VisibleLightCommunicationVisible lightcommunication(VLC) isanevolvingcommunication technologyforshort-rangeapplications.Exploitingrecentadvancesinthedevelopmentofhigh-powervisible-light-emittingLEDs,VLCoffersanenergy-efficient,cleanalternative toRFtechnology,enabling the development of optical wireless communication systems thatmake use ofexistinglightinginfrastructure.

Drawing on the expertise of leading researchers from across the world, this concisebook sets out the theoretical principles of VLC, and outlines key applications of thiscutting-edge technology. Providing insight into modulation techniques, positioning andcommunication, synchronization, and industry standards, as well as techniques forimprovingnetworkperformance,thisisaninvaluableresourceforgraduatestudentsandresearchers in the fields of visible light communication and optical wirelesscommunication,andforindustrialpractitionersinthefieldoftelecommunications.

SHLOMI ARNON is a Professor at the Department of Electrical and ComputerEngineeringatBen-GurionUniversity(BGU),Israel.HeisaFellowofSPIE,aco-editorof Advanced Optical Wireless Communication Systems (2012), and has edited specialissuesof theJournal ofOpticalCommunications andNetworking (2006) and the IEEEJournalonSelectedAreasinCommunications(2009,2015).

VisibleLightCommunication

Editedby

ShlomiArnon

Ben-GurionUniversityoftheNegev,Israel

UniversityPrintingHouse,CambridgeCB28BS,UnitedKingdom

CambridgeUniversityPressispartoftheUniversityofCambridge.

ItfurtherstheUniversity’smissionbydisseminatingknowledgeinthepursuitofeducation,learningandresearchatthehighestinternationallevelsofexcellence.

www.cambridge.org

Informationonthistitle:www.cambridge.org/9781107061552

©CambridgeUniversityPress2015

Thispublicationisincopyright.Subjecttostatutoryexceptionandtotheprovisionsofrelevantcollectivelicencingagreements,noreproductionofanypartmaytakeplacewithoutthewrittenpermissionofCambridgeUniversityPress.

Firstpublished2015

PrintedintheUnitedKingdombyClays,StIvesplc

AcatalogrecordforthispublicationisavailablefromtheBritishLibrary

LibraryofCongressCataloginginPublicationdata

Visiblelightcommunication/editedbyShlomiArnon,BenGurion

UniversityoftheNegev,Israel.

pagescm

ISBN978-1-107-06155-2(hardback)

1.Opticalcommunications.I.Arnon,Shlomi,1968–

TK5103.59.V572015

621.382′7–dc23

2014046702

ISBN978-1-107-06155-2Hardback

CambridgeUniversityPresshasnoresponsibilityforthepersistenceoraccuracyofURLsforexternalorthird-partyinternetwebsitesreferredtointhispublication,anddoesnotguaranteethatanycontentonsuchwebsitesis,orwillremain,accurateorappropriate.

Everyefforthasbeenmadeinpreparingthisbooktoprovideaccurateandup-to-dateinformationwhichisinaccordwithacceptedstandardsandpracticeatthetimeofpublication.Nevertheless,theauthors,editorsandpublisherscanmakenowarrantiesthattheinformation,including,butnotlimitedto,anymethods,formulae,andinstructions,containedhereinistotallyfreefromerror.Theauthors,editorsandpublishersthereforedisclaimallliabilityfordirectorconsequentialdamagesresultingfromtheuseofmaterialcontainedinthisbook.Readersarestronglyadvisedtopaycarefulattentiontoinformationprovidedbythemanufacturerofanyequipmentthattheyplantouse.

Contents

Listofcontributors

Acknowledgment

1 Introduction

ShlomiArnon

2 Modulationtechniqueswithlightingconstraints

JaeKyunKwonandSangHyunLee

2.1 InversesourcecodingindimmableVLC

2.1.1 ISCforNRZ-OOK

2.1.2 ISCforM-aryPAM

2.1.3 Comparisonswithrespecttodimmingcapacity

2.2 Multi-leveltransmissionindimmableVLC

2.2.1 Multi-leveltransmissionscheme

2.2.2 Asymptoticperformance

2.2.3 Simulationresults

2.3 Colorintensitymodulationformulti-coloredVLC

2.3.1 Colorspaceandsignalspace

2.3.2 Colorintensitymodulation

3 PerformanceenhancementtechniquesforindoorVLCsystems

Wen-DeZhongandZixiongWang

3.1 Introduction

3.2 PerformanceimprovementofVLCsystemsbytiltingthereceiverplane

3.2.1 SNRanalysisofVLCsystemwithasingleLEDlamp

3.2.2 ReceiverplanetiltingtechniquetoreduceSNRvariation

3.2.3 MultipleLEDlampswiththereceiverplanetiltingtechnique

3.2.4 Spectralefficiency

3.3 PerformanceimprovementofVLCsystemsbyarrangingLEDlamps

3.3.1 ArrangementofLEDlamps

3.3.2 BERanalysis

3.3.3 Channelcapacityanalysis

3.4 DimmingcontroltechniqueanditsperformanceinVLCsystems

3.4.1 BipolarOOKsignalunderdimmingcontrol

3.4.2 AdaptiveM-QAMOFDMsignalunderdimmingcontrol

3.5 Summary

4 Lightpositioningsystem(LPS)

MohsenKavehradandWeizhiZhang

4.1 Indoorpositioningandmeritsofusinglight

4.1.1 Introductiontoindoorpositioning

4.1.2 Spectrumcrunchandfuturemobilesystem

4.1.3 AdvantagesofVLC-basedpositioning

4.2 Positioningalgorithms

4.2.1 Triangulation

4.2.2 Triangulation–circularlateration

4.2.3 Triangulation–hyperboliclateration

4.2.4 Triangulation–angulation

4.2.5 Sceneanalysis

4.2.6 Proximity

4.2.7 Comparisonofpositioningtechniques

4.3 Challengesandsolutions

4.3.1 Multipathreflections

4.3.2 Synchronization

4.3.3 Channelmulti-access

4.3.4 Serviceoutage

4.3.5 Privacy

4.4 Summary

5 Visiblelightpositioningandcommunication

ZhengyuanXu,ChenGong,andBoBai

5.1 Introduction

5.1.1 Indoorlightpositioningsystem

5.1.2 Outdoorlightpositioningsystem

5.2 Indoorlightpositioningsystemsbasedonvisiblelightcommunicationandimagingsensors

5.2.1 Systemdescription

5.2.2 LPSwithknownLEDpositions

5.2.3 Monte-Carlosimulationresults

5.3 OutdoorlightpositioningsystemsbasedonLEDtrafficlightsandphotodiodes

5.3.1 Lightpositioningsystem

5.3.2 Calibrationoferrorinducedbynon-coplanargeometry

5.3.3 Numericalresults

5.4 Summary

6 Thestandardforvisiblelightcommunication

KangTae-Gyu

6.1 ScopeofVLCstandard

6.1.1 VLCserviceareacompatibility

6.1.2 VLCilluminationcompatibility

6.1.3 VLCvendorcompatibility

6.1.4 Standardcompatibility

6.2 VLCmodulationstandard

6.2.1 VariablepulsepositionmodulationVPPM

6.2.2 Linecoding

6.3 VLCdatatransmissionstandard

6.3.1 Wiredtransmissionprotocol

6.3.2 Wirelesstransmissionprotocol

6.4 VLCilluminationstandard

6.4.1 LEDlightingsourceinterface

6.4.2 Fixtureinterface

6.4.3 LEDintelligentsystemlightinginterface

6.4.4 VLCservicestandard

7 Synchronizationissuesinvisiblelightcommunication

ShlomiArnon

7.1 Introduction

7.2 VLCmodulationmethodsinthetimedomain

7.2.1 Onoffkeying(OOK)

7.2.2 Pulsepositionmodulation(PPM)

7.2.3 Inversepulsepositionmodulation(IPPM)

7.2.4 Variablepulsepositionmodulation(VPPM)

7.3 Biterrorratecalculation

7.3.1 OOKBER

7.3.2 PPMBER

7.3.3 IPPMBER

7.3.4 VPPMBER

7.4 TheeffectofsynchronizationtimeoffsetonIPPMBER

7.4.1 TheeffectofclockjitteronIPPMBER

7.5 Summary

8 DMTmodulationforVLC

Klaus-DieterLanger

8.1 Introduction

8.2 Indoorapplicationscenarios

8.3 Aspectsofhigh-speedVLCtransmission

8.3.1 LEDmodulationbandwidth

8.3.2 Channelcapacity

8.3.3 Considerationsonhigh-speedLEDmodulation

8.4 DMTmodulationandvariants

8.4.1 DC-biasedDMT

8.4.2 AsymmetricallyclippedopticalOFDM(ACO-OFDM)

8.4.3 Pulse-amplitude-modulateddiscretemultitone(PAM-DMT)

8.4.4 DMT/OFDMperformanceandmitigationofdisruptiveeffects

8.5 PerformanceenhancementofDMTmodulation

8.5.1 CombinationofACO-OFDMandDC-biasedDMTmodulation

8.5.2 SpectrallyfactorizedOFDM

8.5.3 Flip-OFDM

8.5.4 UnipolarOFDM

8.5.5 PositionmodulatingOFDM

8.5.6 Diversity-combinedOFDM

8.5.7 Furtherapproaches

8.6 Systemdesignandimplementationaspects

8.6.1 Aspectsofsystemdesign

8.6.2 DMT/OFDMapplicationinadvancedsystems

8.6.3 Practicalimplementationissues

8.6.4 Implementationanddemonstration

8.7 Summary

9 Imagesensorbasedvisiblelightcommunication

ShinichiroHaruyamaandTakayaYamazato

9.1 Overview

9.2 Imagesensors

9.2.1 CCDimagesensor

9.2.2 CMOSimagesensor

9.2.3 ComparingCCDimagesensors,CMOSimagesensors,andphotodiodes(PD)

9.3 ImagesensorasaVLCreceiver

9.3.1 Temporalsampling

9.3.2 Spatialsampling

9.3.3 Maximalachievabledatarate

9.4 DesignofanimagesensorbasedVLCsystem

9.4.1 Transmitter

9.4.2 Receiver

9.4.3 Channel

9.4.4 Field-of-view(FOV)

9.4.5 Effectofcommunicationdistanceandspatialfrequency

9.5 Massivelyparallelvisiblelighttransmission

9.5.1 Concept

9.5.2 Systemarchitecture

9.5.3 Linkestablishment

9.5.4 Prototypeofamassivelyparalleldatatransmissionsystem

9.6 Accuratesensorposeestimation

9.6.1 Overview

9.6.2 Singleviewgeometry

9.6.3 Poseestimationusinglights

9.6.4 Lightextraction

9.7 Applicationsofimagesensorbasedcommunication

9.7.1 Trafficsignalcommunication

9.7.2 Positionmeasurementsforcivilengineering

9.8 Summary

Index

Contributors

ShlomiArnon

Ben-GurionUniversityoftheNegev,Israel

BoBai

NorthwesternPolytechnicalUniversity,China

ChenGong

UniversityofScienceandTechnologyofChina

ShinichiroHaruyama

KeioUniversity,Japan

MohsenKavehrad

ThePennsylvaniaStateUniversity,USA

JaeKyunKwon

YeungnamUniversity,Korea

Klaus-DieterLanger

FraunhoferHeinrichHertzInstitute(HHI),Germany

SangHyunLee

SejongUniversity,Korea

KangTae-Gyu

ElectronicsandTelecommunicationsResearchInstitute(ETRI),Korea

ZixiongWang

TheHongKongPolytechnicUniversity

ZhengyuanXu

UniversityofScienceandTechnologyofChina

TakayaYamazato

NagoyaUniversity,Japan

WeizhiZhang

ThePennsylvaniaStateUniversity,USA

Wen-DeZhong

NanyangTechnologicalUniversity(NTU),Singapore

Acknowledgment

Iwouldliketoexpressmyspecialappreciationandthankstomybelovedwife,whohashelpedmeforsomanyyears,andfortheinterestingdiscussionswithmykidsthatalwayschallengeme.Ialsowanttothankallthebookcontributors,whohavemadeitwhatitis–withoutthemthebookwouldneverhavebeenrealized.

1 Introduction

ShlomiArnon

Visible light communications (VLC) is the name given to an optical wirelesscommunicationsystemthatcarriesinformationbymodulatinglightinthevisiblespectrum(400–700nm)thatisprincipallyusedforillumination[1–3].Thecommunicationssignalis encodedon topof the illumination light. Interest inVLChasgrown rapidlywith thegrowth of high power light emitting diodes (LEDs) in the visible spectrum. Themotivationtousetheilluminationlightforcommunicationistosaveenergybyexploitingthe illumination to carry information and, at the same time, to use technology that is“green” in comparison to radio frequency (RF) technology, while using the existinginfrastructure of the lighting system. The necessity to develop an additional wirelesscommunication technology is theresultof thealmostexponentialgrowth in thedemandfor high-speed wireless connectivity. Emerging applications that use VLC include: a)indoorcommunicationwhereitaugmentsWiFiandcellularwirelesscommunications[4]whichfollowthesmartcityconcept[5];b)communicationwirelesslinksfortheinternetof things (IOT) [6]; c) communication systems as part of intelligent transport systems(ITS)[7–14];d)wirelesscommunicationsystemsinhospitals[15–17];e)toysandthemeparkentertainment[18,19];andf)provisionofdynamicadvertisinginformationthroughasmartphonecamera[20].

VLCtoaugmentWiFiandcellularwirelesscommunicationinindoorapplicationshasbecomeanecessity,withtheresultthatmanypeoplecarrymorethanonewirelessdeviceat any time, for example a smart phone, tablet, smart watch, and smart glasses and awearable computer, and at the same time the required data rate from each device isgrowingexponentially.It isalsobecomingincreasinglyclearthatinurbansurroundings,humanbeingsspendmostoftheirtimeindoors,sothepracticalityofVLCtechnologyisself-evident.Itwouldbeextremelyeasytoaddextracapacitytoexistinginfrastructurebyinstalling a VLC system in offices or residential premises. In Fig. 1.1 we can see anexample of a VLC network that provides wireless communication to a laptop, smartphone,TV,andwearablecomputer.

Figure1.1 VLCwirelessnetwork.

Thedownlink includes illuminationLED,Ethernet power line communication (PLC)modem,andLEDdriver,whichreceivesasignalbyadedicatedordonglereceiveraspartofthedevice.Theuplinkconfigurationcouldbebasedforexampleon:a)aWiFilink;b)an infrared–IRDA link; or c) a modulated retro reflector (Fig. 1.2). A modulated retroreflectorisanopticaldevicethatretroreflectsincidentlight[21,22].Theamplitudeoftheretroreflectedlightiscontrolledbyanelectronicsignal,asaresultmodulationofthelightcanbeachieved.Inthecasesofaninfrared–IRDA[23]linkoramodulatedretroreflectorthe receiver could be part of the illumination LED. In that case the uplink receiverincludesphotodiode, transimpedanceamplifierandmodem.Inthisway,anoperationalwirelessnetworkcouldbecreatedinnexttonotime.

Figure1.2 Awirelesscommunicationnetworkbasedonmodulatedretroreflector.

In the near future, billions of appliances, sensors and instrumentswill havewirelessconnectivity,ascanbeanticipatedfromtherevolutionaryconceptoftheinternetofthings(IOT).This technologymakes it possible to have ambient intelligence and autonomouscontrolwhichcouldadapttheenvironmenttotherequirementsandthedesiresofpeople.VLCcouldbeavery relevantwireless communication technology that is cheap, simpleandimmediateanddoesnotencroachonanalreadycrowdedpartoftheelectromagneticspectrum.

Intelligent transport systems (ITS) are an emerging technology for increasing roadsafety and reducing the number of road casualties as well as for improving trafficefficiency(Fig.1.3).VLChavebeenproposedas ameans forproviding inter-vehicularcommunication and for establishing connectivity between vehicular and roadinfrastructure,suchastrafficlightsandbillboards.Thesesystemsprovideone-wayortwo-wayshort- tomedium-rangewirelesscommunicationlinksthatarespecificallydesignedfortheautomotivesphere.Thetechnologyusestheheadlightsandtherearlightsofcarsastransmitters, and cameras and detectors as the receivers. The traffic lights are thecounterpartofatransmitterinthissphere.

Figure1.3 AnintelligenttransportsystemusingVLC.

Themedicalcommunitypursueswaystoimprovetheefficiencyofhospitalsandatthesame time to reduce hospital-acquired infections, which are very costly in money andhuman life. One way to upgrade the communication infrastructure is by wirelesstechnology.Thetechnologymakesitpossiblefordoctorstoaccessandupdatepatientdatausing tablet computers at the patient’s bedside instead of manually keeping paperdocumentsthatarekepteitheratthebedsideorinthenurses’backofficestation.Anotherapplicationisadeviceusedtomonitorpatientwell-beingandvitaldataremotely.Duetothe fact that RF, like WiFi and cellular, is a best effort technology, meaning that thetransmissionofinformationisnotguaranteed,interferencefromnearbydevicescouldjamthecommunication.Thissituationisunacceptableformedicalapplicationsandthereforeaswitch toVLC technology is self-evident. It is clear that VLC technology can providelocalizedsolutionsimmunetointerferenceandjammingforthemedicaldomain.

Thetoysandthemeparkentertainmentsectorisaveryinterestingapplicationthattakesadvantage of VLC technology due to two main characteristics (Fig. 1.4). The firstcharacteristic is theability tocommunicateby lineof sightor semi lineof sight, so thecommunication is localized to a specific volume. This makes it possible to providelocation based information in the theme park so the audience will have a multi-dimensional andmulti-sensory experience.The same concept could be used in the toysmarket to communicate between toys, using the already present LED. The secondcharacteristic is the low cost required to implement the technology in toy and parkentertainment. One example of a way to reduce the cost of toys is using the toy LEDsimultaneouslyastransmitterandphotodiodereceiver.

Figure1.4 VLCintoysinthethemeparkentertainmentindustry(courtesyofS.Arnon).

Dynamicadvertisingthroughasmartphonecameraisanewareathatusesabillboardand illumination to transmit information that is detectedby the camera, and thenby anappropriate algorithm the communications data are extracted from the video. Thistechnology could be used to add an extra and dynamic layer of information toadvertisementsinthestreet,shoppingcenterandsubway.

TheIEEEhasdefinedanewstandard,IEEE802.15.7,whichdescribeshighdata rateVLC,up to96Mb/s,by fastmodulationofoptical light sources.Meanwhile, in severalexperiments around the world data rates of more than 500 Mb/s have been reported[24–27].Inadditionmanynewmethodsarebeingdevelopedtomaximizebitratesandtoprovide algorithms to manage interference and subcarrier re-use [27–33], methods toanalyzesynchronizationissues[34]andmethodstodesignanasymmetriccommunicationsystemusingmodulatedretro-reflectors(MRR)basedonnanotechnology[35].Inthelightof these developments a newmethodological book covering this subject is required inordertohelpprogressthetechnology.Theaspirationofthisbookistoserveasatextbookforundergraduateandgraduatelevelcoursesforstudentsinelectricalengineering,electro-opticalengineering,communicationsengineering,illuminationengineeringandphysics.Itis also intended to serve as a source for self-study and as a reference book for senior

engineersinvolvedinthedesignofVLCsystems.

The book includes nine chapters that cover the important aspects of VLC scientifictheory and technology. Following this introductory chapter, Chapter 2 deals withmodulation techniquesunder lightingconstraints,and iswrittenbyJaeKyunKwonandSangHyunLee.TheauthorsexplainthatthephysicallayerdesignofVLCsystemsisofsubstantiallydifferentcharacterthanconventionalwirelesssystems,notablywithregardtoanewaverage intensityconstraint.Thisnewconstraint,whichwillbe referred toas the“dimmingtarget”of theilluminationsystem, introducesanewdomainofsystemdesignthathasrarelybeenconsideredinexistingcommunicationmedia.Chapter3,byWen-DeZhongandZixiongWang,describesmethodssuchasreceiverplanetiltingandaspecialLEDlamparrangementasperformanceenhancementtechniquesforindoorVLCsystems.The next two chapters, 4 and 5, deal with very important aspects of VLC, which takeadvantage of the characteristics of light to obtain location information. Chapter 4, byMohsenKavehradandWeizhiZhang,outlinestheapplications,investigatescurrentaccesstotheradiospectrumandstudiesindepththeshiftingneedsofindoorpositioninginthevisible light spectrum.At the end of the chapter, challenges and potential solutions arediscussed. Chapter 5, by ZhengyuanXu, ChenGong, andBoBai, presents indoor andoutdoor lightpositioningsystems(LPS).For indoorLPS,combiningVLCwithpositionestimationmethods is presented and an optimal estimation algorithm is implemented atthe receiver to provide an unbiased estimate of the camera position. For outdoorautomotiveLPS,thelightsignalcouldbeemittedfromatrafficlight,carryingitspositioninformation.Then, thepositionof thevehicle couldbe estimatedbasedon the receivedposition informationof the traffic lightand the timedifferenceofarrival (TDoA)of thelightsignalattwophotodiodes.Chapter6,byKangTae-GyudescribesthestandardsforVLC. In this chapter, the lamp and electronic power are presented in terms of electricsafety in IECTC34.Later, other standards forVLC, such asPLASAE1.45 and IEEE802.15.7,whichneedanumberofprotocolsbetweenasendingpartyandareceivingpartyarediscussed,aswellaselectricsafety.ThischapteralsodiscussesthecompatibilitiesoftheVLC service area, illumination, vendor considerations and standards.Chapter 7, byShlomiArnon,presentsdifferentmodulationmethodsusedinVLC,suchasonoffkeying(OOK)pulsepositionmodulation(PPM), inversepulsepositionmodulation(IPPM)andvariable pulse position modulation (VPPM). Later, explanations are given on how tocalculate thebiterror rate (BER)foreachmodulationscheme.Adetaileddescriptionofhow to calculate the effect of synchronization time offset and clock jitter on the BERperformance is also presented. Chapter 8, by Klaus-Dieter Langer, describes discretemultitone (DMT) modulation for VLC and presents advanced and highly spectrallyefficient solutions for this scheme, such as DC-biased DMT, asymmetrically clippedopticalOFDM(ACO-OFDM)andpulse-amplitude-modulateddiscretemultitone (PAM-DMT). Chapter 9, by Shinichiro Haruyama and Takaya Yamazato, deals with imagesensor based VLC and introduces two unique applications using an image sensor: (1)massively parallel visible light transmissions that can theoretically achieve amaximumdata rate of 1.28 Gigabits per second; and (2) accurate sensor position estimation thatcannotberealizedbyaVLCsystemusingasingle-elementphotodiode(PD).Applicationsofimagesensorbasedcommunicationfortheautomotiveindustry,positionmeasurementsincivilengineeringandbridgepositionmonitoringarealsopresented.

References[1] ShlomiArnon,JohnBarry,GeorgeKaragiannidis,RobertSchober,andMuratUysal,eds.,AdvancedOpticalWirelessCommunicationSystems.CambridgeUniversityPress,2012.

[2] SridharRajagopal,RichardD.Roberts,andSang-KyuLim,“IEEE802.15.7visiblelight communication: Modulation schemes dimming support.” CommunicationsMagazine,IEEE50,(3),2012,72–82.

[3] IEEEStandard802.15.7forlocalandmetropolitanareanetworks–Part15.7:Short-rangewirelessopticalcommunicationusingvisiblelight,2011.

[4] Cheng-XiangWang,FouratHaider,XiqiGao,etal., “Cellular architecture and keytechnologies for 5G wireless communication networks.” IEEE CommunicationsMagazine52,(2),2014,122–130.

[5] Shahid Ayub, Sharadha Kariyawasam, Mahsa Honary, and Bahram Honary, “Apractical approachofVLCarchitecture for smart city.” InAntennasandPropagationConference(LAPC),2013Loughborough,106–111,IEEE,2013.

[6] TetsuyaYokotani,“ApplicationandtechnicalissuesonInternetofThings.”InOpticalInternet(COIN),201210thInternationalConference,67–68,IEEE,2012.

[7] Fred E. Schubert, and Jong Kyu Kim, “Solid-state light sources getting smart.”Science308,(5726),2005,1274–1278.

[8] Shlomi Arnon, “Optimised optical wireless car-to-traffic-light communication.”TransactionsonEmergingTelecommunicationsTechnologies25,2014,660–665.

[9] SeokJuLee,JaeKyunKwon,Sung-YoonJung,andYoung-HoonKwon,“Evaluationof visible light communication channel delay profiles for automotive applications.”EURASIPJournalonWirelessCommunicationsandNetworking(1),2012,1–8.

[10] Sang-YubLee,Jae-KyuLee,Duck-KeunPark,andSang-HyunPark,“Developmentofautomotivemultimedia systemusingvisible lightcommunications.” InMultimediaandUbiquitousEngineering,pp.219–225.Springer,2014.

[11] S.-H.Yu,OliverShih,H.-M.Tsai,andR.D.Roberts,“Smartautomotivelightingforvehiclesafety.”CommunicationsMagazine,IEEE51,(12),2013,50–59.

[12] Shun-HsiangYou,Shih-HaoChang,Hao-MinLin,andHsin-MuTsai,“Visiblelightcommunications for scooter safety.” In Proceeding of the 11th Annual InternationalConferenceonMobileSystems,Applications,andServices,509–510,ACM,2013.

[13] Zabih Ghassemlooy, Wasiu Popoola, and Sujan Rajbhandari, Optical WirelessCommunications:SystemandChannelModellingwithMatlab®.CRCPress,2012.

[14] Alin Cailean, Barthelemy Cagneau, Luc Chassagne, et al., “Visible lightcommunications: Application to cooperation between vehicles and roadinfrastructures.”InIntelligentVehiclesSymposium(IV),1055–1059,IEEE,2012.

[15] Ryosuke Murai, Tatsuo Sakai, Hajime Kawano, et al., “A novel visible lightcommunication system for enhanced control of autonomous delivery robots in a

hospital.”InSystemIntegration(SII),2012IEEE/SICEInternationalSymposium,510–516,IEEE,2012.

[16] Seyed Sina Torkestani, Nicolas Barbot, Stephanie Sahuguede, Anne Julien-Vergonjanne, and J.-P.Cances, “Performance and transmission power bound analysisfor optical wireless based mobile healthcare applications.” In Personal Indoor andMobileRadioCommunications(PIMRC), 22nd InternationalSymposium,2198–2202,IEEE,2011.

[17] YeeYong Tan, Sang-Joong Jung, andWan-YoungChung, “Real time biomedicalsignal transmission of mixed ECG signal and patient information using visible lightcommunication.” In Engineering in Medicine and Biology Society (EMBC), 35thAnnualInternationalConferenceoftheIEEE,4791–4794,IEEE,2013.

[18] Nils Ole Tippenhauer, Domenico Giustiniano, and Stefan Mangold, “Toyscommunicatingwithleds:Enablingtoycarsinteraction.”InConsumerCommunicationsandNetworkingConference(CCNC),48–49,IEEE,2012.

[19] StefanSchmid,GiorgioCorbellini,StefanMangold,andThomasR.Gross,“LED-to-LEDvisiblelightcommunicationnetworks.”InProceedingsofthefourteenthACMinternational symposiumonMobile ad hocNetworking andComputing, 1–10,ACM,2013.

[20] RichardD.Roberts,“UndersampledfrequencyshiftON-OFFkeying(UFSOOK)forcamera communications (CamCom).” In Wireless and Optical CommunicationConference(WOCC),645–648,IEEE,2013.

[21] Etai Rosenkrantz, and Shlomi Arnon, “Modulating light by metal nanospheres-embedded PZT thin-film.”Nanotechnology, IEEE Transactions on 13, (2), 222–227,March2014.

[22] EtaiRosenkrantzandShlomiArnon,“An innovativemodulatingretro-reflector forfree-space optical communication.” In SPIE Optical Engineering + Applications,p.88740D.InternationalSocietyforOpticsandPhotonics,2013.

[23] RobOtte,Low-PowerWirelessInfraredCommunications.Springer-Verlag,2010.

[24] Yuanquan Wang, Yiguang Wang, Chi Nan, Yu Jianjun, and Shang Huiliang,“Demonstrationof575-Mb/sdownlinkand225-Mb/suplinkbi-directionalSCM-WDMvisible light communication using RGB LED and phosphor-based LED.” OpticsExpress21,(1),2013,1203–1208.

[25] AhmadHelmiAzhar,T.Tran, andDominicO’Brien, “A gigabit/s indoorwirelesstransmission using MIMO-OFDM visible-light communications.” PhotonicsTechnologyLetters,IEEE25,(2),2013,171–174.

[26] Wen-YiLin,Chia-YiChen,Hai-HanLu,etal.,“10m/500MbpsWDMvisible lightcommunicationsystems.”OpticsExpress20,(9),2012,9919–9924.

[27] Fang-MingWu,Chun-TingLin,Chia-ChienWei,etal., “3.22-Gb/sWDMvisiblelightcommunicationofasingleRGBLEDemployingcarrier-lessamplitudeandphasemodulation.”InOpticalFiberCommunicationConference,OTh1G-4.OpticalSocietyofAmerica,2013.

[28] Giulio Cossu, Raffaele Corsini, Amir M. Khalid, and Ernesto Ciaramella, “Bi-directional400Mbit/sLED-basedopticalwirelesscommunicationfornon-directedlineof sight transmission.” In Optical Fiber Communication Conference, p. Th1F–2.OpticalSocietyofAmerica,2014.

[29] DimaBykhovskyandShlomiArnon,“Multipleaccessresourceallocationinvisiblelightcommunicationsystems.”JournalofLightwaveTechnology32, (8),2014,1594–1600.

[30] DimaBykhovskyandShlomiArnon,“Anexperimentalcomparisonofdifferentbit-and-power-allocation algorithms forDCO-OFDM.” Journal of LightwaveTechnology32,(8),2014,1559–1564.

[31] Joon-hoChoi,Eun-byeolCho,ZabihGhassemlooy,SoeunKim, andChungGhiuLee, “Visible light communications employing PPM and PWM formats forsimultaneousdatatransmissionanddimming.”OpticalandQuantumElectronics1–14,May2014.

[32] Nan Chi, Yuanquan Wang, Yiguang Wang, Xingxing Huang, and Xiaoyuan Lu,“Ultra-high-speed single red-green-blue light-emitting diode-based visible lightcommunicationsystemutilizingadvancedmodulationformats.”ChineseOpticsLetters12,(1),2014,010605.

[33] LianeGrobe,Anagnostis Paraskevopoulos, JonasHilt, et al., “High-speed visiblelight communication systems.”CommunicationsMagazine, IEEE51, (12), 2013, 60–66.

[34] Shlomi Arnon, “The effect of clock jitter in visible light communicationapplications.”JournalofLightwaveTechnology30,(21),2012,3434–3439.

[35] EtaiRosencrantz andShlomiArnon, “Tunable electro-optic filter based onmetal-ferroelectricnanocompositeforVLC,”OpticsLetters39,(16),2014,4954–4957.

2 Modulationtechniqueswithlightingconstraints

JaeKyunKwonandSangHyunLee

The physical layer design of visible light communication (VLC) systems is ofsubstantiallydifferentcharacteristicsfromstandardRFcommunicationsinthatitinvolvesanewconstraint,namelyalightingconstraint.Thiskindofconstraint is imposedontheaverage intensityand flickerof the light emission.Since the flickerhas little impactonhumaneyeperceptionwhenlightpulsesblinkat200Hzorhigherfrequency,theaverageintensity constraint ismainly addressed in this chapter.While this constraint is usuallygivenasaninequalityinopticalwirelesscommunication,itisrepresentedasanequalityinVLC.Inaddition,comparedtoradio-frequencycommunication,where thesignalpower,the squared value of signal level, is usually constrained, the intensity, the signal levelitself,isconstrained.Inotherwords,thelightingconstraintisdefinedwithrespecttotheaverage (the first-ordermoment)of thesignal, insteadof thevariance (thesecond-ordermoment).Therefore,thisnewconstraint,whichwillbereferredtoasthedimmingtarget,introducesanewdomainofsystemdesignwhichhas rarelybeenconsidered inexistingcommunicationmedia.

In this chapter, several ways of communicating a message subject to the averageconstraint are addressed. This chapter is far from comprehensive but attempts to offerseveral promising ways of achieving such a goal. To satisfy the lighting constraintrepresentedbytheaverageconstraint,severalapproacheshavebeenaddressed,andtheycan be categorized as to shift signal levels, to compensate in time, and to change leveldistribution. Some of those schemes are simply realized, and some provide improvedthroughput.First,theshiftofthesignallevelisoneofthesimplestapproaches.Atypicalnon-return-to-zero (NRZ) on-off keying (OOK) has 50% average intensity for uniformprobability of binary symbols. For the lighting constraint of 75% dimming, it offers asimple solution of moving the OFF symbol level from 0% intensity to 50%, which isreferred to as analog dimming. Although this is conceptually simple, a non-linearcharacteristic of LEDs poses some technical difficulties, and the reduced level spacingdegradesdetectionperformance.

Second,compensationoftheintensitydifferenceintimeisanotherapproachthatcanbesimply realized.Forgeneral data transmissionwith50%average intensity, i.e., uniformsymbol probability,when the lighting constraint of 75%dimming is targeted, the sameduration ofdummy ON transmission time as the data transmission time is appended tomeet the target. If the target isbelow50%,adummyOFFsymbol interval isapplied toresolve the difference. Those dummy transmissionsmay be either appended after eachsingledataframeorinsertedbetweeneverysymbol.Anexampleoftheformeristhetime-multiplexedOOK[2,3]presented in theIEEEstandard.Ontheotherhand,pulsewidthmodulation(PWM)isatypicalrepresentationofthelatter.Severalstudies[4,5,6]basedonPWMpresent some simple solutions toprovide amarginal rate enhancement.PWMcan also be superposed with OOK and pulse position modulation (PPM) for dimmingsupport [7,8].Variable pulse positionmodulation (VPPM) [2] is another approach thatusesPWM.Thiscombines2-PPMandPWMforadimmingcontrolasshowninFig.2.1.Pulse dual slope modulation [10] is a variant of VPPM and offers improved flicker

mitigation.

Figure2.1 PPMwithadimmingcontrol(after[9]).

Third, change of the distribution of symbol levels can be considered a sophisticatedapproach,whichpromisesadditionalrateenhancement.Inversesourcecoding(ISC)[11,12]convertstheuniformdistributionofONandOFFsymbollevelsto75%ONsymbolsand 25%OFF symbols for achieving a binaryOOKwith 75% target. This can also beextendedtoM-arymodulationandprovidesatheoreticaldatarateboundasymptoticallyinanoise-freeenvironment.Multiple-PPM(MPPM)[3,13,14]isanotherschemethatfallsin this category. This uses all possible combinations ofON andOFF symbolswithin aspecifiedintervaltorepresentdistinctmessages,whiletheratioofONsymbolsandOFFsymbols isadjustedtomeet thedimmingtarget.AlthoughISCandMPPMprovidehighthroughputinalow-noisechannel,bothstillhavesomechallengestoovercometocoexistwith channel coding in a high-noise channel. To this end, several practical schemes[15–18]toaccommodatechannelcodingindimmableVLChavebeenproposed.

TypicalLEDlightingprovideswhiteillumination.However,someapplicationsrequiremulti-colored LEDs, such as light therapy, display, and lighting with a high colorrendering index. In such applications, the lighting requirement does not simply give aconstraint on scalar average intensity but a vector of average color and intensity. Twoapproachesforthecoloredcaseareaddressedhere.Colorshiftkeying(CSK)[2,19,20]separatesconsiderationofcolorandintensity.Theintensityisfixedtothetargetintensity,and thedata transmission iscarriedoutusingan instantaneousvariationof thecolor,asshowninFig.2.2.Thesignalconstellationliesinatwo-dimensionalcolorspacewiththeidenticalintensity.Ontheotherhand,colorintensitymodulation(CIM)[21]changesboththe color and intensity simultaneously. Therefore, this can provide throughputenhancementoverCSK.

Figure2.2 SymbolconstellationofCSKwiththreesymbolsinCIExyYcolorspace(©2012IEEE,reprinted,withpermission,from[21]).

In the following sections, the detail of three schemes is considered: ISC,multi-levelapproaches, and CIM. In addition, another physical consideration should be taken intoaccount in the implementation of the system. VLC requires instantaneous variation oflightingtoavoidhumandetection.Thus,whenthevariationofsymbolsisfastenoughthisensures no flicker. In addition, careful attention should also be given to physical colortemperatureandchromaticityshiftofLEDlighting.ThisresultsmainlyfromthechangeininputcurrentlevelandtemperatureoftheLED,andmulti-levelschemesaresusceptibletothisshift.

2.1 InversesourcecodingindimmableVLC

2.1.1 ISCforNRZ-OOK

The ISC scheme for binary modulation is introduced first. Let d denote the dimmingtarget.Toachievethistarget,theOOKmodulationshouldbeformedusingONandOFFsymbolsintheproportionofdand1−drespectively.Ifthecommunicationiscarriedoutusingthismodulation,thedatarateisupper-boundedbythebinaryentropy,whichisgivenby

Ep=−dlog2d−(1−d)log2(1−d). (2.1)

Thus,toachievethemaximumtransmissionefficiency(datarate)withthedimmingtargetd,thecompositionofmessagesymbolsshouldbeadjustedsothatONandOFFsymbolsoccurwithprobabilitydand1–d, respectively, inasingledataframe.Sincethesourcecoding (compression) operation is used to maximize the entropy by changing thecomposition of symbols as uniformly as possible, the inverse operation of this can beapplied to adjust the composition of symbols to an arbitrary proportion. Thus, thisoperation is referred to as inverse source coding (or dimming coding) and can beincorporated in the transmitter of aVLC system as inFig.2.3. Since the proportion ofinputbinarysymbolsiskepttobeeven,abinaryscramblingoperationmaybeappliedtomaintaintheuniforminputprobabilityinthecasethatthesymbolcompositionoftheinputmessage is non-uniform.The binary scrambling can be realized by taking symbol-wisemodulo-2operationsafteraddinga randombinary sequence toa streamof themessagesymbol.

Figure2.3 ThetransmitterofaVLCsystemwithinversesourcecoding(©2010IEEE,reprinted,withpermission,from[11]).

Figure 2.4 shows the transmission efficiency increase of ISC over existing time-multiplexing-baseddimmingsupport.ThetransmissionefficiencyoftheISC,denotedbyEp, is consistently improvedover theexistingscheme,denotedbyE0, andbothbecome

equalwhend=0,0.5,and1.Also,theincreaseinefficiency isexpressedas

(2.2)

as illustrated in Fig. 2.4. As the dimming target deviates from 0.5, the efficiencyimprovement becomes larger. A 50% improvement in efficiency is induced when thedimmingtargetis29%(or71%)and100%for16%(or84%).

Figure2.4 EfficiencyimprovementwithISC(©2010IEEE,reprinted,withpermission,from[11]).

Here,therealizationofISCisconsideredandexemplifiedusingaHuffmancode.Letthedimming targetdbe0.7.That is, theproportionofONandOFFsymbolsshouldbe70%and 30%, respectively.Thus,Huffman encoding for this condition is first applied.Since the probability of an ON symbol is larger than that of an OFF symbol, the ONsymbol is jointly considered with the consecutive symbols. Thus, the resultingprobabilitiesfornewsymbols“0,”“10,”and“11”aregivenasinTable2.1.TherightmostcolumninTable2.1listscodewordsthatareencodedbytheHuffmancode.Theaveragelengths of uncoded and encoded symbols are 1.7 and 1.51, respectively. Thus, the

compressionratioisgivenas .Sincetheentropycanbeevaluatedas–0.3log2

0.3 – 0.7log2 0.7≈ 0.881, more than 94% of the compression ratio isobtained compared to the maximally achievable compression ratio. Inverse Huffmancodingisusedtoconvertadatastreamwithuniformbinarysymbolstothatwith70%oftheresultingsymbolsbeingON.ThisisshowninTable2.2,whichisrealizedsimplyviathe inverseof themappingpresented inTable2.1.The average lengths of uncoded andinversely-Huffman-encoded symbols are 1.5 and 1.75, respectively. Thus, the

decompressionratiois .Theresultingdimmingrateisgivenby

(2.3)

which is close to the dimming target of d = 0.7. Elaborate Huffman coding and itsassociated inverse Huffman coding using a larger number of symbols can yield animprovedfittothedimmingtarget.

Table 2.1 Huffman encoding (© 2010 IEEE. Reprinted, with permission, from[11]).

Symbol/length Probability Codeword/length

0/1 0.3 00/2

10/2 0.21 01/2

11/2 0.49 1/1

Table2.2 InverseHuffmanencoding(©2010IEEE.Reprinted,withpermission,from[11]).

Symbol/length Probability Codeword/length

00/2 0.25 0/1

01/2 0.25 10/2

1/1 0.5 11/2

Lastly,theconflictbetweenchannelcodingandinversesourcecodingisaddressed.Letan example sequence 00 01 1 01 1 be inversely Huffman-coded. Then, the flow ofencodingoperationsproceedsas:

inputsequence:00011011,inversely-Huffman-encodedsequence:010111011,sequencecorruptedbyerroneouschannel:000111011,sequencealignedforrecovery:000111011,recoveredsequence:0000001011.

Inthisexample,thenumberofsymbolsincreasesby2andthefourthsinglesymbol“1”isdecodedtothree-tuplesymbol“000.”Thus,thechannelcodingthatmayfollowishighlyliketofailtorecovertheoriginalsequence.Thisindicatesthatanordinaryinversesource

code may not seem to work well with the usual channel coding. Therefore, for VLCtransmissionovererroneouschannels,twotopicsforISCremainopen:theexistenceofanISC approach which can work well with channel coding, and the design of a channelcodingschemethatcanproducecodewordsofunequalbinaryprobabilityofONandOFFsymbolsthatadapttothedimmingtarget.

2.1.2 ISCforM-aryPAMInISCwithOOKmodulation,thedimmingtargetstraightforwardlydeterminesthebinarysymbolprobabilitybyadjustingthedutycycleofadataframe,theproportionofONandOFFsymbols.However,non-binarymodulation,suchaspulseamplitudemodulation,mayemployvariousalternativesfordimmingsupport,eachofwhichleadstoadifferentvalueof the spectral efficiency. In this context, the distribution of non-binary symbols thatmaximizes the spectral efficiency is considered. For the spectral efficiency, the entropycanbeconsidered.TheentropyforM-PAMissimplygivenby

(2.4)

wherepiistheprobabilityoftheithlevelofPAM.IntheequidistantM-PAM,thespacingbetweentwoadjacentlevelsisequalandthedimmingtargetdisexpressedas

(2.5)

sincetheithlevellieson ofthemaximumlevelanddisthenormalizedaverageoflevels.Thesymbolprobabilitydistribution{pi}thatmaximizes(2.4)undertheconstraint(2.5) isobtainedviaanoptimization formulation.Since thisoptimization isconcave, itssolutionisguaranteedtoprovidetheglobalmaximum.Toobtainaclosed-formsolutionoftheoptimization,thedualformulation[1]isderivedas

(2.6)

whereλ1andλ2areLagrangemultipliers.Then,bysomealgebra,thedimmingtargetcanbeexpressedas

(2.7)

wherea=1/ln2+λ1and .Thus,afeasiblepairof(λ1,λ2)ischosensuchthatthedistribution{pi} iswell-defined andmaximizes the entropy.To implement theobtained

distribution, the inverse Huffman codemay be used. Figure 2.5 shows the normalizedentropyof3,4,8-PAMISCandtime-multiplexdimming.ISCconsistentlyoutperformsthetime-multiplexdimmingschemeregardlessofthemodulationorder.ForM-PAMISC,thenormalized entropies are all of similar trend regardless ofM. Therefore, ISC remainsefficientforanylargeM.

Figure2.5 Normalizedentropy(©2012IEEE,reprinted,withpermission,from[12]).

2.1.3 ComparisonswithrespecttodimmingcapacityHere,ISC,analogdimmingandtheirhybridapproacharecompared.Theanalogdimmingapproach changes the intensity level of symbols: the intensity level is raised if thedimmingtargetisgreaterthan0.5,andreducedifitissmallerthan0.5.Theintensitylieswithintheinterval[0,A]andthesymbolspacingisequidistant.Thelargestshiftinsymbolintensity is 2A(d − 0.5). Hybrid dimming is achieved for an intensity shift in analogdimmingsmallerthan2A|d−0.5|andISCisusedtoadjusttheremainingshifttowardsthedimmingtarget.Theentropyofanalogdimmingremainsconstantforthedimmingtarget.Sincetheentropycurveof3-PAMhybriddimmingresultsfromthehorizontalscalingof3-PAM ISC, 3-PAM hybrid dimming has the same maximum entropy as that of 50%dimming for an arbitrarydimming target.Note that otherPAMs such as 3-PAMand6-PAMareused in this chapter for explanatorypurposes, though2n-PAMsare practicallyused.Since95%dimmingisexpectedtoyieldalowerdataratethan50%dimminginanoisychannel, special care is required for considerationof theperformancedegradationcaused by noise in comparisonwith analog or hybrid dimming.Theminimumdistancebetween symbols is considered here. If theminimum distance is identical, the order ofperformancedegradationcanbeassumedtobeequal.Thus, theentropyof thedimmingmethods can be compared when the minimum distance between symbols is identical.

Figure2.6depictstheentropyof4-PAMISCand3-PAMhybriddimmingwiththesameminimumdistance.4-PAMISCisconsistentlysuperiorto3-PAMhybriddimming.Thisisstraightforward because the symbol set of 3-PAM hybrid dimming {S2,S3,S4} can beconsideredaspecialcasewheretheprobabilityofS1iszerointhesymbolsetof4-PAMISC,{S1,S2,S3,S4}. Therefore,M-PAM ISC is better than or equal toN-PAM hybriddimming in performancewhenM >N with the sameminimum distance.However, thecomparison of ISC, analog dimming, and hybrid dimming is not straightforward fordifferent minimum distances. Thus, the dimming capacity is introduced to comparebetweendimmingmethodsfordifferentminimumdistances.

Figure2.6 EntropyofISCandhybriddimming(©2012IEEE,reprinted,withpermission,from[12]).

DimmingstrategiesarecomparedunderadditivewhiteGaussiannoise(AWGN):

Y=X+Z, (2.8)

whereXisatransmittedsignal,Yisareceivedsignal,andZisAWGNwithzeromeanandvarianceσ2. The dimming capacity, denoted byCd≡ I(X;Y), is defined as the mutual

informationbetweenXandYsubjecttothedimmingconstraint .Then,I(X;Y) isgivenby

(2.9)

wheref.(.)istheprobabilitydistributionfunction,sothat

(2.10)

(2.11)

where

(2.12)

andDs is thelevelshiftforanalogandhybriddimming.Thus, theintensityrange[0,A]canchangeintoeither[0,A–Ds]or[Ds,A]accordingtothedimmingtarget.

Figure2.7showsthedimmingcapacityofISCby2,3,4,8-PAMfordimmingtarget0.5.ThebestmodulationorderthatachievesthemaximumvalueofthecapacitychangeswithrespecttochannelqualitymeasureA/σ.Sincetheaveragepowercannotbeadjustedfreelyunder agivendimming target, analogdimmingcan reduce the intensity rangeof [0,A]only.Thus,thedecreaseindimmingcapacitybyanalogdimmingisdirectlyindicatedinFig.2.7byaleftwardshiftalongthehorizontalaxis,A/σ.Forexample,3-PAMisusedanddimmingchangesfrom0.5to0.8.Theintensityrangechangesfrom[0,A]to[0.6A,A]inordertosatisfy0.8analogdimming,wheretherangeislimitedtoonly40%oftheoriginalrange. This results in a horizontal shift of −3.98 dB≃ −4.0 dB. If A/σ is 9 dB, thisbecomesequivalently5dBforanalogdimming.Then,thedimmingcapacityof0.8analogdimmingcorrespondsto0.69,decreasedfrom1.47inthetwofilledsquaresshowninFig.2.7.

Figure2.7 ThedimmingcapacityofM-PAMwith0.5dimming(©2012IEEE,reprinted,withpermission,from[12]).

Here,theperformancecomparisonamongISC,analogdimming,andhybriddimmingisdiscussed.Themodulation level is restricted to2,3,4,8,and16-PAM.Figure2.8depictsthedimmingcapacitywith respect toA/σ anddimming target. Figure2.9 illustrates thedimmingmethodthatyieldsthemaximumcapacity.Figure2.9exhibitsthebestdimmingmethodwithrespecttoA/σandthedimmingtarget.ISCandanalogdimmingarechosenwhen the normalized DC shift (intensity shift) is zero and one, respectively. Hybriddimmingcanbeappliedformid-rangevalues.ISChasbetterperformanceexceptwhenA/σisgreaterthan20dBandwhenA/σisaround11.5dBwiththedimmingtarget97%.IfhighermodulationsareallowedforaregionA/σgreaterthan20dB,ISCstillremainsthebest dimming method. Another exception occurs around an A/σ of 11.5 dB, wheremodulations between 4-PAM and 8-PAM are not allowed. Figure 2.10 shows how todeterminethey-axiscoordinateinFig.2.9whenA/σis11.5dBandthedimmingtargetis97%. The maximum dimming capacity occurs for 6-PAM and ISC. If 6-PAM is notallowed, 4-PAMwith slightDC shift (hybrid dimming) yields themaximum.However,thedifferenceofthecapacityisnegligiblebetweenISCandhybriddimmingin4-PAMinthiscase.Therefore,ISCisthebestmethodwhenallmodulationlevelsareavailable,andISC is of performance better than or similar to the others when only a subset ofmodulationlevelsisallowed.

Figure2.8 Dimmingcapacityof2,3,4,8,and16-PAM(©2012IEEE,reprinted,withpermission,from[12]).

Figure2.9 Dimmingmethodselection(©2012IEEE,reprinted,withpermission,from[12]).

Figure2.10 DimmingmethodselectionwhenA/σis11.5dBandthedimmingtargetis97%(©2012IEEE,reprinted,withpermission,from[12]).

2.2 Multi-leveltransmissionindimmableVLCIn this section, a multi-level transmission scheme is presented for visible lightcommunicationsystemsofdimmingsupport.Toprovideadimmingcontrolofthemulti-level modulation, the concatenation of different pulse-amplitude-modulated symbols isused to yield the overall signal with an average amplitude that matches the dimmingrequirement. This scheme also aims at arbitrary dimming adaptation by adjusting thecompositionofconcatenateddifferently-modulatedsymbols.To thisend, theproblem isformulated into linear programming with the objective of maximizing the transmissiondatarateandsatisfyingthedimmingrequirements.

The need for the improved spectral efficiency leads to the use of multi-levelmodulations, such as PAM, and poses a new challenge for themulti-level transmissionschemedesignthatadaptstothedimmingrequirement.

Tothisend,thissectionpresentsapracticalmulti-leveltransmissionapproachthatusesaconcatenatedcodewhereeachofthecomponentcodesisencodedandconstructedwithdifferent modulations. To obtain high error-correcting capability with a simple codestructure, agroupof linear codes isusedandconcatenated.However, a linear codecanonly generate a set of codewords with a uniform number of symbols. The average

intensitylevelofalinearcodewordmodulatedusingPAMofthepeaklevelintensityAis

always equal to , which corresponds to the dimming of 0.5. In other words, thestraightforwardconcatenationoflinearcomponentcodesdoesnotallowfortheadaptationto non-trivial dimming requirements. Therefore, the design of an efficient transmissionschemesatisfyingthedesireddimmingrequirementispursuedbyadjustingtheproportionofsymbolslinearlyencodedandmodulatedwithdifferentPAMsandconcatenatingthosesymbols.Sincetheproportionof linearcodesisadjustedarbitrarily, theadaptationtoanarbitrary dimming requirement can be accomplished. In what follows, a linearoptimizationformulationthatdeterminestheoptimalcompositionoflinearcodedsymbolsforthehighestspectralefficiencyisintroducedandsolved.

