Visible Light Communication, Networking and Sensing: A ...169.237.7.61/pubs/vlc-comsoccst.pdf1 Visible Light Communication, Networking and Sensing: A Survey, Potential and Challenges

1

Visible Light Communication, Networking andSensing: A Survey, Potential and Challenges

Parth H. Pathak∗, Xiaotao Feng,†, Pengfei Hu,∗, Prasant Mohapatra∗∗Computer Science Department, †Electrical and Computer Engineering Department,

University of California, Davis, CA, USA.Email: {phpathak, xtfeng, pfhu, pmohapatra}@ucdavis.edu

F

Abstract—The solid-state lighting is revolutionizing the indoorillumination. Current incandescent and fluorescent lamps arebeing replaced by the LEDs at a rapid pace. Apart from ex-tremely high energy efficiency, the LEDs have other advantagessuch as longer lifespan, lower heat generation and improvedcolor rendering without using harmful chemicals. One additionalbenefit of LEDs is that they are capable of switching to differentlight intensity at a very fast rate. This functionality has given riseto a novel communication technology (known as Visible LightCommunication - VLC) where LED luminaires can be used forhigh speed data transfer. This survey provides a technologyoverview and review of existing literature of visible light com-munication and sensing.

This paper provides a detailed survey of (1) visible lightcommunication system and characteristics of its various com-ponents such as transmitter and receiver, (2) physical layerproperties of visible light communication channel, modulationmethods and MIMO techniques, (3) medium access techniques,(4) system design and programmable platforms and (5) visiblelight sensing and application such as indoor localization, ges-ture recognition, screen-camera communication and vehicularnetworking. We also outline important challenges that need tobe addressed in order to design high-speed mobile networksusing visible light communication.

1 INTRODUCTIONThe indoor lighting is going through a revolution. Theincandescent bulb that has been widely used to lit oursurroundings since its invention over a century ago isslowly being phased out due to its extremely low energyefficiency. Even in the most modern incandescent bulbs,no more than 10% of the electrical power is convertedto useful emitted light. The compact fluorescent bulbsintroduced in 1990s have gained increasing popularity inthe last decade as they provide a better energy efficiency(more lumens per watt). However, recent advancementsin solid-state lighting through Light Emitting Diodes(LEDs) have enabled unprecedented energy efficiencyand luminaire lifespan. Average luminous efficacy (howmuch electricity is used to provide the intended illu-mination) of best-in-class LEDs is as high as 113 lu-

mens/watt in 2015 [1], and is projected to be around200 lumens/watt by the year 2020. This is a many foldincrease compared to current incandescent and fluores-cent bulbs which provide an average luminous efficacyof 15 and 60 lumens/watt [1] respectively. Similarly, thelifespan of LEDs ranges from 25,000 to 50,000 hours- significantly higher than compact fluorescent (10,000hours). Apart from the energy savings and lifespanadvantages, the LEDs also have other benefits like com-pact form factor, reduced usage of harmful materials indesign and lower heat generation even after long periodof continuous usage. Due to these benefits, LED adoptionis on a consistent rise and it is expected that nearly 75%of all illumination will be provided by LEDs by the year2030 [1].

The rapid increase in the usage of LEDs has provideda unique opportunity. Different from the older illumi-nation technologies, the LEDs are capable of switchingto different light intensity levels at a very fast rate. Theswitching rate is fast enough to be imperceptible by ahuman eye. This functionality can be used for commu-nication where the data is encoded in the emitting lightin various ways. A photodetector (also referred as alight sensor or a photodiode) or an image sensor (matrixof photodiodes) can receive the modulated signals anddecode the data. This means that the LEDs can servedual purpose of providing illumination as well as com-munication. In last couple of years, VLC research hasshown that it is capable of achieving very high data rates(nearly 100 Mbps in IEEE 802.15.7 standard and uptomultiple Gbps in research). The communication throughvisible light holds special importance when comparedto existing forms of wireless communications. First, withthe exponential increase of mobile data traffic in last twodecades has identified the limitations of RF-only mo-bile communications. Even with efficient frequency andspatial reuse, the current RF spectrum is proving to bescarce to meet the ever-increasing traffic demand. Com-pared to this, the visible light spectrum which includeshundreds of terahertz of license free bandwidth (seeFig. 1) is completely untapped for communication. The

2

300 MHz 300 GHz 430 THz 790 THz 30 PHz 30 EHz

Microwave Infrared Visible Ultraviolet X-ray Gamma

105 m 1 mm 750 nm 380 nm 10 nm 0.01 nm Wavelength

Frequency

VioletBlueGreenYellowOrangeRed

Radio

3 KHz

1 m

Fig. 1: Human eye can perceive the electromagnetic signals between the frequency range of 430 THz and 790 THzwhich is referred as the visible light spectrum.

Visible Light Communication (VLC) can complement theRF-based mobile communication systems in designinghigh-capacity mobile data networks. Second, due to itshigh frequency, visible light cannot penetrate throughmost objects and walls. This characteristic allows oneto create small cells of LED transmitters with no inter-cell interference issues beyond the walls and partitions.It can also increase the capacity of available wirelesschannel dramatically. The inability of signals to penetratethrough the walls also provides an inherent wirelesscommunication security. Third, VLC facilitates the reuseof existing lighting infrastructure for the purpose ofcommunication. This means that such systems can be de-ployed with relatively lesser efforts and at a lower cost.This untapped potential of visible light communicationhas motivated us to compile this survey.

The pioneering efforts of utilizing LEDs for illumi-nation as well as communication date back to year2000 when researchers [2] in Keio University in Japanproposed the use of white LED in homes for buildingan access network. This was further fueled by rapid re-search, especially in Japan, to build high-speed commu-nication through visible light with development of VLCsupport for hand-held devices and transport vehicles.This led to formation of Visible Light CommunicationsConsortium (VLCC) [3] in Japan in November of 2003.VLCC proposed two standards - Visible Light Commu-nication System Standard and Visible Light ID SystemStandard - by 2007. These standards were later acceptedby Japan Electronics and Information Technology In-dustries Association (JEITA) [4] as JEITA CP-1221 andCP-1222 respectively. The VLCC also incorporated andadapted the infrared communication physical layer pro-posed by international Infrared Data Association (IrDA)[5] in 2009. In parallel, hOME Gigabit Access project(OMEGA) [6], sponsored by European Union, also de-veloped optical communication as a way to augmentthe RF communication networks. In 2014, VLCA (VisibleLight Communications Associations) [7] is established asa successor of VLCC in Japan for further standardizationof VLC. The first IEEE standard for visible light com-munication was proposed in 2011 in the form of IEEE802.15.7 [8] which included the link layer and physical

layer design specifications. In last couple of years, theachievable VLC link capacity has surpassed 1 Gbps, andincreasing research efforts are being directed towardsrealizing the full potential of VLC.

In this survey, we provide a systematic view of VLCresearch and identify important challenges. Specifically,we provide technology overview and literature reviewof

1) Visible light communication system componentsand, details of transmitter and receiver characteris-tics,

2) Physical layer characteristics such as channelmodel and propagation, modulation and cod-ing schemes, and Multiple-Input Multiple-Output(MIMO) techniques,

3) Link layer, multiple user access techniques andissues,

4) System design and various programmable VLCplatforms,

5) Visible light sensing and applications such as vis-ible light indoor localization, human computer in-teraction, device-to-device communication and ve-hicular communication applications.

Based on the review, we then outline a list of challengesthat need to be addressed in future research to realizefull potential of VLC.

The growing interest in VLC has resulted in a fewsurveys in past couple of years. This article differs fromthese surveys in many ways. In [9], authors discussedLED-based VLC where the primary focus of discussionwas on design of physical layer techniques (modulation,circuit design etc.) that can enhance the performanceof VLC. Compared to [9], this article focuses on abroader discussion about VLC, covering other aspectsof networking such as medium access as well as sensingusing visible light. Medium access protocols for VLChave been surveyed in [10], however, no comprehen-sive overview and comparison of networking techniqueshave been provided. Also, in this paper, we show thatthe usage of smartphone camera and light sensor forreceiving visible light signals extend the VLC to otherrelated fields of mobile computing and sensing. Multipleresearch topics in this area such as indoor localization

3

and smartphone screen-camera communication are notsurveyed in any earlier work before this paper. In thispaper, we provide a comprehensive survey of thesetopics with additional focus on visible light sensing. Com-pared to [11] and [12] where authors surveyed free-space communication along with other forms of opti-cal wireless communications, the primary focus of thissurvey is narrower and more detailed towards visiblelight communication. In another related survey, authorsprovided a detailed overview of how optical wirelesscommunication can be used for cellular network designin [13], with different aspects of outdoor environmentand its impact on the communication performance. Com-pared to this, our primary focus in this paper is onvisible light communication primarily in indoor settings.Authors provided a brief survey of VLC applications in[14] with some discussion on vehicular networks andindoor broadcasting. However, in this paper, we surveya growing body of literature since the publication of [14]focusing on novel applications of VLC such as indoorlocalization, screen-camera communication etc. We alsodetail various practical aspects of communication systemdesign by reviewing currently available programmableplatforms and LED transmitters/receivers. This will en-able researchers with RF communication backgroundto easily extend their expertise in visible light wirelessaccess networks.

The rest of the survey is organized as follows. We startby providing an overview of various components of avisible light communication system with introductionto LED luminaires and different types of receivers inSection 2. In Section 3, we survey the physical layerproperties of VLC with details on channel and propaga-tion, modulation methods and MIMO techniques. It alsoincludes an overview of VLC standard IEEE 802.15.7 [8].This is followed by Section 4 where various link layerand medium access protocols are discussed. Section 5describes various aspects of VLC system design andsurveys available programmable platforms that can beused for research. Section 6 reviews a wide variety oftopics in visible light sensing and applications whichincludes indoor localization, screen-camera communica-tion, vehicular communication and human-computer in-teraction. Based on the review, Section 7 outlines variouschallenges that need further research in order to buildhigh-capacity, mobile VLC networks. We have compiledthe acronyms used throughout the paper and presentedthem with their full forms in Table 1.

