedvoltage. Thecomparisoninthis caseinvolvesusingtheexactEP-correctedcurveforaspecificVWC,whichistheoutputof theAlgorithm 1, with thecase we use the genericfitting equation curveproduced afterthat algorithmforanyvalueofVWC.Weendedupconcludingthatthereiserrorattheanalysis oftheproblem,asexplainednext. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 63 Fig.1.22EP-correctionforthedielectricconstantk ′ accordingtoAlgorithm1. Considerthevalueoftheeffectiveimaginarypartofthepermittivitybasedon(1.20): ε ′′ e =ε ′ tanδ e (1.57) k ′′ e =k ′ tanδ e (1.58) Sofar,basedonthefactthatEPeffectsdonotstronglyimpacttheeffectivelosstangent (tanδ e =DF), the EP-correction only considered k ′ . This procedure is sufficient if this pa- rameter is enough for the analysis of the problem, which is the case for some biochemistry applications. However, this is not the case for MI signal attenuation models. If just one VWC case is considered, the linear interpolation for the values of k ′ automatically corre- spond to EP-corrections for k ′′ e by using (1.58) and using the mentioned fact about the DF andEPrelation. However,whenweinterpolatevaluesofk ′ (realplane)fordifferentVWCs we are in fact significantly distorting the values of DF which are supposed to be the same, before and after the EP correction. This is what was happening with our first approach for theVWC-frequencygenericfittingmodel. To solve the above problem, we hypothesize based on the physical interpretation of the problem that the solution would be the realization of a second fitting process at k ′′ e values (imaginary plane) once these values were recalculated using the new sets of EP- corrected k ′ curves for different VWC cases. In other words, what this process is actually CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 64 doing is conserving DF values as close as possible to their original values. As a result, a relatively accurate MI soil attenuation model is achieved, as will be discussed in detail at the next chapter. The second EP-correction procedure, now associated with k ′′ e is presented inAlgorithm2. Algorithm2ElectrodePolarization(EP)CorrectionProcedurefork ′′ e Require: k ′′ e -set: effectivedielectriclossvaluesforaspecificvolumetricwatercontent(VWC)level andforaspanoffrequenciesbetween1kHzandanupper-frequencypoint(UFP),suchas200kHz Require: k’-corrected: setofEP-correctedk ′ values(outputofAlg.1) Require: DF-set: dissipation loss (or effective loss tangent) values for the mentioned frequency range Ensure: Outputisk ′′ e setwithnewEP-correctedvaluesforthefrequencyregion1kHz-UFP index←0 whilenotatendofthe (k ′′ e ,DT)datasetdo index←index+1 k ′′ e,index←k ′ index endwhile 1.5.3 SimplifiedMI-SoilDielectricModelforMI-WUSNs Our ultimate goal for the next chapter is to have a MI soil attenuation model. From the user’s perspective, given the soil class, the frequency, and the water content (VWC%), the model provides the soil path signal attenuation (in dB) or, more conveniently, the RX in- duced voltage level (typically, in μV) provided that additional circuitry details are given. Internally, such model will eventually need to have the complex propagation constant for thesoilmediumγ,givenby(1.11). Thatequationrequirestheangularfrequencyω,σ e ,and ε ′ . The angular frequency ω is known, but σ e and ε ′ are also function of VWC, besides frequency. This is the point where a soil dielectric model is necessary and this is the goals ofthissection. From the two algorithms presented at the last section, we already have a way to calcu- late EP-corrected version ofε ′ andε ′′ e , given the soil class (for what we collected dielectric measurements), the frequency, and certain values of VWC. The conversion from ε ′′ e to σ e is straightforward by using the relation σ e = ωtanδ e ε ′ based in (1.15). The values for tanδ e ε ′ =DF directly come from the empirical measurements and ε ′ = kε o . Although it seemsthatallaspectstosolvetheproblemareaddressed,westillneedtorememberthatthe available EP-corrected data sets for k ′ and k ′′ e are valid for certain discrete frequencies and CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 65 VWClevels. Therefore,amoregenericempirically-basedsoildielectricmodelisnecessary. Itwasalreadyhighlightedthatourfirstattempttosolvethisproblemwasnotsuccessful because we initially did not consider k ′′ e at the EP-correction procedure, but now this issue was addressed. The next step is to find a way to reach an interpolated fitting curve which has VWC as an parameter besides the frequency. From now on, we will always assume thatthesoilclassisfixedforthediscussionand,naturally,alltheproceduresinthischapter canberepeatedagainforanynumberofsoilclasses. Thementionedfittingproblemwould be very complex to be solved if any of the k ′′ (also k ′′ e ) dispersion curves, each one related to a specific VWC, has non-linear behavior. Fortunately, the EP-correction procedure also producedlinearcurves. Forinstance,a k ′ curverelatedto20%VWCforTONZIdatahasitsspecificvaluesof b andmintheequationk ′ =m∗ f +b,wheremistheslopeofthelineandbisthey-intercept (in this case, the k’-intercept). Therefore, if we have a collection of 8 curves for k ′ , one for VWC(e.g.,5,10,..40%),webasicallyhavetwo8-elementvectors. Next,weplotacurveof VWC vs. b-vector and another plot for VWC vs. m-vector. Then, the goal is to any avail- able linear interpolation tool to interpolate these two curves, independently. The process is thenrepeatedfor k ′′ e . Whatthisinterpolationproducesarevaluesof bandmgivenVWCas input: one pair for k ′ and one pair for k ′′ e . Next, we use the logarithm conversion relation to conclude the problem: k ′ (f,VWC)=10 b ∗ f m and k ′′ e (f,VWC)=10 b ∗ f m , where the pair b,m of k ′ is distinct from the pair for k ′′ e and these pairs are the output of the linear interpo- lationjustdiscussed. Aftersomeevaluationswithdifferentpolynomial-fittingalgorithmswithdifferentorders for b and m fittings, we observed that the most accurate results at the final application (MI signal attenuation) were achieved by simply having these fitting curves as lines connecting thebestcase(lowVWC)withtheworstcase(highVWC).Inalmostallevaluatedcases,the Soilattenuationmodelusingthisfittingmethodpresentedaminorbiastowardspessimistic previsions(seeTable1.2),whichisaconservativedesignapproach. Thebandminterpola- tioncurvesfork ′ (TONZIcase)areshowninFigs. 1.23and1.24. Thebandminterpolation curvesfork ′′ e (TONZIcase)areshowninFigs. 1.25and1.26. Thevaluesofthecoefficientbandmforallsoilclasses,andtheoverallequationsrelated to each MI-Soil model are listed below and the comparison between the original k ′ and k ′′ e CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 66 Fig.1.23Fittingforthebcoefficientforthedielectricconstantk ′ ,TONZIclass. Fig.1.24Fittingforthemcoefficientforthedielectricconstantk ′ ,TONZIclass. curves and the EP-corrected ones are shown in Figs. 1.27 and 1.28 and such curvey obey the equations below. It is important to remember that although only 8 corrected curves are being shown for each case, the MI-Soil models based on the listed equations may use any value for f and VWC (i.e., infinite number of curves), provided that these values are inside CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 67 Fig.1.25Fittingforthebcoefficientforthedielectricconstantk ′′ e ,TONZIclass. Fig.1.26Fittingforthemcoefficientforthedielectricconstantk ′′ e ,TONZIclass. the range of frequencies and VWC values related to the interpolation processes. The accu- racyofthesemodelscanbeverifiedattheirfinalapplication,theMIsoilattenuationmodel, whichisthetopicofthenextchapter. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 68 b real (v6) =0.052(VWC)+3.4 (1.59) m real (v6) =−0.0054(VWC)−0.385 (1.60) b imag (v6) =0.0278(VWC)+4.46 (1.61) m imag (v6) =0.00380(VWC)−0.526 (1.62) ε ′ (v6) =10 b real (v6) f m real (v6) (1.63) ε ′′ e(v6) =10 b imag (v6) f m imag (v6) (1.64) b real (v7) =0.0385(VWC)+3.375 (1.65) m real (v7) =−0.0049(VWC)−0.398 (1.66) b imag (v7) =0.0195(VWC)+3.965 (1.67) m imag (v7) =0.00225(VWC)−0.352 (1.68) ε ′ (v7) =10 b real (v7) f m real (v7) (1.69) ε ′′ e(v7) =10 b imag (v7) f m imag (v7) (1.70) b real (v8) =0.054(VWC)+4.2 (1.71) m real (v8) =−0.0051(VWC)−0.535 (1.72) b imag (v8) =0.0362(VWC)+4.72 (1.73) m imag (v8) =0.00425(VWC)−0.595 (1.74) ε ′ (v8) =10 b real (v8) f m real (v8) (1.75) ε ′′ e(v8) =10 b imag (v8) f m imag (v8) (1.76) wherethevolumetricwatercontent(VWC)isgivenin%andfrequency f isgiveninHz. CHAPTER1. MI-SOIL:SOILDIELECTRICMODELFORSUB-MHZWUSNS 69 Fig. 1.27 Comparison between the measured values for k ′ and the resulted curves due to MI-SOILv6model(TONZIsoilclass). Fig. 1.28 Comparison between the measured values for k ′′ e and the resulted curves due to MI-SOILv6model(TONZIsoilclass). Chapter2 TowardanAccurateSub-MHzSoilDiel. Spectroscopy At the previous chapter, an empirical sub-MHz soil dielectric model is proposed. As will bedemonstratedsoonatthischapter,suchmodelpotentiallyhasenoughaccuracyregarding its original application: a MI-soil signal attenuation model and this one will be developed at Chapter 3. Nonetheless, the main drawback of this soil model is the potential errors in frequencies smaller than 50 kHz, in particular at 1-10 kHz range. This is the case because the adopted strategy of the MI-soil proposed model was to focus on the measurements per- formed above 100 kHz and, based on this set of data, extrapolate on the impedance values below100kHz. Suchapproachisactuallynotnovelandithasbeensuccessfullyappliedin manybiologicalscenariosformanydecades. Theworkin[60]isagoodreferenceregarding the origins of such technique. Nonetheless, the novelty of the approach proposed in Chap- ter 1 is the application of such technique for soils. The validity of the technique assumes that no dispersion phenomenon actually occurs for the material under analysis and for the frequencyrangeofinterest. While it is possible that strong dispersion effects do not actually occur for certain com- binationsofsoilstypes,moisturelevels,andfrequencyranges,theapproximationapproach used at the proposed soil dielectric model will indeed produce significant errors (particu- larly for ε’, as will be discussed soon) if the specific soil under analysis presents one or morerelaxationtimesatthelowerfrequencyband. Fortunately,aswewillverifyatChapter 3 by means of extensive tests at 10 kHz, significant errors are not detected regarding the proposed MI signal attenuation model in conjunction with the soil dielectric model derived inChapter1. Fortheseexperiments,differentoutdoorsitesareemployedwithmultiplesoil 70 CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY71 conditions. Therefore, considering those preliminary empirical results, it is apparent that theproposedsub-MHzsoildielectricmodelcanbeusedforMIsignalattenuationmodels. As a consequence, one can consider the possibility of using this soil model also for soil dielectricmeasurementsingeneral. Unfortunately,weanticipatethattheproposedmodelin Chapter1doesnothavesufficientaccuracyforthisspecificpurpose. Inthischapter,westart with an explanation for this statement and the steps toward the development of an accurate sub-MHz soil dielectric spectroscopy instrumentation are discussed. Next, the EP problem isdeeplyinvestigatedandasurveyofdifferentmethodstoeliminatetheIPimpedancefrom the dielectric measurements is provided. A novel approach to identify the EP impedance from the impedance measurements is presented. Finally, the preliminary analysis of the empirical results associated with this technique and the required special instrumentation is presented. 2.1 TheDielectricMeasurementInversionProblem This research work primarily belongs to the Magnetic induction-based Wireless Commu- nication area. Nonetheless, at the sub-MHz, there are strong bounds connecting this area totwootherones: ElectrochemicalImpedanceSpectroscopy(EIS)andMagneticinduction Impedance Spectroscopy (MIS). In this section, these relationships among these areas will bediscussed. To better understand how a soil dielectric model can be accurate enough for a complex signal attenuation model but not sufficient for simple dielectric measurements, numerical examples are provided. The Table 2.1 presents the simulated RX induced voltage V C es- timations based on our MI-Soil signal attenuation model (to be studied in Chapter 3) for a distance of 15m, disregarding any contribution due to circuitry aspects, and considering differentvaluesforfrequency,dielectricconstantk ′ andeffectiveconductivityσ ef f . TheV C valuesarenormalizedregardingtheover-the-air(OTA)mediumcase. Themaingoalofthis evaluation is to determine what would be the variation of V C if a soil dielectric model has negligible estimation error regardingσ ef f but almost one order of magnitude error regard- ingk ′ . Thisanalysisisagenericoneandcanbeappliedforthestudiesofanysub-MHzsoil dielectric model. As shown in Table 2.1, even with a large estimation error regarding the dielectric constant of soils, the error on the MI signal attenuation is smaller than 4% for all CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY72 Table2.1Signalattenuationerrorduetothedielectricconstantestimationerror. Freq σ ef f ε ′ r ork’ V C (OTA-normalized) V C error AvgV C error 5kHz 0.02 60 0.709452 - - " " 600 0.709989 0.08% - " 0.002 60 0.906015 - 0.40%(for5kHz) " " 600 0.909237 0.36% - " 0.0002 60 0.970999 - - " " 600 0.978518 0.77% - 10kHz 0.02 60 0.598711 - - " " 600 0.599785 0.18% - " 0.002 60 0.868386 - 0.81%(for10kHz) " " 600 0.875369 0.80% - " 0.0002 60 0.961053 - - " " 600 0.975020 1.45% - 30kHz 0.02 60 0.359176 - - " " 600 0.359534 0.10% - " 0.002 60 0.776186 - 1.34%(for30kHz) " " 600 0.798389 2.86% - " 0.0002 60 0.940018 - - " " 600 0.950761 1.14% - 50kHz 0.02 60 0.228748 - - " " 600 0.224772 1.74% - " 0.002 60 0.715744 - 2.51%(for50kHz) " " 600 0.739387 3.30% - " 0.0002 60 0.929070 - - " " 600 0.905965 2.49% - analyzed cases. Also, this error becomes smaller as the frequency decreases. These results have a good agreement with the empirical analysis to be discussed at the next chapter and they indicate that a hypothetical soil dielectric model with such level of inaccuracy is still potentially adequate for fair MI-WUSN communication models. Nonetheless, considering the inverse problem in this scenario, the results show that a sub-MHz soil dielectric model cannot be fully validated for generic soil dielectric measurements only based on signal at- tenuationmeasurementsofaMIwirelesscommunicationcaseortestbed. As will be verified in Chapter 3, the phase information is dropped from the final ex- pression for V C , the induced voltage level at the RX side. This is the case because only the magnitude of the signal is actually important for basic signal attenuation models and communicationmodulationschemes. Therefore,byjustanalyzingthemagnitudeoftheRX CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY73 signal it is not possible to fully characterize the electrical properties of the soil. In other words,atestbedinvolvingMIcommunicationinsoilwherejustthemagnitudeofthesignal is evaluated does not solve the inverse problem of finding k ′ andσ ef f for that soil medium. This discussion is illustrated in Fig. 2.1(b) where one can observe that, as expected, the electricalpropertiesofthesoilarerequiredtoestimatethesignalmagnitudeattheRXside. However, the inverse of this statement does not typically hold and the full solution of this inverseproblemcanbeobtainedbyadoptingtheinstrumentationsetupshowninFig. 2.1(c), wherealock-inamplifierisusedtodetectbothmagnitudeandphaseinformationattheRX signal. RegardingtheFig. 2.1(c),theassociatedinstrumentationareaiscalledMagneticinduc- tionImpedanceSpectroscopy(MIS)and,unfortunately,veryfewMISdevicesandresearch worksareavailable. OneexampleofsuchMISinstrumentationistheHPE5050Acolloidal probe (to be connected to an impedance analyzer). This solution was temporarily available at the market but it became discontinued. The main challenge in MIS is not the accurate measurement of phase (nonetheless, this one is still critical), but the very tiny variation of the signal magnitude at the RX side if TX (excitation) and RX (probe) coils are separated by only few centimeters of millimeters. The theory behind this observation is presented in Chapter 3 when the Table 3.1 is discussed. For the current discussion, it is sufficient to statethatthecombinationoflowfrequenciesandsmalldistancesmaketheMIsystemvery resilient (or insensitive) to the electrical properties of the medium. Therefore, the achieve- ment of good measurement sensitivity on MIS systems that are required to operate at low frequencies is still an open research topic. For instance, the mentioned HP E5050A oper- atedat75kHz-30MHzandformanydielectricspectroscopystudies,the75kHzlimitisstill consideredtoohigh. AlthoughthedielectricmeasurementsdiscussedatChapter1areperformedbymeansof commercialimpedanceanalyzers(orLCRdevices),thebasicinstrumentationsetupusedin thischapterisdifferent,asshowninFig. 2.1(a)andthepurposeofthischangeistoincrease the measurements accuracy. Nonetheless, this setup is essentially embedded at the major- ity of the available commercial impedance analyzers. However, the main advantage of the illustrated scenario is the possibility to change many of the input variables, such as source voltage,currentlevels,andvaluesforR1. Moreover,additionalequipmentscanbeattached to the reference resistor R1 in order to measure the total harmonic distortion (THD) and this provision is paramount for the proposed EP-identification technique proposed in this CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY74 chapter. ObservethatthelabelforFig. 2.1(a)isElectrochemicalImpedanceSpectroscopy(EIS). Iftheelectrode-sampleinterfacedoesnothaveanyimpactonthemeasurements,onecould properly use the electrical term for the instrumentation technique. This would be the case for any ideal solution that fully eliminates the EP effects. Nonetheless, such idealistic in- strumentation does not exist and the maximum that a EIS solution can achieve is a very strong mitigation of the EP effects to a level that the EP impedance is negligible in com- parison with the impedance of the bulk sample under analysis. Also, observe in Fig. 2.1(a) an important aspect requirement that is not present in MIS solutions: the dielectric cell and its specific geometry. Such requirement may also be source of potential errors in EIS mea- surements: the accuracy of the measurements are strongly dependent on how the physical measurements of the dielectric cell are precisely performed. Moreover, the measurements are greatly impacted by the existence of any air gap (or other insulation) between the elec- trode and the sample. On the other hand, MIS is a non-contact dielectric measurement and theexistenceoftinyinsulationlayersbetweenthecoilsandthematerialunderinvestigation typicallyhasnegligibleimpact. Based on the previous information regarding EIS and MIS systems, one can conclude thattherearechallengesassociatedtobothtechniques,althoughtheformeroneisthetypical choice in dielectric spectroscopy. Our ongoing work beyond the scope of this dissertation is to connect the links between sub-MHz dielectric instrumentation and communication. Specifically,wealreadystartedtheinvestigationofhowtoevolvethesolutioninFig. 2.1(b) into(c). SomeofthedetailsofthisongoingprojectarediscussedinAppendixA.Weexpect to achieve good sensitivity for a MIS solution even at frequencies smaller than 10 kHz by employingrelativelargedistances,rangingfrom30to200cm. If a very accurate MIS solution is eventually achieved, one can conclude that this one canbeperceivedasacompletesubstituteforanEISsystem. However,webelievethatsuch expectation is not correct for two main reasons. First, EIS and MIS techniques are actu- ally a dual system. It means that one approach is an excellent way to verify the validity of the measurements of the other technique considering the fact that they do not share many aspects. Second, the EP effects which are only present at the EIS systems may be very valuable at the future and it is necessary time and data collection/analysis to confirm if this expectation is valid. As well known, the EP effects are typically considered instrumenta- CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY75 tion artifacts that must be eliminated. However our vision is different: the EP data must be identified and properly isolated in order to achieve correct dielectric measurements of the bulk material. Nonetheless, the EP data by itself can potentially contain very important information about the composition of the bulk material. Therefore, for the investigation of mixtures, including soils, it is possible that the EP data provides insights that are not pos- siblebyonlyusingMISsystems. Atthenextsection,theEPeffectswillbestudiedindetail. 2.2 TheElectrodePolarization(EP)Effects The dielectric instrumentation literature is extensive and no attempt is made at this chap- ter to be comprehensive regarding this topic. In particular, the EP topic was superficially addressed at the previous chapter and additional information is added here in order to un- derstandtheimportanceofanEP-identificationmethodology. Withoutawaytoremovethe EP impedance from the EIS measurements, such data can be strongly distorted [86]. The goalofthissectionistounderstandthecausesofsucheffectsandhowthementionedmea- surementdistortionoccurs. The nature if the EP effects is related to the presence of a charge distribution at the electrode-sample interface, assuming that the material under analysis contains ions. The mentioned charge distribution at the interfaces are due to the mobility characteristics dif- ferences between the electrons passing through the wires of the instrument and the ions movement inside the sample. Usually, it is easy to understand why the EP effects occur using an example where a DC current is applied. However, even for AC excitations, the EP effects may also be present depending on the material and how small the frequency is. The reason why charge distributions may be formed when, for instance, a sinusoidal signal at a low frequency is applied to the sample lies on the time period necessary to change the polarizationofthedistributionsaccordingtothephaseofthesignal. Ifthefrequencyofthe signalis high enough, thetime necessary toestablish thechargedistributionatthe material is higher than period of the signal. Therefore, there is not sufficient time for the realization ofthechargedistributionandtheEPeffectswillnotoccur. In general, the literature mentions the occurrence of the EP effects for frequencies smaller than 100 kHz [60, 86–88], but such frequency boundary actually depends on the electrical properties of the material, in particular its electrical conductivity. For instance, a CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY76 higher conductivity material, such as a wet soil sample, will start to present the EP effects at a frequency higher compared to other sample with moderated conductivity, such as soils with 10% of moisture. Moreover, it is possible that the EP effects will not occur (or they will not impact the measurements) if the sample material does not provide a way to allow theionsmovement,suchasinmanysolidsorwhenverydrysoilsamplesareinvolved. Sim- ilarly, very good dielectric materials will not present EP effects. On the other hand, with the addition of water in soil samples, the EP effects may be promptly observed due to the formationofanelectrolytemixture. Redox (oxidation-reduction) reactions involving the electrode strongly complicate the analysis of the EP effects because multiple chemical processes are occurring simultane- ously. Moreover, the redox reactions imply that the electrode itself is being physically affectedwithtimeandtherepeatabilityoftheexperimentsbecomesdifficulttobeachieved. Therefore, hereafter, it is assumed that redox does not occur at the electrodes. In fact, this assumption is typicallyvalidbecause it is usual in dielectric measurements to employelec- trodesthataremadeofaninertmaterial,suchas316-gradestainlesssteel,thepreferredone inourempiricalevaluations. ThefirststudiesregardingthenatureoftheEPeffectsarerelatedtotheeffortofmeasur- ingtheresistivityofelectrolytesattheendof19thcentury,whenmanyresearchersobserved a significant potential drop at the electrodes whenever an alternating current was employed for such measurements [89]. The reason for this phenomenon was not cleat at that time, but Kohlrausch (1899) was the first to confirm that the issue was significantly minimized if platinum electrodes were employed [86, 89]. Perhaps the first one to try to model that ob- servedelectrode-electrolyteinterfacephenomenonwasWarburg(1899)[58]whoproposed the so-called Warburg impedance: a RC series model where the values of both R andC are frequency-dependent. According to Warburg, if a very low current density is assumed, both values of R and C vary inversely as the square root of the frequency with a constant phase angle of 45 o [58, 86], as shown in Fig. 2.2. Observe that the constant k shown in Fig. 2.2 is dependent on the materials of both electrode and electrolyte. Simulated results for a specific case of k is shown in Fig. 2.3. Soon, we will discuss how small or high values forC and R in such figuresimpactthedielectricmeasurements. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY77 Based on empirical evaluations, Frickle (1932) [90] realized that the Warburg model doesnotholdinmanyscenarios. Therefore,heproposedanextensiontotheWarburgmodel whichisshowninFig. 2.4,whicharecallingheretheFrickemodel. Inthismodel,twocon- stantsassociatedtotheimpactofthefrequencyontheEPimpedanceareemployed: aslight modification of k and a new factor m. When m=0.5, the Fricke and Warburg model have a match,asshownbythesimulatedresultsfortheFrickemodelinFig. 2.5. Observe that both mentioned EP models do not take into account the DC case [89]. In other words, if the real interface model includes such capacitanceC in series with the sam- ple (in its both sides), then a direct current could not pass through the electrolyte, which is not the real case. Considering this limitation, Geddes and Baker (1968) [91] added the so-called Faradic resistance (R f ) shunting the Warburg resistance and capacitance series circuit where R f is considered specifically for the DC case. The same work also proposed theadditionofthehalf-cellpotential(E 1/2 )inserieswiththeoverallcircuit. In the studies in this chapter, we will not explicitly consider both R f and E 1/2 because wewillintentionallyavoidfrequenciesbelow100Hzorscenarioswhere E 1/2 dramatically changesduetotherectificationprocess. Thisprocessoccursduetothenon-linearEPeffects when, for instance, the current density through the electrode is very high. In fact, the main goal of the novel EP-identification method proposed in this chapter is to identify this EP non-linear region and potentially maintain the dielectric measurements close to the begin- ning of this region but definitely out of it. In this sense, E 1/2 may be considered as a fixed factorwithnegligiblevariationsforthedynamicrangeoftheinstrumentationparameters. As evident by the long-term studies involving EP, such phenomenon is not a kind of research topic that is still questioned regarding its nature and effects. For instance, Pol- lak (1896) [89] quickly patented what we know as aluminum oxide electrolytic capacitor as soon he figured out the huge value of capacitance it is possible to achieve involving a metallic electrode/interface and electrolyte solutions. Moreover, many industrial process involving the assessment and physical protection of parts also employ EP studies. How- ever,whenthegoalisdielectricmeasurementsinlowfrequencies,theEPeffectsareusually considered a huge issue, an instrumentation artifact to be eliminated, if possible. Next, we will discuss how the mentioned seriesC and R at the electrode interface indeed can distort impedancemeasurements. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY78 Before proceeding with the explanation about the effects of EP on the dielectric mea- surements, it is important to highlight that a system can have heterogeneous electrodes, for instanceonewhichhasminimumEPeffectsandotheronewithmediumorhighEPeffects. Inthissense,RandC(relatedtotheEPimpedanceattheinterface)havedifferentvaluesfor each electrode. At the electrochemical literature, one electrode may be called the reference andtheother workingone,respectively. Duringourempiricalwork,twokindsofdielectric cells were used, emulating ideal parallel-plate and coaxial capacitors. The former poten- tially has homogeneous EP effects in its electrodes because they have the same dimensions and shapes. However, the coaxial-capacitor has heterogeneous EP effects that can be very different between the electrodes. This is the case because the internal pin has a higher cur- rentdensityduetoitssmallersurfaceareacomparedtotheouterelectrode,asshowninFig. 1.8. Therefore, when dealing with coaxial-capacitor dielectric cells, it is important to give attentionprimarilytothedimensionsoftheinnerelectrodebecausethisonewillbethefirst to enter into the EP non-linearity regions if the applied current significantly increases. For the following explanations, consider that C and R reflect the net effect of both electrodes, nomatterthescenario. Consider the case in Fig. 2.5 for two distinct frequencies: 1 and 100 kHz and value of m=0.5, that is, both Warburg and Fricke models can be applied with the same results. In order to illustrate the effects of the EP impedance on the dielectric measurements, we will assume two parallel-plate dielectric cells with the same plate area but with different distancesbetweentheirplates. Inthisway,thecellconstantsareK1=30andK2=60m −1 ,for the smaller and bigger dielectric cells, respectively. Next, values for the dielectric constant and effective conductivity of the sample are used for the evaluation observing that they are based on actual EP-free measurements. For this example, we use the values k ′ =ε ′ r =500 and σ ef f =0.01 S/m for 1 kHz and k ′ =ε ′ r =50 andσ ef f =0.007 S/m for 100 kHz. Simulations are performedaccordingwiththefollowingsteps: • For each of the 4 combinations involving the frequencies f 1 =1 kHz and f 2 =100 kHz andthecellconstantsK1=30m −1 andK2=60m −1 ,calculatethetruevaluesoftheca- pacitanceandresistanceofaparallelRC circuitwithvaluesfor R parallel sample andC parallel sample using(1.53)and(1.54)andalsotherelationK x =d/Aonthoseequations. • Convert R parallel sample and C parallel sample into R series sample and C series sample by using well-known circuit theoryexpressions(theseexpressionsalsoappearattheAlgorithm3inthischapter). • CalculatetheapparentvaluesofRandC foraseriesRC circuitmodelwherebothEP CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY79 impedanceandtruesampleimpedancevaluesareinvolvedinthisway: R series appar. =R series sample +R series EP C series appar. = C series sample ·C series EP C series sample +C series EP • ConverttheapparentvaluesofRandCforaseriesRCcircuitintovaluesforaparallel RCcircuit: R parallel appar. andC parallel appar. . Again,usewell-knowcircuittheoryexpressionsfor thisseriestoparallelRC circuitconversion. • Calculate the values of the apparent dielectric constant (k ′ (appar.) ) and effective con- ductivity(σ (appar) ef f )ofthesamplebasedonthevaluesR parallel appar. andC parallel appar. . ThesimulatedresultsaretabulatedinTable2.2. Theconclusionsbasedontheseresults arelistedbelow: (a) Thedielectricmeasurementserrorsindeedincreasewithadecreasingfrequency. This resultisexpectedbasedonthephysicalinterpretationoftheEPphenomenon. (b) The error related to the real relative permittivity (or dielectric constant k ′ ) is much largerthantheconductivityerror. Therefore,itisusualtofindmanyEPrelatedworks thatdisregardtheanalysisoftheeffectiveconductivityerror. Nonetheless,inthisdis- sertationresearchwork,weavoidtosimplifytheanalysisoftheEPchallengesandthe conductivity error analysis is not removed. In fact, at the proposed EP-identification method the conductivity values are verified in order to identify potential solutions. On the other hand, while comparing results regarding the mitigation/elimination or identification of the EP effects, it is usual to primarily compare the k ′ results because thesearetheoneswithexpectedoneswithverylargeerrors. (c) If the dimensions of the dielectric cell increases (i.e., the cell constant K increases), potentiallytheEPeffectscanbemitigated. (d) The analysis of the problem is complicated by the fact that back-and-forth RC se- ries/parallelconversionsarerequiredtoobtaincorrectinsightsoftheEPchallenges. Regarding the latter aspect, it is worth to mention that the majority of the dielectric works convert the initial problem into an admittance circuit model in order to mathemati- cally simplify the exposition of the topic. Nonetheless, this is not the approach adopted in CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY80 this dissertation work. Although the way one models the circuit problem (i.e., by means of aseriesorparallelRC circuit)ismathematicallyirrelevant,itisourbeliefthatthecloserwe canadoptamodelregardingtherelatedphysicalrepresentation,itbecomeseasiertocapture the challenges of the problem. The drawback is the need of large number of mathematical manipulationsinordertosupportthisapproach. Inthissense,whenweuseadielectriccell, the physical model is indeed a lossy capacitor, that is, a parallel RC circuit. Accordingly, thevaluesforR parallel sample andC parallel sample appearatthetableasinputvalues. Ontheotherhand,as discussed previously, in general the proposed EP models adopt a series RC representation which is the closest one to the reality. To this end, the Table 2.2 also allocates input fields forthevaluesofR series EP andC series EP . However, the combined effect of both impedances (EP and sample) requires the adop- tion of a temporary single RC model for the calculations. We conveniently use a series RC circuit for these calculations in order to achieve an apparent impedance which can be promptly compared with the output of regular impedance instrumentation (i.e., magnitude and phase of the impedance) without the need of any additional conversions. Nonetheless, if a commercial impedance analyzer is employed as the single instrument, such discussion may be irrelevant because the device potentially can display the results in series or parallel arrangement with high precision. Finally, once the apparent values forC and R are calcu- lated or measured for a series RC configuration, the conversion of these values to dielectric entitiesrequiresagainaseries-to-parallelconversion. Asexpected,thefinalvaluesfork ′ andσ ef f arestronglydistortedbytheEPimpedance in many cases, as one can conclude by analyzing the cases in Table 2.2. The most critical observed scenario is the first one with combines small values for both frequency and cell constant K. For this case, the apparent dielectric constant is more than 3 times the actual valueofthesample. Atthenextsectionwewilldiscusswhatareintuitivewaystoeliminate the EP effects, what are the state-of-the-art in this area, and also the drawbacks of each so- lution. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY81 Signal generator CH1 CH2 Lock-In amplifier Ref Sig Electrode Soil Electrochemical Impedance Spectroscopy (EIS) V signal (MAG, PHA) ´ , of soil (a) TX Coil RX Coil MI node Soil MI node Magnetic induction-based Wireless Communication RX signal (MAG) ´ , of soil (b) Signal generator CH1 CH2 TX Coil RX Coil Lock-In amplifier Ref Sig Front-end Soil Front-end Magnetic induction Impedance Spectroscopy (MIS) RX signal (MAG, PHA) ´ , of soil (c) Reference Resistor R1 V R1 Dielectric cell geometry R1 } RX signal (MAG) ´ , of soil Fig. 2.1 Conceptual vision of comprehensive dual EIS-MIS system for sub-MHz dielectric spectroscopy where a MI-based wireless communication solution is an important compo- nent. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY82 C R R = (2 pi f k) sqr(f) -1 . . . . C = k / sqr(f) tan = X / R = 1 C Sample Fig.2.2EquivalentcircuitmodelsfortheEPimpedance: Warburgmodel(1899). 10 Hz 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 1 uF 100 uF 10 mF 1 ohm 100 ohms Electrode polarization effects: Warburg impedance model, k=0.0002 Frequency Warburg resistance R / capacitance C Warburg capacitance Warburg resistance Xc = R R C Fig.2.3SimulatedresultsfortheWarburgEPmodelandk=0.0002. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY83 C R R = X / tan(m pi/2) . C = k´ / (2 pi f) = m pi/2 Sample Fig.2.4EquivalentcircuitmodelsfortheEPimpedance: Frickemodel(1932). 10 Hz 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 1 uF 100 uF 10 mF 1 ohm 100 ohms Electrode polarization effects: Fricke impedance model, k=0.0002 * sqr(2*pi) Frequency Fricke resistance R / capacitance C Fricke capacitance m=0.5 (= Warburg cap.) Fricke resistance m=0.5 (= Warburg res.) Fricke capacitance m=0.3 Fricke resistance m=0.3 Fricke capacitance m=0.7 Fricke resistance m=0.7 Xc = 1 / k(2 pi f) 1m C R Fig.2.5SimulatedresultsfortheFrickeEPmodelandk=0.0002* √ 2π. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY84 Table2.2ExampleoftheimpactoftheEPimpedanceondielectricmeasurementsatlowfrequencies. Freq k ′ σ ef f K R series EP C series EP R parallel sample C parallel sample R series sample C series sample R series appar. C series appar. R parallel appar. C parallel appar. k ′ (appar.) σ (appar) ef f k ′ err σ ef f err 1kHz 500 0.01 30 25.2Ω 6.3μF 3000Ω 147.6pF 3000Ω 19.1μF 3025.1Ω 4.8μF 3025.5Ω 582.7pF 2001.4 9.916e-3 300.3% 0.84% " " " 60 " " 6000Ω 73.78pF 6000Ω 9.5μF 6025.1Ω 3.8μF 6025.4Ω 183.5pF 1243.6 9.958e-3 148.7% 0.42% 100kHz 50 0.007 30 2.52Ω 632nF 1500Ω 14.8pF 1499.7Ω 76.29nF 1502.2Ω 68.1nF 1502.6Ω 16.5pF 55.8 6.988e-3 11.6% 0.17% " " " 60 " " 3000Ω 7.38pF 2999.4Ω 38.1nF 3001.9Ω 35.97nF 3002.6Ω 7.8pF 52.9 6.999e-3 5.8% 0.09% CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY85 2.3 MethodstoMitigatetheEPEffects SincethebeginningoftheEPstudies,manydifferentmethodstoremovetheEPimpedance from the dielectric measurements have been proposed. A recent survey on these methods is provided in [92]. Before analyzing the most promising EP-reduction methods, it is im- portant to return to the analysis of the Table 2.2 in order to capture what are potential ways to reduce the EP effects. Accordingly, it is apparent that a potential way to reduce the EP effectsisbysimplyincreasingthephysicaldimensionsofthedielectriccell. Inapreliminaryanalysis,itisimportanttorememberthatthismethoddoesnotactually changetheEPimpedanceateachoftheelectrodes,onlytheimpactofthisimpedanceonthe apparent impedance Z ′ sample which is measured as Z EP +Z sample . For the 1 kHz example in Table 2.2, by doubling the distance between the electrodes, the k ′ error is reduced from 300.3% to 148.7%. Therefore, one can consider a case where such distance between the electrodesisincreased10foldsinordertoachieveevenbetterresults. Inthiscase,theerror wouldindeeddropto29.9%andforacertainhigherdistancebetweentheelectrodes,theEP effectswouldbeeventuallynegligible. Althoughthissolutionseemstobereasonable,there are at least 3 critical aspects to be considered: a) a bigger dielectric cell would also require abiggersamplevolumeandinsomecasesthisisnotpossible,b)withanincreasingsample volume, the total bulk resistance will increase to a point that the current passing through the system may not be accurately detected by the instrumentation, and c) larger physical dimensions of the dielectric cell may strongly increase the parasitic capacitance of the cell which is called stray capacitance and, at the same time, the sample capacitance will also decreaseduetothebiggercellconstantK leadingtoascenariothattheparasiticcapacitance mayhavethesameorderofmagnitudeofthesampleinsidethedielectriccell. The latter aspect is usually the most critical one and the first one to actually constrain the upper size limit of the cell. For instance, at the previous example involving a 10-fold distance increase, the sample capacitance drops from 147.57 to 14.757pF. Unfortunately, values around few pF are reported for parasitic capacitance in dielectric cells [60]. A typ- ical procedure in such cases is to adopt a calibration procedure using a standard solution, suchasKCl. However,itisverycriticaltoemploycalibrationprocedureswhenthequantity tobesubtracted/removedisverysimilar(intermsoforderofmagnitude)totheactualquan- tity which is being measured. Therefore, it is not a surprise that this EP-reduction method isnotusuallymentionedattheliterature. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY86 WenowreturntothefirstEP-reductionmethodwhichhasbeenusedformorethan120 years: the adoption of platinum electrodes. A rich source of information regarding the na- ture and effectiveness of this method is provided in [86]. In particular, starting at pp.32, a detailedevaluationoftheplatinizedplatinum(Pt-Black)electrodeispresented. Pt-Blackis one of the most common options for electrodes in sub-MHz dielectric spectroscopy. The effectivenessofPt-Blackisusuallyassociatedwiththefactthatthesurfaceareaoftheelec- trode increases by more than 2 orders of magnitude when the platinization process occurs. Note that the smooth surface of a stainless steel electrode has the contrary effect, that is, it has the smallest surface area possible for that electrode and consequently it can potentially favor the impact of the EP effects. On the other hand, if the EP data, besides the dielectric dataofthebulksample,isalsoimportant,astainlesssteelispotentiallyagoodchoice. Such kindsofelectrodesareusuallycalledpolarizableincontrastwiththenon-polarizableones, such as Pt-Black. In our proposed EP-identification method, both kinds of electrodes are supported. To understand how a higher surface area contribute to reduce the EP effects, the first step is to remember that seriesC EP will increase its value with the surface area of the elec- trode becauseC is a linear function of the surface area as indicated in (1.53). The next step is to demonstrate that when the value of the series C EP increases, C parallel appar. decreases and approaches the valueC parallel sample . That is, the k ′ error decreases until the point it is negligible. This interesting analysis, which is usually omitted at the literature, is usually difficult to be capturedbecauseitinvolvestheseries-parallelconversions. Anumericalexamplemayhelp to illustrate the point. To this end, we return to the worst scenario in Table 2.2: 1 kHz, cell constant K1=30. We can realistically assume that the PT-Black platinization process can achieve different increases of the electrode surface area: 1, 2, and 3 orders of magni- tude. Disregarding the impact at R EP (this one will actually be reduced, also contributing to a smaller EP impedance), we want to analyze specifically the measurement error on the dielectric constant k ′ of the sample. The results are tabulated in Table 2.3, where the first lineistheoriginalcasewithoutPt-Black. Based on the analysis of the results on Table 2.3, we can conclude that by increasing the surface area of the electrode, the EP effects can be virtually eliminated. Although the justification for this statement has been provided by a numerical instance, an analytical explanation is also simple to be achieved. Assume thatC series sample ≪ series-C EP , which is val- CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY87 Table2.3ExampleoftheeffectivenessofthePt-BlackelectrodeinreducingtheEPeffects. Surfaceareaincreasefactor C series EP C parallel sample C parallel appar. k ′ err 1e0(noplatinization) 6.32μF 147.6pF 582.7pF 300.3% 1e1 63.2μF " 188.9pF 28.0% 1e2 632μF " 149.5pF 1.3% 1e2.25 6320μF " 147.6pF 0% idated by the fact that we have found a way to strongly increase the value of series-C EP . Therefore,theseriesassociationbetweenC series sample andseries-C EP isapproximatelythevalue ofthesmallercapacitance,series-C EP . Asaresult,theapparentseriescapacitanceistheac- tual capacitance of the dielectric cell and it is then proved that, for this case, the EP effects donotimpactthedielectricconstantmeasurements. The Pt-Black option is an excellent one for EP-mitigation when fluids are involved, which is the typical scenario in many biological and biomedical studies. Unfortunately, this method presents many issues when applied to solids, particularly with soil samples with different levels of soil moisture. Although our first results involving soil and platinum electrodes(notPt-Black)wereverysuperiorregardingthemitigationoftheEPeffectscom- pared to the results of stainless steel electrodes, it was not possible for us to satisfactorily reproduce the experiments. In particular, tiny clay particles penetrate the rough surface of the electrode resulting in a different surface area. Unfortunately, the process of cleaning the electrodes is not sufficient. In short, based on our experience with multiple electrodes, once the initial experiments with soils are performed with brand new platinum electrodes, the same results may never be repeated again with the same pair of electrodes. Therefore, weconcludedthatanysolutionthatemployselectrodeswithroughsurfaceispotentiallynot compatiblewithaccuratemeasurementsofsoilsamples. AnotherEP-reductionmethodwhichisverypopularindielectricspectroscopyistheuse of a 4-electrode (4E)instrumentation, as discussed at the previous chapter. We also evalu- ated many options regarding this technique that involved the design and implementation of differentfront-enddevicestobeusedinconjunctionwithcommercialimpedanceanalyzers and lock-in amplifiers. Although the preliminary tests involving calibration samples, such asDI-water,NaCl/KClsolutions,wereindeedsuccessful,problemsarisewithsoilsamples. Specifically, similar same problems in reproducing soil experiments that we faced with the platinum electrodes occurred again with 4E instrumentation. For instance, one of our de- CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY88 signs involved needle electrodes that were employed as potential probes of the 4E system. Inthis case, a smallmovementofanyof theneedles wasenough toresult inerrors ofmore than 30% on the dielectric constant measurements, but not on the conductivity ones which wereverystable. After a deeper investigation regarding the 4E method [60, 93], it became apparent that the potential probes cannot be placed close to the virtual boundary lines of the dielectric cellcapacitor. Infact,in[60],H.Schwan(1963)providedaverysuccessfulsolutionforthe problembyproposingawaytoplacethepotentialprobesrelativelydistantfromthedielec- tric cell. Unfortunately, for dielectric soil measurements, the proposed solution cannot be applied since it involves a coupling fluid and the sample is also expected to be a fluid. We continuedtheinvestigationregardingthecombinationof4Eandsoilmeasurementsandwe concluded that the existing few related work do not provide enough information to allow the reproduction of the experiments. Moreover, to the best of our knowledge, a particular solutionforthementionedproblemhasnotbeenreportedandweeventuallyconcludedthat suchchallengeisstillanopenresearchproblem. ThelastmethodrelatedtotheEP-reductionwhichisdiscussedinthissectionistheelec- trode distance variation. This method was proposed by H. Fricke and H. Curtis in 1937, which is indeed a mathematically accurate solution for 2E instrumentation. The electrodes canbemadeofstainlesssteelandaregularimpedanceanalyzer(orsimilarinstrument)can be employed. Two different dielectric measurements of the same specimen are performed bymeansofdifferentcellconstantvalues(forinstance,differentdistancesbetweentheelec- trodes). Themathematicalfoundationforthismethodisprovidedatthenextsectionsinceit isalsoemployedforourproposed technique. Nonetheless, thismethodis rarelymentioned attheliteratureandwebelievethattheexplanationisthefollowing: thetechniqueassumes that both measurements are taken in scenarios where the EP impedance is the same while, in practice, this condition is very hard to be satisfied. This is particularly a big challenge becausewhenoneemploysdifferentcellconstants,itisexpectedtohavecellswithdifferent resistancesandthisfactimpliesthatthecurrentdensitymaypotentiallychange. For instance, when one increases the distance between the electrodes of a parallel-plate dielectric cell, the resistance of the sample increases. Because the Fricke’s method do not specify any other form of process control, we can assume that all test parameters are the same including the excitation voltage or current levels. Therefore, a change of the sam- CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY89 ple resistance can potentially vary the current density through the electrode. According to [60, 61, 87], a change of the current density is potentially associated to different EP impedances. More specifically, at these studies, curves EP-impedance vs. current density aregivenforfixedvaluesoffrequency. Basedonthesestudies,weconcludedthatmeasure- ments at different distances with a fixed set of other parameters are potentially associated to different EP impedances and, if this is indeed the case, the Fricke’s method cannot be employed. To the best of our knowledge, this is the first time that such conclusion is being reportedregardingthiscriticallimitationofthemethodofthedistancevariation. To prove the previous statement, we performed a series of experiments involving both parallel-plate and coaxial capacitor dielectric cells, two commercial impedance analyzers (Quadtech 1910 and HamegHM81180), and a customized impedance measurement instru- mentation describe in Appendix A. We also employed different voltage source levels and frequencyrangebetween100Hzand200kHz. OurgoalwastoemploytheFricke’smethod foravarietyofwell-controlledscenariosandcomparetheresults. Theconclusionofthisset ofexperimentsisthefollowing: basedonthepreliminarytests,thereisnowaytoconclude if the proposed method works only based on the information of the original paper. More specifically, in some cases, the application of the method resulted in negative values for bothC corrected sample and R corrected sample . For some other cases, reasonable smaller corrected values for RandC arefound,butwithoutanyindicationofaconvergenceatthecorrectvalues. These resultspartiallyexplainthelackofempiricalresultsregardingtheFricke’smethod. One rare exception is the work performed by P. Hoekstra and H. O’Brien (1969) [94] involving clay suspensions and the distance variation method. It is interesting to mention that this work employed platinum black-coated electrodes in association with the distance variation method. Therefore, there is a high probability that the EP effects were strongly reducedbeforetheapplicationofthevariationmethodresultinginexperimentswithsimilar EP impedances, a proper scenario for the application of the Fricke’s method. This obser- vation was the starting point for our work toward a systematic methodology that could be appliedattheexperimentsinvolvingtwodistinctdielectricmeasurements(i.e.,bymeansof different cell constants) in order to achieve the same EP impedance for both experiments. Oncesuchconditionisachieved,theFricke’stechniquecanbeaccuratelyemployed. Atthe nextsection,ournovelmethodologybasedontheFricke’smethodofdistanceisdeveloped. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY90 2.4 NovelEP-IdentificationTechnique: LinearEP-Match A novel approach to identify, rather than to mitigate/eliminate, the EP impedance from the impedance measurements is presented. It is called Linear EP-Match methodology for EP- identification in 2-electrode systems. As previously mentioned, we hypothesize that once the same EP impedance constraint of the Fricke’s technique is achieved, that distance vari- ation method can be applied with accurate results, that is, with the separation of the EP impedance from the apparent measured impedance, resulting on corrected values for the dielectricconstantandeffectiveconductivityofthematerial. The first step toward the development of the proposed method is to present the original work of Fricke [59] which is also discussed in [60]. Nonetheless, at the following devel- opment no simplification of the expressions is performed. It is worth to highlight that we observed errors larger than 1% at the evaluation of certain scenarios when the approxima- tions proposed in [60] are adopted. Moreover, final expressions forε ′ r andσ ef f are derived fromthedistancemethodandprovidedforthefirsttime. Assumethattwodielectricmeasurementsina2-electrodesystemareperformedforthe samematerialandfrequency. Thedifferenceismainlytheuseoftwodistinctcellconstants. Foraparallel-platedielectriccell,itmeansdifferentdistancesbetweentheelectrodes. How- ever, for a coaxial-capacitor dielectric cell, the different cell constants usually refer to two different lengths or heights (as opposed to diameters) of the cells. Assume that Z 1 (ob) and Z 2 (ob) are the observable or apparent measurements provided by the instrumentation for the different cell constants. Also, Z 1 (sa) and Z 2 (sa) are the actual impedances of the sam- ple material and these are the unknown values that we are interested to determine. Finally, Z 1 (ep)=Z 2 (ep)=Z(ep) is the EP impedance and, as clearly indicated, it is assumed that the same EP impedance occurs in both measurement scenarios. Therefore, we can conclude that: Z 1 (ob)=Z 1 (sa)+Z(ep) (2.1) Z 2 (ob)=Z 2 (sa)+Z(ep) (2.2) Subtracting (2.1) from (2.2), it is possible to remove the EP impedance effect. The accuracy of this step is only constrained by: a) the measurement errors regarding Z1(ob) andZ2(ob)andb)theerrorontheassumptionZ 1 (ep)-Z 2 (ep)=0: CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY91 Z 1 (ob)−Z 2 (ob)=Z 1 (sa)−Z 2 (sa) (2.3) Next,wedeveloptheleftsideof(2.3)basedonmeasurementsofaseriesRC configura- tion: Z 1 (ob)=R 1 (ob)− jX 1 (ob) (2.4) Z 2 (ob)=R 2 (ob)− jX 2 (ob) (2.5) Z 1 (ob)−Z 2 (ob)=[R 1 (ob)−R 2 (ob)]− j·[X 1 (ob)−X 2 (ob)] (2.6) The next step is to develop the right side of (2.3) considering that the dielectric cell containing the sample is actually modeled as a parallel RC (lossy capacitor). That is, the resistance R sa is in parallel with the capacitanceC sa and the its corresponding reactance is Xc sa . Rather than repeating the same development for Z 1 (sa) and Z 2 (sa), we develop an expressionforZ(sa): Z(sa)= R sa ·Xc 2 sa R 2 sa +Xc 2 sa − j R 2 sa ·Xc sa R 2 sa +Xc 2 sa (2.7) Thefollowingexpressionsareassociatedtothephysicaldimensionsofthedielectriccell and how they are related to the actual values ofε ′ r andσ ef f of the material. Such relations are accurate provided that the electrical field lines are homogeneous inside the dielectric cell and stray capacitance is negligible. The term K refers to the cell constant and its value depends if the dielectric cell is a parallel-plate or coaxial-capacitor configuration. For the former case, d is the distance between the plates and a is the surface area of the plate. For thelattercase,aistheradiusoftheinternalelectrode,bistheradiusoftheouterelectrode, andhisthelengthorheightofthecoaxial-capacitor: K(parallel−plate)= d a (2.8) K(coaxial−capacitor)= lnb/a 2πh (2.9) R(sa)= K σ ef f (2.10) C(sa)= ε ′ r ·ε o K (2.11) Theaboveexpressionscanbeappliedin(2.7),resultingin: CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY92 Z(sa)= K σ ef f · 1 ω 2 C(sa) 2 K 2 σ 2 ef f + 1 ω 2 C(sa) 2 − j K 2 σ 2 ef f · 1 ωC(sa) K 2 σ 2 ef f + 1 ω 2 C(sa) 2 (2.12) Z(sa)= K σ ef f · 1 ω 2 ε ′2 r ·ε 2 o K 2 K 2 σ 2 ef f + 1 ω 2 ε ′2 r ·ε 2 o K 2 − j K 2 σ 2 ef f · 1 ω ε ′ r ·εo K K 2 σ 2 ef f + 1 ω 2 ε ′2 r ·ε 2 o K 2 (2.13) Z(sa)= K σ ef f · 1 ω 2 ε ′2 r ·ε 2 o 1 σ 2 ef f + 1 ω 2 ε ′2 r ·ε 2 o − j 1 σ 2 ef f · 1 ω ε ′ r ·εo K 1 σ 2 ef f + 1 ω 2 ε ′2 r ·ε 2 o (2.14) Z(sa)= K·σ ef f σ 2 ef f +(ωε ′ r ε o ) 2 − j K·(ωε ′ r ε o ) σ 2 ef f +(ωε ′ r ε o ) 2 (2.15) Z(sa)=K[A− jB] (2.16) Z 1 (sa)−Z 2 (sa)=K 1 ·[A− jB]−K 2 ·[A− jB] (2.17) Next,wewanttoequaterightandleftsideof(2.3)using(2.6)and(2.17): [R 1 (ob)−R 2 (ob)]− j·[X 1 (ob)−X 2 (ob)]=K 1 ·[A− jB]−K 2 ·[A− jB] (2.18) [R 1 (ob)−R 2 (ob)]− j·[X 1 (ob)−X 2 (ob)]=A(K 1 −K 2 )− j·B[K 1 −K 2 ] (2.19) [R 1 (ob)−R 2 (ob)]− j·[X 1 (ob)−X 2 (ob)]=A(K 1 −K 2 )− j·B[K 1 −K 2 ] (2.20) [R 1 (ob)−R 2 (ob)]=A(K 1 −K 2 ) (2.21) A= σ ef f σ 2 ef f +(ωε ′ r ε o ) 2 (2.22) [X 1 (ob)−X 2 (ob)]=B[K 1 −K 2 ] (2.23) B= ωε ′ r ε o σ 2 ef f +(ωε ′ r ε o ) 2 (2.24) CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY93 By plugging (2.22) into (2.21), it is possible to achieve an expression for the corrected value of ε ′ r , the dielectric constant of the material, potentially without the influence of the EPimpedance: [R 1 (ob)−R 2 (ob)]= σ ef f σ 2 ef f +(ωε ′ r ε o ) 2 ·(K 1 −K 2 ) (2.25) E = [R 1 (ob)−R 2 (ob)] (K 1 −K 2 ) (2.26) E = σ ef f σ 2 ef f +(ωε ′ r ε o ) 2 (2.27) σ ef f =[E·σ 2 ef f ]+[E·(ωε ′ r ε o ) 2 ] (2.28) ε ′2 r = σ ef f −E·σ 2 ef f E·ω 2 ·ε 2 o (2.29) ε ′ r =sqrt σ ef f E −σ 2 ef f ω 2 ε 2 o (2.30) ε ′ r =sqrt σ ef f R 1 (ob)−R 2 (ob) ·(K 1 −K 2 )−σ 2 ef f ω 2 ε 2 o (2.31) ε ′ r = q σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f ωε o (2.32) Similarlytotheε ′ r case,byplugging(2.24)into(2.23),itispossibletoachieveanexpres- sion for the corrected value of σ ef f , the effective conductivity of the material, potentially withouttheinfluenceoftheEPimpedance: [X 1 (ob)−X 2 (ob)]= ωε ′ r ε o σ 2 ef f +(ωε ′ r ε o ) 2 ·[K 1 −K 2 ] (2.33) F = [X 1 (ob)−X 2 (ob)] (K 1 −K 2 ) (2.34) F = ωε ′ r ε o σ 2 ef f +(ωε ′ r ε o ) 2 (2.35) ωε ′ r ε o =F·σ 2 ef f +F·(ωε ′ r ε o ) 2 (2.36) Next,weplug(2.32)into(2.36)inordertoreachanexpressionforσ ef f : CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY94 ω· q σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f ωε o ·ε o =F·σ 2 ef f +Fω 2 ε 2 o · σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f ω 2 ε 2 o (2.37) s σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f =F·σ 2 ef f +F· σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f (2.38) s σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f =F·[σ 2 ef f + σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f ] (2.39) F = q σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) (2.40) Byequating(2.40)and(2.34),wehave: q σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) = [X 1 (ob)−X 2 (ob)] (K 1 −K 2 ) (2.41) q σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f σ ef f R 1 (ob)−R 2 (ob) =[X 1 (ob)−X 2 (ob)] (2.42) [X 1 (ob)−X 2 (ob)] 2 = σ ef f (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ 2 ef f σ 2 ef f (R 1 (ob)−R 2 (ob)) 2 (2.43) [X 1 (ob)−X 2 (ob)] 2 = (K 1 −K 2 ) R 1 (ob)−R 2 (ob) −σ ef f σ ef f (R 1 (ob)−R 2 (ob)) 2 (2.44) [X 1 (ob)−X 2 (ob)] 2 = (K 1 −K 2 )(R 1 (ob)−R 2 (ob)) 2 σ ef f (R 1 (ob)−R 2 (ob)) −(R 1 (ob)−R 2 (ob)) 2 (2.45) [X 1 (ob)−X 2 (ob)] 2 = (K 1 −K 2 )(R 1 (ob)−R 2 (ob)) σ ef f −(R 1 (ob)−R 2 (ob)) 2 (2.46) (2.47) Continuingwiththedevelopmentoftheexpression(2.46): CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY95 σef f[X 1 (ob)−X 2 (ob)] 2 =(K 1 −K 2 )(R 1 (ob)−R 2 (ob))−σ ef f (R 1 (ob)−R 2 (ob)) 2 (2.48) σef f = (K 1 −K 2 )(R 1 (ob)−R 2 (ob)) [X 1 (ob)−X 2 (ob)] 2 +[R 1 (ob)−R 2 (ob)] 2 (2.49) Therefore,ifweassumethatthetwomeasurementsarerelatedtothesameEPimpedance, theoverallprocessforcalculatingthecorrectedvaluesofε ′ r andσ ef f bymeansoftheLinear EP-matchmethodcanbesummarizedasfollows: • Measure sample #1 with a dielectric cell with cell constant K 1 in order to get the expressionZ 1 (ob)=R 1 (ob)-jX 1 (obj) • Measure sample #2 with a dielectric cell with cell constant K 2 in order to get the expressionZ 2 (ob)=R 2 (ob)-jX 2 (obj) • CalculateK 1 -K 2 • Calculate [R 1 (ob)−R 2 (ob)] • Calculate [X 1 (ob)−X 2 (ob)] 2 • Calculateσ ef f using(2.32)andthepreviouscalculationsresults • Calculateε ′ r usingσ ef f andthepreviouscalculationsresultsin(2.49) So far, with the proposed method it has been possible to obtain the corrected values of the electrical properties of the material under investigation. However, the actual value of the EP impedance was not actually determined: it was eliminated while subtracting Z 2 (ob) from Z 1 (ob). The following novel algorithm is been proposed as a way to determine Z(ep) bymeansofaninverseproblemapproach: The next goal behind the development of the EP-Match method is to find the necessary conditions for two experiments to have the same EP impedance. The first step toward the direction of the proposed solution was a very short but remarkable work of H. Schwan in 1965. As already mentioned, H. Schwan and J. Maczuk published an novel investigation CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY96 Algorithm 3 Determination of the EP impedance using the Linear EP-Match EP- identificationmethod Require: ω -angularfrequency Require: Z 1 (ob)-measuredimpedancewithdielectriccell#1 Require: Z 2 (ob)-measuredimpedancewithdielectriccell#2 Require: ε ′corrected r -EP-correcteddielectricconstantcalculatedusingequation(2.32) Require: σ corrected ef f -EP-correctedeffectiveconductivitycalculatedusingequation(2.49) Require: K 1 -cellconstantofthedielectriccell#1 Require: K 2 -cellconstantofthedielectriccell#2 Ensure: OutputpotentiallyistheactualEPimpedanceifR 1 (ep)=R 2 (ep)andC 1 (ep)=C 2 (ep) C 1 (sa)←ε ′corrected r ·ε o ·K 1 C 2 (sa)←ε ′corrected r ·ε o ·K 2 R 1 (sa)← K 1 σ corrected ef f R 2 (sa)← K 2 σ corrected ef f C 1 (sa) series ←C 1 (sa)+ 1 ω 2 R 2 1 (sa)C 1 (sa) C 2 (sa) series ←C 2 (sa)+ 1 ω 2 R 2 2 (sa)C 2 (sa) R 1 (sa) series ← R 1 (sa) 1+[ωR 1 (sa)C 1 (sa)] 2 R 2 (sa) series ← R 2 (sa) 1+[ωR 2 (sa)C 2 (sa)] 2 Xc 1 (sa) series ← 1 ωC 1 (sa) series Xc 2 (sa) series ← 1 ωC 2 (sa) series Z 1 (sa)←R 1 (sa) series − j·Xc 1 (sa) series Z 2 (sa)←R 2 (sa) series − j·Xc 2 (sa) series Z 1 (ep)←Z 1 (ob)−Z 1 (sa) Z 2 (ep)←Z 2 (ob)−Z 2 (sa) R 1 (ep)←ℜ[Z 1 (ep)] R 2 (ep)←ℜ[Z 2 (ep)] C 1 (ep)← −1 ω·ℑ[Z 1 (ep)] C 2 (ep)← −1 ω·ℑ[Z 2 (ep)] CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY97 about the linearity limits of the EP impedance in [61]. It is reported that the current den- sity plays a very important role regarding the EP effects. Based on that work, we perform tests where the current density is the same for two different cell constants. As expected, to achieve this solution is necessary to control the excitation voltage or current. For our instrumentation,thecontrolwasonthevoltagesource. Our initial expectation was that if two experiments employ the same sample and also the same current density (although involving distinct dielectric cells), then they potentially also have the same EP impedance. Unfortunately, the empirical results did not converge: forsimilaranddifferentcurrentdensities,thesuccessiveapplicationofthedistancemethod was not successful in finding a convergent EP impedance. We also observed significant variationoftheresults: sometimesthebetterresultsoccurredwitharelativelyhighersource voltage level, sometimes with a smaller voltage level. The investigation proceeded in order tofindthereasonsforthementionedresults. Inshort,weeventuallyconcludedthatdifferent aspectswereimpactingtheexperiments: thementioned sameEPimpedanceconstraintand alsoaccuracyissuesregardingthemeasurements. In 1974, C. Gabrielli and M. Keddam investigated the generation of harmonics when a highersinusoidalvoltagelevelisemployedinasystemwhereEPispresent[62]. Thiswork wasveryimportantforusbecausea)itshowedthatthehysteresiseffectsareexpectedonce the system enters at the non-linear region of the EP effects and b) it showed that the reduc- tion of the EP impedance with an increasing current density actually was not monotonic as initially suggested by Schwans’ studies: if one continues increasing the current density, he will observe oscillations regarding the EP impedance. The hysteresis effects were actually briefly mentioned by Schwan in 1963 [60], but details were not provided. Nonetheless, in 1984, B. Onaral, H. Sun, and H. Schwan published a detailed study about the linear and non-linearelectricalpropertiesofbioelectrodes[63]. The relation of such studies with the topic of this chapter is the following: considering that the EP impedance is a non-linear function of the current density and it is also possible that the measurement system presents some sort of hysteresis, the effort to find measure- ments conditions under the same EP impedance - what we call hereafter EP-match - must be guided by the mentioned aspects. We formalize our conclusions regarding this topic as follows: Assume that two impedance measurement systems A and B are used to take measure- CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY98 ments of the same sample and the basic parameters of these systems are also the same. The main difference between A and B is the dielectric cell: each system has a distinct cell constant. In order to apply the method of impedance subtraction, as proposed by Fricke in 1937, the systems A and B must operate under the same EP impedance, that is, under EP-match. Moreover,thesystemsmusthaveenoughaccuracyforthemeasurements. As exposed, two conditions are paramount for the application of the proposed EP- identificationmethod: EP-matchandenoughaccuracyforbothsystems Aand B. Infact,in some cases, the EP-match condition is achieved but the measurement accuracy of A, B or bothisnotsufficientfortheapplicationoftheimpedancesubtractionmethod. Theproblem is particularly critical when the EP impedance has a value closer or much higher than the actual sample impedance. As already discussed, the use of any EP-reduction method can mitigate this issue. However, considering that we are investigating an EP-identification so- lution particularly targeting soils, the adoption of bare stainless steel electrodes apparently is the best approach although potentially favoring the increase of the EP impedance. In or- dertodealwiththesechallenges,wedevelopedguidelinestoachieveEP-matchandalsoto identifymeasurementaccuracyissuesandotherdistortions: • VerifythatthesystemsAorBhavenotenteredatthenon-linearregionwithpotential hystereticbehavior. • Toverifyifthehystereticbehaviorispresent,checktherepeatabilityofthemeasured impedance while increasing and decreasing the current density. If a impedance is measured with a smaller value than expected and this one is persistent even with a decreasing current density, the system potentially is facing hysteresis and measure- mentscannotbetaken. Insuchscenario,removetheinstrumentationcablesandshort circuit the electrodes for a certain period of time (lower frequencies require higher timeintervalsforthisprocedure). • Ideally, the measurements of the system A and B are performed under small current density. Inthiscase,smallvariationsofthecurrentdensitydonotstronglyimpactthe EP-matchcondition. However,asmallcurrentdensitymayalsoresultinbadmeasure- ments accuracy. Remember that the proposed method of subtraction of impedances will fail for any of these two reasons: the EP-match condition is not achieved or the accuracy level while measuring A is very different from measuring B or accuracy errorsimpactforsystemsAandB. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY99 • Ifsmallcurrentdensityisnotpossibleduetothelimitationsoftheinstruments,search foraregioncloseto(butoutof)thenon-linearity. Insuchregions,theEPimpedance issignificantlyreducedandEP-matchispotentiallypossibletobefound. • One technique to achieve EP-match is to use the same current density of system A at system B and vary the voltage source in different closer points, above and below, to thatreferencecurrentdensity. Thissweepoperationisimportantbecauseitmaycover small differences between system A and B that can also impact the EP impedance besidesthecurrentdensity. Besides the above techniques, we also developed a methodology involving the mea- surement of harmonics while any dielectric measurement is being performed. By increas- ing/decreasingtheexcitationsignallevelandinvestigatingwhenthesystementersorleaves its non-linear region due to the EP effects, it is possible to identify a pattern of the total harmonic distortion (THD) level involving hysteresis. As already stated, it is essential to take measurements at the linear region of the EP phenomenon. Unfortunately, the main challenge is that such procedure requires a kind of instrumentation which is not available commercially as a single solution or product. Therefore, we designed a preliminary in- strumentation configuration comprised of a sinusoidal signal generator, a 70V-line power amplifier,signalgalvanicisolators(100kHztransformers),a0.01-degreeresolutionlock-in amplifier, a precise AC digital multimeter (300kHz) , and a FFT Spectrum Analyzer (100 kHz). With the above instrumentation setup, the mentioned procedure finds the exact excita- tionvoltage-levelsnecessarytoachievethesameEPimpedancewhendistinctdielectriccell geometries are employed, that is, the EP-match. This methodology extends the Fricke and Schwan’sworksinthreeways. First,itisnotlimitedtoparallel-platedielectriccells. Infact, thebestresultsareactuallyachievedwithcoaxial-capacitordielectriccellsduetothesmaller parasitic capacitance effects. Second, as already commented, both Fricke and Schwan pre- sented a simplified mathematical evaluation of EP for the distance variation technique and these simplifications are justified by the characteristics of the samples they investigated. In thisdissertationwork,anexactevaluationisperformed,asalreadydiscussed,duetothevery wide dynamic range of dielectric properties of soils under dry/wet conditions. Third, both FrickeandSchwanreportedempiricalevaluationsforthedistancetechniqueinvolvingnon- polarizableelectrodes(i.e.,Platinum-Black). Inthischapter,itwasalreadyreportedthatthe Frike’s distance technique typically fails when regular electrodes (i.e., stainless steel) are CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY100 employed. Non-polarizableelectrodestypicallyhaveirregularsurfacesandwhiletheyarea good option for the dielectric spectroscopy of fluids, these electrodes are a challenge when soil samples are evaluated. For the proposed EP-match method, polarizable electrodes can also be employed without restrictions (assuming only that they are relatively inert) which increasestherepeatabilityqualityofthestudiesinvolvingsoilsamples. 2.5 InstrumentationSetup,Experiments,andDiscussion In this section, the instrumentation setup used to evaluate the Linear EP-match method is presented. Preliminary experiments are performed and the related results are discussed. The instrumentation setup is shown in Fig. 2.6. Although similar to some reported dielec- tricmeasurementinstrumentationschemes,therearesomenovelaspectsinthissetup: • Special ground scheme: because some measurements involve signals smaller than 10μV,theproposedgroundschemeisessentialfortherealizationoftheexperiments. • 70V power amplifier: typically AC dielectric measurements involve very small sig- nals and the voltage levels rarely are higher than 1V. Nonetheless, in order to better investigatethenon-linearregionoftheEPphenomenon,suchequipmentisemployed with a special care to avoid damage at the remaining equipments due to the high- voltagelevel. • FFT analyzer: this is not a typical instrument for this scenario. In fact, to the best of our knowledge, this is the first work that report the use of such instrument associ- atedwithdielectricmeasurements. Thisinstrumentisusedtoevaluatethegeneration of harmonics whenever the system is entering in its non-linear region. Two total harmonic distortion (THD) measurements are always taken (each measurement set consists of the averaging of 500 distinct measurements): one with the actual sample and the other with a resistive load R load (check Fig. 2.6) temporarily substituting the dielectric cell. The value of R load corresponds to the magnitude of the sample impedance. The actual sample THD is the difference between these two THD mea- surements. In this way, the noise due non-EP related sources, such as the wiring, environment,andinstrumentationisremovedfromtheanalysis. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY101 Fig.2.6InitialinstrumentationsetupusedfortheevaluationoftheLinearEP-Matchmethod. • Phase drift avoidance: each impedance phase measurement is performed considering the difference between R load and sample, similar to the previous THD procedure. In this way, there is no need to perform calibration due to the cabling and other sources ofphasedrift. • R 1 selection: thesensitivityofthemeasurementsiscompletelycontrolledbytheuser, onceallnecessaryinputparametersfortheexperimentsareundercontroloftheuser. AtFig. 2.7screenshotsoftheFFTanalyzerareshown. Inthosemeasurements,theTHD levelofthesystemwitharesistiveloadinplaceofthedielectriccelliscomparedtotheTHD measurement with the dielectric cell. In this way, it is possible to determine the generation of harmonics at the sample. The advantage of selecting a measurement point close, but before,thenon-linearregion(whereTHDstartsincreasing)isthattheEP-impedanceisnat- urally reduced at this point due to the high current density. The trade-off of this option is thatthehighestvariationsofEPimpedancealsooccurinthisregion. InordertoachieveEP- match, usually it is necessary to take measurements with the system A and take note about its current density J A and THD level THD A . Next, take multiple measurements of system B with current density J B which is centered at J A but it is also varied. For each of these points, THD B levels are registered. The final step is to use a computer program to execute thecalculationspresentedinthischapterconsideringthemeasurementin Aandeachofthe mentionedmeasurementsfromBuntilthebestEP-matchpointisdetermined. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY102 Fig.2.7NoveluseofFFTAnalyzerforTotalHarmonicDistortion(THD)analysisindielec- tric measurements: the goal is to determine when the system enters in its non-linear region duetotheEPphenomenon. The Linear EP-Match method has this name because the points of measurements re- garding both systems A and B must occur in a linear region regarding the different current density levels that can be employed for both systems. The detection of such EP-linearity is very difficult to be achieved due to the very small magnitudes of the harmonics generated at the beginning of the non-linear region. The FFT analyzer we are employing has enough sensitivity for this task. As exemplified in Fig. 2.8, the evidences of hysteretic behavior of the system can be identified by two forms: a) a smaller measured impedance for a current densitypointwhereahigherimpedancewasalreadymeasuredandb)smallerTHDlevelfor apointwhereahigherlevelwasalreadymeasured. The proposed EP-identification method has been tested for two different kinds of soils undertwoextremescenarios: dryandwet(saturated)conditions. RegardingtheEPeffects, the most critical scenario is the one involving wet soils that also present high conductiv- ity. To this end, the NHS-SAT soil which was already discussed at the previous chapter CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY103 Fig.2.8NoveluseofFFTAnalyzerforTotalHarmonicDistortion(THD)analysisindielec- tric measurements: the goal is to determine when the system enters in its non-linear region duetotheEPphenomenon. is selected for the following demonstrations. Another aspect to be highlighted is that a coaxial-capacitor is used as dielectric cell for both systems A and B. Moreover, no special EP-reduction method is used, such as Pt-Black electrodes, besides the presented methodol- ogy. ThefirststepoftheEP-Matchmethodistotakemultiplemeasurementsusingbothsys- tems A and B where the voltage level is varied in order to achieve different current density levels for the experiments. This procedure must be repeated for each frequency of interest. Moreover, the procedure described here may be eventually encapsulated in a single system or instrument. As shown in Fig. 2.9, these set of measurements of A and B also have a related set of THD values. Whenever a THD value of the system A is very close to the one of B, a potential EP-Match point is identified. Such point may or may not be associated to the same current density or to the same source voltage level. The main goal is to select thepotentialEP-Matchpointsandapplythemethodoftheimpedancedifferenceinorderto CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY104 0 90 180 270 360 450 540 630 720 815 905 1,000 0 0.001 % 0.002 % 0.003 % 0.004 % 0.005 % 0.006 % 0.007 % 0.008 % EP Difference Method :: Total Harmonic Distortion Evaluation :: NHS SAT soil, VWC=40%, f=10 kHz Voltage source level for the Dielectric Cell (V src ) (mV) Total harmonic distortion (THD) (%) Dielectric Cell #1 (length 9.41mm) Dielectric Cell #2 (length 5.53mm) Vsrc1=266.48mV, Vsrc2=451.11mV Vsrc1=717.13mV, Vsrc2=816.75mV Vsrc1=266.48mV, Vsrc2=725.18mV Vsrc1=266.48mV, Vsrc2=269.56mV Vsrc1=266.48mV, Vsrc2=269.56mV D B C A E Fig.2.9LinearEP-MatchMethod: definingpotentialpairsofEP-matchpoints. analyzethebestEP-correction. BasedontheselectionofthebestEP-matchpairsofAandBsystems,thecorrectionfor thedielectricconstantandconductivityisperformed. Theresultsforthecaseunderanalysis areshowninFig. 