2.2.1 Multi-leveltransmissionschemeAmodelmulti-level transmissionscheme is first introducedanda linearoptimization isformulatedforobtainingthebestconfigurationofthescheme.Forthisscheme,NmessagesymbolsthatconstituteasingledataframeareeithermodulatedinoneofM−1differentPAMs (2 toM-PAM) or punctured. The spacings between levels are uniform for allmodulations.Thatis,thedifferencebetweenanytwoadjacentlevelsofaPAMisidentical.Without loss of generality, the dimming targetd iswithin [0,0.5], i.e.,d∈ [0,0.5].Anexampleofsuchamulti-leveltransmissionschemeisdepictedinFig.2.11.Thehorizontalandverticalaxesareassociatedwiththesymbolconfiguration(orthesymbolorder)andthe intensity of the transmitted signal, respectively. Message symbols modulated indifferentmodulationsareseriallyconcatenated.Thus,amessagesymbolmodulatedinoneoftheMPAMsistransmittedateachsymboltimeinstant.However,themessagesymbolsarenotnecessarily transmitted in thisorder,because the transmissionof the symbols inthiswaycausesaperiodicgradualdecreaseoftheintensityinadataframe,whichresultsin a severe flickering effect. Therefore, the symbols are interleaved in a random wayknown a priori to transmitter and receiver so that the order of the intensity levels israndomizedandtheflickerisaveragedout.AlevelofonePAMisalignedwiththesamelevelofdifferentPAMs.Forexample,theithlevelofj-PAMisofthesameintensityastheithlevelofk-PAMfori≤j<k.TheunitleveloftheverticalaxisinFig.2.11isuniform.Thus, the transmission scheme can be thought of as anM-ary PAMwith non-uniformsymbol level probability or alternatively as the superposition ofM − 1 different OOKmodulations.Thevaluesof effectivepoweraredifferent for themessagesmodulated indifferentPAMssincethemeansymbolpowerincreaseswithincreasingnumberoflevelsofPAM. If themessage is assumed to be randomandmodulated in (i + 1)-PAM, eachlevelinaPAMsymboloccurswithuniformprobability1/(i+1)andtheaveragesymbollevel is equal to i/2 times the intensity of a single level.By changing theproportionofsymbolsmodulatedinMdifferentPAMs,thedimmingratecanbeadjustedtoanarbitrarytarget.Theproportionsofleveli=0,…,M−1andthatofeachlevelin(i+1)-alphabetsymbols are denoted bypi andqi, respectively. It is defined thatqM−1 =pM−1. Then, itfollowsthat

(2.13)

Notethat andthesumof{pi}isequaltoone.Also,itisobviousthat

(2.14)

That is, the (i + 1)-PAM has the proportion of (i + 1)qi, which corresponds to theproportionof the ith column inFig.2.11.A symbolmodulated in (i + 1)-PAMhas theamount of information proportional to log2(i + 1) bits per symbol. Since the dimmingtargetisassociatedwiththeaverageintensity,distribution{pi}canbechosentomeetthedimming target. Therefore, the dimming target is expressedwith respect to distribution{pi}as

(2.15)

Whend∈(0.5,1],adistribution canberedefinedas bysymmetry.Forthis purpose, the proportion for (i + 1)-PAM qi is found such that the overall spectralefficiency ismaximized.Since the(i+1)-PAMsymbolcanconvey log2(i+1)bits, thespectralefficiency(orentropy)ischosentobeanobjectivefunctionoftheoptimization,whichisgivenby

(2.16)

Sinceallexpressionsassociatedwith{pi}and{qi}arelinear,theoptimizationisgivenina linear convex optimization formulation, i.e., linear programming. Bymerging (2.13),(2.14), and (2.15), the resulting linear optimization, with the objective of maximizing(2.16),isgivenby

subjectto

(2.17)

To obtain a closed-form solution, the dual formulation [1] is derived because only two(dual)variablesaresufficientandthesolutioniseasiertoobtainthantheoriginal(primal)formulation.Therefore,(2.17)isreformulatedas:

subjectto

λ+iμ≥log2(i+1),i=0,…,M–1. (2.18)

To see this, theLagrange function [1] of (2.17) can be considered. Since there are twoconstraints, two Lagrangemultipliers denoted by λ and μ, respectively, are introduced.Since the Lagrange function provides the original formulation with a concave upperbound, the dual formulation turns out to be theminimization of the bound in order tomatchtheboundtothemaximum.Forthisgoal,theLagrangefunctioncanbeexpandedwith respect toM variables .Then, the fact that the coefficients of thoseM variablesbecomenegative in theminimization leads to thecorrespondingMdifferentconstraints,and the remaining terms that are independent of the variables provide the objectivefunctionof (2.18).Theobjective functiondependson symbol alphabetM anddimmingtargetd.Also,eachconstraintisassociatedwitheachofMPAMs.Sincebothformulationsaresimplelinearprogramming,thestrongdualityholdsbySlater’scondition[1].Inotherwords,theoptimalsolutionsofbothformulationsareidentical.Thedualformulationcanalso be interpreted in a geometric way as illustrated in Fig. 2.12. The M differentconstraints generate a convex feasible region over the (λ,μ) plane. In addition, theobjectivefunctionisassociatedwithalineacrossthefeasibleregionwithitsslopebeing1/(2(M–1)d)anditsx-interceptequaltotheobjectivevalue.Therefore,thedualproblemturns out to determine a line which touches a point of the feasible region and has thesmallestx-intercept.Accordingtotheconstraintsin(2.18),thereareMdifferent linesofdecreasingslopesthatdefinetheboundaryofthefeasibleregionintheplane.Therefore,M –1 intersecting points can occur on the boundary of the feasible region. Since thefeasibleregionisconvex,thelineassociatedwiththeobjectivefunctionintersectsonlyasinglepointofthefeasibleregion.Byinspection,thesolutionoccursatapointwhereonlytwoconstraintsin(2.18)holdtheequality.Therefore,considerationoftheintersectionofthose points and the line associated with the objective function suffices. By thecomplementary slackness, all variables associated with strict inequalities have zero

values,i.e., forisuchthatλ+iμ>log2(i+1).Atmost,twoprimalvariables cantake non-zero values, and only two (consecutive) PAMs are needed for the optimaltransmission.Here,theeffectivemodulationparameterisdefinedasThen,thesolutionisexpressedas

(2.19)

Thecorrespondingobjectivevalueisgivenby

(2.20)

Inaddition,theassociateddistributionisobtainedas

(2.21)

Thisimpliesthatonlytwomodulationsof -PAMand -PAMareusedintheoptimal transmission.Thiscanbeconsidered inageneral settingwhereanarbitrarysubsetofPAMalphabetsisallowedinsteadofallalphabets(1toM-PAM).Sinceasingleconstraintin(2.18) isassociatedwithadistinctmodulation, thedualformulationanditsgeometricinterpretationareintactforanychangeinPAMs.Sincethenumberofmessagesymbols is practically chosen to be power-of-twos, 2k-PAMs are preferred for positiveintegerk.Forexample,only1,2,4,and8-PAMareused.Then,thecorrespondingfeasibleregion becomes a polygon with at most four vertices. The intersections and thecorrespondingdata ratesare listedalongwithvalid rangesof thedimmingrate inTable2.3. The upper triangular part (i.e. above the table diagonal) shows the intersectionsassociatedwithtwoPAMscorrespondingtorowandcolumnindices.Thelowertriangularpartrepresentsthecorrespondingobjectivevalue.Thelastrowandcolumncorrespondtothelowerandupperlimitsofachievabledimmingtargets.Forexample,inthecaseofonly2-PAM and 4-PAM used, the line associated with the objective function crosses point

onthe(λ,μ)plane.Thecorrespondingobjectivevalueis for

Figure2.11 Multi-leveltransmissionscheme(©2013IEEE,reprinted,withpermission,from[15]).

Figure2.12 Geometricinterpretationofthedualproblem(©2013IEEE,reprinted,withpermission,from[15]).

Table2.3 Solutionfor thesystemconsistingof1,2,4,and8-PAM(©2013IEEE,reprinted,withpermission,from[15]).

(PAM,PAM) 1 2 4 8 B

1 (0,1) 0

2 14d

4

8 6d

A 0

To ensure that all levels of the (i + 1)-PAM symbol occur uniformly, (i + 1)-aryscramblingcanbeappliedbytakingsymbol-wisemodulo-(i+1)operationsafteradditionofan(i+1)-aryrandomsequence.Forexample,4-PAMisusedandtransmittedsymbolsareassumedtobe01122231.Thus,thesymbollevelprobabilityisnon-uniform.Inaddition,thescramblingsequenceisgivenas32103210.Thesymbolsobtainedafterscramblingaregivenas33225441mod4=33221001.Therefore,theresultingtransmitted symbols are uniform. To retrieve the original symbols at the receiver, thescramblingsequenceissubtractedfromthereceivedsymbolsandtherecoveredsymbolsaregivenas0112–2–2–11mod4=01122231.LetNibethenumberofsymbolsmodulatedin(i+1)-PAMinadataframe.Thatis,Niistheclosestintegerto .Then,theoveralldimmingrateisconsideredasarandomvariableDsuchthatE[D]=dandisgivenby

(2.22)

where Uij is a discrete random variable uniformly distributed over interval [0, i],representingthejthsymbolmodulatedin(i+1)-PAM.Also,A is the largest levelofM-

PAM and the spacing between two adjacent levels is equal to . Therefore, thevarianceofDdenotedbyvar[D]isgivenby

(2.23)

Sinceonly -PAMand -PAMareusedfortheoptimaltransmission,thevarianceissimplygivenby

(2.24)

wheretheequalityholdswhend=0.5.Sincethelowestopticalclockrateis200kHz[22],thevariationcanbeboundedwithinapproximately1%bytakingthedataframelengthNgreaterthan1000forM=8.Therefore,flickerdoesnotoccurbecausethebrightnessof

visiblelightchangesataratemuchfasterthan150–200Hz[2].

2.2.2 AsymptoticperformanceThe performance of the multi-level transmission scheme is addressed in variousconfigurations, such as all consecutive alphabets and power-of-two alphabets. For faircomparison,thenormalizedcapacityisevaluatedbydividingtheaveragenumberofbitspersymbolbylog2(i+1)for(i+1)-PAM,asillustratedinFig.2.13.Thiscorrespondstothe relativedata rate introduced inorder to comparevarious transmission schemeswithdifferent data rates. ML-(i + 1)PAM denotes the multi-level transmission scheme thatallowsconsecutive(i+1)differentPAMs(1to(i+1)-PAM),andrML-4PAM(1,4)denotesatransmissionschemewithonly1-PAMand4-PAMused.AlthoughISC[12]establishesatheoreticalupperbound,apracticalcapacity-approachingschemeachievingthisboundhasnotbeenfound.ThespectralefficiencyofM-PAMISCincreasesslowlyasthenumberofalphabetsMincreases.Theperformanceoftheschemeapproachestheupperboundinapiecewise linearwayfor the increasingnumberofalphabets.Thus,agoodupper-boundforML-MPAMinFig.2.13isobtained.Thekthpointwheretheslopechangesislocated

at ontheML-MPAMcurve.AsthenumberofalphabetMincreases,the number of such points increases correspondingly, and the set of line segmentsconnectingadjacentpointsprovidesagoodapproximationof thecurve forML-MPAM.

Then,allpointsin liealongthecurverepresented

by .Notethatforx∈(0,0.5],itbecomesthecasethatfM(x)=1asMtendstoinfinity.

Figure2.13 Normalizedspectralefficiency(capacity)(©2013IEEE,reprinted,withpermission,from[15]).

TheperformanceimprovementoverrML-(i+1)PAM(1,i+1)isdepictedinFig.2.14.The improvementdecreasesgradually as thedimming rate approaches0.5.TheoptimalcompositionofPAMsisshownforvariousdimmingratesinFig.2.15.ThisisconsistentwiththesolutionoftheoptimizationwhereatmosttwoPAMsarechosenfortheoptimaltransmission.SinceeightdifferentPAMsareallowed(1to8-PAM),twolinesassociated

with k-PAM and (k + 1)-PAM cross in interval for k = 1,…,7, i.e., at sevendifferentpoints.Comparedtothis,threecrossesoccurinrML-8PAM(1,2,4,8).Therefore,this figure is used to determine the optimal transmission scheme graphically. Once thetarget dimming rate is known a priori, both the transmitter and receiver can find theoptimalcompositionofthemodulationsimmediately.

Figure2.14 Capacityimprovement(©2013IEEE,reprinted,withpermission,from[15]).

Figure2.15 Levelcomposition(©2013IEEE,reprinted,withpermission,from[15]).

2.2.3 SimulationresultsSimulationresultsofuncodedandcodedtransmissionschemesareshownhere.LetAandMbethelargestlevelandtheorderofthelargestorderedPAM,respectively,i.e.,Aistheintensity of theMth level ofM-PAM. Under the assumption of uniform PAM symbol

level, the kth level of (i + 1)-PAM is expressed as . Then, the symbol errorprobabilityoftheuncoded(i+1)-PAMisgivenby

(2.25)

where ,andσ2isGaussiannoisepower.Forachannelqualitymeasure,the signal-intensity-to-noise-amplitude ratio is used. In opticalwireless communication,theLEDmodulates the instantaneousoptical intensityproportional toan inputelectricalcurrentsignalviaintensitymodulationanddirectdetection[23].Thecombinationofthesetwo conversion steps ensures the validity of (2.25) in VLC. The overall symbol errorprobabilityisgivenby

(2.26)

whereℳ is the set of the largest levels (in [0,M− 1]) of all PAMs allowed in a dataframe, i.e., Since only two adjacent PAMs are usedfortheoptimalschemein(2.21),theerrorprobabilityisgivenby

(2.27)

The decoding error performance depends on two parameters and , and hence ondimming rate d and modulation order M. The equality holds when

and,whend=0,noinformationistransmittedand the errorboundobviouslybecomeszero.Theerrorbound ismaximized ifonlyM-PAMisusedandd=0.5.Thisisobviousbecausethesymbolerrorprobabilityinherentlygrows as thenumberof transmitted symbols increases.Given that for anon-zeroε,spectralefficiencyR(i+1)decreasestozeroasfastas .Therefore,itisnecessarytouseagoodchannelcodeforeachPAMtoobtainhighspectralefficiency.Theresultingspectralefficiencyisexpressedas

(2.28)

Since and arefunctionsofthedimmingtarget,soisthespectralefficiency.

Theperformanceofcodedschemesisnowconsidered.Symbolerrorprobabilityfori∈ℳisobtainedonlybysimulation.IftheinformationisencodedwithcoderateRandmodulatedintoNPAMsymbols,thetwoexponentsof(2.28)arereplacedwithand , respectively. Turbo codes are used for the evaluation of the codedperformance,inthattheuseofpracticalcodingschemesguaranteesthefeasibilityofthetransmission scheme, and turbo codes lead to the capacity-approaching performance. Inaddition, turbo codes are robust against the performance degradation incurred bypuncturing foradaptinganarbitrarydimming target.Figures2.16and2.17 compare thespectral efficiencies for dimming requirement d = 0.1 and 0.4, respectively. Several

different codes of rates can be obtained from a single code of rate bypuncturing. For comparison, the results of other transmission schemes that meet thedimming requirement, using 8-PAM or -PAM are presented. For the uncodedscheme,theresultofarandomschemeispresentedbycalculatingtheaveragewithrespectto theensembleof feasibleprofiles .Fordifferentdimming targets, themulti-leveltransmissionschemeconsistentlyshowssuperiorthroughputperformancetoothers,anditselectsdifferentpairsofPAMsaccordingly,e.g.,(2,3)-PAMford=0.1,(3,4)-PAMford=0.2, (5,6)-PAM for d = 0.3, and (6,7)-PAM for d = 0.4. Between two cases that use

puncturing, the scheme using -PAMoutperforms that using themaximal orderPAM (8-PAM). The performance difference between them is large in a low dimmingregime,becauselargernumbersofsymbolsarepuncturedwithsmallervaluesofdfor8-PAMandtheoverallamountofinformationislowerthanfor

Figure2.16 Spectralefficiencyfordimmingtarget0.1(©2013IEEE,reprinted,withpermission,from[15]).

Figure2.17 Spectralefficiencyfordimmingtarget0.4(©2013IEEE,reprinted,withpermission,from[15]).

2.3 Colorintensitymodulationformulti-coloredVLC

2.3.1 ColorspaceandsignalspaceNow the characteristics of the color space are introduced and the difference from thesignalspace isconsidered.To thisend, themodelof themulti-coloredsystemisbrieflydescribed. N different LEDs and corresponding photodetectors (PDs) with differentwavelengthcharacteristicsareused.LetIi(λ)andrj(λ)denotetheintensityoftheithLEDandtheresponsivityofthejthPD,respectively.Then,theoverallintensityatthereceiversisgivenby .Let , ,and bethenormalizedcolormatchingfunctions[24]associatedwiththecolorperceptioncapabilityofhumaneyes.Then,three

components of light stimuli are given by , , and

.TheseparameterscancharacterizetheperceptioninhumaneyesindicatedinCIEXYZcolorspace[24].Asinglecolorisassociatedwitheachlinepassingtheoriginin CIE XYZ color space, as shown in Fig. 2.18, and its corresponding intensity isrepresented by the distance of a point on the line from the origin. To provide a VLCfeatureunderthesecolormatchinganddimmingconstraints,colorshiftkeying(CSK)[22]canbeused.CSKplacesthesymbolconstellationaroundthetargetcolorinCIExyYcolor

space such that the average of colors associatedwith symbols is the same as the targetcolor.At the same time, the intensityofLEDoutput is controlled tomeet thedimmingrequirement.Figure2.2showsanexampleofCSKachievedusingthreeLEDs,denotedbyLED1,LED2,andLED3,respectively.Messagesymbolscanbeplacedwithinatrianglewith three vertices associated with those three LEDs such that the overall averagecoincideswiththepointcorrespondingtothetargetcolor.Indoingso,messagesymbolsshouldbelocatedasfaraspossiblefromoneanothertominimizethedetectionerror.Thedetection of the signal is carried out in the signal space. Thus, the received signal isrepresentedwithanN-dimensionalvectorS= [S1,…,SN]T,where the ithcomponent, the

outputoftheithphotodetector,is .Lightingconditionsofcolormatchinganddimmingareconsidered in thecolor space,while the spectral efficiency (ormutualinformation) of the communication feature can be addressed in the signal space.Therefore,itisnotstraightforwardtoconsiderbothobjectivessimultaneouslyinasinglespace.

Figure2.18 CIEXYZcolorspace.

2.3.2 ColorintensitymodulationTomitigatethisdifficultyandimprovethespectralefficiencyunderthegivenconstraint,

colorintensitymodulation(CIM)isproposed.Tothisend,asubspace(orapoint)inthesignal space is chosenwhich is associatedwith the subspace of the color spacewherecolorconstraintissatisfied.Forthissubspaceofthesignalspace,thepositionsofsymbolsare determined to maximize the spectral efficiency subject to the constraint that thesymbols’ weighted average belongs to the subspace. For the control of the averageintensityofsymbols,PAMisusedheresinceithasthesamebandwidthregardlessoftheaverage intensity.When the average intensity is achieved usingM-PAM ISC [12], twodifferent approaches can be used: either adjusting symbol probabilities for fixedequidistantsymbolsorcontrollingpositionsandprobabilitiessimultaneously.InFigs.2.19and2.20,thosetwoapproachesareillustratedwithA/σ8dBandthedimmingtarget0.8.Thesecondapproachoffers1.3%improvedperformancewiththesimultaneouscontrolofpositionsandprobabilitiesofmessagesymbols.

Figure2.19 Optimalequidistantsymbolswithmutualinformation0.9373bits/symbolwhenA/σis8dB,andthedimmingtargetis80%(©2012IEEE,reprinted,withpermission,from[21]).

Figure2.20 Optimaladjustedsymbolswithmutualinformation0.9494bits/symbolwhenA/σis8dB,andthedimmingtargetis80%(©2012IEEE,reprinted,withpermission,from[21]).

Nowmulti-dimensionalchannelisconsideredwithmulti-colorLEDs.Forindependentandparallelcontrolofopticalchannels,wavelengthdivisionmultiplexing(WDM)canbeused.ISCofferseachopticalchannelofWDMwithcolormatchinganddimmingcontrol.On theother hand,CIMdoesnot suffer from inter-channel interference, evenwhen thechannels are non-orthogonal. CIM yields improved spectral efficiency over the normalWDM andWDMwith ISC in non-orthogonal channels. To summarize the concepts ofCSK,WDM,andCIM,first,CSKusesonlyatwo-dimensionalregion(apartofaplane)with a specific intensity in the three-dimensions, Fig. 2.18. The plane has a positiveintercept on each axis. On the other hand, WDM and CIM adopt the entire three-dimensionalregion.TheirdifferenceisthatWDMseparatelyuseseachchannelofthree-dimensions and CIM exploits the entire three-dimensional region at once. If the three-dimensional channels are orthogonal to each other and can be separated withoutinterference, WDM supported by ISC yields the same performance as that of CIM,otherwiseCIM is better thanWDM.For analysis of the spectral efficiency, an additiveGaussiannoisechannel isconsideredwith fullyorthogonalopticalchannelswhereeach

receivercandistinguishitscorrespondingsignal,i.e., .Then,thereceivedsignal vectorV = [V1,…,VN]T can be denoted byV =S +W, whereW is an additiveGaussiannoisevector.Bytheconstraintsoflighting,theweightedaverageofSbelongstoa subspace of signal space, which is reduced to a point whenN = 3, with three colorchannels.

Figure 2.21 shows one-dimensional mutual information with respect to A/σ and

dimmingconstraints.Whentheoptimalpointischoseninthesubspace,thesumofI(Si;Vi)istheupper-boundforthecapacityofCIM.Inaddition,atwo-dimensionalexampleisdepicted inFig.2.22forA/σ 8 dBand6dB, anddimming targets 0.8 and0.5 for eachdimension, respectively. Two axes represent the signal communicated betweencorrespondingtransmitterandreceiver.Then,themutualinformationis0.9494+0.9385=1.8879bits/symbol,andthenumberofsymbolsis4×3=12.Thus,theupper-boundforthe capacity is 1.8879. A three-dimensional extension is shown in Fig. 2.23 with thelightingconstraintofA/σ5dB,dimmingtarget0.3fortheadditionaldimension.Thus,themutualinformationisequalto0.9494+0.9385+0.6945=2.5824bits/symbol.

Figure2.21 Mutualinformationofasinglecolorchannel(after[21]).

Figure2.22 CIMsymbolconstellationintwo-dimensionalsinglespacefortheorthogonalchannels:(A/σ,dimming)=(8dB,80%)inR1axis,and(A/σ,dimming)=(6dB,50%)inR2axis(after[21]).

Figure2.23 CIMsymbolconstellationinthree-dimensionalsignalspacefortheorthogonalchannels:(A/σ,dimming)=(8dB,80%)inR1axis,(A/σ,dimming)=(6dB,50%)inR2axis,and(A/σ,dimming)=(5dB,30%)inR3axis(after[21]).

Table 2.4 compares the mutual information of the CIM and CSK. The channelcondition is identical to thecase shown inFig.2.23with the exceptionof thedimming

target 0.2 for the first channelR1 instead of 0.8. The results for CSK1 and CSK2 areobtained to maximize the minimum distance between three symbols without and withconsideration of channel gains, respectively. The following three CIM schemes placesymbols ina three-dimensionalspaceindifferentways.CIM1placeseightequiprobablesymbols centered on the target point so that the set of symbols forms a rectangularparallelepiped.CIM2 locateseight symbolswithdifferentprobabilitiesat thecornersofthe rectangular parallelepiped as in Fig.2.23. CIM3 denotes the scheme shown in Fig.2.23.Thus,itssignalconstellationcorrespondstoamirrorimageofFig.2.23alongtheR1axis.

Table 2.4 Performance comparison of CIM andCSK (© 2012 IEEE, reprinted,withpermission,from[21]).

Modulationscheme Mutualinformation(bits/symbol)

CSK1-general 1.4768

CSK2-optimal 1.5043

CIM1-analogdimming 2.0070

CIM2-binary 2.3247

CIM3-optimal 2.5824

Finally, a non-orthogonal multi-colored channel is considered. Since the signalsreceived at the receivers are not independent, the resulting signal subspace forms aparallelogram and a parallelepiped in the two-dimensional and three-dimensional signalspace,respectively.Figure2.24illustratesthecasewherethefirstreceivermayrespondtothe second transmitter.Thus, the receiver signal at the first receiver is formedas

.Ontheotherhand,thesecondreceivercandistinguishthesignalfromthe second transmitter.Since2dBdifferenceoccursbetweenR1 andR2, themaximallyachieveddataratefrom isgivenas1+10−0.2≈1.63insteadof2.Thecorrespondingmutual information is 1.9258 bits/symbol and no geometrically meaningful pattern isfoundamongsymbols.Figure2.25showsthecasewherethesecondreceiverrespondstothe first transmitter with only a half responsivity, i.e., . Then, thecorrespondingmutualinformationisequalto2.0458bits/symbol.

Figure2.24 Two-dimensionalnon-orthogonalsignalspacewithR′1=R1+R2andR2:(A/σ,dimming)=(8dB,80%)inR1axisand(A/σ,dimming)=(6dB,50%)inR2axis(after[21]).

Figure2.25 Two-dimensionalnon-orthogonalsignalspacewithR′1=R1+R2andR′2=R2+0.5R1:(A/σ,dimming)=(8dB,80%)inR1axisand(A/σ,dimming)=(6dB,50%)inR2axis(after[21]).

References[1] S. Boyd and L.Vandenberghe,ConvexOptimization, Cambridge University Press,2004.

[2] S. Rajagopal, R. D. Roberts, and S.-K. Lim, “IEEE 802.15.7 visible lightcommunication:Modulationschemesanddimmingsupport,”IEEECommun.Mag.,50,(3),72–82,2012.

[3] K.Lee andH. Park, “Modulations for visible light communicationswith dimmingcontrol,”IEEEPhoton.Technol.Lett.,23,(16),1136–1138,15Aug.2011.

[4] G. Ntogari, T. Kamalakis, J. W. Walewski, and T. Sphicopoulos, “Combiningillumination dimming based on pulse-width modulation with visible-lightcommunications based on discretemultitone,” IEEE/OSA J.Opt. Commun. Netw., 3,(1),56–65,2011.

[5] W.O.Popoola,E.Poves,andH.Haas,“Errorperformanceofgeneralisedspaceshiftkeyingforindoorvisiblelightcommunications,”IEEETrans.Commun.,61,(5),1968–1976,2013.

[6] Z.Wang,W.-D.Zhong,C.Yu,etal., “Performance of dimming control scheme invisible light communication system,”Opt. Express, 20, (17), 18861–18868, 13 Aug.2012.

[7] E. Cho, J.-H. Choi, C. Park, et al., “NRZ-OOK signaling with LED dimming forvisiblelightcommunicationlink,”inProc.16thEuropeanConferenceonNetworksandOpticalCommunications(NOC),Newcastle-Upon-Tyne,UK,July2011,pp.32–35.

[8] H.-J.Jang,J.-H.Choi,Z.Ghassemlooy,andC.G.Lee,“PWM-basedPPMformatfordimming control in visible light communication system,” in Proc. 8th InternationalSymposium on Communication Systems, Networks & Digital Signal Processing(CSNDSP),Poznan,Poland,July2012.

[9] S. Arnon, et al., Advanced Optical Wireless Communication Systems, CambridgeUniversityPress,2012.

[10] M. Anand and P. Mishra, “A novel modulation scheme for visible lightcommunication,” in Proc. 2010 Annual IEEE India Conference INDICON, Kolkata,India,Dec.2010.

[11] J.K.Kwon, “Inverse source coding for dimming in visible light communicationsusingNRZ-OOKonreliablelinks,”IEEEPhoton.Technol.Lett.,22,(19),1455–1457,1Oct.2010.

[12] K.-I.AhnandJ.K.Kwon,“CapacityanalysisofM-PAMinversesourcecodinginvisiblelightcommunications,”IEEE/OSAJ.Lightw.Technol.,30,(10),1399–1404,15May2012.

[13] A. B. Siddique andM. Tahir, “Joint brightness control and data transmission forvisible lightcommunicationsystemsbasedonwhiteLEDs,” inProc. IEEEConsumerCommunications and Networking Conference, Las Vegas, NV, Jan. 2011, pp. 1026–1030.

[14] J.Kim,K.Lee,andH.Park,“Powerefficientvisiblelightcommunicationsystemsunderdimmingconstraint,” inProc.23rdIEEEInternationalSymposiumonPersonalIndoor andMobile Radio Communications (PIMRC), Sydney, Australia, Sept. 2012,pp.1968–1973.

[15] S.H.Lee,K.-I.Ahn,andJ.K.Kwon,“Multileveltransmissionindimmablevisiblelightcommunicationsystems,”IEEE/OSAJ.Lightw.Technol.,31,(20),3267–3276,15Oct.2013.

[16] S.H.LeeandJ.K.Kwon,“Turbocode-basederrorcorrectionschemefordimmablevisible light communication systems,” IEEE Photon. Technol.Lett., 24, (17), 1463–1465,1Sept.2012.

[17] S. Kim and S.-Y. Jung, “Novel FEC coding scheme for dimmable visible lightcommunication based on the modified Reed–Muller codes,” IEEE Photon. Technol.Lett.,23,(20),1514–1516,15Oct.2011.

[18] S.KimandS.-Y.Jung,“ModifiedRMcodingschememadefromthebentfunctionfordimmablevisiblelightcommunications,”IEEEPhoton.Technol.Lett.,25,(1),11–13,1Jan.2013.

[19] P.Das,B.-Y.Kim,Y.Park,andK.-D.Kim,“Anewcolorspacebasedconstellationdiagramandmodulation scheme for color independentVLC,”Advances inElectricalandComputerEngineering,12,(4),11–18,Nov.2012.

[20] B.Bai,Q.He,Z.Xu,andY.Fan,“Thecolorshiftkeymodulationwithnon-uniformsignalingforvisiblelightcommunication,”inProc.1stIEEEInternationalConferenceonCommunications inChinaWorkshops (ICCC),Beijing,China,Aug.2012,pp.37–42.

[21] K.-I.AhnandJ.K.Kwon,“Colorintensitymodulationformulticoloredvisiblelightcommunications,”IEEEPhoton.Technol.Lett.,24,(24),2254–2257,15Dec.2012.

[22] IEEEStandardforLocalandMetropolitanAreaNetworks–Part15.7:Short-RangeWirelessOptical CommunicationUsing Visible Light, IEEE Standard 802.15.7–2011,Sept.2011.

[23] S.HranilovicandF.R.Kschischang,“Optical intensity-modulateddirectdetectionchannels:Signalspaceandlatticecodes,”IEEETrans.Inf.Theory,49,(6),1385–1399,2003.

[24] Y.Ohno, “CIE fundamentals for colormeasurements,”Proc. of IS&TNIP16 Intl.Conf.onDigitalPrintingTechnologies,Vancouver,Canada,Jan.2000.pp.540–545.

3 PerformanceenhancementtechniquesforindoorVLCsystems

Wen-DeZhongandZixiongWang

3.1 IntroductionLight-emitting diodes (LEDs) have beenwidely deployed for illumination, due to theirhigh performance and energy efficiency properties compared with conventionalincandescent and fluorescent lamps [1]. In addition, advantages such as high frequencyresponse,freespectrumlicenseandhighsecurityhavemadeLEDsapromisingmeansforwirelesscommunications,wherebypeoplecanaccesstheinternetthroughthesamevisiblelight. Significant research has been carried out to develop indoor high data rate visiblelightcommunication(VLC)systems[1–11].ThedatarateofVLCsystemsinalaboratoryenvironment has been demonstrated to reach the order of Gb/s [12, 13]. In addition,advancedmodulationschemessuchasspatialmodulation[14–17]havebeenintroducedtoVLC systems to enhance the data rate considerably. In 2011, light-fidelity (Li-Fi) wasintroduced by Haas [18], who demonstrated that a VLC system can be leveraged todevelopanalternativemethodofaccessingnetworkresourcesasasubstituteforwirelessfidelity (Wi-Fi). Although there has been significant progress in this area in the pastdecade, there are still some challenges to overcome in order to implement and deployVLC systems [19] on a larger scale. Two major challenges are selection of an uplinktransmission approach, and design of energy-saving receivers for long transmissiondistance.Inthischapter,anumberofrecentlyproposedtechniques[20–25]forenhancingthe performance of indoor VLC systems are discussed alongwith results pertaining totheirperformance.Theseincludeareceiverplanetiltingtechnique[21]andanLEDlamparrangement approach [22,23] to improve the signal-to-noise ratio (SNR) and bit errorrate(BER)performances,andperformanceevaluationofVLCsystemsunderadimmingcontrolscheme[20,24].

3.2 PerformanceimprovementofVLCsystemsbytiltingthereceiverplaneInaVLCsystem,thereceivermaybelocatedfarfromtheLEDlamp,wheretheSNRismuchsmallerthanatthoselocationsthatareclosetothelamp.ThelowerSNRisobtainedasthedistancefromthesourceincreasesandastheincidentangleincreases.Thissectiondescribesareceiverplanetiltingtechnique[21]toimprovetheperformanceof theVLCsystemacrosstheentireroom.Thedimensionoftheroomisassumedtobe5mlength×5mwidth×3mheight.TheSNRswith/withouttiltingthereceiverplaneareanalyzedandcompared.Forsimplicity,thereflectionsofwallsarenotconsideredinanalyzingtheSNRsincethelightfromtheline-of-sight(LOS)isdominant.

3.2.1 SNRanalysisofVLCsystemwithasingleLEDlamp

Figure3.1illustratesthegeometryofanindoorVLCsystemwithoneLEDlamplocatedontheceiling.TheparametersoftheVLCsystemconsideredinthischapteraregiveninTable 3.1. The LED lamp is assumed to be located at the center of the ceiling whoseposition is [2.5m, 2.5m, 3.0m], and the photo-detector (receiver) is on a deskwith aheightof0.85mfrom the floor.Letφbe theangleof radiationwith respect to theaxisnormal to theLED surface (plane). Following [1,8], the emitted light from anLED isassumedtohaveaLambertianemissionpattern,andtheradiationpatternisgivenby:

(3.1)

wherem is theorderofLambertianemission,which is related to the transmitter’ssemi-angleathalfpowerφ1/2asm=ln(1/2)/ln(cosφ1/2).

Figure3.1 GeometryoftheLEDsourceandphoto-detector(receiver)intheVLCsystem.

Table3.1 ParametersoftheVLCsystemconsideredinthischapter[1,20–22].

Roomsize(length×width×height) 5m×5m×3m

Heightofdeskwherethereceiverislocated 0.85m

Transmitter’ssemi-angleathalfpower(φ1/2) 60°

Physicalareaofphoto-detector(A) 10−4m2

Receiver’sfieldofview(FOV) 170°

Responsivityofphoto-detector(R) 1A/W

Backgroundcurrent(Ibg) 5100μA

Noisebandwidthfactor(I2) 0.562

Field-effecttransistor(FET)transconductance(gm) 30mS

FETchannelnoisefactor(Г) 1.5

Fixedcapacitance(η) 112pF/cm2

Open-loopvoltagegain(G) 10

Definiteintegralinvolvedinexpressionofcircuitnoise(I3)[26]

0.0868

ThefrequencyresponseoftheLEDandphoto-detectorisassumedtobeflatwithinthemodulationbandwidthofthesignalsconsideredinthischapter.ConsideringonlytheLOStransmissionpath,thechannelDCgainisgivenby[1,27]

(3.2)

wheredisthedistancebetweentheLEDsourceandthereceiver,Aisthephysicalareaofphoto-detector,andθistheangleofincidencewithrespecttotheaxisnormaltothedeskplanewheretheLEDislocated.Anglesφandθareassociatedwiththelocations/positionsof both LED source and receiver. Let [XS, YS, ZS] and [XR, YR, ZR] be the locations(coordinates)ofsourceandreceiver,respectively.WithreferencetoFig.3.1,theradiationangleφisdeterminedby

(3.3)

where‖X‖isthenormofX.Equation(3.3)indicatesthattheradiationangleφisconstantfor the given locations of the source and receiver. The value of the incident angle θ isdeterminednotonlybythelocationsofsourceandreceiver,butalsobythedihedralanglebetween the receiver plane and the desk plane where the receiver is located. WithreferencetoFig.3.1,letνRSbethevectorfromthereceiver to thesource,andνRbethevectorofthereceiver.Thentheincidentangleθiscalculatedby

(3.4)

where(νRS,νR)istheinnerproductofνRSandνR.SubstitutingEq.(3.4)intoEq.(3.1),the

channelDCgaininEq.(3.1)becomes[21]

(3.5)

AssumethattheLEDlightismodulatedwithamodulatingsignalf(t).Theopticalsignalat the output of an LED can be expressed by p(t) =Pt (1 +MI f(t)), where Pt is thelaunchedpoweroftheLEDlampandMIisthemodulationindex[28],whichisassumedtobe0.2.ThereceivedopticalpowerPrisgivenby

Pr=H(0)Pt. (3.6)

After photo-detection and considering that theDC component of the detected signal isfilteredoutinthereceiver,theoutputelectricalsignalisgivenby

s(t)=RPrMIf(t), (3.7)

whereRistheresponsivityofthephoto-detector.Hence,theSNRoftheoutputelectricalsignalcanbecalculatedby[5],

(3.8)

where is the average power of the output electrical signal, andPnoise is the noisepower.Thenoisepowerconsistsofbothshotnoiseandthermalnoise,whosevariancesaregivenby[1],

(3.9)

(3.10)

where isthetotalreceivedpower,qistheelectroncharge[13],B istheequivalentnoisebandwidth,kdenotestheBoltzmannconstant,andTKrepresentstheabsolute temperature.Theparameters inEqs. (3.1)–(3.10) and other parameters used intheVLCsystemarelistedinTable3.1.

UsingEq.(3.8), theSNRdistributionof thereceiver locatedon thedeskplaneat theheightof0.85mcanbecalculated.IncalculatingtheSNRdistribution,theparametersinTable3.1areused.Figure3.2showstheSNRdistributionforthecasewhenthelaunchingpowerof theLED lamp is5W,and theLED is locatedat thecenterof theceiling.As

showninFig.3.2,themaximumSNRis28.94dBwhenthereceiverislocatedrightbelowtheLED,whiletheminimumSNRis6.23dBwhenthereceiverisplacedatacornerofthe room. Thus, the peak-to-trough SNR difference is 22.70 dB. Note that the SNRverticaltintbarontherighthandsideofthefigureindicatestherelationshipbetweentheSNRvalueandthecolor(blackrepresentsthesmallestvalueofSNR;whilewhitedenotesthe highest value of SNR). The large variation of the SNR in a room can significantlyreducetheoverallsystemperformance[21].ThelargeSNRdifferenceinaroomiscausednotonlyby thedistancebetween theLEDand the receiver,butalsoby thenon-normalincidenceof the light fromtheLEDto the receiver.For thegiven locationsof theLEDandreceiver,thedistancebetweentheLEDandthereceivercannotbechanged.However,theincidentangleofthelightfromtheLEDtothereceivercanbeadjustedtoreducetheSNRvariation.

Figure3.2 SNRdistributionofVLCsysteminaroomwithoneLEDlamplocatedonthecenteroftheceiling.

3.2.2 ReceiverplanetiltingtechniquetoreduceSNRvariationAs described above, the non-normal incidence of the light could result in a large SNRvariation ina room.The incidentangleθ isdeterminedby thevectorsνRandνRS.NotethatthevectorνRisalwaysperpendiculartothereceiverplane.ThevectorνRSisconstantforthegivenlocationsofthesourceandreceiver.AccordingtoEq.(3.4),cosθreachesitsmaximumwhen the two vectors νRS and νR are parallel to each other, i.e., the receiverplane faces the source. In the case where the receiver is not located on the desk rightbelowthesourceontheceiling,especiallywhenthereceiver ispositionedinoneof thecorners of the room, themaximum channelDC gain is greatly reduced as the incidentangleθincreases.However,bytiltingthereceiverplanetowardsthesourcesuchthatthe

two vectors νRS and νR become parallel to each other, the value of cosθ reaches itsmaximumand thusmaximumchannelDCgainof aparticularpositioncanbe attained,whichisonlyassociatedwiththetransmissiondistancedandtheangleofradiationφ[21].

VectorνRScanbeexpressedasνRS= [a,b,c]=[XR,YR,ZR]– [XS,YS,ZS].Hereweassumethattiltingthereceiverplanedoesnotchangethepositionofthereceiver.Inthesphericalcoordinatesystem, thelocationof thereceiver isselectedas theorigin.Beforetiltingthereceiverplane, thevectorVR is [0,0,1],whichmeans that thereceiverplanepoints to the ceiling. After tilting the receiver plane towards the LED source on theceiling, the vectorVR becomes [sinβ·cosα, sinβ·sinα, cosβ], where β is the inclinationangle[29]whichisequaltothetiltingangle,asshowninFig.3.3andtheazimuthangleαisdeterminedbythepositionsofthereceiveraswellasthesourceprojectiononthedesk.IntheCartesiancoordinatesystemwiththereceiverastheorigin,thevalueofangleαisgivenbyEq.(3.11):

(3.11)

Thus,cosθinEq.(3.4)becomes

(3.12)

SubstitutingEq.(3.12)intoEq.(3.5),thechannelDCgainaftertiltingthereceiverplane,denotedbyf(β),becomes,

(3.13)

Since the receiver is located on the desk, the initial inclination angle β is zero. Thisreceiver plane tilting technique can be implemented by electricalmachinery.When theinclinationangle is increasedafter tilting the receiverplane, the twovectorsνR and νRStend to becomeparallel to eachother and thereby the receivedoptical power increases.Theelectricalmachinerywillnotstopchangingtheinclinationangleβuntilthereceivedopticalpowerdoesnotincreaseanymore.

Figure3.3 GeometryoftheLEDsourceandphoto-detectoraftertiltingthereceiverplane.

TheNewtonmethod(afastalgorithmtofindthemaximumvalueoff(β)[30])canbeemployedtosearchfortheoptimuminclinationangleβ.Afterfindingtheoptimumtiltingangleby theNewtonmethod, themaximumopticalpower isobtained foreach receiverposition.Figure3.4 shows the improvedSNRdistribution.ThemaximumSNRremainsunchangedat28.94dB,whiletheminimumSNRateachcorneroftheroomisincreasedto11.92dB,resultinginanimprovementof5.69dBinthepeak-to-troughSNRdifferenceascomparedtothecasewithouttiltingthereceiverplane.

Figure3.4 SNRdistributionofVLCsysteminaroomwithoneLEDlamponthecenteroftheceilingaftertiltingthereceiverplane.

3.2.3 MultipleLEDlampswiththereceiverplanetilting

techniqueTo further reduce the SNR variation, multiple LED lamps could be employed inconjunctionwiththereceiverplanetiltingtechnique[21].Asanexample,consideracasewherefourLEDlampsarelocatedontheceilingatthepositionsof[1.5m,1.5m,3.0m],[1.5m,3.5m,3.0m],[3.5m,1.5m,3.0m]and[3.5m,3.5m,3.0m],respectively.Asthelight is received from all of the four LED lamps, the channel DC gain in Eq. (3.2) ismodifiedasfollows:

(3.14)

wherethesubscriptidenoteslampi.Inthiscase,thetotalradiatingpowerfromthefourLEDsremainsunchangedat5W,i.e.,thelaunchingpowerofeachLEDlampisreducedtoonequarterofthatofoneLEDlampinsubsection3.2.2.Figure3.5(a)showstheSNRdistributionwithfourLEDlampswithouttiltingthereceiverplane.AsshowninFig.3.5(a),themaximumandminimumSNRsare22.72dBand8.95dB,respectively.Thatis,thepeak-to-trough SNR difference is 13.77 dBwithout tilting the receiver plane. It is alsoobservedthat in thearea inside theprojectionsof theLEDsonthedeskplane, theSNRdistributionisalmostconstantandhencethere isnoneedtoadjust theSNRdistributionwithinthisarea[21].

Figure3.5 SNRdistributionofVLCsysteminaroomwithfourLEDlampslocatedontheceiling:(a)before;and(b)aftertiltingthereceiverplane(after[21]).

However,theSNRvariationisquitelargeintheplacesoutsideprojectionsofthefourLEDsonthedeskplane.Insuchplaces,theSNRdifferencecanbereducedbytiltingthereceiverplaneinthesamewayasdescribedforthecaseofasingleLEDlamp.

When the receiver is not equidistant with respect to any of two LEDs, it faces thenearest LED of the four, which determines the value of azimuth angle α. When thereceiver isequidistantfromtwoLEDs, it faces to themiddleof them.ThetotalchannelDCgainaftertiltingthereceiverplane,denotedasf(β),isgivenby

(3.15)

AsinthecaseofasingleLEDlamp,theoptimumtiltingangleβcanthenbeobtainedbytheNewtonmethod.Figure3.5(b)showstheimprovedSNRdistributionwiththereceivertiltingtechnique.AsshowninFig.3.5(b),themaximumSNRremainsat22.72dB,whiletheminimumSNRincreases to13.09dB.That is, thepeak-to-troughSNRdifference isreducedfrom13.77dBto9.63dB.Inotherwords,a4.14dBimprovementinthepeak-to-troughSNRdifferenceisachieved.InthecaseoffourLEDlamps,thestudyshowsthatonlythreesearchstepsarerequiredbytheNewtonalgorithmtoconvergetotheoptimumvalue[21].

3.2.4 SpectralefficiencyTilting the receiver planemakes it possible to attain optimumSNR.A higher SNRpersymbol(Es/N0)meansabetterbiterrorrate(BER)performance.Asdiscussedabove,theSNRcanvary considerablywithin the room.Similarly toRFwireless communications,adaptive advancedmodulation formats such asM-ary quadrature amplitudemodulation(M-QAM) orthogonal frequency division multiplexing (OFDM) can be employed toenhancetransmissioncapacity[31].Thissubsectiondiscussesthespectralefficiencyofasingle user VLC system employing adaptive M-QAM OFDM. Figure 3.6 is a blockdiagram for a VLC system with adaptive M-QAM OFDM, where the value of MrepresentsthenumberofpointsinthesignalconstellationandcanbevariedinaccordancewiththeSNR.Hereitisassumedthatinfrared(IR)oranotherkindofwirelesstechnologyisemployedtoprovidechannelfeedbackaswellasuplinktransmission.

Figure3.6 BlockdiagramofaVLCsystememployingadaptiveM-QAMOFDM.CP:cyclicprefix,P/S:paralleltoserial,S/P:serialtoparallel(adaptedfrom[21]).

Following[32,33]andapplyinggraycoding for themappingof theM-QAM signal,theBERoftheM-QAMOFDMsignalisgivenby

(3.16)

whereQ(·)istheQ-function.NotethattherelationshipbetweenSNRpersymbol(Es/N0)andSNRperbit(Eb/N0)isgivenbyEs/N0=log2(M)×Eb/N0[33].

As shown in Fig. 3.6, when theM-QAM OFDM optical signal reaches the photo-detector, its power is detected and sent back to the sources on the ceiling via the IRfeedbackchannelaftertiltingthereceiverplane.HerethebenchmarkBERissetat10−3,which satisfies the requirement of error-free transmission by applying a forward-errorcorrection (FEC)code [34].A smallervalueofM shouldbe chosen to attain aBERof10−3 at the locations/positionswith lowSNR.However, at theplaceswithhighSNR,alargervalueofMisselectedtoachieveahigherdataratewhilemaintainingasteadyBERof10−3.It isnotedthatthevalueoftheM-QAMOFDMsignalshouldberealinopticaltransmission (this is attained by applyingHermitian symmetry [4]), resulting in a 50%reductioninspectralefficiency[35,36].UsingEq.(3.16),theSNRpersymbolthresholdforachievingaBERof10−3canbecalculatedforagivenvalueofM,asshowninTable3.2,wherethesymbolrateis50Msymbol/s.

Table3.2 ThecalculatedSNRpersymbolthresholdforachievingaBERof10−3fordifferentvaluesofMinM-QAM.

ValueofMinM-QAM 4 16 64 256 1024

SNR(dB)persymbolthreshold 9.8 16.5 22.5 28.4 34.2

LetNbe thenumberofsubcarriersused inOFDM.Assuming that thepulseshape isrectangular,thespectralefficiency(SE)inunitsofbit/s/HzofanM-QAMOFDMsignalcanbeexpressedas[37]

(3.17)

where½representstheSEreductionasaresultofapplyingtheHermitiansymmetry.Inasingle-user VLC system using adaptiveM-QAM OFDM, the value ofM is varied inaccordance with the value of SNR. The average SE across the entire room can beexpressedby

(3.18)

wherep(Mi)istheprobabilitythatMi-QAMisused,whichcanbecalculatedbasedontheSNRdistributionintheroom.

Figure3.7 (a)and(b) show the averageSE for the casesofoneLED lampand fourLED lamps, respectively. The average SE increases with the total LED power. This isbecause SNR is increased with the total LED power, which in turn increases theprobabilityofemployingalargervalueofMinadaptiveM-QAMmodulation.Figure3.7(a)and(b)alsoshowtheaverageSEimprovementattainedbytiltingthereceiverplane.Inthe case of one LED lamp, the average improvement is about 0.36 bit/s/Hz, and themaximumimprovementis0.47bit/s/HzwhentheLEDlamppoweris9W.Inthecaseoffour LED lamps, the average improvement is 0.18 bit/s/Hz, and the maximumimprovementis0.23bit/s/HzwhenthetotalLEDlamppoweris19W.

Figure3.7 Averagespectralefficiencyforthecasesof:(a)oneLEDlamp;and(b)fourLEDlamps,withandwithouttiltingreceiverplane(adaptedfrom[21]).

3.3 PerformanceimprovementofVLCsystemsbyarrangingLEDlampsFor an indoor VLC system, equal signal quality in terms of SNR and BER across theentire room is important; this is particularly true when there are multiple users in the

room.AsdiscussedinSection3.2,theLEDlampsareusuallylocatedaroundthecenterofthe ceiling (called a centered-LED lamparrangement) in a typical room.This centered-LEDlamparrangementmakesSNRvarylargelyfromonelocation/positiontoanother[1,21], which can significantly affect the quality of the received signal across the room.Section 3.2 analyzes and discusses the performance improvement in an indoor VLCsystem by tilting the receiver plane. Tilting the receiver plane can reduce the SNRvariationtosomeextent,butitmayconsiderablyincreasethecomplexityofthereceiverdesign. This section describes an effective LED lamp arrangement reported in [22] tosignificantlyreduceSNRvariationandhenceimproveBERperformanceofVLCsystemsthroughout the entire room, so that multiple users can receive signals of almost equalquality,regardlessoftheirlocation.

3.3.1 ArrangementofLEDlampsInordertoseetheeffectivenessoftheLEDlamparrangement,withoutlossofgenerality,letus first consider a situationwhere there are16 identicalLED lamps that are locatedaroundthecenteroftheceiling.TheintervalbetweenadjacentLEDlampsis0.2m.EachLEDlampemits125mW,andhencethetotalpowerof16centered-LEDlampsis2W.TheotherparametersoftheVLCsystemconsideredhereareprovidedinTable3.1.100positionsaresampledintheroom.Thesesampledpositionsareuniformlydistributedontheplanewherethephoto-detectorislocated.Toevaluatethequalityofsignalreceivedateverylocationintheroom,aparameterdenotedQSNRisintroduced,whichisdefinedas

(3.19)

where is themeanofSNR, andvar(SNR) denotes the variance ofSNR.AhighervalueofQSNR implies that theSNR ismoreuniformlydistributedover theentire room.Figure3.8(a)illustratesthecalculatedSNRdistribution,inwhichthemaximumvariationinSNRisabout14.5dB.Inthiscase,theassociatedQSNRisapproximately0.5dB,whichmeans that the SNR varies significantly throughout the room and the signal quality isstrongly related to the user’s location/position. In calculating SNR distribution, thereflectionsofwallsor the ISIarenotconsideredhere, since theLOS light isdominant.However,theywillbeincludedintheanalysisofBERperformanceinthenextsubsection.

Figure3.8 SNRdistributionofaVLCsysteminaroomwithatotalpowerof2W:(a)16centered-LEDlamps;(b)16circle-LEDlamps(adaptedfrom[22]).

ThepoorSNRdistributionillustratedinFig.3.8(a)arisesfromthecentered-LEDlamparrangement,wherethedistancesbetweenauserinthecorneroftheroomandthelampsare much greater than the distances between a user in the center of the room and thelamps.

IftheLEDlampsarelocatedseparatelyandsymmetricallytothecenteroftheceiling,the SNR distribution is expected to be improved considerably, since the differences indistances between users and lamps are reduced [22]. In [22], a circle-LED lamparrangement is proposed to reduce SNR variation. Figure 3.8 (b) shows the SNRdistribution for the caseof16 lamps,where theLED lampsaredistributed evenlyon acircleontheceilingwitharadiusof2.5mandeachlampemitsthesameamountofpower

asthatinthecaseof16centered-LEDlamps.AsshowninFig.3.8(b),theSNRvariationisreducedlargelyfrom14.5dBto2.4dB,andtheQSNRincreasessubstantiallyfrom0.5dB to 9.3 dB. This demonstrates that the circle arrangement offers much better signalquality and thereby an improved communication system, which is little related to theuser’sposition[22].

AlthoughtheQSNRisashighas9.3dBforthecircle-LEDlamparrangement,theSNRinthefourcorners(whichareequivalentintermsofdistance)isstillsmallerthanthatinotherlocations,asshowninFig.3.8(b).AtechniquetofurtherimprovetheSNRsatthefourcornersistoaddanLEDlampineachcornerasdescribedin[22].Assumethatthedistancesofthelampsinthecornerstotheirnearestwallsareall0.1m,asshowninFig.3.9(a).Inordertomakeafaircomparison,thetotalLEDlamppowerisstillmaintainedat2WandthenumberofLEDlampsplacedinthecircleisreducedto12,sothatthetotalnumberofLEDlampsremainsunchanged.LetPt,circleandPt,cornerbetheemittedpowersofacircle-LEDlampandacorner-LEDlamp,respectively.Byadjustingthepowerofthe4corner-LEDlampsandthe12circle-LEDlampsspreadinguniformlyonthecircle,thevarianceofthereceivedopticalpowerPr,canbeminimized,asfollows[22]:

minvar(Pr)=minE[(Pr,j−E(Pr,j))2], (3.20)

whereE(∙)denotes themeanvalueandPr,j is the receivedpoweratsampledposition j,whichisdescribedby

(3.21)

Next,theradiusoftheLEDcircleandthedistancebetweenacorner-LEDanditsnearestwallsarechangedtofindtheminimumSNRfluctuation.TheresultsaregiveninTables3.3and3.4.AsshowninTable3.3,whenthedistancebetweenthecorner-LEDlampsandtheirnearestwallsis0.1m,theoptimumradiusoftheLEDcircleisbetween2.2to2.3m,whereQSNRislargest.Table3.4showsthattheoptimumdistancebetweenthecorner-LEDlampsandtheirnearestwalls is0.1m,whentheradiusof theLEDcircle is2.2m.It isalso found that the largestQSNR isobtainedwhen thepowersof eachcorner-LED lampand each circle-LED lamp are 238mW and 87mW, respectively. The improved SNRdistributionisshowninFig.3.9(b),wherethemaximumSNRdifferenceis0.85dBandthecorrespondingQSNR is12.2dB.Theabove results reveal that thearrangementof12circle-LED lamps and 4 corner-LED lampswith the given parameters in Table 3.1 canprovide almost identical communication quality to multi-users, irrespective of theirpositionsintheroom[22].

Figure3.9 Arrangementof12LEDlampsinacircleand4LEDlampsincorners:(a)locationsofLEDlampsand100receivers;(b)SNRdistributionwithatotalpowerof2W(adaptedfrom[22]).

Table3.3 SNRandQSNRunderthearrangementof12circle-LEDsand4corner-LEDswith different radii, where the distance between a corner-LED and its nearestwallsis0.1m[22].

Radius(m) 2.1 2.2 2.3 2.5

SNRrange(dB)[min,max] [5.5,6.4] [5.6,6.5] [5.5,6.5] [5.3,6.5]

QSNR(dB) 12.1 12.2 12.2 11.5

Table3.4 SNRandQSNRunderthearrangementof12circle-LEDsand4corner-LEDswithdifferentdistancesbetweenacorner-LEDand itsnearestwalls,where theradiusoftheLEDcircleis2.2m[22].

Distance(m) 0.5 0.25 0.15 0.1

SNRrange(dB)[min,max] [6.0,7.4] [5.9,6.8] [5.7,6.6] [5.6,6.5]

QSNR(dB) 10.7 11.7 12.1 12.2

3.3.2 BERanalysisAsdiscussedabove,thearrangementof12circle-LEDlampsand4corner-LEDlampscanprovidealmostuniformlydistributedSNRtousers,irrespectiveoftheirlocations/positionsintheroom.However,thismaycauseanincreaseinISIsincethephoto-detectorreceivessignalsfromalloftheLEDlampswhosedistancestothereceivervarygreatly,whichmayreducetheBERperformancesignificantly.Withoutconsideringreflections,themaximumdifference of light arrival time under the arrangement of 12 circle-LED lamps and 4corner-LED lamps is 15.9 ns when the receiver is located in a corner position [22];whereasthemaximumtimedifferenceisonly2.34nswhenthe16LEDlampsarelocatedinthecenteroftheceiling.In[22],theBERperformanceofboth100Mb/sand200Mb/sbipolarOOKsignalsisanalyzedandevaluated.

Thissubsectionpresents theBERanalysisofa100Mbit/sbipolarOOKsignalunderthe arrangement of 12 circle-LED lamps and 4 corner-LED lamps. The first order ofreflection is considered in the BER analysis. Let us consider theworst casewhere thereceiverisplacedinacornerwithalocationof[0.25m,0.25m,0.85m]andhencetheISIismostsevere.Thereflectivityandthemodulationindexareassumedtobe0.7and0.2,respectively.