2 VLC SYSTEM OVERVIEWIn this section, we provide an overview of visible lightcommunication system and its transmitter and receivercomponents. We then discuss various modes of VLC.

2.1 VLC TransmitterThe transmitter in a visible light communication systemis an LED luminaire. An LED luminaire is a complete

TABLE 1: Acronyms and their full names

Acronym Full formACO-OFDM Asymmetrically-Clipped Orthogonal Frequency Di-

vision MultiplexingADC Analog to Digital ConverterAoA Angle of ArrivalBER Bit Error Rate

BIBD Balanced Incomplete Block DesignsCAP Contention Access PeriodCCA Clear Channel Assessment

CC Convolutional CodingCCM Code Cycle ModulationCRC Cyclic Redundancy CheckCSK Color Shift Keying

CSMA-CA Carrier Sense Multiple Access - Collision AvoidanceDAC Digital to Analog Converter

DCO-OFDM Direct Current biased Optical Orthogonal Fre-quency Division Multiplexing

DMT Discrete MultiToneDOPPM Differential Overlapping Pulse Position Modulation

DPPM Differential Pulse Position ModulationDSRC Dedicated Short-Range CommunicationEPPM Expurgated Pulse Position Modulation

FEC Forward Error CorrectionFET Field Effect Transistor

FOV Field Of ViewFPS frames per second

Gbps Gigabits per secondGPS Global Positioning SystemGTS Guaranteed Time SlotHCI Human Computer Interaction

HetNets Heterogeneous NetworksIM/DD Intensity Modulation/Direct Detection

JT Joint TransmissionKbps Kilobits per secondLCD Liquid-crystal-displayLED Light Emitting DiodeLOS Line Of SightLTE Long Term Evolution

MAC Medium Access ControlMbps Megabits per secondMCS Modulation and Coding Scheme

MEPPM Multi-level Expurgated Pulse Position ModulationMIMO Multiple Input Multiple OutputMISO Multiple Input Single Output

MPPM Multipulse Pulse Position ModulationMU-MIMO Multiple User - Multiple Input Multiple Output

NFC Near Field CommunicationNRZ Non Return to Zero

OCDMA Optical Code Division Multiple AccessOFDMA Orthogonal Frequency Division Multiple Access

OFDM Orthogonal Frequency Division MultiplexingOMPPM Overlapping Multipulse Pulse Position Modulation

OOC Optical Orthogonal CodesOOK On Off Keying

OPPM Overlapping Pulse Position ModulationPAPR Peak to Average Power Ratio

PDP Power Delay ProfilePPM Pulse Position Modulation

PWM Pulse Width ModulationQAM Quadrature Amplitude Modulation

RC Repetition CodingRF Radio Frequency

RGB Red Green BlueRLL Run Length Limited

RS Reed-Solomon codingRSS Received Signal Strength

RTS/CTS Request To Send/Clear To SendSFO Sampling Frequency Offset

SISO Single Input Single OutputSLM Spatial Light ModulatorSMP Spatial Multiplexing

SM Spatial ModulationSNR Signal to Noise RatioVLC Visible Light Communication

VPPM Variable Pulse Position ModulationWDM Wavelength Division Multiplexing

4

lighting unit which consists of an LED lamp, ballast,housing and other components. The LED lamp (alsoreferred as an LED bulb in simpler terms) can includeone or more LEDs. The lamp also includes a drivercircuit which controls the current flowing through theLEDs to control its brightness. When an LED luminaireis used for communication, the driver circuit is modified(further details in Section 5) in order to modulate thedata through the use of emitted light. For example, in asimple On-Off Keying modulation, the data bit “0” and“1” can be transmitted by choosing two separate levelsof light intensity.

A crucial design requirement for VLC system is thatillumination, which is the primary purpose of the LEDluminaries, should not be affected because of the com-munication use. Hence, performance of the VLC systemis also affected depending on how the LED luminairesare designed. White light is by far the most commonlyused form of illumination in both indoor as well asoutdoor applications. This is because colors of objects(also known as color rendering) as seen under the whitelight closely resemble the colors of the same objectsunder the natural light. In solid-state lighting, the whitelight is produced in following two ways -

1) Blue LED with Phosphor: In this method, thewhite light is generated by using a blue LED thathas yellow phosphor coating. When the blue lighttraverses through the yellow coating, the combina-tion produces a white light. Different variations ofthe white light (color temperatures) are producedby modifying the thickness of the phosphor layer.

2) RGB Combination: White light can also producedby proper mixing of red, green and blue light. Inthis method, three separate LEDs are used whichincreases the cost of LED luminaire compared tousing the Blue LED with Phosphor.

Due to ease of implementation and lower cost, thefirst method with blue LED and phosphor is morecommonly used for designing white LED. However, interms of communication, the phosphor coating limits thespeed at which LED can switched to a few MHz. Aswe will discuss in Section 3.2, various solutions havebeen proposed to alleviate this limitation. On the otherhand, RGB combination is preferable for communicationas it also creates an opportunity of using Color ShiftKeying to modulate the data using three different colorwavelength LEDs.

2.2 VLC Receiver

Two types of VLC receivers can be used to receive thesignal transmitted by an LED luminaire

1) photodetector - also referred as photodiode or non-imaging receiver,

2) imaging sensor - also called a camera sensor.The photodetector is a semiconductor device that con-verts the received light into current. The current com-

mercial photodetectors can easily sample the receivedvisible light at rates of tens of MHz.

An imaging sensor or a camera sensor can also be usedto receive the transmitted visible light signals. Becausesuch camera sensors are available on most of today’smobile devices like smartphones to capture videos andimages, it has the potential to convert the mobile devicesin readily available VLC receivers. An imaging sensorconsists of many photodetectors arranged in a matrixon an integrated circuit. However, the limitation of animaging sensor is that in order to enable high-resolutionphotography, the number of photodetectors can be veryhigh. This significantly reduces the number of framesper second (fps) that can be captured by the camerasensor. For example, the fps of commonly used camerasensors in smartphones is no more than 40. This meansthat direct use of camera sensor to receive visible lightcommunication can provide very low data rate.

The “rolling shutter” property of camera sensor can beused to receive the data at a faster rate. Due to a largenumber of available photodetectors in a camera sensor, itis not possible to read the output of each pixel in parallel.Instead modern camera sensors employ row scanningwhere photodetectors of one row of the matrix is read ata time. This procedure of reading photodetector outputrow by row (or column-by-column) is referred as rollingshutter. Fig. 2a shows how the rolling shutter process canbe leveraged to increase the data rate. For illustrationpurposes, we assume that the transmitter uses ON-OFF modulation. The transmitter can change its state(transmit the next symbol) in a time shorter than the timerequired to scan a row of pixels. As shown in Fig. 2a,the transmitter is in ON state first which results in higherintensity output for pixels of the first column. At the nexttime instance, it changes its state by switching to OFFstate. This can be recorded as low intensity output forpixels of the second column. Once all the columns arescanned, all the columns of the resultant image can beconverted to binary data. It was shown in [16] that multi-kbps of throughput can be achieved using the rollingshutter process of camera sensor.

Note that image sensor can allow any mobile devicewith camera to receive visible light communication.However, in its current form, it can only provide verylimited throughput (few kbps) due to its low samplingrate. On the other hand, stand-alone photodetectors haveshown to achieve significantly higher throughput (hun-dreds of mbps). In this survey, we assume the receiverto be the photodetector unless otherwise mentionedspecifically.

2.3 VLC Modes of Communication

Visible light communication can be classified into twomodes: (1) Infrastructure-to-device communication and(2) Device-to-device communication. An indoor scenariowhere LED luminaires are used to illuminate the roomis shown in Fig. 2b. In this case, the luminaires can

5

ON OFF ON OFFTransmitter LED state

Image sensor readout

(a) The rolling shutter effect observed when receiving data using an image sensor.

Thermostat

LED lightsSafety alarm

Television

(b) An example scenario showing that LEDs cancommunicate to various devices including user’s

mobile devices and other smart devices;reproduced using [15]

Fig. 2: The rolling shutter effect and typical usage scenario of an indoor VLC network

transmit data to various devices inside the room. TheLEDs can also coordinate between themselves to reducethe interference and even enable coordinated multi-pointtransmission to receiving devices. The uplink transmis-sion from the devices are difficult to achieve becauseusing LEDs on end-user devices can cause noticeabledisturbance to users. In such case, RF or infrared com-munication can be used for the uplink transmissions.Similar to the indoor case, the LEDs used in street lampsas well as traffic lights can be used to provide internetaccess to users in cars and pedestrians. We will discusssuch vehicular application in Section 6.3.

Due to omni-present camera sensor for mobile de-vices, the visible light communication can also be usedfor near-field device-to-device communication. Here, theLED pixels on the display of one smartphone can beused to transmit data to the camera sensor of anothersmartphone. With recent advances in design of efficientcodes, such screen-to-camera streaming has been shownto achieve very high throughput. We discuss these tech-niques in Section 6.2. In another form of device-to-devicecommunication, cars and other vehicles on the road cancommunicate with each other to form an ad-hoc networkusing VLC.

Although we discussed the vehicular networking andscreen-camera communication, our primary focus inthis survey is towards design and analysis of indoorinfrastructure-to-device networking using visible light.

3 PHYSICAL LAYERWe start with a comprehensive overview of VLC physi-cal layer by discussing (1) channel model and character-istics, (2) modulation methods and (3) MIMO techniquesfor VLC.