2.10. Multiplepotentialsolutionsareobservedandsuchvariationgiveus aninitialerrorestimationofthemethod. Notallresultsinthisphaseoftheanalysisareactuallyvalidones. Onepotentialwayto determine the upper limit of the true (EP-corrected) dielectric constant is to verify the con- vergencewhenthesystemisdeeperinnon-linearmode. Insidethisregion,whichismarked by very high current densities, it is possible to verify min-max peaks for the dielectric con- stant, as shown in Fig. 2.11. The minimum of these values is a potential upper-boundary for the true dielectric constant because it potentially occurs when the EP effects are poten- tially eliminated or strongly reduced. Nonetheless, we cannot employ such values because the associated system was under non-linear behavior and it is not possible to determine the distortion/errorofthesevalues. ThemainchallengeofthecurrentEP-identificationmethodologyistoperformtinysteps regarding the valuation of the current density for the experiments. This is our current main challengingregardingtheevaluationoftheproposeEP-Matchmethod. Specifically,because the instrumentation is not fully automated, a single evaluation involving only 5 frequencies CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY105 0 1000 2000 3000 4000 5000 6000 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 X: 269.6 Y: 1142 Voltage source level for the DC#2 (Dielectric Cell 2) (V src2 ) (mV) Corrected Real Relative Permittivity ( ‘) EP Corrected Real Relative Permittivity NHS SAT soil, VWC=40%, f=10 kHz DC#1: Vsrc=717.1 mV(rms) DC#1: Vsrc=625.7 mV(rms) DC#1: Vsrc=266.5 mV(rms) B A C E D Fig. 2.10 Linear EP-Match Method: achieving the EP-corrected value for the dielectric constant. 0 1000 2000 3000 4000 5000 6000 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 X: 269.6 Y: 1142 Voltage source level for the DC#2 (Dielectric Cell 2) (V src2 ) (mV) Corrected Real Relative Permittivity ( ‘) EP Corrected Real Relative Permittivity NHS SAT soil, VWC=40%, f=10 kHz DC#1: Vsrc=717.1 mV(rms) DC#1: Vsrc=625.7 mV(rms) DC#1: Vsrc=266.5 mV(rms) B A C E D Fig. 2.11 Linear EP-Match Method: determining upper-boundary for the true value of the dielectricconstant. canlastmanyhours. Therefore,whileperformingtheexperiments,thesourcevoltagelevel is varied in steps of 100mV or higher. Based on the analysis shown in Fig. 2.12, such steps are actually too big causing the highlighted estimation errors regarding the dielectric constant, a fact that we only observed after the experiments. On the other hand, the errors regarding the effective conductivity are much smaller, as shown in Fig. 2.13. These results CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY106 0 1000 2000 3000 4000 5000 6000 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 Voltage source level for the DC#2 (Dielectric Cell 2) (V src2 ) (mV) Corrected Real Relative Permittivity ( ‘) EP Corrected Real Relative Permittivity NHS SAT soil, VWC=40%, f=10 kHz DC#1: Vsrc=717.1 mV(rms) DC#1: Vsrc=625.7 mV(rms) DC#1: Vsrc=266.5 mV(rms) B C A D E AVG: 3233 MIN: 1142, MAX: 6412, maxVAR: 98.33% Fig.2.12LinearEP-MatchMethod: determiningtheestimationerrorofthemethodregard- ingthedielectricconstant. areexpected,asinitiallydiscussedinthischapter. Before discussing the empirical work with the EP-Match method, it is important to ex- plainwhatpreliminarytestsarestillnecessarybeforeproceedingwiththetargetsamples: (a) Tests with sample-phantoms: in this context, the phantom sample refers to the RC circuitthatmimicstheexpectedelectricalbehaviorofthematerialunderanalysisfora certainfrequency. Formultiplefrequencies,asimpleRC circuitmaynotbesufficient and it is usual to employ multiple distinct RC circuits for such evaluation. The main goal of this test is to verify the accuracy of the overall instrumentation without the dielectriccell. Inmanyinstruments,thisstepisconsideredacalibrationoneandtypi- callysub-testscalledshort-open-loadareemployed. However,inourinstrumentation, thecalibrationisperformedonasetofphantomsthatarerepresentativeoftheextreme cases regarding our samples (e.g., soil types/conditions). We employed 14 soil phan- toms for diverse types of soils, water content levels, and frequencies. The values for thecomponentsoftheparallelRC(oracorrespondingseriesRC)circuitimplementing acertainphantomarecalculatedbasedonthedielectriccellsdimensions. Inourcase, because two distinct cells are used, a total of 28 circuits are implemented. Without involving the THD tests (the EP-effects are not present in this scenario), each phan- tom is tested with potential different values for the reference resistor R1. Therefore, CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY107 0 1000 2000 3000 4000 5000 6000 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Voltage source level for the DC#2 (Dielectric Cell 2) (V src2 ) (mV) Corrected Effective Conductivity ( eff ) EP Corrected Effective Conductivity NHS SAT soil, VWC=40%, f=10 kHz DC#1: Vsrc=717.1 mV(rms) DC#1: Vsrc=625.7 mV(rms) DC#1: Vsrc=266.5 mV(rms) B A C D E AVG: 0.4468, MIN:0.4202, MAX: 0.4641, maxVAR: 5.95% Fig.2.13LinearEP-MatchMethod: determiningtheestimationerrorofthemethodregard- ingtheeffectiveconductivity. calibration tables can be created for each test scenario. Alternatively, the maximum measurementerrorcanbeidentifiedandsimplyacceptediftheyarenottoolarge. (b) Co-emptycell-test: this test is used to identify the effective value of the parasitic capacitance (or inductance) of the set cabling/text-fixture/dielectric-cell. Once the dielectric cell is empty, the expected values for k ′ (i.e., 1) and σ ef f (i.e. 0) can be verified for each frequency of interest. A calibration table can be used with the goal to correct the data after the measurements. Alternatively, the maximum errors can be identifiedandaccepted,aspreviouslydiscussed. (c) Standard dielectric material: this test has the same purpose of the previous one. Deionized water and many single-phase substances with known dielectric constant andconductivitycanbeemployedforthiscalibration. Thesetestsareusuallydoneat higherfrequencies(e.g.,>100kHz)inordertospecifictheerrorduetothedielectric cell at these frequencies. At lower frequencies, the EP effects can impact the results. In our empirical work, we performed tests with deionized water involving frequen- cies such as 1, 10, and 100 kHz with error smaller then 1%. The best accuracy was obtainedwithglycol-ethylene. (d) Standard conductivity material: this test is oriented toward a preliminary verifica- tion of the EP-reduction or -identification methods. It is important to highlight that CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY108 such test is typically seen as a necessary condition for the validation of EP-related methods, but they are not sufficient, in particular for multi-phase materials, such as soil samples. We employed a standard KCl 0.1M solution (measured conductivity: 1.289 S/m) and the EP-Match method is verified. The EP-identification and correc- tion is found to be mainly successful at 1 and 10 kHz, with errors smaller than 2%. Therefore,forotherfrequencies,wecouldnotpreliminarilyverifytheeffectivenessof the EP-Match method at least for the 0.1M KCl solution. Higher estimation errors at other frequency points are due to the limitations of our instrument. The lack of accu- racyforfrequenciessmallerthan1kHzcanbeexplainedasfollows: asthefrequency decreases, the EP-linear region also decreases and for a relative small dielectric cell and a high-conductivity material the excitation voltage must be very small (e.g., tens of millivolts), causing strong accuracy errors on the instrumentation and these errors impacttheuseoftheEP-Matchmethod. Nonetheless,ifthematerialtobetesteddoes not have a very high conductivity, these mentioned errors due to our instrumentation arepotentiallysmallerforfrequenciessmallerthan1kHzandhigherthan10kHz. Next, we compare the preliminary results using this novel EP-Match method with the soil dielectric model proposed at the last chapter. In Fig. 2.14, the results of the MI-Soil dielectric model applied to the NHS-Soil are presented. This is our first version of a sub- MHz soil dielectric model which was proposed in Chapter 1. In Fig. 2.15, the critical scenario we want to highlight is shown: a saturated soil with very high conductivity. The EP effects are expected to be very strong for such scenario. To this end, we perform an empirical evaluation for this soil scenario using commercial impedance analyzers and two cell constants for the dielectric cell, as shown in Fig. 2.16. As expected, the EP effects are clearly present at the measurements. In fact, the use of a dielectric cells with two dis- tances/lengthsisasimplewaytoevaluatediftheEPeffectsarepresentatthemeasurements. Finally, the preliminary results of our new instrumentation scheme associated to the novelLinearEP-Matchmethodarepresented. AsshowninFig. 2.17,theproposedmethod hasvaluesofdielectricconstantsmallerthantheestimationsofourpreliminaryMI-Soildi- electricmodel. Nonetheless,thedifferencesarenottoostrong. Theexperimentsfor100Hz wereextremelydifficultduetoourcurrentinstrumentationlimitation: wecannotaccurately measure THD for signals smaller than 300μV. Unfortunately, as the frequency decreases, the limit of linearity also decreases and to achieve measurements at the linear region it is necessary to reduce the current density (i.e., the source voltage level must decrease to very low values in our specific setup). An interesting result is the fact that the preliminary CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY109 10 2 10 3 10 4 10 5 10 6 10 1 10 2 10 3 10 4 10 5 10 6 10 7 Frequency (Hz) EP corrected Dielectric Constant ( ) Actual Corrected NHS SAT soil, vwc: 5..40% vwc: 40% vwc: 5% Empirical (August 2014): original coaxial capacitor, diam. 25.4mm, length 9.41mm, 300mV EP corrected MI Soil v.6 model Fig.2.14MI-SoildielectricmodelproposedinChapter1andappliedtotheNHS-SATsoil. 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 10 2 10 3 10 4 10 5 10 6 10 7 Frequency Real Relative Permittivity ’ (or dielectric constant ) NHS SAT soil, vwc: 40% EP corrected MI Soil v.6 model Empirical (August 2014): original coaxial capacitor, diam. 25.4mm, length 9.41mm, 300mV Fig. 2.15 Critical scenario to be investigated in this context: saturated soil with very high conductivity. TheEPeffectsareexpectedtobeverystrongforsuchscenario. MI-Soil model has a fair agreement with the current empirical results involving the new instrumentationandthenewEP-identificationmethod. In order to compare the results with an EP-reduction solution, we employed mesh elec- trodes typically used in water filtering systems (i.e., electrochemical water disinfection). The tests are performed with virgin electrodes because, as discussed before, the repeata- bility of soil experiments with such electrodes are very difficult to be achieved due to the CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY110 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 10 2 10 3 10 4 10 5 10 6 10 7 Frequency Real Relative Permittivity ’ (or dielectric constant ) Empirical (May 2015): new dielectric cell (coaxial capacitor), diam. 25.4mm, length 9.41mm, 500mV Empirical (May 2015): new dielectric cell (coaxial capacitor), diam. 25.4mm, length 5.53mm, 500mV Empirical (August 2014): original coaxial capacitor, diam. 25.4mm, length 9.41mm, 300mV NHS SAT soil, vwc: 40% EP corrected MI Soil v.6 model Fig. 2.16 Recent empirical evaluation using commercial impedance analyzers and two cell constantsforthedielectriccell. TheEPeffectsareclearlypresentatthemeasurements. 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 10 2 10 3 10 4 10 5 10 6 10 7 Frequency Real Relative Permittivity ’ (or dielectric constant ) Empirical (May 2015): new dielectric cell (coaxial capacitor), diam. 25.4mm, length 9.41mm, 500mV Empirical (May 2015): new dielectric cell (coaxial capacitor), diam. 25.4mm, length 5.53mm, 500mV Empirical (August 2014): original coaxial capacitor, diam. 25.4mm, length 9.41mm, 300mV NHS SAT soil, vwc: 40% EP corrected MI Soil v.6 model Preliminary empirical results using the "equivalent EP points" difference mentod Fig. 2.17 Comparison between the preliminary results of the new instrumentation scheme associatedtotheLinearEP-MatchmethodandtheMI-Soildielectricmodelestimations. impregnation of tiny soil particles in rough surfaces. The results are also presented in Fig. 2.17. From the perspective of a MI-WUSN system, an important frequency point is 1 kHz. The analysis regarding this point shows that the result from the proposed Linear EP-Match methodispotentiallycorrect(orwitharelativesmallerror)sinceitliesclose,butbelowthe result from the platinum-based experiment. This result is expected since the latter method does not guarantee total elimination of the EP effects. However, for the frequency points CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY111 10 and 20 kHz, it is possible that the results from the platinum-based tests are more accu- rate than the ones with the EP-Match method. These errors are mainly due to our current instrumentation limits: while moving to higher frequencies, the number of harmonics to be analyzedatourTHDinstrumentdecreasesandtheEP-Matchmethodisalsoimpacted. ThediscussedresultsonlyprovideapreliminaryandpartialvalidationoftheLinearEP- Matchmethod. Nonetheless,theunderlyingtheorybehindthemethodisindeedsolid: ifthe same EP impedance is present in two measurements related to two distinct dielectric cells, than this EP impedance is eliminated when the observable (measured) impedances of these systems are subtracted. Also, such mathematical operation can only be performed at the linearregionofthesystems,whichisonetherequiredconditionsoftheproposedmethod 1 . Therefore, if the accuracy of these two measurements is the same and sufficient, the Linear EP-match method must work properly. Note that the previous sufficient (accuracy) term is our current challenge in applying the proposed method for very distinct materials: more accurateDHTanalyzersandLock-Inamplifierswouldberequired. Our final discussion for the proposed Linear EP-Match method is related to how im- portant is the use of a DHT analyzer. The answer lies on the region we are targeting to find EP-match pairs. At very small current densities (i.e., small source signal levels), many EP-match pairs can be found and usually the minimum dielectric constant can be found by applying the subtraction method among this large set of EP-match pairs. The trade-off of this option is the accuracy of the measurements before the subtraction task: a small error can imply a large error regarding the EP contribution. However, if the instrument uses a highly-stable and linear sinusoidal source in conjunction with AC microvoltmeters, than this solution may be enough for the Linear EP-match. For this case, note that it is assumed that the involved small current densities leave the system continuously in its linear region andfurthertestsandDHTmeasurementsarenotrequired. 1 During our evaluations of the THD pattern, we subtracted THD values even at the non-linear region. By performing such operation, an unknown error can be present at the results. Nonetheless, such results are not used at the proposed method, but they provide a simple way to identify a pattern with very large maximum and minimum peaks of THD values whenever the system is in its non-linear region which is the expected pattern reported by C. Gabrielli and M. Keddam [62]. More important, from the instrumentation perspective, the algorithm that can potentially identify linear and non-linear EP regions is based on the hysteresis effects on the measured EP impedance, not on the THD values at the non-linear region. Therefore, the mentioned operationisnotactuallyimpactingneitherthesoundness,northeaccuracyofthemeasurements. CHAPTER2. TOWARDANACCURATESUB-MHZSOILDIEL.SPECTROSCOPY112 On the other hand, if we move to higher current densities in order to increase the accu- racy of the measurements, it is possible to look for some EP-Match pairs close and before the non-linear EP region. In this scenario, the EP-Match condition is much more difficult to be achieved compared to lower current densities. The DHT measurements come to the scene mainly as a tool to verify if the system is actually in its linear region. Moreover, the DHTvaluesarealsousedtofindpotentialEP-Matchpairsinamuchmorecriticalscenario forthissearch. Inthissense,theDHTanalyzerindeedseemstobearequirement. Nonetheless, in our ongoing studies, we are trying to find a way to eventually elimi- natetheDHTanalyzerfromtheinstrumentationbasedontheLinearEP-Matchmethod. To this end, we are investigating the use of very small voltage level increments while taking the measurements. The ultimate goal is to apply the difference operation on hundred or thousand of potential EP-Match pairs. As expected, the drawback of this solution which doesnotrequireaDHTanalyzeristhepotentialincreaseoftimetotakethemeasurements. Therefore, we believe that future EP-aware impedance analyzers will come with a feature that allows the user to select the frequency points where an EP-identification method must beemployed(longertimes),whiletheremainingpointsfollowthetraditionalmeasurement procedure (shorter times). In this way, the equipment or the user can employ an extrap- olation technique to achieve EP-free dielectric measurements with excellent accuracy in relativesmallperiodsoftime. PartII CommunicationPhysical(PHY)Layer forMI-WUSNs 113 Chapter3 MISignalAttenuationModelfor WUSNs Theoperatingfrequencyofthemid-rangeMIsystemwhichisbeingpresentedinthiswork varies from 1 KHz to around 200 KHz. The main justification for the limitation to this LF band will be given later, when different real-world scenarios are evaluated. Aspects such asrobustnesstodifferentsoilconditions,energy-efficiency,size,costs,andnetworkperfor- manceofthesolutionwilleventuallydictatethepracticalconstraintsthatleadtothisLFop- eration. Nonetheless,itisimportanttohighlightthatdifferentMIsystemstargetingsmaller distances(e.g.,<15m)canstilloperateattheMHzbandalthoughsuchshort-distanceoper- ationisoutofthescopeofthiswork. In this chapter, our study starts with the Maxwell’s equations in order to determine the fields due to a single loop. The results can be approximated for a circular multilayer (air- cored) component which is our TX coil of the MI system. Next, the induced voltage at the another coil (RX coil), at a distance r from the TX coil, is analyzed. For practical reasons, the discussion is limited to the case where the same coil can be used for transmission and reception in a half-duplex link. After considering how the RX coil receives the signal, the importance of employing resonant LC tanks, rather than a simple coil with inductance L without using a capacitor, is highlighted. Different resonance circuits can be used depend- ing if the purpose is to maximize the energy transfer, such as in Wireless Power Transfer (WPT) solutions [32], or to maximize the received signal - more specifically, the signal-to- noiseratio(SNR)-attheRXside. Then,theessentialspartstoachieveabasicMIwireless communication solution are already in place and we use the MI-SOIL model from the pre- viouschapterandaddtoittheinvestigationsdiscussedinthischapter. Attheend,itwillbe 114 CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 115 Wave equation: z y x r E(r) _ Fig.3.1Coordinatesystemandwaveequationduetoatime-varyingsource. possible to answer basic application questions, such as, given the distance r, the frequency f, the soil class x, the water content level y, and essential circuitry parameters (such as TX current,numberofturns,wiringgaugeanddiameterofthecoil),whatwouldbetheexpected induced voltage level at the RX coil. This is a fundamental question because it allows one to understand the theoretical and practical boundaries of the system. We anticipate that the intuition for an upper-bound frequency of the MI system for a dry soil comes from such evaluations. Unfortunately, due to the non-linear relationships among many of the system input parameters, it is not possible to simply state that the smaller the frequency, the more efficientistheMIsystemforworstmediumconditions(e.g.,saturatedsoil). Thedetailsre- gardingthisdiscussion,manyofthemnon-intuitiveinaquickinvestigationoftheproblem, areprovidedatChapters4to6. Next, empirical experiments in real-world scenarios involving different kinds of soil settingsarediscussedinordertovalidatetheproposeMIsignalattenuationmodelandalso to answer additional questions, such as the impact of the misalignment between TX and RX coils in real deployment scenarios. The chapter in then concluded with the proposal of a new concept called breakpoint distance (BD) to be used in the context of MI-WUSNs, ratherthanthepopularbutambiguoustermnear-field. 3.1 FieldsDuetoaMultilayerCoil(TXside) If we assume that the electrical length of the inductor to be used in MI-communication is smaller than a tenth of a wavelength in soil (λ s ), we can properly model the inductor as a small loop antenna or, using duality, as an ideal short electric dipole [95]. We can start with the Maxwell’s equations to reach the general solution to the wave equation due to a CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 116 time-varying source (our ideal electric dipole) with volume V, as shown in Fig.3.1. The distance vectors r and r ′ refer to the distances origin-observation point and origin-source point, respectively. Considering an unbounded scenario, the solution to the wave equation inFig.3.1assumingalinear,homogeneous,andisotropicmediumisgivenby[18]: E(r)=−jωμ Z V J(r ′ ) e −γ·|r−r ′ | 4π|r−r ′ | dv ′ + 1 jωε ∇ ∇· Z V J(r ′ ) e −γ·|r−r ′ | 4π|r−r ′ | dv ′ (3.1) Ananalyticalexpressionfor(3.1)canbeobtainedforspecificcases. Forinstance,assuming an electrically short dipole (lengthΔl <<λ) aligned with the z-axis and a uniform current I,wecansolve(3.1)withthecurrentdensityJ= ˆ zIΔl inordertofindthecomponentsofE andHasfollows: E r = cosθ 2π IΔl s jωμ σ e + jωε ′ γ 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr (3.2) E θ = sinθ 4π IΔl s jωμ σ e + jωε ′ γ 2 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr (3.3) E φ =H r =H θ =0 (3.4) H φ = sinθ 4π IΔlγ 2 1 γr + 1 γ 2 r 2 e −γr (3.5) Bythedualityprinciple[18],itispossibletocalculatethefieldsofasmallloopcentered attheoriginandplacedatthex-yplanbyusingtherelation I m Δl = jωμSI o , where S isthe areaoftheloop,giveninm 2 ,and I o istheRMSelectriccurrent,givenin A(allunitsinthis workifnotexplicitlystatedareS.I.units): H r = cosθ 2π I m Δl s σ e + jωε ′ jωμ γ 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr (3.6) H θ = sinθ 4π I m Δl s σ e + jωε ′ jωμ γ 2 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr (3.7) H φ =E r =E θ =0 (3.8) E φ =− sinθ 4π I m Δlγ 2 1 γr + 1 γ 2 r 2 e −γr (3.9) CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 117 For the MI method, we are primarily interested in analyzing the magnetic field com- ponents in (3.6) and (3.7). In these equations, if γr >> 1 then the terms with 1/γ 2 r 2 and 1/γ 3 r 3 vanish (i.e., the reactive fields are too small and can be neglected) and we have the radiationasthedominatingphenomenoninthephysicalregioncalledfar-field. Ontheother hand, whenγr<<1, the term 1/γ 3 r 3 dominates in both equations and this physical region is called near-field, which is characterized by strong reactive fields (inductive for a loop, capacitive for an electric dipole). Withγr<<1, the term e −γr is approximately equal to 1 which is an indication that the radiation is practically not occurring at that physical region. Nonetheless, we will not simplify the expressions based on an assumption of near-field op- erationandthelastsectionofthischapterwilldiscussthisaspect. Thecomplexpropagation constantγ intheseexpressionsisgivenby(1.30). Next, two important assumptions in this research are used: 1) typical soils are non- magnetic media and 2) only air-cored coils are employed (ferro-magnetic cored-coils are out of the scope of this research). Hence, we apply the equality μ =μ o at the expressions (3.6)and(3.7). Infact,allexpressionsinvolvingμ areautomaticallyconvertedtoμ o inthis work. Also,becauseweareonlyinterestedatthemagneticfieldcomponents(H r andH θ )for the MI technique, E φ can be disregarded from now on. Finally, we also want to include the numberofturnsN TX ofthetransmittingcoil[96]andapplytherelationI m Δl= jωμ o S TX I TX in order to isolate TX circuitry aspects from other parameters related to the signal soil path attenuation. To this end, the TX circuitry aspects are grouped in a term called G TX which standsforTXgain: G TX ,N TX ·S TX ·I TX (3.10) γ = p jωμ o (σ e + jωε ′ ) (3.11) H r = cosθ 2π G TX ωμ o " j s σ e + jωε ′ jωμ γ 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.12) H θ = sinθ 4π G TX ωμ o " j s σ e + jωε ′ jωμ γ 2 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.13) Next,thetermγ 2 isexpandedandalgebraicmanipulationsleadto: CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 118 H r = cosθ 2π G TX ωμ o " j s σ e + jωε ′ jωμ (jωμ o (σ e + jωε ′ )) 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.14) H r =G TX cosθ(ωμ o ) 3 2 2π " j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.15) H θ = sinθ 4π G TX ωμ o " j s σ e + jωε ′ jωμ (jωμ o (σ e + jωε ′ )) 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.16) H θ =G TX sinθ(ωμ o ) 3 2 4π " j p j(σ e + jωε ′ ) 3 2 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (3.17) Observe that if we makeθ =0, the value for H r in (3.15) is the double of the value for H θ ifθ =90 o in (3.17). This observation leads to the question about the best deployment arrangement and misalignment aspects. The second aspect will be discussed later and the formeroneisassociatedtotheconceptofhorizontalaxesdeployment(coilinverticalposi- tion but the axis of the coil is in horizontal position, like the wheels of the car in relation to theroad)orverticalaxesdeployment(coilinhorizontalposition,thatis,paralleltothesoil surface, but the axis of the coil is in vertical position) [53, 56]. If the TX and RX coils are actually tri-axial coils such as in [50], there is no need to select one angle (θ =0 or =90 o ) and misalignment issues are also eliminated. However, we believe that the costs of instal- lation of the MI system will dramatically increase if the MI device has a cubic form with relative big dimensions (e.g., radius =30cm). Therefore, in our implementation discussed inthisworkweoptedbyasimplecoilwithasingleaxisandthehorizontalaxesdeployment wasselectedduetoitsincreaseofthereceivedmagneticfieldasalreadydiscussed. The trade-off of the horizontal axes deployment, such in Fig. 3.2, compared with the vertical option is the lack of a quasi-omnidirectional pattern for the TX coil in relation to the RX coils [56], assuming that all coils are at the same burial depth. Therefore, the net- work topologies are based on lines of nodes with this kind of deployment. At chapter 9, a networksolutionwillbegivenforthisscenario. Itisinterestingtohighlightthatthereisan optimumθ angle (around 27 o ) associated with a total magnetic field which is still stronger compared to the one at the θ =0 option. However, in general, such angle is not practical forreal-worlddeploymentsduetothepotentialcoilmisalignmentsissues. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 119 Fig.3.2Horizontalaxesdeployment: strongerRXsignal,constrainednetworktopology. Withoutlossofgeneralityforfuturediscussionsinthiswork,wecannowassumeθ =0 andsimplifytheevaluationoftheMIsignalattenuationproblembylimitingtheanalysisto a single field component: H r . Note that, besides the convenience of this approach, there is other important reason behind such decision: the elimination of the electric field E φ which could be a potential noise source at the RX side at the boundaries of the near-field. At the near-field, the electric and magnetic field are potentially not in-phase, not in-quadrature, and attempts to filter the electric field at the RX side may be difficult to implement. Next, consideringθ =0 o ,wetaketherealpartoftheexpressionforH r in(3.15)withtheobjective to determine the exact magnitude of the magnetic field|H r | MI which is associated with an inducedvoltageattheRXcoil(G TX andγ arerepeatedforconvenience): G TX ,N TX ·S TX ·I TX (3.18) γ = p jωμ o (σ e + jωε ′ ) (3.19) |H r | MI =G TX (ωμ o ) 3 2 2π ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (3.20) Ifθ =0, we determine the total magnetic field at the RX side as|H| MI =|H r | MI . How- ever, ifθ is different than 0, one can derive|H θ | MI in a similar way as in (3.20) and deter- mine the total magnetic field as|H| MI =|H r | MI +|H θ | MI . Considering the fact that the MI methodisonlyenergy-efficientiftheterm1/γ 3 r 3 dominatesthemagneticfieldexpression, wecanderiveaverysimpleexpressionfor|H| MI inthecaseswhere1/γ 2 r 2 ≪1/γ 3 r 3 : |H| MI = G TX 2πr 3 ℜ e −γr = N TX ·S TX ·I TX 2πr 3 ℜ e −γr (3.21) CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 120 Table3.1Asymptoticbehaviorofthemagneticfieldin(3.22). Frequency f WatercontentWC) Distancer Attenuationfactorα e −αr 1KHz 5% 10m 0.00046Np/m 0.9954 1KHz 40% 10m 0.00229Np/m 0.9774 1KHz 40% 30m 0.00229Np/m 0.9336 10KHz 40% 10m 0.01539Np/m 0.8574 10KHz 40% 30m 0.01539Np/m 0.6302 100KHz 5% 30m 0.01348Np/m 0.6674 100KHz 40% 30m 0.1020Np/m 0.0469 If we disregard the initial phase due to the distance given by βr in (3.21), we reach a veryusefulexpressiontostudyasymptoticbehaviorofMIsystems: H MI = N TX ·S TX ·I TX ·e −αr 2πr 3 (3.22) whichisvalidonlywhenθ=0 and (γr) −2 ≪(γr) −3 . By using the above expression (3.22), one can verify that the attenuation given by the term e −αr is very small for different values ofα, that is, for different soil conditions, pro- videdthatrelativelysmalldistancesrandfrequency f areconsidered. Someexamplesusing values for α based on the MI-SOIL model v.6 considered at the last chapter are shown in Table 3.1. Observe that frequencies smaller than 10KHz and distances up to 30m, the MI circuit is very resilient to changes at the soil moisture. Note that for f =100 KHz, the as- sumption (γr) −2 ≪(γr) −3 may not hold for mid-range distances such as 30m. This novel observationrelatedtotheresilienceoftheMIwirelesschannelunderspecificcircumstances hasadramaticimpactatthechoiceofthelower-boundfrequency(LBF),aswewillseelater inthiswork. 3.2 InducedVoltageinaMultilayerCoil(RXside) At this point of this study, we already have the necessary parameters to calculate the mag- netic flux density B at the RX coil distant r meters from a circular TX coil which has an alternating sinusoidal current I TX passing through it. The magnetic flux density B can in- creaseiftheRXcoilhasaferromagneticcore. Forthescopeofthisresearch,onlyair-cored CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 121 coilsarestudied,thereforewecanderiveBconsideringμ RX =μ o ,asfollows[18]: |B| MI =|H| MI ·μ RX =|H| MI ·μ o (3.23) where μ o is the permeability of free-space and|H| MI is total magnetic field and (3.21) can beusedifθ=0or,else,(3.21)canbeemployedif (γr) −2 ≪(γr) −3 isalsosatisfied. Next,wewanttofindanexpressioncorrespondingtotheinducedvoltageattheRXcoil due to B. This problem is more complex because the geometry and orientation of the RX coil impact the results. To solve the problem, we start by using the integral form of the Faraday’s law to reach the well-known expression for the induced voltage at a stationary loopinthepresenceofatime-varyingmagneticfluxdensityB[18,96]: V ind (t)=−N RX Z S RX ∂B ∂t ·ds (3.24) whereN RX isthenumberofturnsattheRXcoilandS RX isthesurfaceareaoftheRXcoil. The dot product in (3.24) implies that the orientation of the RX coil and B are related by cosφ, such that the reception angle φ=0 and the angle θ=0 when TX and RX coils are perfectly aligned. In this study, we assume that both coils are properly aligned (or, alterna- tively, triaxial coils are employed). This is a realistic assumption provided that during the deployment of MI coils the coils are adjusted for maximum signal transfer. Next, by eval- uating the integral in (3.24), using (3.23), assuming a sinusoidal TX source, and dropping theminussignalforconvenience,wehave: |V ind |=(N RX ·S RX )·|B| MI (3.25) |V ind |=(N RX ·S RX )·|H| MI ·μ o (3.26) where the|V ind | is the magnitude of the induced voltage across the inductor (given in V), notnecessarilyattheinputoftheRXpre-amplifier(PA)whichwewillcallV C . Fortunately, itispossibletoachieveavoltagegain(orattenuation,ifcareisnottaken)inrelationtothis voltage level|V ind | andV C , the voltage across the input of the PA, by means of resonance andthisisthetopicofthenextsection. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 122 Fig.3.3LCresonancecircuitsforMI-WUSNnodes. 3.3 ResonanceTechniquesforMI-BasedWirelessCommu- nication When a coil with an inductance L does not have any capacitor in series or in parallel, an interesting phenomenon occur. All non-ideal, real-world electronic components present some form of parasitics capacitance (or inductance) which has a value that depends on the frequency. Therefore, when a capacitor is not used in conjunction with an inductor, the existing parasitics capacitance will act as the capacitor C of a tank LC circuit. However, because typically we cannot determine the capacitance value of this hidden C, we end up with a tank LC circuit that can have a very bad performance at the operating frequency of the system. Therefore, if a coil is part of the communication system, such as in MI-WUSN nodes, the design of an explicit resonant LC tank is the best approach. Before continuing this discussion, it is important to remember that, in fact, the term LC tank actually refers to LCR tank because the effective resistance of the coil is always present. The resistance R is a critical factor to determine the merit factor Q of the coil and it is also associated with the bandwidthofthesolution. Therearediversewaystoachieveefficiencywithresonantscircuits. However,typically for the MI-WUSN circuits there are just few options. On the other hand, if the goal is wirelesspowertransfer(WPT)technology,theimpedancematchingatbothTX/RXsidesis paramountandtheresonantcircuitrycanbeverycomplicatedduetothepotentialexistence of many stages of matching circuits. Again, for our MI application, the resonant circuit design is straightforward, at least regarding the TX circuit, as shown at the left side of Fig. 3.3. In this case, the series capacitor configuration is the best solution and the explanation issimple. At resonance, the reactive impedances of L and C reach the same magnitude. Because these reactances are out-of-phase, the reactive impedance of the circuit is canceled, i.e., CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 123 0. Therefore, the current of the signal source is only limited by an existing resistances in serieswiththeinductor,suchastheinternalresistanceofthesourceandtheeffectiveresis- tance of the TX coil, R ef f , which takes into account the wiring DC resistance, the wiring skin-depth resistance, the coil proximity effect resistance, and additional core losses (the last one does not apply for air-cored coils). At resonance, the TX circuit with a series ca- pacitor provides the highest current and according to (3.22) and (3.26), the highest induced voltage is achieved at the RX side. The only issue with this configuration is the need of a low impedance TX driver circuit, which is usually not an issue for LF amplifiers operating at/near audio frequencies. Another aspect in this configuration requires more attention: if the effective resistance of the TX coil is very small, the TX driver may not have enough high-current capability to support the coil load. In other words, although the internal re- sistance of the TX coil, R ef f , must be kept at low values in order to achieve the highest MI channel efficiency, there is still a circuit constraint. For instance, it is very difficult to design a TX amplifier for a load with R ef f =0.1Ω. In this case, R ef f is typically increased bychangingN TX toahighervalue. Turning our attention now to the RX circuit, the LC resonant tank at the RX side can have a series or a parallel configuration, as shown at the right side of the Fig. 3.3. Many WUSN papers show a configuration where the RX load is in series with the coil and ca- pacitor. This configuration is one the ideal configurations for power transfer systems, but it is not a proper one for exclusive data transfer purpose for two reasons. First, in a series configuration,theRXpre-amplifier(PA)wouldhavetopresentlowimpedance,comparable the R ef f of the RX coil, which is not an usual option for high-sensitivity, high-impedance PA devices. Second, this series configuration voids the possibility of having a significant passive voltage gain at the RX LC tank. This argument can be visualized by means of Fig. 3.5. IftheRXPAisinserieswiththecoil,asintheleftsideofthefigure,thevoltageatthe inputofthePAisalwayssmallerthantheinducedvoltageacrossthecoil. Infact,thereisa potentialsignalattenuationduetothisseriesconfiguration. On the other hand, the figure at the right side shows a configuration where it is possi- ble to achieve a very high voltage gain. For instance, if we assume that the frequency is fixed and L increases (forcing C to be smaller to maintain the resonance), the value of V C increases, that is, the RX gain increases. Also, if the effective resistance of the coil R ef f decreases, the gain also increases. Nonetheless, there is an assumption behind this parallel configuration that must be highlighted: the RX PA must have a very high impedance (e.g. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 124 At resonance f = 1 / [ 2 (LC) ] (Hz) RX circuit model = 1 / [ (2 • f)² L ] (F) V / R (A) C ind I ind eff 2.83 I • X c V eff c pp = 2.83V /(2 • f • R • C) (V) ind ind RX G eff = 2 • f • L / R Examples (V =20 V, R =8 , L=1mH) eff Fig.3.4RXcircuitmodelandexamplesofvoltagegains/levelsfordifferentfrequencies. Fig.3.5Rightside: properLCresonanceconfigurationforRXcircuitofMI-WUSNnodes. >1MΩ) and low capacitance. Fortunately, this is the case for the majority of Instrumenta- tionAmplifiers(IA),thetypicaldevicesusedformid-rangeRXMIcircuitsduetotheirhigh sensitivity. Some examples of possible voltage gains due to resonance at the RX side are shown in Fig. 3.4. Observe that relative high frequencies, such as 1 MHz, are associated to very small values of the capacitorC, close to the value of the parasitics capacitance of the coil itself. This constraint is highlighted in [56] and it is reported that the value of C can drastically limit the communication capacity of MI systems. We will see in future chapters thatthisfactandadditionalfactorsforcemid-rangeMI-WUSNstooperateatsub-MHzfre- quencies. AlthoughthementionedRXresonanttankhasapotentialassociatedvoltagegain,G RX , itisalsopossibletooccurthecontraryandtohaveattenuation. Thiscaseusuallyoccursdue to a deviation of the L andC designed values from the resonance frequency. Some typical reasonsforsuchdeviationaretemperaturechange,choiceandvaluetoleranceofthecapac- itor. Moreover, by means of experimental work, we observed that when a coil is buried, a CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 125 slight modification of the L value also occurs. However, depending if the merit factor Q of the LC tank is very high, small deviations of C or L, such as 1%, has dramatic effect on the circuit performance. For instance, assume that at 20 KHz, the calculated resonant capacitorforacertaininductorwithmeasuredL=17.25mHis3.67nFandR ef f =8Ω. How- ever, the measured value of the capacitor is 3.35 nF, still under the 10% tolerance specified by the manufacturer. The question we want to answer is what quantitatively happens with the received signal in such case. In other words, we want a mathematical expression for the potential attenuation or loss of gain in relation to the designed value. To derive such expression,considerthecircuitattherightsideinFig. 3.5andfollowtheseequationsbased onfundamentalsofcircuittheory: V C ≃I coil ·X c =I coil 1 jωC (3.27) I coil ≃ V ind jωL+R ef f + 1 jωC (3.28) V C = V ind jωL+R ef f + 1 jωC 1 jωC (3.29) V C =V ind 1 −ω 2 LC+ jωCR ef f +1 (3.30) G RX , 1 1+ jωCR ef f −ω 2 LC (3.31) V C =V ind ·G RX (3.32) whereV ind isgivenin(3.26). Therefore, if we plug the values of the proposed problem to (3.31), considering both C-value cases, we figure out that G RX for the exact designed value ofC is 271.0, while its value for the deviated value ofC is 11.4. The RX circuit still has a gain, but only 4.2% of theexpectedgain. ThisexampleillustratestheneedofaccuratevaluesforLandCwithpos- sible adjustments during deployment. The higher the designed value of G RX , the higher is thecomponentsdeviationimpacttothesolution. Onepotentialwaytomitigatethisissueis to employ military-grade capacitors at the resonant tank, with value-tolerance smaller than 2%andthatarehighlystabletotemperaturechanges. At resonance, if the values of L and C are exactly the same as the designed ones, the CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 126 reactiveimpedancewillbezeroandG RX in(3.32)assumesitsmaximumvalue,whichisthe oneshowninFig. 3.5: Atresonance: jωL= −1 jωC (3.33) Atresonance: ω 2 LC =1 (3.34) G max RX = 1 1+ jωCR ef f −(1) (3.35) G max RX = 1 2πfR ef f C = 2πfL R ef f (3.36) 3.4 MISignalAttenuationModelforWUSNs So far, the MI-TX and RX circuits have been analyzed separately while such approach is not allowed in the inductively-coupled circuits theory [96]. From circuit fundamentals, it is well-known that a secondary of an inductive circuit appears to the primary circuit as an additional in series impedance and similar effect to the secondary circuit. However, the novelapproachweareproposinginthisresearchistoseparatetheTXandRXcircuitsasif they were completely independent. This approach has the important advantage of allowing a better analysis of the overall MI signal attenuation because it is easier to identify, sepa- rately, the impact of each part: TX circuit, soil medium, and RX circuit, which is the way we did since the beginning of this chapter. Moreover, we will see at the following chapters that the optimization of the solution, such as for the coil design and also for the selection oftheoperationalfrequencyrange,issignificantlysimplifiedwiththementioneddecoupled TX/RXcircuitsapproach. Nonetheless,westillneedtoverifyifthisapproachisacceptable intermsoferrorsordistortionsforthemodel. Thepresentedtechnique,whichcompletelyseparatesTXandRXcircuits,indeedholds ifweassumemid-rangedistancesbetweenTXandRXcoils. Toquantitativelyanalyzethis statement, the coupling factor k (not to be confused with dielectric constant) is introduced and k= M/(L 1 ·L 2 ) 1/2 [96]. M is the mutual inductance, given in H (henry) units, which is defined as the number of magnetic flux lines that appear at a second coil per unit current passingthroughthefirstcoil[97]. Alternatively,M canbeexpressedastheinducedvoltage atthesecondcoilduetoanalternatingcurrentpassingthroughthefirstcoil: E 2 =−jωMI 1 [97]. However, for strongly coupled magnetic circuits, such as in transformers, M is typi- cally given only in terms of the geometry aspects regarding the two coils, number of turns, CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 127 andfrequency. Fordifferentcoilphysicalforms,distinctformulasforM arepromptlyavail- able at the literature [98]. The bottom line is that M exponentially decreases with an in- creasing distance r between the coils and when it is very small the primary and secondary circuitsaresaidtobelooselycoupled. The actual impact of M for a pair of resonant TX/RX tanks is the decrease of the max- imum current I TX due to the presence of the RX coil at the physical neighborhood of the TX coil. By decreasing I TX , the value of G TX in (3.10) also decreases, meaning that the actualmagnitudeofthemagneticfieldH MI in(3.21)andin(3.22)issmallerthanpresented in these formulas. Our goal is to quantitatively determine this error in order to evaluate the impact of this approach. For lossy medium, such as wet soil, such calculation is complex, but we can limit our analysis to the free-space case where the mentioned error due to the decoupled TX/RX circuits approach is even stronger than in soil medium. Therefore, if we figure out that such error err is acceptable in free-space, it is secure to also consider err as the maximum error for UG settings. We use first principles in circuit theory [97] and re-define (just for this analysis), the induced voltage at the RX coil at resonance conditions withidenticalTX/RXcoils(sameR ef f ),asfollows: V ind =−jωMI TX (3.37) M = j V ind ωI TX (3.38) I max TX = V TX src R ef f (3.39) I actual TX = V TX src R ef f + ω 2 M 2 R ef f (3.40) err = I max TX −I actual TX I max TX (3.41) err =1− R ef f R ef f + ω 2 M 2 R ef f (3.42) It is clear by inspection in (3.42) that err is maximum when (ω 2 M 2 )≫R ef f . We will now plug some values into the previous equations that come from our empirical real-world field experiments. The case is for the soil NHS-DRY with WC=5%, r=20m, f=10 KHz, R ef f ≈7Ω,I TX =103.7mA,andV ind =3.2μV.Proceedingwiththecalculationsanddropping theimaginarypartin(3.38),wehave: CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 128 M = 3.2E−6 2π·10E3·103.7E−3 =4.9E−10 err =1− 7 7+ (2π10E3) 2 (4.9E−10) 2 7 =1.9344E−011 For this specific case, clearly the error due to the decoupled TX/RX circuit approach is negligible. Verysmallvaluesoferr willalsooccurfortypicalmid-rangeMI-WUSNcases. However, the error increases if the distance r is significantly reduced and/or the frequency increases. For very small values of k=M/L (e.g., k<10 −4 ), the exact circuit analysis has a value forV C (the voltage level across the RX PA’s input) which differs from our simplified model by less than 1% for the majority of the practical mid-range cases we investigated. Sucherr=1%simplymeansthatV C isactually99%ofthemodelvalue. Comparesucherror with the impact of 1% variation at the nominal value of the capacitance C of the resonant capacitor, as previously discussed. The latter capacitance deviation may cause order errors atV C andprohibittherealizationofthecommunication. Therefore,consideringthenegligi- ble error effect of the decoupled TX/RX circuit approach, from now on in this research the termsrelatedtomutualinductanceM andcouplingfactork willnotbeused. Thissectionisconcludedwithaself-containedversionoftheMIsoilattenuationmodel. Itappliesfor3categoriesofsoil,accordingtotheMI-SOILmodel(v6,v7,v8). Consulting theTable1.1,thev6modelistypicallyusedformediumloamandsandyloamsoils,suchas NHS, LOA1, LOA2, TONZI, NHS-DRY1, NHS-WET, and NHS-DRY2 soil textures. The v7modelisusedingeneralforsandysoils,suchasSAN1. Thev8modelisusedforhighly- dispersive soils, such as NHS-SAT which has silty loam texture with minimal amount of sand. As discussed in Chapter 3.1, new MI-SOIL models can be derived for specific cases ofsoilsfollowingtheguidelinesinthatchapter. ThefollowingaretheassumptionsfortheproposedMIsoilattenuationmodel: (a) Homogeneousandunboundedmedium: thepreliminaryempiricalworkweperformed indicatethatsuchassumptionsarenottoostrongforMIoperationandlowfrequencies (e.g.,<200KHz). (b) Non-magnetic soil: magnetic soils are very rare and typically not related to the main applicationsofWUSNs. (c) TXandRXcoilsareidentical: air-cored,circularshape,singleormulti-layer. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 129 (d) TXcoilhasaseriesresonantconfigurationasintheleftsideofFig. 3.3. (e) RXcoilhasaparallelresonantconfigurationasinFig. 3.4. (f) TheRXPAhashighinputimpedance(i.e.,>1MΩ)andlowcapacitance(i.e.,<5pF). Theultimatemodeloutputparameterofinterestistheinducedvoltageattheinputstage oftheRXPA,V C . Notethatthisapproachisprettydistinctfromthetypicalvoltagelevelat- tenuationorenergyattenuationgivenindBvalues. Thereasonforthisapproachisrelatedto thefactthatnostandardimpedance,suchas50Ω,isadoptedforMIcircuits. Therefore,the sensitivityofaRXMIdevicepotentiallywillbegivenintermsoftheminimumacceptable V C level considering the existing noise levelV noise C . This scenario is different from the one forRFdeviceswherespecificationsaregivenintermsofdBmforbothTXandRXcircuits. Inthislatterscenario,itispracticaltoprovidethesignalpathattenuationandotherlossesin term of dB to simplify the calculations. On the other hand, for MI circuits in MI-WUSNs, itispotentiallybettertoencapsulateTXandRXcircuitryaspectsintermssuchasG TX and G RX . Inourproposedapproach,wegetthevaluesofG TX andG RX and,next,weconsiderthe soil path attenuation model in order to verify if the V C level is enough for a reliable com- municationunderaspecificsetofdeploymentandenvironmentalparameters. Nonetheless, it is relatively simple to derive the total energy loss of the TX/RX MI system by consider- ing the supplied energy at the TX side and the received input energy which is equal to the product of time operation and the input power given by P input = (V C ) 2 R ef f , assuming resonant circuits. Then, the ratio of these energy values can be converted to dB. If the goal is to numericallyachievethedBattenuationforspecificallythesoilpath,itisimportanttoadopt thesamemethodologyusedinwirelesscommunication: defineareferencedistanced o (e.g., 1m) away from the source and calculate the H field at that point [19]; then, calculate the H fieldattheRXpoint. Hence,thesoilpathloss(dB)canbecalculatedasfollows: Atten soil (dB)=10log H MI d=r H MI d=do (3.43) where d in this context is the distance between the source and the observation point and H MI can be calculated using (3.20), assuming θ =0. If θ6=0, such as when coils with 2 or 3 aves are employed, the H θ must also added to the previous H r field component using (3.17). CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 130 NotethattheabovesoilpathattenuationcalculationdoesdonotdependonTXandRX circuitry parameters, as expected. Once such guidelines have been provided, we return to ourinitialgoalofforecastingtheinducedvoltageV C bymeansoftheproposedmodel. This valueforthealreadydiscussedθ =0caseisgivenby: V C =V ind ·G RX (3.44) G partial RX , 1 1+ jωCR ef f −ω 2 LC (3.45) V C =V ind · 1 1+ jωCR ef f −ω 2 LC (3.46) Atresonance: V max C =V ind · 2π ˙ f·L R ef f (3.47) V ind =(N RX ·S RX )·|H| MI ·μ o (3.48) V max C =(N RX ·S RX )·|H| MI ·μ o · 2π ˙ f·L R ef f (3.49) G full,max RX ,N RX ·S RX ·μ o · 2π ˙ f·L R ef f (3.50) V max C =G full,max RX ·|H| MI (3.51) G TX ,N TX ·S TX ·I TX (3.52) Forθ=0:|H| MI =G TX (ωμ o ) 3 2 2π ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (3.53) V max C =G full,max RX ·G TX · (ωμ o ) 3 2 2π ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (3.54) CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 131 The values for the soil parameters γ, σ e , and ε ′ to be used in (3.54) depend on the frequency f in Hz, on the volumetric water content level (VWC) in %, and on the version oftheMI-SOIL,asfollows: b real (v6) =0.052(VWC)+3.4 (3.55) m real (v6) =−0.0054(VWC)−0.385 (3.56) b imag (v6) =0.0278(VWC)+4.46 (3.57) m imag (v6) =0.00380(VWC)−0.526 (3.58) ε ′ (v6) =10 b real (v6) f m real (v6) (3.59) ε ′′ e(v6) =10 b imag (v6) f m imag (v6) (3.60) b real (v7) =0.0385(VWC)+3.375 (3.61) m real (v7) =−0.0049(VWC)−0.398 (3.62) b imag (v7) =0.0195(VWC)+3.965 (3.63) m imag (v7) =0.00225(VWC)−0.352 (3.64) ε ′ (v7) =10 b real (v7) f m real (v7) (3.65) ε ′′ e(v7) =10 b imag (v7) f m imag (v7) (3.66) b real (v8) =0.054(VWC)+4.2 (3.67) m real (v8) =−0.0051(VWC)−0.535 (3.68) b imag (v8) =0.0362(VWC)+4.72 (3.69) m imag (v8) =0.00425(VWC)−0.595 (3.70) ε ′ (v8) =10 b real (v8) f m real (v8) (3.71) ε ′′ e(v8) =10 b imag (v8) f m imag (v8) (3.72) σ e =ω·ε ′′ e =2πf·ε ′′ e (3.73) γ = p jωμ o (σ e + jωε ′ ) (3.74) CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 132 Observe that, so far, the unknown parameters of the model are circuitry-related: N TX , S TX , I TX , N RX , S RX , R ef f , and L. The value of the resonant capacitance C can be derived from L using the relation C = 1 ω 2 ·L . These unknown parameters are analyzed at the next chapters. Thenextstepinourresearchstudyistovalidatetheproposedmodelbyempirical work. 3.5 ExperimentsandValidation In order to validate the MI soil attenuation model proposed at the last section, field experi- mentsareperformedwith2pairsofcoils. Theinstrumentationusedfortheexperimentsare listedbelow: (a) CustomizedanalogRMSmicrovoltmeter: mainlybasedontheinstrumentationam- plifier AD8421 (Analog Devices Inc.) which has a configurable voltage gain of up 10,000. ThesameRXcoilusedattheexperiments,oranyothercoil,canbeattached to this instrument. The AD8421 circuitry is then calibrated for a fixed gain of 1,000 and its output is connected to a RMS-to-DC converter in order to feed or an analog- to-digital(ADC)lineofamicrocontroller(MCU),orananalogpanel(afterthesignal passes through a current buffer). Again, the overall circuit is calibrated to have an error smaller than 1% for input signals smaller than 100μV. This instrument is used for prompt evaluations of the magnitude of magnetic field and it serves as an prelim- inary instrumentation before employing the MI digital instruments. The circuitry of thisdeviceiseventuallyusedattheRXcircuitrypartofourMInode. (b) TX node: refers to the TX part of the MI node. Initially, TX and RX parts are sep- arated and only unidirectional communication tests are performed. Later, both parts are migrated in a single board allowing the MI node to perform bi-directional com- munication. For the tests presented in this chapter, it is sufficient and convenient to consider TX and RX functionalities separately. The TX node comprises by a MCU, a precise sinusoidal signal generator (DDS), a 1W power amplifier (actual consump- tionwassmallerthan500mWduringthetests),theTXcoilanditsresonantcapacitor. ThesystemisDC-poweredwith5..6V.TheMCUhastwoconfigurations: continuous- wave(CW)modeorregulardatatransfermode. AtCWmode,theDDScontinuously generates a sinusoidal signal, while at regular mode, the on-off-keying (OOK) mod- ulation is used for data transfer at around 1,000 bits per second (bps). The induced voltage measurements to validate the model are performed with the TX node in CW CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 133 mode. The packet error rate (PER) evaluation is performed with the TX node in reg- ularmode. (c) RXnode: referstotheRXpartoftheMInodeanditismainlybasedonthementioned analog microvoltmeter above. The difference lies on the use of a MCU to perform package error detection and received signal strength (RSS) tests. A simple cyclical redundancy check (CRC) is employed to detect error on data streams and the ADC line is used to evaluate RSS. The critical part of the circuitry is the automatic gain control(AGC)algorithmnecessaryforthementionedOOKmodulation. (d) Coil 1A: circular, multilayer, and air-cored coil with an approximated square cross- section (details on Chapters 5 and 6). Diameter=12cm, 29AWG wiring, N =116. At 10KHz,thecoilhasmeasuredvaluesforself-inductanceL=3.012mHandR ef f =11.7Ω. (e) Coil 2A: circular, multilayer, and air-cored coil with an approximated square cross- section. Diameter=12cm,29AWGwiring,N=150. At10KHz,thecoilhasmeasured valuesforself-inductanceL=5.01mHandR ef f =15.2Ω. (f) Coil 1B: circular, multilayer, and air-cored coil with an approximated square cross- section. Diameter=30cm,26AWGwiring,N =56. At10KHz,thecoilhasmeasured valuesforself-inductanceL=2.692mHandR ef f =7.615Ω. (g) Coil 2B: circular, multilayer, and air-cored coil with an approximated square cross- section. Diameter=30cm,27AWGwiring,N =74. At10KHz,thecoilhasmeasured valuesforself-inductanceL=3.922mHandR ef f =11.71Ω. 3.5.1 SetofExperimentsA This experiment was performed at the garden of a residential place at Pasadena, CA, USA on 25 June 2014. The measured VWC is 1.6%. The nodes TX and RX described above are used, as well as Coil 1A as TX coil and Coil 2A as RX coil. Resonant capacitors with capacitance values close to the design values are employed for f=10 KHz. The measured currents are I TX =134mA(rms) and I TX =137mA(rms) for dry and saturation soil conditions, respectively. The burial depth is 25cm, that is, there is a soil layer of around 13cm above both coils. The soil type is LOA1, as described in Table 1.1. The distance r between the properlyalignedcoilsis5m. Thecoilsaremountedinaplasticsupportandaretestedwith- out any kind of enclosure. Two soil conditions are tested: dry (VWC=1.6%) and saturated (VWC=35.5%). Thesoilsaturationconditionisachievedafteraround1hourofirrigationat CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 134 Fig.3.6SetofexperimentsA:soilsettingsnotimpactingtheMIchannel. all soil point nearby and between the coils. Later, the soil saturation was verified at diverse pointsdownto35cmdepth. For the data transfer test (each one involving 200 of 70-byte messages), the pair of nodes performed unidirectional communication at 1,000bps with an average PER of 8.6% and 7.7% for dry and saturated soils, respectively. Regarding the RSS tests, the TX node is configured for CW mode and the RX node is configured to monitor the RSS level which is given in terms of voltage level at the input stage of the RX PA. These results are shown in Fig. 3.6, where it is possible to verify a good match between the proposed model and the empirical results. One can also observe that for relatively small distances and frequencies, the soil settings have a small impact on the RX induced voltage, as previously highlighted inthischapter. 3.5.2 SetofExperimentsB The second set of experiments was performed at a native land (restricted access) owned by the Bureau of Land Management of the California government, Jackson, CA, USA on 5 August 2014. The main goal is to verify the severity degree of TX/RX coils misalignment. Because this problem is only associated with the geometry of the deployment scenario, we opted for over-the-air tests to address this research problem. The node TX and the analog RMS microvoltmeter are used, as well as Coil 1B as TX coil and Coil 2B as RX coil (the latterisattachedtothemicrovoltmeter). Theresultsaresummarizedbelow: CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 135 (a) Small vertical (height, depth) misalignment: at a distance of 15m, the RX coil is placedinverticalposition(alsocalledhorizontalaxesdeployment)exactlyabovethe soilsurface. TheTXcoilhasitssurfaceparalleltotheRXcoil’ssurface,butitscenter isshiftedup1minrelationtotheRXcoil. Forthiscase,theθ angleisapproximately 3.8 o and cosθ=0.9978. The φ angle is also changes in this scenario and it follows similar deviation. The combined effect on the induced voltage V c is expected to be around0.5%. Themeasuredvaluesshowavariationof0.52%. Fordistancesr higher than15m,theangledeviationsareevensmaller. Thesameexperimentisperformedat theExperimentSetCwithonenodeburiedandtheothermovingfromthesoilsurface to 1m up for the same distance r=15m and the signal variation is smaller than 1%. Therefore,itisconcludedthatformid-rangedistancesthiskindofcoilmisalignment isnotcriticalprovidedthatthedeviationisnotabnormallyhigh. (b) Small lateral misalignment: at a distance of 15m, the RX coil is placed in vertical position exactly above the soil surface. The TX coil has its surface parallel to the RX coil’s surface, but its center is shifted to the right (also left) 1m in relation to the RX coil. This case is basically the same geometric problem as the previous one but theresultswillbesignificantlydifferentifthesoilinterfacehasastrongeffectonthe MI signal. The measured values show a variation of 1.8%. If the TX coil deviates 1mbothlaterallyandvertically,theresultisasmallerinducedvoltagewith96.7%of its value for a perfect alignment. Therefore, it is also concluded that for mid-range distances both vertical and lateral coil misalignments are not very critical, provided thatthedeviationisnotabnormallyhigh. (c) Big lateral misalignment: at a distance of 15m, the above test is repeated with a lateral deviation of 3m and it is performed for left and right cases. While for the previous tests, the actual distance r changed very little (from 15m to 15.033m), this newscenariochangesr moresignificantly,besidestheθ andφ anglesdeviation. The actual value of r in this scenario is 15.3m and the newθ is 11.4 o and cosθ=0.9804. The combined effect ofθ andφ deviations is expected to be almost 3.9%. However, the change at r is expected to cause a signal reduction of around 5.8% and the total combined signal reduction is expected to be 9.5%. However, the measured values show a variation of 19.9%, almost the double of the expected value. The reason for this mismatch is potentially the lack of accuracy on the way the above φ angle deviation calculations are being performed which requires further investigation. It is concluded that significant lateral misalignments, as the one demonstrated here, can impacttheinducedvoltageattheRXside. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 136 3.5.3 SetofExperimentsC This set of experiments was performed at the New Hogan Lake, Valley Springs, CA, USA on 6-7 August 2014. All the tests are UG experiments and 4 distinct sites at the same lo- cation are used: NHS-DRY1, NHS-DRY2, NHS-WET, and NHS-SAT. The characteristics of each of these kinds of soil are shown in Table 1.1. The nodes TX and RX previously discussed at the set of experiments A are also used here but in conjunction with Coil 1B as TX coil and Coil 2B as RX coil. Resonant capacitors with capacitance values close to the design values are employed for f=10 KHz. The measured currents vary according to thebatterylevelandrangefromI TX =103.7to153.6mA(rms). Noartificialirrigationisper- formed and different VWC levels are achieved by selecting sites closer and closer to the lake,exactlyatthisorder: NHS-DRY1,NHS-DRY2,NHS-WET,andNHS-SAT.Theburial depthvaryfrom38cmto41cmandsuchsmallvariationispotentiallynotcriticalaccording tothetestsofthesetB.Thedistancesr betweentheproperlyalignedcoilsare10,15,17.5, and20m,accordingtothesite/soilconditions. Theveryflatcoilsaremountedincardboards and a thin plastic tape provides the minimum physical protection to the coils (no abnormal behaviorisobservedinalltestsusingthesamepairofcoils). For the data transfer test, the same procedure described in set A is used, but this time onlyinvolving20messages(70-byteeach)inordertospeed-uptheexperimentsduetovery harshconditionsofthesesites. Moreover,bi-directionalcommunicationat1,000bpsisper- formed to verify a possible asymmetric behavior of the MI UG channel. This is not the case and we report for the first time that, based on empirical work, the MI UG channel be- tweentwoidenticalcoilshasindeedasymmetricbehaviordespitestheheterogeneityofthe soil. Asexpected,thisstatementdoesnotholdifundergroundmetallicobjectsareinvolved. The average PER is smaller than 5% in all cases and the main noise source at the RX side is identified as the input stage of the RX PA, not undesirable signals at the UG channel. This observation points to a potential reliable UG channel, if water content is initially dis- regarded. Regarding the RSS tests, the TX node is configured for CW mode and the RX node is configured to monitor the RSS level, similarly to the procedure at set A. The results for the first site, NHS-DRY1, are shown in Fig. 3.7, where the induced voltage V C at the RX PA input is given as a function of the distance r. The VWC level for this site is fixed and measured as 5.0%. It is possible to conclude based on this figure that all the models, in- cluding Scott’s 67 in [16], have a good match with the empirical results. All the MI-SOIL CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 137 Fig.3.7SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS-DRY1. sub-models appear as a single line due to the proximity between the results of each model. Moreover, although not shownin this figure, the signal variation is alwayssmaller than 3% in relation to the displayed average. Therefore, the variance interval is too small to appear atthefigure. Clearlyobservedinthefigure,theRXinducedsignal,whichisdirectlyrelated tothemagneticfieldattheRXpoint,presentsa 1 r 3 decayaccordingtothedistancer. The results for the second site, NHS-DRY2, are shown in Fig. 3.8, where the induced voltage V C at the RX PA input is given as a function of the distance r, although a single experiment is performed at r=15m. The VWC level for this site is fixed and measured as 5.9% on the average, but there are point of the path between the nodes where the VWC reaches values close to 2%. The models have a fair match with the empirical data and a potential reason for this slight worse performance is the observed VWC inhomogeneity of thissite. OnecanalsoobservethattheScott’ssoildielectricmodelshowsminordifferences in relation to our soil models at distances higher than 10m. We will see shortly that such differencewilldramaticallyincreasewithanincreasingVWClevel. The results for the third site, NHS-WET, are shown in Fig. 3.9, where the induced CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 138 Fig.3.8SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS-DRY2. voltage V C at the RX PA input is given as a function of the distance r, although a single experiment is performed at r=15m. The VWC level for this site is fixed and measured as 18.1%. Observe that the MI-SOIL sub-models start to show differences among them, in particular for distances higher than 22m for this VWC level. Moreover, all these MI-SOIL sub-modelspresentagoodmatchingwiththeempiricaldatabuttheScott’smodeldoesnot provideagoodmatchingforthisscenario involvingarelativelyhighVWClevel. Basedon the fact that significant mismatch occur at higher VWC levels, it is possible that the dielec- tric measurements in Scott’s work have distortions due to the EP effects, in particular for thislowfrequency(10KHz). The results for the last site, NHS-SAT (this one is very close to the lake), are shown in Fig. 3.10, where the induced voltage V C at the RX PA input is given as a function of the distance r and two distances are empirically evaluated: 10 and 15m. The VWC level for thissiteisfixedandmeasuredas32.8%. AllMI-SOILsub-modelspresentagoodmatching with the empirical data for both distances r while the Scott’s model continues to signifi- cantly deviate from the empirical results. For this high VWC level, it is now possible to identifydistinctbehaviorfortheMI-SOILsub-modelscurves. CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 139 Fig.3.9SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS-WET. TheerroranalysisrelatedtothepreviousresultsisprovidedinFig. 3.11,wherethenor- malizedinducedvoltageV C attheRXPAinputisgivenasafunctionoftheVWClevelfora fixeddistancer=15m. Thisfigureshowsdetailsthatarenotvisibleatthefoursemilogarith- mic plots previously discussed. It is observed that the MI-SOIL sub-models present a very good matching for VWC levels higher than 15% but they are significantly pessimistic for low values of soil moisture. On the other hand, the Scott’s model monotonically presents a pessimistic behavior, that is, they predict a RX signal level much smaller than the actual one. One can also observe that our proposed model, for the average case, improves the accuracyofthepredictionsincomparisonwiththeexistingScott’smodel. Inparticular,our proposedmodelpresentsaverygoodaccuracyfortheworstscenariosinMI-WUSNs: high soil moisture levels. Nonetheless, it is also possible to conclude that the model can still be improvedinrelationtothedrysoilscenarios. The proposed model was also validated in set A, when the sub-model v6 is adopted to predict the signal attenuation for LOA1, a different kind of soil not evaluated when v6 was developed. ThisfactindicatesthehighpotentialityofapplyingtheproposedMIsignalatten- CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 140 Fig.3.10SetofexperimentsC:ValidationoftheMIattenuationmodelforsiteNHS-SAT. Fig.3.11SetofexperimentsC:ErroranalysisoftheMIsoilattenuationmodel. uationmodelforalargevarietyofsoilsnotspecificallyconsideredinthisstudy. Preliminary investigationsrelatedtothisaspectoftheresearchshowsthatalthoughthesub-modelsMI- CHAPTER3. MISIGNALATTENUATIONMODELFORWUSNS 141 SOIL v6, v7, and v8 can lead to relatively significant deviations of ε ′ for certain kind of soils, the prediction of the MI signal attenuation is not impacted at the same level. There- fore,basedontheresultsinFig. 3.10,theaverageerrorofthemodelappliedtogenericsoil casesisaround10%and,ingeneral,themodelisslightlypessimistic. Despitesthisrelative large error (particularly worse for dry soils), the mentioned characteristic of the model is desirable in order to achieve successful real-world implementations. Nonetheless, these er- rors can still lead to non-optimum performance of the system. To address such challenges, the proposed MI node employs a dynamic frequency-switching scheme. In this research work,suchswitchingalgorithmisdefinedatthedesigntime. However,futureMInodescan employ on-the-fly schemes based on actual signal measurements. For such cases, the pro- posedmodelpresentedinthischaptercanprovideagoodinitialoperationalfrequencypoint fortheoperationofanodebeforetheoptimumpointiseventuallyachievedconsideringthe existingsoilconditions. Chapter4 OperatingFrequencyRangefor MI-WUSNs This chapter is mainly based on the contents of our paper [99]. The MI technique has been reportedasapotentialoptionforthephysicalcommunicationlayerinWUSNs[11,42,56]. Nonetheless,afundamentalquestionregardingthebestoperatingfrequencyforMI-WUSNs is still not fully answered. We anticipate that this answer, besides the electrical properties of the soil, is strongly related to the distance between the MI coils. In this chapter, this question is addressed for mid-range distances (e.g., 15..30m) and, to this end, the practical MIsignalattenuationmodelforWUSNswhichwasdevelopedinChapter1isused. 4.1 FieldsDuetoaMultilayerCoil(TXside) In this section the expressions for the magnetic fields due to a multilayer coil are repeated forconvenience. NotethattheywerederivedinChapter1. Two important assumptions are used here: a) typical soils are non-magnetic media and b) only air-cored coils are employed. The former assumption is reasonable because most soil constituents are nonmagnetic [2]. Regarding the latter assumption, our preliminary investigation of ferromagnetic cores shows two critical aspects. First, the potential signal gain at the RX coil is usually voided by the attenuation of the TX signal due to the eddy currents formed at the TX core. For this case, the same coil may not be used for the TX and RX roles. Second, the quality factor Q of the LC tank dramatically increases with a ferromagnetic core. Unfortunately, because the nominal value of L varies around 1% 142 CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 143 (according to our empirical investigation) as a function of the soil wetness, a high value of Q would require a dynamic adaptation of the C value. Because such aspects make our initial communication problem even more complex, in this paper we limit the discussion to the basic scenario involving only air-cored coils. Hence, we apply the equality μ = μ o (permeability of free space) at the expressions (3.6) and (3.7). In fact, all expressions involving μ are automatically converted to μ o in this work. Also, because we are only interested at the magnetic field components (H r and H θ ) for the MI technique, E φ can be disregarded from now on. Finally, we also want to include the number of turns N TX of the transmitting coil and apply the relation I m Δl = jωμ o S TX I TX [96] in order to isolate TX circuitry aspects from other parameters related to the signal soil path attenuation. S TX and I TX aretheTXcoil’sareaandtheRMScurrentpassingthroughthiscoil,respectively. Next, theTXcircuitryaspectsaregroupedinatermcalledG TX whichstandsforTXgain: G TX ,N TX ·S TX ·I TX (4.1) γ = p jωμ o (σ e + jωε ′ ) (4.2) H r = cosθ 2π G TX ωμ o " j s σ e + jωε ′ jωμ ·γ 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.3) H θ = sinθ 4π G TX ωμ o " j s σ e + jωε ′ jωμ ·γ 2 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.4) Expandingthetermγ 2 using(4.2),weget: H r = cosθ 2π G TX ωμ o " j s σ e + jωε ′ jωμ ·(jωμ o (σ e + jωε ′ )) 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.5) H r =G TX cosθ(ωμ o ) 3 2 2π " j p j(σ e + jωε ′ ) 3 2 · 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.6) H θ = sinθ 4π G TX ωμ o " j s σ e + jωε ′ jωμ ·(jωμ o (σ e + jωε ′ )) 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.7) H θ =G TX sinθ(ωμ o ) 3 2 4π " j p j(σ e + jωε ′ ) 3 2 · 1 γr + 1 γ 2 r 2 + 1 γ 3 r 3 e −γr # (4.8) Based on the above expressions, an asymptotic analysis of MI systems in soils were CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 144 performedinChapter1resultinginthisusefulexpression: H MI = N TX ·S TX ·I TX ·e −αr 2πr 3 (4.9) whichisonlyvalidwhenθ=0 o and (γr) −2 ≪(γr) −3 . By using (4.9), one can verify that the attenuation given by the term e −αr is very small for different values ofα, that is, for different soil conditions, provided that relatively small distances r and frequencies f are considered. This fact may explain why sometimes one find at the MI literature the misleading concept that the MI method is not impacted by the characteristics of a lossy medium such as soil. To better understand these aspects, consider someexampleswithα valuesempiricallydeterminedattheNewHoganLake(SpringsVal- ley, CA), listed in Table 3.1. Observe that for low frequencies (1, 10 kHz) and distance r≤30m, the MI solution is indeed very resilient to drastic soil moisture variations which is not the case for 100 kHz. Nonetheless, f=100 kHz can still be employed for mid-range MI-WUSNswhilethesoilisdry,thusincreasingthebandwidthoftheWUSNapplications. ThispreliminaryanalysiswhichwasperformedinChapter1partiallyanswerstheques- tion behind the title of this chapter: audio frequencies are the proper option for the com- bination mid-range distances and high soil moisture. This novel observation related to the resilience of the MI wireless channel under specific circumstances has a dramatic impact on the way WUSNs may be designed. For instance, MI-WUSNs can employ a dynamic frequency selection scheme to better adapt the solution to the soil conditions. Nonetheless, theexpression(4.9). Hence,theChapter1completestheMIsignalattenuationmodelbyre- turningtotheexactexpressionsforthemagneticfieldandalsoincludingTX/RXresonance andothercircuitaspects. 