DEGREE PROJECT, IN , SECOND LEVELCOMPUTER SCIENCESTOCKHOLM, SWEDEN 2015

Indoor Positioning Utilizing BluetoothSmartA COMPARATIVE STUDY BETWEENTRILATERATION AND FINGERPRINTING

NIKLAS NYGÅRD

KTH ROYAL INSTITUTE OF TECHNOLOGY

CSC

IndoorpositioningutilizingBluetoothSmart

InomhuspositioneringmedBluetoothSmart

Acomparativestudybetweentrilaterationandfingerprinting

NiklasNygård

nnygard@kth.se

Master’sThesisattheSchoolofComputerScienceandCommunication

ComputerScienceandCommunicationandIndustrialEngineering

SupervisoratMobiento:AndreasBergman

SupervisoratCSC:MichaelMinock

Examiner:JohanHåstad

December03,2015

AbstractThis thesis explores the possibilities of constructing an indoor positioning system based onBluetooth Smart technology. Two non-trainable trilateration approaches and two trainablefingerprintingwereimplementedandevaluatedatMobiento'sofficesinStockholm,Sweden.

A trilateration approach is based on finding a sought location based on known distancestowards know locations, at least three locations and distances are needed. A fingerprintingapproach is based on creating a radiomap, which describes transmission signals within theroom, towardsdifferent transmitters.A set amount of coordinates are assigned a fingerprint.Thesearethenusedasreferencepointsforasoughtlocation.

For eachmajor approach, trilateration and fingerprinting, aweighted approach is conducted.These approaches are evaluated in a disturbance free environment in term of accuracy,implementation and setup. In terms of accuracy, the non-weighted fingerprinting approachperforms slightlybetter than theweighted fingerprintingapproaches.Bothof thesearemoreaccurate than the trilateration approaches.When it comes to implementation and setup, thetrilateration algorithms impose less cost. These allow for better scalability when the indoorenvironmentbecomeslarger.

SammanfattningDetta examensarbete undersökermöjligheterna att skapa ett inomhusnavigeringssystemmedhjälp av Bluetooth Smart tekniken iBeacons. Två trilaterations algoritmer samt tvåfingerprintingalgoritmerkonstruerasochutvärderaspåMobientoskontoriStockholm,Sverige.Entrilaterationsalgoritmbyggerpåatthittaensöktplatsirummetbaseratpåattmanharettkänt avstånd till kända sändare i rummet, åtminstone tre sändare och avstånd behövs. Enfingerpriting algoritm bygger på att skapa en radiokarta över rummet. Denna beskriversignalstyrkatillolikasändareutifrånolikareferenspunkterirummet.Närmansökerenpunktjämförssedantestpunktenmeddendatabasavreferenspunktermanharsattuppföratttaredapåpositionen.Trilaterationochfingerprintingalgoritmerernaprövasävenmedenvarsinviktadmetod.Utvärderingen av alla algoritmerna sker i en störningsfrimiljö, ochutvärderaspå kriteriernanoggrannhet, implementation och installation. När det gäller noggrannhet presterade bådafingerprintingalgoritmernabättreäntrilaterationsalgoritmerna.Trilaterationsalgoritmenvarfördelaktig när det handlade om implementation och installation. System baserade påtrilaterationärenklareattinstalleraochgerbättraskalbarhetnärdeblirstörre.

AbbreviationsAngleofArrival-AoA

BluetoothSpecialInterestGroup-BluetoothSIG

Decibel-milliwatts-dBm

DeviceAccessCode-DAC

GlobalPositioningSystem-GPS

IndoorPositioningSystem-IPS

Industrial,ScientificandMedicalband-ISM

LineofSight-LoS

LocationBasedServices-LBS

Non-LineofSight-NLOS

ReceivedSignalStrengthIndicator-RSSI

TimeDifferenceofArrival-TDoA

TimeofArrival-ToA

UniversallyUniqueIdentifier-UUID

Tableofcontents1Introduction.................................................................................................................................11.1Background.........................................................................................................................................11.2Purpose.................................................................................................................................................21.3Delimitations.......................................................................................................................................21.4Researchquestions...........................................................................................................................31.5Outline...................................................................................................................................................3

2Literaturereview........................................................................................................................42.1Conceptualframework....................................................................................................................42.1.1Bluetooth..........................................................................................................................................................42.1.2BluetoothSmart.............................................................................................................................................52.1.3Commercialsystems....................................................................................................................................72.1.4Receivedsignalstrengthindicator........................................................................................................82.1.5Positioningtechniques................................................................................................................................82.1.6Relatedfields................................................................................................................................................10

2.2Theoreticalframework................................................................................................................102.2.1GSM/Mobile..................................................................................................................................................112.2.2WirelessSensorNetworks.....................................................................................................................122.2.3Wi-Fi.................................................................................................................................................................132.2.4Bluetooth.......................................................................................................................................................152.2.5BluetoothSmart..........................................................................................................................................162.2.6Theoreticalframeworksummary.......................................................................................................17

3Implementation........................................................................................................................183.1Testbed..............................................................................................................................................183.1.1Roomspecifics.............................................................................................................................................183.1.2Hardware.......................................................................................................................................................213.1.3Beaconplacement......................................................................................................................................22

3.2Positioningtechniques.................................................................................................................223.2.1Trilateration.................................................................................................................................................223.2.2Fingerprinting..............................................................................................................................................25

3.3Datacollection.................................................................................................................................273.3.1Concept...........................................................................................................................................................273.3.2Software.........................................................................................................................................................28

4Results..........................................................................................................................................294.1Trilateration.....................................................................................................................................294.1.1Regulartrilateration.................................................................................................................................294.1.2Weightedtrilateration.............................................................................................................................31

4.2Fingerprinting.................................................................................................................................324.2.1Regularfingerprinting.............................................................................................................................324.2.2Weightedfingerprinting..........................................................................................................................33

5Discussion...................................................................................................................................375.1Trilateration.....................................................................................................................................375.2Fingerprinting.................................................................................................................................38

5.3Trilaterationandfingerprinting...............................................................................................395.4AdvantagesofBluetoothSmartandbeacons........................................................................405.5Errorfactorsandsources............................................................................................................41

6Conclusionsandfuturework...............................................................................................436.1Conclusions.......................................................................................................................................436.2Futurework......................................................................................................................................44

References......................................................................................................................................45

Appendix.........................................................................................................................................47AppendixA...............................................................................................................................................47AppendixB...............................................................................................................................................48AppendixB1............................................................................................................................................................48AppendixB2............................................................................................................................................................49AppendixB3............................................................................................................................................................50

FiguresFigure2.1BluetoothSmartstackarchitecture............................................................................................................6

Figure2.2Basicsoftrilateration........................................................................................................................................9

Figure3.1Testbed:Roommeasurments...................................................................................................................19

Figure3.2Testbed:Roompicture1.............................................................................................................................20

Figure3.3Testbed:Roompicture2.............................................................................................................................20

Figure3.4Testbed:Roompicture3.............................................................................................................................21

Figure3.5Testbedspecifics.............................................................................................................................................24

Figure3.6Trilaterationapproach..................................................................................................................................25

Figure3.7Weightedtrilaterationapproach..............................................................................................................27

Figure3.8Fingerprintingapproach..............................................................................................................................26

Figure3.9Weightedfingerprintingapproach..........................................................................................................27

Figure3.10Recordingsoftware......................................................................................................................................28

Figure4.1Pullpercentage,testpointone..................................................................................................................34

Figure4.2Pullpercentage,testpointtwo..................................................................................................................35

Figure4.3Pullpercentage,testpointthree...............................................................................................................36

TablesTable3.1Beaconplacement.......................................................................................................................................22

Table4.1Trilateration:Beaconstrength..............................................................................................................29

Table4.2Trilateration:Pathlossconstant..........................................................................................................30

Table4.3Trilateration:Calculateddistances.....................................................................................................30

Table4.4Trilateration:Calculatedpositions......................................................................................................30

Table4.5Weightedtrilateration:Intialvalues,K-vales,roomconstants..............................................31

Table4.6Weightedtrilateration:Calculateddistances.................................................................................31

Table4.7Weightedtrilateration:Calculatedpositions..................................................................................32

Table4.8Weightedtrilateration&trilateration:Comparisontable........................................................32

Table4.9Fingerprinting:Testpointone..............................................................................................................33

Table4.10Fingerprinting:Testpointtwo...........................................................................................................33

Table4.11Fingerprinting:Testpointthree........................................................................................................33

Table4.12Fingerprinting:Calculatedpositions...............................................................................................33

Table4.13Weightedfingerprinting:Calculatedpositions...........................................................................36

Table5.1AverageRSSIatreferencepointscomparedtoreferencevalues..........................................41

Introduction

1

1 Introduction

1.1BackgroundInrecentyears,sincetheintroductionofsmartphones,manysuccessfullocationbasedservices(LBS) formobile deviceshave emergedon themarket.Among these services there is a greatamount of different navigation applications, but also other applications that for instance liststhenearestATMsorcafes.Acommondenominatoramongtheseimplementationsisthattheyhavebeenenabledbytheglobalpositioningsystem(GPS)andthedevelopmentofsmartphones.ApplicationprogramminginterfacessuchasGoogleMapsorMapsfor iOSallowdeveloperstoeasily implement LBS into their applications. Implementing LBS in a mobile application hasbeensimplifiedandcanbeviewedasatriflewhencomparedtootherprocessesinthemobileapplicationdevelopment.

The global positioning system uses time of arrival (ToA) to calculate a signal into distance.Embeddedwithineachtransmissionfromthesatellitesisthetimeoftransmission.Thedistancefroma satellite to the receiver can thenbe calculated since the signals travel at the speedoflight.Usingtrilaterationinthreedimensions,anintersectioncanbeextractedfromfoursurfaceareasgivenby thespheres formedby thecalculateddistances from foursatellites.Using foursatellitesisrequiredtoproducealongitude,latitudeandaltitudeposition.Withthreesatellitesone can produce a longitude and latitude position, and using more than four satellites willproduceamorepreciseposition.Whenthe lineofsight isclear, thesystemworkswelland isveryprecisewiththehelpofsupportsystemssuchasDGPSandEGNOS.

However, in non-line of sight (NLOS) conditions, for instance in urban areas, calculatingdistance from the satellite signal can become unreliable. In areas like these, the signal isexposedtomanydisturbances,likebouncinguponstructuresaswellaspenetratingawall.Bothbouncing on a wall and traveling through a thick wall will cause a spoof on the perceivedEuclideandistancebetweenthesatelliteandreceiverwhencalculatingit.Conditionslikethesemake it problematic to utilize GPS indoors when precision is desired. An Indoor positioningsystem (IPS) is a solution designed to locate objects and people indoors. There are manydifferentapproachesandsolutionstohowthesecanbedesigned.

The range ofmobile applications that utilizes indoor positioning is small when compared toapplicationsthatutilizingtheglobalsatellitenavigationsystem,GPS.IfLBSindoorsistobecomeaswellestablishedandcustomaryasLBSoutdoors,thereneedstobesomeleapsintechnologyin termsof standardization,priceandprecision.Therearemany localdeploymentsof indoorpositioningimplementationsbuteachindoorenvironmentusuallyhasitsownuniquetopology,whichentailsdifferenttermsofconditionswhendeployingtransmitters.Furthermore,thereisnoacknowledgedstandardonwhichsystemthatshouldbeusedwhenconstructinganIndoorpositioningsystem.TherehasbeenalotofresearchinutilizingWi-FiandalsosomeinutilizingBluetooth. In addition to easy deployment, an IPS should also be cheap to deploy. If theconditionsstandardization,pricingandprecisioncanbemetbyasystem,thiscouldallowthemarketforLBSindoorstomaturelikeithasforLBSoutdoors.

OnaninitiativestartedbyNokia,thefirstdevelopmentofwhatwouldbecomeBluetoothSmartwas launched in 2001. The technology derived from the ordinary Bluetooth stack but wasdevelopedforthepurposeofminimizingthepowerconsumption.Itwasfirstreleasedunderthe

Introduction

2

nameBluetooth LowEndExtension and later renamed toWibree. In 2010, itwas integratedcompletely in the new Bluetooth specifications after agreements with the Bluetooth SpecialInterestGroupunderthenameBluetoothSmart.Apartfromtheadvantagesofabetterpowerconsumption compared to regular Bluetooth, Bluetooth Smart also has a new broadcastingfeaturewhichallowsaBluetoothSmartunittosendshortbroadcastingmessages.Thisfeatureallowsatransmittertosendmessagesouttoanyreceiverthatmightbelistening.

WiththereleaseoftheBluetoothSmartstack,manycompanieshavestartedtoproducebeaconsutilizing this technology. A beacon is a transmitter that advertises its universally uniqueidentifier (UUID) on a set interval. Apple has created and patented a protocol forcommunicationwith Bluetooth Smart devices called iBeacons. The primary purpose of thesebeacons is proximity-based services like triggering a certain action when entering a certainrange of a beacon, but they can also be used for the purpose of indoor positioning. In thisMaster’sThesis,itisinvestigatediftheBluetoothSmartstackissuitedforthepurposeofindoorpositioning.Morespecifically,thestandardsandspecificationsintroducedinthecommerciallyestablishedprotocoliBeaconareinvestigated.

1.2PurposeWiththeemergenceofmanynewbeaconproducers,thestrongpotentialoftheBluetoothSmartstack andquick establishment for the relativelynewprotocol on themarket, thedemand forbeaconservicesonmobilefirmssuchasMobientohasincreasedrapidlythepastyear.Severalclientswishtoincorporatebeaconswithinapplicationsthattheyorderbuttheyusuallydonothave any specifics onwhatorhow these services shouldbe implemented.There is aneed toinvestigate the possibilities with beacons in order to meet these demands. InvestigatingBluetooth Smart’s abilities towork as an indoor positioning system, input and knowledge toMobiento concerning how the company can meet its customers’ demands as well ascontributingtotheresearchfieldofIPSisprovided.

In this thesis, the performance of a non-trainable trilateration approach versus a trainablefingerprinting approach in a disturbance free environment utilizing the Bluetooth Smarttechnology known and trademarked as iBeacons is investigated. To implement thesepositioningtechniques,receivedsignalstrengthindicator(RSSI),aninheritedBluetoothSmartpropertythatisaccessibleinxcodeisutilized.