3.1 Channel Model and Propagation Characteristics

In this section, we describe the channel model for prop-agation of visible light. Based on the channel model, it

is possible to choose an LED with appropriate specifica-tions and estimate its communication link performance.Note that the notations symbols used throughout thissection are listed in Table 2 with their meaning.

3.1.1 Transmitted Power of an LED - Luminous Flux

An LED transmitter serves dual purpose of illuminationand communication. Therefore, it is necessary to firstestablish an understanding of relevant photometric andradiometric parameters. Using these parameters, we willbe able to calculate the Luminous Flux which is thetransmitted power of an LED transmitter. First, we willcalculate the transmitted power, path loss and receivedpower of a Line-Of-Sight (LOS) link and then analyzethe multipath impact of reflected paths.

Photometric parameters quantify the characteristics oflight (such as brightness, color etc.) as perceived by thehuman eye. They are useful in understanding the illumi-nation aspects of LEDs. Radiometric parameters measurethe characteristics of radiant electromagnetic energy oflight. They are useful in determining communicationrelated properties of LEDs. There are two ways of cal-culating the Luminous Flux - using spectral integral orusing spatial integral. Depending on which parametersare available for a given LED transmitter, one of the twomethods can be chosen for calculation of luminous flux.

Spectral Integral: The spectral integral method usesluminosity function of human eye and spectral powerdistribution of an LED to derive the luminous flux.

Luminosity Function V (λ): The photopic vision of humaneye allows humans to distinguish different colors, mak-ing it a crucial factor in designing lighting technology[17]. It was shown in [18] that human’s photopic visionexhibits different levels of sensitivity to different wave-lengths of visible light spectrum. This aspect is shownin Fig. 3 using the luminosity function V (λ). The functionshows that human eye can see the colors within therange of 380 nm to 750 nm with the maximum sensitivityat wavelength of 555 nm (the yellow-green region).

6

TABLE 2: Symbols and their meaning

Symbol Meaningλ Wavelength

V (λ) Luminosity functiuonST (λ) Transmitter spectral power distribution function

FT Transmitter luminous fluxFR Receiver luminous flux

gt(θ) Luminous intensity distributionI0 Axial intensity

θmax Half beam angleΩmax Full beam angleLL Luminous path lossLP Optical power path lossD Distance between transmitter and receiverr Radius of the receiver apertureα Incident angleβ Irradiation angleAr Receiver aperture areaΩr Receiver solid angle from transmitterm Order of Lambertial emission

φ1/2 Semi-angle at half illuminanceRf (λ) Spectral responsivity functionPRo Received optical power

SR(λ) Receiver spectral power distribution functionλrL Lower wavelength cut-off for optical filterλrH Higher wavelength cut-off for optical filterPR(i) Received optical power from LOS link of ith LED

PR(total) Total received optical powerρ(λ) Spectral reflectranceN Number of LED transmittersk Number of bounces of light

h(t) Power delay profileδ Dirac delta functionc Speed of light

FOV Acceptance angle of receiverΓ(k)n Power of reflected ray after kth bounce

σshot Standard deviation of shot noiseσthermal Standard deviation of thermal noise

x Number of photons collected in unit timeκ Boltzmann’s constantIB Photocurrent due to background noiseGol Open-loop voltage gainTk Absolute temperature

Cpd Capacitance of the photodetector per unit areaη FET channel noise factor

gm FET transconductanceI2, I3 Noise-bandwidth factors

Spectral Power Distribution ST (λ): The ST (λ) of an LEDis the function representing the power of the LED atall wavelengths in the visible light spectrum. The LEDvendors typically publish the distribution to explainhow different colors will be rendered in the presenceof the LED. It is a radiometric parameter measuredin Watts/nm. The spectral power distribution of threedifferent colored LEDs are shown in Fig. 4. It can beobserved that all three LEDs have high radiant powerat two wavelengths - blue and yellow. As describedin Section 2.1, most current LEDs produce white lightby combining blue light emitted by a blue LED withyellow phosphor coating. Depending on the desiredtype of white color (warm, natural or cool), blue andyellow light emissions are controlled using the phosphorcoating. For example, more yellow light is allowed inwarm and natural white compared to the cool whiteLED.

Luminous Flux: The luminous flux combines luminos-ity function and spectral power distribution to calculate

Hu

ma

n e

ye

se

nsitiv

ity V

(λ)

Wavelength (λ) (nm)

0.0001

0.001

0.01

0.1

1

300 400 500 600 700 800 900

Vio

let

Blu

eC

ya

n

Gre

en

Ye

llow

Ora

ng

e

Re

d

Visible spectrum

Maximum sensitivity at 555 nm

Fig. 3: Luminosity function representing human eye’ssensitivity to different wavelengths in the visible

spectrum.

No

rma

lize

d R

ad

ian

t P

ow

er

(%)

Wavelength (λ) (nm)

Warm white

Natural white

Cool white

0

10

20

30

40

50

60

70

80

90

100

400 450 500 550 600 650 700 750

Fig. 4: Power spectral distribution for LED of threecolor types - warm white, natural white and coolwhite. Warm white and natural white have more

radiated power for green-yellow-orange wavelengthscompared to cool white which provides a more bluish

illumination; Figure reproduced from [19].

the “perceived” power emitted by the LED. It weighs theST (λ) function with V (λ) (the sensitivity of human eyeto different wavelengths) because we know from Fig. 3that human eye does not respond to all wavelengthsequally. The luminous flux of the transmitter LED (FT )is measured in lumens and it can be calculated as

FT = 683 (lumens/watt)

750 nm∫380 nm

ST (λ)V (λ)dλ (1)

The constant 683 lumens/watt is the maximum luminousefficiency. The luminous efficiency is the ratio of luminousflux to the radiant flux, which measures how well theradiated electromagnetic energy and required electricityof an LED was transformed to provide visible lightillumination. We know from Fig. 3 that human eyeis most sensitive to detect the wavelength of 555 nm(green). The electrical power necessary to produce onelumen of light at the wavelength of 555 nm is derived tobe 1/683rd of a watt [20]. This means that for any other

7

color source, the power necessary to produce one lumenof light is always higher than 1/683rd of a watt. Hence,the maximum luminous efficiency is 683 lumens/wattwhich occurs at 555 nm wavelength.

Spatial Integral: Another way of calculating the lumi-nous flux is to utilize LED’s spatial emission properties.For this, we will use luminous intensity and axial inten-sity as described next.

Rela

tive

Lu

min

ou

s I

nte

nsity (

%)

Angle

LED 1LED 2

20

40

60

80

100

-100 -80 -60 -40 -20 0 20 40 60 80 100

(a)

90

45

0

300

600

900

1200

Lum

inou

s in

tens

ity (c

ande

la)

Angle

I0= 987 candela

494

max=47o

(b)

Fig. 5: (a) Luminous intensity distribution for two LED- (1) Cree XLamp XP-E High-Efficiency White [19] (2)

Cree XLAMP XR-E [21] (b) Luminous intensitydistribution of Cree LMH6 in polar coordinates [22]

and its half-beam angle; Figures reproduced from [19],[21], [22].

Luminous Intensity gt(θ): While luminous flux mea-sures the total amount of light emitted by an LED, theluminous intensity measures how bright the LED is ina specific direction. It is measured in Candela which isluminous flux per unit solid angle (1 steradian). Thisallows us to understand where the LED directs its light.Fig. 5 shows the luminous intensity distribution of threedifferent LEDs. In Fig. 5a, both the LEDs emit lightat wider angles allowing better illumination in manydirections, while in Fig. 5b, it can be observed that LEDemits light in a narrower beam (much like spotlighting).Most LED sources have Lambertial beam distribution[23] which means that the intensity drops as the cosineof the incident angle.

There are two important parameters to be derivedfrom the intensity distribution

Axial Intensity (I0) is defined as the luminous intensityin candelas at 0o solid angle. For LED in Fig. 5b, the axialintensity is 987 candela. Typically, the luminous intensitydistribution provided by the vendors are normalizedwith the axial intensity as shown in Fig. 5a.

Half Beam Angle (θmax) is the angle at which the lightintensity decreases to half of the axial intensity. For theLED in Fig. 5b, the half beam angle is 47o. For theLambertian sources like LEDs, the half beam angle iscalculated from the entire beam angle (Ωmax) as follows

Ωmax = 2π(1− cos θmax) (2)

The luminous flux can now be calculated by integrat-ing the luminous intensity function over the entire beam

solid angle Ωmax. Different from Equ. (1) which was aspectral integral, here the flux is calculated using spatialintegral as below

FT =

Ωmax∫0

I0gt(θ)dΩ (3)

where gt(θ) is the normalized spatial luminous intensitydistribution. Combining Equs. (2) and (3), we get

FT = I0

θmax∫0

2πgt(θ)sinθdθ (4)

3.1.2 Path Loss and Received PowerBased on the luminous flux calculated above, we will

now derive the value of path loss. It was proven in[24] that the path loss in photometric domain (referredas luminous path loss LL) is the same the path lossin radiometric domain (referred as optical power pathloss LP ). This is due to the fact that in line-of-sightfree space propagation, the path loss can be assumedto be independent of the wavelength. Therefore, we cancalculate LL using the luminous flux derived in theprevious section. Specifically, LL is the ratio of luminousflux of the receiver (FR) and the transmitter (FT ). FT canbe calculated as Equ. (4).

max

radius (r)

Distance (D)

Transmitter

Receiver

Tran

smitt

er r

Ar

Receiver

Fig. 6: Relative position of transmitter and receiver inLOS settings; reproduced from [24].