4.2 InducedVoltageattheRXCoil The following expression were also derived in Chapter 1 and are repeated here for conve- nience. When TX and RX coils are perfectly aligned, the reception angle φ=0 o and the angleθ=0 o . For the following analysis we assume that both coils are properly aligned (or, alternatively, triaxial coils are employed). This is a realistic assumption provided that dur- ing the deployment of MI coils they are adjusted for maximum signal transfer. Hence, an expressionfortheinducedvoltagecanbederived: CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 145 |V ind |=(N RX ·S RX )·|H| MI ·μ o (4.10) where the|V ind | is the magnitude of the induced voltage across the inductor (given in volts), not necessarily at the input of the RX pre-amplifier (PA) which we call V C . Fortu- nately,itispossibletoachieveavoltagegain(orattenuation,ifcareisnottaken)V C /|V ind | by means ofresonance. Because the techniques studied in this thesis do not consider phase deviationattheRXside,fromnowonthetermV ind willbeusedreferringtothemagnitude valuegivenby(4.10). 4.3 ResonanceattheRXSide WithoutassumingaperfectLC-tankresonanceattheRXside,anexpressionforthevoltage acrosstheterminalsoftheRXinputamplifiercanbederived: G RX , 1 1+ jωCR ef f −ω 2 LC (4.11) V C =V ind ·G RX (4.12) whereV ind isgivenin(4.10). Atresonance,ifthevaluesofLandCareexactlythedesignones,thereactiveimpedance iszeroandG RX in(4.12)reachesitsmaximumtheoreticalvalue: G max RX = 1 2π· f·R ef f C = 2π· f·L R ef f (4.13) 4.4 MISignalAttenuationModelforWUSNs ThepreviousexpressionsforTXandRXsidescannowbegrouped. Tothisend,weadopted a decoupled TX/RX circuit approach. Typically, the error due to this approximation is in- deed very small as discussed in detail in Chapter 1. More specifically, such error is poten- tially more than one order of magnitude smaller compared to the error due to the deviation ofthecapacitancevalueoftheLCtank. Thelattererrorisindeedexpectedatleastduetothe soil temperature changes even when a very high-grade capacitor is employed. Therefore, CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 146 consideringθ =0,theChapter1concludedwiththefollowingMIsoilattenuationmodel: G full,max RX ,N RX ·S RX ·μ o · 2π· f·L R ef f (4.14) G TX ,N TX ·S TX ·I TX (4.15) V max C =G full,max RX ·G TX · (ωμ o ) 3 2 2π ·ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (4.16) 4.5 OperatingFrequencySelectionStrategy Observe how practical the expression (4.16) is: the first term contains RX circuit aspects, the second term contains TX circuit aspects, and the last term is related to the soil path at- tenuationandthedistancer. Therefore,oncemaximumvaluesforthecircuitparametersare established for the TX/RX circuits, such as coil diameter, coil wiring thickness and num- ber of turns, maximum TX current, RX sensitivity (minimumV c ), etc., the proper selection of f is dictated by the evaluation of the soil conditions for a certain target distance r. By plugging these values in (4.16) for different values of f, the accurate answer regarding the operating frequency range for MI-WUSNs is finally achieved. This process is illustrated in astudy-caseinTable4.1forsiltyloamsoiland30cm-diameterTX/RXcoils. Before starting the analysis of the simulated results in Table 4.1, it is important to re- member that the usual term LC tank used in this work actually refers to LCR tank because theeffectiveresistance(RorR ef f )ofthecoilisalsoalwayspresent. ThevalueofR ef f com- priseswiringaspectsofcoils,suchasDCresistivity,skindepthandproximityeffects;R ef f expressions can be found elsewhere [97, 98]. The hidden (and strong) effects of R ef f will be considered in Chapter 6. These effects are particularly important if we desire to design aMIsystemwhichisrobustandefficient(energyandbandwidth)forbothdryandwetsoil cases. CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 147 Table4.1Study-case: siltyloamsoil,targetV c =100μV,and30cm-diameterair-coredcoils. # Freq. f Water r AWG Nturns R ef f [97] I TX V DC TXpower G RX V c μV Comments 1 10kHz 15% 15m 26 56 7.1Ω 75mA 1.5V 40mW 24.5 100.1 Feasiblelow-powersolution 2 10kHz 40% " " " " 92mA 1.8V 61mW 24.5 100.8 Feasiblelow-powerandrobustsolution 3 26kHz 15% " " " " 12mA 0.3V 1mW 63.6 100.1 Moreenergy-efficientsolutionthan#1 4 26kHz 40% " " " " 21mA 0.4V 3mW 63.6 104.0 Moreenergy-efficientsolutionthan#2 5 10kHz 15% 30m " " " 650mA 12.9V 3000mW 24.5 100.8 NOTalow-powersolution 6 10kHz 40% " " " " 1.02A 20.3V 7387mW 24.5 100.0 NOTalow-powersolution 7 26kHz 15% " " " " 115mA 2.3V 93.9mW 63.6 100.9 Feasiblesolution 8 26kHz 40% " " " " 505mA 10.1V 1811mW 63.6 100.5 NOTalow-powersolution 9 26kHz 40% 30m 26 100 12.6Ω 93mA 3.3V 109.0mW 108 100.3 Veryenergy-efficientandrobustsolution 10 150kHz 5% " 34 " 80.7Ω 6mA 1.4V 2.9mW 119.1 103.7 Higherbandwidth+low-powersolution CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 148 AsshowninTable4.1,thecases1-4clearlyhighlightanovelandveryimportantaspect of MI-based communication systems: a lower frequency may not necessarily provide the best solution. For instance, for a wet soil (cases 2 and 4) and the specific scenario under analysis, a frequency of 26 kHz is much more energy-efficient and 10 kHz. The reason for this result lies on the value of G RX which is almost 2.6 folds higher compared to the 10 kHz case. One can directly check the explanation by visually inspecting the equation (4.14): suchrateofincreaseisalsotherateof26kHzand10kHzifweassumethatR ef f is basicallythesameforbothfrequencies. Infact,thisassumptionholdsforthewirethickness (i.e.,AWGatthetable)andnumberofturnsconsideredincases2and4. However,assoon as f increases even more, the mentioned assumption may not hold anymore and R ef f may stronglyincrease(nonlinearly)inawaythatforcesG RX todecrease. Besides the effects of R ef f , there are also the effects of the resonance detuning, that is, the actual decrease of G RX compared to its design value due to the deviation of the res- onance frequency. As already explained, such deviation can be potentially caused by the effect of soil temperature variations on the nominal capacitance value of C, the resonance capacitor at the RX side. To better understand this issue, consider again the case 4 but now assumethatforagivenfrequency f higherthan26kHzwecanachieve200asthevaluefor G RX . Annaivedecisionwouldbequicklyemploythisnewfrequencybecauseitapparently provides higher level of efficiency in terms of energy (and also bandwidth in some cases). Unfortunately,thisdesignmayleadtoveryunstable(nonreliable)communicationsystems. This happens when we assume a non-realistic value of G RX taking for granted that L andC values do not change with soil conditions. One practical example of this potential design trapwasgiveninChapter1. Therefore,itisimportanttorememberthatthemodelerrorcanberealisticallyestimated only if the empirically-determined effects of the resonance detuning are also considered. Basically, there 3 ways to minimize the impact of the mentioned detuning. First, the use of capacitors and coils that are very resilient to the environmental conditions. Second, we canestablishupperlimitsforG RX duringthedesignphasethusincreasingthebandwidthof theLCtanks. Inotherwords,withalargerbandwidth,smalldeviationsoftheresonantfre- quencymaynotimpacttheactualvalueofG RX . Thelastwaytodealthedetuningchallenge istheonethatleadstothebestenergyresultsbutwiththecostofamuchhighercomplexity: theauto-tuningfeature. Forthiscase,thevalueofCisautomaticallyelectronically-adjusted. While this solution may be properly employed for point-to-point communication regarding CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 149 apairofMInodes,itcanbeverycomplexwhenmultipleMInodesareinvolved. Asarule-of-thumb,considerthatthehigherthedesignvalueofG RX is,thehigheristhe energy efficiency and smaller is the application bandwidth. Basically, we can increase G RX by simply increasing the frequency. However, this action is eventually constrained by the valueofR ef f whichalsoincreaseswithfrequency. Returningtothecases1-4inTable4.1,observethatwithoutchanging f,butonlychang- ingI TX ,wecanstillhaveanenergy-efficientsolutionforbothdryandwetsoils. Thisresult isachievedincase2(comparetocase1)andcase4(comparetocase3). Now,let’sgiveour attention to cases 5-8. The scenario is the same regarding cases 1-4 with the exception of thedistancer whichincreasesfrom15to30m. Forsuchcases,itisnotanymorepossibleto haveasolutionwithlessthan100mW(providedthatthedimensionsofthecoilsdonotin- crease). Thebestenergy-efficientsolution,case7,isstillrelativelypower-hungry: 260mW. Thisexamplesgivesapreliminaryindicationthatitmaynotbepossibletohavelow-power MIsolutionsinsoilfordistancesmuchlongerthanthemid-rangeonesdefinedinthiswork (i.e.,around15to30m). However,in[53],alow-powersolutioninvolvingpassiveMIrelays is proposed to extend the communication range. Nonetheless, practical demonstrations of suchsystems(calledMI-waveguides)involvingsoilmediumandmid-rangedistancessofar havenotbeenprovided. By comparing cases 5-8, it is possible to verify that a dynamic frequency adaptation alone cannot solve the energy and robustness challenges in these cases. In other words, when the soil conditions vary drastically, the dynamic change of the operating frequency is still a necessary condition, but not a sufficient one, in order to achieve both robustness and energy-efficiency. Forinstance,considerhowthepowerissueofthecase8issolvedincase 9. Ratherthandrasticallychanging f orI TX ,agoodsolutioncanstillbeobtainedbysimply increasing the number of wiring turns N: from 56 (case 8) to 100 (case 9). As expected, suchN increasewillalsoincreaseR ef f which,inasuperficialinvestigationof(4.14),would decrease G RX . However, this does not occur in case 9 because the N increase leads to the increase of L in such way that the increase of R ef f has a smaller impact causing G RX to have a slight increase. At the TX side, G TX strongly increases due to the increase of N, as evident by visual inspection of equation (4.15). For case 9, both increases of G RX and G TX are the reason why this case is more energy-efficient than case 8 by more than one order of magnitude. CHAPTER4. OPERATINGFREQUENCYRANGEFORMI-WUSNS 150 Finally, we analyze case 10 which is one for dry soil conditions. In this case, f can be even higher than 100 kHz assuming that a thinner wiring for the coils (i.e., higher AWG). If the wire thickness of case 8 is preserved (26 AWG), the R ef f would strongly increase mainly due to the proximity effect [97]. If only the skin depth effect is considered here, the effective value R ef f would be much smaller and the proposed solution of adopting a single wire thickness (e.g., 26 AWG) would work. However, this is not the case as we will see in Chapters 5 and 6. This analysis leads to the novel conclusion that MI-WUSN systems potentially need to employ 2 wire gauges (different AWGs) at the same TX/RX coil (i.e., one for dry soil operation and other for wet soil) in order to efficiently adapt the operating frequencyaccordingtothesoilconditions. In this chapter, we have concluded that audio frequencies are the best option for mid- range MI-WUSNs if we consider that the solution must be robust enough for both regular and very wet conditions. This conclusion justifies the study of a sub-MHz soil dielectric model as discussed in Chapters 1 and 2. The MI communication in underground settings is not necessarily constrained to audio frequencies and higher values around 200kHz can still be employed if the system dynamically is able to adapt itself for dry soil conditions. Besidesthefrequencyswitching,inordertoachieveahigherenergyefficientthementioned adaptationalsorequiresthechangeoftheothersystemsparameters,suchasthetransmitting current,numberofturnsN,andwirethickness. Thelatteroftheseparametersisparticularly difficult to be taken into account in a dynamic system because it is fixed by design. This challenge is addressed at the Chapter 6. However, before starting this discussion, it is also necessarytodevelopamethodologythathelpsustofindtheoptimumdesignparametersfor theMInodesgiventheelectricalpropertiesofthesoilandacertainsetofdesignconstraints andthisisthetopicofthenextchapter. Chapter5 FrequencyAdaptationForExtremeSoil Conditions This chapter is mainly based on one of our recent papers [100]. At the previous chapter, it was shown by numerical examples based on the MI Signal Attenuation model (this one was proposed in Chapter 1) that the MI channel in soil medium is basically constrained to operational frequencies at the 1kHz-200kHz range considering a) mid-range distances (i.e.,15..30m),b)differentkindsofsoilcompositionandmoisturelevels,c)relativelysmall physical volumes for the MI nodes, and d) energy-efficient MI devices that are expected to operateduringmultipleyearswithoutbatteryexchange[99]. ThementionedMI-WUSNfrequencyrangefindingisinagreementwithoneofthefew empirical MI works tailored to soil/rock medium and mid-range distances [50]. In that work, it is reported that audio frequencies are a feasible option in contrast with frequencies higherthan1MHz. Asaresult,thefirstgenerationofMInodesisexpectedtobetailoredto relatively low data-rate applications due to the low operational frequency, at least from the peer-to-peercommunicationperspective(collaborationinMI-WUSNscanstillimprovethe application performance). Moreover, multichannel and special modulation techniques can beinvestigatedinordertoincreasethedatathroughputofMI-WUSNs. Thischapterisorganizedasfollows: apracticalMI-signalattenuationmodelwhichwas proposed and validated (at 10 kHz) in Chapter 1 and in [99] is discussed in Section 5.1. However, for this time, additional expressions for self-inductance L and R ef f are included in the attenuation model. Next, the optimization problem of selecting the proper opera- tional frequency according to a given set of parameters is discussed and a novel algorithm 151 CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 152 c b c a b=c b: axial length (cross-section) a: mean radius of the turns c: radial thickness (cross-section) square cross-section Fig. 5.1 Scenario investigated in this chapter: practical and optimized performance aspects targetinglow-powerandmid-range(15..30m)MI-WUSNs. is proposed in Section 5.2. Simulated results for diverse scenarios are then presented in Section 5.3 and it is concluded that the use of a single lower-bound frequency FL aiming the worst-case scenario has several penalties in terms of bandwidth and power-efficiency. Accordingly, in Section 5.4, it is recommended the use of an upper-bound frequency FH whenever the medium conditions are very favorable. The determination of FL and FH occurs at the design phase and the frequency-switching decisions occur while the system is operating. Unfortunately, for the majority of the cases in mid-range MI-WUSNs, if the same coil wiring is used for the frequencies FL and FH, the solution is far from optimum. Based on this observation, that section is concluded with a novel strategic design approach which employs two distinct coil wirings hosted at the same physical enclosure: one for FL operationandtheotherfor FH, resultinginaproperbalanceintermsofrobustness/power- efficiency/high-bandwidth. 5.1 PracticalMI-SignalAttenuationModel Inthissection,ourMI-signalattenuationmodelproposedin[99]isintroduced. Thismodel is considered practical because it does not employ the mutual inductance parameter M typically found in the WUSN-MI literature. This is possible because, for some loosely inductive-coupledsystems,theimpedanceofonecircuitisnotsignificantlyimpactedbythe impedance of the other circuit (and vice-versa). The model error due to this approach is verysmall(e.g.,<0.1%)forthedistanceandfrequencyrangesevaluatedinthisstudy. A set of additional practical assumptions for the MI node design are considered in this chapter,mainlytargetingtheapplicationsdiscussedinChapter9. Nonetheless,withoutloss ofgenerality,themajorityoftheconclusionsinthischaptercanbealsoappliedtodifferent scenarios, such as for MI communication in collapsed buildings [24]. These assumptions CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 153 arelistedbelow: FNon-magneticsoilmedium: typicalreal-worldscenario. FTransmitting(TX)andReceiving(RX)coils: identicaldesign;symmetricbidirectional communicationisdesired. FAir-coredcoils: baselinecaseforMI-WUSNevaluations. FMultilayercircularcoils,squarecross-section: asshowninFig. 5.1speciallytargeting mid-rangeMI-WUSNs[97,99]. FResonance: optimum performance for communication purpose is achieved with the res- onance configuration shown in Fig. 5.1; it is initially assumed that the system is operating preciselyattheresonantfrequencyselectedbythedynamicfrequency-switchingalgorithm presentedinthischapter. FImpedance: ideally,lowoutputimpedancefortheTXsignalsource(e.g.,<20Ω,typical) andhighinputimpedancefortheRXpre-amplifier(e.g.,>1MΩ). F Deployment: axis of the coils in horizontal position as shown in Fig. 5.1; stronger RX signalforsingle-axiscoils[99]. F TX and RX perfectly aligned: our preliminary empirical investigation showed that for mid-range distances, misalignment errors smaller than 3% in relation to the inter-node dis- tancer arenotsignificant. FEffectiveresistanceofthecoilR ef f : theexpressionforR ef f inthischapteronlyconsid- ers the DC conductivity and the skin depth effects of the coil wiring. The proximity effect is neglected and this approach is only valid when the wire diameter is smaller than 4 times the skin depth [97], otherwise the additional resistance due to the proximity effect must be calculated or measured. Also, the value of R ef f can be smaller than the one provided in this work if Litz wires are used causing drastic reduction of the skin depth effect. In this case, the adjusted value of R ef f can be calculated considering the number of strands (the insulatedconductorsthataretwistedformingasinglestrandedwire)[97]. CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 154 The complex propagation constant γ considered in this chapter for the soil medium is givenby[18]: γ = p jωμ(σ e + jωε ′ ) (5.1) σ e =ωε ′ tanδ e (5.2) where f istheoperatingfrequency,ω istheangularfrequency(=2π· f),μ isthemagnetic permeability of the medium, ε ′ is the real part of the complex permittivity of the medium (i.e.,soil),σ e istheeffectiveconductivityofthemedium,andδ e istheeffectivelosstangent. The parametersε ′ andδ e are particularly adopted because these are the ones derived from soil measurements when impedance analyzers are employed. In this chapter, we will use valuesforγ derivedfromthelow-frequencysoildielectricempiricalmodelin[16]andalso fromourempirically-determinedsoilmodel[99]basedon9differentsoiltypesfordifferent frequenciesandvolumetricwatercontent(VWC)levels. Considering the previous assumptions, an expression can be derived for the voltageV C acrossthecapacitorC,thatis,attheinputoftheRXpre-amplifier(PA),asgivenbelow[99]. Here, the expression for the self-inductance L of circular coils with square cross-section is the Grover’s formula (adapted from [98], pp.95) which we empirically verified as having less than 1% error for all MI-coils we designed so far. All units are SI, if not explicitly stated: CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 155 V C =G TX ·G RX · (ωμ o ) 3 2 2π ·ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (5.3) G TX ,N TX ·S TX ·I TX (5.4) G RX ,N RX ·S RX ·μ o · ω·L R ef f (5.5) L=4π·10 −7 ·a·N 2 h 1 2 1+ 1 6 c 2a 2 ·ln 8 ( c 2a ) 2 −0.84834+0.2041 c 2a 2 i (5.6) (forφ wire <δ wire ) R ef f = ρ wire ·l wire π·δ wire ·(φ wire −δ wire ) (forφ wire ≥δ wire ) R ef f = ρ wire ·l wire π·( φ wire 2 ) 2 (5.7) l wire =2π·a·N (5.8) δ wire = r ρ wire π·μ o · f (5.9) S=S TX =S RX =π·a 2 (5.10) c=(1+isol fact )·φ wire · l √ N m ,N=N TX =N RX (5.11) where r is the inter-node distance and G TX /G RX are the maximum theoretical gains regard- ing TX/RX circuit aspects. N TX and N RX refer to the number of turns of the TX/RX coils and,becauseweareassumingidenticaldesigns,N TX =N RX =N inoursimulations. Similarly, S TX =S RX =S refer to the surface area of the TX/RX coils. I TX is the rms value of the sinu- soidal current passing through the TX coil and μ o is the permeability of vacuum. L is the self-inductance of the TX/RX coils,φ wire is the diameter of the wire, and R ef f is the effec- tiveresistanceofthecoils. a,carecoilphysicaldimensionsshowninFig. 5.1. isol fact isthe isolation factor which varies from 0 to 1 and corresponds to the thickness of the isolation layer of the wire. In our simulations, we used annealed copper wires (ρ wire =1.724·10 −8 Ω·m)andisol fact =5%(20%forLitzwires). The sensitivity of the RX side in relation to the input voltageV C is mainly a function of the noise level at the input stage of the RX PA and also the internal noise figure of this am- plifier. For instance, in outdoor experiments at 10 kHz, our designed RX MI node showed goodperformance(e.g.,packetreceptionrate>90%)whenV C wasatleast100μVwhilethe totalinputnoisewassmallerthan25μV(i.e.,SNR=12dB).Weobservedthatthenoiselevel for the MI signal in ground medium was very low and stable, similar to the behavior of the CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 156 underground radio channel [10]. Therefore, the RX sensitivity can be potentially enhanced by a careful design involving very-low signal amplifiers. To this end, we used commercial InstrumentationAmplifiers(IAs)thatarepotentiallyaproperchoiceforthefirstgeneration of real-world MI nodes. In our simulations, we establishV min C =100μV as the limit for suc- cessfulcommunication. We turn our attention to the expression (5.3) which is basically comprised of 3 blocks: tworelatedtoTX/RXcircuitsaspectsandtherightmostonewhichisafunctionoffrequency f, distance r, and the electrical properties of the soil medium (e.g.,γ,ε ′ , andσ e ). G TX and G RX are not significantly impacted by the soil conditions 1 and all 3 blocks are functions of f. Although the expression for G TX does not directly show this fact, an increase of the magnitude of the voltage source (V src ) is not a linear function of f mainly due to the effect ofR ef f attheTXsidewhichhasasignificantandnon-linearincreasewhen f increases. For the G RX expression, an increase of f does not necessarily increases G RX also due to R ef f valuedispersion. Finally,forthelastexpressioninvolvingarealpart,itisnotwisetomake any kind of conclusion regarding its value associated to the variation of f. This complex expressionisalsohighlydependentonthevaluesofthedistancer andthepropagationcon- stantγ and the latter depends on the soil electrical properties that non-linearly vary with f [2]. 1 We empirically verified small variations (i.e. <1%) of R ef f and L values when a coil is in free-space or immersed in dry/wet soils even for the frequencies smaller than 100 kHz. These small errors can be safely neglected if the TX/RX circuits are capable of fine-tuning their resonance frequencies accordingly. However, high values of G RX may imply a very high value of the quality factor Q of the LC tank and such resonance adjustment becomes critical in practice. On the other hand, by having G RX not very high, a higher bandwidth can be achieved and relatively small variations of the values of L and C may not be source of issues or significantmodelerrors. CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 157 5.2 OptimizationProblemStatement Initsmostreducedform,theoptimizationproblemwewouldliketoinvestigatehas9input parameters: f, r, a (the average coil diameter is 2a), I TX , N, φ wire , ρ wire , ε ′ , and σ e . Note thatthesoilconditions,suchasthevolumetricwatercontent(VWC)level,areembeddedat the two latter parameters and, similarly, all other expressions in this chapter are eventually derived from this 9-set of parameters. Moreover, there is a list of hard constraints (to be discussedsoon)thatinvolves7additionalengineering-relevantparameters- I max TX , R min ,etc. Therefore, the optimization problem in this section involves 16 parameters. Due to its very highcomplexity,weoptedtofixsomeoftheseparametersinordertoachievepracticalcase analysis for different scenarios. The default values for some of the fixed parameters are shownbelow: FixedparametersFSET fortheoverallstudy: F r: specific inter-node distance (e.g., 30m). The results of the optimization process do not follow linearly with r because the underlying equations also do not follow. In fact, one MIsystemdesignedtooperateproperlyat30mmaynotoperatefordifferentsmaller/larger distances. This is remarkably different in comparison with typical over-the-air radio sys- tems. AsageneralbehavioroftheMIsystem,thesmalleristhedistancer,thehigheristhe frequencybandthatcanemployed. Fρ wire : copperwireresistivity(e.g.,1.724·10 −8 Ω·m). Fε ′ ,σ e : dielectric constant and effective conductivity of the medium, respectively. These valuesareduetoactualsoilmeasurementsorprovidedbyanexistingsoildielectricmodel. For the determination of FL and FH, these values are defined for the worst-case (wet soil) and best-case (dry soil) scenarios, respectively. We anticipate that a system at FL properly operatesunderawiderangeofε ′ andσ e values. F a: average coil radius, as shown in Fig. 5.1 (e.g., 15cm). The surface area of the coil (S=2π·a 2 ) is a key-parameter in the sense that it increases simultaneously G TX and G RX . Preliminary simulations showed that the parameter a has so strong impact on the power- efficiency of the solution that it must be selected with its maximum allowable value. This choicemustconsidermanufacturingandinstallationcostsoftheMInodeduetoitsphysical volume. CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 158 FI max TX : max. outputTXcurrent(rms)(e.g.,300mA). FR min : min. allowedresistancefortheTXload(e.g.,3Ω). FC min : min. capacitance of the resonant capacitorC which must be much larger than the parasitic capacitance of the TX/RX coil. Based on our experiments, 20pF is a conservative valuefortypicalMI-WUSNcoils. FV max src : max. DC voltage level of the TX power supply (e.g., 6V). This parameter in con- junction with I max TX limit the maximum allowable value of R ef f even if the number of turns N isnotconstrained. FV min C : min. RXvoltage-levelsensitivity(e.g.,100μV).Becausetheinputimpedanceofa RX MI circuit has high impedance (not a fixed low-value, such as 50Ω), we will not define theRXsensitivityintermsofdBm,whichistypicallythecaseforradioreceivers. F BW min : min. bandwidth supported by the application (e.g., 100 Hz). This parameter is typically very small in order to allow a minimum level of functionality at worst soil con- ditions. We anticipate that for mid-range low-power MI-WUSNs, the bandwidth is usually smallerthan2kHz. F MIRAD min : min. MI-to-RAD induced voltages ratio (e.g., 3dB). Observe that the equa- tion (5.3) has terms involving γ and r that have powers 2 and 3. The former is associated to the radiative (RAD) part of the signal and the latter with the inductive (MI) part. While the MI part is the signal of interest, the RAD part is not in-phase with MI and it can poten- tially distort the MI signal (similar to a non-coherent noise source). The MIRAD min value can be empirically-determined considering the capabilities of the RX decoding circuit, the existenceofinputfiltersand/orsignalprocessingtechniques. Fixedparameterforanindividualcase: f Each convergent simulation for a frequency-candidate f provides a local-minimum value for the power function (to be defined soon). Both FL and FH are simply selected by ana- lyzingallindividualcasesof f andchoosingtheonewhichrepresentstheglobal-minimum CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 159 value of power. Besides the power option, it is also possible to implement bandwidth and signal-level functions that have the same behavior as the power function but with different optimizationgoals. Variableparameters: φ wire ,I TX ,andN The proposed methodology for the optimization problem is eventually reduced to the anal- ysis of just 3 parameters for each individual case f. As a result, the overall investigation is also significantly simplified. Note that the coil’s radius a is assumed to be fixed. However, in our first investigations, we considered a as one of the variable parameters resulting in a very complex analysis. In that case, a very high number of distinct f-φ wire -I TX -N configu- rations were offered as excellent power solutions with order of μW. However, the majority ofthoseresultsdonothaveanypracticalsignificanceduetotheenormoussizes/weightsfor thecoiland/orverysmallbandwidths. The optimization process can be summarized as follows: with FSET, select an indi- vidual frequency f i from a frequency range-set, and vary the parameters φ wire , I TX ,and N in order to minimize the TX power function (the RX power consumption is very small compared to the TX operation in MI nodes). After evaluating all cases involving distinct frequencies f i , select the frequency (and its associated φ wire -I TX -N configuration) with the smallest power function value. The formal definition of the process is given by Algorithm 4 and the optimization problem for an individual frequency case is shown below. The cor- rectnessoftheAlgorithm4canbeeasilyverifiedbecauseitemploysabrute-forceapproach withoutanycomputational-optimizationeffort: min power i (FSET, f i ,I TX ,N,φ wire ) = I 2 TX /R ef f s.t. φ wire < 4·δwire I TX ≤ I max TX R ef f ≥ R min C = 1 L·ω 2 ≥ C min V src =2.82·I TX ·R ef f ≤ V max src V C ≥ V min C BW ≥ BW min MIRAD ≥ MIRAD min CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 160 Algorithm4Op. FrequencyDeterminationforMI-WUSNs Require: FSET fixedparameters Require: FREQ SET : frequencyrangeset(asc. order) Require: WIREDIA SET : wiringgaugeset(order: thickesttothinnest) Require: OPTIM SET : bestconvergentsolution f, power,etc. Ensure: Bestfrequencyintermsofpower-efficiencyforFSET OPTIM SET (power)←∞ S←calcEqn10(FSET) isol factor ←0.05(materialcharacteristics) N max ←300(user-defined,convergencecontrol) I min TX ←1mA(user-defined) indexF←length(FREQ SET ) indexW←length(WIREDIA SET ) whileindexF>0do f←FREQ SET (index) δ wire ←calcEqn9(FSET, f) whileindexD>0do ifφ wire <(4·δ wire )then I TX ←I min TX whileI TX ≤I max TX do N←1 whileN<=N max do G TX ←calcEqn4(N,FSET,I TX ) l wire ←calcEqn8(FSET,N) R ef f ←calcEqn7(FSET,l wire ,δ wire ,φ wire ) ifR ef f ≥R min then c←calcEqn11(isol factor ,φ wire ,N) L←calcEqn6(FSET,N,c) C← 1 L·(2πf) 2 ifC≥C min then V src ←2.82·R ef f ·I TX ifV src ≤V max src then G RX ←calcEqn5(N,FSET, f,L,R ef f ) MI←calcEqn3(G TX ..)(onlyMIpart) RAD←calcEqn3(G TX ..)(onlyRADpart) V C ←MI RADMI←20·log MI RAD (RADMIindB) ifV C ≥V min C then BW← R ef f (2π·L) ; ifBW≥BW min then ifRADMI≥RADMI min then power←I 2 TX ·R ef f if power <OPTIM SET (power)then saveConfigParameters(OPTIM SET ) endif endif endif endif endif endif endif N←N+1 endwhile I TX ←increment(I TX ) endwhile endif indexD←indexD−1 endwhile indexF←indexF−1 endwhile ifOPTIM SET (power)6=∞then FL←OPTIM SET (f) else FL←0 endif CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 161 3 KHz 4 KHz 5 KHz 6 KHz 7 KHz 8 KHz 9 KHz 0 20 40 60 80 100 120 140 160 Frequency TX Power (mW) and Comm. Bandwidth (Hz) TX Power, 30cm diameter coil Communication Bandwidth, 30cm diameter coil TX Power, 40cm diameter coil Communication Bandwidth, 40cm diameter coil FL Selection Fig.5.2Selectingthebestfrequency/configurationforFL(worstscenario). 5.3 PreliminaryResultsandDiscussion Based on the signal attenuation model introduced in Section 5.1 and the Algorithm 4 pre- sentedattheprevioussection,simulationsareperformed. Defaultvaluesfortheparameters are considered, as well distance r=30m and φ wire (coil diameter) = 30cm and 40cm. Four soil moisture levels (VWC) are evaluated: 1, 5, 15, and 40%. The ultimate goal is to de- termine what is the best frequency/configuration for the worst case (40%) and compare the impact of adopting this single-frequency design for better soil conditions (i.e., VWC=1, 5, and15%). TheanalysisstartswithFLinordertocheckthefeasibilityofthesolutiongiven theavailabledesignconstraints. TheAlgorithm4isusedwithoutmodification,thatis,opti- mized for the smallest TX power-consumption configuration. The power consumption and bandwidth curves for VWC=40% are shown in Fig. 5.2 where 13 distinct configurations are represented. For each configuration, a certain set of values for f, N, I TX , andφ wire (re- lated to the wiring gauge, in AWG) is associated. Note that there are theoretical frequency boundaries for both 30 and 40-cm diameter coils. For instance, a 30-cm MI system facing soil moisture of 40% can only operate at the 4-9kHz. For 40-cm coils, this range changes to 3-9kHz. It is very important to highlight that these frequency boundaries are specifically tied to this scenario, including the assumed RX sensitivity of V min C =100μV at r=30m. An CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 162 infinitenumberoffiguressimilartoFig. 5.2exists,dependingontheinputparameters. Such discussionillustratesthechallengesassociatedwiththedesignofMI-basedcommunication systemsincomparisonwithtraditionalradiolinks. As shown in Fig. 5.2, the bandwidth presents a trend to increase with the frequency. Similarly, for the power consumption curve, there is a trend to decrease its value with the increasingfrequencyandtoflataftercertainpoint. Keepinmindthatsuchnon-linearcurves arenotassociatedthefrequencyresponseofacertainsystem/configuration. Eachfrequency point actually is associated with a distinct solution in terms ofφ wire , N, and I TX . Our main contributioninthischaperistoshowthatsuchapproachleadstopracticalrobustandpower- efficient solutions. For instance, one can observe that the highlighted region around 8 kHz is potentially the best FL choice (for both 30 and 40cm coils) because it balances a higher bandwidth with a relatively smaller power consumption. It is recommended to avoid oper- ational frequencies close to any frequency boundary because drastic variations of the soil conditionscanstillleadtoasystemthatcannotoperate. Inthissense,7kHzisstillanother goodoption. Thecharacteristicsforthe40cmdiametercoilsystemoperatingat8kHzare: 2.7mWTX power,136Hzbandwidth,N=280,28AWG,I TX =6mA,4.8dBRAD/MIratio,253gweight, and1.3V DC battery. Forthe30cmsystematthesamefrequency: 17.3mWTXpower,139Hz bandwidth, N=298, 28 AWG, I TX =17mA, 4.8dB RAD/MI ratio, 202g weight, and 2.9V DC battery. Note that by using a 33.3% larger coil, 84% power saving is achieved. Nonethe- less, the overall results are impressive: if we disregard the very low bandwidth (roughly, 8 characterspersecond),andfocusonthepoweraspects,MIsystemsformid-rangedistances are potentially very competitive in relation to radio-based systems and much less prone to interference due to their physical communication bubbles (after a certain distance, the MI signaldrasticallydropsandmulti-pathandsimilareffectsarenegligible). Assume that the above selected configuration for a 30-cm coil operating at 8 kHz is usedfordifferentVWClevels,thatis,withoutadoptingafrequency-switchingscheme. We wanttoanalyzetheimpactofsuchdecisionand,tothisend,newsimulationsareperformed considering VWC 1, 5, and 15%. For each case, the most power-efficient configuration is found. ThecostsofusingthementionedFL=8kHzanditsassociatedconfigurationincom- parison of using tailored and efficient configurations for each VWC case are shown in Fig. 5.3. Forfairness,thevalueofI TX atthe8kHzconfigurationisdecreasedforeachcase(due CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 163 to smaller soil path attenuation) resulting in TX power values smaller than the mentioned 17.3mWforVWC=40%. However,beforediscussingtheresults,itisimportanttohighlight atechniquethatwasusedtoachieveabetterpowervs. bandwidthbalancewhenselectinga configuration. The problem with the Algorithm 4 is the fact that, for instance, it can trade 50% less bandwidth by 1% less power. Such non-balanced decision can potentially hide very good configurations. A technique that can be employed to mitigate this issue is to first employ the algorithm considering the power criteria, as usual, and hold the results. Then, re-run the same algorithm modified to bandwidth criteria (this modification at the Algorithm 4 is simply achieved by exchanging the world power by BW at the most internal if-decision). With both power and bandwidth results, it is possible to evaluate a balanced solution. We used this approach to select the best configurations for VWC 1, 5, and 15% shown in Fig. 5.3. The use of a single frequency/configuration designed for the worst-case scenario po- tentially limits the bandwidth, as shown in Fig. 5.3. In some cases, there is a power penalty, such as for 1 and 5% VWC scenarios shown in that figure. However, the main issue is the limited bandwidth: without the possibility of using higher frequencies, the bandwidth cannot increase even under better soil conditions. For instance, for 5% VWC, the achieved bandwidth is only 7.8% of the one achieved with a proper configuration and this achievement is also more power-efficient. Therefore, it is clear that an adaptive fre- quency/configurationschemeispotentiallydesirableformid-rangeMI-WUSNs. 5.4 FL-FHFrequency-SwitchingScheme One can propose that the mentioned adaptation challenge can be solved by changing the frequency without modifying the configuration of the coils/circuits, except by adjusting C to the new resonant frequency. Unfortunately, this scheme may not be a proper solution duetothecoilwiring: theoptimizationprocessselectsthebestwiringschemeandsomeof these parameters, particularly the wiring thickness φ wire (i.e., AWG#) cannot be modified by software once the MI node is deployed. Unfortunately, the choice of the AWG# for FL maybenotaproperonewhenhigherfrequenciesareconsidered. Thereare3mainsjustifi- cationsfor thisstatement: first, athickerwire usedat FL maynot pass theproximity effect CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 164 1% VWC 5% VWC 15% VWC 40% VWC 20% 40% 60% 80% 100% 120% 140% Soil volumetric water content (VWC) Variation of Power Consumption and Bandwidth 7.8% bandwidth 15.3% bandwidth 113% power 139% power 72% power Baseline 6.5% bandwidth Fig. 5.3 Energy and bandwidth penalties when using a fixed frequency (designed for the worst-case40%VWC)fordifferentsoilmoisturelevels. constraint (φ wire <4·δwire). Second, R ef f for the thicker wire can significantly increase at higher frequencies. Third, and the most important (and typically hidden) reason that justi- fiestheneedofdistinctconfigurationsforFL-FH operation: thedetuningeffect. Oneofthe assumptions we made is that TX and RX circuits are ideally tuned at the design resonant frequency. However,incaseofdetuning,theexpectedvalueforG RX givenby(5.5)mustbe adjustedasfollows: G RX =N RX ·S RX ·μ o ·ω· 1 1+ jωCR ef f −ω 2 LC (5.12) When f increases, the term ω 2 LC requires a drastic decrease of the C value in order to achieve perfect resonance (the term equals 1). Because we are trying to use the same FL configuration for a higher frequency FH, L has the same value and R ef f may increase moderatelyincomparisontothenecessaryvaluechangeforthenewCunderfrequencyFH. Unfortunately,aslightvariationoftheLorC valuecandrasticallydecreaseG RX . Thesame potential issue can also happen for our original configuration at FL, but we had used one constraint that significantly mitigated the detuning effect: we limited Q of the LC tank to notexceed60. Inthatconfiguration,Q=57.8;butforthesameconfigurationatFH=80kHz, Q=451withC=58pF.Withthissetup,avariationofaround1%attheC valuewillcausethe decrease of the RX signal by almost one order of magnitude and the solution is clearly not robust. CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 165 50 55 60 65 70 75 80 85 90 95 10 0 10 1 10 2 10 3 10 4 Frequency (kHz) TX Power Consumption (mW) and Bandwidth (Hz) Power, 30cm diameter coil, 5% VWC Bandwidth, 30cm diameter coil, 5% VWC FH Selection 28 AWG 29 AWG 30 AWG 31 AWG Fig.5.4Selectingthebestfrequency/configurationforFH (bestscenario). In general, for all cases we investigated so far, a proper solution requires 2 wirings, one forFLandotherforFH. Bothcoilscanbeenclosedtogether,oneinsidetheother. Usually, the coil with thicker wiring must be the external in order to not decreasing too much the values of the radius a for both coils. Such scheme requires an electronic switching circuit and a resonant capacitor C for each coil. The MI node can dynamically switch between the coils by employing a soil moisture sensor or, alternatively, by analyzing the communi- cation performance. It is important not to confuse the FL and FH coils with the TX/RX coils. In our design, we assumed that the same coil used for transmission is also used for reception. Therefore, in dry soil conditions, the FH coil will be used for both TX and RX roles. However,anotherelectronicswitchingschemeisnecessarywhenthehalf-duplexMI node switches between TX and RX. This switching is related to the coupling of the TX or RX circuits, each one with a distinct resonance configuration, as shown in Fig. 5.1. For Fig. 5.4,theAlgorithm4isusedwithBandwidthcriteriaandtargetingabest-casescenario with 5% VWC. The selected FH frequency is highlighted in the figure and it corresponds to 72kHz (73 is also a good option). The characteristics of this configuration are: 6.2mW TX power, 1774Hz bandwidth, N=274, 30 AWG, I TX =3mA, 5.6dB RAD/MI ratio, 117g weight,Q=40.6,and5.8V DC battery. A novel approach for the determination of the best operational frequencies for mid- CHAPTER5. FREQUENCYADAPTATIONFOREXTREMESOILCONDITIONS 166 range MI-WUSN systems is presented in this chapter. It is concluded that at least two independent wirings for the coil is necessary for optimal operation under a variety of soil conditions. This recommendation does not impact the soil irrigation and leak detection system novel architectures proposed in Chapter 9. A very important finding in this chapter isthefactthatminimumvariationsoftheLandCvaluesoftheMI-node’sresonanttankscan drastically impact the solution if the exact optimum configurations are applied. Therefore, it is recommended to avoid non-realistic high values of Q for the LC tanks in order to accommodate L and C value variations on the order of 1%. Empirical work considering higherfrequencies(FH)isalsonecessarytofullyvalidatetheproposedframework. Chapter6 Dual-WireSchemeAddedtoFrequency Adaptation At the previous two chapters, the sub-MHz range is identified as the adequate one for low- power/mid-range distance MI nodes, in particular audio frequencies when the soil is rela- tively wet [99]. A sub-MHz MI signal attenuation model is then proposed which has good agreement with preliminary outdoor experiments at 10 kHz. This model is the first one to employ a practical modular approach: with good accuracy, it is possible to separate trans- mitting (TX) and receiving (RX) circuit aspects from the attenuation due to the medium (e.g.,soil)electricalcharacteristics[99],asshownbyinspectingtheexpression(6.1): V C =G TX ·G RX · (ωμ o ) 3 2 2π ·ℜ ( j p j(σ e + jωε ′ ) 3 2 1 γ 2 r 2 + 1 γ 3 r 3 e −γr ) (6.1) G TX ,N TX ·S TX ·I TX (6.2) G RX ,N RX ·S RX ·μ o · ω·L R ef f (6.3) (φ wire <δ wire )R ef f = ρ wire ·l wire π·δ wire ·(φ wire −δ wire ) (6.4) (φ wire ≥δ wire )R ef f = ρ wire ·l wire π·( φ wire 2 ) 2 l wire =2π·a·N δ wire = r ρ wire π·μ o · f (6.5) BW(bandwidth)= R ef f 2πL (6.6) At the previous chapter and in [100], an important conclusion is reached: an energy- 167 CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 168 efficient mid-range MI system potentially requires an adaptive frequency scheme. This mechanism may not be necessary for small distances, such as 5m, but it is typically neces- saryforhigherdistancesinordertoachieveabalancedsolutioninvolvinga)robustnessand b) energy-efficient. This environmental-aware solution takes into consideration the typical variationsoftheelectricalpropertiesofsoils,forexampleduetorainordrydowneventand, accordingly,changesitsoperationalfrequency. Tothisend,adesignoptimizationalgorithm is proposed [100] in order to determine the best frequency and coil configuration for the worst (wet soil) and best (dry soil) cases. The analysis of the simulated results shows that although the idea of using two operating frequencies for dry and wet soil conditions seems to be a strategic one, the main challenge is still the fact that an important design parameter, the wire thickness of the MI node, is fixed and cannot be dynamically optimized once the MInodeisdeployed. 6.1 MI-CoilDesignOptimizationProblem In this chapter, the adoption of a dual-wire coil for MI-nodes is proposed and a dual-wire system is implemented and tested. The analysis of the results supports the proposed ap- proachandtheoverallMIdesignstrategyiseventuallypresented. Thechapterisorganized asfollows: atthefirstpart,theresearchproblemregardingtheoptimizationofaMI-WUSN design is identified and discussed. Next, preliminary empirical evaluations of the proposed scheme are discussed and a novel sub-MHz indoor testbed is employed (the details of this testbed are provided in Appendix A). Finally, the combination of the adaptive frequency mechanismwiththedual-wiresolutionisanalyzedandconcludingdesignremarksarepro- vided. Asalreadyhighlightedatpreviouschapters,theMI-Soilsignalattenuationmodelwhich is represented by (6.1) is comprised of two parts: a) the TX (G TX ) and RX circuit aspects (G RX ) and b) the medium path attenuation (the term with γ). We start the analysis by fo- cusing exclusively on the effects of the medium path. The associated simulated results are shown in Fig. 6.1, where 7 frequencies are investigated for free-space and two extreme volumetric water content (vwc) levels in soil. For distances r smaller than 5m, the medium typehasarelativesmallimpact. However,atr=10mthemediumimpactisverystrong. For instance, the magnitude of a 100kHz signal in soil is attenuated by more than 31dB when vwcchangesfrom1to40%(soilstypicallysaturatebefore40%). Forallcases,exceptfree- CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 169 5m 10m 15m 20m 25m 30m 32m 34m 40m 40 50 60 70 80 90 100 110 120 130 Inter node distance r Attenuation of the RX induced voltage (V ) (dB) PA C soil vwc 40% 100kHz 40% 75kHz 40% 50kHz 40% 25kHz 40% 10kHz free space 2MHz free space 100kHz soil vwc 1% 100kHz Fig.6.1MI-signalattenuationexclusivelyduetothemediumpath(referencedtor=1m). space and 1%-vwc soil, there are specific breakpoint distances (bd) that mark the critical distance where the signal quickly falls and the associated strong signal attenuation cannot besimplymitigatedbycircuit-relatedimprovements(again,itisassumedinthisdiscussion theuseoflow-powerdevices,e.g.,<500mW). Observe that by fixing the soil moisture value (e.g., 40-vwc in Fig. 6.1), the bd dis- tance becomes smaller as the frequency increases. For instance, if r=20m is the target distance, only frequencies smaller than 25kHz are energy-efficient options. On the other hand, when the soil is very dry (e.g., 1%-vwc), frequencies even higher than 100 kHz can be employed. Note that, besides a potential higher application bandwidth, the adoption of ahigherfrequencymayalsoincreasetheenergy-efficiencyoftheMIsolutionprovidedthat thefrequencyisnotinaprohibitiverangeasjustdiscussed. RegardingtheG TX andG RX termsin(6.1),itismoredifficulttohaveananalysissimilar totheoneshowninFig. 6.1duetothenumberofvariables. Therefore,assumethatN andS are fixed and we would like to investigate how G TX varies with f. When f increases, R ef f also increases according to (6.5) and (6.4a). If the TX circuit does not have an automatic gaincontrol,thecurrentI TX willdecreaseleadingtoaG TX decrease. Inshort,if f increases, potentiallyG TX decreases. Next, we perform the same analysis for G RX : when f increases, G RX potentially in- CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 170 creases due to ω in (6.3). Nonetheless, the R ef f term in (6.3) plays an inverse role in relation to ω: it reduces G RX . This is not necessarily a negative aspect because the band- widthBW increaseswhenR ef f increasesaccordingto(6.6)). Infact,thisobservationisone the most important one regarding MI-WUSN designs: the trade-off energy-efficiency and applicationbandwidthisdirectlycontrolledbythevalueofR ef f . Basedonthispreliminary analysis, one cannot expect a monotonic function of the induced voltage V C regarding the frequency because two conflicting frequency-dependent factors are impacting the value of V C . Maxima and minima sets are indeed expected for a wide valuation of the input param- eters. Accordingly, in Chapter 5 and also in [100] an algorithm is proposed to get the local V C maximum points considering design and deployment constraints. The results of such evaluation depend on the value of the frequency which being selected, FL or FL, but is apparent from Fig. 5.2 that the important R ef f is strongly impacted by the wire thickness (i.e.,#AWG). Again, a key parameter to be closely observed is R ef f . For instance, when we need to reduce the energy of the solution, the typical design effort is to find a way to reduce R ef f . However, when one tries to drastically reduce R ef f without changing N, the typical option is significantly increase the wire thickness φ wire . However, this approach can lead to unacceptable very small values for BW. For instance, a relatively higher power mW- solution with a BW of 1000Hz may be more desirable than a muW one with BW of 50Hz. On the other hand, a 10mW solution at 15% soil moisture level may drastically increase to hundreds of mW when the soil is saturated. In fact, this is the scenario where two coils with differentwire thicknessesmay represent a better balanced solution. Morespecifically, a robust MI-based solution which continuously operates no matter the soil conditions may requireaswitchingscheme,notonlyfortheoperatingfrequencyasproposedattheprevious chapter[100],butalsoforφ wire ,thatis,byusingtwocoilswithdistinctvaluesforφ wire : one coil for a lower frequency - FL - used at wet soil conditions and another one for a higher frequency - FH - used at dry soils. In this chapter, we want to investigate the dual-wire approach: itsfeasibility,advantages,andscenarioswhereitcanbestrategicallyadopted. 6.2 PreliminaryEmpiricalEvaluations In order to validate the feasibility of the dual-wire solution, an indoor MI-soil testbed is constructed, as discussed at Appendix 9.2. Part of the testbed is shown at Fig. 6.2. Some ofthesimulatedresultsinthischapterwerealsoverifiedbyempiricalevaluationsusingthe CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 171 5cm-diameter PVC pipe 0.5m 1m 1.5m 2m 0.9m RX RX RX RX TX 0.15m 0.16m 2.1m FL coil 10 kHz FH coil 26 kHz 28 AWG N=121 36 AWG N=80 Air Twinax connectors Rotation angle mark 2cm 4.9cm 1cm 4.9cm Soil Air 0m Fig.6.2IndoorSub-MHzMI-Soiltestbedusedfortheempiricalevaluationsinthiswork. mentionedindoortestbed. AsshowninFig. 6.2,twoair-filledPVCpipesareusedtohosttheFL(10kHz)andFH (25kHz) loop antennas. I TX is fixed at 64.52 and 56.21mA for FL and FH, respectively. Although the same antenna can be used for TX and RX roles, we assigned the PVC pipes with their specific roles for all experiments. The FL coil is placed on top of the FH coil and such order is arbitrary (this is not the case if one coil is placed inside the other). Even when the idle coil is disconnected from the circuitry, eddy currents can still be formed depending on the physical geometry and the current I TX passing through the active coil. Nonetheless, for the parameters used in our experiments, the proximity between the coils didnotsignificantlyimpactthesystemperformance. AsshowninFig. 6.3,forthedistance r=2m,theagreementbetweensimulatedandempiricalresultsisgood(averageerrorsmaller than 5%) and this is an indication that the dual-wire scheme is feasible. Fig. 6.3 also providesthefollowingimportantinsights: (a) Themeasurementsensitivityinthistestbediscriticalforalldistances. (b) Themodelerrorincreasesatsmallerdistances(e.g.,<1m). (c) Based solely on this figure, one can conclude that the signal is not significantly im- pacted by the soil conditions. However, comparing this figure with Fig. 6.1, we concludethatthissituationapparentlyonlyholdsforsmalldistances. 6.3 AdaptiveSystem: Analysis&Discussion Thenextresearchquestiontobeaddressedisrelatedtotheadvantagesoftheproposeddual- wirescheme. Thesametestbedconfigurationsareusedwiththepracticalconstraintofhav- CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 172 ingaminimumV C of25μVatr=2m. Tothisend,I TX willvaryfordifferentconfigurations. We want to test 4 basic combinations where 2 of them are related to a single-coil design. That is, we want to evaluate the penalties in maintaining the FH coil even at high vwc lev- els (or FL for low vwc values). For these cases, 10kHz and 25kHz are always associated with 40% and 1%-vwc, respectively, as already discussed regarding Fig. 6.1. Additional 4 casesareinvestigatedfora20%-vwcscenario. ThesimulatedresultsareshowninTable4.1. Before proceeding, it is important to understand why R ef f did not change, for instance incases#1and#2,whendifferentfrequenciesareusedatthesamecoil. Thisoccursbecause both cases fall at the conditionφ wire <δ wire in (6.5). Basically, the AC resistance effects are neglectedbutthiswouldnotbethecaseifaslightlythickerwireisused. AsshowninFig. 6.3,asignificantseparationexistsbetweenthesignalstrengthlinesfor FLandFH. Thisisanindirectindicationthatthesolutionsbelongtodifferentclasses. First, FL has smaller values for R ef f and BW; it is tailored to the worst case scenario (wet soil) and it is mainly an energy-oriented configuration. Second, FH is intentionally associated withahighervalueofR ef f inordertoincreaseBW accordingto(6.6);itisdesignedforthe bestcase(drysoil)anditismainlyaBW-orientedconfiguration. Thisanalysiscorresponds tothecases#1and#4,respectively,atTable6.1. Based on case #2, the single coil FL can be used but the solution can never switch to a higher BW. Similarly, based on case #3, the single coil FH can also be used for both dry and wet soil conditions, but the solution can never switch to a lower power consumption level. This example is not an exception: typically the MI-based communication design is dominatedbysuchenergy/BW balancedecisions. In summary, an adaptive solution for mid-range MI-based undergroundcommunication systems is proposed. It involves not only the frequency adaptation but also a coil switch- ing. Extreme soil conditions are then associated with specific configurations of frequency and coil. Nonetheless, the best advantage of this solution is the possibility of dynamically changing the operational mode of the MI node taking in consideration the soil conditions. This system feature is exemplified at cases #5 to #8 for a 20%-vwc case. For this scenario, itispossibleforthedual-wireMInodetoselectthemostefficientconfigurationintermsof energy, bandwidth, or both. For this example, cases #6 and #8 seem to be the most promis- ingones. CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 173 0.5m 1m 1.5m 2m 20 uV 100 uV 1 mV 6mV Distance r between the MI nodes Induced voltage at the RX side (V ) PA C FL setup (10kHz), dry soil (1% volumetric water content) model FL setup (10kHz), dry soil (1% volumetric water content) empirical FL setup (10kHz), wet soil (30% volumetric water content) model FL setup (10kHz), wet soil (30% volumetric water content) empirical FH setup (25kHz), dry soil (1% volumetric water content) model FH setup (25kHz), dry soil (1% volumetric water content) empirical FH setup (25kHz), wet soil (20% volumetric water content) model FH setup (25kHz), wet soil (20% volumetric water content) empirical FH setup (25kHz), wet soil (30% volumetric water content) model Worst signal variance: 0.61 µV (2.36% smallest signal) Distance between 1% and 20% vwc cases: 3.12 µV (1.02 dB) Distance between 20 and 30% vwc cases: 0.92 µV (0.31 dB) Worst signal variance: 0.92 µV (1.55% smallest signal) Distance between 1% and 30% vwc cases: 4.42 µV (0.63 dB) Fig.6.3RXinducedvoltagelevelasafunctionofdistance,frequency,andsoilconditions. Table6.1Study-case: dual-wirevs. singlecoil. # Coil vwc& f R ef f I TX TXpower BW 1 FL 40%,10kHz 4.0Ω 19.0mA 1.4mW 564Hz 2 FL 1%,25kHz " 2.8mA 31μW " 3 FH 40%,10kHz 16.8Ω 276.2mA 1279mW 3580Hz 4 FH 1%,25kHz " 40.3mA 27.2mW " 5 FL 20%,10kHz 4.0Ω 18.2mA 1.3mW 564Hz 6 FL 20%,25kHz " 3.0mA 36μW " 7 FH 20%,10kHz 16.8Ω 264.0mA 1168mW 3580Hz 8 FH 20%,25kHz " 43.5mA 31.7mW " 6.4 MI-WUSNAdaptiveSystems: FinalRemarks Based on what was presented at the previous chapters, the research optimization problem regardingthedesignofmid-rangeMI-WUSNscanbeinitiallyformulatedasfollows: Givenasetofcircuitconstraints,includingmaximumI TX ,coildiameter,expectedmini- mumvoltageV C ,etc.,whatisthebestoperatingfrequency(andtheassociatedcoilconfigu- ration)consideringacertainsoilscenarioandaweightedbalancebetweenenergyandBW CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 174 performance? Theaboveproblemdoesnothaveaclosed-formsolutionandwehaveproposedanopti- mizationalgorithmtospecificallyaddressthischallengeinChapter5. Thatstudycombined with this chapter show that due to the significant variation of the electrical properties of soils (e.g., due to rain and dryout event), the use of a single frequency (maybe aiming the average soil moisture case) may not be a proper choice if a robust and energy-efficient MI- WUSN solution is required. Therefore, it is proposed at least the use of two frequencies in conjunctionwiththefollowingstrategy: • Best-casesoilscenario(e.g.,drysoil,1%vwc): UseFH (higherfrequency)-selectedindesigntimeusingtheAlgorithm4 • Worst-casesoilscenario(e.g.,wetsoil,40%vwc): UseFL(lowerfrequency)-selectedindesigntimeusingtheAlgorithm4 • Othercases: Select FL or FH according to the energy cost, or application bandwidth (BW), or a weighted balance between energy and BW The FL/FH selection is also based on theAlgorithm4,buttheresultsfordifferentvwclevelscanbesavedatthenon-volatile memoryoftheMInode AsalreadydiscussedbasedonFig. 6.3,theFL/FH frequencyadaptationseemstobean important technique in any mid-range MI-WSUN if the soil is expected to have significant variationsofthesoilmoisturelevel(vwc). AnadditionalmessagebehindFig. 6.3isthatfor each pair (f, vwc), a maximum distance r (called break point distance or bd) is associated and circuit-related efforts after that distance may potentially have negligible or no effect at all at the MI communication performance. Therefore, this analysis is the starting point of anyMI-WUSNdesign. The study in Chapter 5 initially evaluated the FL/FH frequency adaptation scheme as- sumingthatthesamecoilconfigurationismaintainedforbothFLandFH cases. However, the reported simulated results show that although the dual adaptation is more efficient than the adoption of a single frequency as expected, the overall solution is still not optimum because when FL changes to FH (or vice-versa) the wire thickness of the coil cannot be adjustedaftertheMInodeisdeployed. Morespecifically,theargumentliesonthefactthat CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 175 R ef f is strongly impacted by both f andφ wire which is the diameter of the wire. Nonethe- less, in a single-coil setup, when f changes,φ wire maintains the same value and this aspect is the potential optimization opportunity that can be exploited. Accordingly, the Chapter 5 is concluded highlighting the need of the extension of that research problem which can be formulatedasfollows: Considering that the coil configurations for FL and FH employ the optimum (and po- tentially distinct) φ wire values, is it still practical/feasible to have a MI-WSUN node with twodistinctcoils(dual-coilscheme),oneforFLandotherforFH operation? Also,arethe energy/BW optionsfornon-extremesoilcasesbetterthantheonesassociatedwithacertain optimumvalueofφ wire whichcouldbemoreefficientlyusedbybothfrequenciesFLandFH inasingle-coilsetup? Theinitialpartofthischapterindirectlyinvestigatedtheabovequestions. Itisconcluded thattheuseoftwocoilsatthesamephysicalenclosurepotentiallyisnotanimpactingfactor if a) the idle coil is completely disconnected and b) one coil is placed on top of the other with a certain distance which must be investigated case-by-case. In fact, for the empirical parametersusedinourindoorstestbed,nointerferenceofthe FLcoilisdetectedatthe FH coilcausingaperformancedegradation(andvice-versa). Regarding the second part of the just stated extended research problem, one can inter- pretthatitisasimplequestionregardingthedecisionofasinglecoilortheadoptionofthe dual-wireschemewithtwocoils. However,besidesthatquestionthereisalsoaproposalof selecting an optimum value for φ wire which could provide an excellent solution if a single coilschemeis usedinconjunctionwiththe FL/FH adaptation. Thisquestion wasnotfully answeredinourresearchworkalthoughsofarwehavenotfoundsuchcase. Nonetheless,it ispossiblethatifbothenergyandapplicationbandwidthrequirementsarestronglyrelaxed, suchoptimumsingle-coilsetupcanbeeventuallyfound. What is apparent by means of the numerical examples presented so far is that the dual- wire scheme indeed provides better dynamic configuration options no matter the soil con- ditions. ThisimportantpartofourstudycanbebetterdemonstratedinTable6.1,wherewe want to give particular attention to cases #6 and #8. The difference of TX power between these cases is almost 3 orders of magnitude. This is the cost for that scenario associated to increasing the application BW by almost 6 folds. These two specific cases indeed exploit CHAPTER6. DUAL-WIRESCHEMEADDEDTOFREQUENCYADAPTATION 176 the mentioned opportunity of having the bestφ wire associated with f=FL (or FH) in order toachievethebestenergysolution(orthebestapplicationBW). Nonetheless,allthecasesatTable6.1stillhaveaverysmallBW comparedtotraditional WSNapplicationsandthisispotentiallyacharacteristicofmid-rangeMI-WUSNswhichis veryhardtochange. Accordingto(6.6),BW canstillbeincreasedbeyondthetypicalvalue offewkHzif: • Ldecreases,whichisultimatelyconstrainedby: Themaximumallowableincreaseofpowertosustainthecommunication: bothcircuit gains - G TX and G RX - are strongly impacted by the physical dimensions of the coil andI TX mustdrasticallyincreasetobalancethedecreaseofbothG TX andG RX • R ef f increases,whichisultimatelyconstrainedby: Thecombinationof f andφ wire accordingto(6.4)and(6.5) Therefore,thenovelconclusionprovidedinthischapteristhatanincreaseof f maynot increaseBW bytheexpectedamountifφ wire isthelimitingfactor(viaR ef f ). However,ifthe MI coil has two or more coils with distinctφ wire values, the frequency adaptation becomes extremely powerful because more combinational options are available in order to increase or decrease the values of R ef f and such options are in actually possible pairs of (TX power , BW) that can be employed according to the application needs, energy state, and soil condi- tions. Ontheotherhand,ifarelativelysmallbandwidth,suchas564Hzgivenattheprevious example, is enough for the continuous operation of the WUSN application, the dual-wire schemeisnotactuallynecessary,justthefrequencyadaptation. Forthecase#3inTable6.1, if the same coil configuration is used when the soil becomes saturated (vwc=40%), the MI nodeinthissingle-coilsetupjustneedstochangeitsfrequencyto10kHzinordertomain- tain excellent communication with the cost of only 1.4mW. This case was investigated and validated in our 2m-testbed. On the other hand, once deployed this solution is permanently constrained to such lower BW value. The bottom line of this chapter is that the frequency adaptation is indeed a very important mechanism for mid-range MI-WUSNs and the dual- wire mechanism is proposed as an optional add-in solution whenever it is known at design timethatcriticalbandwidthrequirementsmustalsobesatisfied. PartIII Energy,Reliability,andNetworking AspectsofMI-WUSNs 177 Chapter7 Energy-ManagementFrameworkfor SpecialWSNs Thischapterismainlybasedonthecontentsofthreeofourpapers:[4],[21],and[101]. We developedannetworkingarchitecturecalledRipple-2toprimarilytargetingtheSoilSCAPE project. Designed to support in-situ soil moisture measurements for the National Aeronau- ticsandSpaceAdministration(NASA)SMAPMission,theSoilSCAPEprojectstartedmore than 6 years ago and currently (August 2015) it has more than 141 sensor nodes and 500 soil probes spread in 7 sites, each one with a Wireless Sensor Network (WSN) connected to a central data server by means of cellular text (SMS) and 3G links. These unattended WSNsareamongthelargestexistingoutdoornetworks(intermsofnodesparsity)andthey are operating continuously for more than 2 years. All regular sensor nodes employ non- rechargeable batteries and the failure rate due to software and electronic issues is steadily zeroduringtheyears. Although the SoilSCAPE project was developed specifically for terrestrial WSNs, the characteristicsofthissolutionareremarkablyverysimilartotheonesofWUSNsregarding theneedofverylevelsofresilienceandenergy-efficiency. Regardingtothefewenergyhar- vesting opportunities associated to WUSNs, the study in [34] provides additional insights. Nonetheless, in this study it is assumed that energy harvesting is not currently a practical optionforthetargetMI-WUSNsunderanalysishere. Therefore,webelievethattheRipple- 2 architecture can provide a time-proved foundation for MI-WUSNs because the battery lifetime of the buried nodes can realistically reach multiple years. Nonetheless, it is impor- tanttohighlightthatextensionsoftheRipple-2architecturearestillrequiredforacomplete MI-WUSN solution and this effort is part of our ongoing research. In particular, 2 novel 178 CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 179 aspects must be investigated: a) the power requirements of MI-WUSN devices and b) the modificationoftheselfishnodeconcept(tobeexplainedinthischapter)toallowsomeform oflightcollaborationamongnodes,anaspectthatwasintentionallyremovedattheoriginal Ripple-2architecture. 7.1 FrameworkForSensorNodesWithConstrainedEnergy Profiles Despites the possible existence of power scavenging sources for WSN (Wireless Sensor Networks) nodes, this fact does not necessarily imply a longer lifetime or a high reliability level for such nodes. For instance, when a photovoltaic cell is used, the energy harvesting process is typically not continuous or stable. Moreover, when rechargeable batteries are part of the energy harvesting system, the lifetime of the node is ultimately dictated by the age or by the number of recharging cycles of those batteries, among other factors. Even with very well controlled charging cycling, typical secondary cells (rechargeable batteries) for WSN nodes have a lifetime smaller than 3 years. A potential solution to achieve a 5 to 10-year maintenance-free solution is the adoption of a battery-free design, as proposed in [102],wheresupercapacitorsareusedastemporaryenergyreservoirs. However,besidesthe assumption of an available form of energy harvesting, the main challenge of such solution is still the need of sustaining a certain level of reliability even when the current capability ofthepowersourceisinsufficientforthenodeoperation. Theaboveissueisaggravatedwhentheduty-cycleofthenodeisnotsolelygovernedby the main application. This is the case when the node must also actively collaborate in the network. Moreover,theefforttoincorporateenergyconsumptionmetricsonexistingphysi- calandhigher-layernetworkingprotocolsisstillachallengeandtypicallysuchprovisionis not implemented in commercial WSN solutions. As a result, it is very hard to achieve very long lifetime of a node associated with a relatively high reliability level. In this chapter, an adaptive and flexible framework for sensor nodes with constrained energy scavenging profilesispresented. Thisenergy-managementframeworkhashardwareandsoftwarecom- ponents and a significant emphasis on integration is given here in order to facilitate the partial or integral adoption of the proposed framework on existing WSN platforms. At the contextofthisthesis,thepresentedguidelinesprovideasoundfoundationforthedesignof CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 180 power-managementsub-systemsofMI-WUSNnodes. The motivation for the work presented in this chapter is associated with the goal of having a reliable WSN solution (including MI-WUSNs) with a lifetime between 5 and 10 years. That is, during this period of time, no human intervention is expected due to power depletion of a node. It will be shown that, to achieve this goal, the complexity level of the solution at the design-time is relatively high. Also, the initial cost of a node is expected to be relatively high compared to traditional off-the-shelf WSN nodes (e.g., 30% higher cost, based on our design experience with Ripple-2 nodes). However, the practical functionality and the total ownership cost (TCO) of the system at long-term can be very attractive. Note that this work diverges with the traditional idea that a WSN has hundreds of very-low cost nodes with a relatively high probability of failure. On the contrary, the focus of this work is on the achievement of a very high-quality and controlled solution. Note that this goal is particularly important for MI-WUSNs nodes because the maintenance costs can be poten- tiallymultipletimesthevalueofthedeviceitself. This chapter starts with the presentation of the energy effort tripod and energy control loop concepts in Section 7.2. It is highlighted that effective energy savings for WSN nodes typically lies on a balanced solution in terms of hardware, network, and application de- mands. In Section 7.3, the foundations of the proposed framework are discussed: a) the optionalbutrecommendeduseofprimarycellsassociatedtoharvestingsystems,b)thead- vantages of a distributed system inside a node, and c) the adoption of the dual duty-cycle operation(DDC)forWSNnodes. Thecorepartoftheproposedframeworkisacross-layer network protocol which is presented in section 7.4. This protocol is implemented as an application-layer overlay on top of existing WSN solutions. Such overlay mechanism can be dynamically activated and deactivated in order to allow the network to achieve the best performance while satisfying existing energy constraints. Many of the components of the proposed framework are in fact part of a long-term and ongoing project involving one of thelargestoutdoorsWSNdeploymentsstillinoperation[5,17,20]. Thefieldresultsofthis projectinconjunctionwithsimulatedoutcomesarereportedinSection7.5andthischapter is concluded in Section 7.6 where specific comments regarding MI-WUSNs are also in- cluded. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 181 Table7.1MainTermsandAcronymsMainlyIntroducedinThisChapter Term Explanation EnergyEffortTripod Fundamental framework’s concept: an energy-balanced WSN solution involves hardware,network,andapplication. EnergyControlLoop Fundamental framework’s concept: the operation of a WSN node must be regu- latedbyitsenergystate. SelfishNodeConcept Ultra-light behavior of an node: wake-up, sense-and-send, sleep. The expected networkoverheadis<1%. DDC(Dualduty-cycle) 2modesofWSNoperation: a)current(ortraditional)WSNsolution(RDC)and b)constrained,buthighlyenergy-efficientmode(LDC). RDC(Regularduty-cycle) Provides the same network performance as the existing regular WSN solution with potential energy penalties. For a MI-WUSN node, this mode means higher throughputbetween2nodes. LDC(Lowduty-cycle) Maintenance mode with excellent energy performance. It can also be used for lowduty-cycledata-collectionapplications. DDCSwitching Framework provision that allows that application’s needs (or an event) automat- icallyswitchtheDDCmode. DDCSystem Part of the proposed energy-management framework which involves hardware andsoftwareadditionstotheWSN. DDCNode Itisadistributedsystem(multipleMCUs)insideasingledevice. ThemainMCU runstheBETSprotocol. PowerGating Technique extensively used in a DDC node which allows the hardware modules ofanodetobeactivated/deactivated. BETSProtocol Best-Effort Time-Slot allocation protocol is the core DDC module that controls thebehaviorofthenetworkinLDCmode. EnergyReservoir InaDDCsystem,theenergyreservoiristypicallysupercapacitors,primarybat- teries,oracombinationofthem. WOR(Wake-UpOnRadio) Micro or nano-power technology that allows the DDC node to switch to RDC modewhenaneventofinterestoccurs. BeaconTX Radio transmitter used to trigger remote WORs. Typically, it is separated from theregularradiotransceivermodule. IntelligentSensor ItisnottheWSNnode,butreferstoasensorprobethathascapabilitytowake-up byinterruptionaMCU. ED(EndDevice) TheregularsensornodeintheBETSarchitecture(LDCmode). Notbeconfused withEDterminIEEE802.15.4/ZigBee. CH(ClusterHead) The data-collector node in the BETS architecture (LDC mode). ED nodes only communicatewithaCHnode. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 182 7.2 EnergyManagementinWSNs Inthissection,typicalpitfallsandchallengesinthedesignofenergysystemsforWSNsare discussed. Next, the important energy effort tripod and energy control loop concepts are introduced. 