Figure3.10(a)illustratesthepulseshapeofthereceivedbit“1”underthearrangementof12circle-LEDlampsand4corner-LEDlamps,asshowninFig.3.9(a).AscanbeseeninFig.3.10(a), thedurationof thereceivedbit“1”ismorethan30ns,morethanthreetimesthetransmittedbitperiodT=10ns.Leth=[1a1…ak]bethenormalizedchannelresponse.Itisnotedthatai(i=1,2,…,k)representstheISIcontributionfromthepresentbittothesubsequentithbit.LetImbethepresentreceivedbitwithamplitude ,andIm−1,…,Im−k be thek preceding receivedbits.Considering the ISI from thek precedingbits,thepresentreceivedsignalymisexpressedby

(3.22)

wheren is the additive white Gaussian noise (AWGN) with power spectral density ofN0/2.Let betheconditionalerrorprobabilitywhenthepresentreceivedbitis“1,”i.e.,amplitudeis .Then canbeexpressedby[32],

(3.23)

where Im−1,…,Im−k is a combination of the k preceding received bits, and,i.e.,eachofthekprecedingreceivedbitsiseither1or

0. Note that bit 0 is mapped to −1 for a bipolar OOK signal. P(Im−1,…,Im−k) is theprobabilityofonesuchcombination. istheconditionalerrorprobabilitywhenthepresentreceivedbitis1andoneofthecombinationsofthekpreceding receivedbitsoccurs.Forexample,whenall thekpreceding receivedbits andthepresentreceivedbitare1,theconditionalerrorprobabilityinEq.(3.23)isgivenby

(3.24)

whereQ(∙)istheQ-function.Sincetheoccurrenceofbit1andbit0isequal,theoverallBERperformanceisgivenby

(3.25)

Due to the ISI from the k preceding bits, BER performance is expected to be reducedsignificantly without performing equalization on the received signal. Several signalequalizationtechniqueshavebeendevelopedtomitigateISI[32].Here the timedomainzero-forcing(ZF)equalizationisusedtosuppressISI[32,38].Let{cn}bethecoefficientoftheZFequalizer,and{qn}betheoutputoftheequalizer.Then{qn}istheconvolutionof{cn}andthechannelresponseh.Ideally,{qn}shouldbegivenby

(3.26)

Foranon-idealequalizerwithafinitenumberoftaps,qn≠0whenn≠0, i.e., residualISIexists.Replacingh=[1a1…ak]inEq.(3.22)with{qn}andsubstitutingymintoEq.(3.25),theimprovedBERperformancewithtimedomainZFequalizationisattained.

Figure3.10 BERperformanceofa100Mbit/sbipolarOOKsignalwhenthereceiverislocatedatthecorner:(a)thepulseshapeofreceivedbit“1”withISI;and(b)BERwithandwithoutZFequalization,whenthetotalLEDpoweris2W(adaptedfrom[22]).

Figure3.10(b)showstheBERperformanceofboththetheoreticalanalysisandMonte-

Carlo (MC) simulation with and without ZF equalization. The theoretical results agreewith thesimulation resultsverywell (theyarecompletelyoverlapped in the figure); theBER performance is significantly improved after applying ZF equalization. As anexample,therequiredtotalLEDpoweris6.2WforachievingaBERof5×10−4withoutZF equalization, while the required total LED power is reduced to 3.0 W with ZFequalization.ThusZFequalizationprovides3.2dBimprovementinpowerreduction.ItisalsoobservedthattheBERperformancewithZFequalizationisalmostthesameasthatwithout ISI (the twocurvesarecompletelyoverlapped).Thismeans that the ISIcanbetotallymitigated byZF equalization.By applying an FEC code, error-free transmissioncouldbeattainedunderthisBERrequirement[34].

3.3.3 ChannelcapacityanalysisFor a noisy communication channel, the channel capacity is defined as the maximummutualinformationbetweentheinputandoutputofthechannelovertheinputdistribution[38].Foradiscrete-inputandcontinuous-outputVLCchannel,thechannelcapacitycanbeexpressedby

(3.27)

wherePX(.)istheinputdistribution,xisthediscrete-inputsymbolinthesetofX,fY(y)isthe probability density function of the continuous-output signal y, and fY|X(y|x) is theconditionalprobabilitydensityfunctionofywhenthegiveninputsymbolisx.Assumingthattheinput-discretesignalisabipolarOOKsignal,fY|X(y|x)isgivenby,

(3.28)

where is the variance of noise. Considering the ISI discussed in Section 3.3.2, theconditionalprobabilitydensityfunctioninEq.(3.28)canbeexpressedby

(3.29)

whereh=[1a1…ak]isthechannelresponse,xmisthepresenttransmittedbitandxm−1,…,xm−k are the k previous transmitted bits. Substituting Eq. (3.29) into Eq. (3.27), thechannelcapacityofabipolarOOKsignalwithISIinaVLCsystemcanbecalculated.

Figure3.11showsthecalculatedchannelcapacityofa100Mbit/sbipolarOOKsignalversustotalLEDlamppowerunderthearrangementof12circle-LEDlampsand4corner-LEDlampswithandwithoutZFequalization.AsshowninFig.3.11,theZFequalizationimprovesthechannelcapacitysignificantly.ThemaximumchannelcapacityimprovementbyZFequalizationis0.17bits/symbol,whichoccursatthetotalLEDpowerof2W.Itisalsoshown in the figure thatwithZFequalization, thechannelcapacity remainsalmostthesameasthatofthechannelwithoutISI.Asareference,theShannoncapacityisalsodepictedinFig.3.11.

Figure3.11 Channelcapacityofa100Mbit/sbipolarOOKsignalunderthearrangementof12circle-LEDlampsand4corner-LEDlamps(adaptedfrom[22]).

3.4 DimmingcontroltechniqueanditsperformanceinVLCsystemsIlluminationandcommunicationare the twomajor functionsof theLEDlamps inVLCsystems. The brightness of the LED light needs to be adjusted in accordance with therequirements and comfort of users. In addition, adjusting the brightness of LED lamps

helps save energy [39]. Pulse width modulation (PWM) is widely used as a dimmingcontroltechnique[39,40],wherethebrightnessoftheLEDlightischangedbyadjustingthedutycycleofthePWMsignalwithoutvaryingtheLEDcurrent[20].

AsshowninFig.3.12(a),theLEDcurrentismodulatedbyaPWMsignaltocontrolitsbrightnessbychangingtheondurationwithinthewholeperiod.Sothelightisdimmingduringthewholeperiodof thePWMsignal.Thedataaremodulatedontothedimming-controlled light in the on time only, and no light is transmitted during the off time asdepicted in Fig. 3.12 (b). Since the LED current remains constant throughout, thebrightness of the LED light is varied by applying the PWMdimming control signal toadjustthedurationofitsonperiodasafractionofthewholeperiod.

Figure3.12 (a)Dimmingcontrol;(b)signalwaveformwithdimmingcontrol,dutycycle=0.6(adaptedfrom[20]).

WiththedutycycleofthePWMsignalsetto1,alltheLEDlightistransmittedandthe

lightobtainedisofthehighestbrightness.Whenthedutycycleisreduced,theLEDlightintheoffperiodisblockedandhencethelightisdimmingduringthewholedurationofthePWMsignal.ItisworthmentioningthatthefrequencyofthePWMsignalshouldbereasonablyhigh, sayhigher than200Hz [41],otherwise itwillcause flickerandwouldhaveadverseeffectsonuserhealth[42].

3.4.1 BipolarOOKsignalunderdimmingcontrolAsdiscussedabove,with thedimmingcontrol, thedurationofdata transmission inonePWM dimming control signal period (T) is reduced, compared with the case withoutdimmingcontrol.AlthoughtheBERperformanceintheonperiodofthePWMdimmingcontrolsignalisnotchangedforagivenmodulationformat,thenumberoftransmittedbitswouldbereducedinthewholePWMperiod,whichinfactreducestheaveragedatarate.Todealwith thisproblem, thedata rateshouldbe increasedaccordinglywhile theLEDlight isdimming,so that thenumberof transmittedbitsremainsunchanged.That is, thefollowingequationshouldhold:

R1TD=R0T, (3.30)

whereRrepresentsthebitrateofabipolarOOKsignal,andDdenotesthedutycycleofaPWMdimmingcontrolsignal.Thesubscripts“0”and“1”inEq.(3.30)correspondtothesituationswithout andwith dimming control, respectively.The adaptive data rate underdimmingcontrolisshowninFig.3.13(a).Inthefollowinganalysis,theoriginaldatarateR0ofanOOKsignalisassumedtobe10Mbit/swithoutdimmingcontrol.AsdepictedinFig.3.13(a),underthedimmingcontrol,theadaptivedatarateisinversely-proportionaltothe duty cycle of the PWM dimming control signal in order to make the number oftransmitted bits unchanged.Hence, the adaptive data rateR1 is higher than the originaldata rateR0 and is increased as the duty cycle is reduced. For example,when the dutycycle is 0.1, the adaptive data ratewould be 10 times as high as the original data rate,whichwouldmakethesystemhardtoimplementontheoriginalcircuit[20].

Figure3.13 (a)Adaptivedatarateversusthedutycycle;(b)therequiredLEDlamppowertoachieveBERof10−3withoutapplyingdimmingcontrolinanOOKVLCsystemversusthedutycycle(adaptedfrom[20]).

Following[43],theBERperformanceofabipolarOOKsignalisgivenby

(3.31)

whereQ(∙)istheQ-function.Asdescribedabove,thesignalpoweriskeptconstantwhilechanging duty cycle; however, the noise power is increased as the data rate grows [1]whenthedutycycledecreases,whichreducestheBERperformanceintermsofSNRasshown inEq. (3.31).Under the dimming control, theBER of less than 10−3 should beguaranteedtoachieveerror-freetransmissionbyapplyingtheFECcode[34].SubstitutingEq.(3.7)intoEq.(3.31),wehave

(3.32)

Notethattheaveragepower ofthebipolarOOKsignalinEq.(3.7)isunityandthenoisevarianceσ2isrelatedtothetotalreceivedopticalpowerPrintermsoftheLEDlamppowerPt.BysolvingEq.(3.32),therequiredLEDlamppowertoachieveaBERof10−3

without applying dimming control can be obtained, which is associated with bothmodulationindexMIandthenoisevariance,whenthelocationsofLEDlampandreceiverarefixed.

Inthefollowingscenario,thelocationsofanLEDlampandreceiverareassumedtobe[2.5 m, 2.5 m, 3.0 m] and [3.75 m, 1.25 m, 0.85 m], respectively. In addition, theilluminanceisassumedtobeproportionaltotheLED’sdrivingcurrent.ItisnotedthattherequiredLEDlamppowertoachieveaBERof10−3shouldbekeptunchangedunderthedimming control aswell [20]. Figure 3.13 (b) shows the required LED lamp power toachieveaBERof10−3withoutapplyingdimmingcontrolversusthedutycycle.Whenthedutycycleisvariedfrom1to0.3,i.e.,theilluminanceofLEDlightisreducedto30%oftheinitialilluminance,therequiredLEDlamppowerswithoutapplyingdimmingcontrolincreaseslowlyto0.35Wand0.24Wformodulationindicesof0.2and0.3,respectively.However,asthedutycycleisfurtherreducedfrom0.3to0.1,i.e.,theilluminanceoftheLEDlight isonly10%of the initial illuminance, therequiredLEDlamppowerwithoutapplyingdimmingcontrolgrowsdrasticallyfrom0.35Wto0.72Wwhenthemodulationindexis0.2andgrowsfrom0.24Wto0.48Wwhenthemodulationindexis0.3.SincetheLEDlamppowerhastobekeptconstantthroughoutthedimmingcontrolscheme,inorder toachieveaBERof10−3 for theentiredutycycle rangeof0.1 to1, the requiredLED lamp power should be set as high as 0.72W if themodulation index is 0.2, and0.48Wifthemodulationindexis0.3.Inthisway,whilethedimmingcontrolisapplied,theLEDlamppower ismaintainedconstantandbothaveragedata rateand theBERof10−3 can be guaranteed. The above results reveal that in an OOK VLC system withdimming control, not only the data rate but also the lamp power need to be increasedsignificantly inorder toprovideaguaranteeofcommunicationquality in termsofBERandaveragedataratefortheentiredutycyclerangeof0.1to1.

3.4.2 AdaptiveM-QAMOFDMsignalunderdimmingcontrolThissubsectionanalyzesanddiscusses theperformanceofanadaptiveM-QAMOFDM

signalunderthedimmingcontrolscheme,whereMrepresentsthenumberofpointsinthesignalconstellation[44].SinceonesymboloftheM-QAMsignalcarrieslog2(M)bits,thetotal number of transmitted bits could be kept the same by increasing the symbol rateand/orthevalueofMwhenhigherlevelM-QAMisused.LetM0betheinitialnumberofpoints in thesignalconstellation,andM1be theadaptivenumberofpoints in thesignalconstellation.Hence,wehave

(3.33)

whereR0istheoriginalsymbolrateoftheM-QAMsignalwithoutdimmingcontrol,andR1 is the adaptive symbol rate of theM-QAMsignalwithdimming control.HereR0 isassumedtobe10Msymbol/s.SubstitutingSNRpersymbol inEq.(3.8) intoEq. (3.16),theBERperformanceofanM-QAMsignalwithoutapplyingdimmingcontrolisobtainedfrom,

(3.34)

The average power of theM-QAM signal in Eq. (3.8) is also unity. Solving Eq.(3.34),therequiredLEDlamppowertoachieveaBERof10−3withoutapplyingdimmingcontrol of theM-QAM signal is obtained.Note thatwhenOFDM is applied in aVLCsystem, the bandwidth should be at least twice as large as the symbol rate, due to theapplication of Hermitian symmetry. In addition, Eqs. (3.16) and (3.34) are the BERperformanceforanM-QAMwithasquareconstellation.ForanM-QAMwithnon-squareconstellation, thethresholdofSNRpersymboltoachieveaBERof10−3 isobtainedbyusingMCsimulation.

BasedonEqs. (3.33)and(3.34), thevaluesofR1andM1underdifferentdutycyclescan be calculated.Table 3.5 shows the relationship betweenM1 and the duty cycle foradaptiveM-QAMwith dimming control.As expected, the value ofM1 increases as thedutycycleisreduced.Figure3.14(a)showstherelationshipbetweentheadaptivesymbolrateR1andthedutycycle.Whenthedutycycleislessthan0.3,theadaptivesymbolratehas to increase significantly to satisfy the communication quality. The highest adaptiverateis2.5timesashighastheoriginalsymbolratewhenthedutycycleis0.1,whichisthesame as that of anOOK signalwhen the duty cycle is 0.4, as shown in Fig. 3.13 (a).Hence,theincreaseinthesymbolrateofanM-QAMOFDMsignalismoderatecomparedwiththatofanOOKsignal.

Table 3.5 Relationship betweenM1 and duty cycle for adaptiveM-QAM withdimmingcontrol.

Dutycycle 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

M1 256 128 64 32 16 16 8 8 8 4

Figure3.14 (a)AdaptivesymbolrateofM-QAMsignalversusthedutycycle;(b)therequiredLEDlamppowertoachieveBERof10−3withoutapplyingdimmingcontrolversusthedutycycle.

Figure3.14(b)showstherelationshipbetweentherequiredLEDlamppowerandtheduty cycle. When the duty cycle is greater than 0.4, the required LED lamp powerincreasesslowlyasthedutycycledecreases.Whenthedutycycleis0.9,therequiredLEDlamppowerswithoutapplyingdimmingcontrolare0.97Wand0.65Wforthemodulation

indicesof0.2and0.3,respectively,whicharelargerthantherequiredLEDlamppowerofanOOKsignalwitha0.1dutycycle.However, therequiredLEDlamppowerincreasesrapidlywhenthedutycycleisreducedbelow0.4.Whenthedutycycleis0.1,therequiredLED lamppowersofM-QAMOFDMsignalsare9.5Wand6.3Wfor the0.2and0.3modulation indices, respectively, which are more than 13 times larger than thecorrespondingrequiredLEDlamppowersofanOOKsignal.Hence,whenthedutycycleislargerthan0.4,theadaptiveM-QAMOFDMsignalremainstobeagoodchoicetobeincorporatedwithadimmingcontrolscheme,sincetherequiredLEDlamppowerwithoutapplyingdimmingcontrolisabout2Wandtheadaptivesymbolrateisnotlargerthantheoriginalsymbolrate.

3.5 SummaryThischapterdescribesthreerecentlydevelopedapproachestoimprovetheperformanceofvisible light communication systems. Section 3.2 discusses a receiver plane tiltingtechniquetoreducetheSNRfluctuationinaroom.Thisschemecanprovide5.69dBand4.14dBimprovementforthepeak-to-troughSNRperformance,forthecasesofoneLEDlampandfourLEDlamps,respectively.Thecorrespondingmaximumspectralefficiencyimprovementsare0.47bit/s/Hzand0.23bit/s/Hz, respectively.Section3.3describes anLED lamparrangement technique to improveSNRandBERperformances. It is shownthatthearrangementof12LEDlampsinacircleand4LEDlampsatcornerscanattainmuchbetterperformancethanotherpossiblearrangements.Byapplyingthetimedomainzero-forcingequalization,thisLEDlamparrangementisabletoprovidealmostidenticalcommunication qualities in terms of SNR and BER to all users at different positionsthroughouttheroom.InSection3.4,theperformanceofaVLCsystemunderadimmingcontrol scheme isdiscussed for twodifferentmodulation formats,namelyOOKandM-QAMOFDM.The results show that the adaptivedata rateof anOOKsignal is alwayslarger than theoriginaldata rate,while the requiredLED lamppower is less than1W,when theoriginal data rate is 10Mbit/s.The adaptivedata rateof theM-QAMOFDMsignal isnot larger than theoriginal symbol ratewhen thedutycycle is larger than0.4,however,theLEDlamprequiredwithoutapplyingdimmingcontrolisalwayslargerthanthatrequiredbytheOOKsignal.

References[1] T.KomineandM.Nakagawa,“Fundamentalanalysisforvisible-lightcommunicationsystemusingLEDlights,”IEEETransactionsonConsumerElectronics,50,100–107,2004.

[2] D.C.O’Brien,G.Faulkner,K. Jim,et al., “High-speed integrated transceivers foropticalwireless,”IEEECommunicationsMagazine,41,58–62,2003.

[3] J.Vucic,C.Kottke,K.Habel,andK.D.Langer,“803Mbit/svisiblelightWDMlinkbased on DMT modulation of a single RGB LED luminary,” in Optical FiberCommunication/National Fiber Optic Engineers Conference (OFC/NFOEC), 2011,pp.1–3.

[4] H. Elgala, R. Mesleh, and H. Haas, “Indoor broadcasting via white LEDs andOFDM,”IEEETransactionsonConsumerElectronics,55,1127–1134,2009.

[5] J.M.KahnandJ.R.Barry,“Wirelessinfraredcommunications,”IEEEProceedings,85,265–298,1997.

[6] K. D. Langer and J. Grubor, “Recent developments in optical wirelesscommunications using infrared and visible light,” in International Conference onTransparentOpticalNetworks(ICTON),2007,pp.146–151.

[7] S.Hann,J.-H.Kim,S.-Y.Jung,andC.-S.Park,“WhiteLEDceilinglightspositioningsystems for optical wireless indoor applications,” in European Conference andExhibitiononOpticalCommunication(ECOC),2010,pp.1–3.

[8] L. Zeng, D. O’Brien, L.-M. Hoa, et al., “Improvement of data rate by usingequalization in an indoor visible light communication system,” in InternationalConferenceonCircuitsandSystemsforCommunications(ICCSC),2008,pp.678–682.

[9] G.Ntogari,T.Kamalakis,andT.Sphicopoulos,“Performanceanalysisofspacetimeblockcodingtechniquesforindooropticalwirelesssystems,”IEEEJournalonSelectedAreasinCommunications,27,1545–1552,2009.

[10] D. Bykhovsky and S. Arnon, “An experimental comparison of different bit-and-power-allocation algorithms forDCO-OFDM,”JournalofLightwaveTechnology, 32,1559–1564,2014.

[11] D. Bykhovsky and S.Arnon, “Multiple access resource allocation in visible lightcommunicationsystems,”JournalofLightwaveTechnology,32,1594–1600,2014.

[12] G.Cossu,A.M.Khalid,P.Choudhury,R.Corsini,andE.Ciaramella, “3.4Gbit/svisibleopticalwirelesstransmissionbasedonRGBLED,”OpticsExpress,20,B501–B506,2012.

[13] D. Tsonev,H. Chun, S. Rajbhandari, et al., “A 3-Gb/s single-LED OFDM-basedwirelessVLClinkusingagalliumnitrideμLED,”IEEEPhotonicsTechnologyLetters,26,637–640,2014.

[14] T.Fath,M.DiRenzo,andH.Haas,“Ontheperformanceofspaceshiftkeyingforoptical wireless communications,” in IEEE GLOBECOM Workshops (GC Wkshps),2010,pp.990–994.

[15] R.Mesleh,R.Mehmood,H.Elgala,andH.Haas,“IndoorMIMOopticalwirelesscommunication using spatial modulation,” in IEEE International Conference onCommunications(ICC),2010,pp.1–5.

[16] R.Mesleh,H.Elgala,andH.Haas,“Opticalspatialmodulation,”IEEE/OSAJournalofOpticalCommunicationsandNetworking,3,234–244,2011.

[17] T. Fath and H. Haas, “Performance comparison ofMIMO techniques for opticalwireless communications in indoor environments,” IEEE Transactions onCommunications,61,733–742,2013.

[18] http://www.ted.com/talks/harald_haas_wireless_data_from_every_light_bulb.html.

[19] J.Chen,C.Yu,Z.Wang,J.Shen,andY.Li,“Indooropticalwirelessintegratedwith

white LED lighting: Perspective & challenge,” in 10th International Conference onOpticalCommunicationsandNetworks(ICOCN),2011,pp.1–2.

[20] Z.Wang,W.D.Zhong,C.Yu,etal.,“Performanceofdimmingcontrolschemeinvisiblelightcommunicationsystem,”OpticsExpress,20,18861–18868,2012.

[21] Z.Wang,C.Yu,W.D.Zhong,andJ.Chen,“Performance improvementby tiltingreceiverplane inM-QAMOFDMvisible light communications,”OpticsExpress, 19,13418–13427,2011.

[22] Z.Wang,C.Yu,W.D.Zhong,J.Chen,andW.Chen,“PerformanceofanovelLEDlamparrangementtoreduceSNRfluctuationformulti-uservisiblelightcommunicationsystems,”OpticsExpress,20,4564–4573,2012.

[23] Z.Wang,W.D.Zhong,C.Yu,andJ.Chen,“AnovelLEDarrangement to reduceSNR fluctuation for multi-users in visible light communication systems,” in 8thInternational Conference on Information, Communications and Signal Processing(ICICS),2011,pp.1–4.

[24] Z.Wang,C.Yu,W.D.Zhong,J.Chen,andW.Chen,“PerformanceofvariableM-QAMOFDMvisiblelightcommunicationsystemwithdimmingcontrol,”in17thOpto-ElectronicsandCommunicationsConference(OECC),2012,pp.741–742.

[25] Z.Wang,J.Chen,W.D.Zhong,C.Yu, andW.Chen, “User-orientedvisible lightcommunication system with dimming control scheme,” in 11th InternationalConferenceonOpticalCommunicationsandNetworks(ICOCN),2012,pp.1–4.

[26] H.Kressel, Semiconductor Devices for Optical Communication, Springer-Verlag,1982.

[27] J.R.Barry,WirelessInfraredCommunications,KluwerAcademicPublishers,2006.

[28] I. Neokosmidis, T. Kamalakis, J. W. Walewski, B. Inan, and T. Sphicopoulos,“Impact of nonlinear LED transfer function on discrete multitone modulation:Analyticalapproach,”IEEEJournalofLightwaveTechnology,27,4970–4978,2009.

[29] C.H.EdwardsandD.E.Penney,Calculus,PrenticeHall,2002.

[30] M.T.Heath,ScientificComputing–AnIntroductorySurvey,McGraw-Hill,2002.

[31] A.Svensson,“AnintroductiontoadaptiveQAMmodulationschemesforknownandpredictedchannels,”IEEEProceedings,95,2322–2336,2007.

[32] J.Proakis,DigitalCommunications,3rded.,McGraw-Hill,1995.

[33] F.Xiong,DigitalModulationTechniques,2nded.,ArtechHouse,2006.

[34] R.J.Essiambre,G.Kramer,P.J.Winzer,G.J.Foschini,andB.Goebel,“Capacitylimitsofopticalfibernetworks,”IEEEJournalofLightwaveTechnology,28,662–701,2010.

[35] M.Z.Afgani,H.Haas,H.Elgala,andD.Knipp,“VisiblelightcommunicationusingOFDM,”inInternationalConferenceonTestbedsandResearchInfrastructuresfortheDevelopmentofNetworksandCommunities(TRIDENTCOM),2006,pp.6–134.

[36] J.Armstrong, “OFDM for optical communications,” IEEE Journal of Lightwave

Technology,27,189–204,2009.

[37] U. S. Jha and R. Prasad,OFDM towards Fixed and Mobile Broadband WirelessAccess,ArtechHouse,2007.

[38] A.Goldsmith,WirelessCommunications,CambridgeUniversityPress,2005.

[39] H. Sugiyama, S. Haruyama, andM. Nakagawa, “Brightness control methods forilluminationandvisible-lightcommunicationsystems,”in3rdInternationalConferenceonWirelessandMobileCommunications(ICWMC),2007,pp.78–83.

[40] J.-H.Choi,E.-B.Cho,T.-G.Kang,andC.G.Lee,“Pulsewidthmodulationbasedsignal format for visible light communications,” in 15th Opto-Electronics andCommunicationsConference(OECC),2010,pp.276–277.

[41] S. Rajagopal, R. D. Roberts, and S.-K. Lim, “IEEE 802.15.7 visible lightcommunication: Modulation schemes and dimming support,” IEEE CommunicationsMagazine,50,72–82,2012.

[42] G. Ntogari, T. Kamalakis, J. Walewski, and T. Sphicopoulos, “Combiningillumination dimming based on pulse-width modulation with visible-lightcommunications based on discrete multitone,” IEEE/OSA Journal of OpticalCommunicationsandNetworking,3,56–65,2011.

[43] J.ProakisandM.Salehi,ContemporaryCommunicationSystemsUsingMATLAB,PWSPub.,1998.

[44] A.J.GoldsmithandS.-G.Chua,“Variable-ratevariable-powerMQAMfor fadingchannels,”IEEETransactionsonCommunications,45,1218–1230,1997.

4 Lightpositioningsystem(LPS)

MohsenKavehradandWeizhiZhang

One of the most promising applications of visible light communication is indoorpositioning, also referred to as indoor localization. Indoor positioning has manyapplications in real life. For instance, the development of this technique will makepossible accurate location-based services (LBS), which have increased dramaticallytogetherwith thepopularizationofmobilecomputingdevicessuchassmartphonesandtablets. Therefore, it is increasingly attracting researchers’ interest. Current research onvisible light communication (VLC) technology has provided a new approach tocommercialhigh-qualityindoorpositioningsystems.Throughthischapter,wefirstlistalltheapplications,investigatecurrentaccesstotheradiospectrum,andcoverthemeritsofshifting theneedsof indoorpositioning to thevisible lightspectrum.AsurveyofVLC-based indoor positioning techniques and related works follows. Finally, challenges andpotentialsolutionsarediscussed.

4.1 IndoorpositioningandmeritsofusinglightIndoorpositioning is akey technologywhichcanbebeneficial formany industries andcustomers(Fig.4.1).These include locationsensingandmanagementofproducts insidelargewarehouses,indoornavigationservicesforpedestriansinsidelargebuildingssuchasmuseums and shopping malls, location-based services (LBS) and advertisements forconsumersetc.

Figure4.1 (a)Indoornavigationforpedestrians;(b)locationanalyticsbasedonindoorpositioning.

Accordingtoareportby theFederalCommunicationsCommission in2012[1],mostresearchagreesthatthelocation-basedservicesmarketwillbetripledby2015compared

to its size in 2012. The report also stated that Foursquare, a location-based socialnetworking company, had 10million users, but this has been increasing since then andwas40millionin2013[2].ResearchalsopredictsthatthetotalmobileLBSmarketwillbeworthmorethan10billionUSdollarsin2015[3].

What is maybe even more important, with the standardization of femto-cells in themobilenetwork,isthatlocationawarenessofuserswillbecriticaltoaddresshand-offandresourceallocationissues,whicharediscussednext.

4.1.1 IntroductiontoindoorpositioningAlthough theGlobalPositioningSystem(GPS) iswell-developedandperformswell foroutdoorapplications,itisstillverydifficulttouseitindoors(seeFig.4.2)becauseofthepoorcoverageofsatellitesignals.Themainreasonbehindthisisthemultipatheffectsofradio-wavesontheGPSsignals.

Figure4.2 IndoorcoverageproblemsofGlobalPositioningSystem(GPS).

Signals emitted from satellites may reflect off surfaces around the receiver, such astrees, roofs,walls,orevenhumanbodies.Thesignalssubject tomultipathwillarriveatthe receive sensor as delayed components, therefore creating positioning errors. For anindoor environment, this induced inaccuracy is so large that theperformanceofGPS isdowngradedtoanunacceptablelevel.Apartfrommultipath,therearepotentiallyartificialsourcesofinterference,whetherintentionalornot,furtherbringingdownthepositioningaccuracy. As a result of the poor performance of the GPS system, currentlycommercializationofindoorpositioningsystemsisstillinitsinfancyandthereisaneed

todevelopareliableandaccuratesystem.Sofar,differentproposedcandidatesystemstryto perform accurate indoor positioning by utilizing radio-wave, acoustic and opticaltechnologies[4–9].

4.1.2 SpectrumcrunchandfuturemobilesystemWith the increasing popularity of multimedia services supplied over the cellular radiofrequency(RF)networks[10], includingdataservicessuchaswebbrowsing,audioandvideo on demand, it is certainly only a matter of time before users will face extremecongestion while trying to connect to avail themselves of these services. Advances indisplays, battery technology and processing power have made it possible for users toaffordandcarryaroundsmartphonesandtablets.Hence,asweareenteringaneweraofalways-on connectivity, the expectation from users for not only ubiquitous but alsoseamless voice and video services presents a significant challenge for today’stelecommunicationssystems.Theprospectsfor thedeliveryofsuchmultimediaservicesto these users are crucially dependent on the development of low-cost physical layerdeliverymechanisms[11].Itisrecognisedthattheelectromagneticspectrumhasbecomeextremely crowded [12].For thebenefit of obtainingmorebandwidth, the frequency atwhich mobile bands are regulated has already shifted from 450~800MHz in the firstgenerationofcellularsystems,toashighas2.6~2.7GHzintoday’s4Gtechnology.

As frequency increases, the path loss increases, perhaps as frequency squared tofrequencytosomehigherpower,dependingonenvironment.Thereisnocliffasfrequencyincreases;itjustbecomesmoredifficultgraduallytodeliverahighquality-of-serviceovera non-line-of-sight (NLOS) path. This issue becomes less important as the base stationspacing has decreased to increase system capacity and that is the best way to do it.Decrease spacingbya factorof twoandcapacity increasesbya factorof four, i.e. twosquared.Close-spaced base stations need less linkmargin to overcomepath loss so theincreased path loss with higher frequency tends to be offset by the closer spacing.Allocatingmorebandwidthtocellularwouldhelpalittlewiththespectrumcrowding,butprobablynotenoughtotakecareofthemajorissue.Evenifyouhadtwicethebandwidthavailable,thatisasmallhelp.

Therefore,decreasingbasestationspacing(increasingbasestationdensity) isabetterway toprovidemorecapacityand that iswhathasbeendone for the last20years.Thewireless industry is talkingabout femto-cells forahighcapacitybackhauling.However,smaller cells can bring problems that were not noticeable before, e.g. more frequenthandovers,morecomplex resourceallocationand re-use,etc.Precise indoorpositioningtechniqueswillbenefittheoverallnetworkmanagementsincefemto-cellsaredesignedtoserveindoorenvironments.Byusinglocationinformationfromusersthebasestationcanmakebetterfrequencyallocationtovirtuallyprovidemoreavailablebandwidthtousers,as well as better prediction about when handover will happen (Fig. 4.3), to realize aseamlessservice.

Figure4.3 Handoverpredictioninmobilenetworks.

4.1.3 AdvantagesofVLC-basedpositioningAs noted in the previous chapters, VLC technology has great potential and will helpoffload traffic from the highly congested radio-wave band. First, the visible light bandtheoretically provides 400 THz bandwidth (375–780 nm), which is a much largerbandwidththanRFtechniquescanutilize.Moreover,lightwavesintheopticalwavelengthrangeareconfinedtothewallsinaroomandgenerallydonotpenetratesolidmaterials.Hence,practicalandusablenetworkscanbereadilyrealizedthatutilizethisself-limitinglinkdistance.Wecallsuchsystemsthatemploythispropertyhigh-bandwidthislands.Themotivationforoperatorstoactuallychoosetotransferdatathroughthisopticalbandisthatby doing so, the entire huge bandwidth can be re-used next door, free of interference.Optical wireless is a future proof solution since additional capacity far beyond thecapabilitiesofradiocouldbedeliveredtousersastheirneedsincreasewithtime.

Lightemittingdiodes(LEDs)willreplacetoday’sincandescentandfluorescentlampsused for lighting in due course. Compared to traditional light devices, LED has higherpower efficiency, longer lifetime and higher tolerance to environmental hazards. SinceLED is a semiconductor light source, it can be easilymodulated formany applicationsotherthanlighting,suchasindoorbroadbandcommunication,smartlighting,etc.[11,12].Thisfeatureenablesresearchers toexplore thepossibility touse it toaddress the indoorpositioning problem. Recently, VLC using LEDs or other light sources has beenconsideredasoneofthemostattractivesolutionsforindoorpositioningsystemsbecauseofthemanyfeaturesitbrings,asdescribedbelow:

BetterpositioningaccuracyThere aremany proposed indoor positioning solutions based on radio-wave techniques.Wirelesstechnologiesassociatedwiththeminclude,butarenotlimitedto:wirelesslocalarea network (WLAN), radio-frequency identification (RFID), cellular, ultra-wide band(UWB),Bluetooth,etc.[4–7,21].Thesemethodsdeliverpositioningaccuraciesfromtens

ofcentimeterstoseveralmeters[13],seeTable4.1.VLC-basedsystemsareexpected toprovidebetterpositioningaccuracythanradio-wavesolutions,sincetheysufferlessfrommultipatheffectsandinterferencefromotherwirelesshandsets,whichwillbeverifiedinthefollowingsections.

Table4.1 Positioningaccuraciesofradio-wavetechniques[13].

Positioningmethod(technology) Accuracy(m)

SapphireDart(UWB) 0.3

Ekahau(WLAN) 1

TOPAZ(Bluetooth+IR) 2

SnapTrack(AGPS) 5–50

WhereNet(UHFTDOA) 2–3

LANDMARC(RFID) 2

Generatenoradio-frequency(RF)interferenceAside from the relatively lower accuracies radio-wave approaches can provide, the RFelectro-magnetic (EM) interference brought in is always a concern for many indoorenvironments.Ontheonehand,EMradiationgeneratedbythesetechniqueswilloccupythe already-congested limitedmobile band, further degrading the performance of otherwirelessdevices.Ontheotherhand,sinceradio-frequencyinterferencecandisablecertainkindsofmedicaldevices,withresultsrangingfrominconveniencetoaccidental injuries,radio-frequencyradiationisrestricted,orevenprohibitedinhospitals,aswellasinmanyotherplaceshavinginterferenceconcerns.

Incontrast,VLCsystemsforcommunicationorpositioningpurposes,donotgenerateanyRFinterferenceandthereforearesafetouseinsidehealthcarefacilities.LEDscanbeusedtocarrydifferentformsofinformation,e.g.biomedicalinformationfrommonitoringdevices,textinformationfrompatients,etc.tomedicalstaff[14],[15].

Re-usecurrentlightinginfrastructuresPositioningtechniquesbasedonultrasoundandotheracousticwavesofferaccuracyuptoseveralcentimeters[13],however,theyrequireadenseandcalibratedgridoftransmitters,which sometimes may dramatically increase the cost of the system. In contrast, VLC-based systems re-use the current light infrastructures and thus widespread coverage isassured.Atthesametime,littleornorenovationisneededtoprovideaservice,thereforeVLC-based systems also offer very economical solutions for indoor positioningrequirements.

4.2 PositioningalgorithmsInvestigationswithin this field demonstrate that positioning algorithms proposed so farcanbecategorizedintothreetypes:triangulation,sceneanalysis,andproximity.

4.2.1 TriangulationTriangulation is the general name for positioning algorithms which use the geometricpropertiesof trianglesfor locationestimation.Triangulationhastwobranches: laterationandangulation[13]. In laterationmethods, the target location isestimatedbymeasuringits distances frommultiple reference points. For all the VLC-based indoor positioningsystems proposed, the reference points are light sources and the target is an opticalreceiver.Thedistancesarealmost impossible tomeasuredirectly;however, theycanbemathematically calculated from other measurements such as received signal strength(RSS), time-of-arrival (TOA) or time-difference-of-arrival (TDOA).On the other hand,angulationmeasuresanglesrelativetoseveralreferencepoints(angle-of-arrival(AOA)),afterwhich location estimation is carried out by finding intersection points of directionlineswhichareradiifromreferencepoints.

4.2.2 Triangulation–circularlaterationCircularlaterationmethodsmainlymakeuseoftwokindsofmeasurements:TOAorRSS.

As light travelsataconstantspeed inair, thedistancebetween thereceiverand lightsource is proportional to the travel time of the optical signals. In TOA-based systems,time-of-arrivalmeasurementswithrespecttothreelightsourcesarerequiredtolocatethetarget,giving intersectionof threecircles for2-D,or threespheres ina3-Dscenario.AverygoodexampleofTOA-basedsystemsisthewidelyusedGPSsystem.Inthesystem,navigationmessagessentfromsatellitescontaintimeinformation(intheformofarangingcode)andEphemeris(orbitinformationforallthesatellites).Aftersuccessfullyreceivingnavigation messages from more than three satellites, circular lateration (referred to astrilateration) is performed to determine the receiver’s location. However, all the clocksusedbythetransmittersaswellasbythereceiverhavetobeperfectlysynchronized.Forindoorapplications,thepositioningaccuracyshouldrangefromsub-metertocentimeters,whichmeans thedifferentclocks inTOA-basedsystemshave tobesynchronizedat thelevelofafewnanoseconds,orevenhigheraccuracy.Asaresult,thecomplexityandcostof such systems are impracticable. Therefore, research on the TOA-based positioningtechnique has been very limited. Cramer–Rao bound analysis of a TOA-based VLCpositioningsystemconsideringonlyshotnoiseisgivenin[16],andtheresultsshowthataccuracyaround2~5cmcanbeachieved,dependingondifferentsystemsettings.

RSS-based systems calculate the propagation loss the emitted signal has experiencedbased onmeasurements of received signal strength. Range estimation is thenmade byusing a proper path loss model. As in TOA-based systems, estimation of the target’spositionisobtainedbycircularlateration,showninFig.4.4.Becauseoftheavailabilityofline-of-sight (LOS) channels for most indoor environments, it is considered that RSS-based methods using VLC will deliver a good performance. According to simulationsperformedinpreviouswork[17],thetargetcanbelocatedwithapositioningerroraround

0.5mm.In[18],takingtherotationandmovingspeedofthereceiverintoconsideration,the authors show that an overall accuracy around 2.5 cm can be obtained, assuming atypicalmovingspeedofthereceiver.In[19],theauthorsaccountforpossibleinstallationerrorsoftheLEDlightingbulbsaswellastheorientationangleofthetargetandshowthata precision of 5.9 cm with 95% confidence can be expected, given indirect sunlightexposureandproperinstallationoftheLEDbulbs.

Figure4.4 Positioningusingcircularlateration.

Now let us introduce a mathematical expression for circular lateration in two-dimensionalspace.Theexpressionforthree-dimensionsissimilar.Let(Xi,Yi)representtheposition of the ith transmitter (reference point) on a two-dimensional plane and (x,y)denote the position of the receiver (target). If the measured distance between the ithtransmitterandreceiver isRi, theneverycircleas shown inFig.4.4 isa setofpossiblelocationsofthereceiverdeterminedbyasinglerangemeasurement,whichis:

(4.1)

wherei=1,2,…,nandn is thenumberof transmitters (referencepoints) involved inrangemeasurements.Theoretically,ifrangemeasurementsarenoise-free,theintersectionof circlesgivenbyEq. (4.1) should yield the position of the receiver as a single point.However,thiscannotberealisticinrealmeasurements.NoisyrangemeasurementsleadtomultiplesolutionstooursystemdescribedbyEq.(4.1).Inthisscenario,theleastsquaressolution,whichisdiscussedin[20,21],providesastandardapproachtotheapproximatesolutionofthispositioningsystem.

Notethat:

(4.2)

(4.3)

wherei=1,2,…,n.Thenwecanrewritetheequationsdescribingthesystemintomatrixform[21]:

AX=B, (4.4)

where

X=[xy]T (4.5)

(4.6)

and

(4.7)

Thentheleastsquaresolutiontothesystemisgivenby:

X=(ATA)−1ATB. (4.8)

4.2.3 Triangulation–hyperboliclaterationHyperbolic lateration methods are usually associated with TDOA measurements. InTDOA-basedsystems,lightsignalsfromdifferentLEDsaredesignedtobetransmittedatthesameinstant.ThiscanbeeasilyachievedsincealltheLEDsareincloseproximitysotheycansharethesameclock.Thereceivermeasuresthedifferenceintimeatwhichthesesignalsarrive.Ontheotherhand,thereceiverdoesnotneedtobesynchronizedwiththetransmitters,sinceitisnottryingtoextracttheabsolutetimeofarrivalinformation.

AsinTOAandRSSbasedsystems,threelightsourcesareneededtoenable2-Dor3-Dpositioning. Because a single TDOA measurement with two light sources involvedprovides a hyperboloid on a 2-D plane or a hyperbola in a 3-D space, two TDOAmeasurementsarerequiredtolocatethetargetbyusinghyperboliclateration(referredtoasmultilateration). Instead of taking TDOAmeasurements directly, wemay take othermeasurements, and compute TDOA information by using them. In [22], sinusoidalcomponents of signals emitted from two LEDs generate an interference pattern at thereceiverbecausetheyhavethesamefrequency.Thus,thesinusoidalpeak-to-peakvalueofthereceivedsignalcanbeutilizedtogetequivalentTDOAmeasurements.In[23],TDOAinformationisobtainedbydetectingphasedifferencesbetweenthreesignalswithdifferent

frequencies.Computersimulationsshowanoverallpositioningaccuracyof1.8mm.

For mathematical expression of hyperbolic lateration in two-dimensional space,followingthenotationusedforcircularlateration,everyhyperbolaasshowninFig.4.5isa setofpossible locationsof the receiverdeterminedbya singlemeasurementof rangedifference.Everyhyperbolacanberepresentedby:

(4.9)

whereDijdenotesthedifferencebetweentherangesRiandRj,withrespecttotheithandjthreferencepointsandi≠j.Notethat:

(4.10)

(4.11)

wherei=1,2,…,n.Wecanthenrewritetheequationsdescribingthesystemintomatrixform[21]:

AX=B, (4.12)

where

X=[xyR1]T, (4.13)

(4.14)

and

(4.15)

Thentheleastsquaressolutiontothesystemisgivenby:

X=(ATA)−1ATB. (4.16)

Figure4.5 Positioningusinghyperboliclateration.

4.2.4 Triangulation–angulationIn AOA-based systems, the receiver measures angles of arriving signals from severalreferencepoints.The target’s location is thendeterminedas the intersectionofdirectionlines (Fig.4.6).Typically two light sourcesareneeded to realize2-Dand three for3-Dpositioning. Interestingly, we can find a historical trace back to older photogrammetrytechniques[24]withmanysimilarities.ThegreatestadvantageofAOA-basedsystemsisthatnotimesynchronizationisneeded.Anotheradvantageisthatitisrelativelyeasiertodetect theAOAof the incomingsignals in theopticaldomainwithan imagingreceiver,compared to employing complex antenna arrays used in radio-wave approaches. Thewidelydeployed front-facingcameras,whichare inherently imaging receivers,onsmartphonesand tabletshavebroughtopportunities for thismethod tobecomepracticableonmobileconsumerelectronics.However,toachieveagoodperformanceforarealsystem,the lighting infrastructuremay have to be adjusted sincemost of these cameras have avery limited field-of-view(FOV).Moregenerally, thepositioningaccuracyofanAOA-basedsystemwilldowngradewhenthetargetgetsfartherfromthelightsourcesduetothelimited spatial resolution of the imaging receivers. A positioning accuracy of 5 cm isreportedwhenusinganimagingreceiverwitharesolutionof1296×964pixels[25]. In[26], researchers proposed a two phase positioning algorithmwhichmakes use of bothRSS and AOA measurements, taking multipath effects due to reflections intoconsideration.Computersimulationsshowedthatamedianaccuracyof13.95cmcanbeachieved.

Figure4.6 Positioningusingangulation.

Toobtain the least square solution foranAOA-basedsystem, supposeαidenotes themeasuredangle-of-arrivalwithrespecttotheithtransmitter,whichisgivenby:

(4.17)

wherei=1,2,…,n.Thenwewillhave,

(x−Xi)sinαi=(y−Yi)cosαi. (4.18)

Thenwecanrewritetheequationsdescribingthesystemintomatrixform[21]:

AX=B, (4.19)

where

X=[xy]T, (4.20)

(4.21)

and

(4.22)

Thentheleastsquaresolutiontothesystemisgivenby:

X=(ATA)−1ATB. (4.23)

4.2.5 SceneanalysisSceneanalysisreferstopositioningalgorithmswhichmakeuseoffingerprintsassociatedwitheveryanchorpointinthesysteminsideascene,asshowninFig.4.7.The target isthen located by matching real-time (on-line) measurements to these fingerprints.Measurements that can be used as fingerprints include all measurements mentionedearlier, i.e.TOA,TDOA,RSSandAOA.RSSis themostusedformoffingerprint.Thetimerequiredtomatchfingerprintsisusuallyshorterthanperformingatriangulation,thussaving a lot of time and powerwhichwill otherwise be used for computing.However,scene analysis solutions also have a significant disadvantage. They cannot be deployedinstantly insideanewscenario sinceaccurate systempre-calibration isneeded.A sceneanalysismethodutilizingRSSinformationfromfourLEDlightsasfingerprintsshowsthatapositioningaccuracyof4.38cmcanbeattained[27].

Figure4.7 Positioningsystemusingsceneanalysis.

4.2.6 ProximityProximity-based systems (Fig.4.8) rely on a dense grid of light sources, eachhaving aknownpositionandauniquehardwareID.Whenthetargetreceivessignalsfromasinglelightsource,itisconsideredtobeco-locatedwiththesource.Whensignalsfrommultiplesources are detected, averaging will be performed. Proximity systems using VLCtechnology theoreticallyprovideaccuraciesnohigher than the resolutionof the lightinggriditself.Notethatwhendensegridsareemployed,anarrowilluminationbeamfromthelight sources is required in order to prevent interference and impaired locationdetermination. In [9], theauthorsexperimentallydemonstratedaproximity-based indoorpositioning system. The positioning is provided by visible light LEDs,while a ZigBeewirelessnetworkisemployedtosendthelocationinformationtothemainnode,extendingtheworkingrangeofthesystem.

Figure4.8 Proximitypositioningsystem.

4.2.7 ComparisonofpositioningtechniquesTocomparethesystemswehavementioned,weadoptthefollowingperformancemetrics:accuracy,spatialdimension,andcomplexity,seeTable4.2.

Table4.2 Comparisonofsolutionsmentioned.

Referenceno.

Positioningalgorithm Accuracy Space

dimensionComplexityHardware/Algorithm Comments

17 RSS 0.5mm(Simulation)

2-D/3-D Moderate/Moderate 3-Dpositioningisrealized

18 RSS 2.5cm(Simulation)

2-D Moderate/Moderate Rotationandmovingspeedofthereceiverareaccountedfor

19 RSS 5.9cm(Simulation)

2-D Low/Moderate Asynchronoussystemdesign,installationerrorsofLEDbulbsandorientationangleofthetargetaretakenintoconsideration

23 TDOA 1.8mm(Simulation)

2-D High/Moderate Frequency-division-multi-access(FDMA)

protocolemployed

25 AOA 4.6cm(Experiment)

3-D High/Moderate DifferentcolorsareusedtodistinguishLEDlights

26 RSS+AOA 13.95cm(Simulation,medianvalue)

2-D/3-D High/High HighLambertianordersourcesareassumed(mmultipatheffectsandreceiverorientationaretakenintoconsideration

27 Sceneanalysis(RSSbased)

4.38cm(Experiment,meanvalue)

2-D Moderate/Low Pre-calibrationneeded

9 Proximity Roomlevel(Experiment)

2-D Low/Low 4MHzcarrierisemployedtoachievebetteropticaldetection

Accuracy is usually referred to as the mean value of positioning error. Here, thepositioning error is defined as the Euclidean distance between the real location of thetargetandtheestimationofit.Thesmallerthemeanvalueis,thehigheraccuracyasystemcanachieve.Sonaturallythisisthemostimportantfactorwhenweevaluateapositioningsystem.

Spatialdimensionisthenumberofdimensionsapositioningsystemisabletoprovideon location information.Many proposed solutions are only capable of 2-D positioninghorizontally,inwhichcaseiftheheightofthetargetchanges,thepositioningaccuracyofthesystemwilldropasaresult.Systemscapableof3-Dpositioning,on theotherhand,provideabetterperformanceinthisscenario.

Thecomplexitydefinedinthischaptercontainstwocomponents.Thefirstisthesystemrequirementonhardware,i.e.howmanydevicesareinvolvedandhowcomplicatedistheoverallsystemconfiguration.Hardwarecomplexitymainlydetermineswhatwouldbethe

deployment cost of an indoor positioning system. The other component is thecomputationalcomplexityofthepositioningalgorithm,orthedelaybeforethesystemcanupdatethecurrentlocationofthetarget.Inmostofthesystems,dataprocessinghappenson the target side, e.g.mobile handsets in real scenarios.Algorithm complexity is alsoimportant,becausebatterylife isstillahugeconcernforhandsets,eventhoughtheyarecurrentlycapableofpowerfulcomputing.

4.3 Challengesandsolutions

4.3.1 MultipathreflectionsAsmentioned earlier, a path lossmodel is employed in anRSS-based system.Currentopticalpathlossmodels,whichonlytaketheLOSpropagationintoconsideration,donotalwaysholddue tomultipatheffects causedby reflectionsof lightondifferent surfaces(e.g. walls). The optical energy contained within these reflected components would bedirectly translated into extra positioning errors, thus positioning accuracy will bedowngraded.

Oneavailable solution is tousea fly-eye receiver [28].As an imaging receiver, it isabletoresolvelightcomponentscomingfromdifferentdirections.Therefore, theeffectsof reflections can be significantly reduced by necessary signal processing after opticaldetection.Furthermore,thereceivercanprovideangleinformationwhichcanbeusedtoimprovethepositioningaccuracyincombinationwithAOApositioning.

4.3.2 SynchronizationIn the systemsbased on timemeasurements (TOA,TDOA), synchronization is amajorsource of positioning error. In TOA-based systems, it is extremely difficult to get alltransmitters and the receiver precisely synchronized, if not impossible. This becomeseasier inTDOA-based approaches.The clock used by the receiver does not have to bepreciselysynchronizedwiththetransmittersandwecanmakeuseofothermeasurementsasmentionedin[19,20].Nevertheless,thesignalsfromalltransmittershavetobeemittedsimultaneously.Otherwise,theinitialphasesoftransmittedsignalsmustbemeasuredinaveryaccurateway.Therefore,synchronizationhastobeperformedbetweentransmitters,which can lead to potentially high system deployment costs and limited applicablescenarios.

4.3.3 Channelmulti-accessAsimpliedbytheprinciplesofthepositioningalgorithmsmentionedabove,inallsystemsexcept proximity-based ones,multiple transmitters are required to estimate the receiverposition. This means we have to solve the channel multi-access problem to avoidinterference. Even for proximity-based systems, the problem has to be handled tominimizethepossibilityofdetectionfailurewhenmultipletransmittedsignalsfallinthereceiver field-of-view, if no channelmulti-access protocol is employed and transmitterskeepsendingoutsignalsallthetime.

InGPS systems, channelmulti-access is addressed by using the code-division-multi-access (CDMA) technique [29]. In aTDOA-based systemproposed in [23], frequency-division-multi-access (FDMA) is selected as a solution. Time-division-multi-access(TDMA)isanothertechniqueusedbymanysystems;itrequiressynchronizationbetweentransmitters, resulting in higher deployment cost and time. In [19], a new protocol isproposed to eliminate the need for synchronization. The adoption of the framed slottedALOHAprotocolsuccessfullyenablestheLEDstoworkasynchronously,thusnophysicalconnectionsbetweenthemarerequired,loweringsystemcomplexityandcost.

4.3.4 ServiceoutageOneveryrealisticproblemwithlightpositioningsystemscanbeexpressedintheformofa simple question: What will happen when there is no light? Indeed this is a seriousconcernsincetheremightbeoccasionswhereusersdemandapositioningservicewithoutthe lights being turned on. Also, light can be blocked by furniture and human bodies,resultinginserviceoutage.Therefore,thisproblemmustbeaddressedtoensurethatuserswillhaveaccesstothepositioningservicemostofthetime.

Twoapproachescanofferhelpregardingthisissue.First,infraredlaserorLEDcanbeintegratedintotheLEDbulbsusedinthefuturepositioningsystem,tocontinueofferingaservice when visible light illumination is no longer desired. Secondly, to eliminate thetemporaryserviceoutagemainlyduetoobstruction,sensorfusiontechnology[30]makinguseoftheinertialsensorsinsidethehandsetcanbeemployed.Itwillhelpthesystemtogive the best performance by integrating inertial navigation, which will guide the userthrough the obstruction period, and optical positioning, which generally gives moreaccurateestimation.

4.3.5 PrivacyAlthoughitdoesnotseematechnicalissue,privacyisahugeconcernforlightpositioningsystems. Indeed, better positioning accuracy makes people more worried about thesecurityof the real-time location information. In thevisionary report from theFCC[1],the committee identified several privacy issues as follows: notice and transparency,meaningfulconsumerchoice,thirdpartyaccesstopersonalinformation,anddatasecurityand minimization. So we suggest that research on access control and encryption isnecessaryforthelong-termdevelopmentoflightpositioningsystems.