In order to calculate the FR, it is necessary to specifythe relative positions of the transmitter and the receiver.This relative positioning is shown in Fig. 6. Here, thedistance between the receiver and the transmitter isD, and radius of the receiver aperture is r. The anglebetween the receiver normal and transmitter-receiverline is α (also referred as incident angle). The transmitterviewing angle is β (also referred as irradiation angle). Letthe receiver solid angle as observed from the transmitterbe Ωr and receiver’s area Ar as shown in Fig. 6, then

Arcos(α) = D2Ωr (5)

From Fig. 6, the receiver flux FR can be calculated as

FR = I0gt(β)Ωr (6)

8

The optical path loss LL can be calculated using Equa-tions (4), (5) and (6) as

LL =FRFT

=gt(β)Arcosα

D2θmax∫

0

2πgt(θ)sinθdθ

(7)

Most LED sources have Lambertial beam distributionwhich means that the spatial luminous intensity distri-bution is a cosine function

gt(θ) = cosm(θ) (8)

where m is the order of Lambertial emission. The valueof m depends on the semi-angle at half illuminance Φ1/2of the LED

m =ln(2)

ln(cosΦ1/2)(9)

Substituting Equ. (8) and θmax in Equ. (7), we get thepath loss value for a Lambertian LED source as follows

LL =(m+ 1)Ar

2πD2cos α cosm(β) (10)

If the LED emission can not be modeled using theLambertian cosine function, it is necessary to measuregt(θ) for the given LED, and use it to calculate LL fromEqu. (7).

The received optical power can be now calculatedusing the path loss. It is typical that the receiving pho-todetector is equipped with an optical filter. Let Rf (λ)denote the spectral response of the optical filter. Fig. 7shows Rf (λ) of a typical photodetector. Using Rf (λ), the

0.1

0.2

0.3

0.4

0.5

0.6

300 400 500 600 700 800 900 1000 1100 1200

Responsiv

ity (

A/W

)

Wavelength (nm)

Fig. 7: Spectral response of a typical photodetectorreceiver; responsivity (measured in A/W) is the ratio of

output photocurrent in amperes to incident radiantenergy in watts; reproduced from [25].

received optical power PRO for the direct line-of-sightoptical link can be calculated as

PRO =

λrH∫λrL

SR(λ)Rf (λ)dλ (11)

where SR(λ) = LPST (λ) = LLST (λ) and λrL and λrHare lower and upper wavelength cut-off values for theoptical filter respectively.

Considering Equations (10) and (11), the receivedpower is dependent on three factors - the transmitter-receiver distance (D), incident angle (α) and irradiation

angle (β). These three factors are independent of trans-mitter and receiver hardware, and depend on receiver’smovement and orientation. As an example, if the receiveris a smartphone equipped with a photodiode, the threefactors will change based on user’s movement and de-vice orientation. It is crucial to understand the impact ofthese factors on received power in order to evaluate theachievable capacity. Authors in [26] studied the impactusing a smartphone photodiode as the receiver. Fig. 8show how the normalized received power (measuredas light intensity on smartphone photodiode) varieswith changes in D, α and β. Fig. 8a shows how thereceived power attenuates with D as inverse square low(Equ. (10)). The incident angle measures the changesin smartphone’s orientation (0o means photodiode isdirectly facing the LED). As the incident angle (α) in-creases, the energy at which the photons strike the pho-todiode decreases, which in turn results in decrease ofreceived power. Similarly, the received power decreaseswith increase in the irradiation angle (β) confirmingthe lambertian emission pattern of the LED. The impactof these three factors have important implications onguaranteeing high SNR in VLC access networks andmanaging inter-cell interference as we will discuss inSection 4.2.

3.1.3 Multipath Propagation with Reflected PathsAs we saw in Section 2, typically there are more than oneLED in a luminaire. The receiving photodetector can si-multaneously receive (intensity modulated) signals frommultiple LEDs as shown in Fig. 2b. The received opticalpower of the receiver can be calculated by summing thereceived power of each LOS link within receiver’s field-of-view (FOV) can be expressed as

PR(total) =

N∑i=0

PR(i) (12)

where N is the total number of LEDs and PR(i) isthe received optical power from LOS link of ith LEDcalculated from Equ. (11).

Since the majority of the indoor surfaces are more orless reflective of visible light, it is necessary to under-stand the impact of reflected paths on the performanceof communication. Spectral reflectance (ρ(λ)) representsreflectivity of a surface (such as wall, ceiling etc.) asa function of wavelength. It was noted in [27] thatreflectivity of Infrared signal is higher compared to thevisible band. The spectral reflectance of commonly usedbuilding materials like plaster wall, ceiling etc. wasmeasured in [27] using a spectrophotometer. Fig. 9 showsthe results of measured reflectivity. It can be observedthat plastic wall has the least reflectivity while the plasterwall has the highest reflectivity.

Because of the reflections, the receiver receives signalfrom many different paths. Such multipath propagationcan be characterized using Power Delay Profile (PDP).The PDP gives the distribution of received power as a

9

0.96

0.76

0.57

0.38

0.19

01 2 3 4 5 6

Nor

mal

ized

rece

ived

pow

er

Distance - D (m)

Incident angle = 0Irradiance angle = 0

(a)

10.80.60.40.2

0 90 75 60 45 30 15 0

Nor

mal

ized

rece

ived

pow

er

Incident angle

Distance = 4 mIrradiance angle = 0

(b)

0.2

0.4

0.6

0.8

1

0 15 30 45 60 75

Nor

mal

ized

rece

ived

pow

er

Irradiance angle

Distance = 4 mIncident angle = 0

(c)

Fig. 8: Impact of (a) transmitter-receiver distance, (b) incident angle (α) and (c) irradiation angle (β) on thereceived power; reproduced from [26]

0

0.2

0.4

0.6

0.8

1

350 400 450 500 550 600 650 700 750

Sp

ectr

al re

fle

cta

nce

Wavelength (nm)

Plastic wallCeilingFloorPlaster wall

Fig. 9: Different indoor surfaces exhibit different levelsof spectral reflectance depending on the wavelength;

reproduced from [27].

function of propagation delay. A non-LOS signal canbe bounced from many surfaces before it reaches thereceiver photodetector as shown in Fig. 10. Authors in

Transmitter

Receiver

1

2

01

k

k+1

2

0

k+1 FOV

D1

2Dk+1D

Firstbounce

k-thbounce

0D

Fig. 10: A non-LOS signal can bounce off the surfacesmany times before reaching the receiver; β and α

denote the angle of irradiation and incidentrespectively; reproduced from [27].

[27] modeled the PDP of multiple bounces for a total of

N LEDs at time instance t as

h(t) =

N∑n=1

∞∑k=0

h(k)(t;Sn) (13)

where Sn is the spectral power distribution of nth LEDand k is the number of bounces. When k = 0, theresultant PDP [27] is that of an LOS path as

h(0)(t;Sn) = L0Pnrect

(α0FOV

)δ

(t− D0

c

)(14)

where L0 = LL is the path loss for the LOS case (derivedin Equ. (10)), δ is a dirac delta function, D0 is the distancebetween the LED and the receiver and c is the speedof light. Because the photodiode can only detect thelight whose angle of incidence is smaller than its FOV, arectangular function [27] is used where

rect(x) =

{1 for |x| ≤ 10 for |x| > 1

This means that when if a ray does not reach within theFOV of the receiver after k bounces, its effect on the totalreceived power is considered 0.

When k ≥ 1, the PDP after k bounces (refer Fig. 10)for the nth LED can be calculated [27] as

h(k)(t;Sn) =

∫s∈S

[L1L2 · · ·Lk+1Γ(k)n rect

(α0FOV

)× (15)

δ

(t− D1 +D2 + · · ·+Dk+1

c

)]dAs (16)

where

L1 =As(m+ 1) cos α1 cos

mβ1

2πD12 (17)

For the path loss of the first bounce L1, the ray origi-nated from the LED which we have previously modeledas a Lambertian emitter (Equ. (8)). For the remainingbounces, we can calculate the path loss of each path as

L2 =As cos β2 cos α2

πD22 (18)

Lk+1 =AR cos βk+1 cos αk+1

πDk+12 (19)

10

The integration in Equ. (16) for each surface s of allreflectors S where As is the area of the surface. For Lk+1,AR is the area of the photodiode receiver. Γ

(k)n is power

of the reflected ray after kth bounce. It is calculated [27]as

Γ(k)n =

∫λ

Sn(λ)ρ1(λ)ρ2(λ) . . . ρk(λ)dλ (20)

where ρk(λ) is the spectral reflectance of the surface ofkth bounce.

43.5

32.5

21.5

10.5

00 5 10 15 20 25 30

Rec

eive

d po

wer

(W)

10-5

Time (ns)

Plaster wallPlastic wall

Fig. 11: Power delay profile for 4 LED transmitters in acubic room with plaster or plastic walls; reproduced

from [27].

Fig. 11 shows the power delay profile in a realisticscenario where four LED luminaires are deployed ina square topology on a ceiling of a cubic room witheither plaster or plastic walls [27]. It can be observedthat the first peak is due to the direct received signal(LOS) from the LED. The other peaks are due to multiplereflections from the wall as calculated using Equ. (16).As expected, the received power due to reflection multi-path is relatively lesser compared to the LOS power.

Most of the power delay profiling [27]–[30] of visiblelight communication rely on simulations. However, de-tailed measurement-based studies in realistic scenarios(such as indoor places with many different reflectingobjects, different LED arrangements etc.) are necessaryfor improved understanding of multi-path in VLC anddeveloping the techniques to combat it.

3.1.4 Receiver Noise and SNRThere are three major sources of noise in indoor visiblelight optical link (1) ambient light noise due to solarradiation from windows, doors etc. and noise due toother illumination sources such as incandescent andfluorescent lamps, (2) shot noise induced in the pho-todetector by the signal and the ambient light and (3)electrical pre-amplifier noise (also known as thermalnoise) of the photodetector.