7.2.1 DesignChallengesandPitfalls Manywell-designedprojectsfailduetosmalldetailsandincorrect(butgenerallyaccepted) assumptions. Therefore,beforepresentingtheproposedframework,itisimportanttohigh- lightsomeaspectsassociatedwiththecurrentstate-of-the-arttechnologyonenergyharvest- ingsystemsforWSNs. WSN design: besides long lifetime, reliability: One critical pitfall associated to the energy aspect of WSN designs is to perpetuate the original vision of a WSN with hundreds to thousands of very cheap nodes [103] where a high rate of node failures is actually ex- pected. Althoughsuchvisioncanstillcorrespondtotheneedsofsomeapplications,aquick investigationatthecurrentWSNdeploymentsaroundtheworldrevealsadifferenttrendfor WSNs. For instance, few existing long-term networks actually have more than 50 nodes. More impressive is the ongoing success of the infrastructure-based WSN solutions (star or tree topologies), such as the ones based on IEEE 802.15.4/ZigBee [104, 105]. A higher node reliability is typically crucial when WSNs move from ad-hoc to infrastructured archi- tectures. Accordinglytothiscurrenttrend,asignificantemphasisinthischapterisgivento thereliabilityofthenodes. Inthiscontext,thetermreliabilityisassociatedwiththegoalof havingnodesthatrarelybecomeunavailableduetoanon-controlled powerdepletion. Energy scavenging does not imply a perpetual lifetime: Another pitfall associated to the design of WSNs is to consider that the adoption of an energy scavenging system is automatically associated to an endless node lifetime. Besides the need to consider the life expectancy of the sensing components, such as a humidity sensor or a soil moisture probe, typical power systems rarely can achieve a 5-year lifetime due to a plurality of reasons dis- cussed in this section. Therefore, the first step toward a successful low-cost WSN solution (in terms of functionality + reliability + long lifetime) is the investigation of the expected lifetime of each of the components of an node. In general, the reported premature death cause of WSN nodes is the energy subsystem, in particular the batteries. Primary (non- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 183 rechargeable) batteries typically have a very short lifetime in WSNs [4, 106]. However, it is also important to have in mind that secondary (rechargeable) batteries potentially have a lifetimesmallerthan2-3years. Possibility of adopting a non-realistic energy model: A significant number of WSN papers present three regular inaccuracies regarding to the way the energy model is pro- posed or adopted. First, the transients, such as due to the activation/deactivation of a radio transceiver, are typically neglected as pointed out in [107]. Second, many values used as input parameters for the models are directly imported from the datasheets of the compo- nents without further consideration of the effects of integrating these components together. Forinstance,basedonitsdatasheet,aradiotransceivermodulehasanominalsleepingcur- rent of 10µA. However, it is observed that, once it is attached to a MCU, leak currents are detected and they are many folds higher than the nominal sleeping current. Similarly, a voltage regulator can be included in the design of a WSN node, in particular when energy harvesters are also employed. However, many reported energy models do not include the energy cost of a possible voltage regulator. As a result, an energy model that disregards the existence of voltage conditioners can be drastically distorted (non-linearly) when it is adopted in a node. The severity of this statement can be illustrated by the following real- case. When both MCU and radio modules are sleeping, usually it is not possible to put the voltage regulator in sleeping mode (typically shutdown mode in the context of regulators). Therefore,ratherthananexpectedamountinµWasthepowerconsumptionforthenode(as expressedinmanyWSNenergymodels),thenodecanpotentiallyhaveaneffectivesleeping consumption on the order of mW. Therefore, as one varies the application duty-cycle, the adoptedenergymodelrevealsitsnon-lineardistortion. The third pitfall commonly observed in the WSN literature is associated with the use of batteries. In general, it is assumed that 100% (or a close value) of the initial nominal energywillbeactuallyavailablefortheoperationofthenode. Somepapersevenclaimthat this assumption because both MCU and radio support a low voltage level, such as 1.5V. In practice, it is very hard to achieve related values close to 80%. Factors that invalidate the mentioned assumption are: the self-discharging current, the aging of the battery, temper- ature, discharging regime, etc. As a rule-of-thumb, when the battery reaches its terminal state, a significant amount of energy (e.g., >25%) still remains inside the cell. However, only very low discharging currents are typically possible from that moment on. Note that even if the load affords a low voltage level, the bottom line is actually the constraint of the CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 184 loaddrainingonlyverytinycurrents. Thisishardlythecaseinparticularforradiomodules. Therefore, if the hardware/software solution embedded on that node does have any provi- sion to use this remaining energy at the cell, the adopted energy model must only consider conservative values for the actual initial energy stored at the battery. In many cases, such conservativevalueislessthanhalfofthenominalenergyofthebattery. As observed, realistic energy models for WSNs are inherently complex but can be sim- plified if conservative values are adopted. Moreover, every time a new WSN platform is designed, a significant number of experiments involving the final hardware, different kinds of batteries, and realistic discharging regimes must be considered before an energy model can be adopted for that node. On the other hand, it is interesting to observe that energy models for batteryless solutions are typically reported as properly matching the application needs [108]. In general, it is the case because detailed empirical investigation is realized to justify the energy model in a critical scenario involving a small amount of available energy attheenergyreservoir(e.g.,asupercapacitor). Lifetime of rechargeable batteries: Typical secondary cells used in WSNs require special attention because, besides their inherent shelf lifetime (e.g., <3 years), there are other factors than can drastically reduce their lifetime. For instance, the maximum num- ber of nominal charging cycles is usually smaller than 1,000 considering the kind of cells typically used in WSN nodes. Therefore, without a careful control of how and when the charging cycles are performed, the lifetime of such cells can be realistically smaller than 1 year. Moreover, temperature is a critical factor in particular for secondary cells. In general, extreme temperature can drastically degrade the cell’s performance. For instance, in [4] it is reported that at sub-zero temperatures, many solar-powered nodes stopped the charging process followed by long periods of network inactivity, as shown in Fig.7.1. For this spe- cific case, a solar panel completely covered with snow is one of the reasons for the issue. However, for the majority of the cases, the inability of the secondary cell to be charged at sub-zerotemperaturewasthemainreasonbehindthefunctionalfailures. Itisalsoobserved inFig.7.1thatnodespoweredbyaprimarycell(LithiumThionylChloride,inthisexample) arenotimpactedbyextremetemperatures. Solid-state batteries and WSNs: Solid-state batteries are a recent technological ad- vance that can impact the design of future WSN solutions. These secondary cells are claimed to have a lifetime between 5 to 10 years and a maximum number of charge cycles CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 185 2 NiMH batteries 1Li-SOCl2 battery Fig.7.1Effectofsub-zerotemperaturesonsecondary(top)andprimary(bottom)cells(Ann Arbor,MI,USA).Therechargingprocessofthesecondarycellsisimpactedbylowtemper- aturescausingnodefailures(linesinthefigure). Primarycellsaremoreresilienttoextreme temperatures. between 5,000 to 10,000 [109]. Based on these preliminary values, it is possible to envi- sionareliableWSNsolutionwithaverylonglifetimebasedonsuchbatteries. Nonetheless, these cells have three significant drawbacks: high cost, low energy density, and low power density. Whiletheformeraspectcanbeonlyamatteroftimeasafunctionoftheindustrial scale, the remaining aspects reinforces the need of a careful designed energy-management systemifsuchcellsareexpectedtobeusedinWSNnodes. Lifetime of low-cost outdoor energy harvesters: The life expectancy of low-cost energy-harvestersforoutdoorsistypicallynotinformedbythemanufacturers. Forinstance, to date, manufacturers of micro-wind turbines do not provide such information. Similarly, although relatively big solar panels (e.g., >30cm x30cm) are typically robust and have a realisticlifetimeofmorethan3years,itisnotthecaseinrelationtosmallsolarpanelsused CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 186 in WSN nodes. We performed outdoor tests during more than 2 years with different types and models of small solar panels. Unfortunately, the results were very disappointing: the majority of the panels presented a significant performance deterioration in less than 1 year. The majority of them changed their glossy surface by a white porous surface where dust easily accumulates. In one of the sites which experiences high temperatures (e.g., >40 ◦ C), more than 10% of the panels cracked. To the date, we did not find off-the-shelf small so- lar panels with the typical robust encapsulation found at bigger panels. Another critical aspect in relation to small solar panels left unattended in outdoors is the action of birds. In our outdoor deployments involving sites in three states of USA, we observed the same phenomenon: a small solar panel mounted on top of a pole is a typical place where birds choose to temporarily rest. Without a proper protection against the birds, such solar panels potentiallyrequireperiodiccleanness. 7.2.2 TheConceptsofEnergyEffortTripodandEnergyControlLoop In the previous section, some aspects related to the energy subsystem of a WSN node are highlightedanditwasshownthattheachievementofamaintenance-freesolutionforperiods of more than 5 years is not a trivial task. At that analysis, the network and application as- pectsarenotconsidered. However,theadoptionofanenergymanagementsolutionrequires theintegrationofenergyeffortintermsofhardware,includingthereportoftheenergystate of the nodes, the network, and the application. Accordingly, the focus of this section is to discuss how these three aspects are properly integrated in an energy-management frame- work. The term framework is defined as a broad outline of interrelated items, not a detailed step-by-step set of strict guidelines. The advantage of this design approach is manly the gains in terms of flexibility: one is exposed with some underline concepts and ideas and can easily adapt them to his problem or environment, in this case, a certain WSN platform. Accordingly, the basic concepts available at the WSN energy-management literature are summarizedbytwopicturespresentedinthissection. The first concept, the energy effort tripod, is illustrated in Fig.7.2. Assuming a certain limitedenergylevelforanode,anefficientwaytoachievefunctionalityandreliabilityfora verylongperiodoftimeistobalanceeffortsintermsofhardware,networkalgorithms,and CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 187 applicationdemands. Forinstance,ifadvancesinthehardware/softwareofanodeallowitto reduceitssleepingenergyconsumptionbyoneorderofmagnitude,sucheffortispotentially voided if the effective duty-cycle of the node (due to the network, the application, or both) is still very high. Similarly, a significant reduction of the network overhead can have little energy impact compared to a very high and frequent application demand (e.g., multimedia datatraffic). Therefore,thestartingpointistolimitthedemandofthemainapplicationand to define possible acceptable levels of Quality of Service (QoS) for the nodes, a group of nodes, and the network. As expected, service metrics are required in energy-management systems. Such metrics involve data latency, volume of data traffic, frequency of data bursts (e.g.,schedulingindata-drivenapplications),datalossacceptance/levels,etc. Oncetheapplicationdemandsareclearlydefinedandrealisticallyconstrainedconsider- ingtheenergysystemsandpowersourcesavailableforthenodes,thenextstepistoevaluate howthenetworkandthehardwareofthenodescanbeimprovedor,inotherwords,balanced withtheapplicationdemands. Ingeneral,theadoptionofaveryflexibleWSNsolution(not tailored to a certain category of application) is also associated with energy-hungry network protocols. For instance, consider a WSN application that monitors every 5 minutes the in- frastructure of a bridge. One strategic question to be considered in this case is related to how the network protocols can be optimized considering a static topology and also a fixed monitoringschedule. Similarly,sometimesaveryenergy-efficienthardwaresolutionisnotproperlybalanced with the overall solution. It is possible that such energy-saving mechanisms in such hard- warecanactuallyimpactthefunctionalityandreliabilityofthenetworkandultimatelyalso the main application. For instance, consider the case of a batteryless node which adopts a combination of solar panel and supercapacitors. While during the daylight the hardware of the node is functional and energy-efficient, during the night its behavior can change and impact the application. In this case, a batteryless node is potentially not capable of per- forming transmissions multiple times per second, even under a very low duty-cycle (e.g., <1%). Nonetheless, many current WSN network protocols operate assuming this network patternanditisclearthatthehardwareandthenetworksolutionsarenotproperlymatched balancedinthisexample. Insomecases,theillustratedbatterylesssolutionforWSNnodesimposesacertainlevel of data latency which is incompatible with the application requirements and again the bal- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 188 Network Hardware Application Functionality Reliability Fig. 7.2 Energy effort tripod concept: coordinated efforts involving hardware, network al- gorithms,andapplicationdemandsleadtoanenergy-balancedandefficientWSNsolution. anceisnotachieved. Themainpointbehindthisdiscussionisnotaboutadvocatinginfavor ornotinrelationtoacertaintechnology. Itisactuallyrelatedtowhatcanbeadjustedinthe WSN design in order to obtain a proper balanced solution in terms of energy, functionality, andreliability. Inmanycases,anoptimizedsolutioniscomplexbecauseitisonlyachieved by combining enhancements in the hardware, in the network, and also in the application (i.e., relaxing the required QoS metrics). For the latter aspect, it is clear that control is nec- essaryand,infact,thisistheroleofenergy-managementmodules,asillustratedbythenext conceptualfigure. Thesecondconcepttobediscussediscalledenergycontrolloop,asillustratedinFig.7.3. Such control can be performed at the node level, in a portion of the network, by means of a central data server, or by a combination of these options. As shown by Fig.7.3, the op- eration of the node, such as the activation of an energy harvester, or the activation of the radio module, or the way the node behaves in the network, is governed by the decisions of an energy-management module. As expected, such actions impact the energy state of the node, such as the remaining energy available for the node. Therefore, a proper design goal is to have an energy-management module that receives feedback related to the operation of the node and also energy-related data. Note that the dashed lines used in the figure are an indication that such feedbacks are optional. Specifically, it is possible that the energy- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 189 Energy-management control Node operation Energy state DECISION IMPACT DATA DATA Fig. 7.3 Energy control loop concept: the operation of a WSN node is regulated by its energy state. The decisions are triggered by an energy-management module that can be implemented internally in the node, at the network level, in a centralized data server, or by acombinationoftheseoptions. management module makes inferences about the energy state of a node without receiving explicit feedback data from the node. Next, we will see how energy-management control effortscanberealizedatnode,network,andcentralsystemlevels. Consider a scenario where part of the energy-management processing is performed in- side the node, as illustrated in Fig.7.4a which is related to the activation/deactivation of an energy-harvester. Such control can be realized purely in the node level without requiring anynetworkactivity. Inthiscase,theenergystateisrelatedtothemeasurementsofvoltage levelsandoutputcurrentsofthepowersource. Themaximumpowerpointtracking(MPPT) technique is one example of such kind of energy control which allows the energy harvester toachieveitsmaximumefficiency. The energy-management can also be realized by means of energy-aware networking protocols, as illustrated in Fig.7.4b. Consider an example of a collaborative protocol that dynamically assigns certain roles (e.g., cluster head, router, data aggregation/fusion master node,etc.) tothebestqualifiednodesaccordingtotheirremainingenergy[110]. Observein Fig.7.4b that two feedback data flows occurs: one related to the application data and basic network functionalities and the other associated specifically with energy metrics. In some cases, the final decision related to the temporary role of a node at the network can be done at the node itself without further network activity. In other cases, the decision is performed byaspecializednodethathasamoreholisticviewofthenetwork. In many cases, the energy decisions performed at the node and network levels can be CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 190 MPPT control Energy Harvester ON/OFF Voltage and current measurements (a) Command * Charges the energy reservoir * Powers the load Measurement values Network protocol (cluster head) Network protocol (node) Radio Module ON/OFF Remaining energy at the node (b) Command App + Network data Energy-related data Discharges the energy reservoir Main application (Data Server) Radio Module ON/OFF Remaining energy at the node (c) Command Application data Discharges the energy reservoir Fig. 7.4 Examples of energy-management efforts. (a) At the node level: optimized control of an energy harvester, (b) at the network level: the remaining energy of a node is used as criteria for its selection in network activities, and (c) at a centralized level: based on the sensing data received from the nodes, the data server defines what nodes will sense accordingtolocation/timescheduling. insufficientfortheachievementoftheexpectedQoSmetrics. Attheotherextreme,thereare caseswherethequantityofthesensingdataisfarbeyondthenecessarylevelofinformation andthereisroomforenergyoptimization. Forinstance,considerthecasewhenitisenough from the viewpoint of the main application that only one third of the nodes monitor a spe- cific event in a certain area. Typically, only the Data Server can take such decision because it is closely associated to the historical data flow from multiple sensor nodes and also with the metrics stored exclusively at the Data Server. Accordingly, as shown in Fig.7.4c, the mainapplicationatthecentralDataServercanissuecommandsassociatedtothescheduled activity for each node at the network [17, 111]. Observe that the feedback line in Fig.7.4c is omitted which seems to be counter-intuitive considering that the discussion is about the control loop concept. However, in many implementations of energy-management systems, the energy state of the nodes can be simply inferred based on the network activity of the nodesandexplicitcontrolfeedbackisnotnecessary. Sincetheultimategoalofthischapteristhepresentationofaframework,notallpossible formsofenergy-managementimplementationsarerepresentedbythepreviousillustrations. Nonetheless,itissafetostatebasedontheinvestigationofrelatedworkthatthemajorityof such systems are actually a combination of efforts at the node, network, and central levels. In general, such approach is also associated with a complex design and higher implemen- tation costs. On the other hand, such approach is also the realization of balanced efforts highlightedbytheenergyefforttripodconcept. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 191 7.3 Energy-AdaptiveFrameworkforWSNArchitectures In this section, the generic discussion in Section 7.2 evolves in more practical and flexi- ble guidelines that will compose the proposed energy-adaptive framework involving WSN nodes with constrained energy scavenging profiles. It is important to highlight that this frameworkisnotbeingproposednecessarilytosubstituteexistingonesbuttooptimizesuch efforts in order to achieve very long maintenance-free periods of time for the nodes in con- junction with high levels of functionality and reliability. This section, which is almost one third of this chapter, starts with a presentation of the three foundations of the framework. Next, a discussion about the characteristics of the energy-efficient Low Duty-Cycle (LDC) operationalmodeareprovidedinconjunctionwithpotentialtargetplatformsfortheframe- work. Finally,high-levelimplementationguidelinesareprovided. 7.3.1 Foundation1: TheStrategicUseofPrimaryCells Primarycellshavehighenergydensities,typically3timesincomparisonwithrechargeable batteries [4]. That is, for the same physical volume, primary cells provide higher energy capacity. Another advantage of a primary cell is the possibility of using almost 90% of its nominal energy capacity by menas of proper techniques [4], in contrast with rechargeable batteries,asalreadydiscussedinSection7.2.1. Moreover,primarycellsareveryresilientto extreme temperature in outdoors. Finally, the typical shelf-aging of primary cells is around 10 years compared to 3 years of secondary cells. Nonetheless, for the best of our knowl- edge,thisisthefirsttimethatprimarycellsarehighlightedasanimportantbasisforaWSN energy-management system that involves energy harvesters, in particular if the goal is to achieve a maintenance-free period of more than 5 years. Also, the inclusion of such cells increases the costs and the size of the WSN node. Therefore, it is very important to un- derstandthecontextandassumptionsbehindthisproposalandsuchanalysisisdividedinto fourtopics: Optionaluseofprimarycells: Theuseofprimarycellsiscloselyrelatedtothegoalof longlifesolutionwhileachievinghighlevelsofreliabilityandfunctionalityforthesolution. However,therearecasesthatsuchprovisionisnotnecessaryandabatterylesssolutionfully satisfytheapplicationrequirements. Forinstance,considersensornodesthatharvestenergy from mechanical vibrations at the engines installed in an industrial plant. Potentially, the energybudgetof these nodes can be sufficientfor a realistic zero-energy, batteryless, WSN CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 192 implementation based solely on the mentioned harvester and supercapacitors. In this case, assumingthattheenergyharvestersalsohavealonglifetime,thereisnoneedtoimplement almost the entire framework proposed in this chapter because it is clear that the nodes do not have a critical constrained energy profile. Similarly, the relative small costs of regular maintenance of nodes installed at indoors can justify the avoidance of energy-management mechanisms. Therefore, the proposed use of primary cells is clearly optional and depends onthecharacteristicsofeachWSNapplicationanditsenvironment. Primary cells have lowpowerdensity: In practice, the goal of using 90% or more of theenergycapacityofprimarycellsisseldomachievedinWSNs. Basedonacarefulstudy regardingthistopic[4],potentially10%to3%ofthenominalenergycapacityofaprimary cell can bea reallistically used by a typical WSN node. The main reason behind this fact is identified in the same work: these cells cannot sustain their nominal energy capacity if fre- quent peak currents (e.g., >15mA) occur. In fact, typical WSN nodes have peak current of morethan>100mA.Therefore,thealreadymentioned3-timesenergycapacityadvantageof primary cells in comparison with secondary cells (assuming the same volume) is cancelled by the peak current effect. One can verify that even in the case of a rechargeable battery with50%ofresidualenergythatcannotbeusedbythenode,thisenergyreservoirwillhave two times the effective energy capacity of the mentioned primary cell. Considering this analysis, it is not a surprise to read frequent reports about the need of exchanging primary cells in WSNs in regular periods of few months or even weeks [106]. As expected, the recommendation of using primary cells in the proposed framework only holds if the peak currenteffectcanbemitigated[4]. Achieving a reliable energy harvesting system: The use of primary cells comes to the framework as a way to guarantee certain levels of QoS considering the occurrence of situationswheretheenergyharvestingresourcesarenotenoughtosustainthefunctionality of the node, part of the network, or even the overall network. We acknowledge that the ideal energy-management system would not have an battery of energy-related component that requires regular maintenance. But, once batteries are used, the remaining energy at these energy reservoir must be controlled. Without any energy control, the primary cells alsobecomeanuncertaintyfactoratthesystemwhichmitigatesitsimportancetoachievea higherreliabilitylevel. Thebottomlineisthatprimarycellsareonlybeingproposedinthis contextifitispossibletomeasure/predicttheircurrentcapacity(atnodelevel). Study-case: Duringaperiodofmorethan2years,weadoptedastandardWSNsolution CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 193 based on ZigBee technology, solar panels, and rechargeable batteries in two outdoor sites with extreme high and low temperatures [17]. The reliability of this solution was clearly impactedduetotheperformanceoftheenergysubsystem. Later,wedesignedaWSNnode called Ripple-2A with significant enhancements in terms of network performance. How- ever, this node maintained the original and traditional design for the energy subsystem al- thoughimplementedbymeansofdifferenthardwarecomponents. AlthoughRipple-2Ahas achieved the design goals in term of efficiency with overall network overhead smaller than 1%,theweakestpartofthesolutionintermsofreliabilityandlong-termlifetimewasagain the energy subsystem. As already discussed in Section 7.2.1, the main issues were related to relative small lifetime for the rechargeable batteries (around 1 year), their performance underextremetemperatures,andthefragilityofsmallsolarpanelsavailableatthemarket. ThenextstepwasthedesignofaWSNnodewhichcanbepoweredbynon-rechargeable batteries. Several months of research were dedicated for the realization of long-term out- door experiments that could validate support this approach. The new node design is called Ripple-2D, leaving room for two additional kinds of nodes: Ripple-2B (solar panel + su- percapacitors) and Ripple-2C (solar panel + supercapacitors + non-rechargeable batteries). The latter one is currently under development and it follows many of the guidelines pro- posed in this chapter. The Ripple-2D solved the pulse current effect by slow-charging of supercapacitors which in turn power the radio module. In short, the current drained from the battery never goes beyond 15 mA and the pulse effect is voided. Based on the results from simulated and empirical (accelerated and very-long term experiments) results, the ef- fective capacity of the battery is found to be higher than 90% in relation to the the nominal value[4]. ThedesignedlifetimeoftheRipple-2Dunderthisprojectisalmost2years. Cur- rently, the majority of the nodes reached half of this period of time in a continuous and reliableoperation[5],provingthatthissolutionsignificantlysuperiorcomparedtotheolder Ripple-1andRipple-2Aarchitecturesthatarebasedonrechargeablebatteries. These results provided the foundation for the realistic achievement of a very long life- time for the nodes in conjunction with a high level of reliability. Moreover, the overall network solution also provides a very deterministic way to forecast the lifetime of each individual node [5], as will be discussed in details in Section 7.4. At the Data Server’s side, it is also possible to dynamically extend the lifetime of the nodes by means of spatio- temporal activation of a subset of the nodes [17]. In this way, the initial life expectancy of 2 years can still be significantly extended. In short, this study-case demonstrates that the energy-management efforts can be realized at node, network, and Data Server levels. The CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 194 Analog Sensor Energy-Management Subsystem Energy Harvester Energy Reservoir(s) Energy-Management Controller Power scavenging source(s) Wake-Up On Radio (WOR) Main MCU Intelligent Sensor Digital Sensor Real-Time Clock (RTC) Data line Power line Customizable MCU Potential Non-Custom. MCU Wake-Up Interrupt Beacon TX Power-Gating Dual Duty-Cycle (DDC) Node Fig. 7.5 The Dual Duty-Cycle (DDC) node has multiple microcontrollers (MCUs) and in- telligent devices. That is, it encompasses a distributed system inside itself. The main goal is to achieve energy savings at unprecedented levels compared to traditional nodes. The existence of multiple power lines (lines without arrows) rather than a single power line is associatedwiththeuseofthepowergatingtechnique[4]anddifferentvoltageconditioning schemesfortheinternalmodules. next generation of WSN node for this project, Ripple-2C (currently under development), combinesasolarpanelwithsupercapacitorsandnon-rechargeablebatteries, exactlyasrec- ommended in this section for energy-constrained scenarios. The ultimate goal is to extend thelifetimeforrealisticvaluesbeyond5yearsinordertoproperlysupportthemainproject behindthiseffort[17,20]. 7.3.2 Foundation2: DistributedSystemInsideaSensorNode At the proposed framework, a node can switch between its regular operation and a very low duty-cycle operation. From now on, we will call a WSN node that supports such dual duty-cycle (DDC) mode of operation as a DDC node. The implications associated with the dual mode of operation at the design of a node are definitely not trivial, in particular when energy harvesters are employed. Significant hardware and software additions at the DDC node are required and the reasons why such changes are necessary are better understood CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 195 whentheoverallframeworkisexplained. Atthissection,abriefpresentationofthemodifi- cations in a DDC node are listed. The fundamental change is related to the evolution from an architecture model centralized in a single microcontroller (MCU) into a solution with multiple MCUs, each one with a distinct role. In other worlds, a DDC node is essentially a distributedsysteminsideanode,asshowninFig.7.5. Thefollowingreal-wordcaseisusedtoillustrateoneofthejustificationsfortheaddition of complexity and costs in a DDC node. Consider a typical scenario where a solar panel is being used to charge a secondary battery. The design goal is to use the maximum amount of the energy stored in the battery. Typically, a DC-DC converter is necessary consider- ing the potential variation of the voltage level on the battery while it is being discharged. Whilethenodeisactive,thementionedDC-DCconvertercanpotentiallyachieveveryhigh levels of efficiency, such as >95%. Therefore, its use is clearly justified considering the mentioned goal. When the node enters in sleep mode, the load current drastically drops, as expected,reachingvaluesontheorderofµA.However,becausetheDC-DCconvertermust becontinuouslyactive,theoverallpowerconsumptionofthenodeiseffectivelydominated by the consumption of this converter. Unfortunately, its power consumption can be 2 or 3 orders of magnitude higher than the sleeping consumption of the MCU and adjacent mod- ules because the a DC-DC converter typically has very low efficiency when subjected to lightloads. Therefore,thedesignofthenodecanbemodifiedinordertovoidtheconverter whiletheMCUisinsleepmode. Suchsimplegoaladdssignificantcomplexitytothedesign. Because the proposed framework is founded on a mechanism that drastically changes theoperationalduty-cycleoftheDDCnode,itmustalsobeenergy-efficientwhenoperating inverylowduty-cyclewhenthesleepingenergybecomessignificant. Therefore,thedesign of this node is expected to increase in complexity. Accordingly, the next design step is to attempt to separate the power lines and avoid the use of voltage regulators in modules that are constantly powered-on. Moreover, it is necessary to discover the power needs of the different modules of the DDC node. For instance, the dynamic voltage level of the main MCUmaynotbethesameastheradiomodule,orasthereal-timeclock(RTC).Suchcom- plex scenario is better understood by means of the Fig. 7.5. For example, note that the concept of having a single shared power line for all the modules gives place to multiple andpower-controlledlines. Whilesomeaspectsofthisfigurearediscussednext,additional detailsrelatedtothesoftwaresideofaDDCnodearegiveninSection7.4. Itisimportantto rememberthattherecommendedguidelinesinthischaptercanbepartiallyfollowed. There- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 196 fore,therealizationofafullDDCnodemaynotbethegoalofaWSNdesignerconsidering the characteristics of his application and specific energy aspects. Nonetheless, some of the concepts underlying the framework and its DDC node can be borrowed and integrated on anexistingWSNplatform. Autonomousenergy-managementcontroller: thecontrolofthepowergatingprocess (activation/deactivationoftheinternalmodules[4])ofthenodeisperformedbythisMCU. Moreover,theactivationofvoltageconditioner(s)andsupercapacitorscharger(s),theselec- tion of the main energy reservoir, and the power reset of the main MCU (watchdog-timer function)aretasksprovidedbythismodule. Autonomous energy harvester: while operating in very low duty-cycles, the Main MCUcanbesleepingwhileanenergyharvestingeventismissed. Therefore,anautonomous energyharvestermodulecanbeanidealsolution. Notethattheenergyharvesterandenergy- managementcontrollerrolescanbeintegratedinthesameMCU. Main MCU: this module runs the software associated with the low duty-cycle mode in a DDC node which is mainly implemented as an application-layer overlay. Typically, the legacy platform that is being ported into the DDC node is actually called Radio Trans- ceiver module. Forinstance,considerthecaseofaTelosBnodewithTinyOS2.xandIEEE 802.15.4-based communication. In this scenario, the TelosB is not the Main MCU, but the RadioTransceivermoduleinFig.7.5. RadioTransceiver: itreferstoanyWSNnodethatisbeingportedattheDDCnode. In the case of a completely customized DDC node, the radio transceiver can be any radio that provides at least point-to-point communication. As expected, the majority of the existing WSN nodes fall into this category of radio devices. If a primary cell is used, in order to avoid the pulse effect, it is recommended that the radio module be powered via superca- pacitors. The main drawback of this approach is the significant data latency associated to thetimenecessarytochargethesupercapacitors. WhenaWSNradiotransceiverisdirectly connectedtonon-rechargeablebatteries,thelifetimeofthesecellsarestronglyreduced. Real-TimeClock(RTC):whileinlowduty-cyclemode,theRTCisusedtowake-upthe MainMCUaccordingtotheacertainscheduling. ThepowerconsumptionoftheRTCmust be significantly smaller compared to the sleeping power consumption of the Main MCU. Typically,theRTCdeviceispoweredbyanon-rechargeablebatteryinaDDCnode. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 197 Wake-Up On Radio (WOR): the WOR is an optional module that allows a DDC to quickly swap between low duty-cycle mode to regular mode. In order to be adopted in a DDC node, the WOR module must consume a very tiny power (e.g., <5µW). In [112], a nanopowerWORisreportedmakingitapotentialcandidateforWORinDDCnodes. AnalogSensor: typically, this kind of sensor requires a stabilized power supply. Also, asapassivedeviceitisnotcapabletowake-upthemainintheeventofchangesattheenvi- ronmentwhereitisinstalled. Digital Sensor: typically, this kind of sensor has its own non-customizable MCU, an internal voltage regulator, and a serial interface such as I 2 C or SPI for communication with the MCU. In general, it is not capable to wake-up the Main MCU in case of detection of events. Intelligent Sensor: this is the newest generation of sensors that has the capability to wake-up the MCU based on the analysis of an external event. Although one can customize its own MCU to achieve the above goal, recent available technology goes one step ahead and provides a way to perform continuous monitoring of an event at the cost of few µW. As expected, this scheme allows the significant reduction of the duty-cycle of the Main MCU in event-driven applications. Very recently, the manufacturer Atmel announced the SleepWalking technology which is basically the integration of the Intelligent Sensor with theMainMCUinthiscontext[113]. Although,notshowninFig.7.5,whenaWSNnodeis portedinaDDCnode,thethreementionedkindofsensors(analog,digital,andintelligent) can be attached at the Main MCU, or at the Radio Transceiver (which is the regular WSN node), or at both. These options are associated with the function of these sensors at the frameworkandalsoatthemainapplication. Multiple power lines: with this provision, each individual module can be powered on/off independently by means of the power-gating technique [4]. Moreover, only the de- vices that require voltage conditioning are connected to these special power lines. In fact, there is an underlying effort to avoid, if possible, the use of voltage regulators [4] in any module that is constantly powered, such as the Main MCU and the RTC. Nonetheless, the multiplepowerlineseffortmustbecomparedwiththesimplersolutionofputtingamodule in standby mode, if this function is available. The baseline for power consumption must be thesleepingpowerconsumptionofthemainMCU.Moreover,byreferenceofcomparisons, CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 198 Regular or High Duty-Cycle (RDC) Mode CH CH Sink To Data Server To Data Server Low Duty-Cycle (LDC) Mode Fig. 7.6 Dual Duty-Cycle (DDC) Operation: the network switches between LDC and RDC modes. InRDCmode,thenetworkmaintainsitsoriginalcharacteristics. InLDCmode,the BETS protocol becomes active, a planned network segmentation occurs, and the network achievesitsmaximumenergyefficiency. thetypicalpowerleakage(loss)duetothepower-gatingissmallerthan0.3µW. Recent technological advances point into the direction of pico and nano MCUs embed- ded in a significant number of electronic devices, many of them considered analog devices for decades. This is the case for power scavenging sources, sensors, battery chargers, and even bulb lamps. Therefore, the second foundation of the proposed framework, the adop- tion of a distributed system inside a node, is actually well aligned with the industry trends. Nonetheless, based on the energy effort tripod concept discussed in Section 7.2.2, besides thehardware,alsothenetworkandtheapplicationcharacteristicsmustalsobeconsideredin ordertoachieveatrulyfunctional,reliable,andlong-termmaintenance-freeWSNsolution. So far, this chapter gave a significant emphasis on the energy subsystem and hardware as- pects. Fromnowon,thefocuswillbemainlyorientedtowardenergy-managementsoftware mechanisms. Such software modules assume the existence of the hardware for DDC node with the features presented in this section. Important questions associated to the need of having dual duty-cycle modes and how a DDC node actually operates will be answered in thenextsection. 7.3.3 Foundation3: DualDuty-Cycle(DDC)Operation Atrulyenergy-balancedandefficientWSNsolutionsubjectedtoconstrained/irregularenergy resources depends on coordinated optimization efforts in terms of hardware, network algo- rithms,andflexibleapplicationdemands. Suchcoordinationisthemainfocusofthissection and it is provided a discussion on how a DDC system can achieve the mentioned goals. A WSN node that follows the guidelines provided in this section, that is, a DDC node, can be CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 199 designedentirelyfromthescratchor,alternatively,canbetheresultoftheintegrationofan existing WSN with additional hardware and software modules. Additional details of how to achieve a DDC node by porting an existing WSN platform are provided later in Section 7.3.5. Similarly to the hardware guidelines so far provided, one can decide to implement some of the underlying concepts in his WSN design without fully implementing a DDC operationalsystem. Intuitively, the expression dual duty-cycle operation transmits the idea of having a sys- tem that operates in low and high (or regular) duty-cycle regimes. Therefore, a natural question that raises is: Why not designing toward uniquely a low duty-cycle operation, such as <1% if it is more energy-efficient? The answer is related to two of the legs of the energy-effort tripod concept in Fig. 7.2: the network and application. Starting with the applicationconstraints,insomeWSNsystemsahighnetworkthroughputisexpected,such asinasurveillancesysteminvolvingvideoandaudiotraffic. Asexpected,thereareperiods of time when the nodes are subjected to a very high duty-cycle operation. However, it is possiblethatnotallnodesareinvolvedsimultaneouslyinahighdatatraffic. Moreover,such intensive usage of the network typically occurs in bursts, such as when an event of interest isdetected. Therefore,manyWSNapplicationsalreadypresentsomeformofdualmodeof operation. However, not all WSN solutions optimally exploit this fact in order to achieve maximumenergysavings. Theaboveexampleinvolvingasurveillancesystemisoneofthetarget-scenariosforthe proposed framework and the idea is relatively simple: while in regular mode, the existing WSN solution is fully used as it is, that is, without significant changes due to the proposed framework. Therefore, it is expected maximum network performance (as provided by the current WSN solution) with potential sacrifice on the energy performance. However, when theapplicationdoesnotrequiresuchhighnetworkthroughput,theDDCnodescanbecom- manded to switch from regular duty-cycle (RDC) mode to low duty-cycle (LDC) mode. TheoverallprocessisillustratedinFig.7.6wherethreeadditionalaspectsoftheLDCmode are also shown: the 2-tier architecture, the network segmentation, and the protocol called BETS.Theseaspectswillbediscussedlaterinthenextsections. It is important to highlight that nodes in LDC mode are not necessarily inactive or con- tinuously sleeping. Instead, such nodes are actually following a predefined low duty-cycle maintenancescheduling. AnodeinLDCmodedoesnotneedtowake-upmultipletimesper CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 200 second as usually is the case in traditional WSN protocols. In LDC mode, the nodes only need to regularly wake-up after long and continuous sleeping periods (e.g., 15 minutes) for sending measurements or for just updating their status in a central Data Server. While in themiddleofitsdeepsleepingperiod,orhibernation,anodeinLDCmodecanbeforcedto returntoitsregularorRDCmodeofoperation. Forinstance,inanevent-drivenapplication, if an event of interest is detected by one of more nodes, the network quickly resume to its maximum performance operation. Note that, the Intelligent Sensor previously discussed is animportantcomponenttoturnthisscenariofeasible. AnotherexampleisrelatedtoaWSN application for commercial buildings where the nodes employ indoors PV panels. In this case, the networkswapsfrom LDCto RDCmodeas soonone personstartshis activitiesin hisofficeoratthebuilding. Whenthehumanpresenceisnomoredetected(e.g.,duringthe night), the network returns to its LDC mode in order to save energy. In this way, many of the functionalities of this WSN application are still provided during nights, weekends, and holidays. Forsomedata-collectionmonitoringapplications,actuallythereisnoneedtoswitchbe- tweenLDCandRDCmodesbecauseitisfoundthattheapplicationunderanalysisactually does not require significant amount of data traffic and is delay-tolerant (DTN) [114, 115]. For instance, a soil moisture monitoring system, as well many other environmental moni- toring applications, have typically an application duty-cycle smaller than 1% and only re- quires measurements every 15 minutes or more. Therefore, DDC nodes in permanent LDC mode can potentially satisfy the needs of the application. Nonetheless, one can argue that a traditional WSN solution can also satisfy this application and still be an answer for the mentionedsurveillanceapplication. Consideringthisargument,doubtscanraiseinrelation to the real need of adding complexity and costs to a WSN by implementing DDC nodes. The answer lies at the extreme energy savings achieved when an node operates in LDC mode. In fact, the simulated and empirical results in Section 7.5 are related to LDC mode only. Therefore, the bigger is the period of time that a node stays in LDC mode, the higher is the energy-efficiency of the solution. While the network is operating in RDC mode, its energy-performance is essentially the same as the original WSN platform. The term essen- tiallyisusedherebecausetheenergyoverheadduetotheadditionalhardwarenecessaryfor implementingtheDDCnodeisverytiny,aswillbediscussedlaterinSection7.3.5. Inshort,thedualduty-cycleswitchingmechanismisprovidedtoobtainthebestofboth worlds(energyandnetworkperformances): CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 201 • In regular (RDC) operation, the WSN node has essentially the same performance as it would have if the framework was not applied, but with the cost of potential not optimumenergy-efficiency: Networkperformance⇑,Energyperformance⇓. • Inlowduty-cycle(LDC)operation,theWSNisdrasticallyreformulatedinaprocess similar to virtually removing all existing nodes and deploying a new network but maintainingthe same physicallayer (radio transceiver)of the nodes. The LDC mode has excellent energy performance achieved with some level of network performance sacrifice: Networkperformance⇓,Energyperformance⇑. 7.3.4 CharacteristicsoftheLDCMode This section is a very important part of this chapter because it is explained why the energy saving mechanisms of a DDC node in LDC mode hardly can be achieved in regular WSN nodes. Nonetheless, the achieved very high energy-efficiency has its price in terms of sig- nificantpenaltiesondatalatencyandthroughput. Althoughsuchdrawbacksareactuallyex- pected for a node operating in LDC node (otherwise, it would be operating in RDC mode), there are other constraints that can impact the adoption of a DDC system for all WSN sce- narios. Accordingly,adiscussiononthelimitationsoftheDDCsystemisprovided,suchas the lack of support for node mobility. Also, characteristics of potential target applications thatpermanentlyoperateinLDCmode(LDC-Onlyapplications)areprovided. Motivation To better analyze the energy savings associated with the LDC mode, two discussions are provided,onethatintroducesthepreliminaryintuitionandotherthatgivesadditionalquan- titative intuition. It is important to highlight that the following discussion is crucial for a comprehensiveunderstandingofthegoalsbehindthischapter. In a typical data-collection application, the nodes regularly wake-up, sense their envi- ronment, and transmit the sensing data to a central point. In many cases, the amount of the data is relatively small and the sampling rate is also pretty small resulting in a very low duty-cycle operation. However, this conclusion is exclusively based on the viewpoint of the application. Unfortunately, the overhead of the networking protocols can be suffi- ciently high and dominate the energy consumption of the nodes. On the other hand, if the CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 202 Table7.2Powerprofileusedinthesimulations. MCU Sensors Radio Active 5mW 30mW(5s) 70mW(3s) Sleeping 0.01mW 0.1mW 0.1mW application’s duty cycle is relatively high, the overhead of the network is typically negligi- ble. Therefore,inordertoeffectivelycomparetwosolutions,suchastwodifferentnetwork protocols, it is necessary to understand the effective network overhead caused by each one of the evaluated protocols. Besides the typical overhead added by the networking proto- col in the form over additional bits or bytes at the message payload, it is also necessary to understand the impact of that solution in the time line. Specifically, it is well known that, evenwithoutanyapplicationactivity,manyWSNssustainsomesortofnetworkinfrastruc- ture traffic to maintain the nodes synchronized, to detect the state of the nodes, and so on. Therefore,evenunderaverylowapplicationduty-cycle,theeffectivedutycyclecanstillbe relativehigh. Naturally,becausetheradioistypicallythemostpower-hungrymoduleinthe node, the term effective duty-cycle is related to the use of the Radio Transceiver module in Fig.7.5. The important question to be answered at this phase of the discussion is related to themagnitudeofsuchoverhead. Consider that an anti-mold WSN-based solution is being designed for a commercial building and the sensor nodes will be strategically installed inside the walls. Due to the fact that the nodes will be installed in areas of difficult access and without energy scaveng- ing opportunities, non-rechargeable batteries are required and, due to economical reasons, the exchange of such batteries must occur only after 5 years. Sensing measurements must occur every 20 min, but cycles of up 60min are still acceptable. The power profile of the nodes, typical WSN nodes, is shown in Table 7.2. The sensor node is composed of a pro- cessor, a sensing module, and a radio transceiver. It periodically wakes up, performs some processing (1s), takes measurements (5s), sends/receives data to/from the base station (3s), performs more processing (1s), and finally sleeps again. Assume that the communication performance of the nodes and their reliability are very high and a fixed topology for the nodesisdefined. Inthisscenario,differentstackofWSNprotocolsaretestedandanyaddi- tional network overhead is assumed to be exclusively due to the characteristics of the stack ofprotocols. Also,differentsamplingratesareused. Someprotocolsrequirethatthenodes beactiveformoretimeincomparisonwithothers. Moreover,someprotocolsareassociated withasignificantnumberofredundantdatapathsbetweentwonodes. Allthesedifferences CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 203 impact the overhead of these protocols. As a result, after careful measurements, it is found that the effective duty-cycle of sensor node increases by 0.75%, 1%, 2%, 5%, and 10% according to the choice of the protocol stack. Assuming that the initial energy of a node is 245 KJ, the goal is to run simulations that relate the lifetime of the node with the choice of the protocol stack and application duty-cycle. It is assumed that 100% of the initial stored energy is effectively used by the node (ideal hardware) because we want to focus only on thenetworkoverheadeffects. The results of these simulations are shown in Fig.7.7. As expected, when the network overhead increases, the lifetime of a node decreases. However, such impact is particularly stronger for low duty-cycles applications. For instance, if the network overhead increases from 0.75 to 5%, the life expectancy of a node decreases around 75% and 50% for mea- surement cycles of 60 and 3.2min. But caution is required in this analysis: although the difference between 75% and 50% does not seem to be very drastic, it is not actually the case. These percentages are relative to different lifetime goals and when one translates thesevaluesintoyears,itisfoundthatthelifetimeofthenodesdecreasesbyaround6years when the network overhead increases from 0.75 to 5% and 60min schedule is used. For the same increase in terms of network overhead, the lifetime of the nodes decreases by 1 yearifthe3.2minscheduleisused. Theconclusionisclear: arelativeslightdecreaseatthe performance of the stack of networking protocol dramatically impacts the lifetime of node forlowduty-cyclesapplications. Therefore,inourongoingexample,ifweoptfor60minscheduleinordertosaveenergy in term of active power, the choice of the lighter stack of protocols is the only way to make the solution feasible. Improving the quality of the solution, we can also move to measure- mentscyclesof20min,butinthiscasetheonlywaytoachievetherequiredlifetime>5years istohaveannetworkoverheadnothigherthan1%. Unfortunately,thetypicalnetworkover- head is much higher than 1%. For instance, these are some reported radio duty-cycles of MAC protocols [116]: LPL 10%, T-MAC 2.5%, S-MAC 23%, B-MAC 11.4%. Note that theoverheadduetootherprotocols,suchasnetworkandtransport,arenotincludedinthese values. Therefore, the final answer in relation to the illustrative project is that it is poten- tiallynotfeasibleusingcurrentWSNtechnology. Inthecontextofourframework,thegoal isthedevelopmentofasolutionforthisscenario: whileinLDCmode,theDDCnodemust experienceaverysmalleffectivenetworkoverhead(e.g.,<1%). CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 204 10% 5% 2% 1% 0.75% 0 1 2 3 4 5 6 7 8 meas.every 3.2min (5.5% app.duty cycle) meas.every 5min (3.3% app.duty meas.every 10min (1.7% app.du meas.every 20min (0.8% app.du meas.every 60min (0.3% app.du Network overhead added to the overall duty cycle 3.6V, 19Ah battery lifetime (years) Fig.7.7LifetimeofaWSNnode(seeTable7.2)fordifferentnetworkoverheadsandappli- cationduty-cycles. As a rule of thumb, the higher is the level of flexibility provided by a WSN protocol, thehigheristhenetworkoverhead. Therefore,intheformulationoftheoperationofanode in LDC mode, it is important to consider specific application needs and avoid unnecessary networkfeatures. Notethatinthepreviousexampleanad-hocdeployment,mobility,multi- hopping, and even collaboration are not required features. Considering that the topology is static, it is possible to organize the network into star-based segments and, at the cen- tral point of each segment, a special node (i.e., cluster heads) can collect the data from the sensors and transfer data to/from a base station or Data Server. With such ideas in mind, a 2-Tierasynchronousnetworkcomesnaturallyasafeasiblearchitecture,asshowninFig.7.8. Observe that, in terms of topology, this is the original and still dominant way to organize computers in corporative networks. Moreover, the majority of the IEEE 802.15.4/ZigBee networksalsofollowssuchscheme,sometimesaugmentedwithlow-heighttreestopologies [104, 105, 117]. In short, in order to achieve overall network overheads smaller than 1% (in terms of duty-cycle), the DDC system operating in LDC mode must adopt a very sim- ple and efficient network topology and protocol(s). How this goal is achieved and also the drawbacksofthisapproachareconsiderednext. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 205 NetworkTopologiesConstraints The DDC system operating in LDC mode is based on a cross-layer protocol called Best- Effort Time-Slot Allocation (BETS) protocol. It is implemented as an application-level overlay because this is the simplest way to have a DDC node switching between RDC and LDC modes without changing the software layers implemented at the WSN platform that is been ported. However, if one designs a DDC node from the scratch, the main function- alities BETS protocol can be potentially implemented at the data link layer. The details of BETS are discussed in Section 7.4. For this moment, it is enough to understand its general operationandtherequirementsassociatedwiththisprotocol. The BETS design is guided by simplicity behind the concept called selfish node pre- sented in [5]. A regular sensor node acts selfishly in the sense that no message relaying is actuallyperformedbythenode. Asthenameimplies,thisideacomesintheoppositedirec- tionin relation to ad hoc networksand manycurrentWSNs with emphasis on cooperation. The sensor node, which is called End Device (ED) simply wakes-up, takes measurements, and sends the data to a specific collecting or cluster-head (CH) node. After successfully sendingitsdata toCH, theED receivesanacknowledgmentfromCH withthe precisetime for the next cycle and goes to sleep. In order to avoid the need of message relaying among ED nodes, the network is segmented and each segment has a star-like topology with a pre- defined maximum number of EDs. The CH can communicate with all EDs in that segment anditsroleissimilartothetypicalsinkattheWSNliterature. Themethodofthecommuni- cationbetweenCHandacentralpointisacompletelyopenaspectinthisframework. Itcan beimplementedbymeansofanywirelesstechnology. Itisalsopossibletocreateanetwork ofCHsinvolvinglong-rangelinksamongthemandselectingoneofthemasaBaseStation (BS) which would be in charge of sending the data packets to a Data Server (DS). Such very flexible architecture is possible because BETS is an asynchronous protocol in relation to the message delivery between ED and DS. Once the ED’s data reaches the CH node, the transaction is concluded from the viewpoint of the ED node. When and how CH compacts theEDmessages,ifthisisthecase,tosendtoDSisalsoanopenimplementationaspect. The network architecture supporting BETS (called Ripple-2 in [5]) is also hybrid: any wireless communication technology can be used provided that a simple point-to-point link can be implemented. For instance, instead of using an off-the-shelf WSN node as the Ra- dio Transceiver for the DDC node (vide Fig. 7.5), one can adopt 900-MHZ point-to-point CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 206 !"# ! "$$" Fig.7.8DifferenttopologiesforthenodesinLDCmode[5] radio modules without any networking layer besides the physical layer. Moreover, distinct network segments can adopt different wireless link technologies. Similarly, the technology usedfortheED-CHlinkcanbedifferentfromtheCH-BS,CH-DS,orBS-DSlinks. There- fore, the BETS protocol is a proper choice for the LDC node: it is very flexible, open to integration,andintuitivelyitseemstohaveaverytinyoverheadduetoitssimplicity. Some of the possible topologies for the nodes in LDC mode are shown in Fig. 7.8. Note that the maximum height of the network tree is 2 which highlights the fact that, with time, many enhancementscanbeeasilyaddedtothesolution. Basedontheanalysisofthisfigure,some limitationsorconstraintsoftheproposedsolutioncanalsobeidentified: Constraint 1: No support for mobility: node mobility is not supported by this solu- tionbecauseafixedandwell-planedtopologyisassumedaprioriforeachnetworksegment. When the ED node wakes-up, it simply takes measurements and transmits the data. There is no provision for network setup phase or any kind of search for the location of the CH node: the ED node simply sends the data and expects that CH receives and acknowledges thatmessage. Therefore,amobilenodecouldnotfullyimplementthejustdescribedselfish node behavior. In a DDC system with LDC and LDC modes, the mobile nodes can only communicatebepartofthenetworkwhenthisoneisoperatinginRDC(regular)mode. Constraint 2: Not all network topologies are supported: some physical network CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 207 ! ! " # $ % ! & ' ("! $ # Fig. 7.9 The characteristics of the network while in LDC mode: the BETS protocol is de- signedtoprovidehighenergy-efficiencyforbothEDandCHnodes. topologies can impede the DDC to operate in LDC mode. For instance, consider a line- fashion deployment, such as a sequence of sensor nodes in a bridge. In this case, it is very hardtoimplementlogicalstar-topologiesunlessthecommunicationrangeofanodeencom- passes a significant number of nodes at both directions of the line which is usually not the case. Consequently, another protocol must replace BETS in order to include some level of light collaborationamongEDsinLDCmode. Constraint 3: CH-DS/BS links can pose challenges: when a DDC system is operat- ing with both RDC and LDC modes, all nodes typically use the same wireless technology, such as IEEE 802.15.4 PHY (e.g., TelosB-based WSN). When the networks is switched fromRDCtoLDCmode,thenewformednetworkobeysapredefinedschemewithmultiple star-based segments. The assigned CH nodes now need to send the data to DS by means of the same low-power, short-range wireless links. Therefore, the network must be divided in a significant number of segments and the collaboration among CHs is potentially required becauseonlyshort-rangelinksareavailable. Theproposedframeworkdoesnotofferguide- lines for the CH/BS or CH/DS communication, nonetheless, another potential solution in thiscaseistheadoptionoftworadiomodulesatCH,oneuseonlyforCH-BS/DSlink. TargetApplicationsforLDC-OnlyMode The energy efficiency of the LDC mode is achieved with the cost of network performance penalties in addition to some topology constraints. In fact, we must have in mind that the adoption of the complete proposed framework is actually prevented in some cases. On the otherhand,sometheLDCisidealforsomeapplicationsandtheRDC/LDCswitchingdoes CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 208 not even to be implemented: the network would be always is LDC mode in this case. Such scenario occurs when the WSN application is actually a low duty-cycle data-collection, is delay-tolerant,anditdoesnothavetosupportmobilenodes. Inshort,threecasesinrelation to the application of the DDC system proposed by the framework can occur: a) for many regular WSN applications, the DDC system must operate in dual mode (RDC/LDC), b) for manylowduty-cycledata-collectionapplications,theDDCsystemonlyneedstooperatein LDCmode,andc)forsomefewapplications,theDDCsystemcannotbefullyemployed. Becausethe LDCoperational mode isclosely associated withverysmall networkover- head, the goal of maintaining the DDC system uniquely in LDC mode is actually the best optionintermsofenergy-managementoptimization. However,justbecausethemainappli- cation is low duty-cycle it does not imply that dual mode operation can be removed. There are some factors that prevent the adoption of a LDC-only mode and, thus, the LDC/RDC switching must be preserved. One reason is related to the characteristic of the application: itmustfollowschedules,suchasadata-collectionone. Also,theapplicationmustbedelay- tolerant and without issues to follow a 2-Tier architecture. These reasons are explained in detailnext. DatalatencyinLDC-Onlymodecanbeaproblemforsomeapplications. Thecomplete messagedatalatencyofanED֌DSmessagecanbeontheorderofsecondsorevenmore. Thelatencyfortheotherway(DS֌ED)isevenworseandcanbeontheorderofminutes. Fortunately, typically for a data-collection application, the ED֌DS message direction is theoneofinterest. Eveninthiscase,thereasonsforasignificantdatadelayare: • Once a measurement-cycle starts, the ED node must wait for its time to transmit the data. • Once the data from this ED node arrives at CH, it is necessary to wait until CH con- cludesthedatacollectionfromallnodesatthatcycleinordertohavethetransmission toDS. • At the end of each cycle, CH must activate its CH/BS or CH/DS link to perform the data transfer. Such activation delay can be significant. For instance, to activate a SMS-based (text), the SMS modem can take almost 1 min to conclude such link activation. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 209 Table7.3TargetApplicationsforLDC-OnlyMode Designspace Option Goal Sense-only Time Periodicdatacollection Deployment Plannedapriori Topology Static DataRate Low Delaytolerance Mustsupport • If multiple CHs send data to a shared device (BS), the delay due to the coordination amongCHsandalsoduetotheactivationoftheBS/DSlinkcanalsobesignificant. A second aspect to be evaluated before deciding by the LDC-Only network operation is related to the energy requirements for the CH nodes. The energy consumption of CH follows,atthebesttheoreticalcase,alinearrelationwiththenumberofEDsinthatnetwork segment. Therefore, the lifetime of CHs can be strongly impacted if the WSN design does not carefully consider the energy challenges in CH nodes. The increase of the number of segments (thus, also the number of CH nodes) can be a technique that alleviates the energy workload on each assigned CH node. The dynamic sharing of the CH role among nodes in the segment, such as in a round-robin fashion, can also be a solution. Finally, the adoption ofahybridpowersourcesystemforCHsbasedonenergyscavengingandnon-rechargeable batteriescanextendthelifetimeofCHs. TheoverallcharacteristicsofaDDCsystemoper- atinginLDCmodearesummarizedinFig. 7.9. Notethattwoofthehighlightedchallenges inthisdesignapproachistheneedofsomeformofpowerhibernationforbothEDandCH nodes and also an efficient time synchronization technique for the ED/CH communication. Both aspects are considered in our previous works: the former one, involving hardware techniques, is considered in detail in [4] and the latter one is presented in a short-format in [5]. The detailed discussion of the BETS protocol, the core component of a DDC system, is discussed in detail in Section 7.4, where its algorithms are presented. The performance evaluationofBETSisprovidedinSection7.4. 7.3.5 DDCSystem: ImplementationGuidelines This section is very important from the engineering perspective. Its main goal is to clarify what are the steps necessary for the partial or full implementation of the proposed energy- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 210 management framework. Also, it is a good place to summarize the exposed concepts and acronyms specifically introduced in this chapter, as shown in Table 7.1. In relation to the fullimplementationoftheframework,thereareatleast3importantscenariostobeanalyzed: Case 1: Node in LDC-Only mode and using ordinary radios: Low duty-cycle data collection application is being used. This application is delay-tolerant and does not have mobilitysupport. Inthiscase,theDDCsystemneverswitchestoRDCmodeandthisisthe simplest scenario to implement. The ordinary radios provide a simple point-to-point com- munication link and, in general, there is no need of further configurations for these radio modules. Case2: NodeinLDC-OnlymodeandusinglegacyWSNnodes: Lowduty-cycledata collection application is being used and the system never switches to RDC mode. It is nec- essarytoconfigurethelegacyWSNnodetoprovideasimplepoint-to-pointcommunication with the minimum possible networking functionalities. Any feature or service beyond the physical and medium access control (MAC) layers are unnecessary and, even worse, can potentially increase the network overhead of the solution. Therefore, such features must be permanentlydeactivated. Forinstance,atargetlegacyWSNnodecanhavehiddendynamic topology control and time-synchronization features that are not necessary for this scenario and must be deactivated. Good ways to discover if this is the case are: a) to observe how quickly two WSN modules exchange messages just after they are initialized and b) to use a radio frequency (RF) sniffer device to monitor the content of the packets. For the latter case, low-cost monitoring tools for the 2.4GHz ISM band are available, such as the device in [118] which is used in our ongoing project. If it is not possible to completely deactivate unnecessary features in legacy WSN nodes, maybe the better option for this specific sce- narioistoexchangethembyordinaryradiomodules. Case 3: Node in dual mode and using legacy WSN nodes: This is the general case wheretheWSNapplicationismorestrictinrelationtonetworkQoSmetrics. WhileinLDC mode,itisnecessarytoconfigurethelegacyWSNnodsetoprovideasimplepoint-to-point communication with the minimum networking functionalities, as discussed above. How- ever, when the node returns to RDC mode, such networking features at the legacy WSN nodes must be activated again. Therefore, the integration effort in this scenario involves the addition of an application-layer module at the legacy WSN module to dynamically ac- tivate/deactivatehigh-levelnetworkingfunctionalitiesofthedevice. Iftheunnecessaryfea- turesinlegacyWSNnodescannotbedeactivatedwhilethenodeisoperatinginLDCmode, CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 211 theoptimumenergyefficiencyprovidedbytheBETSprotocolcannotbeachieved. AtypicalexampleofadevicewiththeroleofDDC’sRadioTransceiver(Fig.7.5)under Case 1 is XBee-Pro 802.15.4 Series 1 (Digi Inc.). For the Cases 2 and 3, the open-source TelosBmoteandthecommercialXBee-PROZBZigBee(DigiInc.) aregoodexamplescon- sidering their popularity, documentation, market availability, and ease of integration. Note thatallthesedevicesoperateat2.4GHzbandandareIEEE802.15.4PHY-compliant,which isthegeneraltrendforcurrentWSNs. The following are implementation guidelines, labeled as Gx that apply for all 3 men- tionedcases,unlessexplicitlyindicated: G1 To implement a DDC node, as illustrated in Fig. 7.5, it is necessary to have the MainMCUseparatedfromtheRadioTransceiverinall3mentionedcases. TheBETSpro- tocolimplementationsoftwareresidesattheMainMCU. G2 When an energy scavenger is employed, the Energy Management Controller is separated from the Main MCU. In this case, it runs the software module associated with local energy-management decisions. One of these decisions is the selection of the energy reservoir used to power the Radio Transceiver or any other power-hungry module. More- over, the charging of supercapacitors are directly controlled by this module, as well the power-gating of many modules. When an energy-harvester is not used, the functions of the Energy-ManagementControllercanbeabsorbedbytheMainMCU. G3 ForCase3,theMainMCUisnotdeactivated(orcontinuouslysleeping)evenwhen the node is in RDC mode. This is the case because to power on/off any module, such as a digital sensor connected to the legacy WSN node, there is an energy-management hier- archy to be followed: Radio Transceiver⇒ Main MCU⇒ Energy-Management controller ⇒ Power-Gating device. However, the Main MCU typically is sleeping for the majority of time when the node is in RDC mode. In this case, the legacy WSN node can wake-up the MainMCUbymeansofaninterruptionline,asshowninFig.7.5. G4 In general, for Cases 1 and 2, the sensor devices are physically connected to the MainMCU.ForCase3,theyareconnectedtotheMainMCU. G5 The power hibernation refers to a continuous and long period of sleeping time CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 212 defined in term of minutes or hours [4]. While hibernating, it is possible to shutdown the majority of the modules of the node which leads to significant energy savings. The draw- backofthisapproachistheresultingdelaytohavethenodereadytoresumeitstasks. G6 The operational schedule of the node in LDC mode is defined by a centralized energy-management module running at the Data Server and this information is sent to the nodes via BETS protocol. For Cases 1 and 2, that is LDC-only mode, such scheduling in- formation is typically defined in term of minutes, such as cycles of 15 or 20 minutes for someenvironmentalapplications[20]. Notallnodesatthesegmentneedtofollowthesame schedule. Infact,heterogeneousschedulingisoneoftheprovisionsofacentralData-Server basedenergy-managementtoevenlybalancetheenergyresourcesinthenetwork. ForCase 3, the schedule of the nodes in LDC mode is potentially defined by the Radio Transceiver, whichisthelegacyWSNnode. Inthiscase,thevalueoftheLDRcycleismainlyassociated withtheexpectednetworkQoSmetricsandthecharacteristicsoftheapplication. G7 The Wake-Up On Radio (WOR) module is connected to the Main MCU. Its func- tionistowake-upthenodewhenitishibernatinginLDCmodeandaneventsofinterestis detected. Withoutsuchprovision,thenodeinLDCmodecouldnothaveitshibernationstate aborted due in case of a critical event. For instance, consider a surveillance system under the Case 3 and assume that the nodes are hibernating while in LDC mode. One intelligent sensorattachedtoaspecificsensornodedetectsthepresenceofanintruderanditpromptly awakes the Main MCU of the node where it is installed. However, the major challenge is to wake-up all EDs of the same segment and eventually all the network in order to switch backtoRDCmodeandprovidetheexpectedfunctionalitiesofthesurveillanceapplication. Therefore, the key-answer to aDDCess this challenge is the WOR module: once the first node is awaken by its intelligent sensor, it transmits a special Beacon by means of the Bea- con TX module shown in Fig.7.5. This beacons triggers, at least, the WOR module of the CH node of that segment. In turn, the CH wakes-up all EDs using the same technique. An ultra-lowpowerWORisarecentandsophisticatedtechnologyanditsuseisrecommended forevent-drivenscenarios. G8 For Cases 2 and 3, one preliminary and critical test to be realized is related to the feasibilityofusingthelegacyWSNnodeastheRadioTransceivermoduleoftheDDCnode. Itisnecessarytoperformapoint-to-pointcommunicationwithtwonodesbymeansoftheir serial ports because this is the typical way that the Main MCU communicates with the Ra- dio Transceiver. For instance, for TelosB, there are the pins UART0RX and UART0TX at CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 213 ! " # $ % &'('))* Fig. 7.10 The cross-layer nature of the BETS protocol: a candidate of choice for the LDC mode. the TelosB’s expansion connector that allow such test. Similarly, the XBee modules pro- vide the serial ports TXD and RXD for the same purpose. For the integration of any WSN nodethatsupportspuread-hoccommunication(i.e.,withoutinvolvinganyformofnetwork hierarchy), the process is relatively straightforward. However, legacy WSN nodes that are nativelybasedoninfrastructuredtopologies,suchasXBee-PROZBZigBee,requirespecial attention. For this specific example, typically the assigned CH nodes are the ZigBee nodes defined with the ZigBee Coordinator or Router profiles and the assigned ED nodes in our frameworkarethenodeswiththeZigBeeEndDeviceprofiles. G9 The BETS protocol assumes that the network segmentation is already in place when it is in operation. It means that for the Case 3, an application-layer software running at the nodes in RDC mode must logically configure the network segment. One ED node canonly belogically attachedto asingle segmentand eachED nodehas assigned alogical aDDCess which is used by BETS to identify each ED. If two nodes with the same logical aDDCess, but at distinct network segments, are at the communication range of one of the CHs, potential errors will occur. The simplest solution is to assign a distinct RF channel for each segment, in particular if they are physically close to each other. However, if the frequency-hopping spread spectrum (FHSS) technology is being employed by the nodes, further investigation is necessary to avoid the mentioned problem. Moreover, for Case 3, it is possible that the network segment of the node in RDC mode does not match with the BETSsegmentinLDCmode,asinthecaseshowninFig.7.6. Thisproblemcanpotentially occur when ZigBee-based nodes are used. In this case, the switching between operational duty-cyclemodes must be precededby a dynamicsetup of the WSNmodules. Forthe Zig- Bee example, potentially the BETS segmentation will be achieved by configuring channel frequency,PANid,orboth. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 214 7.4 Cross-LayerProtocolforVeryLowDuty-Cycle(LDC) Mode Fromnowon,thischapterassumesthattheDDCsystemisoperatinginLDCmode. There- fore,thenetworkhasalreadybeenlogicallysegmentedinmultiplesets,eachonewithaCH node and its associated ED nodes. In this section, we focus our attention is one of these networksegmentsanddiscusshowexceptionalenergy-performanceispossiblewithaDDC node running the Best-Effort Time Slot Allocation (BETS) protocol. BETS is an example ofacross-layerprotocolthatiscompliantwiththegoalsoftheproposedframework. How- ever, one can design a similar cross-layer protocol that can also be efficiently used by the nodes in LDC mode. The section starts with an overview of related work in this context, followedbyanoverviewofBETSanditsownterminology. Next,thedesigngoalsofBETS arepresentedandthefunctionaldetailsarediscussedbymeansofitsalgorithms. 7.4.1 RelatedWork The virtual elimination of collaboration among wireless nodes is not a novelty. The IEEE 802.15.4 standard [104] was introduced as a low-rate, short-range communication solution for Wireless Personal Area Networks (WPANs). One of the network topologies defined in thisstandardisastartopologywhereaPANCoordinatorisinchargeofthecommunication withtheremainingdevices. Similarly,theBluetoothtechnologyisbasedonastartopology with a master node as the central point [119]. Note that such arrangements are similar to the relation CH-EDs in the BETS solution. In fact, the design of the network architecture associated with BETS, Ripple-2, was influenced by these standards and their outstanding success. IEEE 802.15.4 only defines the specifications in relation to the physical and MAC lay- ers. However, upper ISO/OSI layers can be optionally used to allow ad-hoc deployments, multihopping, trees, and mesh networks. One example of such augmentation is the ZigBee standard [105] which defines the network, application, and security layers. Although ini- tially a WPAN solution, the functionalities of ZigBee become pretty similar to the ones in traditional WSNs. As expected, many WSN deployments based on ZigBee devices have beenreported[17],[119],[117],[120]. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 215 In this context, BETS can be seen as an effort to add scalability and extreme energy efficiency to IEEE 802.15.4 (star topology mode) solutions without incurring in the higher overhead and complexity of the ZigBee standard. Also, the overhead similar to the one as- sociated with the PAN association procedure [104] in the 802.15.4 standard does not exist intheBETSsolution. Moreover,BETSisalsodesignedtosupportanyunderlyingpoint-to- pointphysicallayer,notonly802.15.4PHY.Infact,theED-CHlinkimplementationisnot even limited to a radio operation. Finally, differently from a PAN coordinator (or ZigBee router,orZigBeecoordinator),theCHnodeintheBETSsolutioncansleepand,evenbetter, it can hibernate. This fact drastically reduces the energy requirements of the data-collector devicewhichstillisachallengefortypical802.15.4-basedsolutions. A simplified version of WSN based on multiple stars is presented in [121] for forest monitoringbutthedetailsrelatedtotheunderlyingnetworkingprotocolsarenotprovided. A hierarchicalarchitecturefordelay-tolerantnetworksispresentedin[122]andacustomized MAC protocol called LiteTDMA is employed. Hardware specialization of WSN nodes, in particularwiththeintroductionofthepower-gatingtechnique,isproposedin[123]anditis extended in [4]. Slot-based MAC implementations have been proposed, such as, TRAMA [124], PMAC [125], Z-MAC [126], and H-MAC [127]. Although BETS is not a MAC protocol,itscorefunctionalitiesinrelationtothetime-synchronizationamongnodesofthe samenetworksegmenthavesomesimilaritieswiththementionedMACprotocols. To the best of our knowledge, the work in this chapter is the first to propose a non- collaborative model for WSNs by means of the implementation of the selfish node concept discussed in Section 7.3.4. As already discussed in the previous sections, such low-energy modelisexclusivelyadoptedwhiletheDDCnodeisinLDCmode. Evenforsomeapplica- tions,thenodescanpermanentlyoperateinsuchLDCmode(whereBETSlies). 