4.4 SummaryIndoor positioning is one of the most important and promising applications associatedwith visible light communications technology. We have specifically discussed whyaccurate indoor positioning plays a key role in futuremobile networks, considering thefact that themobilebandiscurrentlyheavilycongested.Opticalpositioningsystemsareexpectedtoprovidebetteraccuracycomparedtoradio-waverivals.Theywillnotgenerateradio-frequency(RF)interference,maintainingthecurrentperformanceofotherwirelesshandsetsaswellasmakingthesesuitabletobedeployedinsideRFrestrictedorprohibitedenvironments.Finally,indoorpositioningsystemsbasedonVLCtechnologywillbeable

toprovideserviceswhereverLEDsaredeployedatthepriceofaminimaloverhead.

Lightpositioningsystemswillnotonlybenefittherelatedindustrybutalsoagreatmassofconsumers.Therefore,extensiveresearchhasbeendonewithinthisfield.Thedifferentpositioningalgorithmsproposedhave inherentadvantagesanddrawbacks,and thereforetradeoffscanbemadefordifferentapplications.Wehaveprovidedadetailedcomparisontablebasedonthreemetrics:accuracy,spatialdimension,andcomplexity.

We have also identified challenges for the commercialization of light positioningsystems, which include but are not limited to: multipath reflection, synchronization,channelmulti-access,serviceoutage,andprivacy.Possiblesolutionsareprovidedherein.

Acknowledgment

The authors sincerely thank the National Science Foundation (NSF) for its greatcontributiontowardthewritingofthischapter;NSFAwardno.IIP-1169024,IUCRConopticalwirelessapplications.

References[1] Federal Communications Commission, “Location-based services: An overview ofopportunities and other considerations,” [available online]:http://transition.fcc.gov/Daily_Releases/Daily_Business/2012/db0530/DOC-314283A1.pdf

[2] Foursquare,“AboutFoursquare,”https://foursquare.com/about,2013.

[3] PyramidResearch, “Navigation providers try to find theirway,” [available online]:http://www.pyramidresearch.com/points/print/110624.htm

[4] C. Wang, C. Huang, Y. Chen, and L. Zheng, “An implementation of positioningsysteminindoorenvironmentbasedonactiveRFID,”inIEEEJointConf.onPervasiveComputing,pp.71–76,2009.

[5] J.Zhou,K.M.-K.Chu,andJ.K.-Y.Ng,“Providinglocationserviceswithinaradiocellularnetworkusingellipsepropagationmodel,” in IEEE19th Int.Conf.AdvancedInformationNetworkingandApplications,pp.559–564,2005.

[6] Y.LiuandY.Wang,“AnovelpositioningmethodforWLANbasedonpropagationmodeling,” in IEEE Int.Conf.Progress in Informatics andComputing, pp. 397–401,2010.

[7] L.SonandP.Orten,“EnhancingaccuracyperformanceofBluetoothpositioning,”inIEEEWirelessCommunicationsandNetworkingConf.,pp.2726–2731,2007.

[8] H.Schweinzer andG.Kaniak, “Ultrasonic device localization and its potential forwireless sensor network security,” Control Engineering Practice, 18, (8), 852–862,2010.

[9] Y.U.LeeandM.Kavehrad,“TwohybridpositioningsystemdesigntechniqueswithlightingLEDsandad-hocwirelessnetwork,”IEEETrans.Consum.Electron.,58, (4),1176–1184,2012.

[10] “Ciscovisualnetworking index:Globalmobiledata traffic forecastupdate,2009–

2014,” [available online]:http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-520862.html

[11] M.Kavehrad,“Sustainableenergy-efficientwirelessapplicationsusinglight,”IEEECommunicationsMagazine,48,(12),66–73,2010.

[12] M. Kavehrad, “Optical wireless applications: A solution to ease the wirelessairwaves spectrum crunch,” in SPIE OPTO Int. Society for Optics and Photonics,pp.86450G–86450G,2013.

[13] H.Liu,H.Darabi,P.Banerjee, and J.Liu, “Surveyofwireless indoorpositioningtechniques and systems,” IEEE Trans. Syst., Man, Cybern. C, Appl. Rev., 37, (6),pp.1067–1080,2007.

[14] H.Hong, Y. Ren, and C.Wang, “Information illuminating system for healthcareinstitution,” in the 2nd IEEE Int. Conf. on Bioinformatics and Biomedical Eng.,pp.801–804,2008.

[15] Y.Y.Tan,S.J.Sang,andW.Y.Chung,“RealtimebiomedicalsignaltransmissionofmixedECGsignalandpatientinformationusingvisiblelightcommunication,”in35thAnn.Int.Conf.of theIEEEEngineering inMedicineandBiologySociety,pp.4791–4794,2013.

[16] T.Q.Wang,Y.A.Sekercioglu,A.Neild,andJ.Armstrong, “Positionaccuracyoftime-of-arrivalbasedrangingusingvisiblelightwithapplicationinindoorlocalizationsystems,”J.Lightw.Technol.,31,(20),3302–3308,2013.

[17] Z. Zhou, M. Kavehrad, and P. Deng, “Indoor positioning algorithm using light-emitting diode visible light communications,” J. of Opt. Eng., 51, (8), 085009-1–085009-6,2012.

[18] Y.Kim, J.Hwang, J. Lee, andM. Yoo, “Position estimation algorithm based ontrackingofreceivedlightintensityforindoorvisiblelightcommunicationsystems,”inIEEE3rdInt.Conf.onUbiquitousandFutureNetworks,pp.131–134,2011.

[19] W. Zhang, M. I.S. Chowdhury, and M. Kavehrad, “An asynchronous indoorpositioning system based on visible light communications,” J. of Opt. Eng., 53, (4),045105-1–045105-9,2014.

[20] A.Küpper,Location-BasedServices:FundamentalsandOperation,JohnWileyandSons,2005.

[21] A. Kushki, K. N. Plataniotis, and A. N. Venetsanopoulos, WLAN PositioningSystems: Principles and Applications in Location-Based Services, CambridgeUniversityPress,2012.

[22] K.PantaandJ.Armstrong,“IndoorlocalisationusingwhiteLEDs,”Electron.Lett.,48,(4),228–230,Feb.2012.

[23] S.Jung,S.Hann, andC. Park, “TDOA-based opticalwireless indoor localizationusingLEDceilinglamps,”IEEETrans.Consum.Electron.,57,(4),1592–1597,2011.

[24] http://en.wikipedia.org/wiki/Photogrammetry

[25] T.Tanaka andS.Haruyama, “New position detectionmethod using image sensorandvisiblelightLEDs,”inIEEE2ndInt.Conf.onMachineVision,pp.150–153,2009.

[26] G.B.PrinceandT.D.C.Little,“AtwophasehybridRSS/AoAalgorithmforindoordevice localization using visible light,” in IEEEGlobal Commun. Conf., pp. 3347–3352,2012.

[27] S.Y.Jung,S.Hann,S.Park, andC.S.Park, “Opticalwireless indoor positioningsystemusing light emitting diode ceiling lights,”Microwave andOptical TechnologyLetters,54,(7),1622–1626,2012.

[28] G.YunandM.Kavehrad,“Spotdiffusingandfly-eyereceiversforindoorinfraredwirelesscommunications,” inProc.of IEEEInt.Conf.onSelectedTopics inWirelessCommun.,pp.262–265,1992.

[29] http://en.wikipedia.org/wiki/Global_Positioning_System

[30] W. Elmenreich, “Sensor fusion in time-triggered systems,” PhD thesis, ViennaUniversityofTechnology,2002.

5 Visiblelightpositioningandcommunication

ZhengyuanXu,ChenGongandBoBai

5.1 IntroductionPositioningwithhighaccuracyandreliablereal-timeperformanceisanurgentneedandhas become one of the most exciting features of the next generation wireless systems[1–4]. The global positioning system (GPS) suffers from poor performance in certainareas, for example, in indoor environments, due to the strong absorption of the carrierwave by building materials [5], and in urban environments due to link blockage ormultipath interference by tall buildings. In such application scenarios, a visible lightpositioning system can helpmobile users to obtain their real-time position information,based on the signal sources from local light infrastructures not being affected by thecomplicatedcommunicationenvironment.

5.1.1 IndoorlightpositioningsystemIn indoor environments, assisted by indoor positioning systems, various location-basedservicescanberealized,suchasnavigationandguidanceinshoppingmalls,supermarkets,museums,andhospitals;trackingandmonitoringofvaluableequipmentsinfactories;andimprovementofwirelesspersonalnetworkperformancethroughnetworkserviceadaption.Moreover,multipleoptionaltechniquescouldbeadoptedtoimplementindoorpositioning.OnestraightforwardsolutionistoutilizeexistingWiFiaccesspointsforpositioningortoinstallextrapointstoensurewideanddensecoverage[6,7].Thisareaisstillunderactiveresearch. Another approach is to rely on a densely distributed indoor lightinginfrastructurewhichcanserveaslocationsensingunits,suchasconventionalfluorescentand incandescent lamps.Recently, highly energy-efficientwhiteLEDshave emerged asappealing replacements. Similarly to light positioning systems (LPS) using fluorescentlight[8–10],LPSwithwhiteLEDscoupledwithaninexpensiveimagingsensorsuchasisembedded in a mobile handset [11, 12] is an emerging technology. It can provideconcurrentindoorpositioningandillumination.ThelightsignalemittedfromamodulatedwhiteLED,carryingLEDpositioninformation,isreceivedbyaphotodetectorthroughavisiblelightcommunication(VLC)channel,andthenthepositionofthephotodetectorisestimatedbasedon the received signal attributes suchas amplitude andangleof arrival(AOA). If a receiver is equipped with an image sensor instead of a conventionalphotodiodeorarrayedphotodiodes,spatialrejectionofinterferencelightfromotherlightsourceshelpstoincreasecommunicationsignaltonoiseratio(SNR)[13,14].Meanwhile,duetoalargenumberofsmallsizepixels,highspatialresolutionisachievable,leadingtoapossibly lowcost andhighlyaccurateLPS.SinceanLPSoperates in thevisible lightspectrum,itdoesnotcreateelectromagneticinterferencewithexistingRFsystemscriticalforRFrestrictedenvironmentssuchashospitals;itisalsoimmunetoanyRFinterferencefromWiFiorcellularsystems.

AnLPSusingfluorescentlighttypicallycanachieveanaccuracyofseveralmeters[11]orlessthanonemeter[15,16]inasmallservicearea.Thepositioncalculationalgorithms

requiretheincidentlightsignals’horizontalangle,verticalangleandrotationangle,whichareusuallydifficulttomeasureinpractice.Onthecontrary,anLPSwithwhiteLEDsanda camera could achieve a higher accuracy and much wider coverage area of severalmeters.AnindoorpositioningsystemwithwhiteLEDsandanimagesensorisproposedin[17],but two specificallyplaced image sensors are requiredand imagesof at least fourLEDs are supposed to be captured. These requirements raise the system cost andcomputationcomplexity.Receivedlightintensity-basedpositioningmethodsareproposedin [18, 19], pre-calculated incident light intensity at a given position and at least threewhiteLEDswithoutinter-cellinterferenceneedtobeadoptedforthetrilaterationmethod.Also theunpredictable light sourcesmaybringunavoidable intensity fluctuation,whichwouldmakethesetwosystemslessreliable.

5.1.2 OutdoorlightpositioningsystemInoutdoorenvironments,theprecisepositionofavehicleisurgentlyneededtoprovideareliablenavigationserviceinsomedowntownandurbanlocationswherethemostwidelyused GPS has a poor performance [20] due to link blockage or multipath. Effectivemethodstoimprovedriversafetyandrealizeanintelligenttransportationsystem(ITS)arenecessary.Visionbasednavigationmethods[21,22]andLED/monocularcameramethods[23]havebeenproposedtoachievetherequiredrangefinding.WithmoreandmoreLED-basedtrafficlightswidelydeployedintherealworld, trafficlight-basedlocationbeaconand image processing methods [12, 24–27] have also been proposed to obtain vehicleposition information. All these methods need an expensive high speed camera andcomplex image processing procedure to generate the position of a vehicle. Recently,Roberts et al. have proposed a vehicle tail lights based method for a light positioningsystem[28],butitcanonlygeneratetherelativepositionofavehicletotheoneinfrontofitundersomerestrictiveassumptions.

To cater formore general indoor and outdoor application environments, this chapterpresentsanindoorLPSandoutdoorLPSsystem.FortheindoorLPS,wecombineVLCwithpositionestimationmethods,andletwhiteLEDsemittheirlocation-relevantsignalsto be captured by a receiving camera. Then an optimal estimation algorithm isimplemented at the receiver which can provide an unbiased estimate of the cameraposition. For the outdoor LPS system, the light signal emitted from the traffic light,carryingitspositioninformation,isreceivedbytwophotodiodesmountedinthefrontofavehiclethroughavisiblelightcommunication(VLC)link.Thenthepositionofthevehiclecan be estimated based on the received position information of the traffic light and thetimedifferenceofarrival(TDoA)ofthelightsignaltothetwophotodiodes.MethodsforLPSwithonetrafficlightandtwotrafficlightsarereviewed,andthepositioningerrorduetonon-coplanareffectsandtherelatedcoplanarrotationmethodsforbothcasesarealsodiscussed. To simplify the analysis, the blocking effect of the surroundings, andbackgroundlightnoiseandmultipatheffectintheVLClinkareneglected.

5.2 Indoorlightpositioningsystemsbasedonvisiblelightcommunicationandimagingsensors

There are several feasible indoor positioning solutions, based on third generation (3G)systems,WiFi,andtheUWB.However,themultiplesignalreflectionsfromsurroundingobjects cause multipath distortion resulting in uncontrollable errors. These effects maydegrade all known solutions for indoorpositioningwhichuse electromagneticwaves tocarrysignalsfromindoortransmitterstoindoorreceivers.HerewetakeanotherapproachthatrecordsthepositionsoftheLEDimages,asdescribedbelow.

5.2.1 SystemdescriptionAssumethatacameraisusedasbothanimagesensortocapturetheimagepointsofthewhiteLEDs on the ceiling aswell as a communication receiver to capture the positioninformationof thewhiteLEDscarriedby the lightcommunicationsignal.According toNewton’slawoflensimaging[29],namelytherelationshipofthewhiteLEDsandimagepoints on the image sensor of the camera, the position of the camera could be easilyestimated.CaseswithgivenwhiteLEDpositions in thenavigation frameandestimatedpositionsobtainedbyaVLClinkarebothcovered.ConsideringVLCchannelnoise[30],cameranoise [13,31], and the attenuation effect of the signal through propagation, thewhite LED image point on the image sensor may have a random bias and even theposition information of the white LED may be misinterpreted at the receiver [32].Followingtheconceptoflocalizationundernoisyobservations[33],thecurrentestimationproblem can be modeled as an optimal linear combination of noisy measurements.Performanceofthemeansquareerror(MSE)isstudiedbasedontheCramer–Raolowerbound(CRLB)asabenchmarkmeasurement.

InatypicalLPSwithawhiteLEDandcamera,thewhiteLEDmountedontheceilingcan emit a light signal carrying its position information, usually a unique identity (ID)fromwhich thewhiteLED’spositioncanbeobtained, suchas the floornumberand itslocationwithinafloor.Animagesensor-basedcameraassembledinaportabledevicelikeacellphoneandpersonaldigitalassistant (PDA)couldcapture thesignal light intensityvariation where the white LED’s position information in the navigation frame resides.Furthermore, the white LED can also be captured by an image sensor as a light pointthroughaproperlyconfiguredimagelens,increasingtheFOVofthecameraatthesametime.Aproperopticalfiltermayalsobeusedtoreducethereceivedambientlightinthebackground.

5.2.2 LPSwithknownLEDpositionsConsider an LPS with a camera and N white LEDs mounted on the ceiling whosepositionsinthenavigationsystemareknownasSi=[Xi,Yi,Zi]T,i=1,2,N,showninFig.5.1.ThewhiteLEDimagepoints arecapturedontheimagesensorwiththecenterofthesensorsetastheoriginoftherelativecoordinatesystem.Thefocallengthoftheimagelensisfixedasf,thedistancefromthelensofthecameratotheimagesensorplaneisD,andfromthelenstothefloorisZC.Letusfocusonthe2-dimensionalpositionvectorpC=[XC,YC]TofthecamerapositionrepresentedbythecenteroftheimagelensO,whichisthesameasthepositionoftheimagesensorcenterinthenavigationframe.

Figure5.1 LPSwithmultiplewhiteLEDsandacamera.

NoiselessmeasurementAssume theheightZC of the camera (defined as theheight of the image lens) is given.Accordingto thehomothetic triangle theoryandNewton’s lawof lens imaging[29],wecaneasilyobtainthepositionofthecamerapCi=[XCi,YCi]T frompairsofwhiteLEDSianditsimage ontheimagesensoras

pCi=[XCi,YCi]T=[Xi+λixmi,Yi+λiymi]T, (5.1)

where

(5.2)

and[xmi,ymi]TisthemeasuredpositionofthewhiteLEDimagepoint ontheimagesensor.

Notethat,intheabsenceofnoise,Nvalues(1≤i≤N)ofpCiobtainedfromNwhiteLEDswillbeidenticaltopC=[XC,YC]T.

NoisymeasurementNextweconsiderthepresenceofmeasurementnoisefor[xmi,ymi]T,where thenoise for

xmiandymisatisfiestheGaussiandistribution and ,respectively,for1≤i≤N.Notethatthisisessentiallyaparameterestimationproblemundervariousnoisyobservations,whichcanbeperformedviaoptimallinearcombination.

WeconsiderlinearcombiningtoestimatepC=[XC,YC]TbasedontheNmeasurementresultspCifor1≤i≤N,givenasfollows:

(5.3)

where βxi for 1≤ i≤N are linear combining coefficients. First, note that is anunbiasedestimatorofXC,since

(5.4)

Then,itisdesiredtofindtheoptimallinearcombinationcoefficientsβxi,1≤i≤N, thatminimizetheestimationvariance.Accordingto(5.3),theestimationvarianceisgivenasfollows:

(5.5)

Thefollowingmanipulationaimstofindtheoptimalvaluesof .

Notethat

(5.6)

AccordingtotheCauchy–Schwarzinequality,wehavethat

(5.7)

wheretheequalityisachievedifthefollowingissatisfied,

(5.8)

Then,anoptimalsolution to the linearcombinationcoefficientsβxi, denotedas ,andtheminimumvariancearegiven

(5.9)

Similarly,YCcanbeestimatedviacomputingthefollowingunbiasedestimatorbasedonlinearcombination,

(5.10)

withtheoptimalcombinationcoefficientsandtheminimumvariancegiven

(5.11)

Cramer–Raolowerbound(CRLB)CRLBprovidesa lowerboundon thevarianceofanyunbiasedestimator[34],and thuscan serve as a benchmark for the performance of an unbiased estimator. An unbiasedestimatorisoptimalifitcanachievetheCRLB.InthefollowingweshowthattheaboveoptimallinearcombinationindeedachievestheCRLB.

SinceCRLBprovidesalowerboundonthevarianceofanestimatorforadeterministicparameter, an unbiased estimator which achieves this lower bound is efficient. Morespecifically,wehavethefollowingjointdistribution:

(5.12)

Wehave

(5.13)

andthustheCRLBforXCisgivenby

(5.14)

Similarly,wehavethefollowingCRLBforYC,

(5.15)

It isseen that thevarianceof the linearcombinationcanachieve theCRLB,and thus isefficient.

AsimpleexampleoftheCRLBWe provide a simple example of the CRLB of the position estimation. Assume equalestimationvarianceforallLEDsinbothdimensionsxandy,i.e., forall1≤i≤N;andλi=λforall1≤i≤N.ThentheCRLBsforthepositionsXCandYCcanbesimplifiedasfollows,

(5.16)

TheresultscanbejustifiedbythefactthattheaverageoverNmeasurementsreducestheestimationvariancebyafactorofN.

5.2.3 Monte-CarlosimulationresultsAssumefourLEDswithpositions(1,1),(−1,−1),(1,−1),and(−1,1),andonecamerawithposition(0,0).AssumetheheightofthefourLEDstothecameraZi−ZC=3forall1≤i≤4;andthedistancebetweenthelensofthecameraandtheimagesensorplaneD=0.1.

AssumezeromeanGaussianmeasurementnoisewithvariance for all fourLEDs. Figure 5.2 plots the measurement variance averaged over 1000 000 randomrealizations versus the CRLB for 0.005≤σ≤ 0.05. It is seen that the measurementvariancematchesperfectlywiththeCRLB,whichvalidatesthetheoreticalanalysis.

Figure5.2 EstimationerrorversusCramer–Raolowerbound(CRLB).

5.3 OutdoorlightpositioningsystemsbasedonLEDtrafficlightsandphotodiodes

5.3.1 LightpositioningsystemWe consider an LPS system consisting of one or multiple traffic lights and twophotodiodes [35]. Generalization to incorporate more than two photodiodes isstraightforwardandthusomitted.Thelightsignalemittedfromthetrafficlight,carryingthe lightposition information, is receivedby twophotodiodesmounted in the frontofavehicle.WiththeTDoA∆tofthelightsignaltothetwophotodiodes,thepathdifferencefromthetrafficlighttothetwophotodiodesisgivenas∆s=vL∆t,wherevListhespeedoflight.Sinceahyperbolaisthesetofpointsthathavethesamedistancedifferencetotwofixed points, the traffic light is on a hyperbola determined by the path differences andseparation of the two photodiodes which are the two foci of the hyperbola [15].Consideringthecenterofthetwophotodiodesastherequiredpositionofthevehicle,withtwo hyperbolas determined by one traffic light and two photodiodes in two differentlocations,orbytwotrafficlightsandtwophotodiodes,wecanderivetherelativepositionofthetrafficlighttothevehicle,andfinallytheabsolutepositionofthevehicle.Thesetwo

cases are discussed below.Herewe assume perfect synchronization of the traffic lightsthat transmit the positioning information, and perfect synchronization of the photo-receiversthatreceivesuchinformation.

LPSwithonetrafficlightAsshowninFig.5.3,whenthelightsignalofonlyonetrafficlightT1iscapturedbythetwophotodiodes,wecangenerate thefirsthyperboladeterminedby theobtainedTDoA∆t1andthedistanceofthetwophotodiodesF1,F2infrontofavehicleattimet1.Whilethe vehicle moves towards the traffic light, the second hyperbola is obtained throughTDoA∆t2andthesamedistanceofthetwophotodiodes , attimet2.TherelativepositionofthetrafficlighttothevehiclecanbederivedbythecrossingpointsT1,T2,T3,T4ofthetwohyperbolasandsomerelatedconstraintsgivenbelow.

Figure5.3 LPSwithonetrafficlightattheupperrightcornerandvehiclemovementalongtheY-direction.

Notethattherearedifferentrepresentationsofahyperbola.Onefamiliarrepresentationisthatforanypointonthehyperbolathedistancedifferencetothetwofociisfixed,givenby‖(x,y)−(c,0)‖−‖(x,y)−(−c,0)‖=2a.The second form is givenmathematically asfollows:

(5.17)

wherec2=a2+b2.Parameteradeterminesthecross-pointofthehyperbolatothex-axis.However,inthefollowing,weadoptanotherformbasedonafixedratioofthedistancetoonefocusoverthedistancetooneline,toobtainalowcomplexityestimator.

Ahyperbolacanalsobedefinedasthelocusofpointswhosedistancefromthefocusisproportional to the horizontal distance from a vertical line known as the conic sectiondirectrix, and the ratio is the eccentricity, where c is one half of the two photodiodes’

separationand is onehalf of thepathdifference from the traffic light to the twophotodiodes.Let denotethesquarerootofnormofvectorx.Wecanobtainthefollowingequations:

(5.18)

(5.19)

where(x,y)isthepositionofthetrafficlightinthecoordinatesystemwhoseoriginisthecenterofthetwophotodiodes,∆y=vV(t2−t1)isthedistancethevehiclemovesduringthetimedurationandvVisthespeedofthevehicle,ei=c/ai(i=1,2) is theeccentricity.AsshowninFig.5.3,fourcrossingpointsT1,T2,T3,andT4areobtainedthrough(5.18)and(5.19).TouniquelyspecifythedesiredtrafficlightT1,weutilizesomepriorinformation.Supposingthetwophotodiodesarefront-orientedwithoutanomnidirectionalfieldofview(FOV), the traffic light is located in front of the vehicle, that is y > ∆y. This thuseliminatesthepossibilityofT3andT4.Furthermore,letusdefinetheTDoAasthetimeofarrival(ToA)of the lightsignalfromthetraffic light to therightphotodiodeF2 less theToAto the leftF1, that is= tT,F2− tT,F1.Thex-coordinateof the traffic lightposition isnegative (x < 0) when the TDoA is positive and x > 0 when negative. Supposing theobtainedTDoAispositive,thecrossingpointT1isdeterminedastheuniquetrafficlightposition. Finally, the absolute position of the vehicle is determined as (X0 − x,Y0 − y),where(X0,Y0)istheabsolutepositionofthetrafficlight,whichcanbeobtainedbyVLCfromthetrafficlighttothephotodiodes.

LPSwithtwotrafficlightsInthisarrangement,asshowninFig.5.4,thelightsignalsoftwotrafficlightsT1,T2,forexampleonetrafficlightforvehiclesandtheotherforpedestrians,arecapturedbythetwophotodiodesF1,F2.WecangeneratetwohyperbolasdeterminedbytheobtainedTDoAs∆t1 and∆t2 and the photodiode spacing.With the properties of the hyperbola and theabsolute distance and direction of the two traffic lights, we can obtain the followingequations:

(5.20)

(5.21)

‖(x1,y1)−(x2,y2)‖=‖(X1,Y1)−(X2,Y2)‖, (5.22)

(5.23)

where(x1,y1)and(x2,y2)arethepositionsofthetwotrafficlightsinthecoordinatesystemwith the center of the two photodiodes as the origin, and (X1,Y1) and (X2,Y2) are theabsolute positions of the two traffic lights, respectively. Similarly to the casewith onetrafficlight,someundesirablesolutionscanbeexcluded.Notetheconstraintsofy1>0,y2>0;andx1<0,x2<0whentheTDoAs∆t1and∆t2arepositive,andx1>0,x2>0whentheyarenegative.Finallytheonlypossiblesolutiontotheabsolutepositionofthevehiclebecomes(X1−x1,Y1−y1)or(X2−x2,Y2−y2).

Figure5.4 LPSwithtwotrafficlightsandmeasuredTDoAsatthesametime.

Note that for the case of two traffic lights, one shot of the two TDoAs is taken toestimate the locationof thevehicle,whilefor thecaseofonetraffic light, twoshotsarenecessary.ThisisduetothefactthattheTDoAonlyspecifiesahyperbolaonwhichthelocationofthelightsmaylie,andatleasttwohyperbolasareneededtodefineapoint.

AnumericalexampleonthehyperbolaAssume that two photodiodes are mounted on the front of a vehicle with a 2 meterseparation.Consider a coordinate systemwhere theorigin lies in themiddleof the twophotodiodes,withthex-axisperpendiculartothemovingdirectionofthevehicle,andthey-axisalignedwiththemovingdirectionofthevehicle.Insuchacoordinatesystem,the

positionsofthetwophotodiodesare(−1,0)and(1,0).Let(x,y)bethepositionofalightsource.Let∆t=4.10−9sbetheTDoAofthelightsignalarrivingatthetwophotodiodes,wherethearrivaltimeat(−1,0)isearlierthanthatat(1,0)andsuchthat||(x,y)−(−1,0)||−||(x,y)−(1,0)||=−1.2m.Itimpliesthat(x,y)liesonahyperbolawiththetwofocigivenby(−1,0)and(1,0),whichcanbeexpressedinanotherformby(5.18).Morespecifically,wehavethatc=1,a1=0.6,b1=0.8.Itisseenthatthelocationof(x,y)liesonthelefthalfofthehyperbola

(5.24)

After traveling a distance of 4 m, we have the distances which lead to the distancedifference||(x,y)–(−1,4)||−||(x,y)–(1,4)||=−2m.Thenwehavec=1,a2=1,b2=0,which is an extreme case of a hyperbola. Then, we have y = 4, and

.

Thecaseof two traffic lights isanextensionof thatofone light.The locationof thevehicle canbedeterminedby the intersectionof the twohyperbolasdeterminedby twoTDoAsofthesignalssentbytwolightsatoneshot.Thisisessentiallyofthesamenatureastheexampleoflocalizationusingonelightviaseveralshots,whichisgivenabove.

5.3.2 Calibrationoferrorinducedbynon-coplanargeometryThe two hyperbolas for both LPSwith one and two traffic lights, generated in Section5.3.1,arenotnecessarilycoplanar.Thiswouldbringsomeinevitablepositioningerrorandaggravate the performance of the LPS system, especially when the vehicle is near thetrafficlight,whichwouldbringabiggeranglebetweenthetwohyperbolaplanes.Inthissection,wediscussthecoplanarrotationmethod,whichistorotateoneplanetocoincidewith the other around the intersecting line of the two planes, and generate the actualpositionofthevehiclewiththe3-dimensionaltransformtheory.

CoplanarrotationforLPSwithonetrafficlightForLPSwithonetrafficlight,referringtoFig.5.5,settherelativepositionofthetrafficlightT1as(x,y,z)inthecoordinatesystemXYZO.ThetiltanglesofplanesdeterminedbythetrafficlightT1andphotodiodepositionsA1,A2at time t1,andby the traffic lightT1

and photodiode positions B1, B2 at time t2, are and ,respectively.AsshowninFig.5.5, rotating theplaneT1B1B2 tocoincidewith theplaneT1A1A2aroundtheintersectinglineMN,wecangetthecorrespondingpoints andin plane T1A1A2 with respect toB1 andB2. According to the 3-dimensional transformtheory, the positions of the corresponding points and in the coordinate systemXYZOaregivenby

(5.25)

(5.26)

where

(5.27)

arethetranslationmatrixandrotationmatrix,respectively,T−1istheinversematrixofTand∆α=α2–α1istherotationangle.Notethathereweonlyrotatethecoordinateonthey and z-axes clockwise, such that the rotation via angle ∆α is added on the y and zcoordinates.Therotationanglecanbefoundfromα1andα2.Wehavethat

(5.28)

fromwhichwecanderivethevaluesofsin∆αandcos∆α.Morespecifically,wehavethat

cos∆α=cos(α2−α1)=cosα2cosα1+sinα2sinα1,sin∆α=sin(α2−α1)=sinα2cosα1−cosα2sinα1,

(5.29)

basedonresults,Rx(∆α),obtained.

Figure5.5 Non-coplanarLPSwithonetrafficlight.

WhentheheightHofthetrafficlightandtheheighthof thephotodiodesareknown,we can obtain z =H − h. And the TDoA of the mapped traffic light to the twophotodiodesis

(5.30)

Then,(5.18)and(5.19)canbewrittenas

(5.31)

(5.32)

where istheeccentricityofthenewhyperboladeterminedbythetrafficlightT1and the twomapped photodiode positions and . Combinedwith the constraintsgiveninSection5.3.1,wecandeterminetheactualrelativepositionofthetrafficlightandthentheabsolutepositionofthevehicleas(X0−x,Y0−y).

CoplanarrotationforLPSwithtwotrafficlights

Similarly to theLPSwithone traffic light, for theLPSwith two traffic lights shown inFig. 5.6, rotating the plane T2A1A2 to coincide with the plane T1A1A2 around theintersecting line OX, we can get the corresponding points in plane T1A1A2 withrespecttoT2.TherelativepositionsofthetrafficlightsT1andT2are(x1,y1,h1)and(x2,y2,h2)inthecoordinatesystemXYZO,andabsolutepositionsare(X1,Y1,H1)and(X2,Y2,H2) , respectively. According to the 3-dimensional transform theory, the relative andabsolutepositionsofthecorrespondingpoint aregivenby

(5.33)

(5.34)

andtheTDoAofthemappedtrafficlight tothetwophotodiodesA1,A2is

(5.35)

Then,(5.20)–(5.23)canberewrittenas

(5.36)

(5.37)

(5.38)

(5.39)

where istheeccentricityofthenewhyperboladeterminedbythemappedtrafficlight andthetwophotodiodes.CombinedwiththeconstraintsgiveninSection5.3.1,we can generate the actual relative position of the traffic light and then the absolutepositionofthevehicleas(X1–x1,Y1–y1)or .

Figure5.6 Non-coplanarLPSwithtwotrafficlights.

5.3.3 NumericalresultsIn this section,we evaluate the performance of the aboveLPSmethods, bothwith andwithout coplanar rotation. We adopt the following estimation bias, namely the meanpositioningerrorasperformancemetric:

(5.40)

whereBiasX andBiasY are the LPS bias errors in the X-axis and Y-axis, respectively.Assuming that the light signal propagates in an ideal noise-freeVLC link and the lightsignal iswell synchronized, therewouldbenoerror introduced in theabsolutepositioninformationobtainedonthetrafficlightsourceandtheTDoAofthelightsignalfromthetrafficlighttothetwophotodiodes.ThetrafficlightT1forvehiclesandthetrafficlightT2forpedestriansonacrossingarelocated3mleftand1mrightofthecenterofavehicleintheX-axis,respectively.Thetwophotodiodesaremountedonthefrontofthevehiclewitha2mseparation,and themiddlepointof the twophotodiodepositions is treatedas theposition of the vehicle. The vehicle moves towards the traffic light on a flat road andrecordstheTDoAofthelightsignaltothetwophotodiodesevery0.1s.Theheightsofthetraffic lightsT1 andT2 and photodiodes and some other basic parameters are given inTable5.1.

Table5.1 Basicparameters.

Symbol Quantity Value

H1 HeightoftrafficlightT1 6m

H2 HeightoftrafficlightT2 4m

h Heightofphotodiodes 1m

D Separationoftwophotodiodes 2m

t2−t1 Recordedtimeduration 0.1s

Figure 5.7 shows the LPS bias error performance with one traffic light against thedistancefromthetrafficlighttothevehicleintheY-axis.TheLPSbiaswithoutcoplanarrotationgrowssignificantlywhenthedistancefromthetrafficlighttothevehicleintheY-axisisreduced,especiallywhenthedistanceislessthan20m.Thepositioningbiasgetsbiggerwhenthevehiclespeedsupfrom10m/sto30m/s,andtherearestill0.50mand0.52m positioning errorswith 10m/s and 30m/s evenwhen the distance is up to 50meters.Butthereisnopositioningbiaswithcoplanarrotationasnoerrorisintroducedinthemeasurementasassumed.

Figure5.7 BiasperformanceofLPSwithonetrafficlight.

NowsupposethetwotrafficlightsT1andT2arethesamedistancefromthevehicleinthe Y-axis. Figure 5.8 shows the bias performance of the LPS with two traffic lightsagainstthedistancefromthetrafficlighttothevehicleintheY-axis.ThepositioningbiasoftheLPSwithoutcoplanarrotationincreasesdramaticallywhenthedistanceisreduced.Thebiaserror,reducedto0.4mforthedistanceof50m,growsmoreslowlythantheLPSwithonetrafficlightwhilethedistancedecreases.Itbecomes3.4mwhenthedistanceof

thetrafficlighttothevehicleontheY-axisis5m,wherethebiaserroris5.9mand8.1mforLPSwithonetrafficlightatspeedsof10m/sand30m/s,respectively.AlsothereisnobiasintroducedfortheLPSwithcoplanarrotation.

Figure5.8 BiasperformanceofLPSwithtwotrafficlights.

5.4 SummaryThischapterfirstpresentsanindoorLPSmodelusingwhiteLEDsandacamera,andthecorrespondingoptimallinearcombinationforanunbiasedestimateofthecamerapositionthatachievestheCRLB.ThenitpresentsatrafficlightandphotodiodebasedLPSmodelforvehiclepositioning.Thepositionofavehiclecanbedeterminedbasedonthereceivedposition information of the traffic light and the TDoA of the light signal to the twophotodiodesmountedinthefrontofthevehicle.MethodsforLPSwithoneandtwotrafficlights are both presented. When the non-coplanar effect is introduced, the vehiclepositioningmethodsareimprovedbycoplanarrotation.

This chapter considers indoor and outdoor positioning using visible light. A basicassumptionhereistheline-of-sightlinkfromthevisiblelightsource,withoutanyopticalwireless communication link error. Positioning based on optical wireless multipathreceived signals, especially in the scenarioof indoor positioningwithwall reflectionoroutdoor positioning with joint surrounding object reflection and vehicle movement,remainsasubjectforfutureresearch.

References[1] Gu,Y.,Lo,A.&Niemegeers,I.(2009),“Asurveyofindoorpositioningsystemsforwirelesspersonalnetworks,” IEEETrans.Communications Surveys andTutorials 11,13–32.

[2] Liu, H., Darabi, H., Banerjee, P. & Liu, J. (2007), “Survey of wireless indoorpositioningtechniquesandsystems,”IEEETrans.Systems,Man,andCybernetics,PartC:ApplicationsandReviews37,1067–1080.

[3] Cheok,A.&Li,Y. (2008),“Ubiquitous interactionwithpositioningandnavigation

using a novel light sensor-based information transmission system,” Personal andUbiquitousComputing12,445–458.

[4] Cheok, A. & Li, Y. (2010), “A novel light-sensor-based information transmissionsystem for indoor positioning and navigation,” IEEE Trans. Instrumentation andMeasurement60,290–299.

[5] Khoury,H.M.&Kamat,V.R.(2009),“Evaluationofpositiontrackingtechnologiesforuserlocalizationinindoorconstructionenvironments,”AutomationinConstruction18,444–457.

[6] Woo, S., Jeong, S., Mok, E. et al. (2011), “Application of WiFi-based indoorpositioningsystemforlabortrackingatconstructionsites:AcasestudyinGuangzhouMTR,”AutomationinConstruction20,3–13.

[7] Lashkari, A. H., Parhizkar, B. & Ngan, M. N. A. (2010), “WiFi-based indoorpositioning system,” in 2nd International Conference on Computer and NetworkTechnology,Bangkok,pp.76–78.

[8] Liu,X.,Makino,H.,Kobayashi,S.&Maeda,Y.(2006),“Anindoorguidancesystemfor the blind using fluorescent lights – relationship between receiving signal andwalking speed,” in Proc. 28th Engineering in Medicine and Biology SocietyConference,NewYork,pp.5960–5963.

[9] Liu, X., Umino, E. & Makino, H. (2009), “Basic study on robot control in anintelligent indoor environment using visible light communication,” in 6th IEEEInternationalSymposiumonIntelligentSignalProcessing,Budapest,pp.417–428.

[10] Randall, J.,Amft,O.,Bohn, J.&Burri,M. (2007), “Luxtrace: Indoor positioningusingbuildingillumination,”PersonalandUbiquitousComputing11,417–428.

[11] Yoshino,M.,Haruyama, S.&Nakagawa,M. (2008), “High-accuracy positioningsystem using visible LED lights and image sensor,” in IEEE Radio and WirelessSymposium,Orlando,pp.439–442.

[12] Randall,J.,Amft,O.,Bohn,J.&Burri,M.(2003),“PositioningbeaconsystemusingdigitalcameraandLEDs,”IEEETrans.VehicularTechnology52,406–419.

[13] Bigas, M., Cabruja, E., Forest, J. & Salvi, J. (2006), “Review of CMOS imagesensors,”MicroelectronicsJournal37,433–451.

[14] Castillo-Vazquez,M.&Puerta-Notario,A.(2005),“Single-channelimagingreceiverforopticalwirelesscommunications,”IEEECommunicationsLetters9,897–899.

[15] Liu, X.,Makino, H., Kobayashi, S. &Maeda, Y. (2008), “Research of practicalindoorguidanceplatformusingfluorescentlightcommunication,”IEICETransactionsonCommunicationsE91B,3507C3515.

[16] Sertthin, C., Ohtsuki, T. & Nakagawa, M. (2010), “6-axis sensor assisted lowcomplexity high accuracy-visible light communication based indoor positioningsystem,”IEICETransactionsonCommunicationsE93B,2879–2891.

[17] Shaifur,R.M.,Haque,M.M.&Kim,K.(2011),“High-accuracypositioningsystemusing visible LED lights and image sensor,” in 14th International Conference on

ComputerandInformationTechnology(ICCIT),Dhaka,pp.309–314.

[18] Kim,Y.,Hwang,J.,Lee, J.etal. (2011), “Position estimation algorithmbasedontrackingofreceivedlightintensityforindoorvisiblelightcommunicationsystems,”in3rd InternationalConferenceonUbiquitousandFutureNetworks (ICUFN), pp. 131–134.

[19] Kim,H.,Kim,D.,Yang,S.,Son,Y.&Han,S. (2011),“Indoorpositioningsystembasedoncarrierallocationvisiblelightcommunication,”inLasersandElectro-Optics,PacificRim,Sydney,pp.787–789.

[20] Savasta,S.,Pini,M.&Marfia,G.(2008),“PerformanceassessmentofacommercialGPS receiver for networking applications,” in 5th IEEEConsumer CommunicationsandNetworkingConference,pp.613–617.

[21] Savasta,S.,Joo,T.&Cho,S.(2006),“Detectionoftrafficlightsforvision-basedcarnavigationsystem,”in1stPacificRimSymposium,PSIVT,pp.682–691.

[22] Wang,W.&Cui,B.(2006),“Automaticmonitoringandmeasuringvehiclesbyusingimageanalysis,”inProceedingsofSPIE-ISandTElectronicImaging,pp.1–8.

[23] Watada, S., Hayashi, K., Toda, M. et al. (2009), “Range finding system usingmonocular in-vehicle camera and LED,” in Intelligent Signal Processing andCommunicationSystems,2009,IEEE,pp.493–496.

[24] Nagura,T.,Yamazato,T.,Katayama,M.etal.(2010a),“ImproveddecodingmethodsofvisiblelightcommunicationsystemusingLEDarrayandhigh-speedcamera,”in71stVehicularTechnologyConference(VTC2010-Spring),pp.1–5.

[25] Nagura, T., Yamazato, T., Katayama, M. et al. (2010b), “Tracking an led arraytransmitter for visible light communications in the driving situation,” in 7thInternationalSymposiumonWirelessCommunicationSystems(ISWCS),pp.765–769.

[26] Pang,G.&Liu,H. (2001), “Led location beacon system based on processing ofdigitalimages,”IEEETrans.IntelligentTransportationSystems2,135–150.

[27] Pang,G.,Liu,H.,Chan,C.&Kwan,T. (1998), “Vehicle location and navigationsystemsbasedonleds,”inProceedingsof5thWorldCongressonIntelligentTransportSystems,Seoul,pp.12–16.

[28] Roberts, R., Gopalakrishnan, P. & Rathi, S. (2010), “Visible light positioning:Automotiveusecase,”inIEEEVehicularNetworkingConference(VNC),pp.309–314.

[29] Hecht,E.(2001),Optics(4thed.),AddisonWesley.

[30] Cui, K., Chen, G., Xu, Z. & Roberts, R. D. (2010), “Line-of-sight visible lightcommunication system design and demonstration,” in Proc. of 7th IEEE, IETInternational Symposium on Communication Systems, Networks and Digital SignalProcessing,Newcastle,pp.21–23.

[31] Gow, R., Renshaw, D., Findlater, K. et al. (2007), “A comprehensive tool formodelingCMOSimage-sensor-noiseperformance,”IEEETrans.Electron.Devices54,1321–1329.

[32] Sadler,B.M.,Liu,N.,Xu,Z.&Kozick,R. (2008), “Range-based geolocation in

fadingenvironments,”inProc.ofAllertonConference,Monticello,pp.23–26.

[33] Liu, N., Xu, Z. & Sadler, B. M. (2008), “Low complexity hyperbolic sourcelocalizationwithalinearsensorarray,”IEEESignalProcessingLetters15,865–868.

[34] Poor,H.V.(1994),Anintroductiontosignaldetectionandestimation,Springer.

[35] Bai,B.,Chen,G.,Xu,Z.&Fan,Y.(2011),“VisiblelightpositioningbasedonLEDtraffic light and photodiode,” in IEEEVehicular Technology Conference (VTC Fall),pp.1–5.

6 Thestandardforvisiblelightcommunication

KangTae-Gyu

6.1 ScopeofVLCstandardVisible lightcommunicationVLChastheadvantagethat theemittingsourceofwirelesscommunication uses LED light [1]. The standard for illumination has been developedcoveringconnectionsbetweenlampandelectronicpowerintermsofelectricsafety,assetout in IEC TC 34. The standard for visible light communication needs a number ofprotocolsbetweenthesendingpartyandthereceivingparty,asgiveninPLASAE1.45[2]and IEEE 802.15.7 [3]; as well as dealing with electric safety. We have to considercompatibilitieseventhoughtheVLCservicearea,illumination,vendor,andstandardaredifferent.

6.1.1 VLCserviceareacompatibilityVisiblelightcommunicationservicescanbeprovidedindifferentilluminationspaceareas[12]. The area could be a lighting space for amuseum, shoppingmall, hallway, office,restaurant, etc. There are twoVLC service styles.One is for a specific area, such as acompany or organisation defined location inwhich proprietary equipment can be used.The other is a public area where in order to communicate the equipment must becompatiblewithcommunicationsstandards.Whenwedesignforaspecificarea,wedonotneed any standard forVLC.Design of a specificVLC can be done easilywithout anyconstraintsorlimitationsbecausethedesignisbasedonproprietorytechnologyandnotonadefinedstandardsuchas (4) inFig.6.1.ThisspecificstyleVLChas theadvantageoffast,cheapdeploymentatthefirstopeningstage,buthasthedisadvantageoflackingVLCserviceareacompatibilitysuchas(2),(5),(6)inFig.6.1.

Figure6.1 ExampleofVLCserviceareacompatibility.

We need standards to ensureVLC service area compatibility for any kind of service

areasuchas(1)and(3)inFig.6.1.TheinternationalstandardsareIEEE802.15.7[7]andPLASA E1.45 [11]. The domestic or national standards consist of three documents inJapan[8]and18documentsinKorea[9].

6.1.2 VLCilluminationcompatibilityThere are various LED illumination power capabilities, fixtures, and colors; theircapabilitiesand their shapesvaryaccording to their intendedusage.Although there isawiderangeofLEDilluminationsystems,thevisiblelightcommunicationstandardhastobeapplicabletoallkindsofLEDsystem.

We need to know the LED illumination standards in IEC TC 34, so as to go on todevelop new standards, including those forVLC components in LED illumination, seeFig.6.2.

Figure6.2 VLCilluminationcompatibility.

6.1.3 VLCvendorcompatibilityThere are many LED illumination vendors, who can introduce and withdraw LEDilluminationproductsfreely.Illuminationandreceivingterminalsmayceasetofunctionorreachtheendoftheiroperationallifespan.TheVLCstandardhastosupportVLCvendorcompatibility. We can choose LED illumination and receiving terminals transparentlywithoutselectingspecificvendorsorproducts.Wecanreplaceatanytime,anyproduct,anymanufacturer,andanyvendorsuchas(1)and(2)inFig.6.3.Incase(3)inthefigure,thereisaproblemduetolackofvendorcompatibility.

Figure6.3 VLCvendorcompatibility.

InordertoachieveVLCvendorcompatibility,weneedastandardwithinter-operableprofiles.IfwecannotbuyasuitableproductwithVLCcompatibility,wehavetochoosere-installation of illumination and replace the receiving terminal or abandon the VLCservice and install only illumination lighting.We cannot abandon the lighting function,whichismandatoryinourlives.

6.1.4 StandardcompatibilityThereareseveralstandardsrelatingtovisiblelightcommunication:IEEE802.15.7VLCPHY/MAC; IEC TC 34 LED illumination; PLASA E1.45 DMX-512AVLC; and LEDlighting source Zhaga engine [10]. IEEE 802.15.7 VLC PHY/MAC, issued in 2011,coversVLCPHYinteriorLEDilluminationandVLCPHYreceivers,seeFig.6.4.

Figure6.4 Visiblelightcommunicationstandardcompatibility.

The IEC TC 34 LED illumination intelligent system lighting (ISL) ad hocWorkingGroup is developing standards for digital functional components including visible lightcommunication. PLASA E1.45 DMX-512A VLC issued in 2013 covers visible lightcommunicationdata,andwired transferbetweenLEDilluminationandacontrol server.Zhagaengineapplies toLEDlightingsources for illumination.LEDlightingonandoffswitchingcanbecontrolledusingwirelessnetworks:ZigBee,IrDA,Bluetooth,andWiFi.

International or domestic standard specifications can be developed concurrently bymanystandardorganizationsorworkinggroups,whocansharetheirstandards’activitiesand draft specifications via exchanging documents. This should ensure standardcompatibility.Anewstandardspecificationmightupgradeor“versionup”apreviousone,andmustbewrittentoensureforwardandbackwardstandardcompatibility.Theresultisconsistentprinciplesfordevelopersandendusersandavoidanceofconfusion.

6.2 VLCmodulationstandardIEEE 802.15.7 VLC has three different PHY (physical layers) that depend on theapplication:PHYI,PHYII,andPHYIII.PHYIisintendedforoutdoorusewithlowdatarateapplications[4–6].Thismodeuseson-offkeying(OOK)andvariablepulsepositionmodulation(VPPM)withdataratesinthetenstohundredsofkb/s.PHYIIisintendedforindoorusewithmoderatedata rate applications.ThismodeusesOOKandVPPMwithdataratesinthetensofMb/s.PHYIIIisintendedforapplicationsusingcolorshiftkeying(CSK)thathavemultiplelightsourcesanddetectorswithdataratesinthetensofMb/s.

6.2.1 VariablepulsepositionmodulationVPPMVLCneedsamodulationstrategythatiscompatiblewithdimmingcontrolofillumination.Oneexampleofsuchastrategyisvariablepulsepositionmodulation(VPPM).VPPMisa

modulationschemethatiscompatiblewithdimmingcontrolthatvariesthedutycycleorpulsewidthtoachievedimming,asopposedtoamplitudecontrol.

VPPMcombines2-PPMwithPWMforadimmingcontrol.Bits“1”and“0”inVPPMaredistinguishedbythepositionofapulse,whereasthewidthofthepulseisdeterminedbythedimmingratio.TheprinciplesofVPPMareillustratedinFig.6.5.

Figure6.5 PrincipleofvariablepulsepositionmodulationVPPM.

6.2.2 LinecodingThe4B6BlinecodeexpandseachblockoffourbitsintoanencodedblockofsixbitswithDCbalance.Thismeanstherewillalwaysbepreciselythreezerosandthreeonesineachblockofsixencodedbits.

6.3 VLCdatatransmissionstandardTherearetwotypesofvisiblelightcommunicationdatatransmission.Oneisfixeddata,and the other is changeable data within lighting for visible light communication. Thechangeable data can be changed in accordance with wired transmission protocols andwirelesstransmissionprotocols.

6.3.1 WiredtransmissionprotocolThere are two kinds of candidate wired visible light communication data transmissionprotocol: PLASA E1.45 DMX-512A VLC; and IEC 62386 DALI VLC (Table 6.1).PLASA is a trade association for Professional Lighting and Sound with a technicalstandardsprogram.

Table6.1 Wiredtransmissionprotocolforvisiblelightcommunication.

Standardspecification Organization Functions

E1.45DMX-512AVLC

PLASA DMX512-Adatalinkfordatatransmissionfromthoseluminairesusingvisiblelightcommunication,IEEE802.15.7

IEC62386DALI

IECTC34 Aprotocolforcontrolbydigitalsignalsofelectroniclightingequipment

PLASAE1.45DMX-512AVLCwas issued in 2013 to allow communication of 802datatoluminairesoveranANSIE1.11DMX512-Adatalinkfordatatransmissionfromthoseluminairesusingvisiblelightcommunication,IEEE802.15.7.ANSIE1.11,revisedin2008,describesamethodofdigitaldatatransmissionforcontroloflightingequipmentandaccessories, includingdimmers, color-changers, and relatedequipment.DMX512-AcanbeusedforoutdoormediafaçadeLEDlighting[13].

IEC TC 34 62386 DALI is one of the candidate protocols for visible lightcommunicationtransmissiondata.TheInternationalElectro-technicalCommission(IEC)is the world organization that prepares and publishes international standards for allelectrical,electronicandrelatedtechnologies[12].IEC62386digitaladdressablelightinginterface (DALI) specifiesaprotocol forcontrolbydigital signalsofelectronic lightingequipment.DALIcanbeusedforindoorlightingdimmingcontrol.

6.3.2 WirelesstransmissionprotocolThereare complementaryprotocols forwireless transmission:ZigBee, IrDA,Bluetooth,andwirelessLAN.Visiblelightcommunicationneedsadditionalwirelesscommunicationskillsduetoitsunidirectionalcommunicationprinciples.

ZigBee,definedinIEEE802.15.4,isusedinapplicationsthatrequireonlyalowdatarate,longbatterylife,andsecurenetworkingwith250kbit/stransferrate.ZigBeecanbeusedforwirelesslightswitchesincludingdimmingcontrols.

6.4 VLCilluminationstandardTheadvantageofvisiblelightcommunicationtechnologyisitsusageofLEDilluminationwithout any other transmission media, so we need to apply traditional illuminationstandards.TC34preparesinternationalstandardsforlampsandotherrelatedequipment,andwasestablishedin1948.

6.4.1 LEDlightingsourceinterfaceThe international Zhaga Consortium is developing interface specifications that enableinterchangeability of LED light sources made by different manufacturers. The Zhagaspecificationsknownasbooks,describetheinterfacesbetweenLEDluminairesandLEDlight engines. This will accelerate the adoption of LED lighting solutions in themarketplace.

Eightbooksof specifications cover the technologyas follows:1overview,2, 5, 6, 8socketabledrum-shapedLEDsources,3circularLEDmodules,4and7rectangularLEDmodulesaccordingtoshapeofluminaire.

Visible light communicationusesLED light, but theZhagaLEDmodulehasnot yetbeen considered as visible light communication. When we develop a visible lightcommunicationPHYandanapplicationservice,wehavetoconsiderZhagaLEDmodulespecifications.

6.4.2 FixtureinterfaceInternational Electro-technical Commission Technical Committee TC 34 has beendeveloping international standards regarding specifications for lamps including LEDs,lamp caps and holders, lamp control gear, luminaires, and miscellaneous relatedequipmentnotcoveredbyprojectsofothertechnicalcommittees.

Visiblelightcommunicationcanbeoneofaluminaire’sfunctions.IECTC34hasnotyet developed any specifications for visible light communication. They are needed tospecifyhowtomergelightingitselfandwirelesscommunicationusinglight.

6.4.3 LEDintelligentsystemlightinginterfaceTheIECTC34intelligentsystemlightingadhocWorkingGrouphadafirstface-to-facemeeting in January2014. Its remit isnewcreativeconvergence technologybetween thetraditional lightingindustryandinformationandcommunications technology(ICT).Thefunction of LED lighting has introduced use of ICTs such as wireless communication,wired communication, and visible light communications, see Fig. 6.6. WirelesscommunicationtechnologiescanbeadaptedusingZigBee,Bluetooth,IrDA,andwirelessLANaccordingtoapplicationspecificrequirements.

Figure6.6 LEDsystemlightenginewithvisiblelightcommunication.

Visiblelightcommunicationrequiresdevelopmentofco-operativeluminairefunctionsand other information technology such as wireless communication and wiredcommunication.

6.4.4 VLCservicestandardWecanfindservicestandardactivitiesforvisiblelightcommunicationinIEEE802.15.7,IECTC34,PLASACPWG,TTAVLCWG,VLCC,andITU-TSG16.

ThebookAdvancedOpticalWirelessCommunicationChapter14[1]namesasvisiblelightcommunicationapplications:VLCguidancesystems,VLCcolorimaginablesystems,VLCindoornavigator,andVLCautomobiledrivingsupportsystems.

AVLCguidancesystemuseslampsthatilluminateayard,nationalborder,orfacility,for guidance as well as for protection from outside attacks. The lamps have anidentification number (VLC ID or LED ID) and guidance information. A VLC colorimaginablesystemusescolorlampsforcolorinformationitself,whetherfrominstinctoreducation. A VLC indoor navigator uses lamps with visible light communication forindoorsalesareanavigationinlocationswhereGPSisnotsupported.AVLCautomobiledrivingsupportsystemuseslampsincludingheadlights,foglights,turnsignals,andbrakelightsforsafedriving.

TheVLCWorkingGroup(WG4021)inTTAstartedinMay2007.VLCWGdevelopedTTA 5 VLC standard specifications in 2008, covering: Basic Configurations ofTransmitter PHY for Visible Light Communication, Basic Configurations of ReceiverPHY for Visible Light Communication, Basic Configuration of LED Interface forIllumination and Visible Light Communication, Basic Configuration of Light LocationInformationServiceModelusingVisibleLightCommunication,andBasicConfigurationofLightingIdentificationforVisibleLightCommunication.

VLCWGdevelopedTTA23VLCandLEDcontrolrelatedstandarddraftsin2013.The18 specifications are focused on theways to combine visible light communication andilluminationtechnologies.

Visible light communication can deliver creative services, but relevant standardspecificationshavenotyetbeendeveloped.However,theVLCPHYspecificationinIEEE802.15.7,andVLCdatawiredtransmissionspecificationinPLASAE1.45areavailable.Opening of the visible light communications market requires service standardspecifications. Thesewill provide guidance for users from the viewpoint of applicationservicefunctiondevelopment.

References[1] ShlomiArnon,JohnR.Barry,GeorgeK.Karagiannidis,RobertSchober,andMuratUysal, Advanced Optical Wireless Communication, Chapter 14 “Visible lightcommunication,”pp.351–368,CambridgeUniversityPress,2012.

[2] ANSI E1.45, “Unidirectional transport of IEEE 802 data frames overANSI E1.11(DMX512-A),”2013.

[3] IEEEStd802.15.7-2011,“IEEEstandardforlocalandmetropolitanareanetworks–part15.7:Short-rangewirelessopticalcommunicationvisiblelight,”2011.

[4] Sang-KyuLim,“ETRIPHYproposalonVLCbandplanandmodulationschemesforillumination,”IEEE802.15-09-0674-00-0007,2009.

[5] DaeHoKim,“ETRIPHYproposalonVLClinecodeforillumination,”IEEE802.15-09-0675-00-0007,2009.

[6] Youjin Kim, “Analysis of IP-based control networks for LED lighting fixture

communication,”NewTrends in InformationScienceandServiceScience (NISS),4thIEEEConference,2010,pp.307–312.

[7] EunTaeWon,(2009,January),IEEE802.15WPAN™TaskGroup7(TG7)VisibleLightCommunication,Available:http://www.ieee802.org/15/pub/TG7.html.

[8] Masao Nakagawa, (2007), Visible Light Communication Consortium, Available:http://www.vlcc.net.

[9] Leem Chasik, (2014), Telecommunication Technology Association Visible LightConvergence Communication Project Group 425, Available:http://www.tta.or.kr/English/.

[10] “LED light sources interchangeable,” Zhaga Consortium, (2014), Available:http://www.zhagastandard.org/.

[11] KarlRuling,(2014),PLASAStandards,ControlProtocolWorkingGroup,Available:https://www.plasa.org/.

[12] International Electrotechnical Commission Technical Sub-committee 34C,“Auxiliariesforlamps,”(2014),http://www.iec.ch/.

[13] Sang-KyuLim,“Entertainment lightingcontrolnetworkstandardization tosupportVLCservices,”IEEECommunicationMagazine51,(12),pp.42–48,2013.

7 Synchronizationissuesinvisiblelightcommunication

ShlomiArnon

In the preceding chapters, many facets of VLC have been outlined demonstrating thepromising features of this emerging illumination and communication technology. It isanticipated that VLC will become a familiar facility in the office, on the road (V2Vcommunications) and even in toys (Disney Research). The broad range of VLCapplications is driving an exponential rise in the required data rate, while thesemiconductorindustryiskeepingpaceindevelopinglightsourcesthatcouldprovidethenecessaryhighdatarate.ThehighdatarateVLCparadigmopensupnewopportunitiesformodulation methods that are being developed to meet the particular requirements ofsimultaneous illumination and communication. Demodulation of the signal and, hence,extractionofthetransmittedinformationrequirestringentsynchronization.Inthischapter,we present four VLC modulation methods: on off keying (OOK), pulse positionmodulation(PPM),inversepulsepositionmodulation(IPPM),andvariablepulsepositionmodulation(VPPM),anddevelopexpressions thatdescribe thebiterrorrates(BER)foreachmethod.We provide examples for the effect of clock time shift and jitter on thesystemBERperformancefortheinversepulsepositionmodulation(IPPM)method.

7.1 IntroductionThe many different applications of VLC systems, such as V2V [1–5], short rangecommunicationfortoys[6–8]andhighdatarateindoorimplementations[7–13]havebeenelaborateduponintheprecedingchapters.WhiletheunderlyingVLCprincipleissimple–anelectricalsignalfromthetransmitterisconvertedtoamodulatedopticalsignalbythelight source and carried by the illuminating light to the receiver, different modulationmethods that encapsulate the informationwithin the illumination light arepossible.Thecommonmodulationmethodscanbedividedintothreefamilies[11,13,14],basedona)illumination color e.g. color shift keying (CSK), b) many subcarriers e.g. discretemultitone(DMT)ororthogonalfrequencydivisionmultiplexing(OFDM),andc)intensitymeasuredinthetimedomain,includingonoffkeying(OOK),pulsepositionmodulation(PPM), inverse PPM (IPPM) and variable pulse position modulation (VPPM). Thesemethodsmakeitpossibletousetheilluminationlightasatooltocarryinformationinarelatively straightforward way. However, the performance of time domain modulationmethodsisextremelysensitivetotimesynchronizationandclockjitter.Inthischapter,weexplainthefourdifferenttimedomainmodulationmethodsmentionedaboveanddevelopexpressionsfortheBERthatcanbeachieved.Theremainderofthischapterexpandsontheeffectoftimesynchronizationontheperformanceofoneofthesemethods(IPPM).

7.2 VLCmodulationmethodsinthetimedomainIn this section we describe in more detail four different modulation methods thatencapsulate information in the timedomain in thecontextofVLC.Themaindifferencebetween the conventional communication scenarios and VLC is the requirement to

maintain adequate communication performance alongside the necessity to dim the lightintensity.Wenowreviewa)OOK,b)PPM,c)IPPM,andd)VPPM.

7.2.1 Onoffkeying(OOK)Onoffkeying(OOK)isaformofbinarysignalmodulationinwhichabitisencodedbytransmitting optical power for T seconds if the information is “1,” whereas when theinformation is “0,” no optical power is transmitted in the T-second time-slot [13].Dimming of the light is implemented by reducing the transmitted pulse power inaccordancewiththerequiredpercentagedimming(Fig.7.1).

Figure7.1 OOKwithdifferentdimmingpercentages,wheretheencodedinformationis“1011.”

7.2.2 Pulsepositionmodulation(PPM)Pulsepositionmodulation(PPM)[13,14,15]isaformofsignalmodulationinwhichMmessagebitsareencodedbytransmittingapulseforthedurationofTC=T/2Msecondsinoneof the2M possible time-shiftswithina time-slotofT seconds,which is the symbolduration.ThisschemeisrepeatedeveryTseconds,sothatthetransmittedbitrateisM/Tbitsper second.Dimmingof the light is implementedby reducing the transmittedpulsepowerinaccordancewiththerequiredpercentageofdimming(Fig.7.2).

Figure7.2 Pulsepositionmodulation(PPM)withdifferentdimmingpercentages,wheretheencodedinformationis“00101011.”

7.2.3 Inversepulsepositionmodulation(IPPM)Inversepulsepositionmodulation(IPPM)[16]isaformofsignalmodulationinwhichMmessagebitsareencodedbytransmittingpowerforTseconds,exceptfora“hole”inoneof the2M possible time-shifts. The duration of the “hole” isTC=T/2M. This scheme isrepeatedeveryTseconds,sothatthetransmittedbitrateisM/Tbitspersecond.Dimmingof the light is implementedby reducing the transmittedpulsepower inaccordancewiththerequiredpercentagedimming(Fig.7.3).

Figure7.3 Inversepulsepositionmodulation(IPPM)withdifferentdimmingpercentages,wheretheencodedinformationis“00101011.”

7.2.4 Variablepulsepositionmodulation(VPPM)Variablepulsepositionmodulation(VPPM)isaformofsignalmodulationinwhichthemessagebitisencodedbytransmittingapulseatthebeginningofthesymbolfor“0”andattheendofthesymbolfor“1”[11,14,17].Thedurationof thepulse isdetermined inaccordancewiththepercentageofrequiredillumination(dimming).Themainadvantageof thismethod is that the communication is not affected by a change of the amount ofdimming,aslongasthereissomeillumination.ThisschemeisrepeatedeveryTseconds,sothatthetransmittedbitrateis1/Tbitspersecond(Fig.7.4).

Figure7.4 Variablepulsepositionmodulation(VPPM)withdifferentdimmingpercentages,wheretheencodedinformationis“0100.”

7.3 BiterrorratecalculationInthissectionwederiveexpressionsfortheachievablebiterrorrate(BER)withthefourmodulationmethodspresentedabove–OOK,PPM,IPPM,andVPPM.

7.3.1 OOKBEROOK is a form of binary signal intensitymodulation, which can be implementedwithdirectdetectionreceiversinwhichtheopticalpowerisconvertedtoelectronicsignalsbyaphotodetector. The conversion ratio is described by the detector’s responsivity, R. Wedenote the resultant electrical signals, prior to decision-making, by y. The receiverintegratesthereceivedsignalsandthedecisionwhethera1ora0wastransmittedismadeaccording to a given algorithm criterion.We assume that the electronic noise, togetherwithbackgroundnoise,isthedominantnoisesourceanditismodeledbyadditivewhiteGaussian noise that is statistically independent between time-slots. The noise has zeromeanandcovarianceofσ21andσ20forsignals1and0,respectively.Afterintegration,thesignalsaredescribedbythefollowingconditionaldensities:

(7.1)

and

(7.2)

In the next section, synchronization jitter and offset are analyzed; therefore we nowrearrangetheexpressionforthesignalandnoisesothatthebit/symbolduration,T,ispartofthesignal.Thesignalisdescribedbythesquarerootoftheenergy,sothatμ1=ηDimmingR P1T0.5 and σ12 = σ2TH + 2qηDimming RP1 are the received optical signal and theaccompanyingreceivernoisestandarddeviation, respectively.μ0=0andσ02=σ2THarethe received optical signal and the receiver noise standard deviationwhen no power isreceived,respectively,P1 istheopticalpower,σ2TH includestheeffectsof thermalnoiseandofbackgroundshotnoiseandηDimmingistheilluminationdimmingfactor0<ηDimming≤1.Thedecisionalgorithm in thiscase isbasedon themaximuma-posterioriprobability

(MAP)criterion,whichmapsthereceivedsignalaccordingtothefollowing:

(7.3)

whereP(y|s)istheconditionalprobabilitythatifabitsistransmitted(takingoneoftwovalues, 1 or 0), a y will be received,P(s) is the a-priori probability that a 1 or a 0 istransmittedandP(y)isthea-prioriprobabilityofy.

Thedenominatorisidenticalforallsignalsandthereforedoesnotaffectthedecision.Incommunicationsystems,theprobabilitiesoftransmitting1and0bitsare,inmostcases,equal,sowecansimplyinvoke(7.3)andusethemaximumlikelihood(ML)estimator.

Inthatcase,thelikelihoodfunctionisgivenby

(7.4)

In the casewhereσ2TH>> 2qηDimmingRP1, Eq. (7.4) could be simplified by taking thenatural logarithm, ln(x), of both sides of the equation, canceling common factors, and

rearranging. Due to the complexity of this expression, it is common to use anapproximationforthebiterrorprobability(BER)givenby

(7.5)

7.3.2 PPMBERPulsepositionmodulation (PPM) isa formof signal intensitymodulationwhichcanbeimplementedwith direct detection receivers inwhich the optical power is converted toelectronicsignalsbyaphotodetector,asforOOK.Theresultantelectricalsignals,beforeadecisionismade,aregivenbyyi,wherei∈{0…2M−1}.Thereceiver integrates thereceived signals in each of the 2M possible time-shifts. We assume that the electronicnoise, together with background noise, is the dominant noise source when no opticalsignal is transmitted (0) and the signal shot noise, together with electronic noise andbackgroundnoise,isthedominantnoisesourceotherwise.InbothcasesthenoisecanbemodeledbyadditivewhiteGaussiannoisethatisstatisticallyindependentbetweentime-slots.Thenoisehaszeromeanandacovarianceofσ21andσ20ifor1and0,respectively.Afterintegration,thesignalsyi,aredescribedbythefollowingconditionaldensities:

(7.6)

and

(7.7)

whereμ1=ηDimmingRP1TC0.5andσ12=σ2TH+2qηDimmingRP1 are the receivedopticalsignal and the accompanying receiver noise standard deviation, respectively, for atransmitted1andμ0i=0andσ20i=+σ2THarethereceivedopticalsignalandthereceivernoise standard deviation for a transmitted 0, respectively. R is the responsivity of thedetector, P1 is the optical power, σ2TH includes the effects of thermal noise and ofbackgroundshotnoiseandηDimmingistheilluminationdimmingfactor0<ηDimming≤1.Attheendoftheintegrationperiod,thereceivermakesadecisionastowhichslothas

the largest value by comparing the 2M integration results. The decision algorithmcalculates the indices of the minimum values of the signal vector, which is describedmathematically by the function [J,B] = max(A). This function finds the indices of the

maximum values of A and returns them in the output vector J, while B is the largestelementinA.AisgivenbyA=[μ0…μ2M−1].Apracticalimplementationofthedecisionalgorithm could employ a comparator that finds the largest voltages from 2M slotmeasurementsat theendof thesymbolperiod.Thecalculationof theBERcanbedoneeasily ifwe first calculate the probability of a correct decision. The correct decision iscalculatedbyexaminingthereceivedsignalamplitudesoverarangefromminusinfinitytoinfinity.Foreachvalueinthisrangewecalculatetheprobabilitythatthereceivedsignalinthepulseslotwillbelargerthanthevalueofthesignalfromallothertime-slots.Hence,1minus the calculated correct probability is the erroneous probability.We also assumethatallyifori>1areidenticalandindependentlydistributed(i.i.d)randomprocesses,so

theycanberepresentedbyanidenticaldensityfunctionsuchas .

InthatcasetheBERisgivenby

(7.8)

The term 2M−1/(2M − 1) is used to convert the symbol error rate to the bit error rate.

Employingtheerrorfunction,definedas ,weobtainthefollowingBERexpression:

(7.9)

7.3.3 IPPMBERIPPM is a form of signal intensity modulation, which can be implemented with directdetection receivers in which the optical power is converted to electronic signals by aphotodetector,asforOOKandPPM.Theresultantelectricalsignals,beforeadecisionismade, are given by yi where i∈ {0… 2M − 1}. The receiver integrates the receivedsignals in each of the 2M possible time-shifts. At the end of the integration period thereceivermakes a decision as towhich slot has the smallest value by comparing the 2Mintegrationresults.Weassumethattheelectronicnoise,togetherwithbackgroundnoise,isthe dominant noise source during the “hole” and the “signal” shot noise, togetherwithelectronic noise and background noise, is the dominant noise source otherwise. In bothcases the noise can be modeled by additive white Gaussian noise that is statisticallyindependentbetween time-slots.Thenoisehaszeromeanandcovarianceofσ2iandσ20

forasignalandahole,respectively.Afterintegration,thesignalsx,andyiaredescribedbythefollowingconditionaldensities:

(7.10)

and

(7.11)

whereμi=ηDimmingRP1TC0.5 andσi2 = 2qηDimmingRP1+σ2TH are the received opticalsignal and theaccompanying receivernoise standarddeviation, respectively, andμ0 = 0andσ02=σ2TH are the receivedoptical signal and the receiver noise standarddeviationwhenaholeisreceived,respectively.Ristheresponsivityofthedetector,P1istheopticalpower, σ2TH includes the effects of thermal noise and of background shot noise, andηDimmingistheilluminationdimmingfactor0<ηDimming≤1.The decision algorithm calculates the indices of the minimum values of the signal

vectorandisgivenmathematicallybythefunction[J,B]=min(A).ThisfunctionfindstheindicesoftheminimumvaluesofAandreturnsthemintheoutputvectorJ,whileBisthesmallest element inA.A is given byA = [μ0…μ2M−1]. Practical implementation of thedecisionalgorithmcouldemployacomparatorthatfindsthelowestvoltagesfrom2Mslotmeasurementsat theendof thesymbolperiod.Thecalculationof theBERcanbedoneeasily ifwe first calculate the probability of a correct decision. The correct decision iscalculatedbyexaminingthereceivedsignalamplitudesoverarangefromminusinfinitytoinfinity.Foreachvalueinthisrangewecalculatetheprobabilitythatthereceivedsignalin the hole slot will be smaller than the value of the signal from all other time-slots.Hence, 1minus the calculated correct probability is the erroneous probability.We alsoassume that all yi for i∈ {1 … 2M−1} are i.i.d random processes so they can be

representedbyanidenticaldensityfunctionsuchas .

InthatcasetheBERisgivenby

(7.12)

Theterm2M−1/(2M−1)isusedtoconvertthesymbolerrorratetotheBER.

7.3.4 VPPMBERVPPM is a form of binary signal intensity modulation, which can be seen to be acombinationofpulsewidthmodulation(PWM),whichisusedtocontroltheillumination,andPPM,which isused tocarrycommunication information.AswithOOK,PPM,andIPPM, direct detection receivers in which the optical power is converted to electronicsignalsbyaphotodetectorareused.The resultantelectrical signals,beforeadecision ismade,aregivenbyy.Thereceiverintegratesthereceivedsignalsinthefirstslotandthelastslotofthesymbol.Attheendoftheintegrationperiodthereceiverdecideswhichslothasthelargestvalue.Weassumethattheelectronicnoise,togetherwithbackgroundnoise,is the dominant noise source when no power is transmitted, and the signal shot noise,together with electronic noise and background noise, is the dominant noise sourceotherwise.InbothcasesthenoisecanbemodeledbyadditivewhiteGaussiannoisethatisstatistically independentbetween time-slots.Thenoisehaszeromeanandcovarianceofσ2iandσ20forsignalandhole,respectively.Afterintegration,thesignalsy,aredescribedbythefollowingconditionaldensities:

(7.13)

and

(7.14)

When1istransmitted,μR=RPLTC0.5,σR2=2qRP1+σ2TH,μL=0andσL2=σ2THarethereceived optical signal and the accompanying receiver noise standard deviation,respectively,fortherightandleftslots.When0istransmittedμL=RPLTC0.5,σL2=2qRP1+σ2TH, μR = 0 and σR2 = σ2TH are the received optical signal and the accompanyingreceiver noise standard deviation, respectively, for the left and right slots. The BER isgivenby

(7.15)

Theaboveequationcanbesimplified,yielding

(7.16)

7.4 TheeffectofsynchronizationtimeoffsetonIPPMBERWe now examine in depth the effect of clock synchronization in one of the fourmodulationschemesintroducedintheprecedingsections.Inordertodefinetherequiredperformance of the clock we derive a mathematical model that describes the effect ofclock jitter on the performance of the BER model in (7.12). We follow similarassumptionstothoseoutlinedin[16,18,19]inordertoderivethemathematicalmodeloftheBERwithclockjitter.Theprimaryassumptionin(7.12)isthatthetime-slottimingisperfect and the decoder in the switch integrates the signals exactly over the T-secondinterval that constitutes a symbol. If a timeoffsetofΔ secondsoccursduring a symbolperiod,duetotimingerrorsinsynchronization,thentheintegrationoccursoveranoffsettimeinterval.Thatis,thedecoderstartsandstopsintegrationoveraT-secondintervalthatisdisplacedbyΔsecondsfromthatcontainingtheactualsymbolinformation.Assuming,without lossofgenerality, apositive timingoffset (0<Δ<TC) and equiprobable time-slots,thevariouseffectsontheintegrationstatisticsaresummarizedinTable7.1andFig.7.5. As a result of the time displacement, only a portion of the true signal energy isincludedinthesignalintegration,whilesomesignalenergycontributestotheintegrationin the adjacent time-slot, causing inter-symbol interference (ISI) in the form of energyspillover.Theeffectofthisinterferencedependsupontheformoftheadjacenttime-slot,i.e.whetheritcontainssignalenergyornot.Thebiterrorprobabilityforapositivetimingerror, Δ, for each of the four cases is given in the four equations (7.17)–(7.20). TheparametersforeachequationaregiveninTable7.1.

(7.17)

(7.18)

(7.19)

(7.20)

where

(7.21)

AveragingoverallthepossibilitiesgiveninTable7.1,theaverageBERisgivenby

(7.22)

Figure7.6depicts theaverageBERasafunctionofε,whereɛ is thepercentage timingerrorandisgivenbyɛ=Δ/TC.TheIPPMorderis2M=16withM=4.Theresultsshowarelatively fast increase in BER (system performance degradation) as the offset ɛ isincreased.Itiseasytoseethat,definingaBERequalto10−6astheacceptablelimit,the

systemperformanceisessentiallyruinedwhenɛisincreasedabove0.19,0.22,0.25,and0.275forRP1TC0.5/σTHvaluesof14,16,18,and20,respectively.

Table7.1 StatisticalparametersoftheIPPMmethodwithtimeoffset.(CourtesyofIEEE,reprintedwithpermissionfrom[16].)

Subsequentsymbol“hole”islocatedin

Transmittedsymbollocation

Conditionaldensities’parameters

N firsttime-slot

ofthe“hole”(u) Slot μ σ2

I No 0<u<2M−1 “hole” μI1=RP1Δ0.5

σ2I1=2qRP1Δ/TC+σ2TH

Adjacentslotto“hole”

μI2=RP1(TC−Δ)0.5

σ2I2=2qRP1(TC−Δ)/TC

Non-adjacentslotto“hole”

μI3=RP1TC0.5

σ2I3=2qRP1+σ2TH

II No u=2M “hole” μII1=RP1Δ0.5

σ2II1=2qRP1Δ/TC+σ2TH

Adjacentslotto“hole”

μII2=RP1(TC−Δ)0.5

σ2II2=2qRP1(TC−Δ)/T

Non-adjacentslotto“hole”

μII3=RP1TC0.5

σ2II3=2qRP1+σ2TH

III Yes 0<u<2M−1 “hole” μIII1=RP1Δ0.5

σ2III1=2qRP1Δ/TC+σ2

Adjacentslotto“hole”

μIII2=RP1(TC−Δ)0.5

σ2III2=2qRP1(TC−Δ)/T

Non-adjacentslotto“hole”

μIII3=RP1TC0.5

σ2III3=2qRP1+σ2TH

Slotat2Mlocation

μIII4=RP1(TC−Δ)0.5

σ2III4=2qRP1(TC−Δ)/T

IV Yes u=2M “hole” μIV1=0 σ2IV1=σ2TH

Adjacentslotto“hole”

μIV2=RP1(TC−Δ)0.5

σ2IV2=2qRP1(TC−Δ)/T

Non-adjacentslotto“hole”

μIV3=RP1TC0.5

σ2IV3=2qRP1+σ2TH

Figure7.5 FourcasesofatimeoffsetofΔsecondsthatcanoccurduringasymbolperiodduetotimingerrorsinsynchronization(courtesyofIEEE[16]).

Figure7.6 AverageBERasafunctionofɛ(courtesyofIEEE[16]).

7.4.1 TheeffectofclockjitteronIPPMBERAswehave seen in theprevioussection, it is crucial to synchronize the encoder to thebeginningof thesymbol,asotherwise thesystemperformancewill severelydeteriorate.The common method of synchronizing communication systems is based onsynchronizationcircuits, suchasaphased locked loop(PLL).APLL includesavoltagecontrol oscillator (VCO) and a phase detector.However, as systemnoise is part of anypracticalphysicalsystem,clockjitter isunavoidable.Therefore, theerrorprobabilityforthesynchronizedsystemistheexpectationofEq.(7.22)withrespecttothedistributionofclockjitter.Weassumethattheclockjittercanbemodeledbyasymmetricaldistributionaroundzero;inthatcase,theexpectationofBERavgisgivenby:

(7.23)

where f (x) is theprobabilitydensity functionof theclock jitter.Theclock timingerrorjitterisassumedtohaveaGaussiandistributionwithzeromeanandvarianceσ2CL:

(7.24)

Equation(7.23)wasevaluatednumericallyandtheresultingE[BERavg] isplottedinFig.7.7 versus the clock jitter variance. The results show a severe degradation of receiverperformance with increasing timing error variance. It is easy to see that E[BERavg]increasesalmostexponentiallyforalinearincreaseinσ2CL. It is important toemphasizethatE[BERavg]deteriorates forvery smallvaluesof clock jittervariancedue to the factthatE[BERavg]isnotalinearfunction.

Figure7.7 E[BERavg]versustheclockjittervariance(courtesyofIEEE[16]).

AnalyzingthesimulationresultsundertheassumptionthatM=4andinformationdataratesare38.4Mbps[11],96Mbps[11]and1GbpsresultsinvaluesofT=26.04nsec,T=10.04nsecandT=1nsec,aswellasTC=6.5nsec,TC=2.6nsecandTC=250psec,respectively.Inthatcase,forRP1TC0.5/σTH=20andBER=10−6,theσCLshouldbebetterthan 582 psec, 232 psec and 22 psec, respectively, for the given data rate, i.e. nearly 2ordersofmagnitudelowerthanthesymbolperiod,T.

7.5 SummaryIn this chapter we have described modulation methods that can be used in VLC andexpandedon the issueof synchronization.The IPPMsolutionwasexaminedand itwasdemonstrated that proper clock function is critical in order to provide accuratesynchronizationandtoachievetherequiredBER.

ThemainobjectiveinVLCdesignistoachieveenhancedlightingefficiency.However,inmanycasestheefficiencydecreaseswhentheswitchingrateisincreased.Inaddition,

thetimeresponseofphosphor-basedlightingsystemsisquiteslow.Asaresult,thelightsourcemodulationspeedislowandhasalarge,finiteresponsetime.Thesefactorslimitthecommunicationspeedandcontribute to inter-symbol interference (ISI).Withcurrenttechnology,themodulationmethodformatdiscusseddoesnotofferhighdatarate,butitisanticipatedthatanewgenerationofLEDscouldaddressthislimitationandbemodulatedatveryhighspeeds–upto10GHz[20,21].

References[1] Shlomi Arnon, “Optimised optical wireless car-to-traffic-light communication,”TransactionsonEmergingTelecommunicationsTechnologies25,660–665,2014.

[2] SeokJuLee,JaeKyunKwon,Sung-YoonJung,andYoung-HoonKwon,“Evaluationof visible light communication channel delay profiles for automotive applications,”EURASIPJournalonWirelessCommunicationsandNetworking(1),1–8,2012.

[3] Sang-YubLee,Jae-KyuLee,Duck-KeunPark,andSang-HyunPark,“Developmentof automotivemultimedia systemusingvisible light communications,” inMultimediaandUbiquitousEngineering,Springer,pp.219–225,2014.

[4] S.-H.Yu,OliverShih,H.-M.Tsai,andR.D.Roberts,“Smartautomotivelightingforvehiclesafety,”CommunicationsMagazine,IEEE51,(12),50–59,2013.

[5] Shun-HsiangYou,Shih-HaoChang,Hao-MinLin,andHsin-MuTsai,“Visible lightcommunications for scooter safety,” in Proceedings of the 11th Annual InternationalConferenceonMobileSystems,Applications,andServices,ACM,pp.509–510,2013.

[6] StefanSchmid,GiorgioCorbellini,StefanMangold,andThomasR.Gross,“LED-to-LED visible light communication networks,” in Proceedings of the FourteenthACMInternationalSymposiumonMobileadhocNetworkingandComputing,ACM,pp.1–10,2013.

[7] Nils Ole Tippenhauer, Domenico Giustiniano, and Stefan Mangold, “Toyscommunicating with LEDs: Enabling toy cars interaction,” in ConsumerCommunicationsandNetworkingConference(CCNC),pp.48–49,IEEE,2012.

[8] StefanSchmid,GiorgioCorbellini,StefanMangold,andThomasR.Gross,“LED-to-LED visible light communication networks,” in Proceedings of the FourteenthACMInternationalSymposiumonMobileadhocNetworkingandComputing,ACM,pp.1–10,2013.

[9] Nan Chi, Yuanquan Wang, Yiguang Wang, Xingxing Huang, and Xiaoyuan Lu,“Ultra-high-speed single red-green-blue light-emitting diode-based visible lightcommunicationsystemutilizingadvancedmodulationformats,”ChineseOpticsLetters12,(1),010605,2014.

[10] LianeGrobe,Anagnostis Paraskevopoulos, JonasHilt, et al., “High-speed visiblelight communication systems,” Communications Magazine, IEEE 51, (12), 60–66,2013.

[11] ShlomiArnon,JohnBarry,GeorgeKaragiannidis,RobertSchober,andMuratUysal,eds.,AdvancedOpticalWirelessCommunicationSystems,CambridgeUniversityPress,

2012.

[12] AhmadHelmiAzhar,T.Tran, andDominicO’Brien, “A gigabit/s indoorwirelesstransmission using MIMO-OFDM visible-light communications,” PhotonicsTechnologyLetters,IEEE25,(2),171–174,2013.

[13] Zabih Ghassemlooy, Wasiu Popoola, and Sujan Rajbhandari, Optical WirelessCommunications:SystemandChannelModellingwithMatlab®,CRCPress,2012.

[14] SridharRajagopal,RichardD.Roberts,andSang-KyuLim,“IEEE802.15.7visiblelight communication: Modulation schemes and dimming support,” CommunicationsMagazine,IEEE50,(3),72–82,2012.

[15] Joon-hoChoi,Eun-byeolCho,ZabihGhassemlooy,SoeunKim, andChungGhiuLee, “Visible light communications employing PPM and PWM formats forsimultaneousdatatransmissionanddimming,”OpticalandQuantumElectronics,1–14,2014.

[16] Shlomi Arnon, “The effect of clock jitter in visible light communicationapplications,”JournalofLightwaveTechnology30,(21),3434–3439,2012.

[17] IEEEStandard802.15.7forlocalandmetropolitanareanetworks–Part15.7:Short-rangewirelessopticalcommunicationusingvisiblelight.

[18] Chien-Chung Chen and Chester S. Gardner, “Performance of PLL synchronizedoptical PPM communication systems,” Communications, IEEE Transactions on 34,(10),988–994,1986.

[19] Robert M. Gagliardi, “The effect of timing errors in optical digital systems,”Communications,IEEETransactionson20,(2),87–93,1972.

[20] Jin-WeiShi,Che-WeiLin,WeiChen,etal.,“Veryhigh-speedGaN-basedcyanlightemitting diode on patterned sapphire substrate for 1 Gbps plastic optical fibercommunication,” in Optical Fiber Communication Conference, Optical Society ofAmerica,2012,p.JTh2A–18.

[21] GaryShambat,BryanEllis,ArkaMajumdaretal.,“Ultrafastdirectmodulationofasingle-modephotoniccrystalnanocavitylight-emittingdiode,”NatureCommunications2,539,2011.

PartsofthischapterarebasedonthearticlebyShlomiArnon,“Theeffectofclockjitterinvisible light communication applications,” Journal of Lightwave Technology, 30, (21),3434–3439,2012

8 DMTmodulationforVLC

Klaus-DieterLanger

8.1 IntroductionOpticalfreespacecommunicationusingvisibleradiation,i.e.light,hasbeenknownforavery long time. Some early examples are signaling using fire, the Heliograph usingsunlight, which is directed to the receiver by means of a mirror, or the PhotophoneinventedbyGrahamBell(1880).Duetotheoutstandingsuccessofradiotechnologiesandduetotheirintrinsicbenefits,uptonowopticalfreespacecommunicationhasremainedaniche technology.Oneof suchniche applications took advantageof the immunity frominterception, namely the so-called directed transmission formilitary aims duringWorldWar2andlateron.Differentapproachessuchasopticalwirelesscommunication(OWC)using fluorescent tubes are also well recorded in the patent literature but have neverachievedabreakthrough.

A revival of thiswayofwireless communication has come aboutwith the advent ofvisiblelightLEDsofincreasinglyhighopticalpower.Whiletheirapplicationinitiallywaslimitedtosignaling(e.g.telltaleorwarninglight),attheturnofthemillenniumitbecameapparentthatinfuturelightingwouldbedominatedbyLEDs.Henceforth,therehasbeengrowinginterestinapplicationsusingLED-basedOWC,orwhichcombinethefunctionsof lighting and opticalwireless data transmission.At the same time, the common termVisible Light Communications (VLC) was coined for this kind of communication. ThemajorreasonsforthesteadilyrisinginterestinVLCarethelifetimeandimprovedopticalpower of white light LEDs particularly, and their progressive adoption, as well as thesimplicityofLEDmodulationviatheirdrivingcurrentatamodulationbandwidthinthelowerMHzrange,seee.g.[1–3].Moreover,theproliferationofmobileapplicationsusingradio frequencies has accentuated concerns about the adequate availability of radio-frequencybandsand the limitsof transmissioncapacity incurrentwirelessnetworks,aswellastherelateddatasecurityissues.Inthisrespect,VLCcanofferanadditionaloptionforlocalwirelessdatalinkswhereradiolinksarenotdesiredornotpossible[4–6].

The building blocks of a genericVLC system are shown in Fig. 8.1. Regarding thehistoricalbackgroundofVLCanditsusefor transferringmessagesviaschemessuchastheMorsecode,itseemstobeobviousandstraightforwardtoapplyplainon-offkeying(OOK). Indeed, simple experimental VLC systems use OOK realized by intensitymodulation(IM).At thereceivingside,directdetection isappliedusingaphotodetectorfor optoelectronic signal conversion. While image sensors can be used in low-speedsystems(cf.[3]),highdataratescallforaSi-PINoravalanchephotodiode(APD)asthedeviceforopticaldetection.InsuchconfigurationsbasedonOOKandwhitelightLEDsintendedforlighting,transmissionspeedsofe.g.230Mbit/shavebeenachieved[7].

Figure8.1 Buildingblocksofabasicshort-haulvisiblelighttransmissionsystem.

This chapter is focused on indoor applications ofVLC,where LEDs are used as anoptical source for either (pure) high-speed data transmission or for lighting plus datatransmission at bit rates in the upper Mbit/s range (up to Gbit/s speed). Sampleapplications of high-capacity indoor links are presented in the next section. Given themodulationbandwidthofferedbycurrentLEDs,incontrasttotheOOKcasethetargeteddataratesrequireadvancedandhighlyspectrumefficientsolutionsformodulation,suchasdiscretemultitone (DMT),which is addressed in this chapter. In fact, the highest bitratesdemonstratedsofarinasingleVLCchannelusesuchmodulationschemes[8–11].

WhenconsideringVLCsystemsasmentionedbefore, the reach shouldbewithin therange of several meters (say up to ~20 m). Depending on applications such as databroadcast,videostreamingorfile transfer,eitherunidirectionalorbidirectional linksarerequired in point-to-point (P-t-P) or point-to-multipoint (P-t-MP) configuration.Additionalrequirements in thedual-usecaseofLEDsfor lightinganddata transmissioninclude illumination according to the well-established lighting standards and featureswithout restrictions, no flickering light, dimming, etc., cf. [12]. However, it has to bementioned that, for instance, the present version of the IEEE standard on VLC onlyconsidersOOK,variablepulsepositionmodulation(VPPM),andcolorshiftkeying(CSK)asaspecialschemewithrespecttomultipleopticalsourcesofdifferentcolor,whileDMTisnotyetincluded[13–15].

Thechapterisstructuredinthefollowingway.Wecontinuewithabriefdiscussionofsometypicalindoorapplicationscenarios,astheyareofinterestforthegeneralpublicandin industrial sectors.Thenextsection is devoted to the relevant characteristics ofwhitelightLEDsasthekeyVLCelement.Inaddition,theopticalwirelesschannelcapacityisaddressed,aswellasLEDmodulationforexploitingthecapacity,andmajorissuessuchastheeffectofLEDnon-linearity.

The main part of this chapter presents the DMT modulation scheme and variantsthereof,includingrelatedsignalprocessingaswellasbitandpowerloading.Substantialeffectsandconsequencessuchassignalclipping,peak-to-averagepowerratio(PAPR)andthe influence of channel variation are detailed in this section. Subsequently, severalrecently proposed improvements of the DMT modulation scheme are introduced andvariousapproachesarecompared.Onthatbasis,Section8.6examines importantaspectsofsystemdesign,implementationissues,anddemonstrationsinstepwithactualpractice.Asummaryandanoutlookareincludedattheendofthischapter.

8.2 IndoorapplicationscenariosVLCcanbe applied in quite different scenarios. In particular,whenhighdata rates areconsidered the type of link between transmitter (Tx) and receiver (Rx) is crucial.

According to the mode of propagation of light, there are two generic types of indooropticalwirelesslinks.Firstlythedirectedlink,whichreliesonanon-blockedline-of-sight(LOS)betweenahighlydirectedTxandanarrowfield-of-view(FOV)Rx.Secondly,thediffuse link,characterizedbyawide-beamTxanda largeFOVRx,where thenon-LOSlight path relies on numerous signal reflections off the walls and surfaces of objectspresentintheroom[16].

LOSlinksexperienceminimalpathloss,areratherfreefrommultipathinducedsignaldistortion,andareabletodiminishtheinfluenceofambientlight.AslongastheLOSisnot blocked, the link performance only depends on the available power budget.Hence,very high transmission rates are shown to be possible. On the other hand, LOS linksrequirealignmentoftransceiversandgenericallyprovideaverysmallcoverage.

AdiffuselinkoperatesentirelywithoutLOS,whichresultsinanincreasedrobustnessagainstshadowingandsupportofhighusermobilitywithinalargecoveragearea.Thus,adiffuselinkscenarioenablesP-t-MPcommunicationandingeneralismostdesirablefromthe user’s point of view. This is why it evoked much interest from the researchcommunity.However, diffuse links suffer fromhigh optical path loss (i.e., they requirelargeropticalpowers)andareseriouslylimitedbyinter-symbolinterference(ISI)becauseofmultipathdispersion,whichpresentsamajordegradationfactorathighertransmissionrates. Beside the power budget, the achievable transmission rate depends also on roomcharacteristicssuchassize,reflectioncoefficientsofsurfaces,etc.ItshouldbementionedthattheeffectsofmultipathfadingarenegligibleinOWCduetosquare-lawdetectiononaphotodiode(PD)ofhugesizecomparedtotheincomingsignalwavelength.Thisgreatlysimplifiesthelinkdesign.

ManyVLCapplicationscallforcombiningthemobilityofadiffuselinkandthehigh-speedcapabilityofaLOSlink.Inordertobenefitfromtheadvantagesofbothconnectiontypes,anon-directedLOS(NLOS)linkisfrequentlyconsideredasanalternative.Insuchalink,LOSanddiffusesignalcomponentsaresimultaneouslypresentattheRx(assuminga non-blocked LOS path). The equivalent channel response as shown in Fig. 8.2 ischaracterizedbyhighdynamicsinbothbandwidthandgain,dependingaboveall,ontheLOSprominence.Consequently,theratioofLOSanddiffusesignalcomponentsattheRx,oftendescribedbytheRicianK-factor,stronglyinfluencesachievabledatarate,impactofISI,andambientlight,whiletheTx-RxalignmentisrelaxedandP-t-MPcommunicationispossible.

Figure8.2 Channelimpulseresponseh(t)asasuperpositionoftheLOSanddiffusechannelresponses,whereΔTindicatesthedelaybetweenarrivaloftheLOSsignalandthefirstreflectionatthereceiver;1/τisthedecayconstantofthediffusechannel.

Table8.1 summarizes important features of thebasicTx-Rx configurations inOWC.Because in addition to the link type the lighting scenariomay also varywhile used forVLC,adynamicdatarateadaptationappearsnecessary,asalreadyproposedin[17]and[18]. Such a feature would enable efficient use of the channel and its instantaneouscharacter,andthuswouldsubstantiallycontributetomakingVLClinksrobustandreadyforvarioususecases.Thereareseveralapproachesandoptionsbasedonthefundamentalprinciplesofhigh-speedVLCtransmission,whichareaddressedinthischapter.Thus,wewill not dive deeper into thematter of dynamic data rate adaptation. In the following,someillustrativeVLCscenarios,mainlyusinghigh-speedtransmission,arediscussed.

Table 8.1 Comparison of commonly present link configurations for opticalwirelessindoorsystems.

Linktype DirectedLOS Non-directedLOS Diffuse

Linkrate highest high moderate

Beampointing yes coarse no

Beamblocking yes relaxed no

Usermobility low medium high

Dispersion(multipath) none medium high

Pathloss low extended high

Impactofambientnoise low medium high

Purewirelessbroadcastlinkscouldprovidepassengerinformation,etc.,viathegenerallighting,e.g.onundergroundstations(Fig.8.3)orinsidethemetrocars,wherelightsarealwaysonanyway.Suchindoorsystemsrequireunidirectionaldatatransfer(streaming),afewmeterstransmissionreachandawideFOV,whichisinherentlygivenbylighting.The

basic functionality of such an application is quite similar to the traditional and mostcommonlyusedopticalwirelessapplication(atverylowspeed,butusinginfraredlight),namelytheremotecontrol.However,today’stechnologycanprovideveryhighdataratesuptotheGbit/srange.

Figure8.3 Broadcastofmultimediapassengerinformationbyvisiblelightonanundergroundstation.

AnotherexampleofVLCcombinedwithgenerallighting,asshowninFig.8.4,isalsoknownasopticalWiFi orLi-Fi (light fidelity). Inwireless local areanetwork (WLAN)scenarioslikethis, thedownlinkisprovidedinthesamewayasabove,whiletheuplinkfrom the laptop to an access point at the ceiling can be established, e.g. using a LOSinfrared (IR) link (see e.g. [19]). Such a system can offer bidirectional communicationwithwideFOVP-t-MPvisiblelightdownlinksatspeedsintherangeofseveralMbit/sormuchmoredependingonthelinkconditions(LOSordiffuse),andIRdirectedLOSP-t-PuplinksofafewMbit/s,assumingbothfairTx-Rxalignmentandopticalpower[20].

Figure8.4 OpticalwirelessLANscenariowherethedownlinkisprovidedbyVLCwhiletheuplinkisrealized,forexample,byinfraredlight.

Machine-to-machine communication is expected to be a further broad field of VLCapplication,e.g.withrespecttowirelessexchangeofhighdatavolumeswithinsmallanddensecommunicationcells.Itisalsoassumedthatinindustrialenvironmentstheremaybeharshelectromagnetic conditionsgoingalongwith thehighestdemandson security andreliability making it difficult or even impossible to use radio-frequency systems. VLCdirectedLOSlinksontheotherhandcouldprovidesuitablehigh-speedbidirectionallinks.Asanexample,Fig.8.5illustratesascenariowhereperformancetestsofproductsunderassemblyhappenwhile theunits aremovingonaconveyor, anddataexchangewithanassessingcentralserverisrequiredatthesametime.

Figure8.5 BidirectionalVLCappliedinaproductionline,e.g.inordertotriggerthedeviceunderassemblyforperformancetestsandtoreceivetheresultsduringshiftontheconveyor.

8.3 Aspectsofhigh-speedVLCtransmissionInparticular,whencombinedwith lighting,keyenablers forVLCare that theLEDsdonot flicker, that only a minimum of extra power is needed, that VLC works at thecommonlyusedbrightness levels,andthatwell-establishedfunctionsof lightingsuchasdimming are possible without reservation [13, 14, 21]. These aspects are addressed inSection 8.6.1. From the transmission technology point of view, the LED modulationbandwidth is just as crucial as the wireless channel capacity and how to exploit itefficiently.Thesesubjectswillbediscussedinthesubsequentsections.

8.3.1 LEDmodulationbandwidthWhite lightLEDs, askeyVLCelements, aremanufacturedbyappropriatelyadding thelightofthreeorfourcoloredemitters,i.e.red,greenandblue(RGB,trichromatic)orred,yellow, green and blue (RYGB, tetrachromatic) LED devices, respectively. As analternative, a single blue emitter chip coated with a yellowish phosphor layer (YB,dichromatic)isused.TheRGBandRYGBwhitelightsourcesprovidethedesiredspectraloutput,butarehardware-intensiveasmultipleLEDsarerequired.Inaddition,theytendtorender pastel colors unnaturally, a fact which is largely responsible for the poor color-renderingindexofRGBwhitelight.Asaresult,theYB-LEDsarecurrentlythedeviceofchoiceforillumination,andthusalsoforlow-costVLC.

While typical RGB and RYGB LEDs created for lighting purposes offer a 3 dBmodulationbandwidthofsometenMHz,thebandwidthofYB-LEDsismuchlower,i.e.

intheorderofafewMHz,seee.g.[22,23].Thisisbecauseoftheslowtemporalresponseof the yellow phosphor layer. On the other hand, as explained in [24], a bandwidth ofseveraltensofMHzcouldbeexpectedintheabsenceofthephosphorlayer.Accordingly,itisgoodpracticetoplaceablueopticalfilterinfrontofthePDattheRxsideinordertoreceiveonlythebluecomponentofthelightandtotakeadvantageofitslargermodulationbandwidth[25]. This, however, comes alongwith a lower power budget and signal-to-noiseratio(SNR),asthemajorportionofthereceivedlightspectrumisfilteredout[26].Alternatively,equalizationtechniquescanbeusedtocombattheinfluenceofthephosphorlayer[27,28].Tociteanexample,anarrayof16YB-LEDswasmodifiedinthatwaytohaveabandwidthof25MHzwithoutbluefiltering[29].

Regarding the LED modulation bandwidth, in a more general sense it is worthmentioning that according to experimental studies the modulation bandwidth of whitelightaswellasofcoloredLEDscanbeexploitedfarbeyondthe3dBdrop.Forinstance,in [30] a bandwidth of 100MHz has been used for modulation, while the LED 3 dBbandwidthwasabout35MHz.

Because, in addition to the commonly used (inorganic) LEDs in lighting systems,organicLEDs(OLEDs)alsobecomeattractiveforreplacinglargearealuminoussources,it should be noted that OLEDs are considered in VLC too. However, they offer aninherently lowmodulationbandwidth in the rangeof about 100kHz [31,32], and thustheyarenotinthescopeofthischapter.

8.3.2 ChannelcapacityAs briefly discussed in the context of VLC applications above and summarizedqualitatively inTable8.1, the capacityof theopticalwireless channel in agiven indoorscenariostronglydependsonthepresenceofbothLOSanddiffusesignalcomponents.AusefulmeansofdescribingthechannelstatebyasingleparameteristheRicianK-factor,which isdefinedas the ratioof the (electrical)LOSanddiffusechannelgain (loss), i.e.K[dB]=20log(ηLOS/ηDIFF).Followingtheanalysisin[33,34],whereanemptymodelroom and realistic parameters are assumed as an example, the frequency response iscalculated and illustrated in Fig. 8.6 (a). The diagram shows the composite channelfrequencyresponsemagnitudeobtainedfromtheanalyticalmodelforseveralillustrativeK-factors. Clearly, the channel response highly depends on the LOS prominence(described by the K-factor). Where the LOS is weak or blocked, the response isapproximately low-pass and the bandwidth is quite poor. As the LOS gets morepronounced,thechannelresponsevariesuntilitbecomesalmostflatforsufficientlylargeK-factors, renderingbandwidthsup to anorderofmagnitudegreater than in thediffusecase.Thenotches in thechannelcharacteristicaredue todestructive interferenceof thetwofrequencycomponents.Assuming,e.g.arealisticworkingareahot-spotscenario,theK-factorspanofinterestisabout−20to+25dB.

Figure8.6 (a)Frequencyresponseoftheopticalwirelesschannelinamodelroomfor

differentK-factorvalues.Thederivationisoutlinedin[33],whereΔt=10nsdelaybetweenLOSanddiffusesignalcomponentsisassumed.(b)UppercapacityboundCofanadaptiveandanon-adaptiveOWCsystem(bandwidth100MHz)asafunctionofthechannelstateandthemeanopticalpowerPOasparameter.

Thesimplestwaytoobtainreliableconnectivityunderallchannelconditionswithinthecoverage area of a given scenario is to realize a statically designed system, withtransmissionparametersfixedaccordingtotheworstcase.However,suchasystemwouldnotbeabletobenefitfromthechannelproperties.Inordertoillustratethepotentialoftheindooropticalwirelesschannel,theuppercapacityboundintermsoftransmissionratefordifferentchannelstatesandtwoopticalpowerlimits(PO=0.1and0.4W)ispresentedinFig. 8.6 (b). For comparison, the curves of a statically designed system, aimed toguarantee a constant transmission performance in the whole area of interest are alsoshown. From Fig. 8.6 (b) it becomes clear that it is extremely attractive to exploit thechannel capacity, in particular at growingK-factors, where the link becomesmore andmoretransparent.Thegain(withrespecttoaworst-casedesign)growswiththeK-factor,and also by increasing the optical transmit power,where it becomes significant even atdecreasing values ofK. The information rate also depends on the electrical bandwidth,whichmaybelimitedbytheLEDasoutlinedaboveinSection8.3.1.

Thecalculationsweremadewhenthetransmitpowerwasallocatedasbestpossibletothesubcarriersofamultiplesubcarriermodulation(MSM)scheme.Indoingsoitisalsoshownin[35]thatoptimumpowerallocationgivesnegligibleadvancewhentheelectricalbandwidth is limited to about 20 MHz, but offers more improvement at higherbandwidths,particularlyforlowTxpowersunderMSMmethods,whicharethefocusofthischapter.The topicofTxpowerdynamic rangeand its influenceon the informationrateisanalyzedin[36].ItisshowntherethataTxwithawidelineardynamicrangeof20dBormoreprovidessufficientelectricalpowerforOWCwithanopticaltransmitpowerclosetotheboundariesofthedynamicrange,wheretheLEDappearstobeofforpoweredclose to itsmaximum. It is also shown there that an average optical power sweepover50% of the dynamic range can be accommodated using an appropriate modulationscheme,iftheinformationrateisreducedbyonly~10%.

Inadditiontothetheoreticalworkdiscussedabove,resultsofanexperimentalchannelcharacterizationcanbefoundforexamplein[37],whereaVLCchannelbandwidthof63MHzisreportedforacertainroomgeometry.

The cited work and further results indicate that although the well-known Shannonchannel capacity formula for a band-limited, average power-limited additive whiteGaussiannoise(AWGN)channelcannotbeapplieddirectlytotheVLCchannel,itcanbeadaptedinmanycasestofindthelimitofreliableOWC.Asaconclusion,itisnotedthatin non-diffuse situations the channel presents much more capacity than offered by theLED,dependingontheRicianK-factor.Thiscalls foradvancedspectrumefficientLEDmodulation if high-speed communication is desired. In order to maximize the systemthroughput,while reliable operation and full coverage aremaintained, anOWC systemcouldbedesignedthatisbandwidth-andrate-adaptive.Suchanadaptivesystemreducesthe data rate under adverse channel conditions until a desired bit error performance isattained. However, such a feature requires a reliable low-speed feedback link for

transferringthenecessarychannelstateinformationfromtheRxtotheTx.

8.3.3 Considerationsonhigh-speedLEDmodulationWith the aimof developing simple and low-costVLC systems, it is an obvious step tochoosesimpleandwell-knownmodulationformatssuchasOOK,pulse-widthmodulation(PWM), or M-ary pulse-amplitude modulation (M-PAM), which can be applied in astraightforward way using intensity modulation and direct detection. The literatureprovides a lot of proposals and demonstrations on a low-speed level, but also someexampleswhere several100Mbit/shavebeenachievedusingOOK,cf. for instance [7,38].However,ifhighertransmissionspeedsareneeded,theabovementionedmodulationschemesbegin to suffer from theundesired effects of ISI due to the non-flat frequencyresponseoftheopticalwirelesschannel[15].Hence,amoreresilientand,inviewofthelimitedLEDbandwidth,amorespectrumefficienttechniquesuchasMSMisrequired.

In the VLC transmission chain, the LED device is a major source of non-lineardistortion.Thisisduetoanon-lineartransferfunctionwithintheLED’soperatingrange.As the significance of such behavior strongly depends on themodulation scheme used,non-linearityaspectshavebeenextensivelystudiedinnumerouspapers,suchas[39–43].TheLEDnon-linearityisofparticularimportanceifspectrallyefficientMSMmodulationis applied in order to exploit the LED’s bandwidth. Using such modulation schemes,which this chapter is focusedon, the dynamic range and linearity of the emitter devicemayseverelylimit theachievableperformance[4]. Inorder tomitigatesuchrestrictionstheselectionofapropermodulationformatiscrucial[44],[45],whiletheLEDoperatingpointhas tobedefinedcarefully [46],andsignalclippingprior tomodulationhas tobeconsidered[47].

Beyondthenon-linearityissue,furthereffectsontheLEDperformance,suchasLEDdegradationdue toageingor junction temperaturevariationhave tobeconsidered.Firststudiesonthissubjecthavebeenpublished,e.g.in[48],however,furtherinvestigationsontheLEDbehaviorunderVLCoperationconditionsarenecessary.

8.4 DMTmodulationandvariantsMSM techniques are modulation schemes where information is modulated ontoorthogonal subcarriers located in the frequency band considered. The sum of themodulatedsubcarriersisthenmodulatedontotheinstantaneouspowerofthetransmitter.UsuallyMSM is implemented by orthogonal frequency divisionmultiplexing (OFDM),which has been widely employed in both wired and wireless digital communications.Applications of the basic OFDM principle includeWLAN and terrestrial digital videobroadcasting(DVB-T),aswellasdigitalsubscriberlinesystems(xDSL)andpowerlinecommunication (PLC), where a baseband version of OFDM, better known as discretemultitone (DMT) modulation is used. Due to the capability to mitigate ISI, and otheradvantages including the ability to adapt easily to different channels, MSM was alsoconsideredforOWC[49].

ConventionalsystemssuchasWLANorDVB-Tusecoherenttransmission,wherethe

signals are in general complex and bipolar. The same applies to laser-based long-hauloptical transmission systems using single-mode fibers. Concerning optical multimodetransmission as in VLC, it is, however, extremely difficult to collect substantial signalpowerat theRx ina singleelectromagneticmode. In the finalanalysis, thismeans thatintensitymodulationwithdirectdetection(IM/DD),asalsousedforinstanceinOOKandPWM-basedsystems,has tobeconsideredas theonlypracticable transmissionmethod.Thus, only the light intensity (and not the phase) represents the information to betransmitted,i.e.,thetransmittedsignalisofrealandnon-negative(unipolar)value.Atthereceivingend,aphotodetectorproducesanelectricalcurrentproportionaltothereceivedpower,andaccordinglyproportionaltothesquareofthereceivedelectricalfield[16].ThedifferencesbetweentraditionalOFDMtransmissionusingcoherentdetectionandIM/DDsystemsarebrieflysummarizedinTable8.2.

Table 8.2 Contrast between typical OFDM systems and optical multimodetransmissionbasedonintensitymodulationanddirectdetection.

Areaofapplication Typicalexample Carrierof

informationTypeofdetection

Specialreceiverrequirement

ElectricalOFDMtransmission

Radiotransmission,e.g.WiFi,DVB-T

Electricalfield

Coherent Localoscillator

OpticalOFDMtransmission

Long-haulhigh-speedsinglemodefiber(SMF)links

Opticalfield Coherent(onlyoneopticalmode)

Localoscillator

OpticalIM/DDtransmission

Multi-modefiber(MMF,POF)links,OWC

Opticalintensity

Direct(multipleopticalmodes)

None

In consequence of IM/DD transmission, the conventionalOFDMmodulation schemecannot be directly applied in optical multimode systems. Therefore, researchers havedevoted significant efforts to designing OFDM-based modulation schemes, which arepurelyunipolar.UsingOFDMinVLCwasfirstproposedbyTanakaetal.,andtheirbasicstudiescanbefoundin[50].

Thetasktocreateaunipolarsignalfor transmissioniscommonlysolvedbyaddingaDC bias to the bipolar OFDM signal [18, 25]. We call this schemeDC-biased DMT,however,itisalsoknownasDC-biasedOFDMandDC-clippedOFDM.Anothermethodclips the entire negative excursionof theOFDMwaveform. Impairments fromclippingnoise are avoided by appropriately choosing the subcarrier frequencies formodulation.This technique is calledasymmetrically clipped optical OFDM (ACO-OFDM) [51]. A

third method also clips the entire negative signal excursion, but modulates only theimaginarypartsofthesubcarrierssuchthattheclippingnoisebecomesorthogonaltothedesired signal. This technique is well-known as pulse-amplitude-modulated discretemultitonemodulation (PAM-DMT)[52].Thesemodulationschemesand theirpropertieswhenappliedtoVLCarediscussedinthesubsequentsections.

8.4.1 DC-biasedDMTDMTasabasebandversionofOFDMmodulationisakeytechniqueusedonslowlytime-varyingtwo-waychannels,e.g.incopper-basedxDSLandPLCsystems.Itisimportanttoknow thatDMT isalsoof relevance for low-cost short-rangeoptical transmissionusingmultimode silica fibers and plastic optical fibers [52, 53]. Usually, modulation ontomultiple subcarriers of different frequencies for simultaneous transmission, as well asdemodulation,arebasedondiscreteFouriertransformation(DFT)[54,55].Thus,inversefast Fourier transform (IFFT) and fast Fourier transform (FFT) are the main buildingblocksoftheTxandRx,respectively.

On theTx side, first of all the serial data stream is partitioned intomultiple parallelstreams of lower data rates, typically followed by mapping to a quadrature amplitudemodulation (QAM) constellation, as illustrated in Fig. 8.7. A straightforward way toobtainreal-valuedtimedomainsignals,asrequiredinLED-basedVLC,istousethewell-known property that the N-DFT of a real-valued sequence has conjugated symmetriccoefficientsaroundthepointN/2[56].Thismeansthatbyenforcingconjugatesymmetry(oftenreferredtoasHermitiansymmetry)ontheIFFTinputvector[X]inthefrequencydomain, a real-valued time domain signal can be directly obtained at the output of theIFFTblock.

Figure8.7 BuildingblocksofDMT-basedtransmissionoveranopticalIM/DDchannel.Notethatthetasksofsatisfyingnon-negativityanddigital-to-analogconversionmaybeswappeddependingonthehardwareimplementationoftheDCbiasing.

Following this approach, withN / 2 (actually (N / 2) − 1)) independent subcarriersenvisaged in thesystemtocarry information,anN-IFFTblock is required togenerateareal-valued OFDM/DMT symbol. According to the Cooley–Tukey algorithm [56], thecomplexityof theFFToperationscaleswithN log2N.Hence,asamoderatenumberofsubcarriersisconsideredinopticalwirelesssystems,thesizeoftheDFTblocksisnotan

implementation issue.On the other hand,DFT enables digital implementation ofDMTmodulationwithoutprohibitiveanalogfilterbanks.

The IFFT input vectorX = [X0X1⋯XN−1]T consists of the data to be transmitted(elementsX1,X2,⋯,X(N/2)−1)and thefurtherelementsdefinedaccording to theHermitiansymmetryconstraintas

(8.1)

ThefirstinputX0,correspondingtozerofrequency,mustbereal-valuedandisgenerallyleftunmodulated.ItcanbesettozerooralternativelytotheDCleveloftheoutputsignal(seebelow).

Aftercreationoftheinputvector,[X]isfedintotheN-pointIFFTblock,asshowninFig. 8.7. Assuming data symbols of complex-valued modulation formats (usually M-QAM),wheretheinputvectorelementsappearintheformXn=an+jbn,thetimesamplesattheN-IFFToutputare

(8.2)

wherek=0,1,…,N−1denotes the indexof the timedomainsample.Apart fromtheconjugate-symmetry property of the input vector given by Eq. (8.1), the symmetrypropertyoftheDFTtwiddlefactorsej2π(N−k)k/N=e−j2πNk /NhasalsobeenexploitedinEq.(8.2).Obviously,asumof(N/2)−1sampledreal-valuedcosinusoidsisobtainedattheIFFToutput.

InordertomitigateeffectsofthemultipathchannelandtoavoidISI,OFDMandDMTuseaguardinterval,whichisplacedbetweenthetransmittedsymbols(blocks)inthetimedomain.ThisguardbandisformedsimplybytakinganumberofLsamplesfromtheendofeachsymbol,andcopyingthemasitsprefix,asshowninFig.8.8.Itishencereferredtoas cyclic prefix (CP). This (M + L)-point sequence corresponds to the samples of themulticarrierDMTtime-discretesequencetobetransmitted,whichisreferredtoasaDMTsymbol.

Figure8.8 GenerationofthecyclicprefixguardintervalandstructureofaDMTsymbol.CP:cyclicprefix.

InordertoreceiveandproperlydemodulatetheDMTsymbols,twoconditionshavetobesatisfied.Firstly, the lengthof theDMTsymbolwithoutCPshouldbe longer thanorequaltothedurationofchannelimpulseresponseh(t)inordertoavoidISI.Additionally,theCP length shouldbechosen so that itsduration is longer thanor equal to thedelayspreadofh(t).AlthoughtheCPintroducessomeredundancy,andthusreducestheoveralldatarate,iteliminatesbothISIandinter-carrierinterferencefromthereceivedsignalandisthekeytosimpleequalizationinOFDM[55].

WhenassumingalimitedsignalbandwidthBinthebaseband,whichisdividedamongN/2independentsubcarriers,thesubcarrierspacingis∆f=2B/N,whilethefrequenciesin the N-IFFT block cover the bandwidth of 2B because of the Hermitian symmetryconstraint. The period of a DMT symbol, containingN time samples, isTFFT = 1/∆ f,which leads to the sample interval Tsam = 1/2B in the time domain. According to thesamplingtheorem,this issufficient tocompletelydeterminethecontinuoussignalat theoutputofthedigital-to-analog(D/A)converter.AstheCPoflengthTCP=LTsamhastobeaddedaftertheIFFT,theactualdurationofthereal-valuedDMTsymbolisTDMT=TCP+TFFT.ConsideringinadditionEq.(8.2)andtheidentity

(8.3)

wherefn=n∆f,n=1,2,…,N−1andtk=kTsamfork=0,1,…,N−1,thecontinuous-timesignalafterD/Aconversioncanbeexpressedas

(8.4)

Inthisexpression, andφn=arctan(bn/an)aretheamplitudeandtheinitialphaseofeachcosinusoidonthefrequencyoccupiedbythenthsubcarrier,determinedbythe amplitude and phase of the complex-valued symbol Xn at the corresponding IFFTinput.Notethat thesignalx(t)couldalsobeobtainedwithabankof(N /2)−1analogfilters. After D/A conversion, an analog low-pass filter is implemented to suppress the

aliasingspectra.Atthesametime,thisfilterperformsinterpolationofthediscretesignalwaveform.

Figures8.9and8.10illustrateanexampleofthesignalsattheinputandoutputoftheDMTmodulator.Inthisexample,threeoutofN=16orthogonalcarriersinabandwidthofB=20MHzaremodulatedby16-QAMtogenerateaninputvector[X]=[0,(1+j),(3−j),0,(−3+3j),0,0,0]T(beforeHermitiansymmetryenforcement).Theinputofthe16-IFFTblock isshowninFig.8.9,while thecontributionsof individual subcarriersat theoutput aswell as the resulting output signal are illustrated in Fig. 8.10 (a)–(c) and (d)respectively.

Figure8.9 Exampleof16-QAMsymbolsonindividualsubcarriersinf-domainattheIFFTinputaccordingtotheinputvector[X]=[0,(1+j),(3−j),0,(−3+3j),0,0,0]TbeforeHermitiansymmetryenforcement.FurtherparametersareN=16,B=20MHz,andL=2.

Figure8.10 (a)–(c)Theindividualcontributionsofmodulatedsubcarriers;(d)theDMTsymbolinthet-domain.BothsampledandcontinuoussignalsbeforeandafterD/Aconversionareshown.NotethattheamplitudesarescaledbyNtoachievethesamelevelasinthef-domain.Inthisexample,thereal-valuedbipolarDMTsignalwithN=16subcarriersusedintheexample(d)ismadeunipolarbyaddinganappropriateDCbias(e).

IfthebiaswasnotintroduceddirectlyatthesysteminputbysettingX0appropriately,i.e.,ifX0=0waschosen,non-negativityofthetransmitsignalx(t)hastobeachievedafterD/A conversion. A very common and simplemethod is to add a fixed DC bias to thebipolarDMTsignal,as illustrated inFig.8.10 (e), cf. for instance [25,49,52,57].The

requiredDCbiasisequaltothemaximumnegativeamplitudeoftheDMTsignal.Owingto thehighPAPRofDMTsignals, abiasofat least twice the standarddeviationof thebipolar DMT signal distribution is required to minimize clipping [58]. Any remainingnegativevalueswouldbeclipped,whichalsoappliestopositivepeaksuponexceedingthelimitingamplitude.Suchsymmetricalorasymmetricalclippingintroducesclippingnoiseand thus affects transmission. In practice, controlled clipping of high negative peaks infront of the optical source is often permitted, given an unlikely occurrence in a DMTsignal. Then, as long as the effect of clipping on the link performance is tolerable, therequiredDCcomponentcanbereducedtoacertainextent[33],cf.alsoSection8.6.1.AnoptimalDCbiasofasymmetricallyclippedsignalcanbeinferred,e.g.from[59,60].Itisshown there that iterative decoding with clipping noise estimation and subtraction canreducethebiterrorratio(BER)attheexpenseofanincreasedcomputationalcomplexity.

ReturningtoFig.8.7, thesignalreachestheRxafterundergoingtheinfluencesofthetime-dispersivechannelandambientlightasthedominantsourceofnoiseinthechannel.Asimple(low-cost)photodetectorisusedtoconverttheIMsignaltoanelectricalsignal,while the bias component is discarded by AC-coupling. After analog-to-digital (A/D)conversion, theCPis removedandN-pointFFTprocessing isperformed.Thanks to thepreservedorthogonality,thesubcarrierscanbeprocessedseparately.Inprinciple,onlytheoutputs n = 1, 2,…, (N / 2) − 1 need to be further regarded, since they are the onescarryinginformation.Becausethefrequencyresponseoverthesubcarrierbandwidthcanbeconsideredasflatfading,thesignalisusuallyequalizedsimplybymeansofasingle-tap linear feed-forward equalizer with zero forcing or minimum mean squared error(MMSE)criteriabeforedecoding.

Adding the DC bias at the Tx results in extra power consumption to an equivalentextent.Ifthelightsourceisusedatthesametimeforillumination,thisamountofpowerfulfillsitspurposewithrespecttolightingandisthusnotwasted.Onlyifilluminationisnot required, such as in the uplink of a Li-Fi system, the DC bias can significantlyjeopardizeenergyefficiency[15].

8.4.2 AsymmetricallyclippedopticalOFDM(ACO-OFDM)The main objective of improving the preparation of OFDM signals for IM/DDtransmission is to reduce the DC bias, which is indispensable in the DC-biased DMTscheme,toaminimumvalue.NotethattheLEDturn-onvoltageinfactdefinesthelowerboundofthebias.Forthesakeofsimplicitythisparameterisneglectedinthefollowing.Approaches for improving the power efficiency are driven by the basic aim to create areal-valuedbipolartimedomainsignal,wheretheentireinformationispresentatleastinthepositiveparts.Thiswouldenablesimplyclippingthenegativepartsofthesignalwhenmodulatingtheopticalsource.ThemainembodimentofsuchasymmetricclippingisthemethodofACO-OFDMintroducedbyArmstrongetal.[51,58].

WhilethebasicfunctionblocksarethesameasinDC-biasedDMT,theACO-OFDMscheme exploits the properties of Fourier transform by introducing a constraint on thesubcarriersusedformodulation,i.e.,onlyodd-numberedsubcarriersareusedwhereastheeven subcarriers are set to zero. Considering again Hermitian symmetry, up to N / 4subcarriers(CPincluded)areusedformodulation[51,61].AsillustratedinFig.8.11,the

IFFToutputpresentstimesamplesalongthezero-axis,wherethenegativepartisananti-symmetrical copy of the positive part. Finally, thismeans that the information of bothparts is the same. For that reason, clipping of the negative partwill provide a unipolarsignalwithoutanylossofinformation.Relatedproofscanbefoundin[43,51].

Figure8.11 ExampleofanACO-OFDMtimedomainsignalwithN=16subcarriers(theCPisnotconsideredhere).Thepositiveandnegativepartsofthewaveformareanti-symmetricalalongthezero-axis.Thus,hard-clippingthenegativevaluesatthezerolevelonlywilldiscardredundantinformation.

InsertionofaCPandD/Aconversion isdone in thesamewayas in thecaseofDC-biasedDMT.Subsequently,theunipolarsignalisusedformodulatingtheLED.

Theideaofusingonlytheodd-numberedsubcarriersfordatatransmissionincludesthefact that all of the intermodulation products resulting from the asymmetric clipping areorthogonal to the data, i.e., clipping distortion falls on the unused even-numberedsubcarriersanddoesnotaffecttheodd-numberedones.Eventhoughtheeffectofclippingdoesnotresultininter-carrierinterferenceonthedata-carryingoddsubcarriers,itreducesall their amplitudes by half [51, 58, 62]. Moreover, the restriction to use only odd-numbered subcarriers can degrade performance in a frequency-selective channel.Particularlywhen bit and power loading is employed, where the allocation of bits andpowerper subcarrier isadapted to theSNRof the frequency-selectivechannel [54, 63],the subcarrier constraintwill result in a non-optimal adaptation to the channel responseand thus in an unfavorable system performance. This calls for carefully controlled andeffectivelycompensatedsolutions[64],inparticularifthenumberofsubcarriersissmall.

At theRx, thePDdetects the transmitted intensityand theanalogsignal isconvertedinto a digital one. It is well-known that OFDM systems are highly sensitive tosynchronizationerrors.DuetothefundamentallydifferentwaveformsinACO-OFDMandtraditionalOFDMsystems,conventionalsynchronizationcouldfailifapplieddirectly.Atechnique tailored to theACO-OFDMschemeusinganappropriate trainingsequence ispresented in [65].According to [66], the lossofbandwidth efficiencyassociatedwith atraining sequence can be avoided in systems employing ACO-OFDM if the CP-basedmethod is replaced by the presented technique of low-complexity blind timing

synchronization.

In conclusion, ACO-OFDM has a significantly lower detrimental DC componentcomparedtoDC-biasedDMT.Thus,themodulationschemeishighlypowerefficient,butattheexpenseofexploitingonlyhalfofthesubcarriersfordatatransmission.

8.4.3 Pulse-amplitude-modulateddiscretemultitone(PAM-DMT)The concept of pulse-amplitude-modulated DMT (PAM-DMT) introduced in [52] alsoaimsatasymmetricclippingatzerovalueand transmissionofonly thepositivepartsofthe DMT signal as in ACO-OFDM. This means that again anti-symmetry of the timedomainsignalisneededafterIFFT.

Firstly, as in DC-biased DMT and ACO-OFDM, Hermitian symmetry is induced inorder to achieve real-valued signals in the time domain. However, only the imaginarycomponentsofthePAMinputsignalareusedfurther,whiletherealcomponentsaresettozero.TheIFFTthenpresentsreal-valuedtimedomainsamplesthatexhibitanti-symmetry,asdesired.SimilarlytoDC-biasedDMT,uptoN/2subcarriers(includingCP)areused,i.e.theelements(X1,X2,⋯,X(N/2)−1)of[X]carrytheimaginaryPAMsignalcomponents,while X0 = 0. After CP insertion and along with D/A conversion, the entire negativeexcursion of the electricalwaveform is clippedwithout any loss of information, and isthenusedfordrivingtheLED.

In [52] it is shown that the distortion resulting from asymmetric clipping fallsorthogonally onto the real-valued parts of the PAM signal, without influencing theimaginary parts modulated with information, and thus without affecting the systemperformance. Proofs of anti-symmetry and the property of the clipping distortion termscan be found in [43]. At the Rx, demodulation and decoding the data are performedstraightforwardlyandsimilarlytotheschemesdiscussedbefore.

Inconclusion,PAM-DMTusesall subcarriersas in theDC-biasedDMTscheme,butthe modulation is restricted to just one dimension. Hence, PAM-DMT has the samespectral efficiency as ACO-OFDM. Regarding the power efficiency, the PAM-DMTcharacteristicsaresimilartothoseoftheACO-OFDMscheme.

8.4.4 DMT/OFDMperformanceandmitigationofdisruptiveeffectsNumerousanalyticalstudiesoncharacteristicsofthemodulationschemesdescribedaboveandondifferencesintheirperformanceinVLChavebeenpublishedinrecentyears.Themostimportantitemsarediscussedinthissection.

IfreductionofthehighPAPRisrequired,simplenon-lineartechniquescanbeapplied,possibly in combination with some form of predistortion [55, 67]. Various precodingtechniquesforPAPRreductioninasymmetricalclippedsystemssuchasACO-OFDMandPAM-DMTareanalyzedin[68].ItwasshowntherethatprecodingwithsmalleffortcouldgainPAPRreductionofabout3dB,makingthistechniqueattractiveforPAPRreduction

insystemswithasymmetricalclipping.

Severalcomparativeanalyseshavebeenperformedincludingmathematicaldescriptionofeffectsandsimulationsundervariousconditions,cf.inparticular[69–73].Toshedlighton one of the results, it was found that ACO-OFDM and PAM-DMT have virtuallyidenticalperformanceatdifferentbitratesandspectralefficiencies,asdemonstratedinthestudy[70], and is also indicated in several other publications.This is because inACO-OFDM,halfofthesubcarriersarefilled,butinPAM-DMThalfofthequadratureisfilled.Therefore,thesameperformanceisobtainedwhenaflatfrequencyresponseisconsidered.

Inpractice,clippingtheLEDmodulationsignalclearlyaffectsthesystemperformance.Themodulationorderandotherparametersofrelevancesuchasthebiaslevel(includingtheLEDturn-onthreshold)arethereforeofhighsignificance.WhileAWGNisthemaintypeofnoiseinthecaseoflowSNR,clippingdistortiondominatesatlargeSNRvalues.Thus,LEDclippingeffectshavetobeconsideredandthemodulationorderaswellasthepowerofthetransmitsignalshouldbeoptimized.Methodsforoptimizationaregivene.g.in [45].Anotheranalytical studyon the trade-offbetweenbiasingandclippingsuggeststhatratherthaneliminatingallclippings,theSNRisinfactoptimalwithsomedeliberatelyintroduced non-linear clipping distortion because of higher power efficiency at a lowerbias level. A corresponding biasing strategy is proposed in [74]. Thismethod does notrequireanyon-linecalculationandcanbeusedindifferentopticalchannelswithdifferentRxfiguresandmodulationschemes;however,anuplinkbetweenRxandTxisnecessary.

Besidesasymmetricclippingtoachievenon-negativity,thesignalforLEDmodulationoftenhastobedouble-sidedclippedinordertofit intothelinearrangeoftheLED[45,73].This introduces non-linear distortions. In amore advanced approach scaling of thetransmitsignalisproposedinordertoaccommodatethelargePAPRintheLED’slinearrange and to adapt the DC bias accordingly. More precisely, such signal shaping aspresented in [71]conditions the transmit signal tomeet theopticalpowerconstraintsoftheTxbymeansofscaling(realizedbydigitalsignalprocessing,DSP)andbyDCbiasing(realized in the analog Tx circuitry). An optimal signal shaping enables the GaussiantransmissionsignalstominimizetheelectricalSNRrequirement.Asinthecaseofdouble-sided clipping, shaping of the Gaussian time domain signals in ACO-OFDM and DC-biasedDMTresultsinanon-linearsignaldistortion,whichispreciselyanalyzedin[75].This analysis can be used to translate the signal scaling and DC biasing for a givendynamicrangeoftheLEDintoelectricalSNR,andthereforetoBERperformance.

As a summary of the numerous analyses and comparisons, bothmodulation formatsusingasymmetrical clipping, i.e.,ACO-OFDMandPAM-DMT,arebest at lowspectraland high power efficiency [69, 70, 72]. It is also important to note that multipathdispersioncouldbreakthesymmetryatthezerolineforbothoftheseschemes[52].Sincethe symbolperiodusuallyby far exceeds thevalueofdispersion, this effect shouldnotrepresent a serious issue; however, it has to be confirmed by practical experience. Ifmodulation at higher spectral efficiency is needed,DC-biasedDMT performs closer toACO-OFDM. This is because the clipping noise penalty forDC-biasedDMT becomesless significant than the penalty for the larger constellations required if otherwise theACO-OFDM or the PAM-DMT scheme is used. Thus, DC-biasedDMT is expected todeliver the highest throughput in applications where the additional DC bias power

required to create a non-negative signal does notmatter or can serve a complementaryfunctionality, such as illumination. It should be mentioned that the underlyingcomparisons typically assume a perfectly compensated LED non-linearity, e.g. using asuitablepredistortion.Hence,practicalverificationisalsonecessaryinthatrespect.

Beyondperformanceaspects,thecomputationalcomplexityofthemodulationschemesdiscussedisofimportance.Acomparativestudyonthecomputationalcomplexitycanbefound in [70]. The same number of actually used subcarriers in DC-biased DMT andACO-OFDMwaschosenthereinordertomakeafaircomparison.ForPAM-DMT,sinceonlyonedimensionisusedtotransmitdata,thenumberofsubcarriersusedwastwicethatfor theotherschemes. Inotherwords, fromtheDFTpointofview, theDFTsizeof theDC-biasedDMTschemeishalfthatofthefurtherschemes.

Table8.3summarizesthecomputationaleffortneededinthethreecasesatdifferentbitrates and electrical modulation bandwidths. As can be seen, the complexity of ACO-OFDM and PAM-DMT is nearly the same, while DC-biased DMT has the lowestcomputationalcomplexityamongthethreeformats,duetothesmallerDFTsize.

Table8.3 Computationalcomplexityneeded for the threemodulationschemesatlow(upper threerows)andhigh(lower threerows)electricalbandwidth,expressed intermsofrealoperationsperbit.Forparametersindetail,cf.[70].

DC-biasedDMT

ACO-OFDM PAM-DMT

Bitrate(Mbit/s)

Electricalbandwidth(MHz)

Tx Rx Tx Rx Tx Rx

50 25 43.0 45.9 48.5 50.0 48.5 51.5

100 50 43.0 45.9 48.5 50.0 48.5 51.5

300 150 42.4 45.3 48.1 49.6 48.1 51.1

50 50 86.1 91.9 97.0 100.0 97.0 102.9

100 100 85.4 91.2 96.7 99.6 96.7 102.6

300 300 95.3 101.1 106.5 109.4 106.5 112.4

Theoreticalworksuchasmathematicalanalysesandsimulationsincludingraytracinghastobeverifiedandcomplementedbyexperimentsunderconditionsclosetoreality.Atthis stageofdevelopment in thecaseofmodulationschemessuchasDMT, there is thedifficultyof time-consuminghardware implementation includingDSPalgorithms.Asanalternative,thewidelyrecognizedmethodofoff-lineprocessing(seealsoSection8.6.2)isusedinnearlyallcasespublishedsofar.However,thiscomesalongwithrestrictions,such

as focusingon rudimentaryor special functions inVLC transmission.For example, themitigationofbackgroundopticalnoiseproducedbyambientAC-poweredLEDlampsorfluorescent light sources was investigated in an experimental VLC setup using off-lineprocessedDC-biasedDMT, as reported in [76]. The results have shown that the noiseproducedbyAC-poweredLEDshasanegligibleeffect,asthelowest-frequencysubcarrieris far above the 50 or 60 Hz effects, while the noise from fluorescent lamps can beremoved by excluding the impaired subcarriers. Moreover, the mitigation of phosphorlayer effects caused by the YB-LED-based Tx using channel estimation and one-tapequalizationattheRxwasconfirmed.Furtherexamplesofsuchexperimentalworkbasedonoff-lineprocessingandmainlytargetinghighbitratesinalaboratoryenvironmentcanbe found in [8–10,77–79].However,more complex subjects including the influenceofchannelvariationsandadaptivebitratelinkcontrolarehardtoinvestigateunderreal-timeconditions using off-line experiments. If restrictions in DSP speed and bandwidth areaccepted,asimplealternativemaybetoemployprogrammablehardwaresuchasaDSPdevelopmentkit.Anexampleof that kind ispresented in [40],where itwas theaim toinvestigatetheperformanceofDC-biasedDMTtransmissioninthepresenceof(infrared)LEDnon-linearityviaahardwaredemonstratorincludingreal-timeimplementationofthedigitalsignalprocessing.

8.5 PerformanceenhancementofDMTmodulationIn recent years, numerous approaches have been reported aiming at performanceenhancement of the discussedmodulation schemes, in terms of spectral efficiency andpowerefficiencyaswellasPAPR.Inthefollowing,abriefoverviewisgiven.

8.5.1 CombinationofACO-OFDMandDC-biasedDMTmodulationItseemsratherobvioustocombinetheACO-OFDMmodulationscheme,whichonlyusestheoddsubcarriers,withacomplementaryschemeinordertoutilizetheevensubcarrierstoo.Suchan approach, calledasymmetrically clippedDC-biased opticalOFDM (ADO-OFDM) ispresented in [80]. In this technique, theodd subcarriers aremodulatedusingACO-OFDM while the even ones are modulated using DC-biased DMT. The DMTcomponent does not cause interference on the odd frequency subcarriers. Hence, aconventionalreceivercandemodulatetheACO-OFDMcomponentafterdetection.Ontheother hand, the ACO-OFDM signal causes clipping noise, which affects the evensubcarriers.ThisinterferencecanbeestimatedattheRx,andthuscancelledattheexpenseof a 3 dB noise penalty in the DC-biased DMT component. Clearly, ADO-OFDMprovidesbetterbandwidthefficiency thanACO-OFDM,sinceallsubcarriersareused tocarrydata.SimulationresultsforanAWGNchannelhaveshownthatthismethodcanalsoachievebetteropticalpowerperformancethantheconventionalOFDMschemes[72].

8.5.2 SpectrallyfactorizedOFDMAnothermeansforimprovingtheopticalefficiencyofOFDMforIM/DDtransmissionhasbeen proposed in [81]. The formalism for non-negativemultiple subcarrier modulation

describedthereisdenotedasspectrally factorizedopticalOFDM(SFO-OFDM). Insteadof sending data directly on the subcarriers, the autocorrelation of the complex datasequence is performed in this scheme,whichguarantees non-negativitywithout explicitbias.SFO-OFDMcoversallband-limitedOFDMsignals,andunlikeACO-OFDM,itusestheentireavailablebandwidthfordatamodulation.Moreover,itmitigatesthehighPAPR,which is typical inPAM-DMTandACO-OFDMsystems.According to [81],0.5dB inopticalpowerisgainedoverACO-OFDMataBERof10−5.

8.5.3 Flip-OFDMAfurtherunipolarmodulationapproachisknownasflip-OFDM[82].Inflip-OFDM, thepositiveandnegativepartsareextractedfromtherealbipolarOFDMtimedomainsignal,and transmitted in two consecutive OFDM symbols. Since the negative part is flippedbeforetransmission,bothsubframeshavepositivesamples,asrequiredinIM/DDsystems.Thebasic flip-OFDMscheme,aspresented in [83],performsacompressionof the timesamples in order to sustain the duration of the original bipolar symbol frame.Consequently,bandwidthanddataratearedoubled,whilethelengthoftheCPisreducedby50%whencompared toACO-OFDM.Asanalternative, theparametersofanACO-OFDMsystemcanbemaintainedbyomittingthe(notimperativelyrequired)compressionoftheOFDMsubframes.Inthatway,ithasbeenshownbysimulationthatboththeACO-andflip-OFDMschemeshave thesamespectralefficiencyandBERperformance in theelectricaldomain.However,flip-OFDMofferssavingsof50%incomputationalreceivercomplexityovertheACO-OFDMscheme,inparticularforslowfadingchannels[82].

8.5.4 UnipolarOFDMTheso-calledunipolarOFDMmodulationscheme(U-OFDM)hasbeendevelopedwiththeaimofreducingthePAPRandtoclosethe3dBgapbetweenOFDMandACO-OFDMforbipolarsignals,whilstgeneratingaunipolarsignal,whichdoesnotrequirethebiasingofDC-biasedDMT[84].Themodulation process ofU-OFDMstartswith conventionalmodulationofanOFDMsignal.Afterobtainingarealbipolarsignal,itistransformedintoaunipolaronebyencodingtheabsolutevalueofeachtimesampleanditspolarityintoapairofnewtimesamples(absolutevalueononeoutoftwopossiblepositionsdependingon the polarity). At the Rx, reconstruction of the original bipolar OFDM signal isstraightforwardand theprocesscontinueswithconventionaldemodulationofanOFDMsignal.

SinceeachtimesamplefromtheoriginalOFDMsignalisencodedintoasamplepairoftheU-OFDMsignal,thespectralefficiencyisthesameasofACO-OFDM,butU-OFDMhasbetterpowerefficiencyinanAWGNchannel.

8.5.5 PositionmodulatingOFDMBasedonhow thesymbolsareassigned to thesubcarriers, theunipolaroperation formsunipolaropticalOFDMsymbolssuchasthevariousoperationsusedtogeneratetheACO-OFDM, the flip-OFDM, theU-OFDM(seeabove),or theso-calledpositionmodulationOFDM (PM-OFDM) described in [85]. PM-OFDM utilizes the DFT approach but

removestheHermitiansymmetryconstraint.ThisisdoneattheTxbysplittingtheIFFToutput signal corresponding to its real and imaginary components. Their positive andnegativepartsare furtherseparated,and the twonegativesignalsare flippedbypolarityinversion.The resulting four real and positive signals are then sequentially transmitted.BasedonthisTxtechnique,twoRxstructuresarepresentedin[85]targetingeitherhighBERperformanceorlowRx-complexity.

In both, LOS and diffused channels the low-complexity Rx (including time domainMMSEequalizer)achievesthesameBERperformanceasanACO-orflip-OFDMsystem,while the total system complexity is significantly lower. On the other hand, the high-performanceRxincludesafrequencydomainMMSEequalizerandthusanextraFFTandIFFTblock, resulting inamarginallyhigheroverall transceivercomplexitycompared toanACO-OFDMsystem.Asoutlinedin[85],thisextraeffortyields~4dBimprovementataBERof10−4inadiffuseopticalwirelesschannel.

8.5.6 Diversity-combinedOFDMWhenonly theodd subchannels are loadedas inACO-OFDMmodulation, the clippingdistortion fallsexclusivelyonto theevensubchannels, asmentionedabove.Thisnaturalseparationof signal anddistortion indicates that at theRx side, someextentof spectraldiversityamongevenandoddsubchannelscanbeobserved.Basedonthisfact,theideaofanasymmetricallyclipped,diversity-combinedOFDM(AC/DC-OFDM)systemhasbeenpresented in [86].There it is revealed theoretically andby simulation, that the effectiveSNRattheRxofanACO-OFDMsystemcanbeimprovedsignificantlyattheexpenseofoneadditionalIFFT-FFTpairandadiversitycombiningdecodingalgorithmattheRx.Asdiversity combining adds two different signal components, while an additional noisecancellationwouldreducethenoiseattheRx,itmightbeexpectedthatcombiningthesetechniquescouldgivefurtherperformancegains.Thisis,however,nottrueaswasshownanalytically in [87]. On the other hand, it is important to note that noise cancellationgivingamaximumimprovementof3dBismorecomputationallyefficientthandiversitycombining.

8.5.7 FurtherapproachesIn order to enhance the overall performance ofVLC systems and to efficiently exploitspectrumresourcesbymeansofMSM,variousadditionalproposalshavebeenpublished.Twoofthemarebrieflyaddressedinthissection.

Multicarriercodedivisionmultipleaccess(MC-CDMA)isatransmissionschemethatcombinestherobustnessoforthogonalmodulationandtheflexibilityofCDMAschemes.InMC-CDMA,an individualuser’s complex-valueddata symbol is spreadoverOFDMsubcarriersinthefrequencydomainusingaspreadingcode.Thesymbolsfromdifferentusers are aggregated in the frequency domain and passed to the OFDMmodulator fortransformation into the time domain. The further steps of modulation and IM/DDtransmissionarethesameasinthecaseofDC-biasedDMT.AttheRxside,CPremoval,FFT, and de-spreading are carried out, respectively. In [88], an optimized selection ofactivesubcarriersisproposedinordertosignificantlyincreasethepowerefficiencyina

multi-user indoor scenario. The average transmit power reduction is accomplished byselectingthesubcarriersbasedonpre-orpost-equalization,whilepropersettingofafixedDCbiasensureslowsystemcomplexity.A hybridmulti-layermodulation (MLM) scheme designed for optical OFDM-based

IM/DD transmission is introduced in [89]. In optical systems,MLM is expected to becapableofofferingafinegranularity in termsof throughputversusrobustness.Both theconceptofMLMandrelateddoubleturboRxalgorithmsaredetailedinthispublication.In addition, the layer-specific optimumweights that theMLM scheme should obey areconceived.SignificantgainsaredemonstratedbycomparingtheMLM-aidedtechniquetotheACO-OFDMandDC-biasedDMTschemesasafunctionofboththeelectricalandtheopticalenergyperbittosingle-sidednoisepowerspectraldensity.However,thebenefitsareachievedattheexpenseofanincreasedtransceivercomplexity.Asstatedin[89],theapproachisforfurtherstudyunderpracticalopticalchannelconditions.

8.6 Systemdesignandimplementationaspects

8.6.1 AspectsofsystemdesignSofar, researchanddevelopmentonVLCusingadvancedLEDmodulationhasfocusedonphysical transmissionbasics,whereusuallyadirectedLOSisassumed.Ontheotherhand,non-directedLOSlinksaremoreconvenientinpracticaluse,particularlyifterminalmobility is desired. Non-directed indoor wireless IR channels have been extensivelystudied in the past by means of modeling, simulations, and experiments. Consideringexisting knowledge in this regard and similarities to VLC, such studies can provideadequate guidance for designing indoor VLC systems. However, there is only a littleexperience in practical terms of high-speed VLC via non-directed LOS and non-LOSlinks,i.e.,linksindiffusescenarios.

Off-lineexperimentsusingaYB-LEDandDC-biasedDMTmodulation ina realisticnon-LOSbroadcastconfigurationarepresentede.g.in[90].Itisshowntherethatanareaof ~18m2 can be covered with 100–200 Mbit/s, while the brightness level at the Rxequippedwithabluefilterisabout500lux(generatedbyanextraLEDforlighting),andthus complies with workplace requirements. Some further results achieved for the firsttimewitharealbidirectionalsysteminanon-LOSconfigurationarepresentedbelowinSection8.6.4.

Bidirectional links are a necessary condition for any kind of communication beyondbroadcasting. Various approaches for an optical wireless indoor uplink have beendiscussedand thedevelopmentofaqualityuplink iscurrentlyoneof themain researchtopicsinthisfield.Themostcommonapproach,irrespectiveofthemodulationformat,istouseIRlightasillustratedinFig.8.4.Examplesofmediumandhigh-speedsystemsaregiven e.g. in [20] and [91], respectively. However, visible light uplinks are alsoconsidered,asforinstanceshowninFig.8.5,giventhatavisiblespotintheRxregionfitstheapplicationneeds.

Atthesametime,bidirectionallinksenabledynamicadaptationtochangingproperties

ofthelink,e.g.causedbyambientlightorLOSblocking.Thetime-varyingchannelcaneasilybeestimatedusingfrequencydomainchannelestimation,andadaptivemodulationcanbeappliedbasedontherequesteddataratesandqualityofservice.TheideaofOWCusingDMTmodulationwith adaptive bit and power loadingwas proposed for the firsttime in [17] and independently thereof almost simultaneously in [18]. By this method,subcarriers are loaded with the best suitable modulation depending on the channelproperties at individual frequencies, i.e., the power is distributed according to the SNRneeds of each chosenmodulation order or formatwhile keeping the system constraintsregarding BER and transmit power satisfied. This leads to optimal utilization of theavailable resources in a non-flat communication channel even in the presence of non-linear distortion caused by the transfer characteristics of common LEDs. The wholeconcept is analytically presented in [33, 34, 92]. Related algorithms for bit and powerallocationarealsoaddressedin[93],whileexperimentalproofsofconceptcanbefoundin[94–96].Furtherexperimentalstudieshavebeenperformedaccordingto[97]inordertodemonstrate that the benefits of adaptively controlling the resultant QAM modulationordersalsoapplytoYB-LED-basedsystemswherebluefilteringattheRxisomitted.

Inadditiontochanneladaptation,thepossibilitytocombineDMTmodulationwithanymultiple access schemesmakes it an excellent preference for indoorVLC applications.Initialwork directed towardmulti-channel systems is published in [77] (see also [10]),where subcarrier multiplexing (SCM) is used to provide multi-user capability. Thedownlinkcapacityofanexperimentalsetupbasedonoff-lineprocessingwasassembledto575Mbit/sbywavelengthdivisionmultiplexing(WDM)usingthecoloredchannelsofaRGB-LED,whiletheYB-LEDuplinkofthefullduplexsystemoffered225Mbit/s.Pre-and post-equalization in the frequency domainwere adopted to compensate for Tx andchannel distortions. In addition, the transmission capacity was optimized using variousQAMmodulationorders.Byadjustingthebandwidthsandmodulationformatsofthesub-channels,thedownlinkanduplinkcapacitiescanbeeasilyreconfiguredinsuchasetup.

RegardingthefeederorbackhaulnetworkofVLC,acombinationwiththepowerlineinfrastructureandthuswithPLCsuggestsitself,cf.forexample[1,2,19,98,99].FromtheviewpointofDMTmodulationusedinbothsystems,thisideaisofparticularinterestasitcouldsimplifyinterfacesandinterworking.Komineetal.proposedandanalyzedanintegrated system of this kind in [100] based on their earlier work and subsequently,studies addressing that subject have been performed at several places. As an example,broadcastingtheDMT-QAMsignalofaPLCsystembydirectlymodulatingtheYB-LEDofaVLCTxwaspresentedin[101].MorerecentproposalsforanintegrationofPLCandVLCcanbefounde.g.in[102].Forreasonsofcompleteness,ithastobementionedthatopticalfibersandDMTtransmissioncanalsoserveforVLCbackhauling.However,thiscreates a special area of research employing e.g. laser light, and thus will not beconsideredfurtherhere.

The interplay of VLC, backbone network and terminal systems also calls for anappropriatemediaaccess control (MAC)protocol.A specificMAC layerwasproposedfortheopticalwirelesschannel(consideringbothvisiblelightandIRapplications)intheframework of theOMEGAproject [103], see alsoSection 8.6.4. ForVLC broadcast, asimplifiedframeformatwasdescribed.AttheTxsideoftheconsideredsystem,theMAClayer provided a serial data stream at 100 Mbit/s for digital signal processing and

modulationonthePHYlayer.AftertransmissionviaVLClinkandPHYprocessingattheRx side, the retrieved data stream was passed to theMAC layer. An overview of theaccess method, the optical wireless data link layer (OWMAC) and the MAC frameadapted to that VLC prototype is given in [104]. All themodules forMAC and PHYprocessingatboththeTxandRxsidewereimplementedinFPGAtechnology[104,105].Apart fromthatwork,multipleaccessschemesformulticarrier-basedVLCchannelsarealsoaddressedine.g.[106,107].

A major design challenge regarding the commercialization of VLC for joint use inlightingsystemsishowtoincorporatethecommonlyusedPWMlightdimmingtechniquewhile maintaining reliable high-speed communication links. In [108] this problemwasconsidered in terms of VLC power constraints due to lighting requirements in livingrooms for both daytime and night-time scenarios, i.e. even in the worst case ofcommunicationwith the lights off. Itwas shown that very low light emission (virtuallylights-off) is sufficient tomaintain coverage at data rates up to theMbit/s range usingOOKorPPM.ThisleadstotheconclusionthateveniftheaverageopticaltransmitpowerisseverelydimmedandlesspowerefficientmodulationschemessuchasDMTareused,thereshouldbesomeelectricalsignalpowerforcommunicationtoacertainextent.

RegardingVLCsystemsbasedonDC-biasedDMTorACO-OFDM,dimmingcannotbe achieved directly, but the data signal finished for driving theLEDcan be combinedwithadimmingtechniqueonthePHYlayer.However,thesimplewayofmultiplyingthetransmissionsignalandaPWMpulsetrainfordimmingcontrolduringits“on”periodisnotanoption,asthedatathroughputwouldbelimitedtothePWMrate,whichisaslowasa few 10 kHz in commercial LED lighting systems [12]. Achieving high-speedtransmissionwiththisapproachisonlyfeasiblewhenthePWMdimmingsignalisatleasttwicethehighestsubcarrierfrequencyoftheDMTsignaltoavoidsubcarrierinterference[21]. Such a constraint is hardly acceptable, as the LEDmodulation bandwidth shouldinclude the PWM frequency, which cuts the bandwidth for data transmission by 50%.Therefore,dimmingschemesarebeingdevelopedsuchasreversepolarityOFDM(RPO-OFDM),proposedin[109].Thismethodcombinesthehigh-frequencyOFDMsignalwiththe low-frequency PWMdimming signal,while both signals contribute to the effectiveLED brightness. RPO-OFDM utilizes the entire period of a PWM signal for datatransmissionbyadjusting thepolarityof thedatasymbolsbefore theyaresuperimposedon the PWM pulse train, which defines the dimming level, i.e. the symbol polarity isreversed during the “on” periods of the PWM signal. In this way, the data rate ismaintainedforawidedimmingrangeindependentlyofthePWMfrequency.TheapproachalsokeepsthesignalwithinthedynamicrangeconstraintoftheLED.Inconclusion,thedata rate in RPO-OFDM is not limited by the PWM signal frequency, and the LEDdynamic range is fully utilized tominimize the non-linear distortion of themulticarriercommunication signal. This technique can be applied to any version of real-valuedOFDM/DMTsignal,preprocessedfortransmissioninVLC.

OFDM/DMT modulation continually gains in popularity due to its attractivecommunicationperformance,andtheresultsachievedsofarindicatethatitisanexcellentcandidate for bidirectional transmission in VLC systems close to reality. Nevertheless,additional research isnecessary, inparticular toexamineallaspectsofcooperationwithlightingsystemsanditseffectsonlightqualityinpracticalsmartlightingscenarios.

8.6.2 DMT/OFDMapplicationinadvancedsystemsWhen using RGB-type LEDs (or more generally, multi-color LED devices), a VLCsystemenableswavelengthdivisionmultiplexing (WDM).To determine the potential ofWDMin termsof transmissioncapacity,over recentyears several laboratory studiesontheWDMtransmissionperformanceinVLCsystemshavebeencarriedoutandreported.AsademonstrationofbitrateperWDMchannelwasthemaindriver,spectrallyefficientDC-biasedDMTmodulationwasemployedalmostwithoutexception.

Key results of laboratory experiments using WDM in VLC systems based on DC-biasedDMTaregiveninTable8.4,whichindicatesthatbitratesofseveral100Mbit/sperWDMchannelandtotalcapacitiesintheGbit/srangearepossible.

Table8.4 KeyresultsoflaboratoryexperimentsusingWDMinVLCsystemsbasedonDC-biasedDMT.

Aggregatebitrate(Gbit/s)

ColorsofWDMchannels

Modulationbandwidth(MHz)

Numberofsubcarriers Remarks Reference

0.575 RG 50 64 PIN-Rx,2SCMchannelsperWDMchannel,blueLEDonlyusedforlighting;inadditionuplinkviaYB-LED

[10]

0.750 RGB 50 64 PIN-Rx,adaptiveNyquistwindowing

[110]

0.803 RGB 50 32 APD-Rx [111]

1.250 RGB 100 128 APD-Rx [8]

2.930 RGB 230 471 PIN-Rx [79]

3.400 RGB 280 512 APD-Rx [9]

ItshouldbenotedthatalloftheseresultshavebeenachievedbyexploringtheWDMchannels one by one and bymeans of off-line processing. In such a configuration, thetransmit signal is generated by software and an arbitrary waveform generator (AWG),while the received signal is recordedand subsequentlyevaluatedby software.A typicalsetupofsuchanexperimentissketchedinFig.8.12.

Figure8.12 Setupforoff-lineevaluationofsimultaneousDC-biasedDMTtransmissionusingthreeVLCchannelsinWDMfashion.

AsVLCsystemstypicallyusemultipleLEDswhencombinedwithindoorlighting,itisan obvious step to apply the optical wireless multiple-input multiple-output (MIMO)principle in order to boost the overall transmission capacity. MIMO processing cancompensateinter-channelcrosstalk,thusallowingforparalleltransmissionfromanumberofLEDs[112].InacorrespondingopticalMIMOproofofconceptusingAWG-generatedDC-biasedDMTmodulation,a2×2TxmoduleconsistingoffourYB-LEDs,anda3×3-channel imagingRx equippedwith a blue filter, an aggregate bit rate of 1.1Gbit/s hasbeendemonstrated,asreportedin[113].

Optical spatialmodulation (OSM) is another bandwidth and power efficientMIMOscheme forOWC. It is based onmultiple spatially separatedTxunits and utilizes theirlocation to carry extra information [114], i.e., in addition to applying basic signalmodulation,OSMtransmitsfurtherbitsinthespatialdomainbyconsideringtheTxarrayasanextendedconstellationdiagram.AsonlyoneTxisactiveatanygiventimeinstant,theRxunitcaneasilydeterminetheindexofanactiveTxandthuscanbekeptsimple.Inthe case of alignedTx andRx units, itwas shown that the paths of the opticalMIMOchannel are nearly uncorrelated. Thus, an accurate Tx-Rx alignment results in a powerefficiency increase with respect to the modulation scheme used. A comprehensiveoverviewof thestateof theart inspatialmodulation forMIMOtechnologies, includingapplication inVLC, is provided e.g. by [115].A novel unipolarmodulationmethod forOWC based on OSM is introduced in [116]. It combines the basic spatial modulationscheme and OFDM techniques for OWC [69]. Results show that the new approachimproves the power efficiency. For the same spectral efficiency, it exhibits 5 to 9 dBhigher energy efficiency than ACO-OFDM, while in contrast to DC-biased DMT iteliminates the need for a DC bias. Consequently, it exhibits a considerable powerefficiencygainforthesamespectralefficiency.

8.6.3 PracticalimplementationissuesWith the aim of reducing the DSP complexity at both the Tx and the Rx, IFFT/FFTprocessing can be replaced by a discreteHartley transform (DHT) [117,85].DHT is areal-valuedself-inversetransformwithoutHermitiansymmetryconstraintattheinput.Asaresult,thereisnoneedforcomplexalgebraandthesamefastDSPalgorithmcanbeusedfor modulation and demodulation as well, if real constellations are used for subcarriermodulation.Real-timeimplementationsofDHT-basedTxandRxthushaveconfirmedareductionofbothcomplexityandcomputingtime,whiletheperformanceisthesameasinDFT-basedDC-biasedDMTandACO-OFDMsystems,respectively[117].

AnimportantgoalofVLCingeneralistodrivetheLEDatthebestpossibleelectro-optical power conversion efficiency. Additionally in dual use with lighting, theillumination function must be maintained with minimum extra power consumption. Inmost cases, VLC designs try to employ (digital) off-the-shelf high-power LED drivers[109],andalsotousethedimmingcapabilityusuallyofferedbysuchdevices.AnoutlineoftheimportantroleofLEDdriversinsmartlightingsystemscanbefoundin[2],whiletheLEDdriver’senergyefficiencyalongwith theOOK,VPPMandOFDMmodulationschemesisdiscussedin[118].Usually, the largecurrentofhigh-powerLEDscontrolledby thedrivercircuit severelydegrades its response. In [38] it is shown thatOOK-basedsystems nonetheless can achieve rather sound energy efficiency while enablingtransmissionratesof477Mbit/s.Inthiscase,thereddeviceofaRGB-LEDwasdrivenbyaspeciallydesignedcircuitwithpre-emphasis,whichimprovedthe3dBbandwidthoftheopticalTxto91MHz.AnotherapproachtoasimpleandpracticalLEDdriverprovidingahighoverallTx-bandwidthisbasedondrawingouttheremainingcarriersthatexistintheLEDdepletioncapacitanceduringthe“off”stateoftheOOKinputsignal[119].

TheseexamplesforOOKmodulationindicatethatitappearsunrealistictousestandarddriver circuits in OFDM/DMT-based systems, which exploit the bandwidth moreefficiently but need more complex analog circuitry, usually along with higher powerconsumption. Because the optical Tx performance depends, inter alia, strongly on thedriveroutputimpedance,carefulimpedancematchingtotheLEDdeviceormoduleisofutmostimportance.Asatypicalexample,Fig.8.13illustratesthebandwidthofanopticalTxincludinghigh-powerLEDandcustomizedanalogcircuitry,whichcandrivecurrentsupto1.2A.SuchadriverwasemployedinvariousVLCsetupsusingeitheraRGB-LED[8]oraYB-LEDincludingbluefilter[77,120].Comparedtopurelighting,anincreasedpower consumption of about 30% is observedwhen changing toVLC operation at thesamebrightnesslevelattheRx.Inviewofsuchpowervalues,RFleakagecanbestrongerthanthereceivedopticalsignal.ThatiswhyaccurateRFshieldingoftheVLCTxinanycaseisanimportantsubject.

Figure8.13 VLCchannelresponsetakenwithanetworkanalyzeratashortdistancefromahigh-powerYB-LED,whichwasdrivenbyacustom-tailoredanalogdrivercircuitatabiascurrentof0.7A.Theeffectivemodulationbandwidthwasabout180MHz.ForcomparisonsimulatedfirstorderRClow-passLEDamplituderesponsesarealsoshown[120].

Beyond the modulation scheme used in a VLC system, convenient ways forperformance improvementareequalizationat theTx (pre-equalization), at theRx (post-equalization),oracombinationofsuch techniques.Asanexample,pre-equalizationcanbeapartoftheLEDdrivingmodule[29].

Post-equalizationisbrieflyaddressedbelow.AnoisecancellationmethodforanACO-OFDM Rx is presented in [121], where the anti-symmetry of the time domain signalsamplesinherentinACO-OFDMisusedtoidentifywhichsamplesofthereceivedsignalaremost likelytobedueto theadditionofnoise.Thesesamplesare thenset tozero.Amaximum gain of 3 dB in optical power can be achievedwith thismethod. The samepairwisemaximumlikelihood(ML)RxstructurecanbeemployedinPAM-DMTsystemsbyconsideringtheiranti-symmetryproperties.ItisworthmentioningthattheuseofthisRx technique in a system based on flip-OFDM results in the U-OFDM scheme aspresentedin[84],cf.Sections8.5.3and8.5.4.Inanothercase,post-processingisproposedtoeliminate theeffectofnoiseoutside themaximumchanneldelaybymeansofa leastsquare(LS)channelestimationmethod[122].Theschemeisbasedonacomb-typepilotsubcarrierarrangement,whereeachOFDMsymbolhaspilottonesatperiodically-locatedsubcarriers.AfterordinaryLSchannelestimation,thederivedchannelistruncatedinthetime domain according to the maximum channel delay via DFT and inverse DFT,respectively.Inthisway,themethodcansignificantlyimprovetheBERperformance,ashasbeenprovedbysimulationatdifferentQAMconstellationordersinaDC-biasedDMTsystem.

Apart from the limitedmodulationbandwidth,whiteLEDssuffer from intrinsicnon-linearity.ThiseffectisparticularlydetrimentalwhenhighPAPRmodulationformatssuch

as DMT are employed. Much work has been done to overcome this impairment, cf.Section 8.3.3. In [123], the application of Volterra equalization is proposed forcompensatingISIandtheYB-LEDnon-linearityinaVLCsystemthatemploysM-PAMmodulation.ItwasdemonstratedthataRxusingadecisionfeedbackequalizer(DFE)withnon-linearVolterrafeed-forwardsectionuptothesecondordercanefficientlycompensateeffectsofnon-linearityofthetransmittingLEDandperformsupto5dBbetterintermsofoptical power than using a standard DFE. In view of such results, this scheme is alsoattractiveforDMT-basedVLCsystems.

ItishardlysurprisingthattheSNRdistributioninanareacoveredbyindoorVLCcanbe improved,aswellas thespectralefficiencyof theDMT-basedmodulation,by tiltingthe (moving) receiver plane.Such a schemehasbeenproposed in [124] to enhance theperformance in an adaptive system,where a feedback channel is used to adapt bit andpower loading. The optimum tilting angle of the photodetector is determined by theNewtonmethod,which is a fast algorithm requiring a three-step search from the initialstate.