The ambient noise of solar radiation and artificialillumination sources such as lamps results in ambientnoise floor which is a DC interference. The effect ofsuch noise can be mitigated by using a electrical high

pass filter at the receiver. Most of the previous studiesassume that this ambient noise floor remains stationaryover space and time, however, no systematic evaluationis present in the literature. For example, the indoor solarradiation changes at different places depending on win-dows and doors. The radiation also changes dependingon the time of the day (and year) and orientation ofthe windows/doors. Radiation from other illuminationsources will also remain an unavoidable source of noiseuntil we completely transition to LED technology. Itis required that exhaustive indoor measurements arecarried out to accurately account for such noise.

Once the noise due to solar radiation and artificialillumination sources is filtered, the SNR at the receivercan be calculated based on the shot noise and the thermalnoise of the photodetector circuitry as

SNR =PRE

2

(σshot)2 + (σthermal)2(21)

where σshot and σthermal are the standard deviation ofshot noise and thermal noise respectively. The shot noiseis due to inherent statistical fluctuation in the amountof photons collected by the photodetector. It is knownthat the photon counting follows a poisson distributionwhich means that if the mean of number of photonscollected by the photodetector in a unit time is x, thenthe standard deviation of number of photons collectedis√x. This also results in poisson distributed variation

in photoelectrons generated by the photodetector. Basedon this, the variance of shot noise can be calculated [23],[31] as below

(σshot)2 = 2qPREB + 2qIBI2B (22)

The variance of thermal noise [23], [31] is

(σthermal)2 =

8πκTkGol

CpdAI2B2 +

16π2κTkη

gmC2pdA

2I3B3

(23)where B Hz is the bandwidth of the photodetector, κ

is the Boltzmann’s constant, IB is the photocurrent dueto background radiation, Gol is the open-loop voltagegain, Tk is the absolute temperature, Cpd is capacitanceof the photodetector per unit area, η is the FET channelnoise factor, gm is the FET transconductance, and I2 andI3 are the noise-bandwidth factors with values 0.562and 0.0868 respectively. Shot noise and thermal noiseare dependent on the area of the photodetector, anddepending on factors such as room temperature, ambientlight etc. either of them can dominate the overall noise[23] observed by the VLC receiver.

3.1.5 ShadowingThe receiver of a visible light communication link can beshadowed by different objects or humans in the indoorenvironment. For example, if a receiver photodiode ispositioned on a desk, it is possible that movement ofthe nearby chair can result in shadowing of the receiver.Similarly, if a human passes by frequently between the

11

transmitter and the receiver, the link performance isaffected by the frequent shadowing. Authors in [32]studied such case of human mobility using simula-tions and suggested that in multiple spatially sepa-rated LED sources should be used in order to mitigatethe frequent disconnections due to human shadowing.Apart from this preliminary work, shadowing in indoorVLC networks is not studied in literature. Given thatvisible light exhibits significantly different propagationcharacteristics compared to RF (such as no penetrationthrough walls etc.), it is crucial to characterize and modelvisible light shadowing in indoor environment. Thisunderstanding can also provide insights on deploymentaspects of indoor VLC networks and how they shouldbe different than current deployment of LEDs which areprimarily used for illumination purposes.

3.2 Modulation Methods

With the understanding of path-loss, noise and SNR,we now discuss various modulation methods used inVLC. The most striking difference between VLC andRF is that in VLC, the data can not be encoded inphase or amplitude of the light signal [10]. This meansthat phase and amplitude modulation techniques cannot be applied in VLC and the information has to beencoded in the varying intensity of the emitting lightwave. The demodulation depends on direct detectionat the receiver. These set of modulation techniques arereferred as IM/DD (Intensity Modulated/Direct Detec-tion) modulations. In this section, we will discuss theIM/DD modulation techniques used for visible lightcommunication.

Different from other types of communications, anymodulation scheme for VLC should not only achievehigher data rate but should also meet the requirementsof perceived light to humans. These requirements aboutperceived light can be characterized by following twoproperties -

(1) Dimming: It was suggested in [17] that differentlevels of illuminance is required when performing differ-ent types of activities. As an example, an illuminance inthe range of 30-100 lux is often enough for simple visualtasks performed in most public places. On the otherhand, office or residential applications require higherlevel of illuminance in the range of 300-1000 lux. Withthe advancements in LED driver circuits, it has becomepossible to dim an LED to an arbitrary level dependingon the application requirement to save energy.

If an LED can be dimmed to an arbitrary level, it isalso necessary to understand its impact on the humanperceived light. It was first shown in [33] that the relationbetween the measured light and the perceived light isnon-linear. This property is shown in Fig. 12. In otherwords, a human eye adapts to lower illumination byenlarging the pupil to allow more light to enter theeye. The perceived light can be calculated [33] from the

0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60 70 80 90 100

Me

asu

red

Lig

ht

(%)

Perceived Light (%)

Fig. 12: Human eye perceives the actual measured lightdifferently due to enlargement/contraction of pupil.

measured light as

Perceived light(%) = 100×√

Measured light(%)

100(24)

This means that a lamp that is dimmed 1% of its mea-sured light is perceived to be 10% dimmed by the humaneye. This is important in terms of VLC because a usermay choose an arbitrary level of dimming dependingon the application or desired energy savings, but thecommunication should not be affected by the dimming.In other words, the data should be modulated in such away that any desired level of dimming is supported.

(2) Flicker mitigation: An additional requirement forany VLC modulation scheme is that it should not resultin human-perceivable fluctuations in the brightness ofthe light. It was shown in [34] that flickering can causeserious detrimental physiological changes in humans.For this reason, it is necessary that changes in the lightintensity should happen at a rate faster than human eyecan perceive. IEEE 802.15.7 standard [8] suggests thatflickering (or change in light intensity) should be fasterthan 200 Hz to avoid any harmful effects. This meansthat any modulation scheme for VLC should mitigateflickering while providing higher data rate.

The most common cause of flickering is long runs of 0sor 1s which can reduce the rate at which light intensitychanges and cause the flickering effect. Run LengthLimited (RLL) codes are used to mitigate long runs of0s or 1s. RLL codes ensure that the output symbols havebalanced repetition of 0s and 1s. Examples of commonlyused RLL codes include Manchester, 4B6B and 8B10Bcoding. In Manchester coding, a “0” is replaced witha “down” transition (“10”) and “1” is replaced withan “up” transition (“01”). 4B6B coding maps a 4 bitssymbol to a 6 bits symbol that has balanced repetition.Similarly, 8B10B maps a 8 bits symbol to 10 bits symbol.The number of additional bits added is the highest inthe Manchester coding making it a suitable choice forlow data rate services that require better balancing. Onthe other hand, 8B10B reduces the number of additionalbits added (high data rate), however, it performs poorly

12

in terms of the DC balancing.We next discuss four types of modulation schemes

used in VLC (1) On-Off Keying, (2) Pulse modulation,(3) Orthogonal Frequency Division Modulation (OFDM)and (4) Color Shift Modulation (CSK). We describe eachof them along with a discussion on how they providethe dimming support.

3.2.1 On-Off Keying (OOK)In OOK, the data bits 1 and 0 are transmitted by turningthe LED on and off respectively. In the OFF state, theLED is not completely turned off but rather the lightintensity is reduced. The advantages of OOK includeits simplicity and ease of implementation. OOK-likemodulation is widely used in wireline communication.

Most of the early work on using OOK modulation forVLC utilize while LED. As we discussed in Section 2,such LED produces white light by combining the blueemitter with yellow phosphor. The major limitation ofthe white LED is its limited bandwidth (few megahertz[35]) due to slow time response of the yellow phos-phor. It was first proposed by [36] to use NRZ (Non-Return-to-Zero) OOK with the white LED and a datarate of 10 Mbps was demonstrated over a VLC link.To further improve the performance, [35] used a bluefilter to remove the slow-responding yellow component,resulting in a datarate of 40 Mbps. Similarly, [37] and [38]proposed to combine the blue-filtering with analogueequalization at the receiver to achieve data rates of100 Mbps and 125 Mbps respectively. Authors in [39]showed that the performance can be further improvedby using an avalanche photodiode as the receiver insteadof the P-I-N photodiode. The achievable data rate withavalanche photodiode and NRZ-OOK was shown tobe 230 Mbps. Newly available white LEDs combinethe RGB frequencies to produce the white light. Theadvantage of such LEDs is that they do not have theslow-responding yellow phosphor layer. However, suchRGB white LEDs require three separate driver circuitsto realize the white light. A different approach waspresented in [40] where RGB white LED was used butonly the red LED is modulated for data transmissionwhile the other two are provided constant current forillumination. The proposed system can achieve a datarate of 477 Mbps with simple NRZ-OOK modulation anda P-I-N photodiode receiver.

There are two ways proposed in the Standard IEEE802.15.7 [8] to provide the dimming support when usingOOK as the modulation scheme:

1) Redefine ON and OFF levels: To achieve the desiredlevel of dimming, the ON and the OFF levels can beassigned different light intensities. The advantageof this scheme is that required level of dimming canbe obtained without any additional communicationoverhead. It can retain the data rate achievable byNRZ-OOK modulation, however, the communica-tion range decreases at lower dimming levels. Onemajor disadvantage is that using lower intensities

as ON/OFF levels causes the LEDs to be operatedat lower driving currents which in turn has shownto incur changes in color rendering (change inemitted color of LEDs) [41].

2) Compensation periods: In this solution, the ON andthe OFF levels of the modulation remain the samebut additional compensation periods are addedwhen the LED source is fully turned on (calledON periods) or off (OFF periods). The durationof the compensation periods is determined basedon the desired level of dimming. Specifically, ONperiods are added if the desired level of dimmingis more than 50% and OFF periods are added if thedesired level of dimming is less than 50%. Authorsin [42] proposed a way to calculate the percentagetime of active data transmission (γ) within thetransmission interval T to obtain a dimming levelof D as

γ =

{(2− 2D)× 100 : D > 0.5

2D × 100 : D ≤ 0.5(25)

When the desired dimming level is D with OOK,the maximum communication efficiency ED can becalculated [42] using information theoretic entropyas

ED = −D log2D − (1−D) log2(1−D) (26)

This means that communication efficiency is atriangular function of the dimming level with max-imum efficiency at dimming level of 50%. Theefficiency drops linearly when dimming level de-creases to 0% or increases to 100%. The dimmingsupport using compensation periods reduces thedata rate, however, since the modulated ON/OFFsignals have unchanged intensity, the communi-cation range remains unchanged. To address theproblem of lower data rate with compensationperiods, [43] proposed to use inverse source codingto maintain the high data rate while achieving thedesired level of dimming.