7.4.2 ProtocolOverview BETS is a novel cross-layer protocol implemented as an application-level overlay. It is de- signed for low data rate, low duty-cycle, and sense-and-send WSNs. When the DDC node operates in LDC mode, BETS is the protocol of choice. BETS operates at the MAC and upper-levelnetworkinglayers. IfanexistingMACprotocolalreadyexistsinthetargetsen- sor platform, the MAC functionalities of that platform can be disabled or simply ignored if CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 216 Fig.7.11BETSfunctionality(EDside): animplementationoftheselfishnodeconcept. notcausingsignificantoverhead,asdiscussedinSection7.3.5. Inotherwords,theultimate actionsnecessarytoachieveafair,contention-free,andreliablecommunicationchannelare takenbyBETS.AsshowninFig.7.10,BETSassumesthataperiodicsensingapplicationis in place which matches with the way the network behaves while in LDC mode. The proto- colhasaprovisiontocapturetheschedulingdatasentbythemainapplication(DataServer) to the nodes and defines the proper allocation of the wireless channel in the time domain. In this sense, the term schedule refers to the same object for both application and network discussions. AlsoshowninFig.7.10,theenergyefficiencyoftheprotocolismainlyachievedbysac- rificing the network performance in terms of data latency. Therefore, the main application must afford such higher delay which can vary from seconds to hours according to the fi- nal implementation. No routing-related functions are actually provided by BETS and it is assumed that the network is divided into multiple star-like segments, following an asyn- chronous2-tierarchitectureapproach. Another assumption shown in Fig.7.10, static topology, calls our attention to the fact that BETS protocol cannot be easily modified to support mobile nodes. There are optional assumptionstobeconsideredintheBETScontext. AsshownattherightsideofFig.7.10,if theultimatedesigngoalistoachieveveryhighenergyefficiency(i.e.,morethan1orderof magnitude compared to state-of-the-art solutions), the combination of very low application duty-cycles (i.e, <1%) and power hibernation techniques [4], [123] is a required step. As alreadydiscussed,aDDCnodehassuchcapabilities. The adoption of star-like segments potentially reduces the complexity of a network design. However, it is important to verify if the center point (access-point, controller, or CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 217 cluster-head) does not become the real bottleneck of the solution. For instance, the energy issues related to the CH node in WSNs have been studied for a long time and a CH role- rotation scheme has been proposed [110]. In the architectural context where BETS is im- plemented, such role rotation is not easy to be implemented. Therefore, in order to achieve energyefficiencyforbothEDandCHnodes,anon-trivialsolutionisrequiredandthisisthe mainchallengeoftheBETSdesign. Infact,besidestherole-rotation,wedidnotfindinthe literatureanapproachtoefficientlyreduceenergyconsumptionofCH-likenodesinastatic topology. In BETS, we have the goal of having the energy profile of the CH node linearly followingtheaverageenergyprofilesoftheEDnodesinthesegment. From the ED’s viewpoint, the fundamentals goals of BETS are the implementation of the selfish node concept and an efficient way to avoid that two or more EDs try to use the communication channel simultaneously. Because the current MAC protocols do not fully implement the selfish node concept, BETS must have MAC-related functions in order to fulfill the mentioned goals. The messages exchanged by ED and CH nodes are shown in Fig.7.11. Fromtheviewpointoftheselfishnode,theprocessoccursasfollows: (a) EDnodewakesupandtakesmeasurements. (b) Without any channel negotiation, ED sends the ED_MEAS message to the CH node (unicast). Thismessagebasicallycontainsthemeasurementsandfewcontroldata. In somescenarios,insteadofindividualsensingmeasurements,theED_MEASmessage can contain aspects related to the status of the node, compressed data derived from a historicalsequenceofmeasurements,etc. (c) Without any channel negotiation, CH sends back a CH_CTRL message to the ED node(channelbroadcast,logicalunicast). Thismessagecontainsthescheduleforthe nextcyclerelatedtothatnode. (d) Without any channel negotiation, ED sends a ED_CTRL message to the CH node (unicast). Thismessageacknowledgesthereceptionoftheschedule. (e) ED node configures its wake-up circuitry accordingly to the received schedule and sleeps. Asexpected,nocollaborationamongnodesexistsandtheimplementationoftheproto- col becomes significantly simple at the ED side. In fact, such simplicity clearly indicates CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 218 the proper realization of the selfish node concept. One can argue that even the CH_CTRL andED_CTRLmessagescanbeeliminatedforafullimplementationofaselfishnodecon- cept. However,thereliabilityofBETSrequiresthatsomesortofminimumcommunication- quality control exists, as will be explained later in this section. In fact, the small overhead associatedwiththesemessagesbecomesirrelevantinlowduty-cycleapplications. From the CH’s viewpoint, the implementation of the BETS protocol is not so straight- forwardasintheEDcase. BesidesthepropersupportforEDs,itisimportantthatCHsleeps in optimum cycles. Different schedules for the nodes of the same segment can potentially cause energy inefficiencies for the CH node. Even a global schedule may not provide the best energy performance for the CH node. For instance, assuming only 5min application cycles, one ED can follow the sequence 0-5-10-15.. in the line of time, while another ED that was initialized later follows the 3-8-13-18.. sequence. In this case, both EDs have the same schedule but the CH node cannot take a longer sleep while it is cleat that it would be possible. Therefore, BETS must provide a way to accommodate schedules in the best efficient way possible, both for ED and CH nodes. In fact, the first reason for the term Best-Effort in the BETS acronym is related to this aspect. To correct the mentioned issue, which hereafter is called dispersion, BETS adjusts the first cycle the second node which was turned-on at momentt =3. Therefore, this node will have this sequence under BETS: 3-5-10-15.. Note that the first programmed cycle, and only this one, was adjusted in order to group all EDs with the same schedule. It is worth to highlight that such adjustment is fundamental for saving energy at the CH side while such mechanism does not affect the EDs. Inotherwords,BETSisadispersion-freeprotocol. ThesecondreasonfortheBest-Effort termisrelatedtothereliabilityofthesolution. In order to achieve very high end-to-end reliability metrics (in this case, for the ED-CH link), multiple handshake messages may be necessary. However, for every active node, BETS provides a time-slot with a small and fixed length. Bigger and/or dynamic slot lengths can be adopted to increase the communication reliability, but there are energy penalties to be considered. In our simulations to be shown in section 7.5, different communication chan- nel error rates are analyzed under BETS in order to evaluate energy and reliability metrics. Also, in one of our current BETS implementations, the average ED-CH distance is around 210m and for that case, the total data loss was smaller than 1.8% during multiple weeks. The slot length was parametrized to be large enough to allow just a single additional round of ED_MEAS - CH_CTRL - ED_CTRL messages if necessary. Doing so, the solution be- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 219 Fig. 7.12 An example of regular (no errors) operation of BETS from the CH’s perspective. AtinactiveMTSs,allnodes(EDsandCH)aresleeping. However,theCHnodecanstilluse aninactiveMTSfortheCH-BS(orCH-DataServer)communication. came more reliable, but less energy-efficient and clearly the data latency of the solution increases. The bottom line in this discussion is that the provision of large time-slots (mul- tiple messages in sequence) in order to increase the reliability may not be really necessary. Nonetheless,duetotheBETSflexibility,theslotlengthcanbemodified. 7.4.3 Definitions&Terminology Before proceeding with a detailed explanation of BETS, some terms shown in Fig. 7.12 mustbeproperlyintroducedorbetterdefined. Inaddition,somecontextualaspectsaredis- cussedinordertoeasetheadoptionofBETSfortheLDCmodeofDDCsystems. Logical network segment (or simply, segment): a fundamental assumption for BETS is that the network is divided into logical clusters or segments. This division is realized considering the physical topology of the network. Accordingly, it is expected that the CH node be located at the center of a virtual circle where all nodes inside that circle are able to communicate with that CH (unit disk graph approach). Realistically, the communication range of the ED nodes will significantly vary due to many reasons. Moreover, the location ofthenodesmustbeprimarilygovernedbytheapplicationneeds. Therefore,itisveryhard to achieve an ideal division of the network into circles that do not overlap. Due to this fact, additionalcommunicationtechniquesmustbeemployedtoenforcethatanodesolelycom- municateswithasingleCHnodeevenifmorethanoneCHcanbereached. Forinstance,by CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 220 using different channels/frequencies or even by using software filters, it is possible to deal with the overlapping circles issue. This enforced concept of segmentation divides the over- all network into logical network segments: ED nodes of the one (logical) network segment canonlycommunicatewiththeCHnodeofthatsegmentandvice-versa. CH’schildren: all ED nodes of the same networksegmentassociated with a certain CH nodearechildrenofthatnode. Registered children: when CH communicates with DS, the latter can potentially send explicit information about the number of children EDs and their respective schedules. In this case, the ED nodes are considered registered children and the CH can properly calcu- late how much time to spend waiting for the contact of an ED node based on the number of registered children. Such information is easily available in planned deployments with a static topology and can also be modified to reflect possible node failures. The number of registeredchildrenisnotrequiredforthefunctionalityofBETS,butitincreasestheenergy efficiencyoftheCHnode. Major Time Slot (MTS): once CH is initialized (boot), the line of time is divided into fixedperiodsoftimecalledMTSs,eachonewiththelength mtsLength,asoftwarevariable expressed in units of seconds. In our implementation and also in the simulations, the value for mtsLength is 300s (5min.). In order to achieve a collision-free solution and maximum energy efficiency, it is assumed that the application schedules are also given in mtsLength units. Active and Inactive MTS: the CH node does not have to be necessarily active all time, thatis,foreveryMTS.Asexpected,duringsomeMTSs,theCHnodeissleepingbecauseall its children are also sleeping and such MTSs are called inactive MTSs. When CH is ready to hear an ED node, the respective MTS is called an active MTS. The active MTS (AM) is divided into three sequential parts with variable lengths: ETS, BTS, and STS, as explained next. ED Time-Slot (ETS): it refers to the initial part of an active MTS (AM) which is used specificallyforcommunicationwithchildrenEDs. ThedynamiclengthofETScorresponds atleasttothesumoftheassignedtime-slotsfortheactivechildrenatthatMTS.OneBETS CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 221 algorithminvolvingregisteredchildrenisusedtodeterminetheoptimumlengthofETSfor eachAM.ItsultimategoalistoallowCHtosleepassoonaspossible. BSTime-Slot(BTS): it refers to the second part of an active MTS which is used for the CH-BS communication. The length of a BTS varies as a function of the amount of data to be sent to the BS node, CH-BS link throughput, and possible errors at this link. To save energy, ED data from distinct active MTSs can be aggregated. Doing so, the BTS length is 0 for the majority of MTSs and it is maximum for few MTSs. Although the CH-BS data transfer can be divided into multiple MTSs, a BTS transaction cannot conflict with an ETS ofanactiveMTS.However,forinactiveMTS,theBTStransactioncanlastlonger,asshown inFig.7.12. Finally,itispossibletochangetheCH-BStransferschedulingaccordingtothe energy performance metrics at the CH node. Due to the asynchronous nature of the second networklayer(CH-BS),thecommunicationwiththeBSmustnotimpacttheBETSperfor- mancefortheEDnodes. Sleeping Time Slot (STS): it refers to the third part of an active MTS which is actually notbeingused. Duringthisperiodoftime,theCHisinactiveandpotentiallysleeping. Time Slot (TS): the time-slot as seen by each individual ED. The TS has a fixed and unique length tsLength for each segment, a software variable expressed in units of sec- onds. Such parameter basically corresponds to the time necessary for a single ED_MEAS- CH_CTRL-ED_CTRLtransaction,asillustratedinFig.7.12. However,inpractice,tsLength isalittlebiggeranditisinfluencedbymanyfactors,suchasthecharacteristicsoftheED’s radiotransceiverandwirelesschannel,theuseofapower-gatingtechnique,numberofpos- sibleretransmissions,etc. Inourimplementationandalsointhesimulations,tsLengthis8s (1retransmissionissupported). When theretransmissionis notnecessary,which isusually the case, the corresponding reserved time period at the end of a TS window provides a gap betweenTSsandpotentiallymitigatesdriftclock,multi-pathwaves,andchannelcontention issues. Homogeneousscheduling: it refers to the scenario where all ED nodes of the same net- worksegmenthaveexactlythesamesleepingschedule. Heterogeneousscheduling: it refers to the scenario where at least two ED nodes of the CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 222 same network segment have different schedules. A network with multiple segments can havehomogeneousandheterogeneousschedulingschemesatthesametime. Dispersion: in heterogeneous scheduling, the dispersion is defined as an anomaly char- acterizedbyhavingsomeEDnodesusingimproperMTSs,causinganegativeimpactonthe energy-efficiencyoftheCHnode. Inotherwords,thegoalofhavingthemaximumnumber of inactive MTSs is not achieved although it is possible. As already discussed, BETS is designedtobedispersion-free. Emergency Mode (EM): in normal BETS operation, all EDs in a segment are properly synchronized with the CH node. That is, they correctly follow their assigned time-slots. However, when a node is a) deployed for the first time, b) restarted, or c) does not receive CH_CTRLaftersendingED_CTRLmessage,itdoesnothaveanytime-slotassignment. In thiscase,thenodefollowsadifferentalgorithm(EM)inordertocommunicatewiththeCH node. Convergence: when all ED nodes in a segment are in regular operation (not in EM), that network segment is said to be convergent. In a non-convergent segment, one or more nodescantrytocontactCHwhileitissleeping. Also,thenodeinEMcaninterferewiththe current assigned TSs and that segment is no more contention-free (temporarily). While in non-convergentstate,thenetworkhassignificantenergypenalties. Therefore,adesigngoal forBETSistohavesegmentsthatquicklyconverge. 7.4.4 DesignGoals ThemaingoalsoftheBETSprotocolaresummarizedasfollows: (a) Tobefunctionalinrelationto: a)ProvideawaytosendfixedordynamicschedulesforEDnodesthatareoriginated atthemainapplicationrunningattheBSnodeorabove(DataServer). b) Collect sensing and basic control data (network-related errors and energy-related metrics)fromEDnodesandsendsuchdatatothemainapplication. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 223 (b) To implement the selfish node concept and maintain fairness for the wireless channel access. (c) To be energy-efficient (ED side) by imposing a network overhead not higher than 0.75%eveniftheprobabilityoferrorsatthecommunicationchannel(ED-CHlink)is highas5%. Thisoverheadlimitismotivatedbytheanalysisofthescenarioillustrated inFig.7.7specificallyrelatedtoverylowdutycyclesense-and-sendapplications. (d) To be energy-efficient (CH side) by allowing CH to have optimum sleeping cycles evenincaseofheterogeneousscheduling. (e) Toisolatenetworkproblemsbetweensegments. (f) ToisolateproblemsrelatedtoED-CH,CH-BS/DScommunicationlinks. (g) Tosupportnetworkmanagementtasksasfollows: a)verifythereliabilityofindividual EDnodesandtheED-CHcommunicationlinksandb)isolateerraticEDnodes. (h) TomitigatethewirelesschannelcontentionamongtheEDnodesofthesamesegment. BETS must be insensitive to the existence or not of a MAC protocol running above thephysicallayer. (i) Toprovidesupportforoptionaluseofpowerhibernationschemesinordertoachieve veryhighenergysavings. Observe that no specific attention is given to network performance metrics, such as maximum transmit delay or throughput. This fact anticipates the most important trade-off ofBETS:thenetworkperformanceisexpectedtobesacrificedinordertoobtainimpressive energy achievements in conjunction with excellent scalability and reasonable reliability. The main reason for this trade-off is the isolation between ED-CH and CH-BS/DS data- flows (asynchronous approach). Therefore, the possibility to switch from LDC (BETS) to RDCmodesisanecessaryprovisionforcriticalWSNapplications. 7.4.5 BETS:NormalOperation Inordertorealizethedesigngoals2(selfishnodeconcept)and8(contention-free),aTDMA approach is used to avoid contention among nodes and allow a fair usage of the wireless channel. Byreservingatime-slot(TS)foreachEDnodethatwillusethesamefutureactive CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 224 MTS (AM), the wireless channel becomes potentially contention-free. When an ED wakes upinitsassignedTS,itimmediatelystartssensingand,withoutanydelay,sendsthedatato CH.Observethatthissense-and-sendapproachprovidesthebestenergyefficiencypossible because no care is given by the ED node related to the possibility of a busy channel or the availabilityofCH.Clearly,itcorrespondstotheselfishnodeconcept. Infact,thisprocedure isautonomouslyrepeatedbytheEDnodefromtimetotime,eveniftheCHnodeisnotop- erating. In normal operation, once ED_MEAS is received, it is followed by the CH_CTRL message sent by CH. The CH_CTRL message has two purposes. First, this message ac- knowledgestheproperreceptionofED_MEAS.Second,CH_CTRLcontainsconfiguration datatoED,suchasthenexttimethatthenodemustbeactive. OnceEDreceivesCH_CTRL message, it sends back the ED_CTRL message. This message also has two goals. Besides serving as an acknowledgment for CH_CTRL, the ED node uses this message to send its control (log) data: battery status, power shortages, communication errors, etc. At the BS side (or above), the main application can recognize erratic patterns associated with a spe- cific node. By assigning a very long (e.g., hours or days), the erratic node can be virtually isolatedfromthenetwork. The ED_MEAS + CH_CTRL + ED_CTRL transaction forms the core of a handshake- based procedure in BETS. However, in contrast with the traditional usage of acknowledg- ments, the missing of one of the messages does not necessarily trigger a re-transmission. In our implementation of BETS, we provide a second transaction round at the same TS if the first one fails. This explanation helps to clarify why the default tsLength used in our simulations (and real implementations) is 8s while the associated messages only sum up to 3s,accordingtotheTable7.2. Besidestheretrytiming,tsLengthmustcapturetheinvolved timeouts associated with collisions and other communication errors. Observe that such re- dundancy provision is not a formal specification of BETS but tsLength can be increased even more to support multiple retries. Moreover, one critical network segment can have a highertsLengththanthedefaultvalueinordertosupportmorethan2transactionroundsfor eachED. In our real-world BETS implementation, all the 3 mentioned messages are sent twice with a small delay between the messages. Empirically, we figured out that such provision highlymitigatesthepossibilityofamissingmessageinparticularforoutdoors. Thissecond formofredundancyprovidesawaytoincreasethereliabilityofthecommunicationwithout havingtointroducetimeoutsoradditionalcomplexityattheprotocol. Again,sucheffortisa CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 225 designpossibility,butnotaspecificationofBETS.Nonetheless,thisdiscussionisimportant tohighlightthestrengthofBETSintermsofitsadaptabilityfordifferentnetworkscenarios. The acknowledgments provided by CH_CTRL and ED_CTRL messages are primarily used as a network management tool. In other words, it is possible to identify energy and communication problems related to a certain node or a group of nodes. Also, the commu- nication quality is evaluated in both directions. This later aspect is very important because CH and ED nodes may have different antennas. For instance, an omni-directional one for CH and a directional one for EDs. Alternatively, a higher antenna gain (typically a bigger antenna) for CH and a regular one for EDs. The bottom line is that, in all cases, the goal 7 (network management) is fully satisfied in BETS. That is, it is possible to detect and cor- rect reliability issues at the network already deployed. On the other hand, the 3 mentioned messagesonlysatisfythegoals2and8whenthesegmentisoperatinginnormalconditions, that is, it is convergent and all nodes are very well synchronized. Erratic scenarios under BETSareconsiderednext. Algorithm5EmergencyMode(implementedatEDnode) Require: maxContinuousEM:max. timeinEMmodebeforerebooting Require: currTime: timeelapsedsincethebeginningoftheactiveMTS Require: randomTime1: (randomvaluebetween5to11)*tsLength Require: randomTime2: (randomvaluebetween300to720)*tsLength Require: SCchargingTime: timetochargeSCsinEMmode(default:0) Require: tryouts: incrementedforeachunsuccessfulCHtransaction Ensure: AsegmentwithupN=0.8∗(mtsLength/tsLength)EDsconverges tryouts←0 ifcurrTime>maxContinuousEM then Reboot node else ifsupercapacitor(s)is(are)usedthen chargesupercapacitor(s)forSCchargingTime endif iftryouts>5then tryouts←0 backoffTime←randomTime2 sleepduringbackoffTime period else backoffTime←randomTime1 idle(radiooff)duringbackoffTime period endif endif CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 226 7.4.6 BETS:DealingWithErraticScenarios So far, the energy-efficiency, fairness, and collision-free channel characteristics of BETS are achieved provided that all nodes properly have and follow their TS assignments (con- vergence). In this section, the erratic scenarios are considered and the BETS convergence processisdiscussed. The first erratic scenario is related to the well-know clock drift issue. Even when all EDs are properly synchronized, the minimal differences among their internal clocks will eventuallycauseoverlappingbetweenTSs. BETSsolvesthisproblembycontinuouslypro- viding a schedule adjustment for each ED node. Such adjustment occurs every time an ED node receives a CH_CTRL message. Therefore, in general, the clock drift hardly impacts thesolution. However,thecombinationofvery-longschedules(e.g.,>5h)andlow-quality clock modules must be avoided in order to mitigate the risk of clock drift issues. If such schedulesarereallyexpected,twoguidelinescanbeemployed. Firstly,higherqualityclock systems with temperature compensation can be used at CH and ED nodes. Alternatively, a highervaluefortsLengthcanbeusedinordertoenhancetheeffectivegapbetweenconsec- utive time-slots. Note that this parameter is a key one for current and future enhancements ofBETS. The remaining erratic scenarios are basically associated with the same final result and theycanbeanalyzedinasinglescenariounderBETS.Specifically,nomatteriftheED-CH transactionfails(collisionorothercommunicationerror),orCH/EDnoderestarts,oranED nodeisrecentlydeployed,inallcasesthenetworktemporarilyisnon-convergent. Whenan EDnodetimeoutsthereceptionofanexpectedCH_CTRLmessage,itautomaticallyenters inEmergencyMode(EM).ArelatedsolutiontypicallyusedinMACprotocolsisthecombi- nationofchanneloverhearingwithrandomback-offs. InBETS,onlythesecondpartofthis technique(randomback-offs)isused. Thereasonforthisapproachisrelatedtothesupport ofhibernationmodeatEDs,asstatedinthegoal9. The fixed-length nature of a TS (deterministic) highly promotes the adoption of super- capacitorsat theED nodesas partof thehibernation solution. However, because thepower usedtooverhearawirelesschannelisveryhigh,itwouldbenecessarytochargesupercapac- itors with an amount of energy multiple times the expected one for a single TS transaction, thus resulting in drastic energy inefficiencies. Because this design aspect is closely related CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 227 to the hardware characteristics, we empirically evaluated different power management so- lutions for typical WSN nodes and we concluded that instead of overhearing the channel, a more energy-efficient solution would be using the back-off technique alone. The EM procedure,withanoptionalsupportforsupercapacitors,isprovidedbytwoalgorithms(Al- gorithms5and6),onefortheEDnodeandtheotherfortheCHnode. In order to provide support for different hardware platforms, some of the parameters in the Algorithms 5 and 6 are clearly hardware-dependent. Algorithm 5 is executed once ED nodeentersinEManditisre-executedforeachunsuccessfultryoutwhileinthatmode. To generate randomTime1 and randomTime2, a discrete uniform distribution function is used. Algorithm 6 is executed once CH node starts a new active MTS and it is re-executed each timeanewED_MEASmessageisreceived. Inthischapter,wewilllimittheverificationof thecorrectnessofthesealgorithmsbyanalyzingthesimulatedresultsinSection7.5. Inour simulations and context, the goal is to have network convergence achieved in less than the time of (4∗sched min ), where sched min is the smallest of the schedules among the ED nodes thatarealreadysynchronizedwithCH.IfallnodesareinEM,sched min isthedefaultsched- ule used by CH temporarily without children. Because such values are on order of many minutes, it is clear that when a dual system (LDC/RDC modes) is used, the frequency of theswitchingprocessisclearlyimpactedbythelongtime(e.g.,>15min)necessarytocon- vergethenetworkunderBETS.Ifthismodeswitchingoccursfewtimesperday,thisisnot an issue. However, the need of more frequent switching requires a supported not natively supportedbyBETS.Onepotentialandrelativelyeasywaytomitigatethisconvergencetime issue in dual systems is to sequentially add EDs to the segment (rather than all nodes at the sametime)whentheDDCsystemswitchesfromRDCtoLDC. The most important aspect behind these algorithms is the fact that the convergence can be achieved with any number of ED nodes provided that the CH is active for enough time. Because the maximum continuous time the CH can wait for children is limited by mt- sLength, this parameter essentially governs the convergence feasibility. However, tsLength isultimatelytheparameterthatinfluencesmtsLength. Forinstance,withatsLength=8sand 100 ED nodes, the minimum mtsLength is 1000s (16.6min). As a result, the application schedule must be an integer multiple of this value, such as 20, 40, 60min, and so on. In ourimplementation,weoptedbyamaximumnumberof30nodes,leadingtomtsLengthof 5min,whichismoreusefulfromtheviewpointofmanylowdutycycleapplications. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 228 Algorithm6EmergencyMode(implementedatCHnode) Require: expectedNumChildren: #registeredchildren(N) Require: numChildrenCurrentMTS:#pendingEDswithTSsinthisMTS Require: numFutureChildren: #EDswithassignedTSsinfutureMTSs Ensure: AsegmentwithupN=0.8∗(mtsLength/tsLength)EDsconverges return waitEDTime: howmuchtimeCHmustwaithearingchildren if CHjustbooted then waitEDTime←20∗mtsLength else ifexpectedNumChildren=0(CHnevercontactedBSnode)then ifnumChildrenCurrentMTS=0(noassignedTSinthisMTS)then waitEDTime←31∗tsLength else waitEDTime←tsLength∗(numChildrenCurrentMTS+5) endif else{(BShadalreadysentthenumberofEDsinthissegment)} aux←numChildrenCurrentMTS+numFutureChildren ifaux=expectedNumChildrenthen [(convergenceachieved)] waitEDTime←tsLength∗(numChildrenCurrentMTS+2) else{(potentially,oneormoreEDsmissinginthisAM)} waitEDTime←tsLength∗(expectedNumChildren−numFutureChildren+2) endif endif endif CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 229 Algorithm7Dispersion-freeTSAllocation(CHnode) Require: schedED_Table: ED schedules assigned by main app (BS side) fields: (ED_id, x), where xisanintegermultipleofmtsLength Require: MTS_Table: MTS allocation table (old, current, and future AMs) fields: (ED_id, MTS_seq),whereMTS_seqisthenextAMforED_id Require: fixedDelta: readinesstimeforCH(fromwake-uptohearEDs) Require: currMTS:currentvalueofMTS(numberedsequentially) Require: currED:idoftheEDonthecurrenttransaction Ensure: TSoverlappingdoesnotoccur Ensure: Dispersion-freeTSallocation: maximumenergyefficiency return adjSch: nextEDactivationtime(tobesentviaCH_CTRL) //dispersion-free: x←schedED_Table[ED_id] aux← floor(currMTS/x) nextMTS←(aux+1)∗x updateMTS_Table(ED_id,nextMTS) //noTSoverlapping,continuousTSsassignment: adjSch←nextMTS∗mtsLength adjSch←adjSch+fixedDelta numEDsSameMTS←getNumEntriesSameMTS(MTS_Table,ED_id) etsAdj←(numEDsSameMTS−1)∗tsLength adjSch←adjSch+etsAdj CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 230 WithsmallervaluesoftsLength,muchmorenodescanbesupportedinasegmentwhile practical values for mtsLength are still maintained. Therefore, one natural question is as- sociated to the adoption of a high value for tsLength (e.g., 8s) when it is well known that typical RF transceivers can realize the full transaction in less than half second. Besides the additionaltimeforapossiblere-transmissionandforasecuregapbetweenTSs,theanswer lies on the choice of the hardware platform. In our implementation, a single transaction was actually achieved with less than 2s and finally 8s was determined as a secure value based on experiments and taking into account our transceiver choice, power-gating latency [4], reliability aspects, and critical outdoor environments. The latter aspect is important for our ongoing project because we would like to achieve ED-CH distances very close to the informed by the manufacturers of the radio modules. When the messages in the network are separated by significant gaps, the multi-path effects in hilly regions are significantly reduced, as we later proved empirically. Nonetheless, a higher tsLength typically will not affect the energy-efficiency of EDs provided that network errors do not occur frequently. However, a higher tsLength can definitely impact the CH node because it must be active moretimeduringeachcycle. ThenextsectiondiscusstheenergyefficiencyoftheCHnode. 7.4.7 Energy-EfficiencyoftheCHNode ThemaximumenergyefficiencyattheCHsideisachievedwithahomogeneousscheduling. In this case, the nodes of one network segment always use the same AMs whenever they are expected to sense-and-send and the goal of achieving the maximum number of inactive MTSs is trivially achieved. Although this fact does not represent any additional energy- relatedadvantagefortheEDnodes,itdefinitelyextendsthesleepingtimeoftheCHwhich isthenodewiththemostcriticalenergyconstraintinthenetworksegment. Moreover,con- tinuous and long sleeping periods maximize the efficiency of the power-gating technique becausethereareenergypenaltiesassociatedtotheswitchingtransients. However, considering that an energy-aware system running is on top of the BS/DS (whichisaguidelineofthisframework),suchsystemcanpotentiallyselectdifferentsched- ules for EDs. In this case, the already mentioned dispersion problem can occur and must be aDDCessed by BETS. Although the schedules are sent by the main application running on the BS side, due to its cross-layer nature, the CH is effectively the node that controls the distribution of the schedules to its children. Based on this observation, it is possible to CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 231 implementanalgorithmattheCHsidethatcanproperlyadjust(advancingordelaying)the next active-time information (via CH_CTRL message) sent to each ED in order to avoid dispersion. Such approach is used by the Algorithm 7 which is a dispersion-free procedure fortheCHnodeprovidingmaximumenergyefficiencyforthisnode,asstatedinthegoal4. TherearesomeinterestingaspectsthatneedtobehighlightedintheAlgorithm7: • ThetimeoftheentirenetworksegmentisreferenceduniquelybytheCHclock. When the CH is initialized, the moment t 0 is defined and the MTS 0 has its beginning. No real-time information is exchanged through BETS: when an ED node transmits ED_MEAS, CH timestamps it with a real-time value based on its local clock. This schemehelpstomaintainBETSverylightbut,asexpected,measurementswitherrors on the order of seconds can occur. Fortunately, it is rarely an issue for non-real-time applicationsrunningintheLDCmodeofthenetwork. • The least common multiple (LCM) principle is used to avoid dispersion. For in- stance, if currMTS is 13 and an ED has an assigned schedule of 5 (final schedule: 5∗mtsLength), then aux in the algorithm is floor(13/5) = 2. Because newMTS = (2+1)∗5=15,thisnodewillwake-upagain2MTSsahead(MTS15=13+2),not atMTS18. Thisadjustonlyoccursatthefirstassignedcycleforthatnode. Fromthat momenton,thenodecontinuesfollowingtheexpectedscheduling(15,20,25,..). • AftercalculatingthenewAMforthenode,anadjustmentisperformedbyCHinorder to avoid TS overlapping. In other words, if 3 nodes have the same cycles (schedules) tooperate, we do not wantthat all of them send their messages at the same time, that is, at the beginning of that AM. To avoid this issue, the number of current EDs that share the same future AM is recorded. Every time CH allocates a new ED for that AM,itaddsaspecificdelayinordertohaveallthenodesaccessingCHinanefficient, sequential, and contention-free manner. The maximum energy efficiency is achieved fortheCHnode. • Thealgorithmisimplementedwithoutanyhistoricalcontrolorcomplexprocedures/data structures. Such simplicity provides the path to implement the CH side of BETS in hardware platforms already used by regular WSN nodes. This aspect is extremely CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 232 important in a dual mode system (with RDC/LDC modes) because the CH nodes are assigned among regular WSN nodes. If the CH requires a very powerful Main MCU (whichisnotthecase),theassignmentofCHswouldbeimpacted. 7.5 SimulatedandEmpiricalResults(LDCMode) In order to verify the correctness of the BETS algorithms and also to determine the con- straints of the solution, experiments are performed. In addition to simulations, preliminary results from our field deployments are provided. The following are some questions of the particularinterestinthiscontext: • Arethealgorithmscorrectandisgoal3(EDenergy-efficiency)feasible? • AssumingthatCHwasjustinstalled,howmuchadditionaltimeisnecessarytoachieve convergence? • GivenacertainprobabilityoferrorsforthecommunicationchannelbetweenEDand CHnodes,whatistheassociatedenergypenalty? • Whenthenetworkwillnotconverge? 7.5.1 ExperimentalSetup Forthespecificworksinthischapter,wedevelopedanetworksimulatorforBETS(Matlab environment). A single segment is simulated, but an any number of EDs is supported. Due to its asynchronous nature, we omit the CH-BS communication (BTS length=0). Each in- dividualsimulated scenario involvesa minimumof 1,000 iterations. In relation to commu- nication errors, a uniform distribution is considered. Also, such probability is independent amongnodesincreasingthelikelihoodoferrorsinthenetwork. Theprobabilityofcommu- nication channel issues is independent of the probability of collisions, better representing a real scenario. Finally, for the simulations involving 1min-schedule, CH is assumed to not sleep for practical reasons. Specifically, the energy cost associated with the very frequent activation/deactivationofthe(on/offtransients)ishigherthanthesleepingsavings. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 233 Table7.4Defaultparametersforsimulations. ED_MEAStime 1s CH_CTRLtime 1s ED_CTRLtime 1s ED_MEAStimeout 3s ED_CTRLtimeout 2s randomTime1 5..11s randomTime2 5..12min measurementstime 1s maxTimeStartLastED 5min SCchargingTime 11.40s expectedNumChildren valueofN tsLength 8s maxContinuousEM 1h When not specified, the parameters used in the simulations are the same ones used in our real-world implementation of BETS also discussed in this section. Such parameters are listed in Table 7.4. The parameter maxTimeStartLastED requires an explanation. When convergence tests are realized, the EDs are randomly turned-on during a certain amount of time and maxTimeStartLastED represents a limit for this time. For energy-related calcula- tions, Table 7.2 is used and a 3.6V, 19Ah non-rechargeable battery is assumed as the single source of power. Accordingly, many results are given in terms of life expectancy for the node and one can simply convert back to energy values in Joules assuming that 100% of the initial energy is actually used by the node. This assumption is realistic if the power matching technique presented in [4] and supported in this framework is used. With such technique,thecurrentpulseeffectdiscussedinSection7.3isavoided. Byassumingtheuseofnon-rechargeablebatteryasthesinglepowersource,itbecomes straightforwardtoveryifthesolutionwouldreachaverylonglifetimeindependentlyofthe existence of an energy scavenging system for the nodes. Similarly, such estimated lifetime considers the LDC-only mode. Therefore, the energy costs associated with a more critical WSN application (RDC mode) is not computed. In short, both user-defined energy un- knownsfromtheviewpointoftheframework,thepossibleamountofharvestedenergyand the energy spent in a demanding WSN application, are removed in this analysis. However, in a real-world implementation, such information naturally must be included according to the available energy resources and application demands. In our current project, the nodes are being deployed with non-rechargeable batteries and the system is currently LDC-only due to the low duty-cycle characteristics of the application. In this case, the estimated val- uesforlifetimeprovidedinthissectionareexactlytheonesusedintheenergyanalysisand decisionsunderthisproject. Relatedtotheempiricalinvestigation,theBETSsolutionhasbeenimplementedineight CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 234 10 20 30 40 50 0 15 30 45 60 75 Time to achieve full convergence (minutes) Number of ED nodes 1minsched 5minsched 10minsched 20minsched 2hsched Fig. 7.13 Convergence time assuming that all EDs are turned-on randomly during a 5min- period. distinct outdoor networks since Aug 2011. 6 adjacent sites have a total of 150 nodes (76 already deployed) covering a 36km x 36km continuous area). The results here are from the two oldest networks in terms of operation time. The first network features 1 BS, 1 CH, and 26 ED nodes. The CH and EDs are attached to 3 soil moisture sensors. The maximum ED-CH distance is around 150m. The irregular topography and the existence of obstacles (treesandplants)arethemainchallengesforthissite. Thesecondsite,acowfarm,iscom- posed of 21 ED nodes and the maximum ED-CH distance is around 350m. Previously, we had faced problems in this site related to a) solar panels vs. cows, b) rechargeable-batteries vs. extremetemperatures,c)complexsupportforunattended802.15.4-basedrouters,andd) scalability/overhead issues associated with the ZigBee protocol and sparse networks [17], [5]. Accordingly,theBETSdesignwasstronglyinfluencedbythelessonsofthatwork. 7.5.2 PerformanceEvaluation Inthissection,theexperimentalresultsarediscussedinrelationtothreeaspects: a)conver- gencetime,b)impactofcommunicationerrors,c)impactofheterogeneousscheduling. Convergence time: In Fig.7.13, the convergence time is given as a function of the CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 235 numberofnodesfordifferentnode’sschedulesandahomogeneousschedulingisassumed. Also, only communication errors due to the contention collisions are considered. The con- vergence value here represents the additional time, besides the schedule value, for the last ED node to converge. Because all the nodes are randomly turned-on, such scenario repre- sents the worst-case which is associated with a stronger channel contention for the initial moments. Fordualmodesystems(LDC/HDC),suchconvergenceimpactcanbeeliminated byprovidingaprogressiveactivationofthenodes,asalreadydiscussed. For clarity, only the variance of the 20min-schedule case is presented. As expected, higher duty-cycles aggravate the convergence. For instance, a segment with 30 nodes takes around 15min to achieve convergence if 20min-schedule is used. However, if a 5min- schedule is defined for the nodes, the converge can take around 35min. Similarly, a higher number of nodes impact the convergence time. A segment with 20 nodes typically con- verges in less than 15min even if a 5min-schedule is in place. On the other hand, for the same schedule, it is possible that a network with more than 30 nodes only achieve con- vergence after hours. It is important to highlight that if CH randomly restarts, a scenario similar to this simulation will occur. Therefore, such convergence analysis provides impor- tant insights for the parametrization of BETS in a given segment. In our many empirical tests (both sites: 20min-sched), we never observed a convergence higher than 1h and the averagevalueforconvergenceisfoundtobearound40min. Suchvalueisclosetotheupper boundofthevariancelinefor20min-cyclesinthefigure. Because,thissimulationdoesnot includeanycommunicationchannelerrorbesidesthecollisions,thetheoreticalmodelhasa fairagreementwithourempiricalevaluation. Impact of communication errors: Once the convergence at the network segment is achieved,BETSbecomeshighlydeterministicandit’stheoreticalnetworkoverheadisneg- ligible (e.g.,≪ 0.75%) assuming low duty-cycles applications and no communication er- rors. However, for the next simulation, we want to verify if this design goal number 3 also holdswhentheprobabilityofcommunicationerrorsishigh(e.g.,5%). Itisassumedthatthe networkhasalreadyconvergedand30EDsfollow20mincycles. InFig.7.14,theadditional network overhead is given as a function of the channel error probability. This overhead represents the additional number of messages due to the retransmissions and also possible additional collisions. These secondary collisions occur because a non-convergent node can transmit at the TS assigned to other node. As shown in the figure, even a small error rate, suchas1%causesanoverheadof16.3%. Ahighererrorrate,suchas10%,doublesthenet- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 236 0.1 0.5 0.75 1 2 3 4 10% 0 25 50 75 100 Probability of errors at the comm. channel (ED CH link) Additional network traffic (%) Total application+BETS duty cycle: 0.53% 2.8% 8.3% 16.3% 11.6% 26.2% 41.1% 52.3% 111.8% * Simulation based on 20min homogeneous scheduling with 30 ED nodes * Default hardware parameters (Table 1) * Starting point: network already converged lifetime reduction: 38.8% (6.4 to 3.9yr) lifetime reduction: 8.5% Empirical (site#2) lifetime reduction: 46.4% Empirical (site#1) lifetime reduction: 3.4% Fig. 7.14 Impact of the communication channel error (ED-CH link) on the BETS network performance. work traffic. However, when we calculate the effective duty-cycle of BETS for this worst errorcase,itturnsouttobearound0.53%andthedesigngoal3issatisfied. If the mentioned error rates are not temporary, a strong lifetime reduction for the nodes will occur. This result highlights that the lack of extensive error control in BETS has the clear trade-off of making the protocol very sensible to communication channel errors. The lifetimeimpactisslightlysmallerthantheadditionaltrafficrate,butstillveryhigh. Forour empirical investigation, we analyzed the data that arrived at the BS side. Because it is pos- sible to easily detect duplicates, missing data, and the number of retries that each node had experienced,anaccurateenergyprofileforthenodesisfeasible. Theresultsofsuchanalysis arealsoshowninFig.7.15butnowincludingtheoptionsof20and40nodesinthesegment. It is cleat that the mentioned trade-off is aggravated with a higher number of nodes. For instance,considerthe4%errorratecase: thenetworkwith40nodeshasitslifeexpectancy shortenedbymorethan1yearcomparedwiththecasewherethenetworkonlyhas20nodes. Therefore, the significant effect of the channel error on the BETS’ performance is strongly aggravated when more nodes are involved. It is explained by the best-effort approach used by BETS to achieve convergence: the process is relatively slow when the number of EDs is relatively high. In other words, it can takes many minutes for a node that experienced communication error to converge again and, while in EM state trying to contact CH, it is CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 237 wastingenergy. Returning to the analysis of Fig.7.14, it also has empirical results to be discussed. For sites #1 and #2, the average error rates of around 0.6 and 11%, respectively, are calculated based on measured network metrics. These results are in agreement with the fact that site #2 has very high ED-CH distances. Now, using Fig.7.15, it is possible to infer about the expected lifetime of the nodes. For instance, for the site #2 (21 nodes, 11% average error rate),itispossibletoforecastalifetimeofaround4years. ForanexistingWSNsolutionto be superior in terms of energy-efficiency, it needs to have a total network overhead smaller than1.8%(videFig.7.7). Todate,thesite#2hasbeenincontinuousoperationfor15months and,sofar,thesolutioniscomparabletoanyexistingsolutionthathasaneffectivenetwork overheadoflessthan5%. Theseresultsareprettysignificantconsideringthecoveragearea and the number of nodes involved: it is a sparse network and many WSN protocols po- tentially would fail in such site [5]. Moreover, just a single CH node is being used and the energyconsumptionamongEDnodesisguaranteedtobehomogeneous,assumingthatthey havethesameapplicationscheduleandthesameaveragecommunicationerrorrate. Finally, although the energy costs are significant with high communication error rates, our imple- mentationofBETSprovedtoalsohavegoodcommunicationreliability: thedatalossesfor sites #1 and 2 are smaller than 1.4 and 1.8%, respectively. These numbers are relatively superiorthantheaverageforoutdoorsdeploymentsreportedintheWSNliterature. Impactofheterogeneousscheduling: Sofar,thesimulationsconsideredhomogeneous scheduling. However,insomescenariosanadaptivesensingschedulingisdesired[17,111]. Inthenextsimulation,theenergyperformanceofBETSunderhomogeneousandheteroge- neousschedulingisevaluated. InFig.7.16,therelativeenergyconsumptionoftheCHnode for a 24-hour period is shown as a function of different scheduling schemes. The reference case is a 20min-schedule involving 30 nodes. The goal is to estimate how much energy (additional or reduction) is associated if the scheduling scheme changes. The first 3 left- most cases involve the same 20min-schedule and 10, 20, and 30 nodes. Next, 15 nodes in a 10min-schedule. The following case is special: 20 nodes in a heterogeneous scheduling, half following a 10-min schedule and half a 20-min schedule. The last case (rightmost), involves only 8 nodes but with a more frequent 5-min schedule. It is assumed a convergent anderror-freesegment. The number of ED_MEAS messages received per hour by the CH node is already cal- CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 238 0.1 0.5 0.75 1 2 3 4 10% 3 3.5 4 4.5 5 5.5 6 6.37 Probability of errors at the comm. channel (ED CH) link Expected node lifetime (years) 40 ED nodes 30 ED nodes 20 ED nodes Fig. 7.15 Impact of the communication channel error (ED-CH link) on the energy perfor- mance. SimilarscenarioofFig.7.14fordifferentnumberofnodes. culated. NotethatinFig.7.16,thesamenumberofED_MEASmessagesreceivedbyCHin threedistinctschedulingschemes,90meas/h,doesnotimplythesameenergyconsumption of the CH node. The underlying factor that mainly governs the energy performance of the CH node is the number of AMs for a certain period of time, such as 24 hours. If homo- geneous scheduling is in place, the optimum scenario in relation to the number of AMs is achieved. However, for heterogeneous scheduling, the number of AMs can potentially in- creaseandthecalculationoftheenergyspentbytheCHnodeismorecomplex,asexplained next: • IneachMTS,thereisanextratime(dynamicallyvaluedeterminedbysoftware)allo- cated at the ETS as a provision for possible communications failures. The higher the number AMs, the higher is the sum of these time provisions spent by CH and worse isitsenergyefficiency. • Before the ETS beginning, the CH node must be ready for the radio reception and a certain amount of such readiness time is provided. In our implementation, such CPU time has an average of 30s (it also performs sensor measurements) and the cor- responding consumed energy is relatively high. In short, the higher is the number of AMs,thehigheristheenergyspentwiththisprocessingphase. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 239 54.8% 27.4% REF +17.8% +58.8% 30 meas/h 60 meas/h 90 meas/h 90 meas/h 90 meas/h 96 meas/h 10 EDs 20 EDs 30 EDs 15 EDs 10+10 EDs 8 EDs Homogeneous 20minsch Relative energy consumption 10minsch 10/20min 5min-sch Fig.7.16CHenergyconsumptionforhomogeneousandheterogeneousschedulings. • Hardware state transition: the higher is the number of AMs, the higher is also the energy cost due to the transients while turning-on the modules [107]. In particular, suchcostiscriticalwhenpower-gatingtechniquesareused. The basic scenarios associated with an increase of the number of AMs are illustrated in Fig.7.16. An interesting aspect to highlight is related to the three leftmost cases using the same 20-min schedule. A linear relation exists in relation to the number of nodes and energy consumption. However, there is a hidden fixed energy-cost in all cases and it is re- latedto the establishmentof AMs. In thiscase, thenumber of AMsis exactlythesame (72 per day) and the mentioned fixed cost is the same. This analysis indicates that the higher the number of EDs sharing the same AM, the higher is the CH energy efficiency. The next two cases involve 15 nodes (homogeneous scheduling) and 10+10 nodes (heterogeneous scheduling). The number of AMs involved in both cases is the same (144 per day). The heterogeneous scheduling is not adding more AMs because 20 is a multiple of 10 and all AMs related to the nodes that follow the 20-min schedule are also shared with the nodes that follow the 10-min schedule. Therefore, the heterogeneous scheduling is not imposing an additional cost comparing these two cases. Finally, if we compare one of these two just mentioned cases with the one with 30 nodes, the number of ED_MEAS messages received byCHperhourisexactlythesame(90). However,thenumberofAMsforthecasewith30 nodes is smaller (72 per day) and this fact explains why the CH is consuming less energy compared to the other case. In short, to save energy at the CH side, the number of EDs cannot be very high, and similarly for the number of AMs which is directly related to the CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 240 Primary Cell Long-Term Energy Repository Energy-Management Framework Energy-Management Subsystem Distributed System Inside a Node Main MCU separed from Radio/WSN Module Wake-Up On Radio (WOR) Intelligent Sensor LDC RDC Network Segmentation Scheduled Sense-Send-Sleep 2-Tier: CH and ED nodes LDC-Only Mode Dual Mode Operation Dual Duty-Cycle (DDC) System BETS Protocol Energy-Management (Central Data Server) Fig.7.17ProposedEnergy-Framework. employedschedulingscheme. 7.6 Discussion Anopenenergy-managementframeworkforWSNsisproposedinthischapterwithastrong emphasis at the realistic achievement of a functional and reliable solution with a very long maintenance-free lifetime (e.g., >5 years) for the nodes. By means of a detailed and sys- tematic preliminary analysis, it is shown that energy scavenging systems are typically nec- essarytoachievethisgoal. However,inordertoalsoincreasethereliabilityofthesolution, a long-term energy repository is recommended. The traditional candidate for this role is a rechargeable battery but considering the mentioned very long-term lifetime, a primary cell isabetterchoice. Besidestheselectionoftheproperenergyresourcesforthenode,itisalso very important to control the energy used by the nodes and in the network in general. An excellentstrategytoachieveanenergy-efficientsolutionistobalancetheeffortsatthenode, network,andapplicationlevels. Thelatterkindofeffortisrealizedbymeansofanenergy- management system hosted in a central Data Server. By energy measurements received from the nodes, or by estimation techniques, or both, the centralized energy-management systemcanevaluatethecurrentandfutureremainingenergyatthenodesand,dependingon theapplicationQoSmetrics,itcantypicallyactivateanddeactivatesensingnodesbasedon location and time. In this chapter, the emphasis is given on the energy-management efforts atnodeandnetworklevels. CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 241 It is proposed that multiple MCUs compose the sensor node, that is, a distributed sys- tem inside a node. With the technological advances, such proposal does not imply a higher energy consumption but a better way to control individual modules of a node although adding complexity and costs to the WSN project. Inside this node architecture called Dual Duty-Cycle (DDC) node, the traditional WSN node is no more considered the Main MCU ofthenode,butsimplyaRadioTransceivermodule. Moreover,besidesthementionedMain MCU and Radio Transceiver modules, the energy-management subsystem also has its own MCU. In this context, recent availability of intelligent sensor probes are also discussed and thiscomponentcanrepresentanotherMCUinsuchDDCnode. Besides the focus on energy-related hardware aspects, this chapter also provides guide- linesrelatedtotheenergyeffortsatthenetworklevel. ADualDuty-Cycle(DDC)systemis proposed as part of the framework. From the viewpoint of a DDC system, low duty-cycles (LDC)data-collectionapplicationsareclassifiedasbelongingtotheLDC-Onlyclass. How- ever, many WSN applications are not LDC-Only and they are classified as regular or high duty-cycle(RDC)applications. Inordertoachievemaximumenergyefficiency,thenetwork must operate in LDC mode the majority of the time, if possible. LDC-Only class of appli- cations permanently satisfy this goal. However, for RDC applications, it is still possible to have the network operating temporarily in LDC mode and returning back to RDC mode. This is the case, for instance, of indoors WSNs deployed in office buildings and operating during non-office hours. In order to save energy during this period of time, the network can operate in a form of stand-by mode and this is actually what the LDC mode provides for the network. As expected, the network eventually will return to the RDC mode and the mechanismsoftheDDCsystemprovidesuchoperationalmodeswitching. While operating in LDC mode, the nodes are actually running the code at the Main MCU,notthecodeatthelegacyWSNnodes. Suchcodeincludesthenovelcross-layerpro- tocol called Best-Effort Time-Slot allocation (BETS). Under the BETS protocol, the nodes assume a different logical topology according to the BETS logical network segmentation andits2-Tierarchitecture. Bysimulationsandempiricalresults,BETSisprovedtobevery energy-efficient and typically has an effective network overhead smaller than 1%. Its main drawback is a low data throughput which is the reason why this solution is not applied for the RDC or regular mode of operation of many WSN applications. Nonetheless, because it is possible to switch between LDC and RDC modes, many current WSN platforms can CHAPTER7. ENERGY-MANAGEMENTFRAMEWORKFORSPECIALWSNS 242 still be enhanced with the proposed energy framework. For instance, when it is allowed to have the network temporarily operating with low data throughput, the LDC mode can be activatedandthenetworkisthencontrolledbytheBETSprotocol. Inthisway,thenetwork canexperienceitsbestenergyperformanceforthedurationoftheLDCperiod. Chapter8 AchievingHighSystemReliability AsmentionedinChapter7,theunattendedWSNsoftheSoilSCAPEprojectareamongthe largest existing outdoor networks (in terms of node sparsity) and they are operating contin- uously for more than 2 years. All regular sensor nodes employ non-rechargeable batteries andthefailurerateduetosoftwareandelectronicissuesissteadilyzeroduringtheyears. A veryhighreliabilityofthesystemisparamountinordertoreducethetotalcostofownership (TCO)ofWSNsolutionsdeployedinunattendedandharshscenariosbecauseinmanycases themaintenancecostsmaybemultipletimesthevalueofthedevicetobefixed. Therefore, we designed a Watchdog Timer (WDT) solution which is supported by a special software engineering methodology involving software and hardware modules that are WDT-aware. The main focus of this chapter is to describe this novel WDT approach for embedded sys- tems and to analyze its effectiveness for unattended systems, in particular outdoor WSNs. This chapter is mainly based on our recent work on reliability of unattended WSNs [64]. Similarly to Chapter 7, while the original context for this work is terrestrial WSNs, it is apparent that the topic is even more important for MI-WUSNs where the difficulties of re- movingandreinstallingaburiednodecansignificantlyincreasethemaintenancecosts. WSNs were introduced almost 15 years ago as a potential energy-efficient solution for spatially distributed sensors with very constrained energy resources. WSNs have been suc- cessfullyappliedindistinctscenarios,suchasbuildingautomation,industrialprocessmon- itoring,environmentalmonitoring,security,etc. [128]. Astrongevidenceoftheimportance of WSNs is the crescent adoption of the IEEE 802.15.4 and ZigBee standards [104, 105]. Matching with the mentioned energy constraints, WSN nodes typically have few hardware resources in terms of processor and memory capabilities. Due to such restrictions, in gen- eral it has been widely accepted that the resilience of WSNs must be guaranteed by means 243 CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 244 Fig. 8.1 One of the SoilSCAPE sites: the high degree of network sparsity is clearly visible whenvirtuallytransposingthesamenodetopologytotheUniversityofSouthernCalifornia campus. These nodes use off-the-shelf 2.4 GHz radio modules (10-18 dBm transmitting power level) and non-rechargeable batteries. No collaboration among nodes is in place and eachnodedirectlycommunicateswiththedatacollectornode(LC,markedwithadifferent color). of extensive network collaboration and a significant number of nodes since these devices, individually,maynotbeveryreliable. The typical trade-off in WSNs involves network performance (i.e., data throughput and latency) and energy consumption [128]. This aspect is the main difference between WSNs and any other wireless technology, even including modern mobile devices. Therefore, the low-levelnetworkinglayersofWSNsaredesignedtooperateatrelativelowdata-rate(e.g., typically around 100 to 250 kbps, or less) in order to keep the radio module of the devices inactive for the majority of the time. Certain WSN scenarios are even much more complex because the energy profiles of the sensor nodes are heterogeneous. For instance, a WSN node can a) be directly plugged to mains, or b) employ energy-harvesting techniques, or c) have a nominal lifetime dictated by the usage of non-rechargeable batteries [4]. Moreover, in other cases, some of the WSN nodes are mobile ones and ad-hoc schemes must be in place. In order to deal with any of these scenarios, off-the-shelf WSNs typically employ solutions that are strongly dependent on the collaboration among the nodes. Accordingly, it has been observed that the majority of the simulations in the WSN literature involve a significantnumberofnodes(e.g.,>100)withrelativesmallinter-nodedistances(e.g.,10to 100m)[22]. ForthecontextofMI-WUSNs,thesefactscallourattentionregardingtherisks of directly employing in MI nodes the existing reliability approaches used for off-the-shelf WSNs. The SoilSCAPE project started more than 6 years ago and currently it is still evolving CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 245 with more than 141 sensor nodes and 500 soil probes spread in 7 sites, each one with a Wireless Sensor Network (WSN) connected to a central data server by means of cellular text (SMS) and 3G links. The main goal of this project is to provide real-time in-situ soil moisture data for certain sites in order to validate/calibrate the algorithms employed by the SMAP Mission which is expected to provide estimations for soil moisture by means of RadarandRadiometerRemoteSensing[17]. The unattended WSNs used by the SoilSCAPE project are among the largest existing outdoor networks (in terms of node sparsity) and they are operating continuously for more than2years. Veryimpressiveisthefactthatalmostallnodesemploynon-rechargeablebat- teries,thataretypicallyreportedintheWSNliteratureaslastinginpracticeonlyfewweeks or months. For the SoilSCAPE case, the minimum battery lifetime was conservatively de- signed for 15 months of operation while the majority of the nodes already surpassed this goal. The only exception in terms of energy storage is the Local Coordinator (LC) node which employs a regular energy system (with solar panel) typically used in weather sta- tions. ThisisthecaseduetothehighenergycostsfortheLC-DataServerlink(SMSor3G, inourcase). Inthischapter,themainfocusisgiventothedevelopmentofWDTtechniquesthatresult in a very reliable WSN solution for the SoilSCAPE project. As expected, such case study can potentially provide useful guidelines for MI-WUSNs. Moreover, the examples and as- sociateddiscussionsareprovidedinawaytofacilitatetheadoptionofsomeoftheconcepts and engineering guidelines in generic scenarios involving unattended and critical devices, notnecessarilyWSNorMI-WUSNsnodes. ThenoveltyofourproposedWDT-basedsolu- tion is related to two aspects. First, the hardware WDT technique adds a functionality that is time-proved to be much more effective than the regular reset action. Second, in terms of software engineering, we introduce the concept of a WDT-aware system where the choice ofrestarting/rebootingadeviceisnotonlyassociatedwithafrozenstate,butwithabroader definition for erratic states. We believe that this latter systemic aspect is the main contribu- tionofthischapter,withparticularhigherimpactonunattendedoutdoorWSNs. This chapter is organized as follows: in Section 8.1, the motivations regarding the SoilSCAPEprojectisgiveninordertoprovidesomefieldandlaboratorylessonsregarding thedeploymentofoutdoorsWSNs. InSection8.2,multipleWDTtechniquesarediscussed andthemaingoalofthesectionistoshowmissingaspectsatthecurrentapproachesinvolv- CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 246 ing WDTs. The existing work with Virtual Finite State Machines is introduced in Section 8.3 in order to show how to identify erratic states and how to associate this action with the WDT techniques. Finally, in Section 8.4, the impact of using such WDT guidelines at the SoilSCAPEprojectisdiscussedandfinalremarksareprovidedinparticularassociatingthe contentsofthechapterwithMI-WUSNs. 8.1 Motivation The SoilSCAPE project is potentially one of the most energy-efficient solutions so far de- velopedforlargeandsparseoutdoorWSNsinparticularduetothesecharacteristics: • The application itself has very low data-rate: data collection (i.e., soil moisture and temperature)every5to60min. (bydefault,every20min.). • No collaboration among nodes occurs: a regular node only communicates with the LCnode. • Thenetworkoverheadissmallerthan1%. • Withtheexceptionofthedatacollector(LC),noinfrastructurenodeisinplace. • Theenergyconsumptionishighlyhomogeneousamongallregularnodes. • Because there are very big gaps between communication transactions (i.e., 2 to 4 seconds), multipath and similar effects are strongly minimized and it is possible to achieve inter-node distances very close to the theoretical maximum. This feature is particularlyimportantforlow-powerandsparseoutdoorsnetworks. • The network architecture we have developed, called Ripple-2, is open to any com- munication transceiver/module that can provide at least peer-to-peer wireless com- munication. The point-to-point data transfer reliability is ultimately enforced at the application-level,nomatterifthelowernetworkinglayersprovidethissupportornot. InFig. 8.1,itisillustratedonetheremarkablecharacteristicsoftheRipple-2architecture usedattheSoilSCAPEproject: ahighdegreeofnetworksparsity. Toachievesuchsolution with all the previously mentioned features, we had to engage in a preliminary systemic investigation of the problem. The main aspects of this investigation are listed below. The details regarding each of these items, with the exception of the last one, are given at the CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 247 provided references. In this chapter, the main focus is given to the last aspect regarding WDTtechniques: (a) Batteryperformanceunderextremeweatherconditions[4]. (b) Techniques to achieve non-rechargeable battery lifetime close to the expected value consideringitsnominalenergycapacity[21]. (c) Impact of enclosure (water and temperature damage) and solar panel characteristics foroutdoorWSNnodes[17,21,22]. (d) NetworkoverheadoftypicalWSNsolutions[22]. (e) Effects of transients due to voltage regulators and systems involving rechargeable batteriesandsolarpanels[4]. (f) Effectiveenergyconsumptionofanodeinsleepstate[4,21]. (g) Impact of having a star-like network topology and a non-collaborative approach in lowdata-rateWSNapplications[4,21]. (h) Effectsofhavinganapplication-layeroverlayactivatinganddeactivatingoff-the-shelf radiotransceiversandWSNmodules[21,101]. (i) HardwareandsoftwareWatchdogTimer(WDT)techniquestoprovidearobustsolu- tionforunattendednodesinoutdoors. AtthebeginningoftheSoilSCAPEproject,theapproachwastouseoff-the-shelfWSN solutions. OnAugust2010,twoexperimentalsiteswereselected(AnnArbor,MIandCan- ton, OK, USA) and ZigBee nodes [129] with a traditional energy scheme (solar panel + rechargeablebatteries)wereemployedinapreliminaryprototypecalledRipple-1[17]. Af- ter few months, we realized that Ripple-1 and off-the-shelf WSNs in general do not scale wellwhenahighnodesparsityisalsoimportant. WeinvestigatedthelargestoutdoorWSN deployments so far reported and a specific metrics was developed to better understand the sparsity issue [22]. That study shows that high physical coverage area in WSNs typically implytheuseofaveryhighnumberofnodes. In fact, the majority of the WSNs simulations usually employ more than 100 nodes to prove certain aspect related to scalability. Nonetheless, if we assume that only 30 sen- sor nodes are available for a certain WSN project, it can be more important to cover a CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 248 1000x1000m 2 area rather than a 200x200m 2 one. This is particularly true if the measure- mentshavehighspatialcorrelationasinthecaseoftheSoilSCAPEproject. Inotherwords, if the environmental measurements are highly correlated at least at some physical spots, it is typically more strategic to install the available sensing resources in larger areas than to concentrate all sensor nodes in one or few spots. Besides a higher network sparsity, it is alsoimportanttodrasticallyreducetheenergycostduetotheintrinsiccollaborationamong WSNnodes. Thisaspectisnotsignificantforhighdata-rateWSNs,butitisverycriticalfor low duty-cycled applications, as investigated in [101]. Therefore, we started the design of theRipple-2architecture[21,101]. Thisnewarchitecturehasacross-layernetworkingapproachwhichcomprisesapplication- layer software modules and a hardware module for power management and software-rule enforcement. Itisanopenarchitecturedesignedtosupportanykindofradiomodulethatal- lowsatleastpoint-to-pointcommunication(off-the-shelfWSNmodulesarealsosupported). Nothing is assumed in relation to the availability of communication error control at these radio modules. To support these design foundations, a complex system was necessary. The initialcodeoftheRipple-2sensornodehadmorethan3,000lines(currently,1,757). Also, the code of the data collector node (with SMS/3G link) has almost 8,400 lines. Moreover, the complexity of the underlying hardware module is also significant. For instance, our current regular sensor node employs 5 digital switches, 2 watchdog timers, and 1 energy sub-system to charge supercapacitors (the radio module is the load). As shown in Fig. 8.2, whileanoff-the-shelfWSNnodewouldbebasicallycomprisedoftheradiomodule(bigcir- cle),ourRipple-2nodehasmanymorehardwarecomponentsinordertoachieveanenergy efficiency at least 3 folds higher than typical WSN devices. Although this goal was ulti- matelyachieved[101],themaintrade-offistheriskofhavinganon-reliablesolution. That is, such level of system complexity may potentially introduce hidden issues and decrease thesystem’sreliability. Thiscasestudypaperhighlightsourstrategytoachieveaveryhigh reliable solution for the SoilSCAPE project. Accordingly, one very effective way to deal withscenarioswhereWSNnodesare trappedinerraticconditionsis tosimplyrestartthem (we must assume that these actions are rarely necessary) and this is the topic of the next section. CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 249 Radio Module (e.g., XBee PRO) Digital switch External WDT Supercapacitor charger Fig. 8.2 Ripple-2 node: the additional hardware and software modules aiming energy- efficiencycanpotentiallydecreasethereliabilityofthesolution. Accordingly,aWDT-aware design was adopted to allow the node to autonomously deal with problems without human intervention. 8.2 WatchDog Timer (WDT): Initial step toward reliable systems Theadoptionofcomponentredundancyandthewatchdogtimer(WDT)techniquearewell- knowntopicsattheembeddedsystemsliterature[130–133]andtheyaretypicallyregarded as ultimated solutions to achieve high-availability in computer-based systems. The WDT technique is basically a way to automatically reset the main processor after the device has been trapped in an erratic state for a certain amount of time. Clearly, WDT is a drastic so- lution that mimics the power-off/on action performed by a person as a last effort to solve a problem. AsignificantnumberofmodernmicrocontrollershaveaninternalWDTandsuch level of implementation is fairly documented and does not require hardware modifications. Nonetheless, the mentioned solutions may not provide the expected results for embedded systems, in particular unattended WSNs deployed in harsh environments for two main rea- sons. First, a reset action is not always sufficient. Second, after a restarting action, the devicecanstillbetrappedinanerraticstatethatrequireshumanintervention. Bothofthese aspectsarediscussedinthissection. The Ripple-2 architecture evolved from the adoption of a simple and traditional WDT scheme to a mission-critical oriented design and this process was based on progressive and CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 250 long-termlaboratoryexperimentsandnowvalidatedbylong-termfieldresults. Ourcurrent Ripple-2 solution for both regular and LC nodes addressed the previously mentioned two potential issues regarding the use of WDT. Typical questions that we faced during this pro- cessandourconclusionsarelistednext: 1)Isthecomponentredundancythepropersolution? A generic redundancy of components, such as processor or radio module, may potentially decreaseratherthanincreasethereliabilityofthesolution. Thenewswitchingmechanisms thatareeventuallynecessarywhenredundantmodulesareusedmaybepotentialsourcesof new and hidden issues. A more strategic and cost-effective approach is to perform careful testsofthevarioussub-modulesinordertoverifyifanyofthemisparticularlycritical. For instance, we monitored what happened with the energy system of a node employing solar panel and rechargeable batteries at freezing temperatures. In such critical energy-related scenario, we concluded that drastic oscillations of the voltage regulator may stop the pro- cessorevenemployingaregularWDTprotection. Forthisspecificscenario,theredundancy ofcomponentswouldnotmakethenodemorereliable. Conclusions: Theuseofredundantcomponentsisonlyrecommendedformodulesthat have high probability of failure and, in this case, we are assuming that it is not possible to simply adopt a better solution. This is typically the case when long-distance communi- cation links are involved. For this example, a higher cost (or, alternatively, with a lower bandwidth)connectivitysolutioncanbeusedonlyasabackupforthecheaperone(orwith a higher bandwidth). Another example, is the usage of multiple energy reservoirs, such as backups batteries [24]. It is important to highlight that these guidelines (and the remaining ones in this section) are not tied to WSN scenarios. For instance, remote and unattended SCADA (supervisory control and data acquisition) devices is another example of systems thatcanbenefitwiththeproposedguidelines. 2)Howtocharacterizean erraticstatecondition? The typical approach used in WDTs is to add a code procedure that resets the WDT (this action is called WDT refresh) and this procedure is expected to be regularly called. If the main processor freezes, the WDT refresh does not occur and the WDT device eventually resets the main processor. Note that the processor in this case was in an erratic condition, but there are other scenarios where the system may also be in an erratic condition. This is a key observation in our studies regarding reliability: the concept of state must always be considered inside a certain context. Therefore, without a proper formalization of the sys- CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 251 tem, an erratic state may have different interpretations and a system may be trapped in one of such erratic scenarios where the WDT technique is ineffective (i.e., WDT refreshes are stilloccurringbecausethesystemseemstobehaviorcorrectly). Forinstance,considerascenariowhereaWSNnodeisinaloop,movingfromstateAto BandthentoA,alwayswaitingforamessagereplyfromthenetwork. Canweconsiderthis case as an erratic scenario? The answer potentially depends on additional aspects such as, howmuchtimethenodeiswaiting,ifthereisanotherhighprioritytaskthatthenodeisnot performing due to this logical loop, etc. Note that not the necessarily the code programmer knows the answer, but the system designer must provide an answer. Our conclusion is that anerraticconditionofaWSNnodeischaracterizedbythenonconclusionofatransaction for a certain amount of time. The term transaction is associated with an expected sequence ofeventsandtheoverallideaisthatallprocessesarecontrolledbystricttimingsettings. In Section 8.3, software engineering guidelines that were used at SoilSCAPE project will be presented. Conclusions: The interpretation of an erratic state is regularly associated with a) a frozen state (non-responsive processor) or b) code loops. For both cases, it initially seems thatitissufficienttoaddasoftwareroutinethatregularlyrefreshestheWDTthusavoiding the reboot of the device in normal conditions, otherwise restarting it. This solution is very well known and effective for the majority of the real-world cases. However, it is also im- portant to consider cases where the main processor is perfectly operating but the device is an erratic condition from a system point of view. In this case, WDT techniques alone are not sufficient and additional approaches, such as the one to be presented in Section 8.3, are still necessary. Remember that if an unattended device presents such problem only once in ayear,thehumaninterventionstillbenecessarytomanuallyrestartthisdevice. 