1.3DelimitationsThere is a set of delimitations established for this master’s thesis. In this section, these willbrieflybedescribed.

Thereisnohardwareinvestigation,theimpactofchoicesofhardwarearenottested.Differentbeaconproductsandevendifferentbeaconsofthesamemodelmaybehavedifferentlywhenitcomes to transmission strength. This is also the case when it comes to different receivingdevicessuchasphones.Therefore, thereceivingand transmittingdevices isalways thesame.Anotherdelimitation isconductingan investigationofRSSI.The thesisrelieson the literatureforthepurposeofgatheringinformationaboutRSSIbehaviour.

Introduction

3

1.4ResearchquestionsIn order to examine if iBeacons and Bluetooth Smart are suitable for indoor positioning, thefollowingmainresearchquestionhasbeenformulated.

IsBluetoothSmartaviablechoiceforthepurposeofbuildinganindoorpositioningsystem?

Toanswerthemainresearchquestion,thefollowingsub-questionsareformulated.

1. What positioning algorithms can be utilized within Bluetooth Smart and similar radiotechnologies?

2. Howdoes thesepositioningalgorithmsperformagainsteachother in termsofaccuracy,setupandimplementation?

3. WhatadvantagesanddisadvantagesdoesaBluetoothSmart system implementing thesepositioningalgorithmshavecomparedtootherradiotechnologies?

1.5OutlineFirst a literature review is presented with the theoretical and conceptual framework. Theconceptualframeworkpresentsbasicsideasandknowledgethatarenecessarytohaveinorderto grasp the study.The theoretical frameworkpresentsprevious studies in similar fields andhow positioning approaches were implemented in those cases. Furthermore, with thetheoreticalframework,thetoolstoimplementthestudyarepresentedandselected.Followingthe conceptual and theoretical framework, the implementation is presented. Theimplementationbuildsuponwhatislearnedinthetheoreticalandconceptualframework.Afterthe implementations, the results are presented. Following this, a results discussion andfeedbacktoresearchquestions.Finally,conclusionsandthoughtsonfutureworkarepresented.

Literaturereview

4

2 LiteraturereviewThe literature review chapter is split up into two parts, the conceptual and theoreticalframework. The former part, the conceptual framework, provides insight into the underlyingstructures e.g. the Bluetooth stack and the basic principles of positioning techniques. Theseconcepts are fundamental knowledge to the reader if theywish to grasp thewhole setting inwhich this research takes place. The latter part, the theoretical framework describes thedifferentscientificreportsfromwhichthetheoriesonwhichthisworkbuildsuponhavebeengathered.Theseworksaresummarizedandimportantaspectshavebeenhighlighted.

2.1Conceptualframework

2.1.1BluetoothIn1994,engineersatEricssondevelopedthefirstversionofwhatwastobecomeBluetooth.Theoriginal purpose of this development was to create a wireless headset and replace variouscablesassociatedwithacellphone.Sincethen,Bluetoothhasgrownintomuchmorefieldsandis now utilized in many applications, but the common denominator is of course, wirelesscommunication. In 1998 Ericsson, Nokia, Intel, IBM and Toshiba, decided to collaborate andformed the Bluetooth Special Interest Group (Bluetooth SIG).[1] Since then, nobody owns theBluetooth technology. The Bluetooth SIG is a cooperation between over 25000 membercompanies.[2]

TousetheBluetoothintellectualpropertyinitsproducts,acompanymustbecomememberofthe Bluetooth SIG. This group's purpose is to define the standards and specifications in theBluetooth stack. Among these, is a range of different settings and profiles developed andtailored for special purposes such as audio streaming and remote control functionality. Thelatest version of Bluetoothwas released onDecember 02, 2014.[3]The latest update includessomeimprovementstowardssupportingthedevelopmentoftheinternetofthingsconcept.[4]

TheBluetoothcorespecificationprovidesdeveloperswithbothlinklayerandapplicationlayerdefinitions.Bluetoothwirelesscommunicationoperatesintheindustrial,scientificandmedicalband (ISM) at 2.4 - 2.85 GHz. The technology utilizes amethod called frequency hopping tominimize interference fromother radio technologies operating in the sameband.Thesehopsare usually performed at a rate of 1 600 hops per second. This is called adaptive frequencyhopping, thealgorithmappliedsearches forotherdevices inthedefinedspectrumandadaptsits hopping sequence to minimize interference. The range that Bluetooth operates within isrelative short compared to other radio communications. TheBluetooth SIGhasdefined threerangeclassesforBluetoothradios.TheseareClass1,Class2andClass3witharangesetof100meters,10metersand1meterrespectively.[5]

To enable two different devices to communicate with each other, usually a pairing or aconnectionneedstobeperformed.Notalldevicesaredesignedtobeabletobeconnectedwitheachother, forinstanceaBluetoothheadsetmightnotneedtobeabletocommunicatewithaBluetoothkeyboard.For thispurpose, thecommunicationsbetweenBluetoothdevicesalwayshaveonemasteranduptosevenslavedevices.Aslavedevicecanonlycommunicatewith itsmasterandnottoeachother.Somedevicescan,however,switchrolesbetweenbeingamasterand being a slave. Others are only designed to be a slave. In the previous example with the

Literaturereview

5

Bluetoothheadsetandkeyboard, there isreallynoneedforanyof themtobeabletoactasamaster.

TheconnectionbetweentwoBluetoothdevicesisathreestepprocess:[6]

1. Inquiry-Thisisthegreetingprocesswheretwodevicesexchangeaddressinformationwith each other. A Bluetooth radio constantly sends out inquiry messages on the 32different defined channels available forBluetooth communication, the total time for acomplete inquiry scan cycle is 10.24 seconds.[7] After two devices have exchangedaddressestheyareconsidereddiscovered.

2. Paging-Thisisthenextstepinformingaconnection.First,theintendedmastersendsouta“Page”.Thereafter,theintendedslavereplieswiththemaster’sdeviceaccesscode(DAC).Thenthemasterreplieswith itsownfrequencyhopsequencesotheslaveunitcancommunicatewithit.Finally,theslavereplieswithitsownDACandswitchesovertothemaster’schannel.

3. Connection-Afterthedeviceshavefinishedthepagingprocess,theyareconnected.Inthisstate,themasterdevicecancommunicatewiththeslavedevices.Themasterdeviceprovides its slaveswith a threebit identification number,which limits thenumber ofdevicesamastercancontroltoseven.Themastercanputitsslavesindifferentmodes,amongtheseareanactivemode,whichisthenormalmodeforcommunicationandasetofdifferentsleep/powersavemodes.Oneoftheseareholdmode,whichputstheslaveintosleepforasetamountoftime.

Two Bluetooth units can be paired. In this process, the involved units save each otherinformation in termsofaddresses,namesandprofiles.When theyare inproximitywitheachother, they can then be activated automatically. The pairing process can be done in differentways.Somepairingprocessesarejustanautomaticoperationwhileothersrequiresomesortofauthenticationtobeenteredinoneunittoclaimtheotherunit.

2.1.2BluetoothSmartNokia, as described in the introduction chapter, introduced the first version of what wouldbecome Bluetooth Smart, in 2001. This technology derived from original Bluetooth with thepurposeofminimizingpower consumptionandoptimization towards communicationswherelessdataisbeingtransferred.In2010,theBluetoothSIGincludedBluetoothSmartinthenewlyreleased Bluetooth 4.0 stack. Bluetooth Smart is not backwards compatible with regularBluetooth. Consequently, there are three types of Bluetooth devices now; Bluetooth Smart,BluetoothSmartReadyandBluetooth.ABluetoothdeviceandaBluetoothSmartDevicecannotcommunicatewitheachother,butdevicesthatare“BluetoothSmartReady”cancommunicatewithbothtypesofBluetooth.Mostofthenewersmartphonesareforinstance“BluetoothSmartReady”.Laptopsare laggingbehind in introducingBluetoothSmartReadyradios.[8]Applehasnow,however,introducedthetechnologyinalltheirmajorbrandsofcomputers.

TheBluetoothSmartstackhas,asstatedbefore,derivedfromtheregularBluetoothstackandconsists of a controller stack and a host stack. The controller stack handles the lower levelcommunication layers and the physical radio while the host stack handles higher level ofcommunications and protocols. All profiles and applications that are used sits on top of theGeneric Access Profile and the Generic Attribute Profile. A short description of the protocolstack,frombottom-up,ispresentedbelowFigure2.1.[9]

Literaturereview

6

Figure2.1BluetoothSmartstack’sarchitecture.

PhysicalLayer–Consistsof the1Mps frequencyhoppingradiooperating in the2.4GHz ISMband.

LinkLayer - Controls the radio frequency state of the device. There are five different states.Thesearestandby,advertising,scanning,initiatingandconnected.TheadvertisingpartiswhatmakesthetechnologyinterestingforthepurposeofcreatinganIPS.Advertisingiswhatenablesthebroadcastingfunctionalitymentionedintheintroduction.

Host - Controller Interface - Provides standardized means of communication between thecontrollerstackandthehoststack.ThislayercanbothbeimplementedasasoftwareAPIorasahardwareinterfacee.g.USBdependingonthesetup.

Logical Link Control andAdaptionProtocol - Provides data encapsulation services for theupperlayers.

SecurityManager-Definesmethodsforsecureconnectionfortheotherlayerstowardsotherdevices.Definesthemethodsforpairingandkeydistribution.

Generic Access Profile - Interacts with the overhead profiles and applications. Handles theinitiationofthesecurityprotocolsdefinedbythesecuritymanager.

AttributeProtocol-Defineswhattypeofattributesthedeviceisallowedtosharewithotherdevices.

Generic Attribute Profile - This framework defines the procedures for using the AttributeProtocol. It defines what attributes overhead profiles and applications are allowed to use intheircommunicationwithotherdevices.

AdvertisingwithinBluetooth Smart can have several purposes besides just broadcasting. Forinstance, transmitting data to a previously paired device and advertising presence to other

Literaturereview

7

devices. The key enabling factor for this research is however the broadcasting functionality,whereadeviceisjustsimplyadvertisingit’sUUIDatasetintervalcontinuously.ThisiswhatthebeaconandiBeaconconceptsbuilduponwhichisdiscussedinthenextsection.

2.1.3CommercialsystemsThecommercialsystem,onwhichthisresearchhasbeenconducted,buildsupontheBluetoothSmartstack.ThesesystemsareiBeaconandEstimote.iBeaconwasintroducedbyAppleontheWorld Wide Developers Conference in June 2013. Utilizing the iBeacon protocol, a start-upcompany called Estimote, has released their own commercial beacons. In the next twosubsections,iBeaconandEstimoteisbrieflydescribed.

2.1.3.1iBeaconiBeaconisaprotocolforcommunicationswithbeaconsutilizingBluetoothSmart.Withinxcode,severalmethodstohandleandlocateBluetoothSmartbeaconsareprovided.ThemainpurposeofiBeaconsisproximity-basedservices.Aproximitybasedservicetriggersanactionbasedonadevice entering a defined proximity zone. For instance, pushing different offers based onenteringdifferentzoneswhilestrollingthroughastore.Proximitybasedservicesarenotlimitedto retail, there is plenty of potential in other sectors such as healthcare, transportation andevents.

WiththelowenergyconsumptionthatisinheritedfromutilizingBluetoothSmart,abeaconcanadvertise for years. The frameworkprovides three built in rangingmeasurements. These areimmediate(withinacoupleofcentimeters),near(withinacoupleofmeters)and far (greaterthan 10meters). These predefinedmeasurement values derives frommeasured RSSI values.Thisstudydoesnotrelyonthesepredefinedmeasurements,bututilizesRSSI,whichcanalsoberetrievingintheprotocol.

Byutilizingtherangingmechanics,proximityzonesdefinedbythepredefinedproximitiescanissuealertstobepushedtoadeviceandanapplication,eveninthebackground.Suchanalertcan enable triggers such as launching a specific application or just passing a notification. Tocreate an IPS utilizing iBeacon, the RSSI feeds from the beacons needs to be constantlymonitoredandtheapplicationhastobeactiveintheforeground.TrademarkedandpatentedbyAppleInc.,theiBeaconconceptisnotlimitedtoiOSusersonly,sincethereleaseofAndroid4.3thefirstofficialsupportforiBeaconisintroduced.[10]

2.1.3.2EstimoteEstimoteisaproducerandsellerof itsownbeaconsthatarebasedonandutilizetheiBeaconprotocol. The company also provides its own SDK that provides the user with easy tools toutilize these beacons. However, these are not utilized in the report. Estimote is constantlyintroducing new functionalities to this SDK. In fact, Estimote did release its own IndoorPositioning System with a separate SDK this fall. Estimote do not reveal much of how itsalgorithmisimplemented,whichisreasonabletomaintainacompetitiveedge.Thecompanydorevealthatitisacombinationoftrilateration,particlefilteringandsensorfusing.[11]

“Allofthecomplicatedworkisdoneunderthehoodbyasystemourengineersanddatascientistsspent months of engineering time developing, incorporatingmethods like trilateration, particlefilteringandsensorfusion.”

Literaturereview

8

In this comparative study, the performance of a trilateration approach and a fingerprintingapproach is evaluated. An interesting angle would also be to compare it with this recentlyintroducedsystembyEstimote.However,thisiscomparisonisleftoutfortworeasons.Firstly,thereislittleinsightintohowexactlythealgorithmworks.Secondly,thereisanextrachargetopurchasethisseparateAPI.

2.1.4ReceivedsignalstrengthindicatorAs briefly discussed in the introduction, the RSSI parameter can be utilized to determinedistance.Thiscanbedonebyutilizingtherelationshipbetweenthereceivedsignalstrengthatatestpoint,andthereceivedsignalstrengthatthedistanceofonemeterfromthetransmittingbeacon. In [16][17][23] formulas on how to perform these calculations and the overallproperties of this measurement, are presented. Besides the two measurements of RSSI, aconstant that describes the signals’ ability to propagate in the room, this constant can becalculatedifyouuseaknowndistancetoabeacon.

Throughoutthetheoreticalframework,RSSIasadecidingparameterforpositioningtechniquesis described as an uncertain factor. But to be able to construct an IPS built on the iBeacontechnology, RSSI is the only available tool to determine distance and enable utilizing thepositioningtechniquesdescribedinthenextsection.

2.1.5PositioningtechniquesIntheprevioussectionaboutcommercialsystems,itwasdescribedthattheiBeaconsprimarypurposeisproximitybasedservices.TheseservicesarenottobeconfusedwiththetermIndoorPositioningSystem.Inthisreport,referringtoIndoorPositioningSystemreferstoasystemthatactuallydeterminesapositionandnotadistancetoaknownlocation.Intheliteraturestudyitwas revealed that there are two primarymethods that fits this purpose. These are the non-trainabletrilaterationapproachandthetrainablefingerprintingapproach.

2.1.5.1TrilaterationTrilateration isoneof thetwokeypositioningtechniquesthathavebeen implemented in thisresearch. It is interchangeably confused with triangulation, even in research papers. Thedifference between these two positioning techniques is that triangulation utilizes angles todeterminepositionwhileastrilaterationutilizesdistances.

Inordertoconducttrilaterationintwodimensions,oneneedsatleastthreeknowndistancestothreeknownreferencepoints.Fromonereferencepointandaknowndistance,aperimetercanbe identified. This is a circle of possible positions surrounding the reference point.With tworeference points and known distances, two intersecting points can be identified, thus twopossiblepositions.Withthreereferencepointsandknowndistances,thethirdreferencepointanddistancecanruleoutthefalsepossiblepositionfromthecasewithonlytwo.Underperfectconditions, the perimeters formed from the distance to each of the three reference points,intersect at a one point, as seen in Figure 2.2. In this study, exact distances and perfectlyintersectingperimetershaveusuallynotbeenthecase,whichmeansabest-fitapproachneedsto be applied. Throughout the literature review presented in this chapter, least squareestimation has been the dominant approach for this. This is further elaborated upon in thetheoreticalframework.

Literaturereview

9

Figure2.2BasicsofTrilateration.Solve(x,y)tofindlocation.

There are several ways to determine distances towards these references points. In GPS, anapproachcalledToAwasused.Inthisresearch,suchcalculationsweremoredifficultgiventheframework not providing the tools to use a ToA approach. The only option to performtrilaterationistoutilizethelossofsignalstrengthtoestimatedistancetobeacons.

2.1.5.2FingerprintingA fingerprinting algorithm or approach is a trainable positioning technique that has someresemblances with radar technology. Radar repeatedly maps the landscape, and observeschangeswithin it.Thechangescanbemovingobjects,or if theplatformitselfmoves itcanberelative changes within the landscape. A fingerprinting algorithm builds on the principles ofcreatingaradiomap.Thisradiomapisasetofrecordedreferencepointsatdifferentlocations.InthecaseofcreatinganIPSbasedonbeacons,therecordeddataconsistsoftheRSSIfromeachbeaconatthesepoints.Thisradiomapisnotconstantlyupdatedlikeregularradartechnology,butisnowadatabaseofreferencepoints.Whenusersofthesystemwanttolocatethemselveswithinit,theyusethisdatabaseofreferencesystemtofigureoutwhatreferencepointstheyareclosest to. There are several methods to do this, in the theoretical framework two differentapproaches,usedinpreviousstudies,areaccountedfor.[21][23]Thebasicsarethesamehowever,theybothutilizestheEuclideandistancefromthetestedpointtothereferencepoints.Afterthis,theapproachesdifferintermsofhowtheyhandleanddecidewhatpointsaretheclosest.Thereferencesetsaredifferentinthesereports,[21]haveadatasetofeveryrecordedinstanceineach position while [23] utilizes a mean value. This means that in [21] a test point can bematchedtothesamecoordinatesseveraltimesandin[23]itcannot.Inthisthesis,anapproachmoresimilarto[23]isutilized,whichisfurtherelaborateduponintheimplementationchapter.

Literaturereview

10

2.1.6RelatedfieldsIn this section, related fields toBluetooth andBluetooth Smart are described. In these fields,approaches utilizing fingerprinting, trilateration and RSSI are common. Due to the limitedresearch fields of Bluetooth and especially Bluetooth Smart, this literature review has beenconductedinsuchamannerthatthesecloselyrelatedfieldshavebeenexaminedaswell.

2.1.6.1Wi-FiWi-Fi isawireless localareanetworktechnologywhichusuallyutilizesthesame2.4GHzISMbandthatBluetoothutilizes.Anothercommonbandisthe5GHzbandthathasbeenintroducedinmorerecentversions.Thesearethetwomostcommonbands,somerouterscanoperate inbothofthesebands.Thistechnologyiscalleddualband,anditrequirestheroutertohavetwobuilt in radios.[12] The next up and coming commercial system will be 802.11ad, which willdiffersfromthepreviousprotocolsandoperatesinthe60GHzfrequencyband.Thisallowsfortransfers up to 6.75 GBit/s at distances shorter than 10 meters. The development of thisprotocolseekstoeliminatetheneedformanycables.[13]Thereviewedpapersinthetheoreticalframeworkbuildsontheprotocols802.11gand802.11b.Bothoftheseoperateinthe2.4GHzISMband.TheseareasstatethesamebandsasBluetoothandBluetoothSmart.Wi-Fialsohastransmission strength more similar to Bluetooth Smart. These factors make the insights,conclusionsandinformationgatheredfromthisfieldrelevant.

2.1.6.2WirelessSensorNetworksWirelesssensornetworks(WSN)arenetworksofdistributedwirelessnodesandsensorputinplace to monitor some kind of activity. Within this field it is sometimes important for thesensors and nodes to identify their positions towards other sensor within the network. Forinstance, aWSNcanbedeployed for thepurposeofmonitoringanarea. If the sensorsare tomonitor this area, then theyneed toknowhow theyarepositioned in relation to eachother.Therethusismuchresearchonhowthismappingprocessshallbeconducted.Usually,therearesomeidentifiedreferencenodeswheretheexactpositionisknown.Additionally,thereareblindnodeswherethepositionsarenotknown.Ablindnodecanbeidentifiedfromtheinformationreceived from the referencenodes. The approachutilized to identify these canbe, ToA,RSSI,AoAetc,therearenodefinedstandarditissystemspecific.OnescientificreportfromthefieldofWSNisdescribedinthetheoreticalframework.[17]ThisreportfocusesonthepropertiesandreliabilityofRSSItodeterminedistancesbetweennodes.

2.1.6.3GSMIsasecondgenerationmobilecommunicationssystem(2G)firstdeployedinFinlandin1991.[14]Today, it is themostwellestablishedsecondgenerationsystemintheworld.[15]Onescientificreport has been gathered from this field. It presents models to predict path loss in indoorenvironments.[16]Thisistheoldestreportpresentedinthetheoreticalframework.Laterwork,presented in the theoretical framework, build on the same principles when it comes to thepropertiesofRSSI.

2.2TheoreticalframeworkIn this subchapter, the theoretical framework of this study is presented. These articles andresearch papers have been gathered from the fields of Bluetooth, Bluetooth Smart and therelatedfieldsdescribedintheprevioussection.Furtherarticlesandresearchthathavebeenin

Literaturereview

11

interestfocusonthetwointroducedpositioningtechniquestrilateration[18][20][22][23]andfingerprinting,[19][21][23]andalsoworkthataddressesRSSIsuchas[16][17][19][23].

2.2.1GSM/Mobile

2.2.1.1914mhzpathlosspredictionmodelsforindoorwirelesscommunicationsinmultiflooredbuildingsIn [16] the path loss of received signal strength within the 914 MHz band in indoorenvironmentsisinvestigated.Thisbandisformobiletrafficandthepurposeofthepaperistoexamineindoorconditionsofsignalpathdecayforfuturedeploymentofmobileaccesspoints.Eventhoughthemodelsofpathlossareconstructedforthisparticularband,theauthorsarguethatthesemodelscanbeappliedforthewholelowmicrowaveband(1-5GHz).Thisargumentexcludes the multi-floored models for which they suggest that more research need to beconducted.

Intheirtestbedtheyhavefourmulti-flooredstructures,agrocerystore,aretailstoreandtwoofficebuildings.They introducetwotypesofobstacles,concretewallsandsoftpartitions.Thesoft partitions are an arbitrary expression describing a range of obstacles not as thick asconcretewalls, for instance separatorwalls positioned in the office landscape. For each pathlossmeasurementtheycollectasetofdata.Thesevaluesare:

! Medianpathloss! Blueprintsforthemeasuredbuilding! Exactlocationoftransmitter! Exactlocationofreceiver! Transmitter-receiverEuclideandistance! Numbersoffloorsbetweentransmitterandreceiver! Numberofconcretewallsbetweenthetransmitterandreceiver! Numberofsoftpartitionsbetweenthetransmitterandreceiver

Frompreviouswork,theauthorshavesetuparangeofpredictionmodels inordertopredictthepathlossoneachmeasurement.Theysetupadistancedependentpath-lossmodel,afloorattenuationfactorpathlossmodelandasoftpartitionconcretewallattenuationfactormodel.These three models resulted in an overall 5,8 dB standard deviation in predicted versusmeasured path loss. (2.1) describes the same floor path loss predictionmodel.PL(d) is themean path loss, PL(d0) is the path loss at the reference distance d0 and n is the path lossconstant.

!"(!) [!"] = !"(!!)[!"] + !"×!×!"#!" !!!

(2.1)

Fromallmeasurementsitwasconcludedthattheaveragepathlossthroughsoftpartitionswas1.39dBandthroughtheconcretewallswas2.38dB.Intheconcludingarguments,theauthorshighlight these averages for all measured data as well as concluding that site specificinformationcanbeusedtopredictpathlosswithastandarddeviationof5.8dB.Furthermore,theypointoutthatthepathlossconstantriseswhenmoreobstaclesareintroducedtotheroom.

Literaturereview

12

2.2.2WirelessSensorNetworks

2.2.2.1EvaluationofthereliabilityofRSSIforIndoorLocalizationWSNisafieldwheremuchresearchhasbeenconducted.Withtheirarticle,theauthorsof[17],aim towards investigating the properties of RSSI and how reliable it is for the purpose oflocalization.InindoorenvironmentsofWSNdeployment,RSSIrangingisawellused,butnotasufficiently investigated method to locate sensor nodes. This is why the authors intend toinvestigatethereliabilityoftheapproach.

Also introduced in [16], but in [17] better fitted for thepurposeof calculatingdistance fromRSSImeasurements,thisarticleintroducestherelationshipbetweenpathlossinthefreeroomaswellastheobstructedroom,andadistancedependentpath-lossmodel.In(2.2)thepathinoptimalconditionsisdescribed.Pr isthereceivedpowerandd isthedistancetravelledintheunobstructedroom.

!! ∝ !!! (2.2)

(2.2)expressanidealrelationshipbetweenthedistancethepathlossthatisseldomthecaseintherealworld.Thevariablen,calledtheroomconstantcanbeintroducedinordertoimprovethemodel.

!! ∝ !!! (2.3)

(2.4)describeasimplifiedversionof the formuladescribed in(2.1).Sinced0 isset tobeonemeter,itcanbeeliminatedfromtheformula.

!""# = !"×! !"#!" ! + ! (2.4)

Theseequationslaythefoundationofthescientifictestbedinthearticle.Thetestisconductedinonetestsetting.Theroomselectedforthetestisastraightcorridor,withaconcretewallononesideandaglasswallontheother.Measurementsweretakenatseveralpointsamongstthepathbetween the transmitter and the receiverwhile the receiverwasmovingaway from thetransmitteralongthecorridor.

Two reference curveswere produced, one curve is derived from (2.4)and another from thegathereddata.Afteranevaluation,itisconcludedthattheformulabasedreferencecurveisthepreferablechoice.Aftertheestablishmentofareferencecurve,thereliabilityofRSSIasmeansfor localization was evaluated using four approaches for handling the data. These werecomparedtothereferencecurveandtheyare:

! Rawdata! Movingaveragemethod! Weightedaveragemethod! Curvefittingmethod

Themovingaverageandweightedaveragemethodsarethetwomethodsthatbestconformstothereferencecurve.Theauthors’concludingremarksarethatRSSIisunsuitableforthepurposeof locating a mobile node. Worth considering, however, is that this observation applies todetermining distance from one source to one receiver. Observing the graphs displayed it is,

Literaturereview

13

however, obvious that amethod of handling the datawith some type of averaging or fittinggreatlyincreasestheaccuracyofdeterminingdistancefromRSSI.

2.2.3Wi-Fi

2.2.3.1IndoorpositiondetectionusingWi-FiandtrilaterationtechniqueThe authors of [18] suggest using a trilateration technique andWi-Fi technology to performlocation based services in indoor environments where GPS is insufficient. They present agrowing market in LBS outdoors using GPS and suggest transferability into indoorenvironments. This is, nevertheless, not feasible due to the disturbing factors on GPS signalsfromstructures.

Tocircumventdisturbingfactors,theauthorsrecognizesomemethodsasusefuloptionsinIPSdeployment.Withoutacomparativediscussionregardingbenefitsanddisadvantagesbetweenthese, they settle for proposing trilateration as a means for indoor positioning. This can bemotivated by the fact that it is a theoretical article suggesting an approach to a problem,entailingthattheauthorshavenotactuallyinvestigatedtheaspectstoargueforonetechniqueovertheother.

TheauthorssuggestasystemforpositioningintheXY-planeusingthreeWi-Fiaccesspoints.Ifaposition in the XYZ-plane is to be determined another access point is needed. They suggestusingaformulathathasa linearrelationbetweenincreasingdistancesanddiminishingsignalstrengthtocalculatethedistancefromsignalstrength.Thisisopposedtohowthesignaldecayispresentedin[16][17].Fromoneaccesspoint,(2.5)describesthedistancebetweenitandtheprospectedreceiver.

(! − !!) + (! − !!) = !!! (2.5)

Fromthreeaccesspointsasystemofequationsforthesecircleradiusescanbesetup.Satisfyingthis systemwill provide the coordinates in the Cartesianplane for the sought location of thereceiver.Thearticlesuggeststhisapproachforasolutiontotheproblematiccircumstancesofprovidinglocation-basedservicesindoors.