3.2.2 Pulse Modulation MethodsAlthough OOK provides various advantages such assimplicity and ease of implementation, a major limita-tion is its lower data rates especially when supportingdifferent dimming levels. This has motivated the designof alternative modulation schemes based on pulse widthand position which are described next.

Pulse Width Modulation (PWM): An efficient way toachieving modulation and dimming is through the usePWM. In PWM, the widths of the pulses are adjustedbased on the desired level of dimming while the pulsesthemselves carry the modulated signal in the form of asquare wave. The modulated signal is transmitted dur-ing the pulse, and the LED operates at the full brightnessduring the pulse. The data rate of the modulated signalshould be adjusted based on the dimming requirement.

13

Mapping bits to QAM symbols

DMT modulation

PWM

Bitstream

LED

x(t)

p(t)

y(t)=x(t)p(t)

50% dimming with PWM

TPWM

TON

t

y(t)=x(t)p(t)p(t)

DMT+PWM signal

t

Fig. 13: Transmitter block diagram of DMT transmitterwith dimming control (top); An example of how 50%

PWM-controlled dimming signal can be combined witha DMT signal as proposed in [44] (bottom); Figures

reproduced from [44]

Authors in [45] showed that any dimming level from0% to 100% can be obtained with high PWM frequency.One benefit of PWM is that it achieves the dimmingwithout changing the intensity level of pulses, hence itdoes not incur the color shift (like OOK with redefinedON/OFF levels) in the LED. The limitation of PWMis its limited data rate (4.8 kbps in [45]). To overcomethis limitation, [44] proposed to combine PWM withDiscrete Multitone (DMT) for joint dimming control andcommunication. The approach decouples the dimmingbased on PWM and communication based on DMT onthe transmitter side. As shown in Fig. 13, the bitstreamis divided and mapped to symbols using QuadratureAmplitude Modulation (QAM). These QAM symbolsare transmitted on different DMT subcarriers that arespaced by 1/T in frequency where T is the durationof one symbol. The DMT signal x(t) is combined withPWM square wave signal p(t) where the duty cycle isdependent on desired level of dimming. The resultantsignal y(t) = x(t)p(t) is shown in Fig. 13. It was alsoshown that dimming constraint limits the achievablethroughput due to high Bit Error Rate (BER). Authorsin [46] also used QAM on DMT subcarriers to achieve alink rate of 513 Mbps, however, it does not address theissue of LED dimming.

Pulse Position Modulation (PPM): Another pulsemodulation method in visible light communication isbased on the pulse position. In PPM, the symbol du-ration is divided into t slots of equal duration, anda pulse is transmitted in one of the t slots. The po-sition of the pulse identifies the transmitted symbol.Due to its simplicity, many early designs [47], [48] ofoptical wireless systems adapted PPM for modulation.In some of the early works of using PPM for infraredcommunication, authors in [49] proposed the use of rateadaptive transmission scheme where repetition coding isapplied to gracefully reduce the throughput in presence

1

0 1

0 1

S1 S2 S3 S4

S1 S2 S3

PWM

PPM

VPPM

OPPM

MPPM . . .

0

Fig. 14: Schematic diagram showing difference betweenPulse Width Modulation (PWM), Pulse Position

Modulation (PPM), Variable Pulse Position Modulation(VPPM), Overlapping Pulse Position Modulation

(VPPM) and Multipulse Pulse Position Modulation(MPPM); Sn refers to nth symbol.

of poor channel conditions. Authors in [50] designeda rate-variable punctured convolutional coded PPM forinfrared communication. Such a scheme adapts the mod-ulation order of PPM and the code rate of puncturedconvolutional codes based on the channel conditions.For even worse channel conditions, [51] proposed touse rate adaptive PPM transmission with both repeatedand punctured convolutional codes to achieve higher bitrate.

Due to the limitations of lower spectral efficiency anddata rate of PPM (only one pulse per symbol duration),other variants of pulse position-based modulation havebeen proposed over time. A generalization of PPM isreferred as Overlapping PPM (OPPM) which allowsmore than one pulse to be transmitted during the symbolduration [48] and the different pulse symbols can beoverlapping (see Fig. 14). [52] showed that OPPM cannot only achieve a higher spectral efficiency comparedto PPM and OOK but a wide range of dimming levelscan be obtained along with the high data rate. Anothergeneralization of PPM was proposed by [53] which isa scheme referred as Multipulse PPM (MPPM). LikeOPPM, it allows multiple pulses to be transmitted dur-ing the symbol duration, however, the pulses within asymbol duration do not have to be continuous (Fig. 14).It was shown in [48] that MPPM can achieve a higherspectral efficiency compared to OPPM.

Authors in [54] proposed a variation of PPM that com-bines OPPM and MPPM in a scheme called OverlappingMPPM (OMPPM). In OMPPM, more than one pulsepositions are allowed for each optical pulse. It showsthat OMPPM can improve the spectral efficiency ofMPPM without the expansion of bandwidth in noiselessphoton counting channel. Further performance analysis

14

for noisy channels was presented in [55]. It was shown in[56] that OMPPM with fewer pulse slots and more pulsesper symbol duration has better cutoff rate performance.Moreover, Trellis-coded OMPPM was studied in [57],[58] to show its effectiveness in direct detection channelswith background noise. In another set of modulationscheme, Differential PPM (DPPM) was proposed in [59].DPPM is similar to PPM except that the OFF symbolsafter the pulse in a PPM symbol are deleted and thenext symbol starts right after the pulse of the previoussymbol. It was shown in [60] that DPPM requires signif-icantly less average power than PPM for a given band-width in an optical communication channel. Authors in[61] proposed Differential Overlapping PPM (DOPPM)where differential deletion of OFF symbols is appliedto OPPM, and showed that it achieves better spectralefficiency and cutoff performance than PPM, DPPM andOPPM.

Authors in [62] proposed EPPM (Expurgated PPM)where symbols in the MPPM are expurgated to maxi-mize the inter-symbol distance. EPPM achieves the samespectral efficiency as PPM, however, it can be used inVLC to provide dimming support as it can achievearbitrary level of illumination by changing the numberof pulses per symbol (code-weight) and the length of thesymbol (code-length) [63]. With many pulses in a sym-bol, EPPM can also mitigate the flickering as comparedto PPM. MEPPM (Multi-level EPPM) [64] extends theEPPM design with support to multiple amplitude levelsin order to increase the constellation size and spectralefficiency. MEPPM can also support the dimming andprovides flicker-free communication. IEEE 802.15.7 [8]standard proposes a pulse modulation scheme calledVariable PPM (VPPM) which is a hybrid of PPM andPWM. In VPPM, the bits are encoded by choosing dif-ferent position of pulse as in PPM, however, the width ofthe pulse can also be modified as needed. VPPM retainsthe simplicity and robustness of PPM while allowingdifferent dimming levels by altering the pulse width.

3.2.3 Orthogonal Frequency Division Multiplexing(OFDM)One limitation of previously discussed single-carriermodulation schemes is that they suffer from high inter-symbol interference due to non-linear frequency re-sponse of visible light communication channels. OFDMhas been widely adopted in the RF communicationdue to its ability to effectively combat the inter-symbolinterference and multipath fading. Authors of [65] firstproposed the use of OFDM for visible light communi-cation. In OFDM, the channel is divided into multipleorthogonal subcarriers and the data is sent in parallelsub-streams modulated over the subcarriers. OFDM forVLC can reduce the inter-symbol interference and doesnot require complex equalizer, however, there are multi-ple challenges in realizing its implementation. First, theOFDM technique for RF needs to be adapted for appli-cation in IM/DD systems such as VLC. This is because

OFDM generates complex-valued bipolar signals whichneed to converted to real-valued signals. This can beachieved by enforcing Hermitian symmetry constrainton the sub-carriers and then converting the time-domainsignals to unipolar signals.

Depending on how the bipolar signals are convertedto unipolar, there are two types of OFDM techniques: (1)Asymmetrically-Clipped Optical OFDM (ACO-OFDM)and (2) DC-biased Optical OFDM (DCO-OFDM). InACO-OFDM, only odd subcarriers are modulated [66]which automatically leads to symmetric time domainsignal. While in DCO-OFDM [65], [67], [68], all sub-carriers are modulated but a positive direct current isadded to make the signal unipolar. [69] presented acomparison of both the OFDM schemes and showedthat LED clipping distortion is more significant in DCO-OFDM compared to ACO-OFDM. The biggest challengein OFDM VLC system is the non-linearity of LED [70]which is that the relationship between the current andthe emitted light of the LED is non-linear. This especiallyaffects the OFDM-based VLC systems which have higherPeak-to-Average Power Ratio (PAPR). The effect of thisnon-linearity was studied in [71], [72] and a solutionwas proposed to combat it by operating the LED in asmall range where the driving current and optical powerare quasi-linear. Apart from the non-linearity, there isonly a limited support for dimming [73] in OFDM-basedmodulation schemes. Despite these challenges, OFDMfor VLC holds great potential with achievable link ratesin the scale of multiple gbps [74], [75] using only singleLED.

3.2.4 Color Shift Keying (CSK)To overcome the lower data rate and limited dim-ming support issues of other modulation schemes, IEEE802.15.7 standard [8] proposed CSK modulation whichis specifically designed for visible light communication.CSK has attracted increasing amount of attention fromresearch community in last couple of years [76]–[81].As we discussed before, generating white light usingblue LED and yellow phosphorus slows down the fastswitching ability of LED and hinders high data ratecommunication. An alternative way to generating whitelight which is recently becoming more and more popularis to utilize three separate LEDs - Red, Green and Blue(RGB). This combined source with RGB LEDs is oftenreferred as TriLED (TLED). CSK modulates the signalusing the intensity of the three colors in the TLED source.

CSK modulation relies on the color space chromaticitydiagram as defined by CIE 1931 [18] (see Fig. 15). Thechromaticity diagram maps all colors perceivable byhuman eye to two chromaticity parameters - x and y. Theentire human visible wavelength is divided into sevenbands as shown in Table 3 and their centers are markedin Fig. 15. Based on the diagram, the CSK modulation[8], [81] is performed as follows

1) Determine RGB constellation triangle: The con-stellation triangle is decided based on the cen-

15

001

000

010

011

100101110

Constellation triangle

(a)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

y

x

4CSK Symbols

(b)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

y

x

8CSK Symbols

(c)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

y

x

16CSK Symbols

(d)

Fig. 16: (a) RGB constellation triangle (110, 010, 000) (b-d) Symbols of 4-CSK, 8-CSK and 16-CSK.

001

000

010

011

100101110

Fig. 15: CIE 1931 Chromaticity Diagram; The sevencolor codes correspond to the centers of seven bands

dividing the visible spectrum as shown in Table 3;reproduced from [82].

Band (nm) Code Center (nm) (x, y)380-478 000 429 (0.169, 0.007)478-540 001 509 (0.011, 0.733)540-588 010 564 (0.402, 0.597)588-633 011 611 (0.669, 0.331)633-679 100 656 (0.729, 0.271)679-726 101 703 (0.734, 0.265)726-780 110 753 (0.734, 0.265)

TABLE 3: The seven bands used in CSK and their code,center and chromaticity coordinates.

ter wavelength of the three RGB LEDs used inthe TLED source. Table 4 shows the valid colorband combinations as proposed by [81] that canbe chosen as the constellation triangle dependingon the central wavelength of the RGB LEDs. Forthe purpose of illustration, let us assume that wechoose the CSK constellation triangle to be (110,010, 000) as shown in Fig. 16a (example adaptedfrom [81]).

2) Mapping data bits to chromaticity values: De-pending on 4CSK, 8CSK or 16CSK being used,the chromaticity values of symbols can be derivedfrom the constellation triangle. For our example,Figs. 16b, 16c and 16d show how data bits can

Band i Band j Band k1 110 010 0002 110 001 0003 101 010 0004 101 001 0005 100 010 0006 100 001 0007 011 010 0008 011 001 0009 010 001 000

TABLE 4: Valid color band combinations that can bechosen for building the constellation triangle for CSK.

be represented using the symbols for 4CSK, 8CSKand 16CSK. Determining the position of the sym-bols in the constellation design requires solving anoptimization problem where the distance betweenthe symbols should be maximized to minimizethe inter-symbol interference. Note that there is anadditional constraint in the problem which ensuresthat the symbols should be equally distributed inthe triangle so that the combined light emittedwhen transmitting different symbols is perceivedby the human eye to be white light only. Theoptimization problem has been studied in [76]–[79]as we discuss next. Once the symbol coordinatesare decided, each symbol is assigned a bit sequence(e.g. in 4CSK, the 4 symbols are assigned 00, 01, 10and 11 respectively), which is then used to map theincoming bits to the symbols.

3) Determine the intensities of RGB LEDs: Thesymbols are transmitted by varying the intensitiesof the RGB LEDs. The individual intensities ofthe three LEDs (Pi, Pj and Pk) for each symbol iscalculated by solving the following equations:

xs = Pixi + Pjxj + Pkxk (27)ys = Piyi + Pjyj + Pkyk (28)

Pi + Pj + PK = 1 (29)

where xs and ys are the chromaticity values of thesymbol (Fig. 16), and (xi, yi), (xj , yj) and (xk, yk) arethe chromaticity values of the central wavelength of theRGB LEDs being used (three points of the constellationtriangle). The receiver uses the R, G and B intensities to

16

decode the transmitted signal.Dimming support in CSK is simply amplitude dim-

ming where the driving current of the LEDs is variedto change the brightness of resultant white light. Also,different from OOK and pulse modulations, flickeringis not a problem with CSK since no amplitude varia-tion is employed. Due to these advantages, researchershave recently attempted to improve the CSK scheme ofIEEE 802.15.7 by designing its generalized forms witharbitrary constellation. Authors in [76] presented a CSKconstellation design technique based on Billards equiv-alent disk packing algorithm. Similarly, [77] and [78]developed similar techniques with the use of differentoptimization algorithms such as interior point methods.All the constellation design techniques are designed tomeet the color balance requirement where the TLEDsource is required to produce any desired color for illu-mination. The use of four LEDs (blue, cyan, yellow andred) was suggested in [79]. With four LEDs, it is possibleto achieve a quadrilateral constellation shape that allowsQAM-like constellation design. The presented system isshown to be more energy efficient as well as reliable (lessinter-symbol noise) compared to the conventional CSKwith 3 LEDs.

The RGB tri-LED can also be used to implement Wave-length Division Multiplexing (WDM) - a multiplexingtechnique commonly used in fiber optics communica-tion. Authors in [83] proposed modulating separate datastreams on three colors which together multiplex towhite light. With the use of DMT, an aggregate data rateof 803 Mbps was shown to be achievable using singleRGB LED in [83]. Authors in [84] proposed the use ofcarrier-less amplitude and phase modulation on WDMVLC system with RGB LED to achieve a data rate of 3.22Gbps.

IEEE 802.15.7 Physical Layer IEEE 802.15.7 [8] stan-dard has specified three PHY layers for VLC with atotal of 30 MCS (Modulation and Coding Scheme) in-dexes. These MCS levels are shown in Tables 5, 6 and7. Both PHY I and PHY II utilize OOK and VPPMfor modulation. PHY I utilizes Reed Solomon (RS) andConvolutional Codes (CC) for Forward Error Correction(FEC), while PHY II and III mostly reply for RS codesonly for FEC.

As described in [80], “optical clock rate” is an impor-tant parameter for the performance of the PHY layers.PHY 1 utilizes lower optical rate of ≤ 400KHz. This isbecause PHY 1 is designed to be usable in outdoor sce-narios as well where the LED transmitters are typicallyhigh-power and can switch the intensity at a slower rate.PHY II is designed to be used indoors where the opticalswitching rate can be as high as 120 MHz. The opticalrate is 24 MHz for PHY III which is the current feasibleswitching rate for white TriLED.

Depending on the choice of modulation, RLL code,optical clock rate, FEC code, the three PHY modes canprovide different data rates. PHY I can provide data ratesfrom 11.67 Kbps upto 266.6 Kbps. PHY II can achieve

data rates from 1.25 Mbps upto 96 Mbps. PHY III canyield data rates starting from 12 Mbps upto 96 Mbps.Further details of physical layer of IEEE 802.15.7 areprovided in [80].

Table 8 provides a comparison between four majormodulations schemes proposed for VLC. It can be ob-served that OFDM and CSK are more suitable for highdata-rate applications in VLC access networks. As wewill discuss next, OFDM is also more suitable for VLCMIMO design, however, more research is necessary toensure dimming support in OFDM. Another advantageof CSK is that it can provide multi-user access throughwavelength multiplexing as we will discuss in Section 4.Increasing demand of higher data-rates is likely to drivefurther research and development of OFDM and CSKfor VLC-based access networks.

3.3 Multiple Input Multiple Output (MIMO)In order to provide sufficient illumination, most of theluminaires typically contain multiple LEDs. These mul-tiple LEDs can be treated as multiple transmitters thatcan enable visible light MIMO communication. In RFcommunications, MIMO systems are commonly used(in IEEE 802.11n, Long-Term Evolution - LTE) to obtainhigher data rates. Similarly, multiple LEDs can be usedfor higher spectral efficiency in VLC. Note that there arecertain similarities between the VLC MIMO systems dis-cussed in this section and screen-camera links (discussedin Section 6.2) as both of them can use an image sensoras a MIMO receiver. The difference is that unlike smart-phone screens, the multiple LED transmitters consideredhere are also used for the illumination. We will providefurther details of the screen-camera links in Section 6.2.

MIMO systems in VLC are difficult to realize com-pared to RF communications. In RF MIMO systems, thethroughput gains are largely attributed to spatial diver-sity (existence of multiple spatial paths that are diversein nature). However, such diversity gains are limitedin VLC MIMO because paths between the transmitterand receiver are very similar (less diverse) especiallyin indoor scenarios. This limits the available spatialdiversity of VLC MIMO systems. The other challenge inVLC MIMO is the design of the receiver as we discussnext.

3.3.1 MIMO ReceiverAs we discussed in Section 2, there can be two typesof receivers in VLC MIMO systems - photodiode andimage sensor. The performance of the system depends onwhether imaging (image sensor) or non-imaging (photo-diode) receiver is used [85].

Non-imaging receiver in a MIMO system is a set ofindependent photodiodes each with its individual con-centrator optics. The advantage of such a receiver isthat a very high gain can be achieved due to narrowFOV of each photodiode. The disadvantage, however, isthat such a receiver requires careful alignment with the

17

MCS Modulation RLL code Opticalclock rateFEC Data rate (kbps)Outer code (RS) Inner code (CC)

0

OOK Manchester 200 KHz

(15,7) 1/4 11.671 (15,11) 1/3 24.442 (15,11) 2/3 48.893 (15,11) none 73.34 none none 1005

VPPM 4B6B 400 KHz

(15,2) none 35.566 (15,4) none 71.117 (15,7) none 124.48 none none 266.6

TABLE 5: 802.15.7 PHY I operating mode specifications and achievable throughput.

MCS Modulation RLL code Opticalclock rate FECData rate(Mbps)

16

VPPM 4B6B

3.75 MHz RS(64,32) 1.2517 RS(160,128) 218

7.5 MHzRS(64,32) 2.5

19 RS(160,128) 420 none 521

OOK 8B10B

15 MHz RS(64,32) 622 RS(160,128) 9.623 30 MHz RS(64,32) 1224 RS(160,128) 19.225 60 MHz RS(64,32) 2426 RS(160,128) 38.427

120 MHzRS(64,32) 48

28 RS(160,128) 76.829 none 96

TABLE 6: 802.15.7 PHY II operating mode specifications and achievable throughput.

MCS Modulation Optical clock rate FEC Data rate (Mbps)32 4 CSK 12 MHz RS(64,32) 1233 8 CSK RS(64,32) 1834 4 CSK

24 MHz

RS(64,32) 2435 8 CSK RS(64,32) 3636 16 CSK RS(64,32) 4837 8 CSK none 7238 16 CSK none 96

TABLE 7: 802.15.7 PHY III operating mode specifications and achievable throughput.

Modulation Data Rate Dimming Support Flickering Issue CommentsOOK Low to moderate Yes High Low-complexity transceiver designPPM Moderate Yes Low Maximum spectral efficiency with MEPPM

OFDM High No Low Complex design due to LED non-linearity, MIMO supportCSK High Yes Low Requires RGB tri-LED, improved multi-user access

TABLE 8: Major modulation schemes and their characteristics

transmitters because of the narrow FOV, and the capacitycan reduce dramatically even with minor misalignment.

Imaging Receiver: Since an image sensor contains aprojection lens and a large matrix of photodiodes, ithas the potential to create a high data-rate MIMO link.The projection lens ensures a large FOV which nearlyeliminates the alignment requirement. The disadvantageof such as a receiver is that individual photodiodes havelimited gain and advance image processing is requiredto create an efficient MIMO channel. Also, the samplingrate of the image sensor is comparatively lower furtherreducing the achievable throughput.

The channel models of both imaging and non-imagingreceiver MIMO, and their relative benefits and limita-tions were presented in [85]. It was shown in [86] thatan “ideal” MIMO receiver can be a hybrid of imagingand non-imaging sensors which can achieve high gains

of LOS paths using narrow FOV like photodiodes andcan be robust by leveraging non-LOS paths wheneverneeded like an image sensor. Authors in [87] proposedthe design of a spherically-shaped receiver that is madeof a large number of photodiodes. Each of the photodi-ode has a narrow FOV and points in different directionin the room. The photodiodes pointing to transmitterLED can receive the signal with high gain while otherphotodiodes pointing to other directions can establishnon-LOS channels to increase spatial diversity. However,using such a receiver incurs cost for additional hardware.Instead, authors in [88] proposed a way to improve thelower sampling rate of the image sensor. A token-basedpixel selection method was proposed where instead ofconventional row-scanning approach, only the pixels ofinterest are selectively scanned to improve the samplingrate.

18

3.3.2 VLC MIMO TechniquesThere are three types of VLC MIMO techniques pro-posed in literature [89].

Repetition Coding (RC): This is the simplest tech-nique where the same signal is transmitted from all thetransmitters. The transmitted signal from all LEDs meetconstructively at the receivers increasing the overall gain.

Spatial Multiplexing (SMP): In SMP, different datais transmitted from each transmitter to a receiver pho-todiode. With multiple transmitters and receivers, thistype of MIMO creates multiple parallel SISO streams.The challenge is that receiver photodiodes have to beaccurately aligned to the transmitters to avoid any inter-channel interference. SMP MIMO for optical channelshas been studied in some of the early works [90]–[92].In [90], [92], authors proposed optical wireless MIMOcommunication with subcarrier multiplexing where zeroforcing was utilized to cancel the interference from othertransmit antennas. It was shown that for the transmittersemi-angle more than 20o, the transmitter-receiver sepa-ration should be more than 1.5 meters for lower BER. Theimpact of optical beat interference on OMIMO schemeof [90] was studied in [91]. Optical beat interference isthe signal degradation caused by multiple transmitterstransmitting simultaneously on nearby wavelengths.

Spatial Modulation (SM): This MIMO technique wasproposed by [93]–[96] where only one transmitter trans-mits data at any point of time. The constellation dia-gram is extended to include the spatial dimension. Eachtransmitter LED is assigned a specific symbol and whendata bits to be transmitted matches the symbol, theLED is activated. The receiver estimates which LED wasactivated based on the received signal, and uses this todecode the transmitted data. Since the data is encodedin both spatial and signal domain, SM achieves muchhigher spectral efficiency compared to other techniques.

A comparison of all the three MIMO techniques wereprovided in [89]. It was shown that RC is less restrictivein terms of its requirement for transmitter-receiver align-ment but provides only a limited spectral efficiency. SMP,on the other hand, requires more careful alignment oftransmitter-receiver but also provides higher data ratescompared to RC. SM achieves the best of both worldsby being robust to correlated channels and providinghigher spectral efficiency. Also, it was shown in [97]that imaging receivers can obtain much higher SNRwhen using SM or SMP technique compared to the non-imaging receivers.

Due to its advantages over other MIMO techniques,SM has been studied further in recent years. It wasshown in [98], [99] that power imbalance between thetransmitter LEDs can improve the performance of spatialmodulation especially when optical paths between thetransmitter and receiver are highly correlated. Authorsin [100] studied the performance of spatial modulationusing an implementation of 4 × 4 MIMO system andshowed that the challenge in achieving higher through-put with SM is to maintain symbol separation in the

constellation from the receiver’s perspective. Researchersinvestigated the performance of spatial modulation in[101] when only partial channel state information (CSI)is available and concluded that highly accurate CSIestimation is necessary to realize the full potential of SM.The use of generalized spatial modulation was proposedin [102], [103]. Such modulation extends the originalscheme by allowing more than one transmitter to beactive during the a symbol duration. It was shown thatdue to additional flexibility of activating multiple LEDs,the generalized scheme can achieve higher spectral effi-ciency compared to the conventional scheme, however,at the cost of additional complexity in constellationdesign.

Optical MIMO for non-LOS diffuse links has not re-ceived much attention. Authors in [104] showed howbackward spatial filter can be used for optical wirelessMIMO in diffuse channels (no precise alignment oftransmitter and receiver). With user movements, suchdiffuse channel are more likely in practical scenarios andoptimizing MIMO performance for such channels shouldbe investigated further.

3.3.3 Optical Beamforming

Beamforming allows multiple transmitters to concen-trate their signal in a specific direction based on thereceiver location. This type of transmit beamforming iswell studied in RF communication and also utilized byrecent WLAN standards such as IEEE 802.11ac. Similarto RF beamforming, emitting light from multiple LEDscan be focused towards the receiver to create opticalbeamforming. Recently, it was shown in [105] how lightemitted from a single LED can be focused in a specifictarget direction using Spatial Light Modulator (SLM).SLM is an additional device that is required to modu-late the phase or amplitude of the visible light signal.It was shown that significant SNR improvements canbe achieved by using the optical beamforming withany modulation technique. Authors in [106] derivedthe transmit beamforming vectors when multiple LEDsare used to perform the optical beamforming. Opticalbeamforming can improve the performance of a visiblelight communication link significantly, however, thereis only a limited amount of research done towardsthis. Performing optical beamforming while meeting theillumination constraints is an important direction forresearch in VLC MIMO systems.

4 LINK LAYER

When there exists multiple transmitter LEDs and re-ceiver devices connected to them, it is essential to controlthe medium access, device association and device mobil-ity. In this section, we provide an overview of differenttechniques proposed in literature to manage link layerservices.

19

Peer-to-peer Star Broadcast

LED Photodiode

Fig. 17: VLC link layer topologies; reproduced from [8]

4.1 Medium Access Control (MAC)The application scenarios of VLC can be used to identifythe link layer topologies that need to be supported by theMAC protocols. IEEE 802.15.8 [8] proposes three typesof link layer topologies for VLC as shown in Fig. 17 -

1) Peer-to-peer: The peer-to-peer topology involvesone device acting as a coordinator (or master) forthe link between two devices. Both devices cancommunicate with each other since the client hasan uplink to the master. This topology is typicallymore suitable for high-speed Near-Field Commu-nication (NFC).

2) Star: In a start topology, there can be many clientdevices connected to a master device which acts asthe coordinator. A typical use case of this topologyis VLC wireless access networks. The MAC designis especially challenging in the star topology dueto many bi-directional links in the same collisiondomain.

3) Broadcast: Different from the star topology, theclient devices in a broadcast topology can onlyreceive data from the master LED transmitter with-out forming any uplink. Such topology can be usedfor broadcasting information via LEDs throughoutthe network. Since there is no explicit associationneeded, the broadcast topology simplifies the MACdesign.

Three types of multiple access control (MAC) schemesare proposed for VLC - Carrier Sense Multiple Ac-cess (CSMA), Orthogonal Frequency Division MultipleAccess (OFDMA) and Code Division Multiple Access(CDMA).

CSMA: There are two types of random channel accessmechanisms proposed by IEEE 802.15.7 standard. In thefirst type, the beacons from the coordinator are disabled.Such beacon-disabled random access uses an unslottedrandom channel access with CSMA. Here, if a devicewishes to transmit, it first waits for a random back-offperiod and then senses the channel to be busy or not,before transmitting. If the channel is found to be busy,the device waits for another random period before tryingto access the channel again. In the second type wherethe beacons are enabled, the time is divided into beaconintervals. A superframe within the beacon interval con-tains Contention Access Periods (CAP) and ContentionFree Periods (CFP) as shown in Fig. 18(a). If a devi