3)Howthepowermanagementsub-systemisrelatedtothereliabilityofthenode? Oscillations at the voltage rail used by the main processor is extremely critical because it can lead to irreversible damage of components, such as the real-time clock (RTC) chip, the EEPROM, and the system file of an SD card. Voltage regulators alone may not be the so- lution because the specifications of these devices may not hold for certain variations of the input voltage. In this context, we observed that a higher reliability is achieved when we separatethepowerlineofthemainprocessoranditssatellitelow-energydevices(e.g.,RTC chip)fromtheraillinesusedforpower-hungrysub-systems,suchastheradiomodule. CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 252 Conclusions: Becausemodernmicrocontrollerssupportlowvoltagelevels(e.g.,1.8V), itisimportanttoprotectthiscomponentfromissuesregardingenergydepletion. Thisispar- ticularlyimportantfordevicesthatmainlyrelyonenergy-harvesting. Ifthemainprocessor is always protected (e.g., its voltage rail never goes below 1.8V), it is always possible for this processor to log information regarding energy aspects. For instance, when the affected WSN node finally has enough energy to communicate, it is possible to inform the detailed energy-related reason why the node had not communicated before. In this sense, the relia- bilityofthesolutioncaneventuallyincreasewhenanewversionofthedevicehasitsdesign basedonsuchkindofinformation. 4)HowreliableistheWDTsolution? Many modern processors have an internal WDT and it may be used as the first level of the WDTsolution. However,itisalsopossibletouseanexternalWDTdevice,suchasoneRTC chip connected to the reset line of the main processor. This solution is more effective than the internal WDT and it can be employed as a second-level WDT with very small energy consumptiondependingonthechoiceoftheRTCchip. Nonetheless,wealsoobservedthat themostefficientWDTsolutionisanexternalmicrocontroller(MCU)completelydedicated totheWDTfunction. Suchintelligentdeviceallowsahigherlevelofreliabilitycontrolbe- cause it is an independent device that does not require any kind of cyclical setup which is the case when an external RTC is employed. Therefore, this WDT processor may serve as thethird-levelWDT.Whenmorethanoneofthese3WDTlayersareinplace,thereliability ofthesolutioncanbedrasticallyincreased. Conclusions: Inshort,whenmorethanoneWDTdeviceisinplace,thesolutioncanbe veryreliable. ThreeformsofWDTmechanismsarepossible: internal,externalRTC-based, and external MCU-based. Typically, the energy cost of the WDT solution also follows this order, while the reliability level follows the reverse order. A strategic approach is to use the internal or the external RTC-based for conventional WDT actions, such as to restart the processorwhenitistrappedbyafrozenoraloopcausedbyprogrammingissues. Forthese cases, the WDT can potentially solve the problem after a short time (e.g., 1 min.). On the otherhand,theexternalMCU-basedWDTcanbededicatedforlong-termtransactions,such the case when a node cannot communicate with other node after a certain amount of time (e.g., 30 min.). The same external WDT can still be used to enforce rebooting commands that came from higher system levels (main processor, coordinator node, Data Server, user CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 253 Fig.8.3ExternalWDTschemeusedtoincreasethereliabilityofWSNnodes. command,etc.). 5)Howeffectiveistheresetaction? Unfortunately, a reset action on the main processor may not effectively solve a problem withadeviceinanerraticstate. Moreover,duetotheincreasingavailabilityofcommercial sub-modules that have their own MCUs (e.g., RTC chips, radio and data-storage modules, digitalsensorprobes,etc.),potentialsystemreliabilityissuescanincrease. Forinstance,we observed cases where radio modules from diverse manufacturers stopped communicating with the main processor of our Ripple-2 node. The issue was not solved with a reset and, evenmuchmorecritical,itwasnotsolvedwithapower-off/onactionatthemainprocessor: theissuewassolvedwhentheradiomodulewaspowered-off/on. Asimilarissuewasalsoobservedforthemainprocessors. Unfortunately,evenapower- off/oneffectmaynotbeeffectivealltimes. Thisisthecasebecausemodernmicrocontrollers support low voltage levels and under very low-current drain circumstances, it is possible that the existing capacitance at the microcontroller voltage rail maintains the processor ac- tiveandtrappedinafrozen/loopstateevenafterashort-timepower-off/onaction. Tomake the scenario even more complex, when active devices are connected to the I/O lines, the processor also may not be properly restarted because it can still be powered via these I/O lines. In order to address the mentioned issues, we designed a very effective solution that CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 254 properly operated in all scenarios we have investigated. The circuitry is shown in Fig. 8.3 and it comprises of a mechanical relay activated by an opto-coupler. The mechanical relay has the role of switching the voltage rail of the main processor. The novelty in this solu- tionisthefactthatduringactivation(WDTresetaction),thevoltagerail(loadsideonly)is short-circuited to ground and it is maintained in this configuration for a certain amount of time. Therefore,anyresidualvoltagelevelatthelinesoftheprocessoreventuallyvanishes. If peripheral devices are attached to the processor, such lines must be also switched-off or, alternatively, all peripherals must be powered-off. Finally, observe in Fig. 8.3 that even if theWDTdevicefails,thesystemcontinuestooperatenormally. Conclusions: Any form of WDT action involving solely the reset line of a proces- sor/device is highly prone to reliability issues. In particular for unattended remote devices, itisrecommendedthataresetactiontobeproperlytranslatedinhardwarebyapower-off/on actionthateffectivelyaffectsallcomponentsofthesystem,notonlythemainprocessor. We highlightthatthecircuitryinFig. 8.3wasonlytrulyeffectivewhenthepower-offtimewas setto3seconds,whichisaparticularvalueforourRipple-2devicesandmustbeconfigured accordingtoeachcase. ThemainfocusofthissectionistheproperimplementationofWDTtechniques. Nonethe- less,theeventualWDTactionistherebootoftheequipment,aformofdrasticintervention. A way to reduce the need of such interventions is the formal modeling of the sates of the system. Moreover,byanalyzingtheevolutionofthemachinestates,themainprocessorcan identify scenarios where the system is in an erratic/non-expected/unknown state. Once this erratic state is identified, the main processor can issue a command to the external WDT to restart the overall system. We call such system a WDT-aware one and its fully realization requirestheformalizationofthestatesofthesystem,asdiscussedinthenextsection. 8.3 VirtualFiniteStateMachines+WDT The WDT solution can be an effective solution but data loss my be involved in a WDT action. In this section, guidelines will be provided in order to design the system with a proper correctness level where the WDT restarting actions are rarely necessary. One way toachieveabalancedsolutionregardingsoftwaredesignrobustnessandquickdevelopment is the formal modeling of the system states. Unfortunately, even simple cases of real-world CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 255 Fig.8.4Thevirtualfinitestatemachine(VFSM)modelrepresentsthesystembehaviorwith itsexceptionstreatmentassociatedwithWDTactions. machines quickly become very complex due to very high number of state machines (SMs), an issue known as explosion of states. If the purpose of these SMs is to facilitate the anal- ysis of the system by a human being, then the mentioned issue makes the formalization of states almost useless. This fact potentially explains the lack of SMs analysis at the current softwareengineeringtechniquesappliedtoembeddedsystems. A potential way to conciliate the adoption of SMs and the issue of explosion of states is the approach called virtual FSMs (VFSMs) introduced in [134], where a set of possi- ble distinct state machines (SMs) are virtually grouped together. The main purpose of the use of VFSMs in the context of this paper is the visual formalization, usually sufficient for systems with small to medium level of complexity. However, in order to move into the di- rection of computationally verify the correctness of a bigger system, additional tools may benecessary[135]. TheintuitionbehindtheVFSMconceptofavirtualsetofSMsgrouped at state A is the existence of intermediary states A 1 , A 2 , .., and A n that are strongly related toeachotherandcannotbebrokenintosmallerpartsoutofthecontextofthestateA. Note that such concept is relatively similar to the concept of a database (DB) transaction. The fundamentaldifferenceisthattypicallyDBtransactionsareassociatedwithwellcontrolled scenarios and exceptions are relatively straightforward to be managed. On the other hand, embedded systems, such as the ones involving WSN nodes, are normally exposed to many non-controlledscenarios(e.g.,energydepletionissues,communicationfailures,etc.). FewpublishedworksmentiontheuseofVSFMinWSNs,suchasin[101,136],butnot too many details are provided. The following simple example illustrates how we have used CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 256 VFSMs at the Ripple-2 design. Assume that a WSN node is currently at the virtual state A comprisedofSMs: • A 1 : nodereceivedmessagePROT1 • A 2 : nodeidentifiedinthemessagePROT1whatistherequestedmeasurementtypeto beperformed Depending of the A 2 outcome, the node goes to distinct SMs (e.g., B, C, etc.). What makesA 1 andA 2 asinglevirtualSMisthefactthatthenodecannotchangeitsSMwhileA 2 is not completed. But what would happen if a message PROT2 is received while the node makesitstransitionbetweenA 1 andA 2 ? Similarly,whatwouldhappenifanattachedintelli- gentprobeinterruptstheWSNnodebysendingitsmeasurementsrelatedtoacriticalevent? To make the scenario even more complex, assume that the two mentioned exceptions also occurredatthesametimewhenanadditionalexternalRTCchipinterruptsthemainproces- sorregardingaregularscheduledtaskthatmustbeperformed,suchassavingallthelogsat thememoryintoaSDcard? Observethatifthesystemdesignerisnotabletoanswerthese questions, the system may not be correctly programmed and the device can eventually go to a non-function status while operating without any hardware failure. Fortunately, this is a kind of scenario that WDT techniques discussed at the previous section can still help to mitigate, but it also means that the system is not properly designed and it is effectively not very reliable since regular WDT restarting actions can be associated with loss of data and events. For the mentioned example, if the processor moves from A 1 to any state other than A 2 , the VFSM model is clearly violated because a grouped set of SMs cannot be never bro- ken. If however, it is very important to handle each of the mentioned priority cases just after the execution of A 1 , then a new VFSM model must be provided where A 2 is actually another regular machine state, not a sub-state. However, if we allow too many exceptions, thecomplexityoftheVFSMmodelquicklyincreasesanditsusefulnessdecreases. Another important aspect involving SMs in VSFM is the strict adoption of timeouts: the machine cannot be in a certain state/sub-state for more time than expected. These timing values are stronglydependentonuser-definedconstraintsandhardware/networkcapabilities. Inmany cases,theyarebasedonempiricalobservationswhenevermaximumaccuracyisnecessary. Ourbalancedsolutionforasimilarscenariocameintheformofanexceptiontreatment procedurewhichisalwayscalledafterthecompleteexecutionofeachvirtualSMtransition CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 257 (A,B,...),suchasVFSMRULESinFig. 8.4. ThelastcommandatthestateAisthechangeof thenodestatustothenewSM(e.g.,from AtoC). Thisnewstateisthenformallyvalidated accordingtothestateofthesoftwarevariablesasverifiedbytheexceptiontreatmentVFSM RULES.Someofthesevariablesmaybemodifiedduringthepartialtreatmentofanexternal eventwhilethenodewasexecutinginstateA(inthiscase,suchvariablesarenotcriticalfor therunningSM A). TheroleoftheexceptiontreatmentistoconfirmthestateC ortomove themachinetoanewstate X accordingtopredefinedrules. Whenaspecificscenarioisnot addressed by a specific rule, the node can follow two approaches: a) return to the initial START VM where all software variables are reseted (equivalent to a hardware reset) or b) ignore the exception. For instance, this latter approach can be used for the mentioned case ofthereceptionofmessagePROT2whilethemachinetransitionsbetweenA 1 andA 2 . When the VFSM model, such as shown in Fig. 8.4 is evaluated and all expected virtual SM and transitionsbetweenthemareproperlydefined,thesolutionispotentiallyrobust. Again,any non-expectedandcriticalscenario(noexistingrule)canbeproperlyhandledbythedefault treatment(i.e.,softwareorhardwarereset). 8.4 Results As a result of the careful design of the Ripple-2 architecture for the SoilSCAPE project, we have achieved exceptional levels for both energy-efficiency and robustness [101]. No infrastructurenode,suchasroutersorrepeaters,isnecessarybesidesthedatacollector(Lo- cal Coordinator, LC) for very sparse deployments. The regular nodes and even the LC node can hibernate with power consumption smaller than 20μW. This is particularly im- pressive considering the fact that typical infrastructure nodes cannot even sleep. Moreover, theeffectivenetworkoverheadistypicallysmallerthan1%andtheenergy-efficiencyofthe nodes is multiple folds higher than the state-of-the-art in WSNs. Even more important, the energy depletion of the network is highly homogeneous. For instance, in a recent battery replacement trip to one of the sites that was operating for more than 2 years, we observed that the majority of the nodes stopped working at the same 3-month window. As expected, the TCO of the solution is drastically reduced when the maintenance visit can be properly plannedandallnodesareinvolved. Therefore,wehaveachievedthefunctionalgoalsofthe SoilSCAPEproject. Regardingthereliabilityofthesolution,theproblemswefacedwereuniquelyrelatedto CHAPTER8. ACHIEVINGHIGHSYSTEMRELIABILITY 258 external agents, such as animals destroying the soil probe cables, water damage at the soil moisture probes, and expected temporary SMS/3G connection issues due to the high dis- tance between the sites and cellular towers. The failure rate due to software and electronic issuesissteadilyzeroduringtheyears. WeobservedthatfromallLCnodes,5neverhadto use the WDT restarting action and 2 of them employed this drastic WDT action once for a long period of time (1-3 years). These results indicate that the overall system is indeed ro- bust. Nonetheless,forthe2observedcasesthatusedtheWDTrestartingaction,potentially thenodeswouldstopoperatingiftheWDTapproachwasnotused,thusrequiringexpensive humanintervention. Inthischapter,twoimportanttopicsregardingunattendedembeddedsystems(inpartic- ular, outdoor Wireless Sensor Networks) are investigated. First, the well-known Watchdog Timer (WDT) technique is discussed and two main novelties are highlighted: a) the need to fully eliminate any residual voltage level at the system to guarantee an effective reboot of the system and b) the fact that a system can be trapped in an erratic mode even without hardware issues. For the latter case, the combination of software procedures that regularly verify the timeout for important system transactions and an external microcontroller-based WDT solution is very effective. The second important aspect discussed in this paper is the use of Virtual Finite State Machines (VFSM) to identify logical errors that can be avoided without waiting for the drastic rebooting action of the WDT which can imply loss of data and/or environmental events. Although VFSM were introduced more than 20 years ago, few works mention its application in WSN designs. In this chapter, we present a practical example showing some of the problems that can arise when a formal modeling of the sys- temisnotinplace, howitaffectsthereliabilityofthesystem, andhowaVFSMmodelcan be properly employed in a WDT-aware system which accepts commands from upper-layer software to force the restarting of the system when it is trapped in logical loops, such as beinginanunknownmachinestate. Chapter9 High-LevelNetworkingAspectsof MI-WUSNs The main motivation behind this thesis is the practical realization of MI-WUSNs for two classes of applications: a) precise crop irrigation and environmental monitoring and b) pipeline gas/oil leak detection (PLD). These applications are discussed at the next section astheyprovideagoodfoundationregardingdesignconstraintsofMI-WUSNs. Whilewhat was discussed in this thesis is mainly applies to a point-to-point communication between a pairofMInodes,thementionedagricultural/environmentalandPLDapplicationstypically requireanetworkingapproach. Ifthenodesoftheseapplicationsareterrestrial,thesolution can be easily achieved by means of off-the-shelf commercial WSN devices. However, for underground settings involving MI-based communication, novel solutions must be investi- gated. For instance, while for terrestrial WSN nodes it is usual to consider the communication range as a circle-shaped 2D-region with the transmitting node at the center, this is not the case for MI-nodes, in particular if the employ single-axis antennas/coils. Moreover, mis- alignments are more critical in relation to the antennas of terrestrial nodes. Therefore, we believethatoncestandardsforthePHYlayeroftheundergroundMI-basedcommunication areinplace,theproliferationofMI-WUSNswillstillbeconstrainedbyhigh-levelnetwork- ing aspects. In this chapter, we present our preliminary solutions to some of the problems involvingMI-WUSNs. 259 CHAPTER9. HIGH-LEVELNETWORKINGASPECTSOFMI-WUSNS 260 Communication range of the CN node (all MI-WUSN nodes have similar range) CN Fig. 9.1 MI-WUSN application I: precise soil irrigation. The proposed topology allows the employmentofsingle-axiscoilsthusreducingthedeploymentcosts. 9.1 PreciseSoilIrrigationandEnvironmentalMonitoring Agriculturalirrigationandenvironmentalmonitoringapplicationshavemanycommonchar- acteristics [22]. However, in general, the former is more constrained due to the existence of machinery. Therefore, in this section we will mainly highlight this application assuming that many of the proposed guidelines can be promptly applied to the majority of the envi- ronmental monitoring cases. However, before proceeding it is also important to highlight why is not always possible to use terrestrial WSN nodes for environmental monitoring. A more detailed discussion is presented in [22] but for the context of this thesis it is enough to highlight that there are situation that concealment of the sensing devices is a key as- pect for certain project. In some cases, the physical protection of the devices is paramount as in the study of the effects of prescribed burn. For other cases, theft and damage caused byanimalsandotheragentsarethereasonstodeploythesensingnetworkundertheground. Todate,commercialsoilsensingdevicesarenotfullytransparenttotheagriculturalac- tivities(e.g.,plowingandharvesting): typically,theyarewired-connectedtoadatacollector or require posts for antennas and solar panels. The proposed solution allowsthe permanent installation of soil probes at many different points of a crop area with the goal of increas- ing the irrigation effectiveness, one urgent world problem considering the crescent water scarcity and the fact that almost 70% of the water is used for crop irrigation [10]. One approach is the use of a mobile data collector such as demonstrated in [136, 137], where underground nodes communicate with a moving irrigation equipment by means of radio waves(i.e.,EMmethod). Althoughthissolutionallowstherealizationofplowingactivities without the need of removing and redeploying the sensor nodes, the main drawback is the needofhavingtheirrigationmachineoperatinginordertogetthesoildata. The MI-based irrigation solution which is been proposed in this work is shown in Fig. 9.1. Observe that the distance between two nodes vary from 15 to 30m although it is pos- sible to slightly vary these values depending on the design parameters. Nonetheless, the advantagesofusingtheMImethod(i.e.,smallersignalattenuation)quickceasesafter30m CHAPTER9. HIGH-LEVELNETWORKINGASPECTSOFMI-WUSNS 261 Soil Air / Water Pipeline CN CN CN Different sensor technologies MI-Coil Inter-node dist. between MI-Coils: 5..50m Inter-node distance between Coordinator Nodes (CN): few to dozens kilometers CN Central PLD Server Regular MI Node These two routes are formed due to a) a data latency reduction provision (normal case) or b) a way to circumvent a node failure MI connection (low data-rate, small bandwidth) Wired connection CN-Data Server connection (high data-rate/bandwith) (a) (b) (c) Fig. 9.2 MI-WUSN application II: external instrumental (EI) for Pipeline Leak Detection (PLD) systems. (a) the so-called MI-WUSN-PLD can be employed for existing and new pipeline installations and multiple sensor technology can be employed (b) the combination of special nodes (Coordinator Nodes - CN) and a two-routing technique is proposed as a balanced solution regarding reliability and energy-efficiency (c) the CN nodes act as sinks thusalleviatingthedataqueuinginlowdata-rateMI-basedsolutions. (considering audio frequencies) for the majority of soils conditions and moderate coil sizes (e.g., 30-cm diameter) [99]. The use of a single-axis for the MI-coils is strategic for both technical [99] and also economical reasons, the latter mainly due to the ease of installa- tion. The drawback of a deployment based on single-axis MI nodes is the lack of a quasi- omnidirectionalcommunicationpattern[57]resultinginnetworktopologyconstraints. That is, one node can only efficiently communicate with the nodes that are axially behind and in its front, but not with the ones at the laterals, as shown in Fig. 9.1. Nonetheless, such non ad-hoc topology is typically very efficient in terms of network overhead thus resulting in a huge energy efficiency [21]. As shown in Fig. 9.1, this star of lines topology is based on a specialnode(CoordinatorNode-CN)whichhasthedatacollectionrole. Typically,theCN in agricultural applications has multiple coils, each one aligned to a certain section of line ofnodes(centralfigureinFig. 9.1). 9.2 PipelineLeakDetection(PLD) Anotherpotentialapplicationofmid-rangeMI-WUSNsisaPipelineLeakDetection(PLD) system, also called Leak Detection System (LDS). Considering physical sensing aspects, threekindsofLDSshavebeenusedtodetectleaksingas/oilpipelines: a)visualinspection, b)internalpipelineconditions(alsocalledComputationalPipelineMonitoring-CPM),and c) external pipeline monitoring or external instrumentation (EI in this work) [138, 139]. In general, the approaches a) and b) are always in place because they are strictly enforced by CHAPTER9. HIGH-LEVELNETWORKINGASPECTSOFMI-WUSNS 262 governmental laws. While the use of EI devices may still be considered optional (or, at least, the employment of a massive number of sensors), the regulations around the world are quickly changing considering the risks to people and environment with the significant increase of gas/oil pipelines. Different technologies have been employed for EI, such as acoustic and vapor-detection sensors, remote sensing, liquid sensing cables, and fiber op- tic (FO) cables. The latter option - used in a solution called FO-Distributed Temperature Sensing(FO-DTS)-hasbeenparticularlyhighlightedduringthelastdecadeduetothehigh sensitivity of the solution even more small leaks in pipelines lengths of few to dozens kilo- meters[139]. All EI technologies have trade-offs and certain critical drawbacks. For instance, while remote sensing can be accurate, a very high leak detection time can occur if the system is not operating continuously regarding the overall pipeline length. On the other hand, FO- DTScontinuouslymonitorsvirtuallyeverysinglepointofthepipeline(spatialresolutionis typicallysmallerthan1m[139])withveryhighaccuracyandwithouttheneedofinstalling anyadditionalsensor(i.e.,thefiber-opticwhichisinstalledclosetotheexternalwallofthe pipelineistheleaksensor byitself). Nonetheless,FO-DTStypicallycanonlybeemployed for new installations (i.e., redeployment of pipelines may be necessary) and the solution may still present accuracy issues when the pipeline transports multiphase fluids (e.g., oil, gas, water). EI in PLD systems that are comprised of different kinds of sensors and com- municating by means of MI-WUSNs (called MI-WUSN-PLD in this work) can be a very strategicapproachforbothexistingandnewpipelines. Existing pipelines The potential main advantages of MI-WUSN-PLD are a) minimum or no retrofitting issues and b) the fact that multiple kinds of leak detection sensors can be at- tachedtothesystem. TheMI-WUSN-PLDsolutionhasarelativelylow-costofinstallation because the probes and MI nodes are installed in specific points along the pipeline path. Thatis,acontinuouscutofthesoilpathisnotarequirement;otherwise,itcouldbedifficult to conclude the installation if obstacles are found along the path. The distance between the MI-node installation point will be potentially defined by the sensing constraints rather than thecommunicationones. Nonetheless,distancesofup50mbetweentwoadjacentMInodes is still possible although requiring coils larger than 30cm-diameter, as previously exempli- fied for the agricultural case. If the solution is deployed below the seabed, the worst case scenario discussed in this work may be the relevant one to be considered. Our empirical work in [99] actually included one case of naturally saturated soil at the border of a lake CHAPTER9. HIGH-LEVELNETWORKINGASPECTSOFMI-WUSNS 263 wherethesoilpresentsaveryhighlevelofconductivitysimilartotheseabedcase. Newinstallations As new PLD technologies, such as the ones based on FO-DTS, become availableandtheirdrawbacksareprogressivelyminimized,aMI-WUSN-PLDcanstrategi- cally be employed as a backup system for that more accurate EI solution. Moreover, when suchlevelofredundancyiscombinedtoCPManalysis, itmaybepossibletostronglymin- imize false alarms and provide a level of reliability of the system which is not currently available. Besides such advantage, an additional MI-WUSN-PLD system deployed in par- allelwithotherEIsolutioncanprovideenvironmentmeasurementsthatmaynotbepossible with that EI solution alone, such as water or soil pH/conductivity, 10m-range (or similar) vibrationdetection,etc. Inthiscase,suchfunctionalitiesevenwithlimitedspatialresolution areaddedtothePLDsystemthusenhancingitscapabilityofforecastingpotentialthreatsto thepipelinesandenvironment. A practical MI-WUSN-PLD is illustrated in Fig. 9.2, which follows a similar approach as in Fig. 9.1. However, in this case multiple CNs are employed and all nodes, including CNs, employjustasingle coil. Whileall MIcommunication inthiscontextisbidirectional (i.e., the same physical MI-coil has the transmitting (TX) and receiving (RX) roles), the arrows in Fig. 9.1(b) represent the way the sensing data is collected or events are sent to a CN. Potential network segmentations issues are solved by means of a two-routing scheme. Forinstance,ifanodeinaline-of-nodesfails(e.g.,astheonemarkedasstar inFig. 9.1(b), the network is not significantly impacted. The same scheme can also be used to reduce the data latency in a long line of nodes. As shown in Fig. 9.1(c), 3 distinct connectivity approachesareemployed: a)thecommunicationbetweenregularMI-nodes,asdescribedin thiswork,isexpectedtobeconstrained: lowdata-rateandverysmallbandwidth,b)theCN has a wired connection between the existing network infrastructure device (e.g., a FO-DTS repeater) and the local MI node, and c) each CN is expected to have a connection to the centralDataServerbymeansofarelativelyhigh-speedconnection. Therefore,thereliabilityoftheproposedsolutionmainlydependsontwoaspects. First, it is expected the availability of special Coordinator-role nodes (CNs) that are not very far fromeachother(e.g.,every5to30km)andwithahigh-bandwidthconnectiontothecentral PLD system. In this sense, a MI-WUSN-PLD is actually comprised of multiple networks withmultiplegatewaysconnectedtoacentralserver. Second,averyefficientenergyscheme must be provided for the MI nodes because the TX energy cost of a MI node is at least CHAPTER9. HIGH-LEVELNETWORKINGASPECTSOFMI-WUSNS 264 two orders of magnitude higher than the RX cost [99]. Therefore, if the nodes are mainly poweredbynon-rechargeablebatteries,anevent-detectionandmeasure-and-summarizeap- proachesarepreferredratherthanasimpledata-collectionscheme. Conclusions&FutureWork This dissertation work started with two main goals: first, the design of a low-cost, reliable, energy-efficient, and mid-range distance wireless communication solution for the under- ground environment; second, the analysis of the factors that mainly influence the electrode polarization (EP) effects on sub-MHz two-electrode (2E) dielectric measurements systems andalsothedevelopmentofaninitialmethodologytoidentifytheseEPeffects. Inthiswork,ithasbeenshownthatthesetwogoalsweresuccessfullyachieved. Regard- ingtheWUSNsolution,whichisbasedonMagnetic-Induction(MI),themaincontributions arelistedbelow: • Thefirstsub-MHzsoildielectricmodeldevelopedspecificallydevelopedtobeapplied forthedesignofMI-WUSNs • ThefirstcomprehensivesignalattenuationmodelforMI-WUSNs • The novel frequency and dual-wire adaptation schemes for the design and operation ofMI-WUSNs • AcomprehensiveenergymanagementandnetworkingframeworkforsparseWireless SensorNetworks(WSNs)thatcanbeappliedtoMI-WUSNs The proposed sub-MHz soil dielectric model is an empirically-determined one and it is based on 9 classes of different soil types. Moreover, a simple software-based technique is used to reduce the EP effects on the dielectric measurements. This preliminary model still can be improved if more accurate ways to isolate the EP effects are eventually identified. Nonetheless, based on empirical results, the signal attenuation model errors were found to besmallerthan10%inallinvestigatedcases. 265 CONCLUSIONS&FUTUREWORK 266 The proposed MI-Soil signal attenuation is said to be comprehensive because the elec- trical properties of the soil are embedded in this model. For instance, regarding the soil characteristics, the MI-WUSN designer only have to select the frequency, the distance be- tween the MI nodes, the best class of soil that represents the deployment site, and different soil moisture levels. With this set of input data, the soil path attenuation can be calculated without involving any circuitry aspect at this preliminary design phase. This attenuation duetothesoilrepresentstheactuallimitsoftheMI-systemandfromthispointcircuitryas- pectscanbeconsideredandsuchprovisionisalsopresentattheproposedsignalattenuation model. It is concluded that the design of a MI-WUSN is primarily constrained by the soil path attenuation, as discussed above, and also by other factors. Due to the non-linear character- istics of the related optimization problem, the design must be guided by practical engineer- ing and application constraints, such as the maximum levels of allowed transmission (TX) power,minimumlevelofsignalatthereceiving(RX)side,maximumsizeofcoil,frequency resonance drift issues, etc. Nonetheless, it was also demonstrated that the combination of frequency-switching and different wire thickness values for the coils is an important strat- egytobalancerobustness,energy-efficient,andpropervaluesfortheapplicationbandwidth nomatterthesoilconditions. AnotherimportantresearchcontributioninthisworkispresentedinChapter7: ahardware- softwarecross-layersolutionthatallowsaMInodetoextenditsbatterylifetimebymultiple folds. This practical solution, originally designed for terrestrial wireless sensor nodes, has a strong impact on the feasibility of MI-WUSNs. Following that chapter, the next two are also related to some aspects that must be addressed for the proliferation of MI-WUSNs: reliabilityaspectsandnetworkingtopologiesofMI-WUSNs. Regarding the proposed EP-identification methodology, the main conclusions and find- ingsarelistedbelow: • Based on empirical work, repeatability issues are present when the well known four- electrode(4E)method(usedtoeliminatetheEPcontributions)isappliedtosoilmea- surements • Similarconclusionalsoholdsrelatedtotheuseofnon-polarizableelectrodes,suchas Pt-Black CONCLUSIONS&FUTUREWORK 267 • TheEPdata,byitself,maybeanimportantsourceofinformationmulti-phasesamples (mixtures),suchastypicalsoilmaterial • The Fricke’s method (1937) is mathematically accurate but it is very hard to be achieved in practice because it requires that two measurements in distinct dielectric cellshavethesameEPimpedance • The proposed EP-identification methodology is a novel approach that allows the sat- isfaction of the Fricke’s conditions by means of a careful observation of the patterns ofharmonicsduetothenon-linearEPeffects The proposed Linear EP-Match method developed for the EP-identification is essen- tially achieved by satisfying the Fricke’s conditions. To this end, the so called EP-Match conditionisidentifiedbymaintainingthesystems(i.e.,firstandsecondmeasurementswith distinct dielectric cell constants) at their linear regions and measuring the total harmonics distortion (THD). The ultimate goal is to achieve a very small THD which is similar for bothmeasurements. Preliminaryresultswithourworst-casescenariointermsofEPeffects (saturated NHS-SAT class of soil) show that the method is sound and it indeed reduces the high values of dielectric constant for the soil sample in low frequencies and the results are relativelyclosetotheonesobtainedbyusingtheoriginalMI-Soildielectricmodel. As ongoing and future work, we intend to extend the implementation of the Linear EP- Match to more soil types/conditions. In this way, the original sub-MHz MI-Soil dielectric modelcanbealsousedforgeneraldielectricmeasurementsandnotonlyformodelsrelated totheMIsignalattenuationinsoils. Also,weintendtofullyvalidatetheEP-Matchmethod and such effort requires the completion of the current new instrumentation and testbed for thispurpose. ThischallengepotentiallyrequirestoclosetheEIS/MISloop,anovelconcept explainedinChapter2whichcanbeessentiallytranslatedastheneedofdevelopingthefirst MagneticInductionImpedanceSpectroscopy(MIS)systemforfrequenciessmallerthan100 kHz. Abbreviations • 2E:Two-Electrode(impedancemeasurementsystem) • 4E:Four-Electrode(impedancemeasurementsystem) • ADC:Analog-to-DigitalConverter • AGC:AutomaticGainControl • AM:ActiveMTS(usedatBETSprotocol) • AWG:AmericanWireGauge(standardforwirediametere) • BAN:BodyAreaNetworks • BD:Break-PointDistance(maximumdistanceforMItechnique) • BER:BitErrorRate • BETS:Best-EffortTime-Slotallocationprotocol • BS:BaseStation • BW:BandWidth • CH:ClusterHead • CN:CoordinatorNode • CPM:ComputationalPipelineMonitoring • CRC:CyclicalRedundancyCheck • CW:Continuous-Wave • DDC:DualDuty-Cycleoperation • DDS:DirectDigitalSynthesizer(digitalsignalgenerator) • DF:DissipationFactor(sameaslosstangent) • DTN:Delay-TolerantNetwork 268 ABBREVIATIONS 269 • DS:DataServer • ED:EndDevice(e.g.,MInode) • EI:ExternalInstrumentation • EIS:ElectrochemicalImpedanceSpectroscopy • EM:EmergencyMode(usedatBETSprotocol) • EP:ElectrodePolarization • EP-Match: (Linear)ElectrodePolarizationMatch(EP-identificationtechnique) • ETS:EDTime-Slot(usedatBETSprotocol) • FFT:Fast-FourierTransform • FH:upper(High)-boundFrequencyusedforthebestsoilscenario • FL:Lower-boundFrequencyusedfortheworstsoilscenario • FL-FH:FL/FHfrequencyadaptationschemeinMIsystems • FO:FiberOptic • FO-DTS:FiberOptic-DistributedTemperatureSensing • FSET:FixedSETofparametersusedattheMIoptimizationdesignalgorithm • IA:ImpedanceAnalyzer • LC:LocalCoordinator • LCM:LeastCommonMultipleprinciple • LDS:LeakDetectionSystem • LF:Low-Frequency(frequencyband);e.g.,sub-MHzrange • LFP:Lower-FrequencyPoint(usedforEP-correction) • MAC:MediumAccessProtocol • MCU:MicroCOntrollerUnit • MI:MagneticInduction • MIS:MagneticinductionImpedanceSpectroscopy • MI-WUSN:WUSNbasedonMagneticInduction ABBREVIATIONS 270 • MPPT:MaximumPowerPointTracking • MTS:MajorTimeSlot(usedatBETSprotocol) • NFC:NearFieldCommunication • NFMIC:NearFieldMagneticInductionCommunication • OOK:On-Off-Keying(modulationtechnique) • OTA:Over-The-Air(wirelesscommunicationinfree-space) • PA(P.A.): Power-Amplifier(RXfront-endPA,atthiscontext) • PAN:PersonalAreaNetwork • PER:PacketErrorRate • PHY:NetworkingPHYsicallayer • PLD:Pipelinegas/oilLeakDetectionsystem • PV:Photovoltaiccell(e.g.,solarpanel) • QoS:QualityofService • RFID:RadioFrequencyIDentification • RTC:Real-TimeClock • RSS:ReceivedSignalStrength • RX:Receiving(circuit) • SNR:Signal-to-NoiseRatio • SM:SoilMoisture • SM:StateMachine • SMAP:SoilMoistureActive-PassiveMission(NASA) • STS:SleepingTimeSlot(usedatBETSprotocol) • TCO:TotalOwnershipCost • TDMA:Time-DivisionMultiple(medium)Accessprotocol • THD:TotalHarmonicDistortion • TS:TimeSlot(usedatBETSprotocol) ABBREVIATIONS 271 • TTE:Through-The-Earth(wirelesscommunication) • TX:Transmitting(circuit) • UG:UnderGround • UFP:Upper-FrequencyPoint(usedforEP-correction) • UWCN:UnderWaterCommunicationNetworks • VFSM:VirtualFiniteStateMachines • VWC:VolumetricWaterContent • WC:WaterContent • WCUCA:WirelessCommunicationinUndergroundorConfinedAreas • WDT:WatchDogTimer • WOR:Wake-upOnRadio • WPAN:WirelessPersonalAreaNetwork • WPT:WirelessPowerTransfer • WSN:WirelessSensorNetworks • WUSN:WirelessUndergroundSensorNetworks ANovelIndoorSub-MHzMI-Soil Testbed Inthisresearchwork,manyoftheempiricalevaluationsemployedoutdoorstestbedsindif- ferent locations. These preliminary investigations were very helpful in the sense that the ongoing MI-signal attenuation could be tested without the need of special attention to the noise floor, maximum level of the transmitted signal, special equipments, etc. However, as soon we obtained the first validation results, the repetition of the experiments become more difficult because the tests with different frequencies, coil types, and soil conditions wouldrequiremultipletripstothesites. Moreover,theprecisioncontrolofthesoiltemper- ature and moisture is easier to be achieved in a laboratory environment. Considering that such control is a requirement for our ongoing work on sub-MHz soil dielectric model, we decided to investigate the feasibility of developing an indoor MI-Soil testbed. Our previ- ous experiences developing outdoors versions of a radio wave-based [140] and MI-based underground testbeds were important. Nonetheless, the design of an indoor version of a MI-testbedpresentsasignificantnumberofadditionalchallenges. The pioneering work in MI testbed is recently reported in [49] where the 4-9MHz fre- quencybandisusedinconjunctionwithsoftware-definedradiomodulesconnectedtocom- puters. For such frequency range, variable capacitors can be typically employed. Unfortu- nately, for our typical MI nodes operating at/close to audio frequencies, the adjustment of the resonant frequency may need a different approach due to the lack of commercial vari- ablecapacitorsforthetargetrangeofvaluesontheorderofdozenstothousandsofnF(nano Farads). AnotherdifferencebetweentheproposedMItestbedandtheonein[49]isrelatedtothe signalsensitivityandresolution. Forinstance,whilethebackgroundnoisein[49]isaround 7μV(assumingdeviceswith50Ωimpedancein[49]),wearestillabletomeasurevariations 272 ANOVELINDOORSUB-MHZMI-SOILTESTBED 273 5cm-diameter PVC pipe 0.5m 1m 1.5m 2m 0.9m RX RX RX RX TX 0.15m 0.16m 2.1m FL coil 10 kHz FH coil 26 kHz 28 AWG N=121 36 AWG N=80 Air Twinax connectors Rotation angle mark 2cm 4.9cm 1cm 4.9cm Soil Air 0m Fig.A1-1IndoorSub-MHzMI-Soiltestbedusedfortheempiricalevaluationsinthiswork. smallerthan1μ VattheRXMIsignal. Suchresultswereachievedduetoboththeground- ing/shielding design and also the instruments selection/arrangement. We are still able to measure the phase difference between TX and RX signals with accuracy of 0.1 o . Such in- formation is neglected in our current signal attenuation model because its focus is mainly onthemagnitudeofthesignal. Nonetheless,suchphaseinformationisimportantforfuture works regarding an inverse problem. In other words, one can use such MI-communication testbed to infer about the dielectric properties of the medium (soil, in this specific case) basedonthesignalattenuationandphasedrift. Usually at outdoors MI testbeds the interference noise is not an issue because the soil itself is a good attenuator/shield for external signals that come from the surface. More- over, at least for the preliminary tests, it is possible to increase the transmitting signal level withoutmanyconstraintswheneverthemeasurementsensitivitystartstobecomeaproblem. However,atindoorsenvironment,thisscenariodrasticallychanges. Thepresenceofdozens of electronic devices operating at frequencies below 200 kHz (mainly switched power sup- plies) poses a strong challenge for the realization of a sub-MHz MI-soil testbed in terms of signal-to-noiseratio(SNR).Aggravatingthisscenario,usuallyitisnecessarytooperatethe coils/antennasoftheMInodesatlowsignallevel: strongersignalswouldrequirelargersoil volumes in order to support the mentioned important unbounded soil medium assumption. Therefore, the main goal is to confine the major part of the transmitted energy at the soil volumewherebothTXandRXcoilsareembedded. Evenifthisisnotpossible,theenergy notdirectedtotheRXcoilmustnotreturntothesystembyanyotherway,whichistypically whatoccursinareal-worldundergroundenvironment. The first design step in our MI-Soil testbed is to define the minimum acceptable length ANOVELINDOORSUB-MHZMI-SOILTESTBED 274 ofthetestbed,whichisfoundtobe2m,asshowninFig. A1-1. Typically,onewoulddesire amuchsmallervalueforthetestbedlength. However,thisdecisionisactuallyafunctionof manyfactors,suchasthefrequencyrangetobeinvestigated,themaximumTXsignallevel, themaximumsignalattenuation,theminimumsignalvariationstobeidentified,andtheres- olution(amplitudeandphase)oftheRXinstrumentation(aLock-InAmplifierEG&G7260, in our case). Based on preliminary tests with our instrumentation, it becomes apparent that we are constrained to resolutions of around 0.5μV and 0.1 o . For instance, if according to the investigated model the RX signal is expected to vary only 0.1μV for a certain distance r when the soil moisture changes from 1 to 5% vwc, then the adopted testbed length is not acceptable. Oneofthemostcriticalofthefactorsevaluatedtodefinethelengthofthetestbedisthe minimumsignalvariationstobeidentified. AsshowninFig. 6.1inChapter6,disregarding circuit aspects, almost all signal variations involving free-space, dry soil, and saturated soil for some frequencies fall in a window smaller than 10dB. In this case, the instrumentation musthaveenoughresolutionafterconsideringcircuit(signalgain)andsoil(signalattenua- tion) aspects. In our case, the value of 2m for the testbed length was decided based on the mentionedconstraints. Onewaytoverifythequalityofsuchkindoftestbedisbycheckingtheempiricalresults derived from this testbed: if they have a good agreement with the signal attenuation model which is being adopted. Naturally, for this case, it is assumed that the referred model has already been validated by other means. Fortunately, in our case we already had validated our model at outdoors for diverse frequencies smaller than 100 kHz and this step gave us confidence that any significant errors, if found, would be caused by instrumentation arti- facts, not the model itself. Therefore, in our testbed design we proceeded to the next steps knowingbeforehandwhichtheexpectedresults. The next design step is related to the leakage of energy from the testbed and with po- tential residual return to the testbed again. As shown in Fig. A1-1, the quantity of soil surrounding the coil in our testbed is around 5cm. Observe that there are two coils at the PVC pipe (FL and FH), therefore the following set of tests must be repeated for each fre- quencycase. These tests are called preliminary ones because they are necessary conditions to verify ANOVELINDOORSUB-MHZMI-SOILTESTBED 275 Signal generator Siglent SDG1010 CH1 CH2 R C TX Coil RX Coil 250kHz Lock-In amplifier EG&G 7260 Ref Sig C Twinax connectors BNC connector Soil CH1 & CH2 are synchonized * * Keithley 2015-P Fig.A1-2InstrumentationandshieldingschemeusedattheSub-MHzMI-Soiltestbed. the good quality of the testbed, but they are not sufficient (usually a comparison involving an outdoor MI-soil testbed is necessary, as already mentioned). The goal is to define the smallestphysicaldimensionsofthetestbedbesidesthelengthwhichwasalreadyevaluated. Forthisobjective,twotestsareperformed: therotationandthebladetests. Theformertest verifies if the received signal correctly drops if theθ angle changes from 0 to 90 o . The 0 o case corresponds to the ideal alignment, as shown in Fig. 3.2. Observe in Fig. A1-1 that the PVC pipe has a mark to help in this rotation test. For the new angle 90 o (the rotation involves the RX node only), the RX signal must be very close to the value when the TX sourceispowered-off,accordingtothepresentedtheoryinthiswork. Ifthisisnotthecase, part of the TX energy is potentially returning to the testbed and this situation is forbidden: thecauseoftheissuemustbeidentifiedandeliminatedorbiggersoilvolumesmustbeeval- uatedforthetestbed. Thelattertest,thebladeone,employsametallicsheet(weusedanaluminumsheetwith 3mmofthickness)withthedimensionscorrespondingto2to5timesthesurfaceareaofthe MIcoil. Wedefinethedistance d1inthiscontextasthesmallestdistancebetweenanypart ofthecoilandtheinterfacesoil/airatthetestbed,includingitslateralwalls(inourcase,all the walls are made of wood). As shown in Fig. A1-1, the distance d1 in our testbed corre- sponds to approximately 5cm. Therefore, our adopted blade test is associated to a distance of 10cm surrounding the TX/RX coils. Usually, we do not have to worry about the floor ofthetestbedprovidedthatitdoesnotcontainexcessivemetallicorferromagneticmaterial thatcancompromisetheexperiments. Forthebladetest,wewanttoconfirmthatasignificantvariationoftheRXsignaloccurs ANOVELINDOORSUB-MHZMI-SOILTESTBED 276 whenthebladeisplacedveryclosetoTXandRXcoils,thatis,atdistance d1. Inthisway, weareessentiallycheckingtheexistenceofthereactivefieldswhicharespecificallyrelated totheMIcommunicationchannel. Next,wewanttoverifythatthesignalattenuationatthe RX coil does not significantly varies when the blade is placed at distance of 2 times that distance (i.e., 2d1) from both TX and RX coils. If such variation occurs, it means that the fields are still very strong outside the soil volume and this situation is critical. Typically, weneedtoemploylowTXsignaltomitigatethisproblemwhilestillmaintainingrelatively smaller soil volumes for the testbed. Two main effects explain why the signal attenuation occurred at least for the d1 distance case: a) the distortion of the radiation pattern from the TX coil and b) the occurrence of eddy currents at the blade. Summarizing, if strong RX signal variations are not detected at the mentioned distance 2d1, this is a preliminary indi- cationthatthedimensionsofthesoiltestbedinconjunctionwiththesourcesignallevelsare adequate. Despitepassingatthepreliminarytests,themaindesignchallengeindesigntheMI-Soil testbedisstilltheneedoffilteringnoiseatverysmallsignalslevels(ontheorderoffewμV in our case). Besides a careful shielding design as shown in Fig. A1-2, it is also necessary toemployspecialequipments,suchascommerciallock-inamplifiers,thattypicallyemploy coherent filtering techniques by means of signal processing. When such equipments are usedratherthansimpleACmicroornanovoltmeters,evensignalburiedinnoisescenarios can be supported. In this case, the components of the input signal that are not at the fre- quency of interest are efficiently eliminated. Without similar signal processing techniques, the realization of the proposed MI-soil testbed operating at low signal levels may not be feasible. On the other hand, observe that the proposed testbed is still in the analog domain and there additional advantages in this approach. When digital radio modules are directly used at the testbed, this one loses information of the analog domain that can be important dependingifthegoalsofthetestbedarerelatedtobothcommunicationanddielectricsens- ing. One important component at the testbed is the twinax (or twin axial) cable which al- lows the two connections of the FH/FL coil to protected by a frame ground, as shown in Fig. A1-2. This frame ground refers to the earth ground (rather than the analog or logical grounds) of the equipments. Note that any part of the BNC connectors shown in Fig. A1-2 are attached to the frame ground. In this way, regarding to the testbed, the frame ground is a floating one and its only shielding purposes, i.e., no measurement is taken from any ANOVELINDOORSUB-MHZMI-SOILTESTBED 277 pointandtheframeground. Also,thesourcesinusoidalsignalisgeneratedwithoutanyDC bias. Regarding the resistance R shown in Fig. A1-2, its objective is to provide a lower load boundary as perceived by the signal generator. The one we used has a minimum load impedanceof50ΩandR ef f forthecoilconfigurationscanreachmuchsmallervalues,such as 4Ω. In fact, the value of R is transparent to the MI-soil signal attenuation presented in this work because I TX is the main circuit parameter of consideration. Similarly, the detun- ingeffectduetoapossibledeviationfromthedesignresonantfrequencyisnotcapturedby the model at the TX side, only at the RX side. This is the case because, again, only I TX is considered. Nonetheless,inafinalimplementationoftheMInode,I TX willbedynamically monitoredinordertoallowanauto-correctionofthedetuningeffect. As reported in Chapter 6 and in [141], we have obtained success at the majority of the experiments performed based on this first version of MI-Soil testbed in our laboratory. Nonetheless,wealsoobservedsomesensitivitylimitationsathigherfrequencies. Whilewe are able to have fair induced voltage measurement accuracy (error < 5%) even when the difference between the signals is smaller than 1dB, this testbed solution is still constrained to frequencies smaller than 30kHz. We have not precisely determined the reasons for such limitation but we believe that the shielding scheme must be enhanced. One possibility is to employafront-endamplifierclosetotheRXcoilwithactiveguardshieldingbeforesending thesignaltothelock-inamplifier. Weanticipatethatthemaindisadvantageofthissolution, whichiscurrentlyunderevaluation,isthepotentialneedofacalibrationprocedureforeach experimentatthetestbed. The overall good results with this sub-MHz MI-Soil testbed provide additional insights beyond its original application in MI-WUSNs. For instance, one can observe remarkable similarities between the testbed and the HP E5050A colloidal probe connected to a com- mercial impedance analyzer (this equipment was discontinued). That device operated at 75kHz-30MHz which is an expected frequency range due to the small dimensions of the probes. Nonetheless, its applicability for dielectric measurements was limited to fluids. Unfortunately, it is apparent that no commercial instrument based on the electromagnetic induction method is currently available. Therefore, we perceive our sub-MHz MI-testbed as a potential instrument for dielectric measurements in general. Naturally, depending on the frequency range and material of interest the dimensions of the testbed can be strongly reduced. Moreover, the artifacts due to the electrode polarization effects [84] will not be presentinanon-contactinstrumentationsuchasthisMI-testbed. References [1] A. Silva and et. al., The Future of Wireless Underground Sensing Networks Consid- eringPhysicalLayerAspects-TheArtofWirelessSensorNetworks-H.Ammari(ed.), ch.12,pp.451–484. Springer,2014. [2] J.Santamarinaandet.al.,Soilsandwaves. NY:Wiley,2001. [3] K.KleinandJ.Santamarina,“MethodsforBroad-BandDielectricPermittivityMea- surements (Soil-Water Mixtures, 5 Hz to 1.3 GHz),” Geotechnical Testing Journal, vol.20,pp.168–178,June1997. [4] A. Silva and et. al., “Power-management techniques for wireless sensor networks and similar low-power communication devices based on nonrechargeable batteries,” J.Comp.Netw.andComm.,pp.1–10,2012. [5] A. Silva, M. Liu, and M. Moghaddam, “Ripple-2: a Non-Collaborative, Asyn- chronous and Open Architecture for Highly-Scalable and Low Duty-cycle WSNs,” inProc.ACMMiSeNet’12,(NewYork,NY,USA),pp.39–44,ACM,2012. [6] A. Sugden, R. Stone, and C. Ash, “Ecology in the underworld.(Introduction) - Soils -TheFinalFrontier,”Science,vol.304,p.1613,June2004. [7] I. Akyildiz and E. Stuntebeck, “Wireless underground sensor networks: Research challenges,”AdHocNetw.J.,vol.4,pp.669–686,July2006. [8] L. Li and et. al., “Characteristics of underground channel for wireless underground sensornetworks,”inMed-Hoc-Net’07,(Corfu,Greece),2007. [9] I. Akyildiz and et..al., “Channel modeling for wireless underground communication insoil,minesandtunnels,”Phys.Comm.J.,2009. [10] A. Silva and M. Vuran, “Empirical evaluation of wireless underground-to- underground communication in wireless underground sensor networks,” in Proc. IEEEDCOSS’09,(MarinaDelRey,CA),June2009. [11] Z. Sun and I. Akyildiz, “Magnetic Induction Communications for Wireless Un- derground Sensor Networks,” IEEE Trans. on Antennas and Propagation, vol. 58, pp.2426–2435,2010. [12] I. F. Akyildiz and et. al., “Signal propagation techniques for wireless underground communicationnetworks,”Phys.Comm.J.,vol.2,pp.167–183,Sept.2009. 278 REFERENCES 279 [13] N. Peplinski, F.Ulaby, and M. Dobson, “Dielectric properties of soils in the 0.3-1.3- ghzrange,”IEEETrans.GeoscienceandRemoteSensing,vol.33,pp.803–807,May 1995. [14] V. Mironov, L. Kosolapova, and S. Fomin, “Physically and mineralogically based spectroscopicdielectricmodelformoistsoils,”IEEETrans.Geosci.RemoteSensing, vol.47(7),pp.2059–2070,2009. [15] M. Vuran and A. Silva, Sensor Networks: Where Theory Meets Practice - ed. G. Ferrari,ch.CommunicationthroughSoilinWirelessUndergroundSensorNetworks -TheoryandPractice,pp.309–347. Springer,March2010. [16] J. Scott and et al., “Dielectric constant and electrical conductivity measurements of moist rock: A new laboratory method,” Journal of Geophysical Research, vol. 72, pp.5101–5115,October1967. [17] M.Moghaddamandet.al.,“Awirelesssoilmoisturesmartsensorwebusingphysics- based optimal control: Concept and initial demonstrations,” IEEE J. Selected Topics in Applied Earth Observations and Remote Sensing, vol. 3, pp. 522–535, December 2010. [18] C.A.Balanis,Adv.EngineeringElectromagnetics. Wiley,2ed.,2012. [19] T. S. Rappaport, Wireless Communications - Principles and Practice. Prentice Hall PTR,1ed.,1996. [20] M. Moghaddam and et al., “Ground network design and dynamic operation for vali- dation of spaceborne soil moisture measurements: Initial developments and results,” inInProc.ESTF-2010,2010. [21] A.Silvaandet.al.,“AnAdaptiveEnergy-ManagementFrameworkforSensorNodes with Constrained Energy Scavenging Profiles,” Intl J. of Distributed Sensor Net- works,no.272849,pp.pp.1–33,2013. [22] A.Silva,M.Moghaddam,andM.Liu,DesignofLowData-RateEnvironmentalMon- itoring Applications - The Art of Wireless Sensor Networks - H. M. Ammari (ed.), ch.3,pp.51–94. Springer-VerlagBerlinHeidelberg,2014. [23] A. Phocaides, Handbook on pressurized irrigation techniques. Rome, Italy: Food andAgricultureOrganizationoftheUnitedNations,2ed.,2007. [24] A. Silva, M. Liu, and M. Moghaddam, “WSN-SA: Design Foundations for Situa- tional Awareness Systems Based on Sensor Networks,” in IEEE Global Humanitar- ianTech.Conf.(GHTC’13),(SanJose,CA),2013. [25] S. Kim and et al., “Health monitoring of civil infrastructures using wireless sensor networks,”inInProc.IEEEIPSN,pp.254–263,2007. [26] L. Bandyopadhyay, S. K. Chaulya, and P. K. Mishra, Wireless Communication in UndergroundMines: RFID-basedSensorNetworking. Springer,2010. REFERENCES 280 [27] R. King, G. S. Smith, M. Owens, and T. T. Wu, Antennas in Matter - Fundamentals, Theory,andApplications. MITPress,1981. [28] R.King,M.Owens,andT.Wu,LateralElectromagneticWaves: TheoryandApplica- tions to Communications, Geophysical Exploration, and Remote Sensing. Springer, 1992. [29] A. Silva, “Channel Characterization for Wireless Underground Sensor Networks,” Master’sthesis,UniversityofNebraska-Lincoln,2010. ComputerScienceandEngi- neering: http://digitalcommons.unl.edu/computerscidiss/13. [30] A. Silva and M. Vuran, “Communication with aboveground devices in wireless un- derground sensor networks: An empirical study,” in Proc. IEEE ICC ’10, (Cape Town,SouthAfrica),May2010. [31] E.Thomasandetal.,“APowerLinkStudyofWirelessNon-RadiativePowerTrans- fer Systems Using Resonant Shielded Loops,” IEEE Trans. Circuits and Systems I, vol.59,no.9,pp.2125–2136,2012. [32] K. Fotopoulou and B. Flynn, “Wireless Power Transfer in Loosely Coupled Links: CoilMisalignmentModel,” Magnetics, IEEE Transactions on,vol.47, pp.416–430, Feb.2011. [33] A.Lozano-Nieto,RFIDDesignFundamentalsandApplications. CRCPress,2011. [34] S.KahrobaeeandM.Vuran,“VibrationEnergyHarvestingforWirelessUnderground SensorNetworks,”inIEEEICC’13,(Budapest,Hungary),June2013. [35] J.Sojdehei,P.Wrathall,andD.F.Dinn,“Magneto-inductive(MI)communications,” inOCEANS’01-MTS/IEEEConferenceandExhibition,vol.1,pp.513–519,2001. [36] “Magneto Inductive-Remote Activation Munition System (MI-RAMS).” http://www.pica.army.mil/pmccs/supportmunitions/demolitionsys/ramsmi.html. [37] C. Bunszel, “Magnetic induction: A low-power wireless alternative,” RF Design, vol.24,no.11,pp.77–80,2001. vol.24,no.11,pp.78-80. [38] E.Shamonina,V.Kalinin,K.Ringhofer,andL.Solymar,“Magneto-inductivewaves inone,twoandthreedimensions,”Appl.Phys.,vol.92,no.10,pp.6252–6261,2002. [39] E.Shamonina,V.Kalinin,K.Ringhofer,andL.Solymar,“Magneto-inductivewaveg- uide,”Elect.Lett.,vol.38,pp.371–373,2002. [40] N. Jack and K. Shenai, “Magnetic Induction IC for Wireless Communication in RF- ImpenetrableMedia,”inWMED’07-IEEEWorkshoponMicroelectronicsandElec- tronDevices,(Boise,ID,USA),pp.47–48,April2007. [41] M. Sun, S. A. Hackworth, Z. Tang, G. Gilbert, S. Cardin, and R. Sclabassi, “How to pass information and deliver energy to a network of implantable devices within the humanbody,”inEEEEngMedBiolSoc.,pp.5286–5289,2007. REFERENCES 281 [42] Z. Sun and I. Akyildiz, “Underground Wireless Communication Using Magnetic In- duction,”inICC’09-IEEEInternationalConferenceonCommunications,(Dresden, Germany),pp.4234–4238,2009. [43] S. Hoskins, T. Sobering, D. Andresen, and S. Warren, “Near-field wireless magnetic linkforaningestiblecattlehealthmonitoringpill,”inIEEEEngMedBiolSoc.,2009. [44] Z.Sunandetal.,“MISE-PIPE:MagneticInduction-basedWirelessSensorNetworks for Underground Pipeline Monitoring,” Ad Hoc Networks Journal (Elsevier), vol. 9, no.3,pp.218–227,2010. [45] S. Meybodi, P. Pardo, and M. Dohler, “Magneto-inductive communication among pumps in a district heating system,” in Antennas Propagation and EM Theory (IS- APE),20109thIntlSymp.on,pp.375–378,2010. [46] S. Meybodi, M. Dohler, A. Askarpour, J. Bendtsen, and J. Nielsen, “The Feasibility ofCommunicationAmongPumpsinaDistrictHeatingSystem,”IEEEAntennasand PropagationMagazine,vol.55,pp.pp.118–134,2013. [47] Z. Sun and I. Akyildiz, “Deployment Algorithms for Wireless Underground Sensor Networks Using Magnetic Induction,” in IEEE Global Telecommunications Confer- ence(GLOBECOM2010),vol.10,pp.1–5,2010. [48] J.AgbinyaandM.Masihpour,“Excitationmethodsformagnetoinductivewaveguide communication systems,” in Fifth Int. Conf. on Broadband and Biomedical Commu- nications,(Malaga,Spain),pp.1–6,2010. [49] X.Tan,Z.Sun,andI.Akyildiz,“ATestbedofMagneticInduction-basedCommuni- cation System for Underground Applications,” vol. 2015, to appear in IEEE Maga- zineonAntennasandPropagation. [50] A. Markham and N. Trigoni, “Magneto-inductive networked rescue system (MIN- ERS): taking sensor networks underground,” in IEEE IPSN’12, (Beijing, China), pp.317–328,2012. [51] A.Markham, N. Trigoni, S. Ellwood, and D. Macdonald, “Magneto-inductivetrack- ingofundergroundanimals,”inACMSenSys’10,pp.365–366,2010. [52] A. Markham, N. Trigoni, S. Ellwood, and D. Macdonald, “Revealing the Hidden LivesofUndergroundAnimalsUsingMagneto-inductiveTracking,”in ACM SenSys ’10,(Zurich,Switzerland),pp.281–294,2010. [53] Z. Sun and I. Akyildiz, “On Capacity of Magnetic Induction-based Wireless Under- groundSensorNetworks,”inIEEEINFOCOM2012,(Orlando,USA),March2012. [54] B.GulbaharandO.Akan,“ACommunicationTheoreticalModelingandAnalysisof Underwater Magneto-Inductive Wireless Channels,” IEEE Transactions on Wireless Communications,vol.11,pp.3326–3334,September2012. [55] M. Masihpour, D. Franklin, and M. Abolhasan, “Multihop Relay Techniques for CommunicationRangeExtensioninNear-FieldMagneticInductionCommunication Systems,”JournalofNetworks(JNW),pp.999–1011,2013. REFERENCES 282 [56] S. Kisseleff and et al., “Channel Capacity of Magnetic Induction Based Wireless Underground Sensor Networks under Practical Constraints,” in IEEE WCNC 2013, (Shanghai,China),2013. [57] S. Kisseleff, I. Akyildiz, and W. Gerstacker, “Transmitter-side channel estimation in magnetic induction based communication systems,” in IEEE BlackSeaCom’ 14, (May),pp.16–21,2014. [58] E. Warburg, “Ueber das Verhalten sogenannter unpolarisierbarer Elektroden gegen Wechselstrom,”An.Physik.Chemie,vol.67,pp.494–499,1899. [59] H.FrickeandH.Curtis,“TheDielectricPropertiesofWater-DielectricInterphases,” JournalofPhysicalChemistry,vol.41,no.5,pp.729–745,1937. [60] H. Schwan, “Determination of biological impedances,” Physical techniques in bio- logicalresearch,vol.(chapter6),pp.323–407,1963. [61] H. Schwan and J. Maczuk, “Electrode polarization impedance: limits of linearity.,” inProc.18thAnn.Conf.EngngBiol.Med.,(Washington,DC.),1965. [62] C. Gabrielli and M. Keddam, “Progres recent dans al mesure des impedances elec- trochimiquesenregimesinusoidal,”Electrochim.Acta,vol.19,pp.355–362,1974. [63] B. Onaral, H. Sun, and H. Schwan, “Electrical properties of bioelectrodes,” IEEE TransactionsonBiomedicalEngineering,vol.BME-31,no.12,1984. [64] A.Silva,M.Moghaddam,andM.Liu,“CaseStudyontheReliabilityofUnattended OutdoorWirelessSensorSystems,”inIEEESysCon’15,(Vancouver,Canada),April 2015. [65] A.Mohamed,PrinciplesandApplicationsofTimeDomainElectrometryinGeoenvi- ronmentalEngineering. London,UK:Taylor&FrancisGroup,2006. [66] A.vonHippel,DielectricandWaves. ArtechHouse,1995. [67] A.vonHippel,DielectricMaterialsandApplications. ArtechHouse,1995. [68] W. Kuang and S. Nelson, “Low-frequency dielectric properties of biological tissues: areviewwithsomenewinsights,”Trans.ASAE,vol.41,no.1,pp.173–184,1998. [69] V. Rinaldi and F. Francisca, “Impedance Analysis of Soil Dielectric Dispersion (1 MHz - 1 GHz),” Journal of Geotechnical and Geoenvironmental Engineering, vol.125,pp.111–121,February1999. [70] Agilent Impedance Measurement Handbook - A guide to measurement technology andtechniques. AgilentTechnologies,4thed.,2013. [71] B. Sternberg and T. Levitskaya, “Electrical parameters of soils in the frequency range from 1 KHz to 1GHz, using lumped-circuit methods,” Radio Science, vol. 36, pp.709–719,July2001. REFERENCES 283 [72] R. Sengwa and B. Ram, “Dielectric behaviour of shale and calcareous sandstone of Jodhpur region,” Indian Journal of Radio and Space Physics, vol. 333, pp. 329–335, October2004. [73] M. Oh, Y. Kim, and J. Park, “Factors affecting the complex permittivity spectrum of soilat a lowfrequencyrange of 1KHz -10 MHz,” J Environmental Geology, vol.51, no.5,pp.821–833,2007. [74] H. Kaden, F. Koniger, M. Stromme, G. Niklasson, and K. Emmerich, “Low- frequencydielectricpropertiesofthreebentonitesatdifferentadsorbedwaterstates,” J.ofColloidInterfaceSci.,vol.411,pp.16–26,2013. [75] T.LevitskayaandB.Sternberg,“Laboratorymeasurementofmaterialelectricalprop- erties: Extendingtheapplicationoflumped-circuitequivalentmodelsto1ghz,”Radio Science,vol.35,pp.371–383,April2000. [76] D.Gadani,A.Vyas,andV.Rana,“Dielectricpropertiesofwetandfertilizedsoilsat radiofrequencies,”IndianJournalofPur,vol.52,pp.399–410,June2014. [77] A.CeratoandB.Lin,“Dielectricmeasurementofsoil-electrolytemixturesinamod- ified oedometer cell using 400 khz to 20 mhz electromagnetic waves,” Geotechnical TestingJournal,vol.35,pp.1–9,March2012. [78] M. Carrier and K. Soga, “A four terminal measurement system for measuring the dielectricpropertiesofclayatlowfrequencies,”EngineeringGeology,vol.53,no.2, pp.115–123,1999. [79] B. Mazzeo and A. Flewitt, “Two- and four-electrode, wide-bandwidth, dielectric spectrometer for conductive liquids - Theory, limitations, and experiment,” J. Ap- pliedPhysics,vol.102,pp.104106–1–6,2007. [80] M. M. Gomaa, “Relation between electric properties and water saturation for hematitic sandstone with frequency,” Annals of Geophysics, vol. 51, pp. 801–811, Dec2008. [81] K. Klein and J. Santamarina, “Discussion: Polarization and Conduction of Clay- Water-ElectrolyteSystems,”JournalofGeotechnicalEngineering,vol.122,pp.954– 955,November1996. [82] A. Revil, J. Eppehimer, M. Skold, M. Karaoulis, L. Godinez, and M. Prasad, “Low- frequency complex conductivity of sandy and clayey materials,” J Colloid Interface Sci.,vol.398,pp.193–209,2013. [83] C. Hung-Chi and G. Jaffe, “Polarization in Electrolytic Solutions. Part I. Theory,” J. Chem.Phys.,vol.20,pp.1071–1077,1952. [84] P. Ishai, M. Talary, A. Caduff, E. Levy, and Y. Feldman, “Electrode polarization in dielectric measurements: a review,” Measurement Science and Technology, vol. 24, no.10,pp.1–21,2013. [85] J. Scott, “Electrical and Magnetic Properties of Rock and Soil,” tech. rep., United StatesDepartmentofInterior-GeologicalSurvey,1983. REFERENCES 284 [86] L. Geddes, Electrodes and the Measurement of Bioelectric Events. Wiley- Interscience,1972. [87] H. Schwan, “Alternating current electrode polarization,” J Biophysik, vol. 3, no. 2, pp.181–201,1966. [88] H. Schwan, “Elektrodenpolarisation und ihr Einflub Bestimmung dielektrischer Eigenschaften von Flussigkeiten und biologischem Material,” Zeitschrift Fur Natur- forschung,vol.6b,pp.121–129,1951. [89] L. Geddes, “Historical Evolution of Circuit Models for the Electrode-Electrolyte In- terface,”AnnalsofBiomedicalEngineering,vol.25,pp.1–14,1997. [90] H.Fricke,“TheTheoryofElectrodePolarization,”Phil.Mag.andJournalofScience, vol.14,no.7,pp.310–318,1932. [91] L. Geddes and L. Baker, Principles of Applied Biomedical Instrumentation. John Wiley&Sons,1ed.,1968. [92] H. Kalvoy and et.al., “New Method for Separation of Electrode Polarization Impedance from Measured Tissue Impedance,” The Open Biomedical Engineering Journal,vol.5,pp.8–13,2011. [93] H. Schwan and C. Ferris, “Four electrode null techniques for biological impedance work,”inConf.onEngin.inMedicineandBiology,1963. [94] P. Hoekstra and H. O’Brien, “The dielectric properties of clay suspensions in the frequencyrangefrom50hzto20khz,”tech.rep.,CRREL-CorpsofEngineers,U.S. Army,1969. [95] W.StutzmanandG.Thiele,AntennaTheoryandDesign. Wiley,John&Sons,2ed., 1997. [96] F.T.Ulaby,Fund.AppliedElectromagnetics. PearsonP.H.,5ed.,2007. [97] F.Terman,RadioEngineers’Handbook. McGraw-Hill,1943. [98] F.W.Grover,InductanceCalculations. Dover,2009. [99] A.SilvaandM.Moghaddam,“OperatingFrequencySelectionforLow-PowerMag- netic Induction-Based Wireless Underground Communication,” in IEEE SAS’ 15, (Zadar,Croatia),April2015. [100] A.SilvaandM.Moghaddam,“StrategicFrequencyAdaptationforMid-RangeMag- netic Induction-Based Wireless Underground Sensor Networks,” in IEEE SysCon’ 15,(Vancouver,Canada),April2015. [101] A.Silva,M.Liu,andM.Moghaddam,“Ripple-2: anon-collaborative,asynchronous, and open architecture for highly-scalable and low duty-cycle WSNs,” ACM Mobile ComputingandCommunicationsReview,vol.17,pp.55–60,January2013. REFERENCES 285 [102] X. Jiang, J. Polastre, and D. Culler, “Perpetual environmentally powered sensor net- works,” in Proc. 4th International Symposium on Information Processing in Sensor Networks(IPSN’05),(LosAngeles,CA),pp.463–468,April2005. [103] J. M. Rabaey, M. J. Ammer, J. L. da Silva Jr., D. Patel, and S. Roundy, “Picora- dio supports ad hoc ultra-low power wireless networking,” IEEE Computer, vol. 33, pp.42–48,July2000. [104] IEEE 802.15.4 Standard, “Part 15.4: Wireless Medium Access Control (MAC) and PhysicalLayer(PHY)SpecificationsforLow-RateWirelessPersonalAreaNetworks (LR-WPANs),”IEEE,NJ,USA,2006. [105] ZigBee Alliance, “ZigBee Specifications,” ZigBee Standard Organization, San Ra- mon,CA,USA,2008. [106] A. S. Weddell and et al., “Alternative energy sources for sensor nodes: rational- ized design for long-term deployment,” in Proc. IEEE IMTC’ 08, (Victoria, British Columbia,Canada),pp.1370–1375,May2008. [107] A. Ruzzelli, P. Cotan, G. O’Hare, R. Tynan, and P. Havinga, “Protocol assessment issuesinlowdutycyclesensornetworks: theswitchingenergy,”inProc.IEEEInter- national Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing (SUTC’06),(Taichung,Taiwan),June2006. [108] H.Danneels,V.D.Smedt,C.D.Roover,S.Radiom,N.V.Helleputte,C.Walravens, Z. Li, M. Steyaert, M. Verhelst, W. Dehaene, and G. Gielen, “An Ultra-Low-Power, Batteryless Microsystem for Wireless Sensor Networks,” in In Proc. 26th European ConferenceonSolid-StateTransducers,{EUROSENSOR}2012,2012. [109] “EnerChip Smart Solid State Batteries.” http://www.cymbet.com/products/enerchip- solid-state-batteries.php. [110] R. Misra and C. A. Mandal, “Clusterhead rotation via domatic partition in self- organizingsensornetworks,”inSecondIntConf.onCommunicationSystemSoftware andMiddleware(COMSWARE’07),(Bangalore,India),January2007. [111] J.C.LimandC.Bleakley,“AdaptiveWSNSchedulingforLifetimeExtensioninEn- vironmental Monitoring Applications,” International Journal of Distributed Sensor Networks,vol.2012(286981),pp.1–17,2012. [112] S.J.MarinkovicandE.M.Popovici,“NanoPowerWirelessWake-UpReceiverWith Serial Peripheral Interface,” IEEE Journal On Selected Areas in Communications, vol.29,pp.1641–1647,2011. [113] Atmel Corp, “Sleepwalking helps conserve energy.” http://atmelcorporation.wordpress.com/2013/04/16/sleepwalking-helps-conserve- energy/. [114] P. Dutta, D. Culler, and S. Shenker, “Procrastination Might Lead to a Longer and MoreUsefulLife,”inInProc.HotNets-VI,2007. REFERENCES 286 [115] K. Lu, Y. Qian, D. Rodriguez, W. Rivera, and M. Rodriguez, “Wireless Sensor Net- works for Environmental Monitoring Applications: A Design Framework,” in In Proc.IEEEGLOBECOMM2007,(Washington,DC),pp.1108–1112,2007. [116] G.Halkes,MACProtocolsforWirelessSensorNetworksandTheirEvaluation.2009. ISBN9789090244846. [117] S. Farahani, ZigBee Wireless Networks and Transceivers. Oxford , UK: Elsevier, 2008. [118] “Cc2531usbevaluationmodulekit.”http://www.ti.com/tool/cc2531emk. [119] C. Buratti, A. Conti, D. Dardari, and R. Verdone, “An Overview on Wireless Sensor NetworksTechnologyandEvolution,”Sensors,vol.9,pp.6869–6896,2009. [120] H. R. Bogena, J. A. Huismana, H. Meierb, U. Rosenbauma, and A. Weuthena, “Hy- brid wireless underground sensor networks: Quantification of signal attenuation in soil,”VadoseZoneJournal,vol.8,pp.755–761,August2009. [121] Z. G. Kovacs, G. E. Marosy, and G. Horvath, “Case study of a simple, low powerWSNimplementationforforestmonitoring,”in IEEE Electronics Conference (BEC’10), 12th Biennial Baltic Electronics Conference,(Tallinn,Estonia),pp.161– 164,October2010. [122] L. Selavo, A. Wood, Q. Cao, T. Sookoor, H. Liu, A. Srinivasan, Y. Wu, W. Kang, J.Stankovic,D.Young,andJ.Porter,“LUSTER:WirelessSensorNetworkforEnvi- ronmentalResearch.embeddednetworkedsensorsystems(sensys)(november2007) wu,y.,kang„”inInProc.SenSys’07,2007. [123] M. Pasha and et al., “Toward ultra low-power hardware specialization of a wireless sensor network node,” in Proc. IEEE Intl Multitopic Conference (INMIC’ 09), (Is- lamabad,Pakistan),pp.1–6,December2009. [124] V.Rajendran,K.Obraczka,andJ.Garcia-Luna-Aceves,“Energy-efficient,collision- free medium access control for wireless sensor networks,” in Proc. SenSys’03, (Los Angeles,CA),pp.181–192,ACMPress,November2003. [125] T.Zheng,S.Radhakrishnan,andV.Sarangan,“PMAC:Anadaptiveenergy-efficient MAC protocol for Wireless Sensor Networks,” in Proc. IEEE IPDPS’05, (Denver, Colorado,USA),pp.65–72,April2005. [126] I. Rhee, A. Warrier, M. Aia, J. Min, and M. L. Sichitiu, “Z-MAC: a hybrid MAC for wireless sensor networks,” IEEE/ACM Transactions on Networking (TON), vol. 16, pp.511–524,June2008. [127] S. Mehta and K. Kwak, “H-MAC: a hybrid MAC protocol for wireless sensor net- works,” International Journal of Computer Networks & Communications, vol. 2.2, p.108?117,2010. [128] I. F. Akyildiz and et.al., “Wireless sensor networks: a survey,” Computer Networks Journal(Elsevier),vol.38,pp.393–422,March2002. REFERENCES 287 [129] “XBee DigiMesh 2.4 - XBee-PRO - Wireless connectivity using the DigiMesh pro- tocol.” http://www.digi.com/products/wireless-wired-embedded-solutions/zigbee-rf- modules/zigbee-mesh-module/xbee-digimesh-2-4. [130] F. Stajano and R. Anderson, “The grenade timer: Fortifying the watchdog timer against malicious mobile code,” in 7th Intl Workshop on Mobile Multimedia Com- munications(MoMuC’00),(Tokyo,Japan),2000. [131] P. Dutta and et al., “Trio enabling sustainable and scalable outdoor wireless sensor networkdeployments,”inInProc.IEEEIPSN’06,pp.407–415,2006. [132] M. Demirbas and et. al., “INSIGHT: Internet-sensor integration for habitat monitor- ing,” in Intl. Symp. on a World of Wireless, Mobile and Multimedia Networks (WoW- MoM’06),pp.558–563,2006. [133] S. Brown and C. Sreenan, Software Update Recovery for Wireless Sensor Networks - Sensor Applications, Experimentation, and Logistics, vol. 29. Springer Berlin Hei- delberg,2010. [134] F. Wagner, “VFSM executable specification,” in Proceedings of the IEEE Interna- tional Conference on Computer System and Software Engineering, (The Hague, The Netherlands),pp.226–231,1992. [135] R. Turner and et. al., “ExecSpec: Visually Designing and Operating a Finite State Machine-basedSpacecraftAutonomySystem,”in9thIntlSymp.onArtificialIntelli- gence,RoboticsandAutomationforSpace,(Pasadena,CA),2008. [136] A. Silva and M. Vuran, “CPS 2 : Integration of center pivot systems with wireless underground sensor networks for autonomous precision agriculture,” in ACM/IEEE ICCPS’10,(Stockholm,Sweden),2010. [137] X. Dong, M. C. Vuran, and S. Irmak, “Autonomous Precision Agriculture Through Integration of Wireless Underground Sensor Networks with Center Pivot Irrigation Systems,”AdHocNetworksJournal,vol.11,pp.1975–1987,Nov.20132013. [138] A. Aljaroudi and et. al., “Formulation and Analysis of the Probability of Detection andFalseDetectionforSubseaLeakDetectionSystems,”in10thIntl.PipelineConf., (Calgary,Canada),2014. [139] M. Soto and et. al., “Raman-based distributed temperature sensor with 1 m spatial resolution over 26 km SMF using low-repetition-rate cyclic pulse coding,” Optics Letters,vol.36,no.13,pp.2557–2559,2011. [140] A.SilvaandM.Vuran,“DevelopmentofaTestbedforWirelessUndergroundSensor Networks,”EURASIPJournalonWirelessCommunicationsandNetworking,2010. [141] A. Silva and M. Moghaddam, “Adaptive sub-MHz magnetic induction-based sys- tem for mid-range wireless communication in soil,” in to appear in Proc. IEEE-APS APWC’15,2015.