2.2.3.2PropertiesofindoorreceivedsignalstrengthforwlanlocationfingerprintingIntheirwork,theauthorsof[19]aimtoinvestigatethepropertiesofreceivedsignalstrength(RSS)inapositioningsystembasedonfingerprinting.ThestudyraisesanimportantaspectofinvestigatingthepropertiesofRSSfingerprintingbeforebuildingapositioningsystembasedonRSS. The authors have noticed an increased amount of research in the field of fingerprintingbasedpositioningusingWi-Fi,butmostworkinthisfieldhasignoredthepropertiesofRSS.Inanon-staticindoorenvironment,whichisbasicallyanywhereonecouldaspiretodeployanIPS,theRSS isexposed todisturbancesunlikea staticenvironmentwheremost studieshas takenplace.ByinvestigatingthepropertiesofRSS,theauthorswishtocontributeinsightonhowtomodelandconstructfuturefingerprintingbasedIPS.

The authors’ note that it is problematic when utilizing GPS indoors and suggest that afingerprintingmethodiseasytodeploy,whencomparedtoanAngleofArrival(AoA)orTimeDifferenceofArrival(TDoA)method.Thisisduetothefactthatthereisnoneedforspecializedhardware on the receiving end. They do not consider a comparison with Time of Arrival

Literaturereview

14

techniques.Furthermore,theydescribetheimplementationofafingerprintingbasedsystemintheirintroduction.Alocationfingerprintdatabaseoraradiomapneedstobeconstructedbeforesuch a system can be functional. This is a database containing information on a set ofcoordinatesintheroomwherethefingerprintingsystemisdeployed.OnthesecoordinateseachRSSfromallavailablesourcesisrecorded.Eachlocationgetsauniquefingerprintfromthesetof transmitters that is unique to every location. These fingerprints are collected over a setamountoftimetoproduceameanvalue.Intheirstudy,theauthorsusethreeaccesspointstorecordaradiomap.

AmongthesetofpropertiesofRSSthat theauthors test in theirstudy, there is theeffectofauser’s presence on RSS, statistical properties of RSS, properties of multiple RSS at a singlelocation,andpropertiesofRSSatdifferentlocations.Fromthepropertiesofuser’spresenceonRSS,theauthorsconcludethatauser’spresenceincreasethestandarddeviationondBmfrom0.68dBmto3.0dBm.Theorientationoftheuserisofgreatimportancesincethehumanbodyconsists of nearly 70%water, the RSS is absorbed when the user obstructs the signal path.When investigating the statistical properties, they conclude that RSS distribution may bestationary.FromthepropertiesofmultipleRSSatasingle location theauthorsclaimthat theaccesspointsignalsareuncorrelated.TheinvestigationofRSSatdifferentlocationsuggestthatwith a small set ofmeasurement points, two access points is enough to create a radiomap.Thesearesomeofthemanypropertiesinvestigatedinthestudy.

2.2.3.3AWeightedCenterofMassbasedTrilaterationApproachforLocatingWirelessDevicesinIndoorEnvironmentIn[20]theauthorsuggestsandevaluatesaweightedtrilaterationapproachusingtrilaterationandKalman filtering. The suggestedweighting algorithm awards less attenuatedRSSI signalswithhigherweights,thusgivingtheseagreaterimpactwhencalculatingaposition.

In the theoretical background, the author accounts for a range of different positioningtechniques and introduces two major groups of algorithms. These are trainable and non-trainablealgorithms.Atrainablealgorithmisonethatrequiresaninitiallearningphasebeforeit canbedeployed, for instance fingerprinting.Anon-trainablealgorithm isone thatdoesnotrequireanytrainingbeforeitisdeployed,forinstancetrilateration.

The author divides the proposed algorithm in two phases, in the first phase RSSI values areconverted intodistanceusing thepath lossequationexplained in [16][17]aswellas in [20].Thenextphaseconsistof threesteps, in thisphase the locationestimation isexecuted. In thefirst step, circles are drawn fromeach access point givenby the calculateddistance in phaseone.The intersectionof each circlewithall theother formsa setof coordinates in apolygonwith n number of sides depending on the amount of access points and intersections. In thesecondstep,thesepolygonsaredividedintotriangles.Annsidedpolygonformsn-2triangles.ThesetrianglesareassignedcoordinatesfortheircentreofmassbasedontheRSSIstrengthofthe coordinated forming these.When the center of all the triangles have been calculated, theprocesscanberepeateduntilthereisonecentreofmassforthepolygon.Inthethirdstep,theprocessisrepeatedforeachpolygonformedbytheaccesspoints.

AKalman filter is also applied and tested. It is evaluated and applied in bothphases, in eachindividualphase,andinneitherofthephasesasacomparison.AKalmanfilterisamathematicalmodelfornoiseelimination.Itconsistsoftwosteps.Inthefirststepatimeupdate(prediction)

Literaturereview

15

is performed and in the second step a measurement update (correction) is performed. TheKalmanfilterisarecursiveequationthatmemorizessomeattributesfromearlierstageswhilebecomingmoreinvariableforeachiterativestep.

Intotal,theauthorcomparestheperformanceof12differentapproachestoestimateposition.Thealgorithmwiththebestperformance,intermsoflowaverageerrors,wasweightedcenterofmasstrilaterationusingaKalmanfilterinbothstages.

2.2.4Bluetooth

2.2.4.1BluetoothIndoorPositioningThe author of [21] recognizes the acceleratedprogression of locationbased services and thegapbetweenthewellestablishedoutdoorplatformGPSandanindoorcounterpart.Theauthor’ssolution in this particular paper is a trainable fingerprinting algorithm based on regularBluetooth that is toactasa support system tooutdoorGPS.Theapproach that is chosen isafingerprinting algorithmbasedonRSSI. The authordescribes an offlinephase,where a radiomap is created by recording the RSSI from the set of m beacons, for each chosen (x, y)coordinate in the room. During the online phase, the RSSI values received at a location iscomparedtothedatabaseofrecordedRSSIvaluesfromdifferentlocations.Theauthorutilizesthree different algorithms to compare the data. These are k-Nearest Neighbour, regressionbasedonk-NearestNeighbourandNaïveBayesclassifier.

Thek-NearestNeighbouralgorithmisbasedonobservingtheEuclideandifferenceinRSSIfromto the different recorded coordinates RSSI values. The choice of k determines the numbercoordinatestobeselectedinincreasingEuclideandistancefromtherecordedinputdatatothereferencepoints.Whenknumberofcoordinateshasbeenselected fromthereferenceset theselectedsetgetsanequalvoteindeterminingthecurrentlocation.Accordingtotheauthor,thisisadrawback.Buildinguponthek-NNalgorithmaregressionapproachisalsoapplied.

ThethirdalgorithmappliedbytheauthoristheNaïveBayesclassifierbasedBayestheorem.Inthe case of the author’s study, the decision rule inherited by this approach leads to simplyselectingthecoordinateinthetrainingsetthathasthehighestprobability.

Intheconclusion,theauthorstatesthatthek-NNregressionalgorithmprovidedthebestresultsin the study usingk= 7. Furthermore, the author states that Bluetooth has the advantage ofhavingastrongerandmorestablesignal thanWLAN.However, the lackofbeacons inregularBluetoothisaweakness.Theprecisionofthesystemisconcludedtobegoodwitha50%chanceofbeingwithin1.5meterscalculatedpositioninastaticsetting.

2.2.4.2BluetoothIndoorPositioningusingRSSIandLeastSquareEstimationIntheirarticle,theauthorsof[22]suggestanIPSutilizingregularBluetooth,RSSItoobtainthedistance and trilateration as the positioning method. To convert the signal strength intodistance, the authors use the basic assumption of how the signal strength decays in the freeroomasdescribedin2.2.TheyalsoignorethefactthatRSSIhastheinheritedpropertyofbeingalogarithmicscalewhentheyextendthedecayofthesignalintoaformuladesignedtoconvertthe signal strength into distance. This is contrary to [16][17][20] in this literature review,whichhasimpliedthesamebasichypothesisconcerningthesignaldecay.Theauthorsof[22]do refer to theirmethod as being the simplestmethod to describe the relationship betweensignaldecayanddistance.

Literaturereview

16

Judging from their graphs, this implementation has proven to provide some good results inconvertingRSSItodistance.Theypointout,however,thatthereareperfect lineofsight(LoS)conditions. When the distance has been calculated, the authors use linear least squareestimation to solve the overdetermined system formed by the distance to each beaconexpressedasacircleintheXY-plane.TheBluetoothtransmitterswereplacedineachcorneroftheroom.

Aftertheleastsquareestimation,themaximumerrorvalueis2.1metersandtheminimalis0.3meters.Theauthorsleaveoutvaluableinformationsuchasmeanerrorandstandarddeviation.Theydoprovidea3D-graphcoveringthewholemeasurementspace.Thisisperhapsduetothearticlebeingfairlyshort.

Theresultsshowgoodaccuracy fromtrilaterationand linear leastsquareestimationutilizingregularBluetooth. In their final remarks, the authors conclude thatBluetooth technology canprovideafunctioningIPS.

2.2.5BluetoothSmart

2.2.5.1EvaluationofindoorpositioningbasedonBluetoothSmarttechnologyThe purpose of [23] is, according to the authors, to evaluate Bluetooth Smart as a tool toconstruct an IPS. This question is being tackled by answering a set of sub-questions such as,when isBluetooth a viable choice for indoor positioning,what algorithms are best suited forindoorpositioning,whatadvantagesanddisadvantagesdoesthetechnologyhaveversusrelatedtechnologies (suchasWi-Fi and regularBluetooth), andwhat levelof accuracy canBluetoothSmart provide. The algorithms selected for the evaluation are trilateration, iterativetrilateration,particlefilteringandfingerprinting.

Iterativetrilateration isaprocesswherean initialposition is firstcalculatedbasedonregulartrilateration.Thisposition is thereafterupdated repeatedlybasedonnewmeasurements andtheoldcalculatedlocation.Inthisstudy,thisisdonebyaTaylorseriesapproximation,butothermethods, such as the Kalman filtering technique explained in [20] can be applied. Thefingerprintingapproachusedby theauthors isbasedoncalculating theEuclideandistance tothe different reference points similar to [19].After these distances have been calculated thethreenearestreferencepointsareselectedandthemidpointbetweentheseareselected.

BesidespositioningalgorithmsthisstudyalsoexaminesandpresentsseveralcharacteristicsofRSSI.Amongtheseareimpactofangle,behaviouroverdistance,andexpectedbehaviourfrompreviousstudies.Theconcludedexpectedbehaviourcharacteristicsbasedonpreviousstudiesare:

! RSSIvaluesinastaticenvironmentarenotstatic,butvarywithinalimitedinterval.! AcorrelationbetweendistanceandtheaveragemeasuredRSSIexistsinastaticsetting.! ThevarianceofameasuredRSSIincreaseswithdistance.! A formula to convertmeasuredRSSI into distance exists. This is the same formula as

(2.4).The findings from the authors’ study of RSSI behaviour on distance reveals that the RSSImeasurement is only suitable fordistance calculationswithin closeproximity to thebeacons.From theirmeasurements, the authorshave found thatRSSI varianceoscillatewhendistanceincreases.ThesecharacteristicsareunsatisfyingwhenconvertingRSSIintodistance.Afterafew

Literaturereview

17

meterstheauthorsstatesthattheRSSImeasurementsbecomestoounreliableforthispurpose.In the case of angles versus RSSI measurements, the authors’ findings show that the deviceorientationaffectsthemeanRSSIvaluesmeasured.Inaverticalpositiontheanglesorthogonalto the signal was the weakest. It is important to note that the angle of the device is a nonegligiblefactorwhenperformingascientifictest.RSSImeasurementsalsovaryfromdevicetodevice.

Thechosenalgorithmsareevaluatedandgraded in fivedifferentaspects, theseareAccuracy,Calibration complexity, Scalability, Liveliness and response time, and Robustness andadaptability.Nooverallscoresaregiven.Fromtheconclusionof theirwork, theauthorsstatethat the methods relying on distance calculation from RSSI are less desirable. Among theirchosen algorithms, fingerprinting is the only one that does not rely on RSSI distancecalculations.

Overall,thereportcoversalotofaspectsandissuesconcerningutilizingBluetoothSmartasameansofconstructinganIPS.

2.2.6TheoreticalframeworksummaryFrom[16]and[17]propertiesandknowledgeaboutRSSIandhowitcanbeusedtodeterminedistance. It is learned that the RSSI measurements vary in such a manner that it cannot bedescribed as “reliable” for the purpose of determiningwhilemappingwithin aWSN in [17].Methods to improvemeasureddata, such as averaging/fitting/weighting improve results andreliability.ConcretewallsandsoftpartitionsaffecttheoutcomeofmeasuredRSSI.

In [18] an approach to utilize trilateration within Wi-Fi and RSSI to determine distances issuggested for thepurposeof creatingan IPS.From [19] further impactsof softpartitionsaredescribed.TheimpactofthehumanbodyjustbeingpresentgreatlyaffectsthemeasuredRSSIvariance.In[20]anIPSutilizingtrilateration,Wi-FiandRSSIisevaluated.ApproachestoweighthedataandimproveresultsdescribesKalmanfilteringandparticlefilteringassuccessful.

Fingerprinting approaches are evaluated in [21], k-NN algorithms are described where knumberofnearestpointsareselected.In[22]anapproachtoutilizetrilaterationwithinregularBluetoothandRSSItodeterminedistancesistestedandprovenreliable.From[23]moreaboutthe reliability of RSSI within Bluetooth Smart is learnt. Fingerprinting and trilaterationapproachesarealsotestedandevaluated.

In this chapter, answer to sub-question one has been provided. In the following chapter, theimplementationisexplainedwiththistheoreticalframeworkasabasis.

Implementation

18

3 ImplementationInthischapter,theimplementationofthestudyisdescribed.Thischapterisdividedintothreesections.Thefirstsectiondescribesthetestbed,includingtheroomspecificsandthehardware.The second section, positioning techniques, explains the choices of selected algorithms withconnections to the theoretical framework, as well as describing the exact implementation ofthese. Inthethirdsection,datacollection,theapproachandsoftwaresetuptogatherthedataareexplained.

3.1TestbedThis sectiondescribes the roomspecifics, the choicesof hardware and thebeaconplacementwithin the room. The room specifics describe the general layout of the room, the hardwaresectiondescribesthechoicesofhardware,andbeaconplacementpartdiscussesthepositioningchoicesofthebeaconswithintheroom.

3.1.1RoomspecificsTheroomwherethetestwasconductedwithinhaddifferentmeasurementsfromtheblueprintsthan what was actually measured. The dimensions of the room were determined to be 13meterslongand4.7meterwideaftermeasurementswereconducted.

Implementation

19

Figure3.1Sketchoftestsite.Distancesaregivenincentimeters.

The four sidesof the roomweregivennamesafter the cardinalpointsnorth, south, east andwest.Thiswasdoneinordertokeeptrackofthebeaconsandtheirplacementswithintheroom.Atthewestsideoftheroom,therearethreeequallysizedwindowswithassociatedgapsinthewall. The southwall is a plainwall.On the east side, there is first a gap towards an adjacentcorridor,thenapillarandthenanothergaptowardsthesameadjacentcorridor.Afterthisgap,therestoftheeastwallconsistsofapartlyglassedwall.Thenorthwallisconsistsofaregularwallandanotherwindow.Thesearethegeneralattributesofthetestsite.Thecoordinateplanedefined in the roomhas itsorigo in the southwest corner. It is the right corner shown in thepicture inFigure3.4. InFigure3.2 andFigure3.3photos taken from the southeastern andsouthwesterncornerscanbeviewed.

Implementation

20

Figure3.2Picturetakenfromthesoutheasterncorneroftheroom.

Figure3.3Picturetakenfromsouthwesterncorneroftheroom.

Implementation

21

Figure3.4Picturetakenfromthenorthernwall.

3.1.2HardwareInthissection,thechoicesofhardwareandthemotivationbehindthesearedescribed.

3.1.2.1BeaconsInthepreviouschapter,thecompanyEstimoteanditsbeaconswerebrieflydescribed.Severalcompaniesareinthemarkedofdistributingandmanufacturingbeacons.ItwasconcludedthatEstimote had a set of advantages compared to other distributors that made them moredesirable. Three factorsmade the Estimote beacons extra interesting. The firstwas thewell-establishedbrandnamewithinthebusiness.ThesecondfactorwastheprovidedSDKandtheconstantdevelopmentofit.InvestinginEstimotebeaconsprovidesagreatercertaintythatthebeaconscanbeusedinfutureprojects.ThethirdfactorthatmadeEstimotemoredesirablewasthe ability to configure the broadcasting strength. Somemanufacturers shared some of theseattributes,butEstimotewasthemanufacturerthatstoodoutthemost.

Thebroadcastingstrengthofthebeaconsintheexperimentwassettothepredefined-12dBm.Thisistheconsiderednormalstrengththatallowsabeacontobroadcastforanestimatedtimeof three years. This setting was chosen since it is of interest to investigate the systemscapabilitiesatthebroadcastingstrengthdesignedforlongsustainability.

3.1.2.2PhoneIn [23] it is concluded that resultsvary fromdevice todevice.Therefore, itwas important toconductallthetestsonthesamedevicetoavoidbiasprovidedfrommeasurementsconductedon different devices. In [23], it was also established that the device orientation providesdifferent results in average RSSI. Considering these two factors, it was decided that the testdeviceshouldalwaysbethesameduringthewholetestandthatit,also,alwaysshouldfacethesamedirection.TheselectedphoneusedinthistestwasaniPhone4S.

Theorientationofthedevicewithintheroomwasalwaysthescreenfacingtheroofandspeakerpointing inthedirectiondefinedasnorthwithintheroom.In [16][19]effectofobstaclesandhuman bodies are described. Within the room the iPhone was placed at the height of 150centimetersateveryrecordingtestpoint.Thiswasatthesameheightasthebeaconsplacement

Implementation

22

for two reasons. The first reason is that at this height there were perfect LoS conditions,meaningnoobstaclesbetweenthebeaconsandtherecordingiPhonedevice.Thesecondreasonis theenablingof conducting trilateration in the twodimensionalplanedefinedby the lengthandthewidthoftheroom.

3.1.3BeaconplacementThe beacons were placed at the height of 150 centimeters as described above. The distancefrom the floor to the roofmeasured243 centimeterswhichmakes the centerheight at121,5centimeters.Atthisheight,however,thereweresomeobstaclesalongthewallsthatcouldhaveinterferedwithhavingnon-disturbedLoSconditions.Theheight150centimeterswaschosen,astheroomhadnoneinterferencesalongtheXY-planeatthisheight.

In [16][17][19][23] it is stated that RSSI ismore reliable at shorter distance. Consequently,whenplacingthebeaconswithintheroomtheywerepositionedinsuchamannerthatallowedstronger signals to reach a larger area. This means avoiding corners and striving towardsputting the beacons at the center of walls. To represent this, imagine drawing a circle thatrepresentsthedistanceagivensignalhastravelledawayfromabeaconatapointintimeinthetwodimensionalspace.Thewallsinthisspaceintersectthelinesthatrepresentthiscircle.Bydoingso,thewallsminimizetheeffectivesurfacecovered.Ifthesignalonlycanbeconsidered“reliable” at a distance X from the beacons, it is desirable to have that area defined by thedistanceXtobeaslargeaspossible.

Withthisinmind,andhavingfivebeaconstoworkwith,itwasdecidedthatonebeaconshouldbe placed at eachwall. Onewallwould have twobeacons and it should be one of the longerwalls.Toachieveasymmetricsetupitwasdecidedtoputtwobeaconsontheeastwall,sinceatthiswallabeaconcouldnotbeplacedattheexactcenteralongtheY-axis.Thiswasduetoanadjacentcorridorgapbetweentheglassedwallandthepillar.

BeaconName X(m) Y(m)Northeast 4.7 9Southeast 4.7 4South 2.35 0West 0 6.5North 2.35 13

Table3.1Beaconnamesandtheirplacementinthedefinedroom.

3.2PositioningtechniquesThe two compared algorithms in this study are, as mentioned previously, trilateration andfingerprinting.Inthenexttwosubchapters,theimplementationsoftheseareexplained.Withinbothofthesealgorithmsaweightedapproachwherealsoappliedandevaluated.

3.2.1TrilaterationAssuggestedin[18][22],atrilaterationapproachwasselected.Oneregularandoneweightedapproach.Theseareexplainedbelow,astheweightedapproachbuildsupontheinitialstepsforthisapproachisexplainedintheregulartrilaterationapproachsection.

Implementation

23

3.2.1.1RegulartrilaterationapproachByrearranging(2.4),thepathlosspredictionmodel,distancecanbecalculatedwith(3.1).

! = !"!""#!!!"×! (3.1)

Initially,thepathlossconstantneedstobecalculatedwith(3.2).

! = !""#!!!"×!"#!"(!)

(3.2)

In order to utilize these formulaswithin the room, the constantA needs to be calculated foreverybeacon.Thisisthesignalstrengthatonemeterfromthebeacon.Thisreferencevaluecanbemeasuredatanotherdistance.Between[16]and[17] it isobservedthat thechoiceofonemeter simplifies the formula and the calculations. One meter is also applied in[19][20][22][23].

Thepathlossconstantalsoneedstobedecided.Thiswasdecidedbymeasuringwhilelettingd,RSSI, andA be known at 15 reference points. These reference pointswere placed in such amanner in the room that five of them provided themeasurements for the variableA. Thesereference points also provided data for the fingerprinting algorithm, discussed later in thesectionaboutthefingerprintingalgorithms.Ateverypointthepathlossconstantwascalculatedtowardseverybeacon.Atonemeterfromeachpoint,thereferencevaluecannotbecalculatedsince there is a divisionby zero. From this, 14path loss constant valueswere calculatedperbeacon.Themeanpathlossconstantwherethencalculatedforeachseparatebeacon.InFigure3.5theplacementofthebeacons,thereferencepointsandthetestpointsarepresented.

Implementation

24

Figure3.5Sketchdisplayingthereferencepoints(whitecircles),thetestpoints(blacksquares)andthebeacons(blacktriangle)intheroom.

From these measurements, the calculated path loss constant and the reference value at onemeterwereappliedinthetrilaterationalgorithm.Threetestpointsweresetupforthepurposeof conducting the trilaterationapproachand the fingerprintingapproach. In these testpoints,the RSSI values were recorded for a minute and a mean was calculated. From these meancalculations a distance for each test pointwas calculated to all five beacons.With a distancefrom each beacon towards the reference point, the final step was to find the position thatminimizes the distance fromeach radius given by the calculated distances. InFigure3.6 thetrilaterationalgorithmispresented.

Implementation

25

Figure3.6Regulartrilaterationapproach.

3.2.1.2WeightedtrilaterationapproachFrom[20]itislearnedthatfilteringorweightingthedatacanincreaseperformanceandreducedisturbances.Intheweightedtrilaterationapproach,linearleastsquareregressionisappliedtothedatasetbeforeconducinganycalculations.Themeasurementseriescanberepresentedontheform !!,!! , !!,!! ,… !!!! .Asolutiontothefollowingsystemiscalledfor.

!" = ! (3.3)

BymultiplyingbothsidesoftheequationsystembythetransposeofthematrixA,andsolvingthesystem,asolutionontheform! = !! +!canbefound.

!!!" = !!! (3.4)

Inorderfindthesolutionthefollowingmatrixesareconstructed.

! =!!,!!!,!⋮

!!,!

! =!!!!⋮!!

! = !!

(3.5)

Foreachbeacon,aninitialvalueatthedistanceofzerometers(mfromc)andaRSSItrendlineisextracted(k fromc). Theequation foreachbeaconcanprovidemodifieddata for thepathlossconstantaswellasthereferencevalueatonemeter.Inorderthefullyexaminetheimpactof applying least squares, the weighted approach will consist of three different results. Oneresultwillbebasedononlyreplacingthecalculatedroomconstants,oneresultwillbebasedononlyreplacingthereferencevalues,andoneresultwillbebasedonreplacingboththepathlossconstantandthereferencevalueatonemeter.

3.2.2FingerprintingTwo fingerprinting approacheswere selected. The regular fingerprinting approach builds ontheprincipleofselectingthethreenearestreferencepointsandcalculatingamidpointbetween

Implementation

26

these,asapplied in[23].Theweightedapproachseeks to includemorereferencepoints thanjustthethreeclosestone.Theinitialstepswithinboththesealgorithmsarethesame.

3.2.2.1RegularfingerprintingapproachTheinitialstepinthefingerprintingalgorithmistoconstructtheradiomap.Ateachreferencepoint, there are 15 (see figure 3.5), RSSI values are recorded for a minute. The beaconstransmiteverysecond,whichmakesitatotalof60values.Ameanvalueiscalculatedforeachbeaconineachreferencepoint.Thefinalradiomapconsistsof70values,describingthemeanRSSI input from each beacon at each coordinate. When the radio map is complete, the testpoints are recorded. In the sameway as before, data is recorded for 60 seconds. Amean iscalculatedtowardseachbeaconineachtestpoint.Thetestpointsarethencompared,byusingEuclidean distance, towards each of the 15 reference points. This is done by calculating thedifference in RSSI between the beacons in the test points and the beacons in the referencepoints.Thereferencepointsarethensortedbydistanceandtheclosestthreearechosen.Thisissimilar to then k-NN approach described in [21], but one point cannot be selected multipletimes.Dependingonthecalculateddistancethefinalpointisthenplacedbetweenthesethreepointsatthecenterofmass.InFigure3.7thestepsoftheregularfingerprintingapproacharepresented.

Figure3.7Regularfingerprintingapproach.

3.2.2.2WeightedfingerprintingapproachIntheweightedapproach,eachreferencepointsgetanimpactinthefinalresult.Aftersortingthe reference points by Euclidean distance towards the test points, the difference to max iscalculated.

!"#! = !"#$%&'( !""# ! − !"#$%&'([!""#]!"# (3.6)

Difference to max is each reference points’ difference in distance between its own distancecalculatedEuclideandistance,andtheEuclideandistanceofthereferencepointsfurthestawayfrom the test point. That furthest away test pointwill thushave a difference tomax value ofzero.

Thedifferencetomaxisthensquaredforeachcoordinate,inordertogivethecoordinateswitha lower Euclidean distance a higher impact, and the coordinates with a higher Euclideandistancealowerimpact.Thisvalueiscalledpull.

!"##! = !"#!! (3.7)

Implementation

27

Afterthepullhasbeencalculated,thealgorithmstartssteppingfromorigo.Inthefirststep, itstepstowardsthepointwiththehighestpull.Thepullisaddedtoasumcalledtotalmass.Afterthisfirstinitialstep,thissteppingalgorithmislocatedinthecoordinatesofthereferencepointwiththehighestpullandortotalmassequalsthepull.

!"#$% !"##! = !"##! + !"##!…+ !"##! (3.8)

Inthenextstep,itstepstowardsthecoordinatewiththesecondhighestpull.Thepullisaddedtothetotalmass.Butthistimeitdoesnotreachthiscoordinate, itstopssomewherebetweenthelinebetweenthetwocoordinates.Wherethatpointislocated,isdecidedbythecurrentpullofthecoordinate,dividedbythetotalmass.Inthesecondstepitisthepullofthesecondhighestcoordinate,dividedbythesumofthepullofthetwohighestcoordinates.

!"#$ !"#$%&! = !"##!!!"#$ !"##!

(3.9)

Thealgorithmcontinuesandsteptowardseachcoordinate.Witheachstephowever, thetotalmassincreaseswhilethecurrentpullsofthecoordinatesdecreases.TheweightedapproachcanbeviewedinFigure3.8.

Figure3.8Weightedfingerprintingapproach.

3.3DatacollectionInthissection, thedatacollectionapproach isdescribed. In thetwofollowingsubsectionstheconceptandthesoftwareisdescribed.

3.3.1ConceptThedatawascollectedforthebothpositioningalgorithmsatthesametime.Thereferenceandthe test pointswere the same for bothpositioning techniques, as described inFigure3.5. Ineachreferenceandtestpoint,datafromallfivebeaconswererecordedforoneminute.Thedatais recorded and processed in retrospect to the actualmeasurements. Positioning calculationsarethusnotdoneliveintheapplication.Intotal,60RSSIvaluesarerecordedforeachbeaconineachcoordinate,whichsumsupto5400recordedRSSIvalues.Someofthesearefilteredoutofthemeancalculationsincesometimesthebeaconsignalsarenotfoundoraretooweak.Intherecorded data, these are represented by the value zero. This value is higher than the actualrecordeddata,whichcausestroublewhencalculatingamean.Therefore, thesevaluesneedtobetakenoutfromthemeancalculations.Fromatotalof90beaconrecordingsspreadoverthe18 test and reference points, only 12 whole samples did not contain any zero values. Eachrecordedtestandreferencepointproduce6x60matrix.

Implementation

28

3.3.2SoftwareThe software constructed in order to gather the data consists of an iOS application, whichrecords and sends the data to a PHP server. The server utilizes the API Google charts tostructurethedata.Googlechartstablesareeasilyconserved.Fromeachrecording inatestorreferencepoint,aGooglechartwasthecreated.TheiOSapplicationhadtwofunctions.Thefirstinitiated a new recording set over the interval of 60 seconds. The latter cleared previouslyrecorded data on the server. In the background, the app constantly identified beacons andaddedthemtoasetofidentifiedbeacons.Whentheapplicationstartsrecording,theapplicationrecordsthedatafromthepredefinedsetofbeaconsthatexistsamongsttheidentified.Ateverysecond it sends aPOST request to thePHP server containing the current time stamp,beaconnames,andRSSI.Iftheclearbuttonisused,aspecialvariablecalledclearissenttotheserver.Ontheserverside,thedataissavedwithinanarrayinthesessionvariable.Thesessionissettobethesameforeverydevicethatvisitstheserver.Afterarecordingwasdoneusingthemobiledevice, the data was easily copied from the server. Figure 3.9 is a sketch describing therecordingsoftware.

Figure3.9Sketchdescribingtherecordingsoftware.

Results

29

4 ResultsIn this chapter the results of the study are presented. Discussions, comparisons, and generalobservationsare reserved for thediscussionchapter. In the first section, the results from thetrilaterationstudyarepresentedandinthesecondpartthefingerprintingresultsarepresented.MeanRSSIvaluesforthetestandreferencepointsareshowninAppendixA.

4.1TrilaterationIn this part, the results from the trilateration study are presented. Initial data presented in4.1.1.,suchasmeasuredRSSIatonemeterfromeachbeaconandpathlossconstant,isrelevantforbothtrilaterationapproaches.

4.1.1RegulartrilaterationThefirststepofthetrilaterationapproachwastoextracttheRSSIatthedistanceofonemeterfromeachbeacon.TheresultscanbeviewedinTable4.1.AllbeaconsexceptonehadsimilarRSSIvaluesatthedistanceofonemeterfromeachbeacon.Thebeaconthatstoodoutwasthewestbeacon,whichhadasignificantlylowermeanRSSI.Asstatedbefore,thesemeasurementswere taken over the interval of oneminute,with a total of 60measurements per beacon. Toensurethatnothinghadgonewrong,anothermeasurementofthewestbeaconwasconducted.In this re-measurement, the west beaconmeasurement once again indicated a similar meanRSSI.Allbeaconswereset to thesame transmissionstrength.Apossible reason for thishugedifferencecouldbedifferencesinhardware.

Beacon μ(RSSI)North -64.94

Northeast -65.90Southeast -73.83South -66.58West -82.22

Table4.1MeasuredRSSIatonemeterfromeachbeacon.

After the RSSI values at one meter from each beacon were calculated, the next step was todecide thepath lossconstant foreachbeacon.Thepath lossconstantwascalculated foreachbeaconineachreferencepoint.TheresultsofthepathlossconstantcalculationsareshowninTable4.2.

Results

30

X Y Northeast Southeast West North South1 1 -1.64 -1.20 -1.07 -1.62 0.811 4 -1.62 -0.76 0.41 -2.09 -1.071 6.5 -1.43 -0.93 -1.86 -1.451 9 -3.03 -1.64 0.80 -2.57 -1.281 12 -0.85 -1.33 -0.78 -3.52 -1.43

2.35 1 -1.86 -2.31 -0.44 -2.03 2.35 4 -1.27 -0.80 1.42 -1.69 -0.482.35 6.5 -3.09 -1.67 1.25 -2.20 -2.152.35 9 -2.45 -1.29 0.42 -1.27 -0.962.35 12 -2.47 -1.33 -0.11 -0.913.7 1 -2.49 -2.32 -0.42 -2.00 1.083.7 4 -2.26 0.52 -2.22 -0.133.7 6.5 -3.69 -0.64 0.37 -1.92 -0.913.7 9 -1.27 0.29 -1.54 -1.043.7 12 -1.03 -1.99 0.15 -5.80 -1.47

Average -2.09 -1.39 0.20 -2.31 -0.81Table4.2Calculationsofthepathlossconstant(n)towardseverybeaconateveryposition.

With thepath loss constants and theRSSI values at onemeter fromeachbeaconusing (3.1)distancefromeachtestpointtowardseachbeaconcannowbecalculated.ThedataisdisplayedinTable4.3.Thevaluesgivenbythewestbeaconsstandsoutonceagain.

Testpoint Northeast Southeast West North SouthT1(2.35,2.5) 7.79 1.49 0.00 5.96 3.34T2(1.675,6.5) 6.87 1.99 695.58 5.74 5.48T3(2.35,10.5) 1.76 6.82 8.01 2.63 33.01Table4.3Distance(m)calculationsfromeachtestpointtowardseachbeaconusingpathlossconstants

andreferencevaluesatonemeter.

Tobeabletocalculatea locationfromthesedistancesoneneedstosolvetheoverdeterminedsystemofcirclesformedbythebeaconscoordinatedandtheircalculateddistancesasdescribedin(2.5).Thissystemofequationsisoverdeterminedandlacksasolution.Abestfitwassoughtthat minimized the distance towards each of these circles within the defined borders of theroom. The results are displayedTable 4.4. The parenthesis displays the best fit outside thedefinedroom.

Testpoint X Y CalculatedX CalculatedY Difference(m)T1 2.35 2.50 2.21 3.70 1.20

T2 1.675 6.50 4.70(6.96) 2.12(2.96) 5.32(6.36)T3 2.35 10.50 4.70(4.88) 11.82(12.28) 2.69(3.09)

Table4.4Calculatedpositionsfromregulartrilateration.

Results

31

4.1.2WeightedtrilaterationIntheweightedtrilaterationapproachtheinitialstepwastoleastsquarethedatasetinordertoproduceaweightedroomconstantandreferencevalueforthedistancecalculations.Thiswasdonewithlinearleastsquareregressionasdescribedin3.2.1.2.Table4.5describestheinitialvalues(m)andK-values(k)givenbytheleastsquareregressionontheform!""# = !" +!.The reference value at onemeter is given by adding the K-value to the initial value for eachbeacon.Theroomconstantcanbeextractedinthesamewayastheintheregularapproachbyusing the linear equation given by the initial value and the K-value for each reference pointwithintheroom.

Beacon Initial(m) k RoomconstantNortheast -71.23 -1.50 -0.85Southeast -74.82 -1.64 -0.93West -75.92 -1.36 -0.73North -70.78 -1.34 -0.94South -64.86 -1.41 -0.99

Table4.5Initialvaluesatzerometer,K-valuesandroomconstantforeachbeacongivenbyleastsquareregression.

The second step was to calculate the distance given by then weighted room constants andreference values at one meter. In Table 4.6, the calculated distance from these weightedvariablesarepresented.Thedatacolumnindicateswhichweighteddataisusedforthedistancecalculations.A indicatesthattheweightedreferencevalueatonemeterwasused,N indicatedthatweightedroomconstantwasused.A&Nindicatesthatbothwereappliedatthesametime.

Data Testpoint Northeast Southeast West North South T1(2.35,2.5) 24.31 0.95 23.58 13.88 2.89

A&N T2(1.675,6.5) 17.86 1.46 0.79 12.70 4.35 T3(2.35,10.5) 0.63 9.18 2.70 1.85 19.09 T1(2.35,2.5) 3.67 0.97 0.00 2.91 3.63A T2(1.675,6.5) 3.23 1.29 2.40 2.81 5.96 T3(2.35,10.5) 0.83 4.41 0.03 1.28 35.93 T1(2.35,2.5) 154.85 1.82 4.91 80.85 2.70N T2(1.675,6.5) 113.77 2.80 0.16 73.96 4.06 T3(2.35,10.5) 4.01 17.57 0.56 10.76 17.81

Table4.6Calculateddistances(m)towardseachbeaconusingtheweighteddata.

Afterthedistanceshavebeencalculatedthethirdstepwastocalculatethepositionsgivenbythe distances. In Table 4.7 the results of the calculated X and Y coordinates together withdifferencebetweenactualcoordinatesandcalculatedaredisplayed.Onceagainthedatacolumnindicatestheresultsofapplyingthespecificvariablesinthecalculations.Thebestfitlocationisthe locationthatminimizes thesummeddifference incalculateddistancetowardeachbeaconandactualdistance.Thebest-fit locationalgorithmis limitedtotheboundariesdefinedbytheroom,thuspushingsomecalculatedpositiontowardstheseboundaries.Itcanbeobservedthat

Results

32

theapproachresultinginthebestaccuracyistheonethatonlyutilizestheweightedreferencevalue.

Data Testpoint CalculatedX CalculatedY Difference(m) T1(2.35,2.5) 4.70(5.20) 0.00(-0.60) 3.43(4.21)

A&N T2(1.675,6.5) 0.00(-1.88) 3.66(1.04) 3.30(6.52) T3(2.35,10.5) 0.66 12.24 2.42

T1(2.35,2.5) 2.63 5.97 3.48

A T2(1.675,6.5) 2.83 6.36 1.17

T3(2.35,10.5) 3.92 9.30 1.98

T1(2.35,2.5) 0.00(1.28) 0.00(-2.48) 3.42(5.09)N T2(1.675,6.5) 0.00(0.92) 3.31(-3.80) 6.71(10.33) T3(2.35,10.5) 0.00(-8.32) 7.06(14.24) 4.16(11.31)

Table4.7Displaysthecalculatedpositionsfromtheweighteddata.Valuesinparenthesisdisplaystheborderingpositionsbestfitoutsidetheroom.

Asa finalcomparison,Table4.8displaystheaverageerrorof thethreeweightedapproachesandtheregularapproach.

Approach AverageError(m)RegularTrilateration 3.07(3.55)

WeightedTrilateration(A&N) 3.05(4.38)WeightedTrilateration(A) 2.21

WeightedTrilateration(N) 3.73(8.91)Table4.8Approachcomparisondisplayingtheaverageerror.Valuesinparenthesisdisplaysthesummed

errorusingbestfitsoutsidetheroom.

4.2FingerprintingInthispart,theresultsofthetwofingerprintingalgorithmsarepresented.InAppendixB thetotal Euclidean distance from each beacon to each coordinate, the calculated distances fromeachtestpointtoeachreferencespoint,andthepullofeachreferencepointarepresented.

4.2.1RegularfingerprintingIn the regular fingerprinting algorithm, the approachwas to calculate the Euclidean distancetowardsall referencepoints inevery testpoint.Thereafter, the threeclosest referencepointsare selected.Then themidpointof these three referencepoints is calculated. In tables,Table4.9, Table 4.10 and Table 4.11 the three closest calculated Euclidean distances towardsreferencepointsaredisplayed.Fromthesetablesaclearpatternofthereferencepoints,definedas close by Euclidean distance inRSSI, actually being close in reality can be observed. In thetables, thedifferences inRSSI for eachbeacon canbeobserved.Thedistance is calculatedbytakingthesquarerootofthesquaredsumofthebeaconrecordingsforeachcoordinate.

Results

33

X Y Northeast Southeast West North South Distance1 1 -3.1 5.7 3.0 -0.4 -6.1 9.351 4 -5.7 1.9 -6.8 2.2 2.5 9.643.7 6.5 -2.7 0.3 -7.1 -2.1 3.2 8.52Table4.9Thethreeclosestreferencepointsto(2.35,2.5)andtotalEuclideandistanceinRSSI.

X Y Northeast Southeast West North South Distance1 4 -4.6 0.1 4.0 2.6 0.7 6.613.7 4 -1.4 -4.2 2.4 3.8 -5.2 8.163.7 6.5 -1.6 -1.4 3.6 -1.8 1.5 4.77Table4.10Thethreeclosestreferencepointsto(1.675,6.5)andtotalEuclideandistanceinRSSI.

X Y Northeast Southeast West North South Distance2.35 9 4.0 -2.1 -0.4 -2.0 -3.2 5.903.7 9 -5.1 -2.7 0.0 0.0 -2.4 6.243.7 12 0.0 6.4 0.6 3.4 3.6 8.07Table4.11Thethreeclosestreferencepointsto(2.35,10.5)andtotalEuclideandistanceinRSSI.

ThefinaldistancescalculatedfromthesethreeclosestreferencepointsaredisplayedinTable4.12. The calculated X and Y is the describedmidpoint given by the three closest referencepoints. The algorithm does not take the actual distances into consideration. It is a decidingfactortowhichcoordinatesthatshallbepickedout.

Testpoint X Y CalculatedX CalculatedY Difference(m)T1 2.35 2.50 1.9 3.83 1.40T2 1.675 6.50 2.8 4.83 2.01T3 2.35 10.50 3.25 10.00 1.03

Table4.12Finalpositionsfromregularfingerprinting.

4.2.2WeightedfingerprintingThegoalofapplyingtheweightedalgorithmwastotakeallreferencepointsintoconsideration.Thedecidedapproachwastoapplydecreasingweightstothereferencepointwhilehavinganincreasingcounterweight,asdescribed in thesteppingapproach in3.2.2.2. In figuresFigure4.1,Figure 4.2 andFigure 4.3, the test point and each reference points pull percentage aredisplayed. The pull percentage is each coordinates’ calculated pull, as seen in Appendix B,dividedbythetotalmassofthesystem.Fromthesefigures,patternsofhighpullpercentageandcloselocationarequiteclear.Thepullofeachcoordinateisthebasisofthesteppingapproachappliedtocalculatealocation.

Results

34

Figure4.1Pullpercentagemapfortestpointone.(2.35,2.5)

Results

35

Figure4.2Pullpercentagemapfortestpointtwo.(1.675,6.5)

Results

36

Figure4.3Pullpercentagemapfortestpointthree.(2.35,10.5)

TheresultingpositionsfromthesteppingalgorithmaretheonespresentedinTable4.13.Theresultsof theweightedalgorithm indicate that the regularapproachwithmidpoints from thethreeclosestreferencepointsproduceaslightlymoreaccurateresultinthecaseofthesethreetestpoints.

Testpoint X Y CalculatedX CalculatedY Difference(m)T1 2.35 2.50 2.58 4.34 1.85T2 1.675 6.50 2.29 5.42 1.25T3 2.35 10.50 2.26 8.50 2.00

Table4.13Finalpositionsgivenbyweightedfingerprinting.

Discussion

37

5 DiscussionIn this chapter, the general discussion concerning the result and findings of the study isconducted.Inthediscussiontheformulatedresearchquestionsareanswered,excepttheonesthathavenotalreadybeenansweredintheliteraturereview.

5.1TrilaterationTheresultsfromthetrilaterationapproacharedisplayedinTable4.8.Inthetabletheaverageerror of the four different approaches can be observed. Regular trilateration 3.07 meters,weighted trilateration (A&N) 3.05 meters, weighted trilateration (A) 2.21 meters, weightedtrilateration(N)3.73meters.Overall, theresults in theweightedapproachbasedonusingonly thereferencevalue (A)aremore accurate. The results are often skewed towards the borders of the Cartesian systemdefined in the room. This pattern is most visible in the results given by only applying theweightedroomconstant.(N)InTable4.3thehighestdistancevalueproducedbythetrilaterationalgorithmcanbeobservedat695.58meterstowardsthewestbeaconintestpointtwo.Thewestbeaconproducesweirdresults throughout the regular approach but. The best fit location sometimes tangents thebordersoftheroom.Thisistheresultoftheactualbestfitbeingoutsidethedefinedroom.Inthe regular approach, 2/3 coordinates tangents the borders. In the A&N-approach, 2/3coordinatestangentstheborders,intheA-approach0/3coordinatestangentstheborders,andin the N-approach 3/3 coordinates tangents the borders. One of these is also situated in thecorneroftheroom.It isobviousthattheapproachesthatgeneratelessborderingcoordinateswill,ingeneral,bemorelikelytoproducealowersummederror.

Whenstudyingtheseresultsintermsofaccuracy,onemustconsidertheenvironmentinwhichthe test took place. The room, inwhich thesemeasurementswere taken,was kept free fromimpactofhumanbodiesandnon lineofsightconditions.[19]Thissterilesettingmostcertainlypushedtheresultstowardslessunbiasedresultsthanitotherwisewouldhavebeeninthelivecase,where the indoor positioning application somehow interactswith people using it.Withthis inmindandwhen lookingat theresults fromthetrilateration,oneneedstobecareful tomakestrongstatementsaboutit’sfeasibilityinapplication.Itseemslikesomemoremethodsofweighting are needed to boost the results. Applying linear regression and only using thegeneratedreferencevalue,helpedimprovetheresultsquiteabit.

When it come to the implementation, the regular approach and the weighted approach aresimilarintermsofwhatneedstobeimplementedinsoftwarewhencreatinganapplicationthatutilizestrilaterationandiBeacons.Withinanapplication,theweightedapproachonlyneedstoadjusttheRSSImatrix,basedontheprocessdescribedintheimplementationsection,beforeitcansetupthesystemofequationsthatdescribestheposition.The implementationandsetupwork for these two approaches are similar with a slight advantage towards the regulartrilaterationapproach.

Discussion

38

5.2FingerprintingTheresultsfromthetwofingerprintingapproachescanbeobservedinTable4.12andTable4.13.Bothresultsarepromisingbutinoverall,theregularapproachperformedabitbetter.Thetotaldistancesbetweenthecalculatedtestpointsandtheiractualpositionsare in theregularapproach 4.44meters, and in theweighted approach 5.1meters. Theweighted approach didreceive amore accurate result for test point two,which is themost central test point in theroom.

What isworth noting about the regular fingerprinting approach is that in neither of the testpoints, theactualpositionscalculatedwerenotbetween the threestrongest referencepoints.The midpoint of the three strongest reference points did, nevertheless, end up close to theactual position in all three cases. A weighted midpoint taking more reference points intoconsideration would be a solid approach in order to boost the results. This approach washowever the most accurate in terms of Euclidean distance from calculated points to actualpositions.

Intheweightedapproach,itshouldbenotedthatthethreeresultsalongtheX-axisallhappenedtobeclosetothecenteroftheCartesianplanedefinedintheroom.ThecalculatedXpositionswere2.58,2.29and2.26.Thedifference fromtheseXcoordinatesandthecenterof theroomalong the X-axis is 0.23 meter, 0.06 meter and 0.09 meter. In this setting, it seems like theweightedalgorithmcouldhaveapulltowardsthegeometriccenteroftheroom.Anotherwaytolook at it, is to describe it as really accurate since the X coordinate of the test points werelocatedat2.35,1.675and2.35respectively.

If the algorithm indeedhas a pull then thedata amongst the Y-axis should indicate the samepattern. Along the Y-axis, the difference between calculated positions and the center of Y iscalculatedtobe2.42meters,1.58metersand1.50meters.

Thetestpoints’positionsaremorespreadoutalongtheY-axisthantheyarealongtheX-axis.ObservingTable4.13,apullingbehaviourisnotobviousbuttherearesometendencies.Ifthereissuchbehaviourinthealgorithmitcanbeaneffectofeitherthepositioningofthereferencepoints,theselectedweightingapproach,orthesteppingmethodingeneral.

When it comes to the implementation,more referencepoints should increase theaccuracyofthe algorithm, but it also results inmorework of pre recording these.Whendoing a generalapproach, the reference point needs to be multiplied with some constant that adjusts themeasurements toeachseparatedevicesince it isknownthat resultsmayvary fromdevice todevice.Implementingthisalgorithminalivescenariothusaddsacalibrationstepforeachnewdevice that would like to utilize the system. A similar constant would be needed in thetrilaterationcaseaswell.

Overall the fingerprintingalgorithmperformedwellbut itsbigdisadvantage is the setup.Forevery new site in which a fingerprinting algorithm is to be applied, a new radio map withreferencepointsisneeded.

Discussion

39

5.3TrilaterationandfingerprintingIngeneralthefingerprintingalgorithmsperformedbetterintermsofaccuracy.Comparingthetotal Euclideandistancebetween calculatedpoints and actual points the following values areobserved,4.44metersforfingerprinting,5.1metersforweightedfingerprinting,9.22metersfortrilaterationand6.63metersforthebestweightedtrilaterationapproach.A reason why the trilateration approach performed lesser could be the choice of lowtransmissionstrengthonthebeacons.Asstatedbefore,thestrengthofthebeaconsweresettothepredefinedsettingof-12dBm,whichletsthebeaconsbroadcastforroughlythreeyears.Thetrilateration approach may be more sensitive to low signal strengths since RSSI to distancecalculation is considered unreliable at distances over a fewmeters.With a higher broadcaststrength, thisreliable intervalbecomeslarger.Acomparisononahigherstrengthwouldmostlikely improve the fingerprinting results aswell, but thequestionhere iswhether the resultswouldbemorevisibleinthetrilaterationcaseoverthefingerprintingcase.

In terms of accuracy, from the findings of this study, the fingerprinting approach performsbetterthanthetrilaterationapproach.

There are many ways to improve these algorithms, for instance in both these algorithms,implementingmethods thatwould take people into consideration is a possibility. Based on auser’s previousposition, and anewposition, a senseof direction canbe estimated.With thissenseofdirection,consideringthatmostpeopleholdtheirdevicesinfrontofthemwhentheyareactivelyusingtheirphones,theexpectedRSSIreceivedfromdirectionsbehindtheusercanbelowered.

In terms of implementation and method, the fingerprinting algorithms’ disadvantage incomparison to trilateration is of course the setting up of the radio map. The trilaterationalgorithms also need some measurements, such as deciding the path loss constant and thesignalstrengthsatonemeterfromeachbeacon.Buttheoverallworkisless.Whenitcomestosetting up a live software application, an application for the purpose of fingerprinting needsaccesstoradiomapdatabase.

In general, the advantages of the fingerprinting algorithm outweigh the disadvantages incomparison to trilateration. Within the boundaries of this test, this algorithm must beconsidered to be the better approach. Within a setting that is more dependent on goodscalabilityhowever, the regular trilaterationapproachcouldbeagoodchoice since itdoesn’tdiffertoomuchfromthefingerprintingapproachesintermsofsummederror.

Discussion

40

5.4AdvantagesofBluetoothSmartandbeaconsThegeneraladvantagesofbeaconsareasdiscussedbefore,thelowcostsandthefactthattheyare independent from power sources as they run on batteries. In comparison to regularBluetooth, Bluetooth Smart has the broadcasting ability, which allows IPS of over 7 units.RegularBluetoothis limitedtothisnumberbythemaster-slave identificationbits.MoreunitswithinBluetoothcanbeseen,butafullBluetoothscantakes10.24seconds.Ifamasterwishestoreplaceaslavewhilemovingthroughalargersystem,thescancycletimecanbeproblematic.

A disadvantagewithBluetooth Smart beacons that run on batteries is that in a deployed IPSbasednetwork,thesewilleventuallyhavetobereplaced.Theaccuracy,intermsoftransmissionstrength,ofthesystemalsohasatradeofflinkedtothelifeexpectancyofeachbeacon.RegularBluetoothprovidesbetteraccuracybutislimitedtothenumberofunitsthatcanbedeployedinasystem.ComparedtoWi-Fi,Bluetoothsmarthassimilarsignalstrengths.

Whendeploying a fingerprinting IPS, usingbeacons andBluetooth Smart is a solid approach.Having many well-distributed beacons in a larger network would provide a large matrix ofpositionsandsignalstrengths.Inalargesetting,somebeaconsmightnotproduceafingerprintinsomecoordinates.Inordertoretrievebettercomparisonsinsuchasystemwhereallbeaconsarenotviableatonce,onlyconsideringtheEuclideandistancemightprovedisadvantageous.Asystem like this would need to take into consideration which beacons are producing whatdistances. Such a system could match the relevant beacons from the reference points. If areference point cannot see beacon A, then beacon A should not contribute to the Euclideandistancecalculationtowardsthatreferencepoint.

In a trilateration approach, beacons based on Bluetooth Smart are perhaps not the optimalchoice. Since calculating distance from RSSI is considered unreliable, utilizing a radiotechnology with stronger signals might be preferable. For instance, an implementation withregularBluetoothmightprovidebetterresultsorutilizingsystemsalreadyinplacesuchasWi-Firouters.

Discussion

41

5.5ErrorfactorsandsourcesAfirsttopicoferrorfactorsistheproblematicwestbeacon.Fromtheresults,itwaspointedoutthatthewestbeaconhadlowertransmissionstrengththantherest.ItisnotedinTable4.2,thatthewestbeaconproducesapositiveresultof0.2inthecalculationsoftheroomconstantswhileas theotherbeaconsproduceanegativeresultbetween -.81and -2.31.This isaconsequencefromtheaverageof thereferencepointshavingahigherRSSI thenthereferencevalueatonemeter.InTable5.1acomparisonbetweentheaverageRSSIofthereferencepointsiscomparedto the reference value. This behaviour is not visible from the four other beacons where thetrendisthatreferencepointslocatedfurtherawayhasalowerRSSIthenreferencevalue.Thesefour beacons also display a larger gap between reference value and average from thewholeseries.

Beacon Northeast Southeast West North SouthReferencevalue -65.9 -73.8 -82.2 -64.9 -66.6

AverageRSSIofwholeseries -78.4 -82.6 -81.8 -79.7 -74.3Table5.1AverageRSSIatreferencepointscomparedtoreferencevalues.

Overall, the impact of this beaconprobably affects the trilateration algorithmsmore than thefingerprintingalgorithms.Since trilateration reliesonconvertingRSSI intodistance, abeaconwithaconsiderablyweakeroutputbecomesunreliableatashorterdistancethantherest.Inthetrilaterationapproach,thisbeacondirectlyaffectsthedistancetowardsthewestbeacon.Inthefingerprinting approach this error is spread out in 15 comparisons towards each referencepoint.Itcanalsobenotedthatitisthereferencevalue,andnottheaverageofthewholeseriesthatstandsoutinthecaseofthewestbeacon.

It is mentioned in the 3.1.2.2 that the screen was always facing the roof and the speakerpointing in the direction defined as the north wall. From [23] it is know that orientationmatters.However,notoolswereusedtoensureperfectanglestowardsthenorthwallsosmallvariationinanglestowardsthewallmayoccur.

Whenitcomestotheplacementofthedeviceduringtherecordingsintermsofheightabovetheground.Theremayalsobesomevariationsbecausedifferentboxesandobjectswhereusedtoreachtheheightof150centimeters.Thiscombinationofboxesandobjectsmayvariedabitinheightdependingonhowtheywherestack.

In [16][19] the effect of humanbodies andobstacles aredescribed.Nohumanbodieswhereplacedwithinthetestareaorinterferingwiththeperfect lineofsightconditions.However, insome of the recordings employees had to move between a meeting in a room next to thenortheastbeaconthroughtheopenspacebehindthepillarofthesoutheastbeaconandoutintothemainhall.Theydidnotcrossthetestspaceatall.Itishardtodetermineifthiswouldcauseany interference. Each transmission is sent out every second, and only received once. It isunlikely that a reflectionwould be received over a direct signal. This can only happen if thedirectsignalisnotrecorded.

From[16]itisknownthatdifferentmaterialshavedifferentpathlosswhilebeingpenetrated.Thisshouldalsoeffectthereflectionandhowstrongthesignalisafterbouncingofawall.Itishardtodeterminehowthedifferentwalltypeswithintheroominteractedwiththeresult.Like

Discussion

42

inthecaseofpeoplepassingoutsidethetestarea,onlynon-directsignalswouldbeaffected.Acomparisoninsideadefinedareawithzerowallscouldturnouttobeinteresting.

From[23]itisknownthatRSSIvariesbetweendifferentdevices.Inthetrilaterationapproachthis is taking into consideration with different reference and room constants for differentbeacons.Inthecaseoffingerprinting,thisisnotaccountedfor.Inthecaseoftheunreliablewestbeacon,itwasconcludedthattheimplicationsarespreadouttoallthetestpoints.However,ifbeacons seem to demonstrate a wide range of average RSSI series, a normalization functioncouldbeneededtoprovideallbeaconswithsimilarimpact.

Bluetooth technology, with its signal hopping is designed to interact with other radiotechnologies with minimal interferences. It is hard to determine how or what could’veinterfered.Itishoweverknown,thatnootherBluetoothSmartdeviceswhereregisteredintheproximitytothetestsitewhilethetestwasbeingconducted.

Conclusionsandfuturework

43

6 ConclusionsandfutureworkInthischapterconclusionsandpropositionsforfutureworkandresearcharepresented.

6.1ConclusionsOverall, the fingerprinting approach was the most feasible choice of approach within in theboundariesofthisstudy.Asstatedbefore,theunreliabilityofcalculatingdistancesfromRSSIisdisadvantageous to the trilateration approach. The fingerprinting algorithm performedmoreaccurate in both the weighted approach and the regular approach. The best performingalgorithmwhen it comes toaccuracywas the regular fingerprintingapproach. Itwas in total,summinguptheerrors,slightlybetterthantheweightedapproach.

Thereare two factors thatwork in the favourof trilateration; theseare scalabilityand setup.Theseareactuallyconnected.Inafingerprintingsystem,thereismoresetuprequired.Aradiomapneedstobecreatedandhavingmorereferencepointsprovidebetterresults. Intermsofscalability,thecomplexityofthesetupprocessincreasesasthesystemgrowsifsameamountofaccuracyiswanted.Inalargesystem,somekindofweightedtrilaterationapproachmightprovetobeadvantageous,asalargertrilaterationsystemgenerallyjustneedsmorebeacons.

Whenitcomestoscalability,asystemthatusessomekindofweightingandothermethodsofboosting resultswould certainlywork in a trilateration approach favour. In a large system, aradiomapcoveringmanylocationsandmanybeaconswillbedemandingtosetup.Sospeakingintermsofscalabilityatrilaterationapproachisadvantageous.

To roundof thiswork letsbriefly reconnectwithour formulated researchquestions, startingwiththesub-questions.

What positioning algorithms can be utilized within Bluetooth Smart and similar radiotechnologies?

In the literature review, it was concluded that a trilateration approaches and fingerprintingapproachesaresuitableforthepurposecreatinganIPSwithBluetoothSmart.

Howdoesthesepositioningalgorithmsperformagainsteachotherintermsofaccuracy,setupandimplementation?Intermsofaccuracy,afingerprintingapproachhasproventobebetterwithintheboundariesofthisstudy.Whenitcomestoimplementationandsetup,atrilaterationapproachisbetter.What advantages and disadvantages does a Bluetooth Smart system implementing thesepositioningalgorithmshavecomparedtootherradiotechnologies?The advantages of Bluetooth Smart are the low costs and the mobility. A broadcastingfunctionality allows implementation of larger systems than regular Bluetooth. Thedisadvantages are the weak signals, and the fact that beacons will constantly needs to bereplacedinanIPS.

IsBluetoothSmartaviablechoiceforthepurposeofbuildinganindoorpositioningsystem?

Conclusionsandfuturework

44

Overall, Bluetooth Smart is a viable choicewhen constructing an IPS, depending on location,scale, and establishment of other systems such as Wi-Fi and Bluetooth. The fingerprintingapproachwouldbethechoiceforsuchasystem.

As a concluding remark,BluetoothSmart and iBeacon implementations can functionwell butthey will most likely not be the groundbreaking technology that will allow location basedservicestothriveasithaswithGPSinoutdoorenvironments.

6.2FutureworkSeveral topics that couldbeexamined furtherarenoticedduringaprocess like this.Herearesomesuggestionsforfuturework,ideasencounteredduringthisproject.

It is mentioned throughout the discussion and the conclusion that the results from thetrilaterationapproachcouldbemoreaccurateifhighertransmissionstrengthsweretobeused.Inthetradeoffbetweenhavingasystemthatcouldrunforalongerperiodoftimeandselectinghightransmissionstrength,theformerwaschosen.Oneaspecttoinvestigatewithfutureworkwouldbetodothesecomparisonsondifferenttransmissionstrengths.

The fingerprinting algorithms proved to bemore exact by this study,with the disadvantagesthatcomeonadeploymentofa largerscale, therefore,aninterestingselectionof futureworkwouldbe investigating if thedeploymentprocesscouldbe facilitated,orevenautomated.Forinstance,onecouldcreateamodelbasedonmatchingblueprintsandextrapolatingradiomapsofalreadyknownblueprints into similarunknown locations.Anotherexample is a radiomapthatisconstantlyupdatingitselfwhileitisbeingused.Withknownfacts,suchasentrypoints,anddirectiondatamightbecontinuallyevaluatedandupdated.

References

45

References[1]BluetoothOrganization-AbrieftutorialonBluetoothwirelesstechnology-http://www.bluetooth.com/Pages/Fast-Facts.aspx-Lastvisited23/06/2015

[2]BluetoothOrganization-AboutU-http://www.bluetooth.com/Pages/about-bluetooth-sig.aspx-Lastvisited23/06/2015

[3]BluetoothSIG-SpecificationAdoptedDocuments-https://www.bluetooth.org/en-us/specification/adopted-specifications/-Lastvisited23/06/2015

[4]BluetoothOrganization-Bluetooth4.2Smarter,Faster,EnablesIoT|BluetoothTechnologyWebsite-http://www.bluetooth.com/SiteCollectionDocuments/4-2/bluetooth4-2.aspx?_ga=1.103181632.17828467.1430293926-Lastvisited23/06/2015

[5]BluetoothOrganization-ALookattheBasicsofBluetoothTechnology

http://www.bluetooth.com/Pages/Basics.aspx-Lastvisited23/06/2015

[6]A.Franssens,ImpactofmultipleinquiriesontheBluetoothdiscoveryprocess-Anditsapplicationtolocalization,UniversityofTwente(2010).

[7]A.Huang;LRudolph,BluetoothEssentialsforProgrammers,p36,(2007).

[8]BluetoothOrganization-BluetoothSmartandSmartReadyproductsnowavailable-http://www.bluetooth.com/Pages/Bluetooth-Smart-Devices-List.aspx-Lastvisited23/06/2015

[9]TexasInstruments,SimpleLink™Bluetooth®lowenergyCC2640wirelessMCUSoftwareDeveloper’sGuide,(2015),section1.2

[10]ACavallini,iBeaconBible2.0,Gaia-Matrix,p14,(2014).

[11]EstimoteOrganization-IntroducingtheEstimoteIndoorLocationSDK,thenextlevelofcontextformobileapps-http://blog.estimote.com/post/98316374485/introducing-the-estimote-indoor-location-sdk-the-Lastvisited23/06/2015

[12]InstituteofElectricalandElectronicsEngineers-WorkingGroupProjectTimelines-http://grouper.ieee.org/groups/802/11/Reports/802.11_Timelines.htm-Lastvisited23/06/2015

[13]NSaiShankar;D.Dash;H.ElMadi;GGopalakrishnan,WiGigandIEEE802.11adForMulti-Gigabyte-Per-SecondWPANandWLAN,(2007).

[14]A.Huurdeman,TheWorldwideHistoryofTelecommunications,p529,(2003).

[15]CDGWorldwide-WorldwideDatabase-http://www.cdg.org/worldwide/index.asp?h_area=0&h_technology=999&h_frequency=1-Lastvisited23/06/2015

[16]S.Seidel;T.Rappaport,914mhzpathlosspredictionmodelsforindoorwirelesscommunicationsinmultiflooredbuildings,(1992).

References

46

[17]Q.Dong;W.Dargie,EvaluationoftheReliabilityofRSSIforIndoorLocalization,WirelessCommunicationsinUnusualandConfinedAreas(2012).

[18]N.Mahiddin;NSafie;E.Nadia;S.Safei;E.Fadzli,IndoorpositiondetectionusingWiFiandtrilaterationtechnique,FacultyofInformatics,UniversitySultanZainalAbidin(2012).

[19]K.Kaemarungsi;P.Krishnamurthy,PropertiesofIndoorReceivedSignalStrengthforWLANLocationFingerprinting,SchoolofInformationSciences,UniversityofPittsburgh,(MobiQuitous’04),(2004).

[20]N.K.Sharma,AWeightedCenterofMassbasedTrilaterationApproachforLocatingWirelessDevicesinIndoorEnvironment,CAComputerAssociates,(2006).

[21]ABekkelien,BluetoothIndoorPositioning,UniversityofGeneva,(2012).

[22]Y.Wang;S.Shi;X.Yang;A.Ma,BluetoothIndoorPositioningusingRSSIandLeastSquareEstimation,MacaoPolytechnicInstitute;BeijingUniversityofPostandTelecommunications,(2010).

[23]E.Dahlgren;H.Mahmood,EvaluationofindoorpositioningbasedonBluetoothSmarttechnology,ChalmersUniversityofTechnology,(2014).

Appendix

47

Appendix

AppendixAMeanRSSIvaluesfromthe15referencepointsandthethreetestpoints.

X Y Northeast Southeast West North South1.0 1.0 -81.4 -81.9 -90.2 -82.4 -64.81.0 4.0 -78.8 -78.1 -80.5 -85.0 -73.31.0 6.5 -75.2 -79.8 -82.2 -80.2 -78.51.0 9.0 -83.1 -86.8 -78.8 -81.0 -78.91.0 12.0 -71.7 -86.4 -88.0 -72.9 -82.12.35 1.0 -83.0 -87.2 -85.6 -86.9 -66.62.35 4.0 -75.3 -76.8 -74.6 -81.0 -69.52.35 6.5 -82.5 -82.8 -77.6 -82.8 -84.12.35 9.0 -75.0 -83.3 -80.0 -72.6 -75.72.35 12.0 -80.3 -86.0 -83.1 -64.9 -76.43.7 1.0 -88.5 -85.4 -85.7 -86.5 -64.13.7 4.0 -81.9 -73.8 -78.9 -86.3 -67.43.7 6.5 -81.8 -76.6 -80.1 -80.7 -74.03.7 9.0 -65.9 -82.8 -80.4 -74.6 -76.53.7 12.0 -71.0 -91.8 -81.0 -78.0 -82.5X Y Northeast Southeast West North South

2.35 2.50 -84.5 -76.3 -87.2 -82.8 -70.81.675 6.50 -83.4 -78.0 -76.5 -82.5 -72.62.35 10.50 -71.0 -85.4 -80.4 -74.6 -78.9

Appendix

48

AppendixBAppendixBdisplaysthedistancetoeachreferencepointandthecalculatedpulltowardseachreferencepoint.

AppendixB1InAppendixB1thecalculatedpullsanddistancesintestpointonearepresented.(2.35,2.5)

X Y Northeast Southeast West North South Distance Pull1 1 -3.1 5.7 3.0 -0.4 -6.1 9.35 242.331 4 -5.7 1.9 -6.8 2.2 2.5 9.64 233.371 6.5 -9.3 3.6 -5.0 -2.6 7.7 13.80 123.631 9 -1.4 10.6 -8.5 -1.8 8.0 15.91 81.191 12 -12.8 10.2 0.8 -10.0 11.3 22.23 7.23

2.35 1 -1.5 11.0 -1.6 4.0 -4.3 12.62 151.292.35 4 -9.2 0.5 -12.6 -1.8 -1.4 15.76 83.852.35 6.5 -2.0 6.5 -9.7 0.0 13.2 17.74 51.482.35 9 -9.5 7.1 -7.3 -10.2 4.9 17.93 48.782.35 12 -4.2 9.8 -4.2 -17.9 5.6 21.96 8.733.7 1 4.0 9.2 -1.6 3.7 -6.7 12.69 149.563.7 4 -2.6 -2.4 -8.4 3.4 -3.5 10.31 213.383.7 6.5 -2.7 0.3 -7.1 -2.1 3.2 8.52 268.763.7 9 -18.6 6.5 -6.9 -8.2 5.7 23.14 3.153.7 12 -13.5 15.6 -6.3 -4.8 11.6 24.92 0.00

Appendix

49

AppendixB2InAppendixB2thecalculatedpullsanddistancesintestpointtwoarepresented.(1.675,6.5)

X Y Northeast Southeast West North South Distance Pull1 1 -1.9 3.9 13.7 0.0 -7.8 16.38 41.871 4 -4.6 0.1 4.0 2.6 0.7 6.61 263.571 6.5 -8.2 1.8 5.7 -2.3 5.9 11.96 118.601 9 -0.2 8.8 2.3 -1.5 6.3 11.16 136.661 12 -11.7 8.4 11.5 -9.6 9.5 22.85 0.00

2.35 1 -0.4 9.2 9.1 4.4 -6.0 14.93 62.722.35 4 -8.0 -1.2 -1.9 -1.4 -3.1 9.02 191.172.35 6.5 -0.9 4.8 1.1 0.3 11.5 12.51 106.772.35 9 -8.4 5.3 3.4 -9.9 3.1 14.74 65.742.35 12 -3.1 8.0 6.6 -17.5 3.8 20.96 3.583.7 1 5.1 7.4 9.2 4.1 -8.4 15.92 48.043.7 4 -1.4 -4.2 2.4 3.8 -5.2 8.16 215.703.7 6.5 -1.6 -1.4 3.6 -1.8 1.5 4.77 326.633.7 9 -17.5 4.8 3.8 -7.9 4.0 20.49 5.573.7 12 -12.3 13.8 4.5 -4.5 9.9 21.91 0.88

Appendix

50

AppendixB3InAppendixB3thecalculatedpullsanddistancesintestpointthreearepresented.(2.35,10.5)

X Y Northeast Southeast West North South Distance Pull1 1 10.4 -3.5 9.8 7.8 -14.2 21.89 19.831 4 7.7 -7.3 0.1 10.4 -5.6 15.92 108.691 6.5 4.2 -5.6 1.8 5.6 -0.4 9.12 296.421 9 12.1 1.4 -1.6 6.4 -0.1 13.84 156.171 12 0.7 1.0 7.6 -1.7 3.2 8.50 318.15

2.35 1 12.0 1.8 5.2 12.2 -12.3 21.81 20.532.35 4 4.3 -8.7 -5.8 6.4 -9.5 16.05 105.952.35 6.5 11.4 -2.7 -2.8 8.2 5.1 15.48 117.882.35 9 4.0 -2.1 -0.4 -2.0 -3.2 5.90 417.962.35 12 9.3 0.6 2.7 -9.7 -2.5 13.89 154.933.7 1 17.5 0.0 5.3 11.9 -14.8 26.34 0.003.7 4 10.9 -11.6 -1.5 11.7 -11.6 22.91 11.793.7 6.5 10.8 -8.9 -0.3 6.1 -4.9 16.00 106.983.7 9 -5.1 -2.7 0.0 0.0 -2.4 6.24 403.893.7 12 0.0 6.4 0.6 3.4 3.6 8.07 333.72

www.kth.se