Up to now, experimental comparisons of different system approaches and detailsthereof are extremely scarce. Quite recently, the bit rate performances of variousVLCapproaches, including DC-biased DMT, ACO-OFDM, and U-OFDM have beenexperimentallycomparedin[125].Clearly,thebandwidthefficiencyofDC-biasedDMTpresents a higher bit rate performance, but is of course less power efficient thanACO-OFDM.Ontheotherhand,theACO-OFDMaswellastheU-OFDMschemessufferfromtheeffectofbaselinewanderinpracticaltransmission,causedbythemovingaverageoftheasymmetricsignal.Suchfindingsrequirecloserexaminationinfuturework.

8.6.4 ImplementationanddemonstrationThefeasibilityofwhiteLEDintensitymodulationusingOFDMwasvalidatedforthefirsttime by experimental results as published in [126]. Since then, a lot of research andanalyticalwork, aswell as simulations andexperimental studiesonDMT/OFDM-basedVLC, including subsystems thereof as discussed above, have been carried out. Despitethis, there are only a very small number of system implementations which incorporatereal-timesignalprocessingforDMT/OFDMmodulation,etc.Thisishowevermandatoryforsystemverificationunderreal-lifeconditionsandanimportantstepforpavingthewaytorealcommercialapplication.

An early proof-of-concept hardware demonstrator is shown in Fig. 8.14, where thebandwidth of 45 kHz was limited by the DSP development kit used for performingOFDM-relatedsignalprocessing.Theaimofthissystemwastostudytheperformanceofphase-incoherentopticalOFDMundervariousconditions,e.g.againstdifferentelectricalSNRs.Moreover, the systemserved for characterizationof thewireless channel and formodeling.

Figure8.14 Left:LaboratoryVLCdemonstratorforinitialunidirectionalOFDM-basedreal-timeexperiments[127].Right:ReadinglampwithanarrayofninewhiteLEDsinordertoenlargethecoverageareaforOFDMexperiments[128].

In the course of the European research and development project OMEGA, a fully-fledged high-speed VLC system based on DMT modulation has been developed andimplemented, which offers wireless broadcast channels of 100 Mbit/s (net) in homes[129].Underreal-lifeconditionssuchaperformancewasdemonstratedforthefirsttimeinFebruary 2011, by broadcasting up to four HD video streams (80% utilization of the100 Mbit/s channel) simultaneously from a total of 16 YB-LED ceiling lamps to aphotodetectorplacedanywherewithinthelitareaofabout10m2.AnillustrationisgiveninFig. 8.15. TheMAC protocol (OpticalWirelessMAC) developed especially for thispurpose,anddigitalsignalprocessingfunctionalitiesforthePHYlayer,wereimplementedon FPGA boards. The PHY processing encompassed a typical DMTmodulator/demodulator including scrambler and forward error correction (FEC)encoder/decoder with parameters summarized in Table 8.5. The physical layer alsofeaturedfullsynchronizationonbitandframelevels[104,130,131].

Figure8.15 Thefirstpublicdemonstrationofahigh-speedVLCsystemusingDMTmodulationforbroadcastingmultiplevideostreamsatadatarateof100Mbit/s(125Mbit/sgrossdatarateatPHYlayer).(Top)ArrayofVLCtransmittersontheceiling;(bottom)mobilereceiverdeviceconsistingofaphotodetectorandanamplifierforforwardingthesignaltothedemodulatorandfurthertothevideoscreens.

Table8.5 DigitalPHYparametersofthefully-fledgedVLCdemonstrator.

Parameter Value

Signalbandwidth(MHz) 30.5947

Numberofsubcarriers(incl.DC) 32

Lengthofcyclicprefix(samples) 4

QAMorder 16

FEC(ReedSolomon) 187207

MorerecentlyatFraunhoferHHI,abidirectionalhigh-speedreal-timeVLCsystemhasbeenpresented(Fig.8.16),whichoperatesinatimedivisionduplexmode.Rate-adaptiveDC-biasedDMT is implemented by feedback via the reverse link. The transceivers areequippedwithproprietaryVLCTxandRxmodulesprovidingamodulationbandwidthofup to 180MHz.However, the bandwidth used is below100MHz, limited by theDSPchips.Thesetransceivermodulescanoperatewithoutactivecoolingandoffer1000BASE-TEthernetinterfaces,cf.Fig.8.16(bottom).

Figure8.16 (Top):Twobidirectionaltransceiverscommunicatingviaadiffuselink.Differentcolorsareusedfordownanduplinkasanexample,butalsothesamecolorscanbeused.(Bottomleft):Viewofthediffusingspotofthelinkata(standardpainted)wallatadistanceofupto~3mfromthetransceivers.Theoveralllinklengthisabout6m.(Bottomright):Newgenerationofbidirectionaltransceiverwithaformfactorof87×114×42mm3(withoutlenses).

Aparticularadvantageofthisreal-timeVLCsystemisthevariablethroughputofupto500Mbit/swithcontrolledBER,dependingonthequalityoftheopticalcommunicationchannel.ThesystemofferedthefirstmobileVLCexperience.AsshowninFig.8.17, themostimportantparameteristhelightintensityattheRx,leadingtoanalmostproportionaldata rate adaptation. It should be noted that the observed performance is similar withilluminationLEDsofanycolor.IR-LEDscanbeusedaswellaswhitelightphosphorousLEDs,whichbenefitfrombuilt-inpre-equalizationandthusdonotrequirebluefilteringat

theRx.The power consumption of this system is onlymoderately increased – by 30%comparedtotheoriginallightingfunction–duetoanoptimizedLEDdriverdesign.

Figure8.17 (a)Maximumachieveddatarateswiththereal-timerate-adaptiveVLCsystemdependingonthelightintensityattheRx.(b)Maximumdataratesachievedwithdifferentlinktypes:LOSlinkwithnarrowFOV(trianglecurve),LOSlinkwithwideFOV(diamondcurve),andanon-LOSdiffuselink(squarecurve).

Figure8.17revealsanotherimportantresult:high-speednon-LOSdatatransmission.Inthis particular case, lightwas reflected by thewhite-paintedwall to theRx.More than100Mbit/sarepossibleatalinkrangeofabout3m,despitediffusionthroughreflection.Inthisway,thefeasibilityofaflexibleandrobustVLCconnectionwasdemonstratedathighbitratesunderdifferentlinkscenarios.

Asimilarimplementationwaspublishedrecentlyin[91].ThisVLCsystemusedaYB-LEDdeviceforthedownlinkandanIRuplink,withanintegratedOFDM-basedDSPchipforreal-timeoperation.TheLEDbandwidthof~1MHzhasbeenincreasedto~12MHzby an analog pre-equalization technique without using a blue filter. The modulationbandwidth of theDSP chip applied is between2 and30MHzand the adaptiveOFDMmodulationusesformatsfromQPSKto16-QAM.Thissystemoffersadatathroughputupto37Mbit/soverabout1.5mdistance.

Fromalldemonstrationsdiscussedaboveitcanbeconcludedthatrate-adaptivesystemsbased on DMT modulation, as also used in radio and wired transmission, can beimplementedwithoff-the-shelfcomponents. Ithasbeenshownthatsuchsystemsareanexcellent solution for robust optical wireless-link systems with peak data rates up to500 Mbit/s under various lighting conditions. Nonetheless, the commercially availableDSPchips,whichofcoursearenotintendedforapplicationinVLCsystems,orsolutionsextensively tailored on FPGA, respectively, limit the data rate and the overall systemperformance.Moreover, thereare limits inoptimizationof importantparameterssuchaspower consumption. Full custom integrated circuits using current technologywould beabletoremovesuchshortcomingsbutwouldrequireenteringthemassmarket.

These experiments and measurements using real-time adaptive bidirectional VLCsystemsfulfillanimportantintermediatesteptoturntheopticalWiFivisionintoreality.The next steps will integrate VLC technology into real-life room lighting, and willimplementextensionssuchasMp-to-Mpfunctionality.

8.7 SummaryIn this chapter, the use of DMTmodulation schemes inVLC for indoor application ispresented along with the most important features and conditions. Items of VLC arediscussedwithfocusontheprovisionofhigh-speedlinksusingDMT-basedmodulation,i.e. only as far as there are special requirements of system design in that respect. Inaccordance with the present status of research and development on OWC around theworld,thePHYlayerisemphasizedinthischapter.

Against this backdrop, the optical wireless channel is briefly characterized, andmethods to exploit its capacity bymeans of typical high-brightness LEDs intended forlighting in combination with DMT-based modulation and IM/DD transmission arepresented.PropertiesofthebasicDMT/OFDMversions,namelyDC-biasedDMT,ACO-OFDMandPAM-DMTarediscussedandcompared.Moreover,severalvariantsoftheseschemesforuseinVLCsystemsarehighlighted.Inanutshell,DMT/OFDMmodulationis identifiedasamethodofchoice forexploitingVLCchannelsof restrictedbandwidthduetotheLEDdeviceandthedrivercharacteristics.ThemodulationschemesareabletotakeadvantageofthehighSNRthatisavailableinVLCscenariosbasedonroomlightingand typical brightness levels. A further important advantage is the ability to adapt themodulationformatofeachsubcarriertotheSNRofthechannel,commonlyknownasbitandpowerloading.Suchanadvancedexploitationof thechannelcapacityalsoprovidesflexibilityincoverage,butitrequiresafeedbacklink.

In addition, it has been shown that DMT modulation can also be applied to mostadvancedVLC transmission schemesusingWDMandMIMO techniques.BymeansofWDMthetransmissioncapacitycanbemultiplied,however,multi-colorLEDsaswellasmorecomplexorcoloredRxunitsarenecessary insuchascenario.Higherbit ratesarealsopossibleusingMIMOtechnologies.Relatedresearchisinanearlyphasebutatahighlevelofactivity.

ThenumerouscomparativeanalysesoftheDMT/OFDMmodulationformatsshowthatbothofthebasicschemesusingasymmetricalclipping,i.e.,ACO-OFDMandPAM-DMT,are best at low spectral efficiency and high power efficiency, i.e. where powerconsumption rather than bit rate is of importance. Their computational complexity isnearly the same. However, these schemes as well as variants proposed and underdiscussion have not yet left the laboratory stage due to the difficulty of real-timeimplementation, which is mandatory for VLC verification under conditions such asmobility.On the other hand,DC-biasedDMThas the lowest computational complexityamongthethreemajorDMT/OFDMmodulationformats.Theschemeisexpectedtoofferthehighestdatarateinapplications,wheretheadditionalDCbiaspowerrequiredinVLCto create a non-negativemodulation signal can provide a complementary functionality,suchasillumination.Infact,therecord-breakingresultsintransmissionrateaspresented

in[8,9,11,46]haveallbeenachievedusingDC-biasedDMT.Thefirstimplementationsand transceiver prototypes, as far as they have been made public, also use DC-biasedDMT including real-time signal processing, thus enabling a mobile VLC experience.Accordingtothis,thesuitabilityofDMTmodulationforrobustindoorVLCapplicationathigh data rates has been confirmed already by real-life verification and publicdemonstration.

ItisalsoimportanttonotethatthepoormodulationbandwidthofwhitelightLEDsoftheYB-typeisnotanissueanymore.Ithasbeendemonstratedthatthebluefilter,whichmitigatestheadverseeffectofthephosphorlayerbutisunfavorableconcerningthepowerbudget, canbe replacedbyequalization. In thatway, theopticalpower reaching theRxcanbefullyexploitedatbitratesofseveral100Mbit/sorevenmore.

Although DMT-based modulation has continually gained in popularity due to itsattractiveperformance, variousversionsof single-carrier frequencydomain equalization(SCFDE) transmission move into the VLC spotlight. These techniques offer reducedPAPR,whichmayresult inanoverallbetterperformance[132,133].Newly,DC-biasedDMTandSCFDEtransmissionhavebeencomparedinaWDMexperimentalsetup,whereaggregatedataratesof2.5Gbit/sand3.75Gbit/srespectively,havebeenachieved[134].Hence, the SCFDE scheme clearly outperformed DMT in that setup. Another single-carriermodulation scheme, namely carrier-less amplitude and phase (CAP)modulationhasbeenexploredtoo[79].Anaggregatebitrateof3.22Gbit/s,againachievedinaWDMexperiment,confirmsthesignificanceofsingle-carriermodulationasanimportantoptionforhigh-speedVLCsystems.Ways for furtherperformance improvementonPHY levelmay also be opened by a quasi balanced detection scheme forOFDM signals, recentlyintroducedin[135].

SofarworkonbothVLCandLED-basedlightinghasfocusedonvitalissuesofeachindividual domain.More specifically, lifetime, color, luminous efficacy, etc. of deviceshavebeensubjectsinvestigatedinLEDlightingdevelopment,whiletopicssuchasLEDmodulationanddriving,relatedsignalprocessingandlinkcontrolhavebeenaddressedincommunications. True VLC system design, however, needs multi-disciplinary workcoveringcommunicationsand lighting technologyaswell.Forexample,LEDlong-termbehaviorunderVLCoperationincludingapotentiallyhighPAPRneedstobeverified,assofarthereisonlyinitialexperience.RegardingLEDmodulationandpowersupplies,itmaybeusefultodriveLEDlampsasanintegralpartoftheexistinginfrastructuredirectlywithAC-power.BasicresultsonsuchaVLCapproach,yetatlowspeedandusingOOK,havebeenpublished recently in [136].Additional research isalsonecessary toexamineLED modulation effects on light quality in practical smart lighting scenarios. As iscorrectlypointedoutin[12]forthedual-usecase,VLCandlightintensitycontrolareinconflict.Thus,itisimportanttoconsiderDMT-compatible(hybrid)dimmingschemesinextendedVLCstandards.Theyshouldalsotakeintoaccountemergingstandardsonradioandwiredtransmission,whicharecloselyrelatedfromtheVLCsystemspointofview.Itisanopen issuewhether lightingandenergydemandswillbemet inall respects.Thus,this presents a field for further studies. More verification in real environments is alsonecessary,whichcallsforjointworkbythecommunicationsandlightingindustries.

Several VLC techniques have been verified by experiments with reliable off-line

processingandalsobya fewdemonstrationsusingcomponentsoff the shelf.Real low-costsystemsnowneedcustomdesigneddevices,suchasintegratedDSPforoptimizationofVLCperformanceandfeaturestobeconsidered,asstatedinthischapter.Inthismatter,interaction with DSP chip manufacturers is necessary along with ongoing work ofstandardizationatthesystemlevel.

Finally, it must be mentioned that system integration has been considered only at arudimentary level so far andmobility-related issues such as handover inVLChavenotbeenadequatelyaddressed.Moreover,adedicatedstandardizationroadmapisessentialforfuture availability of VLC in portable devices [137]. Standardization activities so faremanate from the Infrared Data Association (IrDA) interest group and from the IEEE.WhereastheIrDAmainlyprovidesspecificationsforwirelessinfraredprotocols,theIEEEhaspublishedafirstOWCstandard,IEEE802.15.7–2011,forVLC.Therecentextensionof the International Telecommunication Union (ITU) g.hn standard (ITU-TRecommendationG.9960,2011) foreseeinganopticalchannel isalsoof importance,cf.[102], concerning integration of VLC, e.g. into home networks and cooperation withpresentinfrastructure.

Even if thereare still challengesasnamedabove,whicharebeing facedby researchand development,VLC presents a realistic and promising supplementary technology toradiocommunication.

References[1] M.Kavehrad,“Sustainableenergy-efficientwirelessapplicationsusing light,” IEEECommunicationsMagazine,48,(12),66–73,2010.

[2] A.Sevincer,A.Bhattarai,M.Bilgi,M.Yuksel, andN. Pala, “LIGHTNETs: Smartlighting and mobile optical wireless networks – a survey,” IEEE CommunicationsSurveys&Tutorials,15,(4),1620–1641,2013.

[3] S. Haruyama, “Visible light communication using sustainable LED lights,”Proceedingsof5thITUKaleidoscope:BuildingSustainableCommunities,2013.

[4] L. Hanzo, H. Haas, S. Imre, et al., “Wireless myths, realities, and futures: From3G/4G to optical and quantumwireless,”Proceedings of the IEEE, 100, 1853–1888,2012.

[5] D.K.Borah,A.C.Boucouvalas,C.C.Davis,S.Hranilovic,andK.Yiannopoulos,“Areview of communication-oriented optical wireless systems,” EURASIP Journal onWirelessCommunicationsandNetworking,91,1–28,2012.

[6] K.-D.Langer,J.Hilt,D.Schulz,etal.,“Rate-adaptivevisiblelightcommunicationat500 Mb/s arrives at plug and play,” Optoelectronics and Communications SPIENewsroom,DOI10.1117/2.1201311.005196,2013.

[7] J.Vučić,C.Kottke,S.Nerreter,etal.,“230Mbit/sviaawirelessvisible-light linkbased onOOKmodulation of phosphorescent white LEDs,”OFC/NFOECTechnicalDigest2010,paperOThH3.

[8] C.Kottke,J.Hilt,K.Habel, J.Vučić, andK.-D.Langer, “1.25Gbit/s visible light

WDMlinkbasedonDMTmodulationofasingleRGBLEDluminary,”Proc.EuropeanConferenceandExhibitiononOpticalCommunication(ECOC)2012,paperWe.3.B.4.

[9] G.Cossu,A.M.Khalid, P. Choudhury, R. Corsini, and E. Ciaramella, “3.4Gbit/svisible optical wireless transmission based on RGBLED,”Optics Express, 20, (26),B501–B506,2012.

[10] Y.Wang,Y.Shao,H.Shang,etal.,“875-Mb/sasynchronousbi-directional64QAM-OFDM SCM-WDM transmission over RGB-LED-based visible light communicationsystem,”OFC/NFOECTechnicalDigest2013,paperOTh1G.3.

[11] D. Tsonev,H. Chun, S. Rajbhandari, et al., “A 3-Gb/s single-LED OFDM-basedwirelessVLClinkusingaGalliumNitrideμLED,”IEEEPhotonicsTechnologyLetters,26,(7),637–640,2014.

[12] J.Gancarz,H.Elgala,andT.D.C.Little,“ImpactoflightingrequirementsonVLCsystems,”IEEECommunicationsMagazine,51,(12),34–41,2013.

[13] R. D. Roberts, S. Rajagopal, and S.-K. Lim, “IEEE 802.15.7 physical layersummary,”Proc.IEEEGLOBECOMWorkshops2011,pp.772–776.

[14] S. Rajagopal, R. D. Roberts, and S.-K. Lim, “IEEE 802.15.7 visible lightcommunication:Modulation schemes and dimming support,” IEEEComm.Mag., 50,(3),72–82,2012.

[15] D. Tsonev, S. Videv, and H. Haas, “Light fidelity (Li-Fi): Towards all-opticalnetworking,”Proc.SPIE9007,Broadband Access Communication Technologies VIII,900702,2013.

[16] J.M.KahnandJ.R.Barry,“Wirelessinfraredcommunications,”ProceedingsoftheIEEE,85,(2),265–298,1997.

[17] J. Grubor, V. Jungnickel, K.-D. Langer, and C. v. Helmolt, “Dynamic data-rateadaptivesignalprocessingmethodinawirelessinfra-reddatatransfersystem,”PatentEP1897252B1,24June2005.

[18] O.Gonzalez,R.Perez-Jimenez,S.Rodriguez, J.Rabadan, andA.Ayala,“OFDMover indoorwireless optical channel,” IEEProc.Optoelectronics, 152, (4), 199–204,2005.

[19] L. Grobe, A. Paraskevopoulos, J. Hilt, et al., “High-speed visible lightcommunicationsystems,”IEEECommunicationsMagazine,51,(12),60–66,2013.

[20] Z.Ghassemlooy,H.LeMinh,P.A.Haigh,andA.Burton,“Developmentofvisiblelightcommunications:Emergingtechnologyandintegrationaspects,”Proc.OpticsandPhotonicsTaiwanInternationalConference(OPTIC),2012.

[21] G. Ntogari, T. Kamalakis, J. W. Walewski, and T. Sphicopoulos, “Combiningillumination dimming based on pulse-width modulation with visible-lightcommunications based on discrete multitone,” Journal of Optical Comm. andNetworking,3,(1),56–65,2011.

[22] J.Grubor,S.C.J.Lee,K.-D.Langer,T.Koonen,andJ.Walewski,“Wirelesshigh-speed data transmission with phosphorescent white-light LEDs,” Proc. European

ConferenceandExhibitiononOpticalCommunication(ECOC)2007,6,Post-DeadlinePaperPD3.6.

[23] C.W.Chow,C.H.Yeh,Y.F.Liu,andY.Liu,“ImprovedmodulationspeedofLEDvisiblelightcommunicationsystemintegratedtomainelectricitynetwork,”El.Letters,47,(15),2011.

[24] Y.Pei,S.Zhu,H.Yang,etal., “LEDmodulation characteristics in a visible-lightcommunicationsystem,”OpticsandPhotonicsJournal,3(2B),139–142,2013.

[25] J.Grubor,S.Randel,K.-D.Langer, and J.W.Walewski, “Broadband informationbroadcastingusingLED-basedinteriorlighting,”J.ofLightwaveTech.,26,(24),3883–3892,2008.

[26] J.Armstrong,R.J.Green,andM.D.Higgins,“Comparisonofthreereceiverdesignsfor optical wireless communications using white LEDs,” IEEE CommunicationsLetters,16,(5),748–751,2012.

[27] D.C.O’Brien,L.Zeng,H.Le-Minh,etal.,“Visible lightcommunications,” inR.Kraemer, M. D. Katz (eds.), Short-Range Wireless Communications, pp. 329–342,Wiley&SonsLtd.,2009.

[28] C.W.Chow,C.H.Yeh,Y.Liu,andY.F.Liu,“Digitalsignalprocessingfor lightemittingdiodebasedvisiblelightcommunication,”IEEEPhot.SocietyNewsletter,26,(5),9–13,2012.

[29] H. Le-Minh, D. C. O’Brien, G. Faulkner, et al., “80 Mbit/s visible lightcommunicationsusingpre-equalizedwhiteLED,”Proc.34thEuropeanConferenceandExhibitiononOpticalCommunication(ECOC),2008.

[30] J.Vučić,C.Kottke,S.Nerreter,K.-D.Langer, and J.W.Walewski, “513Mbit/svisiblelightcommunicationslinkbasedonDMT-modulationofawhiteLED,”JournalofLightwaveTechnology,28,(24),3512–3518,2010.

[31] H. Chun, C.-J. Chiang, and D. C. O’Brien, “Visible light communication usingOLEDs: Illumination and channel modeling,” Int. Workshop on Optical WirelessCommunications(IWOW),2012.

[32] P.A.Haigh,Z.Ghassemlooy,I.Papakonstantinou,andH.LeMinh,“2.7Mb/switha 93 kHz white organic light emitting diode and real time ANN equalizer,” IEEEPhotonicsTechnologyLetters,25,(17),1687–1690,2013.

[33] J. Grubor andK.-D. Langer, “Efficient signal processing in OFDM-based indooropticalwirelesslinks,”JournalofNetworks,5,(2),197–211,2010.

[34] J.Vučić,“Adaptivemodulationtechniqueforbroadbandcommunicationin indooroptical wireless systems,” PhD Thesis at Technische Universitaet Berlin, Germany,2009.

[35] X. Li, J. Vučić, V. Jungnickel, and J. Armstrong, “On the capacity of intensity-modulateddirect-detectionsystemsandtheinformationrateofACO-OFDMforindooroptical wireless applications,” IEEE Transactions on Communications, 60, (3), 799–809,2012.

[36] S. Dimitrov and H. Haas, “Information rate of OFDM-based optical wirelesscommunicationsystemswithnonlineardistortion,”J.ofLightwaveTech.,31,(6),918–929,2013.

[37] X.Zhang,K.Cui,M.Yao,H.Zhang,andZ.Xu,“Experimentalcharacterizationofindoor visible light communication channels,” Proc. 8th International Symposium onCommunicationSystems,Networks&DigitalSignalProcessing(CSNDSP),2012.

[38] N. Fujimoto and H. Mochizuki, “477 Mbit/s visible light transmission based onOOK-NRZ modulation using a single commercially available visible LED and apracticalLEDdriverwithapre-emphasiscircuit,”OFC/NFOECTechnicalDigest2013,paperJTh2A.73.

[39] I. Neokosmidis, T. Kamalakis, J. W. Walewski, B. Inan, and T. Sphicopoulos,“Impact of nonlinear LED transfer function on discrete multitone modulation:Analyticalapproach,”JournalofOpticalCommunicationsandNetworking,1,(5),439–451,2009.

[40] I.Stefan,H.Elgala,R.Mesleh,D.O’Brien,andH.Haas,“OpticalwirelessOFDMsystem on FPGA: Study of LED nonlinearity effects,” Proc. 73rd IEEE VehicularTechnologyConference(VTCSpring),pp.1–5,2011.

[41] S.-B.Ryu,J.-H.Choi,J.Bok,H.Lee,andH.-G.Ryu,“Highpowerefficiencyandlow nonlinear distortion for wireless visible light communication,” Proc. 4th IFIPInternationalConferenceonNewTechnologies,MobilityandSecurity(NTMS),pp.1–5,2011.

[42] D.Lee,K.Choi,K.-D.Kim,andY.Park, “Visible lightwireless communicationsbasedonpredistortedOFDM,”OpticsCommunications,285,(7),1767–1770,2012.

[43] D.Tsonev,S.Sinanovic,andH.Haas,“CompletemodelingofnonlineardistortioninOFDM-based optical wireless communication,” Journal of Lightwave Technology,31,(18),3064–3076,2013.

[44] R. Mesleh, H. Elgala, and H. Haas, “LED nonlinearity mitigation techniques inopticalwirelessOFDMcommunicationsystems,”JournalofOpticalCommunicationsandNetworking,4,(11),865–875,2012.

[45] R.Mesleh,H.Elgala,andH.Haas,“OntheperformanceofdifferentOFDMbasedoptical wireless communication systems,” Journal of Optical Communications andNetworking,3,(8),620–628,2011.

[46] A. M. Khalid, G. Cossu, R. Corsini, P. Choudhury, and E. Ciaramella, “1-Gb/stransmission over a phosphorescent white LED by using rate-adaptive discretemultitonemodulation,”IEEEPhotonicsJournal,4,(5),1465–1473,2012.

[47] B. Inan, S. C.J. Lee, S. Randel, et al., “Impact of LED nonlinearity on discretemultitone modulation,” Journal of Optical Communications and Networking, 1, (5),439–451,2009.

[48] C. Ma, H. Zhanga, K. Cuib, M. Yaoa, and Z. Xu, “Effects of LED lightingdegradation and junction temperature variation on the performance of visible lightcommunication,” International Conference on Systems and Informatics (ICSAI),

pp.1596–1600,2012.

[49] J.B.CarruthersandJ.M.Kahn,“Multiple-subcarriermodulation fornon-directedwirelessinfraredcommunication,”IEEEJ.onSelectedAreasinComm.,14, (3),538–546,1996.

[50] Y.Tanaka,T.Komine,S.Haruyama,andM.Nakagawa,“Abasic studyofopticalOFDM system for indoor visible communication utilizing plural white LEDs aslighting,”8th Int.Symp.onMicrowaveandOpticalTechnol. (ISMOT), pp. 303–306,2001.

[51] J. Armstrong and A. J. Lowery, “Power efficient optical OFDM,” ElectronicsLetters,42,(6),370–372,2006.

[52] S.C.J.Lee,S.Randel,F.Breyer, andA.M.J.Koonen, “PAM-DMTfor intensity-modulated and direct-detection optical communication systems,” IEEE PhotonicsTechnologyLetters,21,(23),1749–1751,2009.

[53] S. Randel, F. Breyer, and S. C.J. Lee, “High-speed transmission over multimodeoptical fibers,” Proc. 34th European Conference and Exhibition on OpticalCommunication(ECOC)2008,paperOWR2.

[54] J.M.Cioffi,“Amulticarrierprimer,”ANSIContributionT1E1,4,91–157,1991.

[55] J. Armstrong, “OFDM for optical communications,” Journal of LightwaveTechnology,27,(3),189–204,2009.

[56] A. V. Oppenheim and R.W. Schafer,Discrete-Time Signal Processing, Prentice-Hall,1989.

[57] S. K. Hashemi, Z. Ghassemlooy, L. Chao, and D. Benhaddou, “Orthogonalfrequency division multiplexing for indoor optical wireless communications usingvisible light LEDs,” Proc. 6th Int. Symp. on Communication Systems, Networks &DigitalSignalProcessing(CSNDSP)2008,pp.174–178.

[58] J. Armstrong, B. J.C. Schmidt, D. Kalra, H. A. Suraweera, and A. J. Lowery,“Performance of asymmetrically clipped optical OFDM in AWGN for an intensitymodulated direct detection system,” Proc. IEEE Global Telecommunications Conf.(GLOBECOM’06),SPC07–4,2006.

[59] S. C.J. Lee, F. Breyer, S. Randel, H. P.A. van den Boom, and A. M.J. Koonen,“High-speed transmissionovermultimode fiberusingdiscretemultitonemodulation,”JournalofOpticalNetworking,7,(2),183–196,2008.

[60] E.Vanin,“Signalrestorationinintensity-modulatedopticalOFDMaccesssystems,”OpticsLetters,36,(22),4338–4340,2011.

[61] X. Li, R. Mardling, and J. Armstrong, “Channel capacity of IM/DD opticalcommunication systems and of ACO-OFDM,” Proc. Int. Conf. on Communications(ICC)2007,pp.2128–2133,2007.

[62] S.K.Wilson and J.Armstrong, “Transmitter and receivermethods for improvingasymmetrically-clippedopticalOFDM,”IEEETrans.onWirelessComm.,8,(9),4561–4567,2009.

[63] S. C.J. Lee, F. Breyer, S. Randel, et al., “Discrete multitone modulation formaximizingtransmissionrateinstep-indexplasticopticalfibers,”JournalofLightwaveTechnology,27,(11),1503–1513,2009.

[64] L. Chen, B. Krongold, and J. Evans, “Performance analysis for optical OFDMtransmissioninshort-rangeIM/DDsystems,”JournalofLightwaveTechnology,30,(7),974–983,2012.

[65] S.Tian,K.Panta,H.A.Suraweera,etal.,“Anoveltimingsynchronizationmethodfor ACO-OFDM-based optical wireless communications,” IEEE Transactions onWirelessCommunications,7,(12),4958–4967,2008.

[66] M.M.FredaandJ.M.Murray,“Low-complexityblindtimingsynchronizationforACO-OFDM-based optical wireless communications,” Proc. IEEE GLOBECOMWorkshops2010,pp.1031–1036.

[67] R.YouandJ.M.Kahn,“Averagepowerreductiontechniquesformultiplesubcarrierintensity-modulated optical signals,” IEEE Trans. Communications, 49, (12), 2164–2171,2001.

[68] B. Ranjha and M. Kavehrad, “Precoding techniques for PAPR reduction inasymmetrically clipped OFDM based optical wireless system,” Proc. SPIE 8645,BroadbandAccessCommunicationTechnologiesVII,86450R,2013.

[69] J.ArmstrongandB.J.C.Schmidt,“ComparisonofasymmetricallyclippedopticalOFDMandDC-biasedopticalOFDMinAWGN,”IEEEComm.Letters,12, (5),343–345,2008.

[70] D.J.F.Barros,S.K.Wilson,andJ.M.Kahn,“Comparisonoforthogonalfrequency-divisionmultiplexingandpulse-amplitudemodulationinindooropticalwirelesslinks,”IEEETransactionsonCommunications,60,(1),153–163,2012.

[71] S.Dimitrov,S.Sinanovic,andH.Haas,“Signalshapingandmodulationforopticalwirelesscommunication,”JournalofLightwaveTechnology,30,(9),1319–1328,2012.

[72] S.D.Dissanayakeand J.Armstrong, “Comparison ofACO-OFDM,DCO-OFDMand ADO-OFDM in IM/DD systems,” Journal of Lightwave Technology., 31, (7),1063–1072,2013.

[73] Z.Yu, R. J. Baxley, andG. T. Zhou, “EVM and achievable data rate analysis ofclippedOFDMsignalsinvisiblelightcommunication,”EURASIPJournalonWirelessCommunicationsandNetworking,(1),1–16,2012.

[74] L. Chen, B. Krongold, and J. Evans, “Theoretical characterization of nonlinearclipping effects in IM/DD optical OFDM systems,” IEEE Transactions onCommunications,60,(8),2304–2312,2012.

[75] S.Dimitrov, S. Sinanovic, andH.Haas, “Clipping noise in OFDM-based opticalwireless communication systems,” IEEE Trans. on Communications, 60, (4), 1072–1081,2012.

[76] C.W.Chow,C.H.Yeh,Y. F. Liu, and P.Y.Huang, “Background optical noisescircumvention inLEDopticalwireless systemsusingOFDM,” IEEEPhot. J.,5, (2),

7900709,2013.

[77] C. Kottke, K. Habel, L. Grobe, et al., “Single-channel wireless transmission at806Mbit/susingawhite-lightLEDandaPIN-basedreceiver,”Proc.14thInt.Conf.onTransparentOpticalNetworks(ICTON),paperWe.B4.1,2012.

[78] Y. Wang, Y. Wang, N. Chi, J. Yu, and H. Shang, “Demonstration of 575-Mb/sdownlinkand225-Mb/suplinkbi-directionalSCM-WDMvisiblelightcommunicationusingRGBLEDandphosphor-basedLED,”OpticsExpress,21,(1),1203–1208,2013.

[79] F.M.Wu,C.T.Lin,C.C.Wei,etal.,“PerformancecomparisonofOFDMsignaland CAP signal over high capacity RGB-LED-based WDM visible lightcommunication,”IEEEPhotonicsJournal,5,(4),7901507,2013.

[80] S. D. Dissanayake, K. Panta, and J. Armstrong, “A novel technique tosimultaneously transmit ACO-OFDM and DCO-OFDM in IM/DD systems,” Proc.IEEEGLOBECOMWorkshops2011,pp.782–786.

[81] K. Asadzadeh, A. A. Farid, and S. Hranilovic, “Spectrally factorized opticalOFDM,”Proc.12thCanadianWorkshoponInformationTheory(CWIT),pp.102–105,2011.

[82] N. Fernando, Y. Hong, and E. Viterbo, “Flip-OFDM for optical wirelesscommunications,”Proc.InformationTheoryWorkshop(ITW),5–9,2011.

[83] Y.-I. Jun, “Modulation and demodulation apparatuses and methods for wired /wirelesscommunication,”PatentWO/2007/064165,2007.

[84] D. Tsonev, S. Sinanovic, and H. Haas, “Novel unipolar orthogonal frequencydivision multiplexing (U-OFDM) for optical wireless,” IEEE 75th VehicularTechnologyConference(VTCSpring),pp.1–5,2012.

[85] A.Nuwanpriya,A.Grant,S.-W.Ho,andL.Luo,“PositionmodulatingOFDMforoptical wireless communications,” Proc. 3rd IEEE Workshop on Optical WirelessCommunications(OWC’12),pp.1219–1223,2012.

[86] L. Chen, B. Krongold, and J. Evans, “Diversity combining for asymmetricallyclipped optical OFDM in IM/DD channels,” Proc. IEEE Global Telecomm. Conf.(GLOBECOM’09),pp.1–6,2009.

[87] S.D.Dissanayake,J.Armstrong,andS.Hranilovic,“Performanceanalysisofnoisecancellation in a diversity combined ACO-OFDM system,” Proc. 14th Int. Conf. onTransparentOpticalNetworks(ICTON),2012.

[88] M.Z.FarooquiandP.Saengudomlert,“TransmitpowerreductionthroughsubcarrierselectionforMC-CDMA-basedindooropticalwirelesscommunicationswithIM/DD,”EURASIPJournalonWirelessCommunicationsandNetworking,(1),1–14,2013.

[89] R. Zhang and L. Hanzo, “Multi-layer modulation for intensity modulated directdetection optical OFDM,” J. of Optical Communications and Networking, 5, (12),1402–1412,2013.

[90] G.Cossu,A.M.Khalid,R.Corsini,andE.Ciaramella,“Non-directedline-of-sightvisiblelightsystem,”OFC/NFOECTechnicalDigest2013,paperOTh1G.2.

[91] C.H.Yeh,Y.-L.Liu,andC.-W.Chow,“Real-timewhite-lightphosphor-LEDvisiblelightcommunication(VLC)withcompactsize,”Opt.Express,21,(22),26192–26197,2013.

[92] J. Grubor, V. Jungnickel, and K.-D. Langer, “Adaptive optical wireless OFDMsystemwithcontrolledasymmetricclipping,”IEEEProc.41stAsilomarConferenceonSignals,SystemsandComputers,2007.

[93] Z.Sun,Y.Zhu,andY.Zhang,“TheDMT-basedbit-powerallocationalgorithmsinthe visible light communication,” Proc. 2nd International Conference on BusinessComputingandGlobalInformatization,pp.572–575,2012.

[94] K.-D.Langer, J.Vučić,C.Kottke,etal., “Advances and prospects in high-speedinformation broadcast,” Proc. 11th Int. Conf. on Transparent Optical Networks(ICTON),paperMo.B5.3,2009.

[95] J.Vučić,C.Kottke,S.Nerreter,etal.,“White lightwireless transmissionat200+Mb/s net data rate by use of discrete-multitone modulation,” IEEE PhotonicsTechnologyLetters,21,(20),1511–1513,2009.

[96] D. Bykhovsky and S. Arnon, “An experimental comparison of different bit-and-power-allocation algorithms forDCO-OFDM,”JournalofLightwaveTechnology, 32,(8),1559–1564,2014.

[97] C.W.Chow,C.H.Yeh,Y.F.Liu,P.Y.Huang,andY.Liu,“Adaptiveschemeformaintaining the performance of the in-home white-LED visible light wirelesscommunicationsusingOFDM,”OpticsCommunications,292,(1),49–52,2013.

[98] K. L. Sterckx, “Implementation of continuous VLC modulation schemes oncommercial LED spotlights,” Proc. 9th International Conference on ElectricalEngineering/Electronics, Computer, Telecommunications and Information Technology(ECTI-CON),2012.

[99] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wireless communication:Potentialandstate-of-the-art,”IEEECommunicationsMagazine,49,(9),56–62,2011.

[100] T. Komine, S. Haruyama, and M. Nakagawa, “Performance evaluation ofnarrowbandOFDMonintegratedsystemofpowerlinecommunicationandvisiblelightwirelesscommunication,”Proc.1stInt.Symp.onWirelessPervasiveComputing,2006.

[101] S.E.Alavi,A.S.M.Supa’at,S.M.Idrus,andS.K.Yusof,“Newintegratedsystemof visible free space optic with PLC,” Proc. 3rd Workshop on Power LineCommunications(WSPLC),2009.

[102] H.Ma,L.Lampe,andS.Hranilovic,“Integrationofindoorvisiblelightandpowerline communication systems,” Proc. 17th IEEE International Symposium on PowerLineCommunicationsanditsApplications(ISPLC),pp.291–296,2013.

[103] K.-D.Langer,J.Grubor,O.Bouchet,etal.,“Opticalwirelesscommunicationsforbroadbandaccessinhomeareanetworks,”Proc.10thInt.Conf.onTransparentOpticalNetworks(ICTON),4,149–154,2008.

[104] O.Bouchet,P.Porcon,andE.Gueutier,“Broadcastof fourHDvideoswithLED

ceiling lighting: Optical-wireless MAC,” Proc. SPIE 8162, Free-Space andAtmosphericLaserCommunicationsXI,81620L,2011.

[105] O.Bouchet,P.Porcon,J.W.Walewski,etal.,“Wirelessopticalnetworkforahomenetwork,”Proc.SPIE7814,Free-SpaceLaserCommunicationsX,781406,2010.

[106] M. V. Bhalerao, S. S. Sonavane, and V. Kumar, “A survey of wirelesscommunication using visible light,” Int. Journal of Advances in Engineering &Technology,5,(2),188–197,2013.

[107] J.DangandZ.Zhang,“ComparisonofopticalOFDM-IDMAandopticalOFDMAfor uplink visible light communications,” Proc. International Conference onWirelessCommunications&SignalProcessing(WCSP),2012.

[108] T. Borogovac, M. B. Rahaim,M. Tuganbayeva, and T. D.C. Little, “Lights-offvisible light communications,” Proc. IEEEGLOBECOMWorkshops 2011, pp. 797–801.

[109] H. Elgala and T. D.C. Little, “Reverse polarity optical-OFDM (RPO-OFDM):DimmingcompatibleOFDMforgigabitVLClinks,”OpticsExpress,21,(20),24288–24299,2013.

[110] R.Li,Y.Wang,C.Tang,etal.,“Improvingperformanceof750-Mb/svisiblelightcommunicationsystemusingadaptiveNyquistwindowing,”ChineseOpticsLetters,11,(8),080605/1–4,2013.

[111] J.Vučić,C.Kottke,K.Habel,andK.-D.Langer,“803Mbit/svisiblelightWDMlink based on DMT modulation of a single RGB LED luminary,” OFC/NFOECTechnicalDigest2011,paperOWB6.

[112] A. H. Azhar, T.-A. Tran, and D. O’Brien, “Demonstration of high-speed datatransmission using MIMO-OFDM visible light communications,” Proc. IEEEGLOBECOMWorkshops2010,pp.1052–1056.

[113] A.H.Azhar,T.-A.Tran,andD.O’Brien,“Agigabit/sindoorwirelesstransmissionusingMIMO-OFDMvisible-lightcommunications,”IEEEPhotonicsTech.Letters,25,171–174,2013.

[114] X.Zhang,S.Dimitrov,S.Sinanovic, andH.Haas, “Optimal power allocation inspatial modulation OFDM for visible light communications,” Proc. IEEE 75thVehicularTechnologyConference(VTCSpring),pp.1–5,2012.

[115] X.DiRenzo,H.Haas,A.Ghrayeb,S.Sugiura,andL.Hanzo,“SpatialmodulationforgeneralizedMIMO:Challenges,opportunities,andimplementation,”ProceedingsoftheIEEE,102,(1),56–103,2014.

[116] Y. Li, D. Tsonev, and H. Haas, “Non-DC-biased OFDM with optical spatialmodulation,” Proc. IEEE 24th Int. Symp. on Personal, Indoor and Mobile RadioCommunications(PIMRC),pp.486–490,2013.

[117] M.S.Moreolo,R.Muñoz,andG.Junyent,“NovelpowerefficientopticalOFDMbasedonHartley transformfor intensity-modulateddirect-detectionsystems,”JournalofLightwaveTechnology,28,(5),798–805,2010.

[118] G.delCampoJiménezandF.J.LópezHernándeza,“VLCorientedenergyefficientdrivertechniques,”Proc.SPIE8550,OpticalSystemsDesign2012,85502F.

[119] T.Kishi, H. Tanaka, Y. Umeda, and O. Takyu, “A high-speed LED driver thatsweeps out the remaining carriers for visible light communications,” Journal ofLightwaveTechnology,32,(2),239–249,2014.

[120] L. Grobe and K.-D. Langer, “Block-based PAM with frequency domainequalization in visible light communications,” Proc. IEEE GLOBECOMWorkshops2013,pp.1075–1080,2013.

[121] K.Asadzadeh,A.Dabbo,andS.Hranilovic,“ReceiverdesignforasymmetricallyclippedopticalOFDM,”Proc.IEEEGLOBECOMWorkshops2011,pp.777–781.

[122] X.Yang,Z.Min,T.Xiongyan,W.Jian,andH.Dahai,“Apost-processingchannelestimationmethodforDCO-OFDMvisiblelightcommunication,”Proc.8thInt.Symp.onCommunicationSystems,Networks&DigitalSignalProcessing(CSNDSP),2012.

[123] G.Stepniak,J.Siuzdak,andP.Zwierko,“CompensationofaVLCphosphorescentwhite LED nonlinearity by means of Volterra DFE,” IEEE Photonics TechnologyLetters,25,(16),1597–1600,2013.

[124] Z.Wang,C.Yu,W.-D.Zhong,andJ.Chen,“Performanceimprovementbytiltingreceiverplane inM-QAMOFDMvisible light communications,”OpticsExpress, 19,(14),13418–13427,2011.

[125] A. H. Azhar and D. O’Brien, “Experimental comparisons of optical OFDMapproaches in visible light communications,” Proc. IEEE GLOBECOM Workshops2013,pp.1076–1080,2013.

[126] M. Z.Afgani,H.Haas,H. Elgala, and D. Knipp, “Visible light communicationusing OFDM,” Proc. 2nd International Conference on Testbeds and ResearchInfrastructures for theDevelopment of Networks andCommunities (TRIDENTCOM),pp.129–134,2006.

[127] H. Elgala, R. Mesleh, H. Haas, and B. Pricope, “OFDM visible light wirelesscommunicationbasedonwhiteLEDs,”Proc.65thIEEEVehicularTechnol.Conf.(VTCSpring),pp.2185–2189,2007.

[128] H. Elgala, R. Mesleh, and H. Haas, “Indoor broadcasting via white LEDs andOFDM,”IEEETransactionsonConsumerElectronics,55,(3),1127–1134,2009.

[129] J.Vučić,L.Fernández,C.Kottke,K.Habel,andK.-D.Langer,“Implementationof a real-timeDMT-based 100Mbit/s visible-light link,” Proc. EuropeanConferenceandExhibitiononOpticalCommunication(ECOC)2010,paperWe.7.B.1.

[130] K.-D. Langer, J. Vučić, C. Kottke, et al., “Exploring the potentials of optical-wireless communication using white LEDs,” Proc. 13th Int. Conf. on TransparentOpticalNetworks(ICTON),paperTu.D5.2,2011.

[131] J.Vučić andK.-D. Langer, “High-speed visible light communications: State-of-the-art,”OFC/NFOECTechnicalDigest2012,paperOTh3G.3.

[132] M.Wolf,L.Grobe,M.R.Rieche,A.Koher, and J.Vučić, “Block transmission

with linear frequency domain equalization for dispersive optical channelswith directdetection,” Proc. 12th Int. Conf. on Transparent Optical Networks (ICTON), paperTh.A3.4,2010.

[133] K.Acolatse,Y.Bar-Ness, and S.K.Wilson, “Novel techniques of single-carrierfrequency-domain equalization for optical wireless communications,” EURASIPJournalonAdvancesinSignalProcessing,2011,articleID393768.

[134] N.Chi,Y.Wang,Y.Wang,X.Huang, andX. Lu, “Ultra-high-speed single red-green-blue light-emitting diode-based visible light communication system utilizingadvancedmodulationformats,”ChineseOpticsLetters,12,(1),010605/1–4,2014.

[135] Y.Wang,N.Chi,Y.Wang,etal.,“High-speedquasi-balanceddetectionOFDMinvisiblelightcommunication,”OpticsExpress,21,(23),27558–27564,2013.

[136] Y.F.Liu,C.H.Yeh,C.W.Chow,andY.L.Liu,“AC-basedphosphorLEDvisiblelight communication by utilizing novel signal modulation,” Optical and QuantumElectronics,45,(10),1057–1061,2013.

[137] G. Dede, T. Kamalakis, and D. Varoutas, “Evaluation of optical wirelesstechnologiesinhomenetworking:ananalyticalhierarchyprocessapproach,”JournalofOpticalCommunicationsandNetworking,3,(11),850–859,2011.

9 Imagesensorbasedvisiblelightcommunication

ShinichiroHaruyamaandTakayaYamazato

9.1 OverviewImage sensors are used in digital cameras and a large number of imaging devices forindustrial, media, medical, and consumer electronics applications. An image sensorconsistsofanumberofpixels;eachpixelcontainsaphotodiode(PD),whichisgenerallyused as a receiver in visible light communications (VLC). Thus, an image sensorconsisting of a number of pixels can also be used as a VLC receiver. A particularadvantageofusing imagesensors,due to themassivenumberofavailablepixels, is theability tospatiallyseparatesources.Owing to thespatial separationofmultiplesources,theVLC receiver uses only the pixels that senseLED transmission sources, discardingother pixels, including those sensing ambient noise. The ability to spatially separatesourcesalsoprovidesanadditionalfeaturetoVLC,i.e.,theabilitytoreceiveandprocessmultipletransmittingsources.

In this chapter, we introduce VLC using an image sensor [1]. After presenting anoverviewofimagesensorsinSection9.2,we introduce theuseofan imagesensorasaVLCreceiverinSection9.3.WeprovideadesignofanimagesensorbasedVLCsysteminSection9.4.InSections9.5and9.6,werespectivelyintroducethefollowingtwouniqueapplicationsofVLCsystemsusing an image sensor: (1)massivelyparallel visible lighttransmission that can theoretically achieve a maximum data rate of 1.28 Gigabits persecond;and(2)accuratesensorposeestimationthatcannotberealizedbyaVLCsystemusinga single-elementPD.Applicationsof image sensorbasedcommunicationare alsopresentedinSection9.7fortrafficsignalcommunication,positionmeasurementsincivilengineering, and bridge positionmonitoring. Finally, in Section9.8,we summarize ourconclusions.

9.2 ImagesensorsAnimagesensorisadevicethatconvertsanopticalimageintoanelectronicsignal.Imagesensorsareusedindigitalcameras,cameramodules,videorecorders,andotherimagingdevices.Aswedescribelater,imagesensorscanalsobeusedasVLCreceivers.

An image sensor consists of n × m pixels, ranging from 320 × 240 (QVGA) to157 000 × 18 000 (line scanner). Each pixel contains a photodetector and devices forreadoutcircuits.Thepixelsizerangesfrom3×3to15×15μm2,limitedbythedynamicrangeandcostof theoptics.Thefractionofpixelareaoccupiedby thephotodetector iscalledthefillfactor.Thefillfactorrangesfrom0.2to0.9;ahighfillfactorisdesirable.Thereadoutcircuitsincludedinpixeldevicesdeterminethesensorconversiongain,whichis theoutputvoltageperphotoncollectedby thePD.The readout speeddetermines theframerate,whichistypically30framespersecond(fps).Highframeratesarerequiredformany industrial andmeasurement applications.Needless to say,VLCmust use a high-frame-rateimagesensor.

Twomajortypesofimagesensorsarethechargedcoupleddevice(CCD)imagesensorand the complementarymetal oxide semiconductor (CMOS) image sensor [2,3].AsidefromCCDandCMOSsensors,a two-dimensionalPDarray isoftenused formassivelyparallelVLC.

9.2.1 CCDimagesensorFigure9.1showstheblockdiagramofaCCDimagesensor.Inthefigure,whenexposureiscomplete,aCCDsequentiallytransferseachpixel’schargepacket.Then,thechargeisconvertedintovoltageanddirectedoff-chip.InCCDs,incidentphotonsareconvertedintoelectric charges and accumulated during the exposure time in a photodetector. BecauseCCDsuseoptimizedphotodetectors,theycanofferhighuniformity,lownoiselevels,andlowdarkcurrentincombinationwithahighfillfactor,highquantumefficiency,andhighsensitivity; however, they do suffer from a number of drawbacks. For example, theycannotbeintegratedwithotheranalogordigitalcircuits,suchasclockgenerators,controlcircuits, or analog-to-digital convertors.Moreover, they require high amounts of power,andduetotherequiredincreaseintransferspeeds, theirframerateislimited,especiallyforlargesensors.

Figure9.1 ACCDimagesensor.

9.2.2 CMOSimagesensorFigure9.2showstheblockdiagramofaCMOSimagesensor.Thesesensorshavegainedpopularity in recent years because of their advances in multi-functionalization, low

manufacturing costs, and low power consumption. The key element of aCMOS imagesensor is thePD,which isonecomponentofapixel.PDsare typicallyorganized inanorthogonalgrid.Duringoperation, light(photons)passingthrougha lensstrikes thePD,where it is converted into a voltage signal and passed through an analog-to-digitalconverter.Theconverteroutputisoftenreferredtoasaluminance.SinceaCMOSimagesensoriscomposedofaPDarray,PDoutputs,i.e.,lightintensityorluminancevalues,arearrangedinasquarematrixtoformadigitalelectronicrepresentationofthescene.

Figure9.2 ACMOSimagesensor.

The primary difference between CCD and CMOS image sensors is the readoutarchitecture; for CCDs, charge is shifted out by the vertical and the horizontal transferCCD, while for CMOS image sensors, charge or voltage is read out using a row andcolumndecoder,similartoadigitalmemory.

9.2.3 ComparingCCDimagesensors,CMOSimagesensors,andphotodiodes(PD)Table 9.1 compares CCD image sensors, CMOS image sensors, and PDs. Forcommunication,sincePDismuchfasterthantheCCDandCMOSimagesensors,asinglePDisattractiveandhasbeenafrequentlyusedreceiver.ThePDiseasytofabricateandhas low production costs; however, it usually has a non-linear response to light. FortransmissionsexceedingGbpsspeeds,atwo-dimensionalPDarrayisusedformassivelyparallelVLC.

Table9.1 ComparisonofCCDimagesensors,CMOSimagesensors,andPD

CCD CMOS PD

Speed moderatetofast fast veryfast

Sensitivity high low high

Noise low moderate low

Systemcomplexity high low verylow

Sensorcomplexity high low verylow

Chipoutput analogvoltage digitalbits analogvoltage

Energyconsumption moderate low low

Spatialseparation yes yes no

Productintegration low high no

Productioncost moderate verylow verylow

AdvantagesofCCDsaretheresultinghigh-qualityimaging,i.e.,CCDshaveoptimizedphotodetectors, very low noise, and no fixed pattern noise; however, CCDs cannotintegratewithother analogordigital circuits, includingclockgenerators, controllers, oranalog-to-digitalconverters.Highpowerconsumptionandlimitedframerates,especiallyforlargesensors,arealsodisadvantagesofCCDs.

The speed of the readout processes for CMOS image sensors could be increased byselecting a specific set of pixels, while discarding others. Sampling a small number ofrelevantpixelsdramaticallyincreasesreadoutspeeds.Usingthistechniqueorotherrelatedtechniques,frameratesof10000fpsormorearepossible[5].

ThepotentialforproductintegrationisanotheradvantageofCMOSimagesensorsoverCCDs.Thispotentialmakespossibletherealizationofacompletesingle-chipcamerawithtiminglogic,exposurecontrol,andanalog-to-digitalconversion.Furthermore,integrationofcommunicationpixelswithconventionalimagepixelshasalsobecomepossible[4].Oflate, most chips found in computers and other electric goods are manufactured usingCMOS technology. Chip manufacturing plants cost millions of dollars to set up, butprovided that they produce chips in sufficient quantities, the resultant price per chip isverylow,particularlywhencomparedwithothertechnologies.Thus,drasticreductionsincostsarepossibleevenforhigh-frame-rateCMOSimagesensors,as longas themarketdemandforsuchchipsishigh.

9.3 ImagesensorasaVLCreceiver

ACMOS image sensor canbeused as aVLC receiver [1,4].A particular advantage ofusingaCMOSimagesensor,duetothemassivenumberofavailablepixels,isitsabilityto spatially separate sources.Here sources include both noise sources, such as theSun,streetlights,andotherambientlights,andtransmissionsources,i.e.,LEDs.

TheabilitytospatiallyseparatesourcesalsoprovidesanadditionalfeaturetoVLC,i.e.,theabilitytoreceiveandprocessmultipletransmittingsources.AsshowninFig.9.3,thedata transmitted from two different LED transmitters can be captured simultaneously.Further, if a source is composed of multiple LEDs, parallel data transmission can beaccomplishedbyindependentlymodulatingeachLED.

Figure9.3 AdvantagesoftheimagesensorbasedVLC.

TheoutputoftheCMOSimagesensorformsadigitalelectronicrepresentationofthescene,whichprovidesuniqueopportunitiesthatcannotberealizedbyasingle-elementPDor radio-wave technology. For example, it offers the ability to simultaneously utilize amultitudeof imageandvideoprocessing techniques, suchaspositionestimation,objectdetection,andmovingtargetdetection,viathedatareceptioncapabilityofaVLClink.

Asanotherexample,considerasituationinwhichareceiverisequippedwithaCMOSimagesensor.Tobeginwith,aVLCsignaliscapturedalongwithitsspatialposition(X,Y)ortheactualrowandcolumnpositionofapixel.ThismeansthattheVLCsignalcanberepresentednotonlybya timedomainsignal,butalsoby thedirectionof the incomingvector from the transmitter to the receiver.Consequently, requisite position data,whichcan be obtained via GPS or another position estimation system, will be availableinherentlythroughVLCtransmission.

Inthefollowingsubsections,wepresentimportanttechnicalfeaturesassociatedwithanimagesensorusedasaVLCreceiver.

9.3.1 TemporalsamplingTheNyquist–Shannonsamplingtheorem,sometimesreferredtoastheShannon–Someyasamplingtheorem,isafundamentaltheoremthatappliestotime-dependentsignals.Thistheorem states that if anLED transmitter generates a discrete-time signalwith the timeintervalTs,i.e.,thedatarateRs=1/Tssample-per-second(bps),thentheframerateofan

imagesensormustbegreaterthanorequalto2Rsfps.Iftheframerateislessthan2Rsfps,then temporal aliasing occurs and the signals become indistinguishable, resulting in asituationinwhichtheoriginalsignalisdifficulttoreconstruct.

Figure 9.4 shows an example of an LED transmitter with on-off-keying (OOK)modulation and an image sensor receiver. The transmitter consists of multiple LEDsgeneratingthesamesignal,whichcanbeusedinthiscase.

Figure9.4 AsingleLEDtransmitterandimagesensorreceiver.

SupposethattheframerateofaCMOSimagesensoris30fps, thenthetransmissionrate,orequivalentlytheblinkingrate,ofanLEDhastobe≤15Hztomeetthesamplingtheorem.Onlyatblinkingrates≤15Hz,canthereceiverrecognizetheblinking.Iftheframerateis1000fps,thentheLEDblinkingratebecomes500Hz.Blinkingat

such a fast rate is invisible to the human eye, thus we recognize LED sources ascontinuous illuminating devices that do not blink. The use of high-frame-rate imagesensorsismandatoryforVLCs.

9.3.2 SpatialsamplingTheabilitytospatiallyseparatesourcesalsoprovidesanadditionalfeaturetoVLCs,i.e.,theabilitytoreceiveandprocessmultipletransmittingsources.AsshowninFig.9.3,datatransmittedfrommultipleLEDs(data1anddata2)canbesimultaneouslycaptured;thisallows parallel data transmission by separately modulating each LED of a sourcecomprisingmultipleLEDs[4].

Figure 9.5 shows an example of a 3 × 3 LED array transmitter modulated in OOKformatwith each LED transmitting a different signal. This is one of the advantages ofusing an image sensor as a VLC reception device: parallel data transmission can beachievedbyseparatelymodulatingeachLED.

Figure9.5 A3×3LEDarraytransmitterandanimagesensorreceiver.

Thesamplingtheoremdescribedaboveisforone-dimensionaldiscrete-timesignals,butitcanbeeasilyextendedtoimagesusingrealnumberstorepresenttherelativeintensitiesofpixels(pictureelements).

Similarly to one-dimensional discrete-time signals, images also may suffer fromaliasing if the sampling resolution or pixel density is inadequate. For example, if thedistancebetweenLEDs is too small, aliasingcanoccur. Ingeneral, fourpixels foreachLED (i.e., two each for the row and the column) are required. The resultant aliasingappearsasaMoirépatternor loss inhigherspatial frequencycomponentsof the image.Thesolutionforbettersamplinginthespatialdomaininthiscasewouldbetomovethereceiverclosertothetransmitter,useahigher-resolutionimagesensor,oruseatelescopiclenstoenlargethetransmitter’simage.InSubsection9.4.5below,wediscusstheeffectofcommunicationdistanceandspatialsampling.

9.3.3 MaximalachievabledatarateInthissubsection,weevaluatethemaximalachievabledatarateϑofimagesensorbasedVLC.Tobeginwith,consideranimagesensorwithN×Mpixels,eachpixelproducingG-levelgrayscalesignals.LetFrbetheframerate.Thenthemaximalachievabledatarateis

obtainedas ,withthe1/8factorastheratereductionbythesamplingofthree-dimensionalsignals.

Forexample,ifweconsideraQVGA(320×240pixels)256grayscaleimagesensorwithaframerateof1000fps, thenthemaximalachievabledataratereaches76.8Mbpsusinga160×120LEDarraytransmitter.ThelatestPhotoronFASTCAMSA-X2captures12-bit gray scale images at a frame rate of 720000 fpswith an image size of 256×8pixels [5]. In this case, the data rate reaches 2.2 Gbps using a 128 × 4 LED arraytransmitter.

Note that the data rate obtained above is for the image sensor designed for digitalimaginginwhichpixelsaredenselyorganized.Ifeachpixelisplacedwithagaptoavoidspatialaliasingand thereceptionspeed ismuchfaster than the transmissiondata rate toavoidtemporalaliasing(asisthecaseinSection9.5belowwithmassivelyparallelvisiblelight transmission), then we may drop the 1/8 factor and the rate is governed by thetransmitter.Forexample,foranN×MLEDarraytransmitterwitheachLEDproducingaG-levelluminancesignalwithdatarateRb,theratebecomesϑ=NMlog2GRb.

9.4 DesignofanimagesensorbasedVLCsystem

9.4.1 TransmitterIn thedesignofan imagesensorbasedVLCsystem,wemayuseasingleoramultipleLEDtransmitter(i.e.,anLEDarray,asshowninFig.9.6).Here,wefocuson the latter,i.e.,theLEDarraytransmitter.

Figure9.6 BasicstructureofamultipleLEDtransmitter(i.e.,anLEDarray).

Consider an LED array transmitter that consists ofM × N LEDs. The transmittergeneratesanon-negativebinarypulsewithwidthTb,whereTbisthebitduration.Thedataratewillbe .Hence,sinceeachLEDtransmitsadifferentbit,thebitrateofthetransmitter becomesMNRb. As shown in the figure, the luminance of the LED can bemodulated by changing the width of Tb or via the pulse width modulation (PWM)technique[6].Forexample, thePWMtechniqueshowninFig.9.7producesa five-levelsetofluminancevalues,i.e.,0,1/4,1/2,3/4,and1(maximumluminance).Omittingthe0luminance signal to avoid an entirely dark period, a four-level signal is produced.Accordingly,thePWMproducesG-levelgrayscalesignalsandthedataratebecomesMNlog2GRb.Finally,thePWMsignalisconvertedintoatwo-dimensionalsignal,andeachLEDtransmitsdatainparallelbyindividuallymodulatingitsluminance.Inotherwords,wetransmitdataasatwo-dimensionalLEDpattern.

Figure9.7 Pulsewidthmodulation(PWM)toassigndifferentLEDluminancevalues.

NotethatthepacketformatisshownbelowthetransmitterinFig.9.6.Auniquecode,

suchastheBakercodesequence,maybeusedfortemporalsynchronization.

9.4.2 ReceiverAsshowninFig.9.8,thereceiverconsistsofanimagesensor,animageprocessingunit,and a data detection unit. The image size of the image sensormust be larger than thetransmitterLEDarray soas to avoid spatial aliasing.Further, the frame ratemustbeatleasttwiceasfastasthedatarateorblinkingrateofeachLED.

Figure9.8 ReceiverforimagesensorbasedVLC.

Thetransmittedsignalarrivesattheimagesensorreceiverthroughtheopticalchannel.Aftertheheaderimage-processingunit,thesignalisfedtothedataimage-processingunit,which consists of an LED array tracking unit, an LED position estimation unit, and aluminanceextractionunit.Simpletemplatematchingcanbeusedforthetracking.

AfterLEDarraytracking,LEDpositionestimationisperformed,locatingthepositionofeachoftheLEDsbasedonthepixelrowandcolumn,aswellastheluminancevalues.ThisisonlypossibleifaccurateLEDarraydetectionisachieved,becauseboththeshapeof theLED array and theLED array tracking are necessary for the output. Finally, theoutputof thedataimage-processingunit isfedtoadecoder,whichoutputs theretrieveddata.

9.4.3 ChannelItisgenerallyacceptedthatVLClinksdependontheexistenceofanuninterruptedline-of-sight (LOS) path between the transmitter and receiver. In contrast, radio links aretypicallysusceptibletolargefluctuationsinreceivedsignalamplitudesandphases.Unlikeradio-waves,VLCdoes not suffer frommultipath fading,which significantly simplifiesthe design of VLC links. Because VLC signals travel in a straight line between thetransmitter and receiver, they canbe blocked easily byvehicles,walls, or other opaquebarriers.Thissignalconfinementmakesiteasytolimittransmissionstoreceiversincloseproximity. Thus, VLC networks can potentially achieve remarkably high aggregatecapacityandasimplifieddesign,becausetransmissionsoutsidethecommunicationrangeneednotbecoordinated.Inotherwords,itisnotnecessarytoconsidersourcesoutsidethevisualrange.

However, VLC has several potential drawbacks. First, since visible light cannot

penetrate walls or buildings, VLC coverage is restricted to small areas and someapplications will therefore require the installation of access points that must beinterconnectedviaawiredbackbone.Further,inadditiontooutrightphysicalblocks,thickfogorsmokecanblurvisiblelightlinksanddecreasesystemperformance.

In short-rangeVLCapplications, the signal-to-noise ratio (SNR)ofadirect-detectionreceiverisproportionaltothesquareofthereceivedopticalpower.Therefore,VLClinkscantolerateonlyacomparativelylimitedamountofsignalpathloss.

In the next subsection, we present unique characteristics of the image sensor basedVLCchannelfromtheviewpointoftheimagesensoritself.

9.4.4 Field-of-view(FOV)Thefield-of-view(FOV)isanimportantparameterfordefiningtheimage-capturingrangeof the receiver. There are three FOV types, namely horizontal FOV, vertical FOV, anddiagonalFOV.WecancalculatetheFOVfromthefocallengthofthelensandtheimagesensor size.As an example, Fig. 9.9 shows the calculation of the diagonal FOV usingfocallengthandsensorsize.ThediagonalFOVisthewidestofthethreeFOVs;thus,wedefinethediagonalFOVasthemaximumFOV(FOVmax)[7].

Figure9.9 Field-of-view(FOV).

Theimage-capturingrangeofthechanneldiffersdependingonthewidthoftheFOV.FornarrowFOVs,thereceiverisequippedwithatelescopiclensthatcapturesamagnifiedimageofatargetsource,usuallylocatedfaraway.Inotherwords,thereceivercapturesthetargetasifthetransmitterisplacedinfrontofthereceiver.Thus,thereceivercaneasilyrecognize each LED having a transmitter. In addition, the receiver is less physicallyaffected by ambient light noise, because light other than the target transmitter hasdifficulty penetrating the lens in the narrow FOV. At first glance, these characteristicsseem favorable; however, the number of transmitters that the receiver simultaneouslyrecognizes decreases, because the receiver limits the number of transmitters used forcommunication.

InthecaseofawideFOV,alargeviewcanbecaptured.UnlikeanarrowFOV,awideFOVallowsthereceivertosimultaneouslycapturenumeroustransmitterswithoutlimiting

thenumberoftargetsources.Ifthesetargettransmitterssenddatausingvisiblelight, thereceiver can obtain these data.More specifically, if the receiver can distinguish visiblelight from transmitters within the FOV, all data from transmitters can be received;however, the size of each target on the image is different and depends on thecommunication distance. The size of the target on the image decreases as the distancebetween transmitter and receiver increases. In addition, a receiverwith awide FOV issusceptible tomore ambient light noise as compared to a receiverwith a narrowFOV,because the probability that the receiver recognizes light sources other than VLCtransmittersishigh.

9.4.5 EffectofcommunicationdistanceandspatialfrequencyAs described in Subsection 9.3.2 above, the size of a target (i.e., the transmitter LEDarray) for a captured imagediffers according to the communicationdistance.Generally,thissizeincreasesascommunicationdistanceof thetargetdecreases.Figure9.10showstheLEDarrayusedtoevaluatetherelationshipbetweencommunicationdistanceandthenumberofpixelsoftheLEDarray.LettheactualdistancebetweenneighboringLEDsbeda.We use anLED arraywithda = 20mm and an image sensor receiverwith a focallengthof35mmandaresolutionof128×128.Here,wedefinepixeldistancedpasthedistancebetweentwoneighboringLEDsontheimagesensor.FromFig.9.10,thenumbersof pixels of the LED array are observed to differ depending on the communicationdistance.WefocusonthenumberofLEDsonthearrayandthenumberofpixelsintheimage.Asdiscussedabove,todistinguisheachLEDonthearrayforanimage,thenumberofpixelsshouldbetwicethenumberofLEDs[8].

Figure9.10 NumberofpixelsoftheLEDarrayinrelationtocommunicationdistance.

TheLEDarrayhas256LEDsarrangedina16×16squarematrix.Inthislayout,theallocation of 32 × 32 pixels is necessary to distinguish each LED on the array. Thisrequirement is satisfiedwhen thecommunicationdistance is20m,as illustrated inFig.9.10.Next,weneedtodeterminethenumberofLEDsrequiredtodistinguisheachLEDfor various communication distances. When the communication distance is 40 m, thenumberofpixels is18×18. In this situation, if the transmitterhasLEDsarranged ina9×9orsmallersquarematrix,thenthereceivercandistinguisheachLEDontheimage.Thisrequirementisequivalenttosendingdatausing8×8LEDsoreveryotherLEDofthe16×16LEDarray.Inthesamemanner,whenthecommunicationdistanceis70m,therequirementis16LEDsarrangedina4×4squarematrixorless,whichisequivalenttosendingdatausing4×4LEDs,oreveryfourthLEDofa16×16LEDarray.

Next,weagain focuson theeffectof communicationdistance from theviewpointofspatial frequency.Spatial frequency refers to thenumberofpairsofbars imagedwithinthe imagesensor.Thespatial frequencymaybe foundbyapplying the two-dimensionalFourier transform. Higher resolutions result in the improved detection of high-spatial-frequencycomponentsoffine-texturedobjects.

As shown in Fig.9.10, the 20mLED array image (32 × 32 pixels) hasmore high-frequencycomponentsthanthe70mLEDarrayimage(9×9pixels).ThereductioninthenumberofpixelsoftheLEDarrayresultsinthereductionofhigh-frequencycomponents.Inotherwords,thelongerthedistance,themorehigh-frequencycomponentsarelost.Thisreflects the fact that the channel is a low-pass channel with a cutoff frequency thatdecreasesasthedistancebetweentransmitterandreceiverincreases[9].

9.5 Massivelyparallelvisiblelighttransmission

9.5.1 ConceptThe speed of visible light communication (VLC) is limited by the frequency responsecharacteristics of a visible light LED and a visible light optical sensor. A visible lightopticalsensorsuchasanSiPINPDhasfrequencyresponsecharacteristicsover10MHz;however,atypicalwhiteLED,whichconsistsofablueLEDandyellowphosphor,hasa3dB cutoff frequency smaller than 10MHz due to the slow response of phosphor [10].Therefore, achieving high-speed data transmission is difficult by using typical whiteLEDs.

One of the methods for improving performance is to use multiple transmitters andreceivers in order to achieve parallel data transmission. A basic concept of massivelyparalleldatatransmissionisshowninFig.9.11.

Figure9.11 Massivelyparalleldatatransmissionofvisiblelightcommunication.

Inmassivelyparalleldatatransmissions,thetransmitterconsistsofanarrayofmultipleLEDs,andthereceiverconsistsofmultiplePDs.Multiplereceiversforfree-spaceopticalcommunicationhavebeenresearchedpreviously. In [11], theauthorsproposeda fly-eyereceiver includingmultipleball lensesandPDsusedtoreceivemultipledata inparallel.Thissystemwasfairlycomplicatedbecauseitwasrequiredtolookindifferentdirectionssimultaneously. In [12], theauthorsproposedan imagingdiversity receiveremployinga37-pixel imaging receiver. This system was designed to be used not only for LOScommunication, but also non-LOS communication by selecting and combining signalsfrommultiple PDs; however, it did not consider parallel data reception. Conversely, inboth[13] and [14], the authors proposedmassively parallel data transmission via a PDarray consisting of 256 PDs.We describe the details of such an approach in the nextsection.

Conventional image sensors have a two-dimensional array of PDs; however, theyusually have slow frame rates, in the range 10–30 fps, due to their mechanism ofsequentiallyreadingdatafromPDs,i.e.,pixels.Thisframerateisnotfastenoughforthemassivelyparallel data transmission to achieve thedesireddata rates that exceedGbps;however, there is a special type of two-dimensional PD array having each signal fromindividualPDsconnectedtoapackagepinsuchthatallPDreceptiondatacanbereadoutin parallel. One example of such a two-dimensional PD arraywith parallel read-out isshowninthemiddlepictureofFig.9.15.

9.5.2 SystemarchitectureFigure 9.12 shows the system architecture of a massively parallel data transmissionsystem.AdatastreamisdividedintomultiplepacketsandtransmittedindependentlyfromeachLEDinthetransmitterLEDarray.Thereceiver,whichiscomposedofalensandatwo-dimensional image sensor, demodulates all the signals in each pixel in parallel.Haruyama’sgroupatKeioUniversityinJapandevelopedahigh-speedmassivelyparalleldata transmission system using a white LED array and a two-dimensional PD array[13,14].Thedetailsofthissystemaredescribedinthissubsection.

Figure9.12 Systemarchitectureofamassivelyparalleldatatransmissionsystem.

Thesystemarchitectureconsistsofaparallelvisiblelighttransmitter,aparallelvisiblelight receiver, and an uplink infrared connection. The link between transmitter andreceivermustfirstbeestablished,asisdescribedinthenextsubsection.Afterthelinkisestablished,data tobe transmittedareconvertedtoMparalleldatapackets.Theaddressthat indicates the position in the two-dimensional LED array is attached to each datapacket, along with an error detecting code of cyclic redundancy check (CRC). Theseparallel packets are allocated toJLEDsand transmittedby theseLEDs.At theparallelvisible light receiver,eachPDof thePDarray independentlydetects the incoming lightsignal, and then performs error detection to check if the received bits are correctlyrecoveredfromtheopticalsignalforeachPD.Ifsomehaveerrors,anAutomaticRepeatreQuest(ARQ)isperformedbysendingtherequestviatheuplinkinfraredconnection.Ifallparalleldataarecorrectlyreceived,theyareconvertedtoserialdata.

TheconfigurationofthesystemdevelopedbyHaruyama’sgroupatKeioUniversityisshowninFig.9.13.

Figure9.13 Systemconfigurationofamassivelyparalleldatatransmissionsystem.

Thetwo-dimensionalLEDarray,composedof24×24whiteLEDs,canbeusedasatransmitter.Thevisible light receiverconsistsofa lensanda two-dimensionalPDarraythat consists of 16 × 16 PIN PDs. The uplink connection consists of an infrared LEDattached to the two-dimensional receiver and an infrared PD attached to the two-dimensionaltransmitter.

9.5.3 LinkestablishmentInparallelwirelessVLCsystems,multipletransmittedsignalsarespatiallyseparatedbyareceiver lens.The imageof theLEDlighting transmitter isprojectedonto thePDarray.Allsignals transmittedfromdifferentLEDsare independentlyreceivedbyeachpixeloftheimagesensor.Beforedatatransmission,wemustfirstknowtherelationshipbetweenthe transmitted signal and the receiving PD. Figure 9.14 shows the link establishmentsequence.

Figure9.14 Linkestablishmentsequence.

The receiver first sets up the zoom of a lens by finding an appropriate pitch of theprojectedLEDs.Detailsof thezoommethodcanbe found in [14].Even if the zoom iscorrectlysetup,caseswithmultipleLEDsprojectedontothesamePDwillbepresent,asshown in Fig. 9.14, resulting in interference if different signals are sent frommultipleLEDs to the samePD.Therefore,LEDshave tobe correctly assigned to correspondingPDstoavoidsuchinterference.Thisisachievedbysendingappropriateinformationfromreceiver to transmitter using an uplink connection. The details of the LED and PDassignmentcanbefoundin[13].ThedatatransmissionbeginsonlyafterthecompletionofLEDandPDassignment.

9.5.4 PrototypeofamassivelyparalleldatatransmissionsystemOuractualprototypeisshowninFig.9.15.The24×24visiblelightLEDarraytransmitterisable tosendindividualdatafromeachLED,eachofwhichiscontrolledbythefield-programmablegatearray(FPGA)atthebackoftheLEDboard.Atthereceiver,thelightfromthetransmittergoesthroughazoomlenswithafocallengthrangingfrom28mmto300 mm; the focused image is projected onto a 16 × 16 PD array device. The two-dimensional PD array chip was made by Hamamatsu Photonics. The rate of thetransmittedsignalfromoneLEDis5Mbps;thus,whenmaximumparallelismisattained,thetheoreticalmaximumdatarateis16×16×5Mbps=1280Mbps.

Figure9.15 Prototypeofamassivelyparalleldatatransmissionsystem.

9.6 Accuratesensorposeestimation

9.6.1 OverviewIn image sensorbasedVLC, it is possible to compute a sensor (i.e., camera) posewithcomputer vision techniques [14–16]. This is one of the major advantages as comparedwithothercommunicationtechniques.Below,weintroducethebasisofcomputervisionforpose estimation, aswell as apose estimationmethodcombinedwithboth computervisionandVLCtechniques.

9.6.2 SingleviewgeometryComputervisionisafieldinvolvinganalysesofthegeometryoftherealworldfromtheimages captured with a camera. In general, research issues are classified into theestimationofthecameraposeandthereconstructionofthethree-dimensionalshapeofanobject (i.e., three-dimensional modeling) from an image or image sequence. Relatedresearch fields of computer vision include image processing, pattern recognition, andmachinelearningforscenerecognitionandunderstanding.

In image sensor based VLC, an image sensor can receive data simultaneouslytransmittedfrommultiplelights,becauseeachpixelonthesensoristreatedasareceiver.Further,thepositionsofthelightsareacquiredasfeaturepointsontheimagesensorandcan be used for camera pose estimation. Since this is a traditional research issue,substantialresearchisreadilyavailableintheliteratureoncameraposeestimationusingpoints [14]. Below, we introduce the basis of single-camera geometry for camera poseestimation.Notethattheterm“pose”hererepresentsthepositionandorientationrelative

tosomecoordinatesysteminthefieldofcomputervision.

As shown in Fig. 9.16, camera geometry is described via three coordinate systems,namely the three-dimensional world coordinate system, the two-dimensional imagecoordinatesystem,andthethree-dimensionalcameracoordinatesystem.Acameraposeisnormallydefinedasthepositionandorientationofthecameracoordinatesystemrelativeto the world coordinate system. The mathematical representation of the position andorientation is equivalent to the parameters of geometric transformation from the worldcoordinatesystemtothecameracoordinatesystemas

where is a homogeneous camera coordinate system, isahomogeneousworldcoordinatesystem,R isa3×3 rotation

matrix(orientation),andtisa3×1translationvector(position).Therefore,acameraposeisequivalentto[R|t].

Figure9.16 Singleviewgeometry.

The camera coordinate system is defined as the systemwith its origin located at thecameracenterand thedirectionofZcperpendicular to the imageplanefromthecameracenter.TheintersectionoftheimageplaneandZcaxisiscalledprincipalpointp=(px,py).Inthepinholecameramodel,athree-dimensionalpointXc=(Xc,Yc,Zc)T in thecameracoordinate system is projected onto a two-dimensional point x = (x, y)T in the imagecoordinatesystemas

wherefisthefocallengthofthelens.ByestablishingcameracalibrationmatrixAas

the projection of a three-dimensional point in theworld coordinate system onto a two-

dimensionalpointintheimagecoordinatesystemisdescribedas

where isahomogeneousimagecoordinate.Thisequationcanbesimplifiedas

P=A[R|t],

wherePisa3×4perspectiveprojectionmatrixthatalsorepresentsacamerapose.

Toestimateacameraposebysolvingtheaboveequations,obtainingmultiplesetsofand ismandatory.Forexample,Pislinearlycomputedfromsixsets,becausethereare12 unknown parameters in P and two equations are determined from one set. IfA isknownbyusingacameracalibration technique [17], it ispossible tocomputeacamerapose with up to four pairs of solutions [18]. Because many solutions under differentconditions have been proposed in the literature, solutions are not limited to the onesdescribedabove.

9.6.3 PoseestimationusinglightsAsexplainedabove,acquiringmultiplesetsofaworldcoordinateanditsprojectedimagecoordinate for camera pose estimation is necessary. In image sensor based visible lightcommunication,suchsetsareacquiredusinglights.

Figure9.17showsanoverviewofposeestimationusinglights.Lightsarefirstplacedina target scene and their three-dimensional world coordinates are measured using anelectronicdistancemeter, such as a total station.Tocompute the image coordinateof alight and receive its transmitted data, a camera captures the lights at a fixed position,becausealightshouldbecapturedatthesamepositionthroughanimagesequence.

Figure9.17 Poseestimationusinglights.

When transmitted data include the world coordinate of the light, we acquire the setcontainingtheworldcoordinateofalightanditsprojectedimagecoordinate;however,the

quantityoftransmitteddatamaynotbeenough.Insuchacase,thetransmitteddatacanbejustanidentificationnumber(ID),andthelistofidentificationnumbersoflightsandtheirworldcoordinatesshouldbestored,asshowninTable9.2.Ifthenumberofacquireddatasetssatisfiestheconditionofposeestimation,acameraposecanbecomputed.Notethattheaccuracyofanestimatedcameraposedependsontheaccuracyofthemeasurementsofthree-dimensionalworldcoordinates.Inthefollowingsubsection,weexplainanefficientmethodforcomputingtheimagecoordinateofablinkinglight.

Table9.2 IDsandworldcoordinates.

ID Worldcoordinate

1 (X1,Y1,Z1)

2 (X2,Y2,Z2)

⋮ ⋮

N (XN,YN,ZN)

9.6.4 LightextractionToextractlightsinimages,oneapproachistouseapredefinedbrightnessthresholdsuchthatapixelisjudgedaslightifthemeasuredbrightnessismorethanthegiventhreshold;however,thisapproachdoesnotworkwell,becausebrightnessissubstantiallyaffectedbymanyelements,suchasimagesensors,environmentallights,thebrightnessofalight,andthedistancebetweenthecameraandlight.Tostablyextractlights,lightextractionusingtherulesofblinkingpatternswasproposedin[17],andaflowchartisshowninFig.9.18(please also refer to the figures in [17] formore illustrativedetails).Below,weexplaineachstepofthelightextractionprocess.

Figure9.18 Flowchartoflightextractionin[17,19].

9.6.4.1 ComputationofathresholdFirst,athresholdforconvertingthebrightnessvalueintobinaryisadaptivelycomputedateachpixel.GivenanimagesequencethatincludesNimages,eachpixelihasNbrightnessvalues;maximumbrightnessLi andminimumbrightnessSi are selected. Then, the newadjustedthresholdateachpixelTiiscomputedas

9.6.4.2 ExtractionoflightcandidatesAftercomputingabinarysequenceateachpixelusingthe threshold,oneapproachis touse an error-checking scheme embedded in the transmitted data to check all pixels;however, its computational cost may be huge; thus, decreasing the number of pixelsbecomes important for error checking.Therefore, to avoid thehugecomputational cost,light candidate pixels are extracted by checkingwhether a binary sequence follows therulesoftheblinkingpattern.

In[17],theblinkingpatternisdesignedsuchthat1bitisrepresentedby4samples,i.e.,0011is0and1100is1.Thismeansthat010and101donotappearintheblinkingpatternof threesamples. Ifapixel includessuchapattern, itcanbe removed.Bychecking thepatternofabinarysequenceofseveral images, thenumberofpixels thatdonotcapture

lightisdrasticallyreduced.

9.6.4.3 ErrorcheckingAfterextractingthelightcandidatepixels,abinarysequenceofallimagesiscomputedforthesepixels.Thenabinarysequenceistestedwithanerror-checkingschemeembeddedinthetransmitteddatatoincreasethereliabilityoflightextraction.Next,theadjacentpixelsthatpassedtheerror-checkingphaseandhadthesamebinarysequenceareconnectedtomake a light area. Finally, by computing the center of each light area, the imagecoordinatesofalightandthetransmitteddataarecomputed.

9.7 Applicationsofimagesensorbasedcommunication

9.7.1 TrafficsignalcommunicationIn this subsection, we introduce a vehicle-to-infrastructure visible light communication(V2I-VLC)systemusinganLEDarraytransmitter,whichisassumedtobeanLEDtrafficlight, and an in-vehicle receiver equipped with a high-frame-rate (HFR) CMOS imagesensorcameraorhigh-speedcamera(HSC)[4].

During our experiments, we placed an LED array horizontally on the ground andmountedthehigh-speedcameraonthedashboardofthevehicle.Thevehiclewasdrivendirectly toward the LED array at 30 km/h, as shown in Fig. 9.19. The communicationdistanceinthesefieldtrialsrangedfrom110to30m.

Figure9.19 Experimentalequipment.

Thetransmitterconsistedof1024LEDsarrangedina32×32squarematrix.TheLEDspacing was 15 mm, and its half value angle was 26 degrees. To compensate for thevibrationsof thecar,werepresentedonedatabitusingfourLEDs(a2×2LEDarray).EachLEDblinksat500Hz,whereasthePWMwasprocessedat4kHz.WeusedR=1/2turbocode forerrorcorrectionand invertedLEDpatterns for tracking.Accordingly, theoveralldataratewas32kbit/s(=500bit/s×256×1/2×1/2).TheinputdatawereaudiodataandwereassumedtobetransmittingsafetyinformationtransmittedfromLEDtraffic

lights.

For the receiver, we used an in-vehicle HSC (i.e., a Photoron FASTCAM 1024PCI100k)withaframerateof1000fpsandaresolutionof512×1024pixelsconnectedtoapersonal computer (PC). The focal length of the lenswas 35mm. In general, the lightsensitivity of high-speed image sensors was set high to provide rapid exposure times,which also means we could set a relatively small lens aperture. For example, the ISOsensitivityoftheHSCwassetat10000,andthelensaperturewassetto11.Inaddition,sinceautofocusingisdifficultwhenavehicleismoving,thefocuswassettoinfinity.

Theheaderimage-processing,thedataimage-processing,andthedecodingtaskswereperformedbythePC.Wealsorecordedanddisplayedagrayscalevideo,obtainedusingthe HSC as a drive recorder, which simultaneously recorded the view in front of thevehicleandthedata transmittedfromtheLEDarray.WeconfirmedtherobustdetectionandtrackingoftheLEDarraywithrespecttocameravibration,alongwithalackoferrorinLEDarraydetectionandtracking.Next,weconfirmedaclearaudiosignal(32kbit/s)receptionfordistancesofupto45mwitherror-freeperformance.

Wealsoconductedanexperimentonthesimultaneoustransmissionoftextinformation.Inthiscase,thedataratewas2kbit/sanderror-freeperformancewasachievedfrom110–20m,whichisdeemedtobeasuitablerangeforintersectionsafetyapplications.

9.7.2 PositionmeasurementsforcivilengineeringAsintroducedinSection9.6above,acameraposecanbecomputedintheframeworkofsingleviewgeometry.Ifmultiplecamerascanbeused,thethree-dimensionalpositionofan object in the real world can be computed [20]. In this subsection, we introducetriangulationwithimagesensorbasedVLCanditsapplicationtocivilengineering[14].

9.7.2.1 BasisoftriangulationTriangulationisamethodforthree-dimensionalpositionmeasurementofanobjectintherealworldwith two known cameras.Because triangulation is closely related to cameraposeestimation,wefirstbrieflyrevisitperspectiveprojectionintroducedinSection9.6.

A three-dimensional point in the world coordinate system is projected onto a two-dimensionalpoint in the imagecoordinate system.When three-dimensionalpoint isprojectedontotwocameras,thisismathematicallydescribedas

wherePiisaperspectiveprojectionmatrixofeachcameraand isahomogeneousimagecoordinate computed by projecting onto each image plane withPi; and areconsideredtocorrespondbetweentwoimages.UseoftwoknowncamerasmeansthatP1andP2areknown.ThisisalsographicallyexplainedinFig.9.20.Intriangulation,thegoalistocompute byfinding and .

Figure9.20 Twoviewgeometry.

9.7.2.2 TriangulationwithimagesensorbasedVLCIfweputablinkinglightat ,itiseasytofind and withimagesensorbasedVLC.Ineachcamera,theimagecoordinatesoflightsandthetransmitteddataarefirstcomputedasintroducedinSection9.6above.Then and canbeacquiredbyselectingthepixelsthatreceivethesamedataineachcamera.

9.7.2.3 ApplicationtobridgeshapemonitoringImage sensor basedVLCworks under different types of illumination fromdaybreak tomidnight,becausealightcanactivelytransmitdatabyblinking.Thisfeatureisveryusefulforbridgeshapemonitoring.Ingeneral,theshapeofabridgeisdeformedwhenthefabricofthebridgeisdilatedandshrinksaccordingtotheambienttemperature.Whenabridgeisbuilt,suchdeformationshouldbemonitoredfortheearlydetectionofproblems.

Figure 9.21 shows an overview of bridge shape monitoring using an image sensorbasedVLCapproach.Onthebridge, twodifferent typesof lightsareplaced, i.e.,pointsformeasuringthedeformationandpointsforcameraposeestimation.Thelatterpointsaretypicallycalled referencepointsandshouldbeplacedatpointswheredeformationdoesnotoccur.

Figure9.21 Bridgeshapemeasurement.

Then, themonitoringprocedureisasfollows.Fromthereferencepoints, theposesofcamera1andcamera2arefirstcomputed.Then,ineachcamera,theimagecoordinatesofthelightsandthetransmitteddataarecomputed.Byselectingthepixelsthatreceivethesamedata, thecorrespondingpixelsof the light in the two imagesarecomputed.Thesepixelsaretriangulatedtocomputethethree-dimensionalpositionofthelight.

9.8 SummaryInthischapter,wehavecoveredvisiblelightcommunicationusingimagesensors,startingwith its fundamental principles, followedby application examples ofmassively parallelcommunication, accurate sensor pose estimation, traffic signal communication, andpositionmeasurementforcivilengineering.Animagesensorisabletospatiallyseparatevisiblelightsources,receiveanddemodulateopticalsignalsfromvisiblelighttransmittersat pixel positionswhere optical light is projected, and detect accurate arrival angles ofincominglight.Byusingtheuniquecharacteristicsoftheimagesensors,theapplicationsdescribed in this chapter aremade possible. For example, the use ofmultiple pixels asreceivers makes it possible to achieve massively parallel communication; further, thecapabilityofveryaccuratearrivalangledetectionmakes itpossible toperformaccuratesensor pose estimation and position measurement for civil engineering; finally, thecapabilityofimageprocessingmakesitpossibletoachievetrafficsignalcommunication.

Many of the examples described above are for special applications and are stillsomewhatexperimental.This isprimarilydueto thehighcostof imagesensorreceiversandthetechnicaldifficultyofhigh-speeddatareception.Conventionalimagesensorsareunabletoperformhigh-speeddatareception;however,high-speedcamerascouldbeusedifhighdataratesarerequired.Thesedevicesarecurrentlyprohibitivelyexpensiveandnotavailable for consumer use, butwhen usefulVLC applications using image sensors aredeveloped – and if these applications arewidely adopted – the cost will go down andmakethesetechniqueseasilyavailableinconsumerapplications.

References[1] S. Haruyama and T. Yamazato, “[Tutorial] Visible light communications,” IEEEInternationalConferenceonCommunications,Jun.2011.

[2] StuartA.Taylor,“CCDandCMOSimagingarraytechnologies:Technologyreview,”TechnicalReportEPC-1998–106,1998.

[3] DaveLitwiller,“CCDvs.CMOS:Factsandfiction,”PhotonicsSpectra,Jan.,2001.

[4] TakayaYamazato, IsamuTakai,HirakuOkada, et al., “Image sensor based visiblelightcommunicationforautomotiveapplications,”IEEECommunicationMagazine,52,(7),88–97,2014.

[5] Photoron FASTCAM SA-X2, http://www.photron.com/?cmd=product_general%product_id=39

[6] T.Nagura,T.Yamazato,M.Katayama,etal.,“Improveddecodingmethodsofvisiblelight communication system for ITSusingLEDarray andhigh-speed camera,” IEEEVehicularTechnologyConference(VTC-Spring2010),May2010.

[7] H.B.C.Wook,S.Haruyama,andM.Nakagawa,“VisiblelightcommunicationwithLEDtrafficlightsusing2-dimensionalimagesensor,”IEICETrans.onFundamentals,E89-A,(3),654–659,2006.

[8] S.Nishimoto, T.Yamazato,H.Okada, et al., “High-speed transmission of overlaycodingforroad-to-vehiclevisiblelightcommunicationusingLEDarrayandhigh-speedcamera,” IEEEWorkshoponOpticalWirelessCommunications, pp.1234–1238,Dec.2012.

[9] S. Arai, S. Mase, T. Yamazato, et al., “Feasibility study of road-to-vehiclecommunication systemusingLEDarray and high-speed camera,”Proceedings of the15thWorldCongressonITS,Nov.2008.

[10] Zabih Ghassemlooy, Wasiu Popoola, and Sujan Rajbhandari, Optical WirelessCommunications:SystemandChannelModellingwithMATLAB®,CRCPress,2012.

[11] G. Yun and M. Kavehrad, “Indoor infrared wireless communications using spotdiffusing and fly-eye receivers,” Canadian Journal of Electrical and ComputerEngineering,18,(4),151–157,1993.

[12] JosephM.Kahn,RoyYou,PouyanDjahani,etal.,“Imagingdiversityreceiversforhigh-speed infrared wireless communication,” IEEE Communications Magazine, 36,(12),88–94,1998.

[13] Satoshi Miyauchi, Toshihiko Komine, Teruyuki Ushiro, et al., “Parallel wirelessoptical communication using high speed CMOS image sensor,” InternationalSymposiumonInformationTheoryanditsApplications(ISITA),Parma,Italy,2004.

[14] Masanori Ishida,SatoshiMiyauchi,ToshihikoKomine,ShinichiroHaruyama, andMasao Nakagawa, “An architecture for high-speed parallel wireless visible lightcommunicationssystemusing2DimagesensorandLEDtransmitter,” inProceedingsof International Symposium on Wireless Personal Multimedia Communications,

pp.1523–1527,2005.

[15] T.YamazatoandS.Haruyama,“Imagesensorbasedvisiblelightcommunicationandits application to pose, position, and range estimations,” IEICE Trans. on Commun.,E97.B,(9),1759–1765,2014.

[16] RichardSzeliski,ComputerVision:AlgorithmandApplications,Springer,2011.

[17] Zhengyou Zhang, “A flexible new technique for camera calibration,” IEEETransactionsonPatternAnalysisandMachineIntelligence,1330–1334,2000.

[18] David Nister, “A minimal solution to the generalised 3-point pose problem,” inProceedings of the IEEE Computer Society Conference on Computer Vision andPatternRecognition,pp.560–567,2004.

[19] HideakiUchiyama,MasakiYoshino,HideoSaito,etal.,“Photogrammetricsystemusingvisible lightcommunication,” inProceedingsof the34thAnnualConferenceoftheIEEEIndustrialElectronicsSociety,pp.1771–1776,2008.

[20] Richard Hartley and Andrew Zisserman, Multiple View Geometry in ComputerVision,CambridgeUniversityPress,2004.

Index

Figuresandtablesaredenotedinboldtypeface.

accuracy(performancemetric)82,83

additivewhiteGaussiannoise(AWGN).

Seealsointerference

andlamparrangement56

channelcapacity141

dimmingstrategies19–21,20,21,22

algorithmcomplexity(performancemetric)83

algorithms(indoorpositioning)75–82

ALOHAprotocol84

analogdimming17–21,20,21,22

angle-of-arrivalmeasurementsystems79–81,79,89

angulation(triangulation)79–81,79

ANSIE1.11standard111,112,112

asymmetricallyclippedopticalOFDM(ACO-OFDM)145,151–152,151,153,154,154,155,160,164,165

automobile(ITSin)4,4,201–202,201

BER.Seebiterrorrate

bidirectionallinks159

biterrorrate(BER)

andLEDdimmingcontroltechnique61–64,62

DMT164,165

IPPM123–124,125–130,127,129,130

lamparrangement56–59,57

M-QAMOFDM64

OOK120–121

planetiltingtechnique49–51,50,51

PPM121–123

VPPM124–125

blinkingpatternlightextraction199–201,200

blockage(light)84–85

bridgestructuremonitoring202,203,203

camerageometry(computervision)196–198,197

capacity

channel59–60,60,140–143,142

dimming17–21,18,20,21,22

channelcapacity

arrangingVLClamps59–60,60

LOS142

chargedcoupleddeviceimagesensor(CCD)182

CIM.Seecolorintensitymodulation

civilengineeringapplications202–203,203,

clippingLEDmodulation153–154

code-division-multi-access(CDMA)84

colorintensitymodulation(CIM)

multi-coloredLED12,36,37

multi-coloredVLC33–37,34,35,36,

colorshiftkeying(CSK)12,33,36,

complementarymetaloxidesemiconductorimagesensor(CMOS)183,183

computervision(singleviewgeometry)196–198,197

constraint.Seelightingconstraint

coplanarrotationmethod99–102,101

Cramer–Raolowerboundperformancemetric(CRLB)93–94,95

CSK.Seecolorshiftkeying

cyclicprefix147–149,147,156

datarates

bandwidthcongestion71,73,73

highVLCscenarios135–139,136,137,139

IEEE802.15.7standard110

increasingaroundworld6

indoorVLCsystems41

real-timemobile169

DC-biased discrete multitonemodulation 144, 145–149,146,147,149,150, 154, 155,160,161,165

diffuselink136,137

dimensionality(performancemetric)

camerageometry(computervision)196–198,197

indoorLPS83

dimmablevisiblelightcommunication

andreliablecommunication160–161

capacity17–21,18,20,21,22

indoor60–66,61,62,65

ISCin13–16,14,15,

multi-leveltransmission21–33

dimmingcoding.Seeinversesourcecoding

directedlink136

discretemultitonemodulation(DMT).

Seealsoorthogonalfrequencydivisionmultiplexing;indoorlightpositioningsystem

advancedmulti-colorapplications161–162,161

enhancementapproaches155–158

implementation163–170,164,166,168

lackofstandards135

systemdesign158–170,162,164,166,167,168

systems143–155

diversity-combinedorthogonalfrequencydivisionmultiplex157,

DMT.Seediscretemultitonemodulation

economics

andlow-costmultimedia71–72

imagesensors204

LBS71

entropy(dimmingcapacity)17–21,18

Ethernetpowerlinecommunication(PLC)

combinedwithVLC159–160

VLCdownlink1

femto-cells73

field-of-viewreceiver(FOV)136–137,136,190

flip-orthogonalfrequencydivisionmultiplex(flip-OFDM)156,164

fly-eyereceiver84

frequency-division-multi-access(FDMA)84

geometry

computervision196–198,197

indoor42

planetiltingtechnique46

singleview196–198,197

globalpositioningsystems(GPS)

asTOAsystem75

channelmulti-access84

difficulttouseindoors70,71,72,88

GPS.Seeglobalpositioningsystems

hardwarecomplexity(performancemetric)83

high-bandwidthislands73,74

Huffmancode13–15,15

seealsoinverseHuffmancode

hybriddimming17–21,18,21,22

hybridmulti-layermodulation(MLM)158

IEC62386DALIVlCstandard111

IECTC34standard107,109,112–113,113

IEEE802.15.4standard112

IEEE802.15.7standard6,107,109,110,114,172

IM/DD.Seeintensitymodulationwithdirectdetection

imagesensors.

Seealsophotodiodes

applications201–203,201,203,

asreceiver185–187,185,186,187

definition181,182

fusiontechnology85,90–94,91,95,133

linkestablishment195

massivelyparallelvisiblelighttransmission192–196,193,194,196

poseestimation196–201,197,199,200

systemdesignusing187–192,188,189,190,191

typesof182–184,182

indoorlightpositioningsystem.

Seealsolightemittingdiodes;discretemultitonemodulation

advantagesof70–75,71,72,73,74

algorithms75–82

arrangingVLClamps51–60

backgroundof41,43,70,

basedonVLCandimagesensors90–94,91,95

channelcapacity140–143

dimmingcontroltechnique60–66,61,62,65

linktypes135–139,136,137,138

multipathreflectionsin82–84

operationinlightblockage84,85

performancecomparison81–82,83

planetiltingtechnique41–51,42

technologies88–89

InfraredDataAssociation172

infraredlaser

andopticalWiFi137,138

operationinlightblockage85

VLCsystem169

infraredlight(asuplink)1,3,4

intelligenttransportsystems(ITS)

asapplicationofVLCtechnology1,4,4

outdoorlightpositioningsystem89,90

passengerinformation138

trafficsignalcommunication201–202,201

intensitymodulationwithdirectdetection(IM/DD)

comparedtoOFDM143–144,144

hybridMLMscheme158

MC-CDMA158,

interference.

SeealsoadditivewhiteGaussiannoise

inter-symbol136,137,143

ISI143

RF5,74

internetofthings(IOT)1,4

inter-symbolinterference(ISI)

diffuselink136,137

inhightransmission143

inverseHuffmancode.

SeealsoHuffmancode

ISC16

M-aryPAM17,17

inversepulsepositionmodulation(IPPM)

andclockjitter129,130,130

andclocksynchronization125–129,127

BER123–124

statisticalparameters126

VLC116

inversesourcecoding(ISC)

andchangeleveldistribution11

dimmableVLC13–21

dimmingcapacity20,21,22

IOT.Seeinternetofthings

IPPM.Seeinversepulsepositionmodulation

ISC.Seeinversesourcecoding

ISI.Seeinter-symbolinterference

ITS.Seeintelligenttransportsystems

ITU-TSG16standard114,172

lateration(intriangulation)75–79,76,78

LBS.Seelocation-basedservices

LED.Seelightemittingdiodes

Li-Fi137,138

lightemittingdiodes(LEDs).

Seealsoon-offkeyingmodulation;indoorlightpositioningsystem

andDMT135–158

andVLC1

forindoorlightpositioningsystems73

illuminationcompatibilitywithVLC108

illuminationstandards112–113,113

intoyandentertainmentarena5–6,5

indoorlightpositioningsystem88–89,91,95

ITSapplications201–202,201

modulationbandwidth139–140

multi-colored12,34,35,35,36,

vendorcompatibility109,109

lightingconstraint10–37

line-of-sight(LOS)

channelcapacity142

non-blocked136,136,137

systemdesignfor158–159

location-basedservices(LBS)

definition70

economicsof71

LOS.Seeline-of-sight

MAC.Seemediaaccesscontrolprotocol

M-aryPAM.SeeM-arypulseamplitudemodulation

M-arypulseamplitudemodulation(M-aryPAM)

ISCin16,17,17

multi-coloredVLC33–34,35

scheme23–28,23,26,143

M-aryquadratureamplitudemodulation(M-QAM)

andLEDdimmingcontroltechnique64–66,64

OFDM64–66,64,

planetiltingtechnique49,50,51

massivelyparallelvisiblelighttransmission192–196,193,194,196

MC-CDMA.Seemulticarriercodedivisionmultipleaccess

mediaaccesscontrolprotocol(MAC)160

modulatedretroreflector1,3,4

modulation

CIM12,33–37,34,35,36,37,38,

DMT133–170,161,164,166,169

IM/DD143–144,144,158,

IPPM119,123–124,125–130,127,129,130

multi-coloredVLC33–37,34,35,36,37,38,

OOK61–64,62,116,117,120–121,133–135,134,143,144,163,164

OSM162,

PAM16–21

PAM-DMT145,152–154,154,164

PPM11,11,116,118,118,121–123

PWM11,60–66,61,143,144,188

VPPM118–119,119

Monte-Carlosimulation57

M-QAM.SeeM-aryquadratureamplitudemodulation

MSM.Seemultiplesubcarriermodulation

multicarriercodedivisionmultipleaccess(MC-CDMA)158,

multi-coloredvisiblelightcommunication

DMT/OFDMapplications161–162,161

LEDcolors140

modulationof33–37,34,35,36,

SCM159

multi-leveltransmission(dimmableVLC)

approachto21–23

performance28,28,29,30,30

scheme23–28,23,26

simulations30–33,32,

multimedianetworkservice71,73,73

multiplesubcarriermodulation(MSM)141,143

multiple-inputmultiple-outputprocessing(MIMO)161–162

noise.Seesignal-to-noiseratio;additivewhiteGaussiannoise

non-orthogonalmulti-coloredchannel37

NRZ-OOK13–16,14,15,

OFDM.Seeorthogonalfrequencydivisionmultiplexing

on-offkeyingmodulation(OOK).

Seealsolightemittingdiodes

andLEDdimmingcontroltechnique61–64,62

BER120–121

IEEE802.15.7standard110

IM/DDin144

LEDdriverefficiency163,164

timemultiplex11

usedinmostVLCsystems133,134,135

VLC116,117,143

OOK.Seeon-offkeyingmodulation

opticalspatialmodulation(OSM)162,

opticalwirelesscommunication.Seevisiblelightcommunication

orthogonalfrequencydivisionmultiplexing(OFDM).

Seealsodiscretemultitonemodulation

andLEDdimmingcontroltechnique64–66

diversity-combined157,

flip156

MSMimplementedby144,144

planetiltingtechnique49,51,52

positionmodulating157

spectrallyfactorized156

unipolar156

OSM.Seeopticalspatialmodulation

outdoorlightpositioningsystem

basicsof89,90

LEDtrafficlightandphotodiode94–104,96,97,99,101,102,103

PAM.Seepulseamplitudemodulation

performance

CIMandCSK37

computervision196–198,197

dimmableVLC28,28,29,30,30

DMT155–158

DMT/OFDM153–155,154

indoorlightpositioningsystem83,93,94,95

indoorLPS81,82

ISCtransmissionefficiency15

outdoorlightpositioningsystem102–104,102,103

planetiltingtechnique52

photodiodes(PD).

Seealsoimagesensors

outdoorLPS94–104

usedinmostVLCsystems135

physicallayer

andlow-costmultimedia71–72

DMTMAC160,166

IEEE802.15.7standard110

VLCdownlink10–37

planetiltingtechnique41–51,46

PLASACPWGstandard114

PLASAE1.4.5DMX-512AVLCstandard109,111

PLASAE1.4.5standard107,

poseestimation196–201,197,199,200

positionmodulatingOFDM157

positioningsystem.Seeoutdoorlightpositioningsystem;indoorlightpositioningsystem

PPM.Seepulsepositionmodulation

protocol.

Seealsostandards

ALOHA84

MAC160

wiredtransmission111,112,112

proximityindoorpositioningalgorithm81,82,82,83

pulseamplitudemodulation(PAM)16–21

pulsepositionmodulation(PPM)

andlightingconstraint11

BER121–123

VLC116,118

withdimmingcontrol11

pulsewidthmodulation(PWM)

andLEDdimmingcontroltechnique60–66,61

andlightingconstraint11

assimplelow-costsystem143

IM/DDin144

imagesensors188

pulse-amplitude-modulated discrete multitone modulation (PAM-DMT) 145, 152–154,154,164

PWM.Seepulsewidthmodulation

radio-frequencytechnology(RF)

andGPS70,71,72

andmachine-to-machinecommunication137

bandwidth73,133

interferencein5,74

lightingconstraint10

notenergysaving1

radio-frequencyidentificationtechnology(RFID)74

receivedsignalstrengthmeasurement(RSS)76,76,81

RF.Seeradio-frequencytechnology

RicianK-factor140–143,142

sceneanalysisindoorpositioningalgorithm81

SCM.Seesubcarriermultiplexing

signal-to-noiseratio(SNR)

andlamparrangement51–60,53,55,

DMT165

imagesensorshelping89

LEDclipping153,154

planetiltingtechnique41–51,45,47,48

simulations

multi-leveltransmission30–33,32,

VLClamparrangement57

spectrallyfactorizedOFDM156

standards.

Seealsoprotocol

donotincludeDMT135

forportabledevices172

IEC107,109,111–113,113

IEEE6,107,109,110,112,114,172

ITU114,172

PLASA107,109,111,114

Zhaga109,112

subcarriermultiplexing(SCM)159

timecompensation(averageconstant)10,11

timeofarrivalmeasurement(TOA)75,76

time-difference-of-arrivalmeasurements(TDOA)

channelmulti-access84

outdoorlightpositioningsystem97–98,99

triangulation77–79

time-division-multi-access(TDMA)84

TOA.Seetimeofarrivalmeasurement

triangulation

indoor75–81,76,78,79

outdoor202–203,203

TTAVLCworkinggroup114,

ultra-widebandradiofrequency(UWB)74

unipolarOFDM156,164,165

variablepulsepositionmodulation(VPPM)

BER124–125

principlesof110–111,111

VLC118–119,119

vehicle-to-infrastructureVLCsystem201–202,201

visiblelightcommunication(VLC)

advantagesofpositioning73–75,74

buildingblocksof133,134

definition1,

high-speedtransmission139–143,142

historyof133,

physicallayer10–37

poseestimation198–199,199,

real-timemobile166,169,169

standards6,107–114,108,110,111,112,113

visiblelightcommunication(VLC)applications

hospital1,5

IOT1,4

ITS1,4,4

machine-to-machine139

smartcityconcept1

smartphonecameraadvertising1,6

toysandthemeparkentertainment1,5,5,6

WiFiandcellularaugmentation1–4,2,3

visiblelightcommunication(VLC)modulation

IPPM116,119

OOK116,117

PPM116,118,118

VPPM116,118–119,119

VLC.Seevisiblelightcommunication

Volterraequalization164

VPPM.Seevariablepulsepositionmodulation

wavelengthdivisionmultiplexing(WDM)34–35,36,161

wiredtransmissionprotocol(VLC)111,112,112

wirelessfidelity(WiFi)

inindoorlightpositioning88

optical137,138

VLCalternativefor41

wirelesslocalareanetwork(WLAN)

opticalWiFiin137,138

radio-frequencytechnology74

wirelesstransmissionprotocol(VLC)112

Zhagaenginestandard109,112

ZigBee112