Abstract (if available)

Abstract Low‐power wireless communication in underground settings and confined areas is considered one of the last frontiers in communications. First, there is the energy challenge: because the nodes are embedded in some kind of medium, such as soil, concrete, or debris of a disaster event, many of the traditional energy solutions for communication devices are ruled out. Moreover, typical radio wave‐based solutions for over‐the‐air communication are significantly impacted in underground settings due to the very strong signal attenuation in lossy medium. ❧ To address such challenges, Wireless Underground Sensor Networks (WUSNs) have been proposed. WUSNs potentially enable a wide variety of novel applications mainly in the areas of precision agriculture (PA), concealed electronic fence (border patrol and residence security), and disaster management (landslide detection, underground structure monitoring). So far, the hope is that the combination of collaborative networking protocols with a high node density can mitigate the mentioned issues of underground nodes regarding energy and signal attenuation. However, the lack of real‐world WUSN deployments is a warning that some WUSN aspects still need to be addressed. ❧ In this research work, the theory and design of a real‐world WUSN based on a technique called magnetic induction (MI) are considered. These studies are divided into three main parts. First, a novel soil dielectric model for low frequencies (i.e., 1 kHz to 200 kHz) tailored to MI-based WUSNs is proposed. This preliminary study in than extended to include a novel methodology to identify and separate the electrode polarization (EP) effects from the dielectric measurements at the sub‐MHz range. The second part of this dissertation work is related to physical (PHY) layer of MI-WUSNS: a MI signal attenuation model is developed and important design strategies involving frequency and coil adaptation schemes are proposed. Such solution allows the MI node to dynamically adapt considering energy resources, application bandwidth, and soil conditions. ❧ While the mentioned parts are related to the PHY layer of a peer‐to‐peer communication system, the concluding part of this work extends the solution by including upper networking layers. Accordingly, a cross‐layer protocol is proposed as a way to achieve very high energy‐efficiency. In this work, both theoretical and empirical results are considered. Preliminary results show a good agreement between the empirical evaluations and the proposed models.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Magnetic induction-based wireless body area network and its application toward human motion tracking

PDF

Solution of inverse scattering problems via hybrid global and local optimization

PDF

Active state tracking in heterogeneous sensor networks

PDF

Multichannel data collection for throughput maximization in wireless sensor networks

PDF

Learning and control for wireless networks via graph signal processing

PDF

Communication and cooperation in underwater acoustic networks

PDF

Relative positioning, network formation, and routing in robotic wireless networks

PDF

Fundamental limits of caching networks: turning memory into bandwidth

PDF

Optimal distributed algorithms for scheduling and load balancing in wireless networks

PDF

Structured codes in network information theory

PDF

Exploiting side information for link setup and maintenance in next generation wireless networks

PDF

Utilizing context and structure of reward functions to improve online learning in wireless networks

PDF

Signal processing for channel sounding: parameter estimation and calibration

PDF

Novel techniques for analysis and control of traffic flow in urban traffic networks

PDF

Compiler and runtime support for hybrid arithmetic and logic processing of neural networks

PDF

Coded computing: Mitigating fundamental bottlenecks in large-scale data analytics

Asset Metadata

Creator Silva, Agnelo Rocha da (author)

Core Title Theory and design of magnetic induction-based wireless underground sensor networks

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 07/15/2015

Defense Date 06/19/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Best‐Effort Time Slot allocation protocol,dielectric properties of soil,dielectric spectroscopy,disaster management,electrochemical impedance spectroscopy,electrode polarization,magnetic impedance spectroscopy,magnetic‐induction,near‐field communication,non‐linear region of electrode polarization,OAI-PMH Harvest,pipeline leak detection,precise irrigation,sub‐MHz impedance spectroscopy,ultra‐low power management,wireless sensor networks,wireless underground communication,wireless underground sensor networks

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Moghaddam, Mahta (committee chair), Hashemi, Hossein (committee member), Jafarpour, Behnam (committee member), Mitra, Urbashi (committee member)

Creator Email agnelors@gmail.com,agnelosi@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-595708

Unique identifier UC11301298

Identifier etd-SilvaAgnel-3604.pdf (filename),usctheses-c3-595708 (legacy record id)

Legacy Identifier etd-SilvaAgnel-3604.pdf

Dmrecord 595708

Document Type Dissertation

Format application/pdf (imt)

Rights Silva, Agnelo Rocha da

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA