UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Ethernet over light Mark, Philip 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2015_february_mark_philip.pdf [ 2.54MB ]
JSON: 24-1.0167659.json
JSON-LD: 24-1.0167659-ld.json
RDF/XML (Pretty): 24-1.0167659-rdf.xml
RDF/JSON: 24-1.0167659-rdf.json
Turtle: 24-1.0167659-turtle.txt
N-Triples: 24-1.0167659-rdf-ntriples.txt
Original Record: 24-1.0167659-source.json
Full Text

Full Text

Ethernet over LightbyPhilip MarkB.Eng., McGill University, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF APPLIED SCIENCEinThe Faculty of Graduate and Postdoctoral Studies(Electrical and Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)December 2014c© Philip Mark 2014AbstractThe advent of high-brightness, fast-switching Light Emitting Diodes (LEDs)has facilitated Visible Light Communication (VLC) as a new form of OpticalWireless Communication (OWC) over the visible light spectrum. In VLC, theseLEDs serve a dual purpose of communication on top of general illumination.The biggest challenge facing VLC lies in finding the “killer application” thatwill propel the technology to widespread adoption. One of the ways we believethis can be achieved is by integrating VLC with pre-existing Ethernet Local AreaNetwork (LAN) backbones. Although there has been some preliminary researchin this area, specifically involving 10Base-T over VLC, none have explicitly dealtwith Fast Ethernet (100 Mbps) over VLC. In this thesis, we investigate theimplementation of analog transmission of 100Base-TX over VLC in an amplify-and-forward approach, which we coin as Ethernet over Light (EoL). We presentthe design of a VLC transmitter and accompanying receiver developed for EoL,and include a comprehensive channel model to analyze this Ethernet-VLC link.Equalization techniques were explored to overcome the various shortcomingsassociated with the EoL channel, and to improve the performance of the overallsystem. This VLC-LAN solution proved to be a viable alternative for providinga wireless broadcast link wherever LEDs would be deployed for illumination inan indoor setting.iiPrefaceThis thesis is original, unpublished, independent work by the author, P. Mark.The hardware design in Chapter 7 was done by B. Stacy.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Recent Advances . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2 Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Outline of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.1 IEEE 802.3 Ethernet . . . . . . . . . . . . . . . . . . . . . . . . 72.1.1 100Base-TX (Fast Ethernet) . . . . . . . . . . . . . . . . 72.1.2 Ethernet Cables for 100Base-TX . . . . . . . . . . . . . . 112.1.3 Power-over-Ethernet . . . . . . . . . . . . . . . . . . . . . 122.2 Light Emitting Diodes . . . . . . . . . . . . . . . . . . . . . . . . 142.2.1 Types of LEDs . . . . . . . . . . . . . . . . . . . . . . . . 152.2.2 LED Model . . . . . . . . . . . . . . . . . . . . . . . . . 182.2.3 LED Nonlinearity . . . . . . . . . . . . . . . . . . . . . . 192.3 Visible Light Communication . . . . . . . . . . . . . . . . . . . . 202.3.1 A Brief History . . . . . . . . . . . . . . . . . . . . . . . 202.3.2 Basics of VLC . . . . . . . . . . . . . . . . . . . . . . . . 212.3.3 Benefits, Trade-offs and Challenges . . . . . . . . . . . . 222.3.4 Photodiode Receiver . . . . . . . . . . . . . . . . . . . . 23ivTable of Contents3 Ethernet over Light Proposal . . . . . . . . . . . . . . . . . . . . 253.1 Our Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263.2.1 Proposed Transmitter . . . . . . . . . . . . . . . . . . . . 273.2.2 Proposed Receiver . . . . . . . . . . . . . . . . . . . . . . 293.3 Advantages, Drawbacks and Challenges . . . . . . . . . . . . . . 304 Channel Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 334.1 Category 5 UTP Channel . . . . . . . . . . . . . . . . . . . . . . 334.1.1 Transmit Jitter . . . . . . . . . . . . . . . . . . . . . . . 334.1.2 Waveform Overshoot . . . . . . . . . . . . . . . . . . . . 344.1.3 Baseline Wander . . . . . . . . . . . . . . . . . . . . . . . 344.1.4 Cable Attenuation . . . . . . . . . . . . . . . . . . . . . . 354.1.5 Flat Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 374.1.6 Return Loss . . . . . . . . . . . . . . . . . . . . . . . . . 384.1.7 Overall Cable Model . . . . . . . . . . . . . . . . . . . . 404.2 VLC Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404.2.1 Wall Reflection Model . . . . . . . . . . . . . . . . . . . . 424.2.2 Spherical Model . . . . . . . . . . . . . . . . . . . . . . . 464.2.3 Hayasaka-Ito Model . . . . . . . . . . . . . . . . . . . . . 474.2.4 Ceiling Bounce Model . . . . . . . . . . . . . . . . . . . . 484.2.5 Gfeller & Bapst Model . . . . . . . . . . . . . . . . . . . 485 Link Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495.1 Signal Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495.1.1 Ethernet Transmitter . . . . . . . . . . . . . . . . . . . . 495.1.2 Cat 5 Cable Channel . . . . . . . . . . . . . . . . . . . . 495.1.3 VLC Front-end (LED Driver) . . . . . . . . . . . . . . . 505.1.4 LED Transmitter . . . . . . . . . . . . . . . . . . . . . . 505.1.5 VLC Channel . . . . . . . . . . . . . . . . . . . . . . . . 515.1.6 VLC Receiver Front-end . . . . . . . . . . . . . . . . . . 515.1.7 Received Signal . . . . . . . . . . . . . . . . . . . . . . . 535.2 Noise Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535.2.1 Noise in the Cat 5 Channel . . . . . . . . . . . . . . . . . 545.2.2 Noise in the VLC Channel . . . . . . . . . . . . . . . . . 555.2.3 Other Electronic Thermal Noise . . . . . . . . . . . . . . 585.2.4 Total Noise . . . . . . . . . . . . . . . . . . . . . . . . . . 595.3 EoL SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605.4 Link Budget Numerical Example . . . . . . . . . . . . . . . . . . 605.4.1 Estimated SNR Values in Typical EoL Configurations . . 656 Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696.1 ISI in EoL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696.2 Equalization Methods . . . . . . . . . . . . . . . . . . . . . . . . 716.2.1 Linear Equalizers . . . . . . . . . . . . . . . . . . . . . . 726.2.2 Decision Feedback Equalizers . . . . . . . . . . . . . . . . 72vTable of Contents6.2.3 Adaptive Equalizers . . . . . . . . . . . . . . . . . . . . . 746.2.4 Equalization for EoL . . . . . . . . . . . . . . . . . . . . 756.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . 766.3.1 Effect of Cable Length and SNR . . . . . . . . . . . . . . 776.3.2 Effect of LED Modulation Bandwidth and SNR . . . . . 806.3.3 Effect of Receiver Location . . . . . . . . . . . . . . . . . 817 Demonstrator Design . . . . . . . . . . . . . . . . . . . . . . . . . 877.1 Transmitter Circuit Design . . . . . . . . . . . . . . . . . . . . . 887.2 Receiver Circuit Design . . . . . . . . . . . . . . . . . . . . . . . 897.3 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 928.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 938.1.1 Uplink . . . . . . . . . . . . . . . . . . . . . . . . . . . . 938.1.2 1000Base-T over VLC . . . . . . . . . . . . . . . . . . . . 94Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95viList of Tables2.1 4B/5B encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . 93.1 Ethernet parameters . . . . . . . . . . . . . . . . . . . . . . . . . 274.1 Cable model parameters . . . . . . . . . . . . . . . . . . . . . . . 405.1 Sample power transfer analysis . . . . . . . . . . . . . . . . . . . 625.2 Sample noise analysis . . . . . . . . . . . . . . . . . . . . . . . . 635.3 Background current values from ambient sources . . . . . . . . . 655.4 SNR simulation parameters . . . . . . . . . . . . . . . . . . . . . 675.5 Typical average SNR levels obtained . . . . . . . . . . . . . . . . 676.1 100Base-TX DFE properties . . . . . . . . . . . . . . . . . . . . . 756.2 BER simulation parameters . . . . . . . . . . . . . . . . . . . . . 776.3 Optimal equalizer performance at 10 m . . . . . . . . . . . . . . . 806.4 Optimal equalizer performance at 100 m . . . . . . . . . . . . . . 807.1 List of components used . . . . . . . . . . . . . . . . . . . . . . . 87viiList of Figures1.1 Global mobile data traffic trend [1] . . . . . . . . . . . . . . . . . 21.2 Installed base estimates of various LED applications in the UnitedStates [5] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.1 OSI reference model and IEEE 802.3 protocol stack . . . . . . . . 82.2 100Base-TX PHY [16] . . . . . . . . . . . . . . . . . . . . . . . . 82.3 MLT-3 equivalent models . . . . . . . . . . . . . . . . . . . . . . 102.4 MLT-3 signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.5 PoE procedure [22] . . . . . . . . . . . . . . . . . . . . . . . . . . 132.6 Basic structure of an LED . . . . . . . . . . . . . . . . . . . . . . 142.7 Yellow-phosphor LED structure . . . . . . . . . . . . . . . . . . . 152.8 Typical phosphor LED properties . . . . . . . . . . . . . . . . . . 162.9 Typical RGBLED properties . . . . . . . . . . . . . . . . . . . . 172.10 OLED structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.11 Lambertian emission pattern . . . . . . . . . . . . . . . . . . . . 182.12 LED nonlinear characteristics . . . . . . . . . . . . . . . . . . . . 192.13 Second-order polynomial P-I curve fit for AVAGO HFBR-1521ZRCLED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.14 Intensity Modulation/Direct Detection . . . . . . . . . . . . . . . 213.1 Ethernet over Light concept . . . . . . . . . . . . . . . . . . . . . 263.2 Proposed EoL high-level system model . . . . . . . . . . . . . . . 273.3 Typical analog LED driver configuration [35] . . . . . . . . . . . 283.4 Typical analog receiver front-end configuration . . . . . . . . . . 304.1 Cat 5 cable model . . . . . . . . . . . . . . . . . . . . . . . . . . 344.2 Worst case baseline wander . . . . . . . . . . . . . . . . . . . . . 354.3 Transmission line model . . . . . . . . . . . . . . . . . . . . . . . 364.4 Cat 5 cable attenuation by cable length . . . . . . . . . . . . . . 374.5 Comparison of the worst case insertion loss (4.18) versus the limit(4.19) at 100 m . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384.6 Frequency response of individual effects in a 100 m cable . . . . . 414.7 VLC IM/DD equivalent channel model . . . . . . . . . . . . . . . 414.8 Phosphor LED emission spectrum compared to the reflectancespectrum of common room surfaces [86] . . . . . . . . . . . . . . 444.9 Wall reflection channel geometry [86] . . . . . . . . . . . . . . . . 444.10 Room configuration and simulation parameters . . . . . . . . . . 45viiiList of Figures4.11 LOS with one reflection . . . . . . . . . . . . . . . . . . . . . . . 454.12 Spherical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475.1 EoL link model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505.2 EoL noise model . . . . . . . . . . . . . . . . . . . . . . . . . . . 545.3 NEXT and FEXT in twisted pairs . . . . . . . . . . . . . . . . . 545.4 Responsitivity of a Thorlabs PDA10A Si-PIN photodiode com-pared to the spectral power distribution of ambient light sources 565.5 Received PSDs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645.6 Link budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.7 SNR distribution in a room . . . . . . . . . . . . . . . . . . . . . 686.1 100Base-TX eye diagrams . . . . . . . . . . . . . . . . . . . . . . 706.2 Family of equalizers . . . . . . . . . . . . . . . . . . . . . . . . . 716.3 FIR LE block diagram . . . . . . . . . . . . . . . . . . . . . . . . 726.4 DFE model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 736.5 FIR DFE block diagram . . . . . . . . . . . . . . . . . . . . . . . 746.6 BER vs SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796.7 BER vs LED modulation bandwidth vs SNR . . . . . . . . . . . 826.8 BER vs photodiode x-y location - NFF = 30, NFB = 20 . . . . . . 836.9 BER vs cable length vs SNR (Non-adaptive DFE) . . . . . . . . 846.10 BER vs cable length vs SNR (Adaptive LE) . . . . . . . . . . . . 856.11 BER vs cable length vs SNR (Adaptive DFE) . . . . . . . . . . . 867.1 VLC transmitter implementation circuit diagram . . . . . . . . . 887.2 VLC receiver implementation circuit diagram . . . . . . . . . . . 897.3 Bode plot of bootstrapped-TIA . . . . . . . . . . . . . . . . . . . 907.4 Transmitter simulation . . . . . . . . . . . . . . . . . . . . . . . . 917.5 Receiver simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 91ixGlossaryAGC Automatic Gain ControllerAOI Active Output InterfaceAPD Avalanche PhotodiodeAWG American Wire GaugeAWGN Additive White Gaussian NoiseBER Bit Error RateBJT Bipolar-Junction TransistorCSK Color-Shift KeyingCSMA/CD Carrier Sense Multiple Access/Collision DetectDFE Decision Feedback EqualizerDPPM Differential Positioning Pulse ModulationDSP Digital Signal ProcessingELFEXT Equal Level Far-End CrosstalkEMI Electromagnetic InterferenceEoL Ethernet over LightFBF Feedback FilterFCC Federal Communications CommissionFET Field-Effect TransistorFEXT Far-End CrosstalkFFF Feedforward FilterFIR Finite Impulse ResponseFOV Field-of-ViewFPGA Field-Programmable Gate ArrayFSK Frequency Shift KeyingFSM Finite State MachineFSO Free Space OpticalGaAsP Gallium Arsenide PhosphideHPF High-Pass FilterIIR Infinite Impulse ResponseIM/DD Intensity Modulation/Direct DetectionIP Internet ProtocolIR InfraredISI Inter-Symbol InterferenceISO International Organization for StandardizationJEITA Japan Electronics and Information Technology AssociationLAN Local Area NetworkLCL Longitudinal Conversion LossxGlossaryLE Linear EqualizerLED Light Emitting DiodeLMS Least-Mean SquareLNA Low Noise AmplifierLOS Line-of-SightLPF Low-Pass FilterLTI Linear Time-InvariantMAC Media Access ControlMDI Medium-Dependent InterfaceMII Medium-Independent InterfaceMIMO Multiple-Input Multiple-OutputMLT-3 Multi-Level Transmit 3MMSE Minimum Mean-Square ErrorNEXT Near-End CrosstalkNIC Network Interface CardNLOS Non-Line-of-SightNRZ Non-Return-to-ZeroNRZI Non-Return-to-Zero InvertedOFDM Orthogonal Frequency-Division MultiplexingOLED Organic Light Emitting DiodeOMEGA hOME Gigabit AccessOOK On-Off KeyingOSI Open Systems InterconnectionOWC Optical Wireless CommunicationPAM Pulse Amplitude ModulationPD Powered DevicePCS Physical Coding SublayerPHY Physical layerPIN Positive Intrinsic NegativePLC Power-Line CommunicationPPM Pulse Position ModulationPMA Physical Medium AttachmentPoE Power-over-EthernetPSD Power Spectral DensityPSE Power Supplying EquipmentPSELFEXT Power Sum Equal Level Far-End CrosstalkPSNEXT Power Sum Near-End CrosstalkPWM Pulse Width ModulationRCLED Resonant Cavity Light Emitting DiodeRF Radio FrequencyRGB Red Green BlueRGBLED Red Green Blue Light Emitting DiodeRLS Recursive Least SquaresRMS Root Mean SquareSISO Single-Input Single-OutputSNR Signal-to-Noise RatioxiGlossarySSL Solid-State LightingTCL Transverse Conversion LossTCP Transmission Control ProtocolTIA Transimpedance AmplifierTP-PMD Twisted Pair-Physical Medium DependentUDP User Datagram ProtocolUTP Unshielded Twisted PairVGA Variable Gain AmplifierVLC Visible Light CommunicationVLCC Visible Light Communication ConsortiumVoIP Voice over Internet ProtocolWDM Wavelength Division MultiplexingZF Zero ForcingxiiAcknowledgementsFirst and foremost, I would like to thank my adviser Dr. Lutz Lampe for hisvaluable guidance and continuous support during my degree. I would also like toexpress gratitude to the Natural Sciences and Engineering Research Council ofCanada for their generous support of this work. I am appreciative to Dr. SteveHranilovic, my fellow colleagues Blake Stacy, Ayman Mostafa, Hao Ma andmy lab mates for their occasional assistance. As well, I would like to express aheartfelt thank you to all the people I have met in Vancouver whom have helpedme stay sane during my Masters degree. Godspeed to the brave and poor soulswho are about to read this.xiiiI dedicate this thesis to my parents, Gim Sing Mark and Xiao Yi Xu.xivChapter 1IntroductionThe number of smart mobile devices and the demand for broadband wirelessconnectivity have been growing consistently and rapidly, year per year,with no signs of slowing down. A global study was recently conducted in 2014by Cisco in order to determine the current and future trends in mobile wirelesscommunication which includes: the transition to smarter mobile devices, theadoption of the Internet of Everything and the dominance of video streamingtraffic [1]. Last year, in 2013 alone, five hundred million mobile devices (77 % ofthem being smartphones) were added worldwide. In fact, it is expected that thenumber of mobile devices will exceed the total population on Earth in the verynear future. Correspondingly, mobile data traffic also increased 81 % during thatsame year, reaching 1.5 exabytes (1.5× 1018 bytes) per month. This trend isexpected to continue growing exponentially upwards of 16 exabytes per monthin 2018 (11 times the amount from 2013) as seen in Figure 1.1. The studyindicated that more than half of that traffic was attributed to mobile videostreaming alone, and is expected to eventually account for more than two thirdsof mobile data by 2018. These same patterns were observed in Canada froma study conducted by Red Mobile Consulting “Study of Future Demand ForRadio Spectrum in Canada 2011-2015” [2].Wireless carriers are struggling to keep up with these types of demands, asvoice traffic now only accounts for but a fraction of the total data traffic on thewireless network. These expanding traffic demands of Radio Frequency (RF)wireless services has reached a tipping point where users are competing againsteach other for capacity. Current wireless RF technology has also matured toa stage where extracting every last bit of capacity and bandwidth is no longerpractical. As a result, the world is trending towards an inevitable spectrum crisisthat is a real concern for wireless telecoms. In addition to this spectrum scarcityissue, studies observe a traffic asymmetry heavily biased towards the downlinkwith a ratio up to 6:1 compared to the uplink [3]. This is mainly attributedto consumer behavior, with strong preference for downlink multimedia such asvideo streaming, web browsing and content sharing. To accommodate this ever-increasing demand for such bandwidth-intensive applications, an evolution to-wards multi-technology and multi-tiered networks with smaller cell sizes (macro,micro, femto, pico . . . ) is currently being investigated to increase network ca-pacity for widespread deployment. Further, since roughly 70 % of wireless trafficusually occurs indoors [4], developing a new type of indoor wireless communi-cation is essential to offload this last mile traffic from the RF network. This hasled both academia and industry to explore new avenues for short to medium1Chapter 1. Introduction1.5 EB2.6 EB4.4 EB7.0 EB10.8 EB15.9 EBExabytes per month05101520201820172016201520142013Figure 1.1: Global mobile data traffic trend [1]distance communication. One of the prime candidates currently receiving pro-gressively stronger consideration is Optical Wireless Communication (OWC).The concept of communicating through the visible spectrum can be broughtback several thousand years ago, when Greeks, Romans and American Indiansused smoke signals as a primitive means to communicate at a distance. Light-houses were another early and basic application of optical communication. Thebeam emitted by the tower was used to assist ships for maritime navigationnear coastal lines. However, the first actual instance of wireless communica-tion using the optical spectrum to transmit data in a controlled manner waspioneered in 1880 by A. G. Bell using an apparatus he conceived as the photo-phone. It used a combination of lenses and mirrors to capture and focus raysof sunlight, and then modulate speech signals onto the beam of light. He wasable to wirelessly transmit a voice telephone message over a distance of 213 m.Decades later, OWC has established itself as a new field of telecommunicationsby making use of the complete range of the optical spectrum, from ultravioletto Infrared (IR). Nowadays, OWC is seen to be a viable wireless alternative toRF in some applications like last mile connectivity, backup/temporary links orcellular backhauls.More recently, the advent of Solid-State Lighting (SSL) technology has pro-pelled Light Emitting Diodes (LEDs) to the forefront of lighting solutions byrealizing high-brightness white or coloured light emission at excellent energy ef-ficiency and cost effectiveness. The replacement of traditional lighting solutions,incandescent and fluorescent lightbulbs, is seen as inevitable with the decliningprices of LED technology. The United States Department of Energy conducteda study to estimate the amount of LEDs installed since 2009. A graph showingthe proliferation of LEDs for different applications is shown in Figure 1.2. Overthe next decade, the adoption rate of LEDs for lighting applications is expectedto continue increasing exponentially for home, office, automobile, outdoor light-ing and commercial signage applications.With the latest technological advances in SSL, OWC, specifically VisibleLight Communication (VLC), can finally take advantage of the untapped visiblespectrum that presents terahertz of free and unregulated potential bandwidth.2Chapter 1. IntroductionMillion01020304050ParkingStreetlightHigh BayTrofferDownlightDecorativeMR16DirectionalA-Type2012201120102009Figure 1.2: Installed base estimates of various LED applications in the UnitedStates [5]VLC is the newest addition to the OWC family, as fast-switching white lightLEDs spawned a whole new branch of OWC and follows in the footsteps of otherOWC variants such as IR and Free Space Optical (FSO), both of which havebeen well-developed over the last thirty years. Communication using visiblelight is accomplished by causing the LED chip to flicker at rates imperceptibleto human vision using a modulated information signal. The fluctuating lightintensity can then be captured by a photosensitive device at a distance away,creating a basic communication channel. VLC intends to reuse LED fixtures de-ployed as light sources to provide optical communication in addition to generalillumination. Seeing that most indoor environments are already illuminated us-ing artificial lighting, VLC can extend the capabilities of these fixtures to enablehigh-speed wireless access through the deployment of VLC-capable LED accesspoints. With the replacement of existing lighting infrastructure currently takingplace, VLC is seen as the perfect candidate for enhancing the last mile broad-band access to end-users in an indoor setting. In fact, VLC has been demon-strated to be a competitive technology capable of complementing the existingRF mobile network and increasing the downlink capacity of wireless networks,by taking advantage of the best characteristics from both transmission medium.It is envisioned that in the close future collaborative heterogeneous networkswill integrate different technologies (RF, Power-Line Communication (PLC),IR, VLC) to develop intelligent networks that provide high-speed connectivitywhile remaining seamless to the end-user. At the same time, such a growingtrend towards multi-tiered network systems is an ideal opportunity for VLC toestablish itself. Since VLC primarily aims to increase system capacity for theshort-medium range, it would usually be provisioned as the last mile connection(attocells) access points in heterogeneous networks. This would also eliminatepotential inter and intra-cell interference with the RF macrocellular network,while providing enhanced indoor coverage as a major benefit of the OWC chan-nel which is its immunity to Electromagnetic Interference (EMI). It is ideal for31.1. Recent Advancesuse in settings with RF restrictions like hospitals, airplanes, or where wirelessis unreachable.1.1 Recent AdvancesRapid development in LED technology has motivated the study of VLC with ex-tensive activity, especially in recent years. Although, current investigation hasbeen predominantly concentrated on exploring techniques of achieving highertransmission speeds, different modulation techniques, Multiple-Input Multiple-Output (MIMO) systems and Organic Light Emitting Diode (OLED)-VLC.Notably, the majority of published material has been focused on the physicalVLC link and developing proof of concepts, while commonly omitting the im-plementation details of the backbone infrastructure. In fact, little VLC researchexplicitly addresses the method of delivery of transmission data from a sourceto the luminaires. Currently, the major candidates for the VLC backhaul areusually either PLC or Ethernet (Power-over-Ethernet (PoE)).In the first case, PLC (IEEE 1901) reuses electrical wiring already installedto power indoor appliances as the carrier for data communication. Since ceilinglamp fixtures are also provisioned by these power lines, PLC enables a low-costnetwork backbone for VLC, eliminating the need for extra cabling infrastructurechanges. In a PLC-VLC system, the voltage signal carried by the power line canbe used in an amplify-and-forward approach to drive the LED. Such a systemwas pioneered in 2003 by T. Komine and M. Nakagawa [6], which at the sametime paved the way for the study of VLC. However, the main disadvantage ofusing PLC as the data delivery method is attributed to the medium itself; powerlines are generally considered hostile environments that are unsecure, noisy andprone to EMI.On the other hand, the twisted pair cable used for Ethernet (IEEE 802.3)offers a better noise immunity environment and improved security over PLC.Ethernet has attained ubiquity in homes and offices and has established itself asthe Media Access Control (MAC) and Physical layer (PHY) protocol of choicefor cabled Local Area Network (LAN). It offers low deployment costs, com-patibility with existing LAN networks and flexible data rates from 10 Mbpsto 10 Gbps. The integration of VLC with an Ethernet backhaul has recentlystarted to gain traction. Notably, 10Base-T over VLC has been accomplishedby several authors in recent years [7–10]. In these works, the authors have suc-cessfully transmitted an analog Manchester-encoded waveform of an Ethernetframe over VLC using commercial off-the-shelf LEDs while providing illumina-tion. In [7], a decode-and-forward strategy is used to extract the EthernetUser Datagram Protocol (UDP) payload, before encapsulating the data witha new custom header which is then retransmitted as a VLC frame using Dif-ferential Positioning Pulse Modulation (DPPM). In [8], the authors designeda low-cost 10Base-T amplify-and-forward duplex system with a VLC downlinkand IR uplink. They were able to demonstrate a UDP and Transmission Con-trol Protocol (TCP) transmission setup at distances of 2 − 3 m with Bit Error41.2. Thesis ContributionsRate (BER) of < 1× 10−9. Similarly, in [9], a 10Base-T VLC-LAN Ethernetlink proof-of-concept was conceived. An additional convenience associated withan Ethernet cable link is the prospect of utilizing PoE as the power backdropfor the LED lamps. Preliminary investigations of VLC systems combined withPoE backhauls were explored in [11–13]. The combination of PoE with illumina-tion has only recently started to increase in popularity, with limited commercialofferings for PoE-powered LED lamps becoming available. We envision extend-ing the original intended usage of PoE switch backbone beyond that of Voiceover Internet Protocol (VoIP) phones or security cameras to create a path forPoE-powered VLC luminaires for homes and offices.1.2 Thesis ContributionsWe see a perfect opportunity for bridging Ethernet/PoE with VLC as mutuallybeneficial to the technologies. Specifically, we look to incorporate a VLC sys-tem with pre-existing LAN infrastructures in indoor applications and developa VLC PHY that would serve as a wireless intermediary between two Ethernetterminals. We are also interested in data rates faster than 10Base-T, with a 100Mbps downlink being more appropriate for the majority of today’s applications,like high-definition video streaming. Prior to this thesis, 100Base Ethernet com-bined with VLC has only been formerly investigated in [14]. In their design,the authors first converted the analog Multi-Level Transmit 3 (MLT-3) signalof 100Base-TX to a 100Base-FX Non-Return-to-Zero Inverted (NRZI) signalbefore transmitting over VLC. The central drawback of this approach is theneed for a TX-FX converter at the transmitter and FX-TX at the receiver. Thecomplexity and size of the transceivers are increased by requiring integratedcircuit or Field-Programmable Gate Array (FPGA) components.In this work, we combine 100Base-TX with VLC to create a broadcast sys-tem which we call Ethernet over Light (EoL). Essentially, EoL would enable apoint-to-point Ethernet connection using a VLC intermediary that is seamlessto the user. Our particular implementation of 100Base-TX over VLC is done atthe Ethernet PHY level, in which the analog MLT-3 electrical signal is transmit-ted directly over VLC as an optical waveform in an amplify-and-forward fashion.This approach minimizes the amount of pre-processing at the transmitter sideand, at the same time, post-processing at the receiver since the transmitted sig-nal is essentially unmodified from the source. The complexity of the transceiverdesign is also greatly diminished for the reason that the system does not requirea complex processing unit (apart from signal processing at the receiver) andcan be realized exclusively using discrete components. Additionally, consider-ing that we are trying to incorporate 100Base-TX over VLC, a natural extensionwould be to use PoE as the main power source for the components of our VLCtransmitter (LEDs, amplifiers). To the best of our knowledge, no other authorhas any published material on an analog amplify-and-forward 100Base-TX PoEVLC system as of the writing of this thesis.Our main contribution in this thesis is an extensive investigation towards the51.3. Outline of Thesisimplementation of an EoL system, and is divided into three parts. First, a com-plete link analysis is presented, with supporting channel models for Ethernetand VLC. This is required in order to examine the transmission characteristicsof our proposed system. In addition, since we are trying to transmit a baseband100Base-TX signal in a bandlimited channel, it is crucial to identify the linkcomponents with the worst bandwidth characteristics that could potentially bot-tleneck the overall system. Using a comprehensive noise model, we establishedtypical Signal-to-Noise Ratio (SNR) levels that would be expected in variousEoL deployment configurations. Second, popular post-equalization techniqueswere investigated as to recover the original information at BERs applicable for100Base-TX. We will also discover that the EoL link introduced several distor-tions that resulted in Inter-Symbol Interference (ISI) and required us to considerthe use of practical sub-optimal equalization. We evaluated the performance ofselect equalizers in our EoL system, subject to varying conditions. For instance,we accounted for assorted Cat 5 cable lengths, LED bandwidths, receiver po-sitionings, as well as different VLC channel models. Third, an early prototypewas developed by M.Eng. student Blake Stacy to complement this thesis. Thetransceivers required careful design considerations to retain the integrity of the100Base-TX signal, while minimizing the effects of noise as much as possiblealong the entire path of the link. A major advantage of our EoL system isthe fact that it can be developed with presently-available LED and photodiodetechnology. We believe that working towards a real-world demonstrator willshowcase the full potential of EoL as an indoor wireless LAN solution.1.3 Outline of ThesisThis thesis is structured as follows. In Chapter 2, we provide the preliminar-ies required to understand the rest of the chapters and includes an overviewof Ethernet (and PoE), LED and VLC technologies. In Chapter 3, we intro-duce our EoL transmitter and receiver designs with an accompanying high-levelsystem model. In Chapter 4, we present the channel models that will be usedsubsequently in our EoL framework. In Chapter 5, an analysis of the overalllink budget along with the noise characteristics is performed. In Chapter 6,we explore equalization strategies and demonstrate how they can be employedto improve the performance of our EoL system. In Chapter 7, we present thedesign of a hardware prototype that was developed in conjunction with thisproject. Finally, Chapter 8 concludes this thesis.6Chapter 2PreliminariesTo start, we provide a background review for the reader on the basic conceptsof 100Base-TX Fast Ethernet, Category 5 Unshielded Twisted Pair (UTP)cables, PoE, LEDs and VLC.2.1 IEEE 802.3 EthernetEthernet, defined in IEEE Standard for Ethernet [15], specifies the data linklayer and PHY of the Open Systems Interconnection (OSI) model stack (Fig-ure 2.1) and describes the functionality of the last mile prior to transmissionover the physical cable medium. The Ethernet MAC sublayer’s main responsi-bilities are data encapsulation and medium access control. It determines howdata trickled down from the upper transport and network layers is formattedinto Ethernet frames, and when these frames are to be transmitted. In turn, thePHY dictates how the frame is sent electrically over the copper or fiber trans-mission medium. As a wide array of Ethernet PHY subtypes exists under IEEE802.3, we will focus on 100Base-TX since our primary goal is to implement a100 Mbps optical link.2.1.1 100Base-TX (Fast Ethernet)100Base-TX (or Fast Ethernet) is defined under Clause 24 and 25 of [15],which supersedes the older ANSI X3.263-1995. Clause 24 defines the func-tionality of the Physical Coding Sublayer (PCS) and the Physical Medium At-tachment (PMA) layers for 100Base-TX and 100Base-FX (fiber variant), whileClause 25 specifies the Twisted Pair-Physical Medium Dependent (TP-PMD)copper cabling for 100Base-TX. The 100Base-TX PHY stack (Figure 2.2) canbe broken down further into multiple sublayers: (1) Medium-Independent In-terface (MII), (2) PCS, (3) PMA, (4) TP-PMD and (5) Medium-DependentInterface (MDI). First, the reconciliation layer provides a connection betweenMAC and the media-dependent PHY layers underneath. The MII presents in-coming, available data in parallel as nibbles to the PCS. The PCS comprisesthe top-most layer of 100Base-X PHY, and performs the tasks of 4B/5B encod-ing, multiplexing and synchronization. The PMA sublayer takes care of framingand clock recovery while the TP-PMD performs scrambling and descrambling aswell as MLT-3 coding. The MDI is the physical interface to the medium and isdistinct for either fiber or twisted pair applications. In 100Base-TX, the ActiveOutput Interface (AOI) converts this ternary signal into an electrical waveform,72.1. IEEE 802.3 EthernetApplicationTransportNetworkLinkPhysicalMediumOSI Reference ModelUpper LayersIEEE 802.3 EthernetLogical Link ControlMedium Access ControlPhysicalMediumFigure 2.1: OSI reference model and IEEE 802.3 protocol stack{ } 100Base-TXHigher LayersLogical Link Layer (LLC)Media Access Control (MAC)ReconciliationPCSMediumTP MDIMIIPMATP-PMD100Base-XFigure 2.2: 100Base-TX PHY [16]which is then coupled onto the twisted pair physical medium as a differentialsignal through a line driver, commonly an isolation transformer with turn ratio1:1. Below, we detail the operations performed by each PHY sublayer in or-der to transmit an Ethernet frame over a twisted pair cable in the form of anelectrical signal.4B/5B EncodingFirst, the PCS sublayer implements a 4B/5B block-encoding that maps a four-bit nibble into a corresponding five-bit code-group. This specific coding schemeguarantees that there is a minimum of two symbol transitions per code-group,and that no more than three consecutive bits are of the same value. 4B/5Bencoding aims to prevent long strings of zeros from occurring and to minimizethe amount of DC content in the signal by allowing a maximum of one leadinglogical zero and two trailing logical zeros. This also eliminates the need for aseparate clock line as timing can be recovered from each rising or falling edgein the signal. With the addition of an extra codeword bit in 4B/5B encoding,82.1. IEEE 802.3 EthernetInput 4b Output 5b Input 4b Output 5b0000 11110 1100 101110001 01001 1101 110110010 10100 1110 111000011 10101 1111 111010100 01010 IDLE 111110101 01011 /J/ 110000110 01111 /K/ 100010111 10010 /T/ 011011000 10011 /R/ 001111001 10110 /H/ 001001010 10111 DEAD 000001011 11010Table 2.1: 4B/5B encodingthe output symbol rate is increased by 25 % from 100 Mbps to 125 Mbps. Ta-ble 2.1 [16, p188] shows the 4B/5B encoding table, where twenty-three of thethirty-two (25) available codewords are used, of which sixteen are used for dataand seven for control (e.g. start-of-stream delimiter, end-of-stream delimiter).The remaining nine unused codewords are invalid and result in symbol error ifdecoded.ScramblingIn 10Base-T, the output data signal is transmitted using Manchester encoding ata rate of 20 MHz with an input symbol rate of 10 MHz. This is indeed adequatefor the case where a 100 MHz standard Category 5 copper cable is employed.But, if the same line code was to be considered for 100Base-TX, a 250 MHz NRZIbitstream (post-4B/5B encoding) would be sent over the wire, which is grosslyinadequate for Cat 5 cables. At the same time, the electromagnetic emissionsfrom the physical link at high frequencies would exceed regulatory limitationssince the Federal Communications Commission (FCC) limits emissions in therange of 30 MHz to 1 GHz [17].In 100Base-TX, scrambling is carried out by the TP-PMD sublayer to ef-fectively flatten the transmit spectrum and spread the energy content of theinput signal uniformly. In doing so, the system makes better use of the avail-able bandwidth and ensures that no single frequency holds too much energy. Asequence length of eleven taps has been proven to attain the necessary reduc-tions in emissions at high frequencies. Scrambling changes the output of thedatastream in a deterministic way such that the original data stream can berestored at the receiver with a simple inversion of the scrambling function by adescrambler. A scrambling sequence (key stream) s(n) is generated using thefollowing recursive relationship, where ⊕ denotes a bitwise XOR (i.e. modulo-292.1. IEEE 802.3 Ethernet11110 0000V0V+1V-1V(a) FSM0+10-10+10-10000111 1(b) Trellis representationFigure 2.3: MLT-3 equivalent modelsaddition) [18, p17]:s(n) = s(n− 11)⊕ s(n− 9) . (2.1)The scrambled output (ciphertext bitstream) c(n) is then obtained by XORingthe key stream s(n) and the plaintext binary data input stream x(n)c(n) = s(n)⊕ x(n) . (2.2)MLT-3 Line CodeOnce more, to distribute the signal energy in the 20 − 100 MHz range towardseven lower frequencies, 100Base-TX adopts a final spectral shaping line code:MLT-3. This specific scheme requires significantly less bandwidth than an equiv-alent binary scheme operating at the same rate. MLT-3 is essentially a modi-fied Pulse Amplitude Modulation (PAM)-3 (pseudoternary) scheme with symboltransition restrictions. In MLT-3, the Non-Return-to-Zero (NRZ) input signalis modulated over three voltage levels {−1 V, 0 V,+1 V}, where a binary 1 in-put indicates a level transition, while a binary 0 indicates a hold. The encoderkeeps a history of the previous state change as it cycles from one level to theother. For instance, if the signal previously transitioned from −1 to 0, the nexttransition will be from 0 to +1, and so on. Direct transitions between +1 and−1 (and vice-versa) are forbidden. This line coding scheme can be representedas a circular Finite State Machine (FSM) (or an equivalent trellis) with fourstates {+0,+1,−0,−1}, one input {1, 0} and one output {+1, 0,−1} as shownin Figure 2.3. The FSM cycles between these four states and changes state whena logical 1 is presented at the input, and remains at the current state for eachlogical 0.Given an input symbol rate of 125 MHz and an input string of consecutivebinary 1’s, we can calculate the maximum fundamental frequency of the signal.Because four level transitions are required to complete a full cycle, the powerthen becomes concentrated at 125 MHz/4 = 31.25 MHz. If the input alternatesbetween 1’s and 0’s, the signal power accumulates at 15.625 MHz [19, p29], andso on. Hence, all higher frequencies above the maximum fundamental frequency102.1. IEEE 802.3 Ethernet0 0.2 0.4 0.6 0.8 1x 10−6−1.5−1−0.500.511.5Voltage (V)Time (s)(a) Sample waveform0 0.05 0.1 0.15 0.2−150−100−50050Frequency (GHz)Power/frequency (dB/Hz)(b) Power Spectral DensityFigure 2.4: MLT-3 signalare harmonics of these lower frequencies. In fact, 80 % of the total signal poweris contained below 62.5 MHz [20], which is also desirable for transmission overCat 5 cables that only supports a maximum of 100 MHz bandwidth. A sample100Base-TX waveform is shown in Figure 2.4(a), where each transition has arise/fall time of 4 ns and a hold time per symbol of 8 ns [18, p31]. From thesignal Power Spectral Density (PSD) in Figure 2.4(b), we notice that the bulkof the signal power is indeed collected around the 16 MHz frequency, and thatthere is a strong dip at 125 MHz. A direct drawback of such a baseband scheme,however, is the presence of a large DC component which can cause the signalto potentially shift towards 0, also known as the baseline wander effect (seeSection 4.1.3).2.1.2 Ethernet Cables for 100Base-TXANSI/TIA-568-C.2-2009 or Balanced Twisted-Pair Telecommunications Cablingand Components Standards [21] specifies the latest standard for the performancerequirements of Category 3, 5e, 6 and 6a balanced UTP cabling for Ethernet.At minimum, a Cat 5 cables or better are required to support the increasedfrequency requirements of 100 Mbps transmission. A standard Category 5UTP cable is rated at a bandwidth of 100 MHz and at a maximum physicallength of 100 m [16]. The cable itself is composed of eight 24 American WireGauge (AWG) thermoplastic insulated, thin copper wires, twisted in pairs witheach other within a thermoplastic jacket, and is terminated with modular RJ-45 plugs. The nominal differential characteristic of the cable cord, includingconnecting hardware, is 100 Ω [16, p232].In 100Base-TX, two of the four available twisted pairs are used separatelyfor duplex transmission and reception (the other two spares are unused). Dif-ferential signalling is employed in twisted pairs, where the regular signal is sent112.1. IEEE 802.3 Ethernetover one wire, and the same signal of inverse polarity is sent over the other.This technique, combined with the twisting property of the pair, allows forgreater resilience to common-mode noise interference and jitter on the wire (seeSection 5.2.1 for more information). ISI and attenuation are the prime chal-lenges that the UTP channel faces as transmission of a baseband signal pulsecan spread out over multiple time symbols due to the phase-delay characteris-tics of the wire. These effects are further magnified the longer the conductor isin length. In Section 4.1, we will describe these impairments analytically andprovide a Cat 5 cable channel model while the topic of noise in twisted pairswill be characterized in Section Power-over-EthernetPoE was developed as a method to safely deliver both power and data through aregular Ethernet cable to any PoE terminals in a LAN, without damaging non-PoE devices. Two IEEE standards were ratified over the years: IEEE 802.3af-2003, which described regular PoE and IEEE 802.3at-2009 PoE+, that was ableto supply increased power. 802.3af can deliver 15.4 W of DC power, 44 V of DCvoltage and 350 mA over a Cat 3/5/5e cable. The enhanced PoE+ can supplyup to 30 W DC, 50 V DC and 600 mA over Cat 5 or better. These amendmentsare now incorporated in the IEEE 802.3 suite [16] as Clause 33 – Data TerminalEquipment (DTE) Power via Media Dependent Interface (MDI), which definesthe PoE requirements for 10/100/1000Base-T technologies. PoE offers a numberof conveniences to the end-user. Primarily, using a single cable for power anddata eliminates the need for a dedicated power cable to the PoE device. Thiscan result in installation cost savings and a greatly simplified deployment. Atthe same time, existing LAN infrastructure can also be reused through a PoEupgrade, foregoing the need to change the existing cabling system.Power Supplying EquipmentThe PSE, commonly a switch, is the apparatus that injects power onto thetwisted pair. It can be implemented as either an endspan, that provides bothpower and data, or as a midspan, an intermediary that injects power between anon-PoE switch and a Powered Device (PD). It is responsible for searching itsend-links for valid PDs, classifying them, supplying and monitoring power, andfinally halting power when required. This power can be delivered safely throughtwo modes of operation: Mode A and Mode B. Mode A delivers power on theactive transmit and receive data pairs of the cable, by applying a common-modevoltage to each pair via the transformer center taps using the phantom powertechnique. Mode B utilizes the unused spare pairs within the cable instead.Powered DeviceThe PD is the device requesting power from the PSE on the terminating endof the link. Common devices can include VoIP phones, Wi-Fi access points,122.1. IEEE 802.3 Ethernet44-5715.5-20.5>10>2.8DetectionPeriodClassificationPeriodStart-UpPeriodOperationPeriodDisconnetionPeriod<500ms 50-75ms >15us 360-400mstVDCFigure 2.5: PoE procedure [22]Internet Protocol (IP) cameras and so on. The PD requires a carefully designedfront-end in order to interface properly with the PSE. The PD front-end ex-tracts power from the twisted pair and forwards the data to the Ethernet PHY.Commonly, a set of diode bridges precedes the front-end with the purpose ofadmitting power, from either the data pairs or the spare pairs, regardless ofsignal polarity. All valid PDs are identified by a signature resistance of 25 kΩ,and are also assigned to one of five power classes (from 0 to 4). This classifica-tion is used to indicate the maximum power the PD is able to consume, rangingfrom very low power (0.44 W) to high power (25.50 W). A DC/DC convertertransforms the received voltage into a lower voltage appropriate for use by thedevice.PoE Power-up ProcedureBefore applying power down the line, a valid PD connected to the PSE hasto be detected before applying power PoE-enabled equipment and to preventdamaging non-PoE devices. An elaborate handshake is then performed betweenthe PSE and PD to determine the nature of the load. The PoE proceduredepicting the stage by stage voltage ramp-up process is shown in Figure 2.5,and the details are listed below.1. Detection: A DC voltage is applied down the line and ramped up from∼ 2 to 10 V in order to detect the required 25 kΩ characteristic impedanceof the PoE device. The impedance is determined by measuring the V-Islope at the PD. Subsequent steps will only be executed if this signatureis detected by the PSE.2. Classification: The PSE continues to increase the voltage up to 20 V.132.2. Light Emitting Diodes++++++- -- -- -Photonsp-type n-typeHoles Electrons e-I+ -VFigure 2.6: Basic structure of an LEDThe PD draws a specific amount of current which is measured to identifyits power class (0/1/2/3/4), and hence the amount of power required bythe PD during normal operation.3. Startup: After the PD has been properly classified, the PSE continuesto ramp up the input voltage to its full capacity.4. Normal operation: The PSE supplies full power to the device duringnormal operation until disconnection is initiated.2.2 Light Emitting DiodesAn LED, as its name implies, is described as a diode capable of emitting lightthrough solid state electroluminescence. The basic structure of an LED is de-picted in Figure 2.6, and is composed of a doped p-type layer built on an n-typesubstrate. When an electrical current is applied through the doped p-n junc-tion of the semiconductor crystal, non-coherent light is produced. At the atomiclevel, when a forward bias is applied, electrons in the n-region diffuse across thejunction, reducing the width of the depletion zone between the two layers. Oncethis zone is thin enough, electrons can tunnel through and combine with theholes (missing electrons) in the p-region, causing a forward electrical current toflow. During this recombination process, electrons drop from a higher energylevel (conductance band) to a lower one (valence band), and a photon of energyproportional to this energy drop is released. The wavelength (i.e. colour) of theemitted photon is determined by the energy band gap of the semiconductor, thedegree of doping and the material.Continuous development of SSL technologies has increased the commercialviability of LED technology. Nowadays, LEDs offer a wide range of advantagesover other primitive lighting solutions, namely: energy efficiency, high bright-ness, reliability, compactness, low heat generation, mercury-free, high humiditytolerance and very long lifetime. One of the areas in which these benefits can142.2. Light Emitting DiodesPhosphorBlueLuminescencePhosphorescenceBond WireLED ChipFigure 2.7: Yellow-phosphor LED structurefully be taken advantage is general illumination. Similar to how compact fluores-cent light replaced incandescent lightbulbs, LEDs are predicted to take over theillumination market and replace these outdated classical lighting solutions. Infact, with prices dropping and consumer adoption rising, energy-efficient LEDsare expected to become the dominant lighting solution for residential and com-mercial use by 2018 [23]. Below, we look at the types of LEDs commonly usedfor VLC applications: (1) phosphor LED, (2) Red Green Blue Light EmittingDiode (RGBLED), (3) Resonant Cavity Light Emitting Diode (RCLED) and(4) OLED.2.2.1 Types of LEDsYellow-phosphor LEDTypically, phosphor-based LEDs are composed of a blue chip and a yellowishphosphor layer. The phosphor is an inorganic host material, doped with anoptically active element. A white light emission is generated when the bluelight passes through the yellow phosphorescent material as seen in Figure 2.7.Phosphor LEDs are currently the most popular option for illumination and com-munication due to their low cost, robustness, reliability and eye safety. However,this type of LED does come with a major drawback when used for VLC applica-tions. The slow temporal response of the phosphor emission (rise and fall time)produces a low data modulation bandwidth of only a few megahertz (about1 − 4 MHz). Though, this limited modulation can be enhanced by suppressingthe slower phosphor component using a blue filter at the receiver. By leavingonly the faster blue emission to be detected, the optical modulation bandwidthhas been reported to improve upwards of 20 MHz in some cases [10, 24–27]. Thisfiltering process does block a significant portion of the emitted energy and canresult in a power penalty of up to 36 dB compared to the white emission [28].The normalized frequency response of a typical blue-filtered yellow-phosphorLED is shown in Figure 2.8(a) (OSRAM DOT-it [29]), along with the emissionspectrum in Figure 2.8(b) (Luxeon III Star DS46 [30]).152.2. Light Emitting Diodes0-10-20Normalized Gain (dB)Frequency (Hz)106 107RawBlue Filtered(a) Normalized frequency responseWavelength (nm)Relative Spectral Power Density Distribution1.       500       600       700(b) Emission spectrumFigure 2.8: Typical phosphor LED propertiesRed Green Blue LEDAn RGBLED consists of three separate colour chips (red, green and blue), emit-ting at their corresponding wavelengths: 650 nm, 530 nm and 460 nm (see emis-sion spectrum of a Seoul Semiconductor F50360 in Figure 2.9(b) [31]). Thesethree channels can be controlled individually to generate white light or a diverseset of colours. In regards to VLC, this enables the use of custom modulationschemes such as Wavelength Division Multiplexing (WDM) [32–34] and Color-Shift Keying (CSK) [35]. Although, additional complexity is required at thetransmitter for colour balancing. Alternatively, one can use a single colour tocarry useful information, while the other two chips are driven with a DC sig-nal to maintain a white illumination balance. While RGBLEDs can sometimeachieve superior switching frequency over phosphor-based LEDs, their opticalefficiency is still currently about half of the former [35]. It is noteworthy how-ever, that the optical power output of each chip is not identical. The greenchip usually has a lower power output compared to the red and blue chips asseen in Figure 2.9(a) ([32]), which is attributed to the fact that it does notefficiently convert power into optical intensity [35, 36]. Additionally, specializedRed Green Blue (RGB) colour sensors are required at the receiver to separatethe individual colour intensities and convert them into voltages. Unfortunately,RGBLEDs are potentially being phased out of the lighting industry, in part,due to their high manufacturing cost and the domination of phosphor LEDs.Resonant Cavity LEDIn RCLEDs, a light emitting region is embedded in a micro-cavity and borderedby two parallel mirrors (Bragg reflectors). The emission wavelength of the LEDis dictated by the thickness of the active region, commonly about half or equalto that of the emitted wavelength, which is also matched with the resonancewavelength of the optical cavity. This resonant cavity, combined with the Braggreflectors, enhances the emitted light (resonant-cavity effect) in a fashion such162.2. Light Emitting Diodes0-10-20Normalized Gain (dB)Frequency (Hz)106 107(a) Normalized frequency response percolourWavelength (nm)Relative Spectral Power Density Distribution1.       500       600       700(b) Emission spectrumFigure 2.9: Typical RGBLED propertiesGlass SubstrateAnodeConductive Layer (Organic/Polymer)Emissive Layer (Organic/Polymer)CathodeFigure 2.10: OLED structurethat RCLEDs can be modulated at very high frequencies, upwards of hundredsof MHz. Compared to classical LEDs, RCLEDs have improved characteristicswhile retaining the benefits of traditional LEDs. The emission beam is highlydirectional with a much narrower spectral distribution, increased brightness,efficiency and bandwidth [37]. These characteristics are typically well suited foroptical fiber applications but not necessarily for VLC, as research in this areais ongoing [38].Organic LEDOLEDs are composed of a carbon-based conductive layer and an emissive layer,lying between the anode and cathode conductors (see Figure 2.10), and operatemuch in the same way like the previous types of LEDs. Compared to phosphorLEDs, OLEDs are more lightweight, have higher contrast ratios and lower powerconsumption [39]. Although, the organic polymer has a lower carrier mobilitythan the silicon counterpart which greatly limits its frequency response to justa few hundred kilohertz, too low for high-speed applications. However, OLEDtechnology for general illumination is still maturing as its efficiency is beingimproved, doubling every two or three years [39]. Though it is not yet consid-ered a suitable replacement for other LED illumination solutions, OLED is anemerging technology in the early stages of development that shows tremendous172.2. Light Emitting Diodesm = 1m = 5m = 50cosm0≤    ≤π/2I0ф0.5I0ффnTXFigure 2.11: Lambertian emission patternpotential. Recent advances in organic device technology and research in OLED-VLC has started to gain traction for indoor lighting applications. References[40] and [41] offer a well-rounded study of using OLED for VLC.2.2.2 LED ModelIn order to analyze the overall VLC link in later chapters, we require a com-prehensive description of a standard LED and its basic illumination properties.LEDs follow a non-isotropic emission pattern, and are typically modeled as Lam-bertian emitters (see Figure 2.11). The radiant intensity angular distributionof a generalized Lambertian emitter with a maximum luminous intensity of I0is given as [42–45]L(φ) ={m+12pi cosm φ, −pi/2 ≤ φ ≤ pi/20, otherwise(2.3)where φ is the angle of irradiance and m is the Lambertian order which dependson the half-power emission angle Φ1/2 property of the LEDm = −ln 2ln(cos Φ1/2). (2.4)The higher the order m, the more directed the radiation beam becomes. m isequal to unity for a half-power angle of 60◦.LEDs exhibit a Low-Pass Filter (LPF) behavior when directly modulated byan AC current source. The frequency response is dictated by the rise and falltime of the LED, which in turn depends on the type and properties of the LEDused. In a Linear Time-Invariant (LTI) system, it is adequate to approximatethe LED as a first-order LPF with a frequency response of [37, 46–51]Hled(f) =11 + j ff3dB(2.5)where the 3 dB frequency f3dB is the maximum modulation bandwidth sup-ported by the LED. Contrarily, if the LED were to be driven by an AC signalvoltage, the LED depicts the characteristics of a second-order LPF [8, p4].182.2. Light Emitting DiodesDriving Current (A)Optical Power (W)Non-Linear RegionsFigure 2.12: LED nonlinear characteristics2.2.3 LED NonlinearityAn important factor to take into account when a practical LED is driven by anelectrical signal is the inherent nonlinear relationship between the input drivingcurrent and the output optical power [52] (see Figure 2.12). This is attributedto electron leakage that occurs at the anode region when the LED is driven athigh currents. The amount of electrons escaping the active region results ina lower number of photons emitted [35]. This nonlinearity effect can lead tosignificant degradation of the original signal, and is especially magnified whenused in conjunction with PAM schemes.To model the effects of nonlinearity, the authors in [53] proposed a Taylorseries expansion model of the optical power emitted versus driving current (P-I)curve that accounts for the low-pass response in addition to the nonlinearity. Afifth order polynomial fitting is usually required to portray an accurate transferfunction but a second order approximation offers a sufficiently fair representa-tion. In [54], the coefficients bn have been experimentally measured and modeledas a second order polynomial fit for an AVAGO HFBR-1521Z RCLED:Pout(t) =∞∑n=0bn[Isig(t) ∗ hled(t)− Idc]n (2.6)≈ b0 + b1(Isig(t) ∗ hled(t)− Idc)+ b2(Isig(t) ∗ hled(t)− Idc)2 (2.7)where Pout is the optical output power, Isig(t) is the input signal driving current,hled(t) is the impulse response of the LED modeled as a first order low pass filter,Idc is the biasing current and bn is the n-th order coefficient approximation.In this case, b0 = 7.3654 is the DC term, b1 = 0.258 mA−1 is the linear gain,b2 = −0.005 mA−2 is the second order nonlinearity coefficient, and Idc = 20 mA.This non-linear P-I curve is shown in Figure 2.13.A variety of techniques have been explored to reduce the amount of distortionand clipping, and effectively compensating for the deficiencies of the LED. First,the DC biasing point can be carefully optimized to keep the signal swing within192.3. Visible Light Communication0 10 20 30 40 50024681012Driving Current (mA)Optical Power (AU)Figure 2.13: Second-order polynomial P-I curve fit for AVAGO HFBR-1521ZRCLEDthe linear P-I operating range as much as possible. Secondly, time domain pre-distortion and linearization techniques have also been considered, in which theinput signal is compensated by linearizing the electrical-optical transfer functionover a certain range [55, 56].2.3 Visible Light Communication2.3.1 A Brief HistoryIn 2003, Visible Light Communication Consortium (VLCC) was established as ajoint venture comprised of Japanese technology companies to promote researchand development in the field of VLC. The first fundamental study of indoorVLC was published by T. Komine and M. Nakagawa in 2004 [6, 57]. All thewhile, Japan led the standardization front by issuing two visible light standards:Japan Electronics and Information Technology Association (JEITA) CP-1221and CP-1222 in 2007. From there, the interest spread to Europe, where thehOME Gigabit Access (OMEGA) project was instituted to further tap into thepotential of VLC. The aim of OMEGA was to provide gigabit connectivityin home LANs using a heterogeneous combination of wired and wireless tech-nologies. Important groundwork was made in the investigation of VLC duringthe this project. More recently, in 2011, the IEEE standardized VLC as IEEE802.15.7-2011 or IEEE Standard for Local and metropolitan area networks - Part15.7: Short-Range Wireless Optical Communication Using Visible Light [58] anddefines the specifications and requirements of the VLC MAC and PHY.202.3. Visible Light CommunicationDrivingCurrent (A)TransmiedPower (W)IncidentPower (W)Photocurrent (A)Figure 2.14: Intensity Modulation/Direct Detection2.3.2 Basics of VLCThe objective of VLC is to reuse current lighting infrastructure deployed inindoor settings to provide short/medium range wireless link of a few meters,by taking advantage of the dual nature of LEDs for communication and illu-mination. Provided with the high switching rate property of LEDs, VLC aimsto provide high-speed wireless optical downlink transmission using the visiblelight spectrum (370 nm ∼ 770 nm) as the communication medium. To transmitinformation with an LED over the VLC channel, the LED is modulated by aninformation signal, which fluctuates in intensity proportional to the driving sig-nal at high switching rates above MHz. This is well beyond the flicker fusionthreshold of the naked eye as the human eye will only perceive the average lu-minance at frequencies more than 200 Hz depending on the individual. On theother hand, a photosensitive detector is capable of capturing these tiny varia-tions in the order of a few µW, and decipher the information signal. Since a VLClink is inherently a localized broadcast system, the transmission is restricted tothe illumination range of the LED fixtures.VLC borrows heavily from IR technology as they both share very similartransmission characteristics. Their radiation behavior and their interaction withmatter are essentially the same considering the fact that the main difference liesin the operating wavelength of the LEDs emission. Just like IR communication,VLC adopts the Intensity Modulation/Direct Detection (IM/DD) technique forcommunication. In IM/DD, a real-valued positive waveform is modulated ontothe intensity of the emitted light, and an electrical current proportional tothe incident optical power is produced at the receiver [59] (see Figure 2.14).Specifically, a forward-biased LED converts a fluctuating driving current intooptical power that is directly proportional to the input signal, while a photo-diode converts the incident radiation back to an output electrical signal. Incontrast to RF communication, it is impossible to control the phase of the light-wave itself. Therefore, IM/DD is required in VLC due to the stochastic nature212.3. Visible Light Communicationof the recombination process inside the LED, resulting in non-coherent emis-sion. As a result, the modulation schemes available for IM/DD are also limited,but include: On-Off Keying (OOK), PAM, Pulse Width Modulation (PWM),Pulse Position Modulation (PPM), Frequency Shift Keying (FSK), OrthogonalFrequency-Division Multiplexing (OFDM), CSK. We refer the reader to [60] formore information on modulation topics.2.3.3 Benefits, Trade-offs and ChallengesIn addition to all the advantages of SSL technology, VLC makes use of a widelyavailable, unregulated, free to use, and nearly unlimited bandwidth of the op-tical spectrum. Compared to IR communication, VLC has access to a broaderspectrum and is also considered safer in terms of eye safety at the same time. Byoperating in the visible spectrum, VLC can take the load off of the highly con-gested RF spectrum without creating additional EMI with radio bands. VLCis thus able to work in areas where EMI sensitive equipment would otherwiseprohibit wireless communication, or where RF is unusable, such as in airplanes,hospitals, industrial plants, mines or even space stations. Other locations thatcould benefit from such a short-range VLC link are in homes, commercial malls,supermarkets, classrooms, museums. Consequently, it can be seen as either areplacement, or complementary system for indoor RF wireless. Although RFtransmission maximizes mobility and allows for transmission through walls, ra-dio links suffer from multipath fading which is defined as large fluctuations inreceived signal magnitude and phase. On the other hand, optical IM/DD utilizesphotodetectors that are several order of magnitudes greater than the wavelengthof visible light (millimeters versus nanometers). Thus, the VLC link gains effi-cient spatial diversity and does not suffer from this multipath fading experiencedin RF links. VLC also provides some security against eavesdropping since thecoverage area is typically confined to the immediate vicinity illuminated by thelamp. In the same vein, this prevents interference from other nearby VLC linksin adjacent rooms, allowing excellent spatial reuse of the available spectrum, andresulting in increased network throughput per unit area. As previously men-tioned, because wireless data traffic is typically bursty and highly asymmetric,biased towards the downlink, this lends itself perfectly for a VLC system.However, the nature of transmitting in the visible spectrum does presentsome unique challenges to overcome as a communication channel. In contrast toRF, optical waves cannot penetrate opaque obstacles, and so, VLC transmissionswould typically be confined to the immediate area being illuminated, typicallya short distance of a few meters. The quality of service of an indoor VLC linkis highly influenced by the configuration of the VLC channel. Specifically, thecommunication link is completely characterized by the surrounding transmissionenvironment, the capabilities of the LED, and the movement and rotation ofthe receiver. The usage of LEDs as transmitting devices also contributes amajor drawback. The amount of bandwidth available is usually not limitedby the optical channel but by the LED itself, as the achievable transmissionrates will depend heavily on the switching frequency supported by the LED.222.3. Visible Light CommunicationIn indoor applications, where wide area coverage is desired, phosphor-basedLEDs and RGBLEDs are the two current preferred SSL types for VLC lightingapplications. The raw bandwidth of these LEDs that are commercially availableare limited to a few megahertz due to the fact that they are manufactured forhigh brightness and not purposefully for high-speed modulation. This introducessome limitations to the performance in a bandlimited system and can becomea significant inhibiting factor (the minimum bandwidth required for our systemwill be explored later in Chapter 6). The main impairments to the quality ofservice in a VLC link are associated with (1) ambient light and thermal noise,(2) blocking/shadowing and (3) multipath propagation.Firstly, a photodiode can capture any and all light within its responsitivityspectrum; the channel then is highly susceptible to other bodies emitting in thevisible light. The major sources of noise and interference in VLC links are at-tributed to ambient sunlight and other nearby artificial lighting fixtures. This ismanifested as shot noise at the photodiode receiver. Optical filters are typicallyemployed to reject these unwanted DC noise components as much as possible.On the other hand, thermal noise is generated by the electrical component ofthe transceivers and can become highly problematic depending on implementa-tion choices. This will be further investigated in Section 5.2.2. Secondly, sincelight cannot pass through solid bodies, the VLC Line-of-Sight (LOS) link issubject to either permanent or temporary shadowing, caused by objects or apassing person. Though VLC links do not necessarily require LOS as it canwork with diffuse reflections, blocking can still immensely disrupt transmissionlink quality. A distributed lighting system can ensure that a VLC fixture alwaysmaintains LOS with the receiver. Lastly, as LEDs employed for illumination aretypically non-directed, the link is subject to multipath propagation. Emittedlight signals by the transmitters can reflect multiple times before finally arrivingat the photoreceiver. Depending on the locations of the transceivers and, moreimportantly, the room configuration, this can result in significant ISI. Somechannel models were developed in literature that account for the contributionsof diffuse components and will be explored later in Chapter Photodiode ReceiverAt the receiving end of the VLC link, a photodiode is employed to convert theoptical signal back to an electrical signal. When an incident photon strikes thesurface of the photosensitive diode, an electron and a positively charged hole pairis formed. The hole will flow towards the anode and the mobile electron towardsthe cathode, generating an electrical photocurrent that is directly proportionalto the instantaneous received optical power. The raw bandwidth of the pho-todetector is dictated by the rate at which it is able to respond to variations inincident optical intensity, usually upwards of hundreds of MHz. In general, twotypes of photodiodes are suitable for VLC applications: silicon-based PositiveIntrinsic Negative (PIN) photodiodes or Avalanche Photodiodes (APDs). Bothare characterized with low noise, low-cost, tolerance to temperature fluctuationsand good sensitivity in the visible light spectrum. Though PIN devices gener-232.3. Visible Light Communicationally have better linearity than APDs [60], the APD’s increased sensitivity fromthe internal gain factor translates to better performance in low light situationsor when pre-amplifier noise is the dominant noise source.In a VLC system, the effective area of the photodiode determines its per-formance in regards to converting light into a photocurrent. A bare photode-tector’s effective area Aeff(ψ) is determined by its aperture size Ar, its Field-of-View (FOV) Ψc [61], and is proportional to the cosine of the angle of incidenceψ from the Lambertian lawAeff(ψ) ={Ar cosψ, 0 ≤ ψ ≤ pi/20, otherwise. (2.8)To increase the performance of photoreceivers in VLC applications, the effectivearea of the photodiode ideally needs to be increased. An enlarged aperture sizeaugments the amount of photons accumulated and consequently, the receivedoptical strength, though at the expense of increased noise and lower bandwidth.Hence, optical filters and concentrators are usually the preferred method of aug-menting the photodiode’s effectiveness instead. The optical filter is primarilyused to remove the DC signal induced from ambient light by passing only mod-ulated signals, while a non-imaging, hemispherical concentrator is used to focusmore incident light onto the photodiode area, compensating for high spatialattenuation. The effective collection area of the photodiode Aeff can then beenhanced to [61]Aeff(ψ) ={Ts(ψ)g(ψ)Ar cosψ, 0 ≤ ψ ≤ Ψc0, otherwise(2.9)or equivalently,Aeff(ψ) = Ts(ψ)g(ψ)Ar cosψ rect(ψΨc)(2.10)where rect(x) is the rectangular function (1 for |x| ≤ 1 and 0 for |x| > 1), Ψc isthe FOV of the concentrator, Ts(ψ) is the gain of the optical filter, and g(ψ) isthe gain of the concentrator (with a refractive index n) defined asg(ψ) ={n2sin2 Ψc, 0 ≤ ψ ≤ Ψc0, ψ > Ψc. (2.11)24Chapter 3Ethernet over LightProposalFrom the ubiquity of Ethernet, and the rapid development of LEDs, we pro-pose to combine VLC with Fast Ethernet and PoE to develop a VLC-LANsystem called EoL. The design of this system would provide full-room illumina-tion while enabling 100 Mbps Ethernet connectivity over a link of a few metersin an indoor residential and office setting while reusing current lighting andEthernet infrastructures. In this chapter, we introduce the concept of EoL anda high-level system block diagram of the major components. We also include adiscussion of the design of the VLC transmitter and VLC receiver, along withthe associated benefits, trade-offs and challenges of our EoL system.3.1 Our ApproachThe general idea of EoL is to develop a broadcast VLC-LAN system with a100Base-TX PoE backbone, merging the technologies of Ethernet with VLC.Figure 3.1 depicts what a real-world implementation of EoL would resemble.A PoE switch located elsewhere in the building connects to set of distributedLED luminaires in a room via standard Cat 5 UTP cables. The PSE deliversboth data and power to the VLC devices. An Ethernet front-end, located at thelamp, converts the data signal directly to a modulated optical signal, which iscaptured by nearby VLC-capable devices (i.e. laptops, tablets, smartphones).This enables a wireless 100Base-TX Ethernet connection, where the VLC linkis essentially seamless to the user. Equivalently, we can think of the VLC linkas a simple RF intermediary between two terminals in an Ethernet LAN. TheVLC transmitter, composed of high-speed LEDs, acts analogously to a Wi-Fiantenna, in which packets are broadcasted wirelessly by the access point overthe air and captured by Wi-Fi receivers.Our EoL implementation follows an amplify-and-forward strategy by trans-mitting the baseband, analog 100Base-TX Ethernet signal “as is” over the VLCmedium. Specifically, the MLT-3 modulated voltage signal received over thewire is converted to a driving current by a simple analog circuit, which is thenlinearly amplified to directly drive an LED using intensity modulation. In ourapproach, no additional pre-equalization, pre-processing, spectral shaping ordemodulation is performed at the VLC transmitter end. This keeps the com-plexity of the transmitter drive circuitry low and eliminates any possible latency253.2. System ModelLaptopCat 5 CablesLED FixturesPoE SwitchFigure 3.1: Ethernet over Light conceptproblems since the signal is forwarded immediately. Instead, the complexity isshifted to the receiver, where signal processing is performed after signal detec-tion. The recovered and decoded Ethernet data signal can then be delivered tothe appropriate Internet application.3.2 System ModelA high-level system model of our proposed system is shown in Figure 3.2, wherethe Ethernet link is depicted in the top half portion of the figure, while the VLClink is shown in the bottom half. To follow this block diagram, we assume aUDP/IP packet is broadcasted by the PoE switch (PSE) located at the top left-hand side of the figure. The configuration parameters of the Ethernet link areshown in Table 3.1. The binary data is encapsulated, encoded and modulatedat the Ethernet PHY sublayers (PCS, PMA, TP-PMD) into an MLT-3 voltagewaveform. The electrical signal is then sent over the data pairs of a regular Cat5 UTP medium and received at the VLC transmitter. At the same time, poweris injected through these same data pairs (or spare pairs) of the cable. TheVLC transmitter then linearly drives the received data signal to modulate anLED, and transmit the information over the VLC channel. The VLC receivercaptures the optical information using a regular photodiode, which then per-forms some amplification and filtering. The original 100Base-TX data is thenrecovered using signal processing (equalization and decoding). Below, we detailthe implementation of an EoL transmitter and receiver.263.2. System ModelHPFEqualization/ DecodingInput4B/5BEncoding(PCS)Scrambling(TP-PMD)MLT-3(TP-PMD) RJ-45PowerAmpRJ-45Power Sourcing Equipment VLC Transmier & Powered Device Front-EndVLC ReceiverDiodeBridgesSignature /Classification IsolationDC/DCConverterLEDDriverDCSupplyBiasTeeLEDPhotodiodeDataCat 5 ChannelVLC ChannelAmp TIAFigure 3.2: Proposed EoL high-level system modelParameter ValueMode Unidirectional, broadcastNetwork layer protocol IPTransport layer protocol UDPEthernet type 100Base-TXData rate 100 MbpsPower-over-Ethernet IEEE 802.3at-2009Cable Cat 5 UTPTable 3.1: Ethernet parameters3.2.1 Proposed TransmitterOur proposed transmitter is two parts: the PoE front-end that handles power,and the VLC side that handles data.PoE Front-endPower is extracted out of the twisted pairs from the middle tap of the receiveisolation transformers. The power available at the output is roughly 13−25.5 Wand 37−42.5 V, because up to 15 % of the originating power is lost along the wire.A set of diode bridges is utilized to receive power in either polarity over a twistedpair and allows for blind compatibility in the case of inverted signal polarity overnon-standard twisted pairs (e.g. crossover cables). The IEEE 802.3at standarddoes not enforce strict polarity transmission requirements over twisted pairs.The PoE power-up procedure (refer back to Section 2.1.3) requires a 25 kΩsignature resistance to correctly identify PoE-enabled devices. Subsequently,another resistor stage is used to determine the power class of our PD. A DC/DCconverter is used to downconvert the available voltage to a necessary level usedfor powering the various circuitry components on the VLC side, in our case:the op-amps and the LEDs. Though this stage can be implemented via analog273.2. System Model+-VinIledRLFigure 3.3: Typical analog LED driver configuration [35]components, we note that commercial PoE controllers are readily available onthe market that can implement the features described above which would greatlysimplify the integration of PoE in our VLC transmitter.VLC TransmitterOn the VLC side, the MLT-3 data signal is received as a differential analogelectric waveform over the twisted pair. The differential signal is combinedusing baluns (transformers or differential amplifiers) by subtracting the invertedsignal output (TX-) from the regular signal output (TX+) to get a single voltagesignal. This signal is then amplified by a Low Noise Amplifier (LNA) to obtaina maximal input swing used for driving the LED. The amount of amplification(i.e. the signal swing) needs to be carefully considered in order to operate theLED in the linear region of the P-I curve, minimizing nonlinear distortions athigh currents. Additionally, the AC signal should not be driven higher or lowerthan its maximal swing to avoid signal clipping at either extremes. A drivercircuit is used to convert this voltage signal into a current signal and a bias teeshifts the transmitted signal above zero to an optimal operating point. Thisbiasing point, generated by the DC source on the PoE side, dictates the amountof illumination in the room and should be kept at a level appropriate for indooruse. The LED driver can be implemented very simply using transistors anddiscrete components as shown in Figure 3.3. The amount of current driving theLED is controlled by the Bipolar-Junction Transistor (BJT) and the negativefeedback op-amp setup.The 100Base-TX PAM signal is transmitted over VLC by the LED, by vary-ing the input drive current, and hence the emitted optical power. For properoperation of EoL, an LED with a sufficient bandwidth range is needed to trans-mit the signal with minimal distortion over the VLC channel. An LED with amatching 3 dB frequency of 100 MHz to that of the cable would be ideal, butis unrealistic due to the current availability and state of LED technology. Ac-cordingly, the LED type and model would then have to be chosen carefully sothat distortions are minimized. The design of the analog drive circuitry at the283.2. System Modeltransmitter will depend on the LED chosen, as a single design will not supportall available LED models. It should also be designed as to not limit the band-width of the overall system though it should not be an issue. Authors in [46]have found that the LED driver can have a sharply limited, wide band-pass,and mostly flat system, depending on the implementation type.3.2.2 Proposed ReceiverAt the receiver end, the incident signal is detected over the VLC channel byan optical front-end. A photodiode semiconductor device is illuminated bythe modulated light intensity a few distance away over the VLC channel, andconverts the optical signal back to an electrical signal. An optical concentratorand optical filter are commonly used to enhance the performance of the receiversystem. As previously discussed in Section 2.3.4, the concentrator enlarges theeffective area of detection, while the optical filter rejects wavelengths outside ofthe desired spectrum. Depending on VLC channel conditions, a large amount ofambient light may, in extreme cases, completely saturate the photodiode. Theselection of the photodetector is also important since its bandwidth should notbe a limiting factor in the system, which is usually not an issue given the factthat photodiodes support a much higher bandwidth compared to LEDs (morethan hundreds of MHz). To convert this photocurrent back to a voltage signal,four types of front-ends can be considered in VLC:1. Low or high impedance terminated voltage amplifier [61, p73]2. Photoparametric amplifier [62, 258]3. Double heterodyne photoparametric amplifier [62, 259]4. Transimpedance Amplifier (TIA)Typically, wideband TIAs are the preferred type for optical applications asthey offer the best compromise between bandwidth and dynamic range. In theTIA setup of Figure 3.4, the photodiode is connected to the inverting terminalof the op-amp and the positive node is connected to ground. A feedback resistoris used to control the gain and to reduce the input impedance of the circuit.We note that the TIA and subsequent amplifier stage are usually realized withField-Effect Transistors (FETs) over BJTs for their superior noise performance[61] (this will be explored later in Section 5.2. Heavy attenuation attributedto VLC propagation loss produces a low level amplitude photocurrent at thereceiver, typically in the order of µA. Though the TIA stage provides somelevel of gain, the received signal will still be very small and require several morestages of amplification along with filtering of noisy interference. Subsequently,analog or digital electrical filters are employed to isolate the desired signal. Atthe output of the pre-amplifier, a first-order High-Pass Filter (HPF) or an ACcoupling capacitor is usually used to suppress the DC content induced by theLED transmitter itself, and ambient light sources like sunlight, fluorescent or293.3. Advantages, Drawbacks and ChallengesRf-+PhotodiodeVoutGIpFigure 3.4: Typical analog receiver front-end configurationincandescent lamps [63]. Subsequently, a pre-amplifier stage is used to boost thesignal back to detectable levels, and can consist of an LNA of fixed gain. Butin our case, ideally it would be a Variable Gain Amplifier (VGA) or AutomaticGain Controller (AGC) unit to accommodate for different EoL configurations(e.g. transmitter-receiver distance, illumination level, cable length). In 100Base-TX Ethernet, an AGC is commonly employed since the length of the cable isunknown to the receiver. As the signal suffers from significant distortion and ISIover the link, post-equalization techniques are utilized to mitigate these effects.We will go in depth in the study of equalization strategies later in Chapter 6.3.3 Advantages, Drawbacks and ChallengesWe see many inherent benefits associated with our integrated Ethernet-VLCsetup. First and foremost, we can take direct advantage of the widespread de-ployment and ubiquity of Ethernet LAN. With 100Base-TX, we have the con-venience of using PoE to power the LED luminaires, eliminating the need for adedicated power source at the VLC transmitter side. The power supplied by thePSE over the twisted pair cable can be used to power all electrical componentsand drive the LEDs. Assuming a PoE backbone and a retrofitted Cat 5 ca-bling network reaching lamp fixtures are available, then the only infrastructurechanges required are the replacements of current lamp fixtures to LED alterna-tives. The low cost of components, manufacturing and implementation of EoLtransceivers can lead to an especially quick deployment. In fact, the transceiverscan be built from scratch with inexpensive off-the-shelf components which willbe shown in the prototype in Chapter 7. This amplify-and-forward implemen-tation facilitates an extremely simple design of the transmitters that minimizesits size and complexity as much of it is shifted to the receiver by opting for post-equalization over pre-equalization. Realizing the receiver also does not requirea complex design; a suitable photodetector, some discrete components and aDigital Signal Processing (DSP) unit are the basic requirements.The issue of flickering is also avoided because 100Base-TX utilizes 4B/5B en-coding (see Section 2.1.1), ensuring that there are at least two signal transitionsper five symbols and thus preventing the signal from staying at a certain level for303.3. Advantages, Drawbacks and Challengesan extended period of time. When there is no data to be sent, the 100Base-TXPHY continuously transmits an IDLE signal to maintain synchronization withthe receiver and the connection alive. An IDLE signal is composed of a con-tinuous string of binary ones (see Table 2.1) resulting in the MLT-3 constantlycycling over all symbols, eliminating potential flickering.Though we do not explicitly deal with dimming in this thesis, it would alsobe possible to support this feature to an extent. Since VLC does not requirefull illumination to work, dimming is feasible down to a certain illuminationlevel. As long as the received optical power satisfies a minimum SNR thresholdand respects the target BER of 100Base-TX. Dimming the illumination levelwhile maintaining operation of VLC link can be performed by simply shiftingthe DC bias current up/down to increase/decrease the luminosity. Although,the level would need to be regulated to avoid the issues related to clipping andnonlinearity. This is especially important in a PAM scheme, where symbols aremapped to individual amplitude levels. We refer the reader to [64], where theauthors performed an analysis of the dimming performance for PAM over VLC.In terms of the transmission efficiency of an MLT-3 signal over VLC, base-band multi-level PAM modulation schemes have high spectral efficiency, albeitat the cost of power efficiency, but are still generally the preferred scheme forLED-based communication [46]. However, PAM schemes usually require a highSNR to perform at acceptable BER seeing that shot noise attributed to ambi-ent illumination dominates at low frequencies [65]. Ideally, signals should notcontain any information at DC, but 100Base-TX employs MLT-3 a basebandsignaling scheme that contains significant frequency components near DC (seesignal PSD in Figure 2.4(b)). As a result, the challenge will lie in maintaininghigh SNR in the VLC link. In addition, since we are trying to transmit a base-band signal in a band-limited system, the performance of the channel will beseverely hampered by the device with the lowest 3 dB modulation bandwidth.This bottleneck can be attributed to either the Cat 5 cable or the modulationcharacteristics of the LED. In the first case, if the lamp fixtures are located faraway from the PoE switch, the length of the cable used may reach up to themaximum 100 m length, which can result in an attenuation loss of up to ∼ 24 dBat 100 MHz (see Section 4.1.4). Again, as mentioned previously in Section 2.2,off-the-shelf LEDs are manufactured for high brightness and not for speed, andcan therefore also become the bottleneck factor in VLC transmissions. Wedetermine the minimum LED bandwidth required for proper operation in Sec-tion 6.3.2.An alternative to this approach, that was initially considered, was a decode-and-forward strategy instead of amplify-and-forward at the VLC transmitter.This implementation of EoL would use a convergence layer and create a bridgeinterface between the Ethernet and the VLC MAC/PHY layers. In this case,a transmitted Ethernet frame over the wire would be decoded and strippedof its headers in order to extract its payload data only. A new VLC frameconforming to the IEEE 802.15.7 standard [58] would be constructed and re-transmitted. The information would then be recovered at the receiver side byextracting the data from the VLC frame before being converted back to an313.3. Advantages, Drawbacks and ChallengesEthernet frame. In this approach, the robustness of the system is increasedcompared to our approach because VLC frames are designed specifically to besent over such a medium, whereas Ethernet frames were obviously not purposedfor OWC. Though, the major trade-off of this system is in the dramatic increasein complexity due to the need for a microprocessing or FPGA unit at the VLCtransmitter to perform the tasks of the convergence layer. The same applies tothe receiver in the conversion of VLC back to Ethernet PHY. Consequently,decode-and-forward was dropped in favour of a simpler amplify-and-forwardstrategy.32Chapter 4Channel ModelingBefore we can analyze the link budget of our EoL system, we need to examinethe characteristics of the overall channel and introduce models that will beused in subsequent chapters. The EoL channel is two parts: the Cat 5 cabled linkbetween the PSE and the VLC transmitter, and the VLC link to the photodiodereceiver. By having a comprehensive understanding of the Ethernet and VLCchannel, we will be able to qualitatively and quantitatively describe the effectsof distortion introduced in the channel to later be able to effectively cope with.4.1 Category 5 UTP ChannelAlong this first leg of the EoL link, the MLT-3 modulated signal is transmittedover a standard Cat 5 UTP cable. Given the fact that this is a bandlimitedchannel, consecutive symbols are bound to spread their energy to adjacent sym-bols, resulting in phase dispersion and ISI at high frequency transmissions [66].The voltage signal suffers from various attenuation and distortion effects as itis transmitted from one end to the other. In [67], the authors provide a MAT-LAB/Simulink model for a 100 m Cat 5 cable. However, since we did not wantto be limited by a model of fixed cable length, we utilized resources provided by[68] instead to model our cable link in this thesis. Individual distortions thatare modeled are the following: (1) transmit jitter, (2) waveform overshoot, (3)baseline wander, (4) cable attenuation, (5) flat loss and (6) return loss. Fig-ure 4.1 shows a block diagram of the Cat 5 cable link model, where the inputsignal is an ideal MLT-3 waveform as in Figure 2.4(a).4.1.1 Transmit JitterTransmit jitter refers to the timing offset between the expected symbol transitiontime and the actual symbol transition time. Specifically, it can be thought ofas a phase delay in MLT-3 modulation. To model this effect, a random timingoffset is applied every time a transition occurs from one level to another. Theamplitude of this timing offset is chosen from a uniform probability distributionwith zero mean and a peak-to-peak value of Jp−p ≤ 1.4 ns as specified in ANSIX3.263 [18, p31]. Although timing jitter does not create noticeable distortions,it can cause timing recovery issues at the receiver by gradually shifting thetransition edge if the amount of jitter is compounded over multiple symbols.334.1. Category 5 UTP ChannelInput TransmitJierWaveformOvershootBaselineWanderInsertionLossFlatLossReturnLoss OutputFigure 4.1: Cat 5 cable model4.1.2 Waveform OvershootWaveform overshoot models the overshooting characteristic at the AOI of theMLT-3 line driver. It defines the amount that the differential signal will exceedthe final voltage value immediately following a signal transition. A maximum of5 % overshoot of the final voltage level is specified by the standard. This effectis modeled as a second-order LPF with transfer function [68]:Hwo(f) =ω2n(j2pif)2 + 2ζωn(j2pif) + ω2n(4.1)where ζ is the damping ratio, and ωn is the undamped natural frequency. ζis required to be positive in order for the system to be stable, and is obtainedfrom the maximum overshoot percentage %OS, which is given by%OS = e−pi ζ√1−ζ2 ; (4.2)while ωn is obtained via the filter’s underdamped step responsehwo(t) = 1− Ce−ζωnt cos(ωntC+ φ)(4.3)C =1√1− ζ2(4.4)φ = tan−1 (Cζ) . (4.5)ωnt can be solved iteratively by setting hwo(t1) = 0.1 and hwo(t2) = 0.9, whichcorresponds to 10 % and 90 % of its steady state voltage and using:ωn =ωnt2 − ωnt1tr. (4.6)With a response time of tr = 3 ns and an overshoot percentage of %OS =5 %, the parameters ζ and ωn were computed to be 0.6901 and 2.8450× 108respectively.4.1.3 Baseline WanderIsolation line transformers are commonly used for coupling the digital waveformonto the physical medium. This magnetic module is intrinsically high-pass innature and rejects frequencies below 50 kHz [69, 70]. Correspondingly, low-frequency components experience more attenuation, causing the signal to decay344.1. Category 5 UTP Channel+10-1Baseline WanderFigure 4.2: Worst case baseline wanderexponentially over time (also known as droop). This is an unfortunate problemfor baseband schemes, in particular MLT-3, which is an unbalanced code thatcontains spectral energy at DC. Baseline wander is described as the signal slowlydrifting away from the average baseline as a result of staying at the same level forprolonged periods of time and is shown in Figure 4.2. Although the combinationof scrambling and 4B/5B encoding in 100Base-TX attempt to minimize baselinewander by preventing long-lasting strings of ones or zeros, it can still have asignificant effect for select strings of inputs. A worst-case scrambler setup andinput stream scenario can result in “killer packets” with 56 consecutive zerosymbols [69, p2171], and can cause the received signal amplitude to almostdouble to 4 Vp−p [20, 71].This effect is modeled as a second-order HPF to replicate the DC rejectionproperties of the channel, where the transmitter and receiver magnetics areassumed to be ideal with a 1:1 turns ratio. The transfer function of baselinewander in Cat 5 cables is given by [68]Hbw(f) =b2(j2pif)2a2(j2pif)2 + a1(j2pif) + a0(4.7)where the parameters that describe the frequency response (b2, a2, a1, a0) aredefined by the cable, transmitter and receiver’s characteristic impedances andinductances (listed under Table 4.1):b2 = L1L2ZL , (4.8)a2 = L1L2(ZS + 2ZCZL) , (4.9)a1 = ZSZL(L1 + L2) + 2ZC(ZSL2 + L1ZS) , (4.10)a0 = 2ZCZSZL . (4.11)4.1.4 Cable AttenuationCable attenuation, also referred to as propagation loss or insertion loss, is theloss of signal energy when travelling in an imperfect conductor. The amountof loss is a function of conductor length and transmission frequency, and is at-tributed to (1) the skin effect and (2) the proximity effect. The skin effect occurs354.1. Category 5 UTP ChannelLdx RdxGdx CdxdxFigure 4.3: Transmission line modelwhen the electron flow migrates towards the skin of the conductor, forming athinner layer of electrons at the outer surface. At high frequency transmis-sion, the conducting layer becomes increasingly thin along the wire. Thus, thecross-section of the conductor is decreased, while the effective resistance of theconductor is increased in turn, inflicting more loss. On the other hand, theproximity effect occurs when the current density in the wire varies as a directconsequence of the closeness of other neighbouring twisted pairs, creating mu-tual repulsion or attraction. In Cat 5 cables, these compounded effects createsevere delay spread over many symbol intervals that is proportional to cablelength and results in significant ISI [72, 73].To model these effects, the twisted pair cable is approximated to a transmis-sion line [74]. The transfer function of a perfectly terminated transmission lineloop of length l is [75]H(f, l) = e−lγ(f) (4.12)where γ(f) is the propagation constant with attenuation constant α(f) andphase constant β(f)γ(f) = α(f) + jβ(f) . (4.13)At high frequency transmissions (> 100 kHz), α and β can be approximated to[74, 76]α(f) ≈ k1√f + k2f (4.14)β(f) ≈ k3f (4.15)where the constants k1, k2, k3 are obtained from the lumped element parame-ters of the transmission line: its series resistance R, series inductance L, shuntconductance G and shunt capacitance C per unit length (Figure 4.3). Typicalvalues can be obtained from datasheets, measured experimentally or computed[77]. The full derivation of the propagation constants γ, α and β for Cat 5 UTPcables can be found in [78, p119].Hence, the transfer function of a Cat 5 UTP cable of length l can be expressedasHins(f, l) = e−l(k1√f+k2f)e−jk3l√f . (4.16)For a Cat 5 UTP, k1 = k3 = 2.265× 10−6 and k2 = 2.648× 10−11 [68]. Thefrequency-selective attenuation as a result of cables with different lengths is364.1. Category 5 UTP Channel104 105 106 107 108−35−30−25−20−15−10−50Frequency (Hz)Magnitude (dB)  10m20m50m100mFigure 4.4: Cat 5 cable attenuation by cable lengthshown in Figure 4.4. The magnitude of this propagation loss isLins(f, l) = −20 log10 |Hins(f, l)| [dB] . (4.17)As a worst case l = 100 m, the insertion loss is commonly approximated to (fin MHz) [66, 78–80]Lins(f) ≈ 1.967√f + 0.023f +0.05√f[dB] . (4.18)The 100Base-TX standard also specifies an upper limit of insertion loss for aCat 5 cable as (f in MHz) [16, p232]Llim(f) < 2.1f0.529 +0.4f[dB] . (4.19)Figure 4.5 compares the EIA/TIA limit with the worst-case loss of a typical100 m length Cat 5 UTP.4.1.5 Flat LossIn a Cat 5 link, flat loss across all frequencies is attributed to the insertion lossof (1) the transmit magnetics, (2) receive magnetics and (3) RJ-45 connectors.These three effects are modeled by the three individual terms in the flat lossequation given byHflat(f) =(1)︷︸︸︷Vout1 V(2)︷ ︸︸ ︷10−M20(3)︷ ︸︸ ︷10−0.2C20 (4.20)374.1. Category 5 UTP Channel100 101 1020510152025Frequency (MHz)Propagation loss (dB)  Worst case lossEIA/TIA limitFigure 4.5: Comparison of the worst case insertion loss (4.18) versus the limit(4.19) at 100 mwhere Vout = 950 mV is the differential output voltage, M = 0.3 dB is the flatloss margin, and C = 4 is the number of connectors in the channel.4.1.6 Return LossReturn loss refers to the amount of power loss due to signal reflections insidethe cable from impedance mismatches. Specifically, return loss measures theamount of reflected energy, relative to the incident signal energy at the receiverand can be represented as the ratio of incident power Pi to reflected electricalpower Pr:RL = 10 log10PiPr[dB] . (4.21)If the load impedance ZL and source impedance ZS are known, the return losscan be expressed as a function of the reflection coefficientΓ =ZL − ZSZL + ZS=VrVi, (4.22)RL = −20 log10 |Γ| [dB] . (4.23)In a cabled link that consists of two terminals, a transmitting source and areceiving load, two pairs of mismatch can be considered: (1) the mismatch be-tween the source and cable, resulting in near-end echo, and (2) the mismatchbetween the cable and load, resulting in far-end echo. The reflection coeffi-cient for the mismatch between the source ZS and cable impedance Z0 and the384.1. Category 5 UTP Channelmismatch between the cable Z0 and load impedance ZL respectively, areΓS =ZS − Z0ZS + Z0, (4.24)ΓL =ZL − Z0ZL + Z0. (4.25)Ideally, in a transmission line scenario, the source and load impedance shouldmatch the characteristic impedance of the transmission line for maximum powertransfer. Some loss in signal power is incurred when a fraction of the transmittedsignal is echoed back towards the transmitter if mismatches between either endare observed. Correspondingly, a slightly attenuated and distorted version ofthe original signal will be received at the load.In the Cat 5 channel, the main causes for echoing are the micro-mismatchesin impedance within the cable in addition to the macro-mismatches betweensource/cable/load impedances. The periodic impedance discontinuities withinthe cable are due to the subtle variation of physical properties of the twistedpair, like the diameter of the copper, the diameter of the insulator or the twistratio [81]. These fluctuations can be attributed to manufacturing imperfec-tions and defects, or sometimes improper installation (i.e. sharp bends). Themacro-mismatches are easier to quantify given that they depend on the physicalseparation of high-level components (e.g. RJ-45 connector). The cable’s nomi-nal characteristic impedances Z0 is 100 Ω, though it can vary between 85−115 Ω(100± 15Ω). Although the effect of return loss is not as significant as insertionloss, the cabling standard for UTP defines a return loss limit requirement forthe manufacturing of Cat 5s. In 100Base-TX, the return loss between two MDIsmust exceed (f in MHz) [16, p232]:RL(f) >{15 dB, 1− 20 MHz15− 10 log10(f/20) dB, 20− 100 MHz. (4.26)The further the measured return loss is above this limit, the less energy isreflected, and the better the power transfer (from (4.21), a high return loss isdesirable). Standard compliant Cat 5 cables will have a return loss well abovethis limit.To account for signal loss due to reflection in our channel model, the transferfunction for return loss is derived from the transmitted voltage at the source Vt,the reflected voltage at the load Vr and the incident (received) voltage at theload Vi given as [82]Hrl(f) =VtVi=Vi − VrVi= 1−VrVi= 1− Γ (4.27)where Γ is the voltage reflection coefficient as before, and Hrl(f) can be thoughtof as the voltage transmission coefficient. Since mismatches can occur at eitherside of the cable end-links, we need to consider both a source return loss HrlS =1− ΓS and a load return loss HrlL = 1− ΓL. In the case where all impedances394.2. VLC ChannelParameter Description Valuel Cable length 1− 100 mJp−p Peak-to-peak transmit jitter 1.4 nstr Response time 3 ns%OS Overshoot percentage 5 %ζ Damping ratio 0.6901ωn Natural frequency of step response 2.8450× 108L1 Transmitter open circuit inductance 350 MHL2 Receiver open circuit inductance 350 MHZC Resistance of each conductor 9.38 Ω/100 mZS Transmitter source impedance 100 ΩZL Receiver load impedance 100 ΩZ0 Cable characteristic impedance 100 Ωk1 Cable attenuation parameter 1 2.265× 10−6k2 Cable attenuation parameter 2 2.648× 10−11Vout Differential output voltage 950 mVC Number of connectors in the channel 4M Flat loss margin 0.3 dBTable 4.1: Cable model parametersin the cable link are perfectly matched (i.e. ZS = ZL = Z0 = 100 Ω), thenΓS = ΓL = 0 and HrlS = HrlL = 1, and return loss can be effectively ignored inthe overall channel transfer function.4.1.7 Overall Cable ModelGiven the all the information gathered above and using the parameters in Ta-ble 4.1, the overall transfer function Hcable(f, l) of a Cat 5 cable model of lengthl includes all the effects depicted in the cable model (Figure 4.1), and is givenasHcable(f, l) = Hwo(f)Hbw(f)Hins(f, l)Hflat(f)HrlS(f)HrlL(f) . (4.28)Figure 4.6 shows the individual frequency responses for four of the six cablemodel effects at 100 m: (1) waveform overshoot Hwo(f), (2) baseline wanderHbw(f), (3) cable attenuation Hins(f, l) and (4) flat loss Hflat(f) for a cable of100 m in length.4.2 VLC ChannelThe second portion of the EoL channel consists of the VLC link. The multipathpropagation property of the VLC transmission medium allows the link to oper-ate in both LOS and diffuse conditions. At the same time, the transmitter can404.2. VLC Channel105 106 107 108−60−50−40−30−20−10010Magnitude (dB)Frequency (Hz)  Waveform overshootBaseline wanderCable attenuationFlat lossFigure 4.6: Frequency response of individual effects in a 100 m cableVLC ChannelNoisePhotocurrentOptical PowerPled(t) Ip(t)n(t)Rhvlc(t)Figure 4.7: VLC IM/DD equivalent channel modelbe categorized as directed, non-directed or hybrid depending on the LED’s prop-erties, establishing a total of six different link configurations [59, 83]. However,the channel we are mostly interested in is the LOS, non-directed channel, whichis appropriate for general illumination. LOS between communicating devices isusually preferred to obtain higher SNR, though the link can still work withNon-Line-of-Sight (NLOS) diffuse conditions in the case of shadowing. There-fore, the LEDs used will also be non-directed in order to be able to providewide illumination coverage. This differs from IR communication since IR LEDtransmitters are usually highly directed. Due to the fact that the locations ofthe lamp fixtures are determined at the time of installation, and the receiversare mostly immobile, the VLC channel is time-invariant with an environmentassumed to be static. Though user movements in the range of tens of centime-ters per second of the receiver can occur, these are relatively slow compared tothe bit rates of the channel [62].414.2. VLC ChannelThe indoor VLC channel is adapted from conventional IR communication[59] that uses IM/DD and is modeled as a baseband LTI system (Figure 4.7)[84]:Ip(t) = RPled(t) ∗ hvlc(t)︸ ︷︷ ︸Ppd(t)+n(t) . (4.29)The drive current is directly modulated into an optical source, producing astrictly positive instantaneous optical power Pled(t). The received optical powerPpd(t) is the convolution of the radiated optical power and the impulse responseof the VLC channel hvlc(t). The photodiode converts this received power intoan an instantaneous electrical photocurrent Ip(t) directly proportional to theinput signal with a responsitivity of R (A W−1). Here, n(t) denotes AdditiveWhite Gaussian Noise (AWGN) introduced by shot noise and thermal noise thatwill be later described in Section 5.2.2). The resulting photocurrent is describedideally as a scaled version of the transmitted signal with some noise content.The VLC channel can be completely characterized by its impulse responsehvlc(t) [62] and is unique to any given VLC configuration. The propagation prop-erties of the link can be entirely defined by four key factors: (1) the emissioncharacteristics of the LEDs, (2) the receiver characteristics, (3) their positionsand orientations and (4) the layout of the environment. As previously describedin Section 2.2.2, the LEDs follows a Lambertian emitter pattern. For this rea-son, the position and orientation of the transceivers and the environment itselfplay a huge role in determining the VLC channel gain which directly affectsthe performance of the link. The impulse response not only characterizes thepath loss, but also the effects of multipath dispersion caused by the features ofthe link. Particularly, reflections of nearby surfaces or objects, which are alsoconsidered ideal Lambertian and purely diffuse, introduce a propagation delayelement. Below we explore some indoor optical wireless channel profiles devel-oped over the years in literature. Although shadowing from obstructing objectscan present an issue, we assume that the receiver always retains LOS path withthe transmitter unless otherwise indicated. Additionally, since we only aim tooperate in indoor environments, the VLC link is not affected by outdoor or at-mospheric factors such as dust, fog, rain, aerosol particles that would otherwisecreate adverse conditions.4.2.1 Wall Reflection ModelThe wall reflection model is most widely used for in-office settings. It considersthe primary LOS component and diffuse components from the reflections ofsurrounding surfaces.LOSThe unobstructed direct LOS component models the free-space path loss andis approximated as a scaled and delayed Dirac delta function. The detectedradiant flux depends on the effective collection area of the photodiode and is424.2. VLC Channelreversely proportional to the square of the distance between the receiver andthe emitter [50, 85]. The delay is attributed to the time it takes for photonsto travel from the lamp to the receiver at the speed of light c. The impulseresponse of the LOS component ishlos(t) =Ard2(m+ 1)2picosm φ︸ ︷︷ ︸L(φ) (2.3)Ts(ψ)g(ψ) cosψ rect(ψΨc)︸ ︷︷ ︸Aeff (ψ) (2.10)δ(t−dc)(4.30)where Ar is the physical area of the photodiode, m is the Lambertian order ofthe transmitter (2.4), d is the LOS distance between the LED and photoreceiver,φ is the angle of irradiance, ψ is the angle of incidence (φ = ψ if the transmitterand receiver share a common normal plane), Ts(·) is the signal transmission ofthe optical band-pass filter, g(·) is the concentrator gain and Ψc is the FOV ofthe photodiode.Alternatively, the channel can also be described in the frequency domain, bytaking its Fourier Transform [45, 84].Hlos(f) = F{hlos(t)} (4.31)=∫ ∞−∞hlos(t)e−j2piftdt (4.32)=Ard2(m+ 1)2picosm φTs(ψ)g(ψ) cosψe−j2pif dc . (4.33)At DC, the frequency response of the LOS is equivalent to the DC channel gainHlos(0) (the propagation path loss of the link) since the Fourier Transform of adelayed Dirac delta is of constant magnitude:Hlos(0) = |Hlos(f)| =Ard2(m+ 1)2picosm φTs(ψ)g(ψ) cosψ . (4.34)NLOSThe NLOS diffuse contribution is described by the multipath dispersion char-acteristics of the environment. It consists of reflections from differing opticalpaths, which is a function of the geometry of the room, spectral reflectance ofsurfaces and obstructing objects. These reflections arrive at the photoreceiverat a slightly delayed time compared to the LOS component while further at-tenuated. This attenuation depend on the transmission wavelength, the surfacematerials and angle of incidence. Figure 4.8 shows the reflectance of differentmaterials in contrast to the spectral distribution of a yellow phosphor LED.Typically, the reflecting surfaces are assumed to be ideal, purely diffuse Lam-bertian reflectors in this model. Specifically, all surfaces in the room are approx-imated as Lambertian-point source emitters that generate a new ray, scaled byits reflectivity, for each incident ray [87]. For the sake of simplicity, we considera primary, single reflection to be the dominant component and ignore shadowing434.2. VLC ChannelSpectral Reflectance of surfaces1. LED WallCeilingPlastic WallFloorWavelength (nm)Relative Spectral Power Density Distribution of LED400                500               600                700              800Figure 4.8: Phosphor LED emission spectrum compared to the reflectance spec-trum of common room surfaces [86]and furniture layout. The impulse response of all contributions from the firstreflection over all reflecting elements R is then computed as [45, 88]:hnlos(t) =R∑i=1Ar(m+ 1)ρi∆A2pid21id21icosm(φ1i) cos(ψ1i)cos(φ2i) cos(ψ2i)Ts(ψ2i)g(ψ2i)δ(t−d1i + d2ic)rect(ψ2iΨc) (4.35)where ρi is the reflectivity of the surface of small area ∆A, φ1i is the angle ofirradiance of the transmitter, ψ1i is the angle of incidence to the surface, φ2iis the angle of irradiance of the surface, ψ2i is the angle of incidence to thephotodiode, d1i is the distance from the transmitter to the reflective surfaceand d2i is the distance from the surface to the photodiode (see Figure 4.9).11dd2d1ΔAфψффψψ22Figure 4.9: Wall reflection channel geometry [86]The complete impulse response of a first-order wall reflection model hvlc(t)sums the contributions from the LOS and all of the individual reflections ashvlc(t) = hlos(t) + hnlos(t) , (4.36)444.2. VLC Channelx yz 5 m3 m5 m0.85 mReceiver PlaneLED TransmiersPhotodiodeParameter ValueRoom size 5× 5× 3 m3ρwall 0.8∆A 0.0667 m2Transmitters (-1.25,-1.25,3),(-1.25,1.25,3),(1.25,-1.25,3),(1.25,1.25,3)Photodiode (0,0,0.85)Figure 4.10: Room configuration and simulation parameters0 2 4 6 8x 10−800. 10−6Time (s)Impulse response h(t)(a) Impulse response106 108−54−52−50−48−46−44−42Frequency (f)Magnitude response |H(f)| (dB)(b) Frequency responseFigure 4.11: LOS with one reflectionwhich can be represented equivalently in the frequency domain asHvlc(f) = F{hvlc(t)} (4.37)= F{hlos(t) + hnlos(t)} (4.38)= Hlos(f) +Hnlos(f) . (4.39)We simulated a sample LOS and NLOS impulse response using the roomlayout in Figure 4.10 with corresponding parameters. The origin of the coor-dinate system is located at the center of the room on the floor. Figure 4.11(a)shows the impulse response with the dominant peak of the LOS transmission,and then the delayed diffuse components. Several smaller peaks can be observedin the diffuse response from the contribution of different delayed reflections. Inthis simulation, the power penalty of the NLOS reflections compared to the LOSwas computed to be as much as 10 dB in difference. The frequency response isshown alongside in Figure 4.11 and has a response that is relatively flat over the454.2. VLC Channelfrequency range of interest, with a periodic component at higher frequencies.Higher order diffuse reflections can be obtained via more extensive modelingtechniques, for example Monte Carlo ray-tracing [89, 90] or 3D CAD modeling[91]. In our case, we consider the first order reflection to capture the effectsof temporal dispersion well enough. This is reasonable since the first reflectionusually composes over 90 % of the received power compared to higher orderreflections [45, 88]. In addition, the computation of higher order reflectionsincreases in complexity dramatically for every extra reflection.4.2.2 Spherical ModelSimilar to the wall reflection model, the spherical model consists of a principalLOS in addition to diffuse components. The response of the diffuse portion isapproximated to an integrating sphere, which is modeled as a smooth, expo-nentially decaying function in the time domain [35, 61]. This exponential decaymay last longer than a symbol duration, and hence can distort baseband signalsat low frequencies. The complete impulse response of this channel model is thesummation of the LOS and the delayed diffuse responsehvlc(t) = hlos(t) + hdiff(t−∆T ) . (4.40)The LOS impulse response hlos(t) is as before in (4.30), while the diffuse com-ponent hdiff(t) is delayed from the LOS signal by ∆T . It decays at a rate of−1/τdiff and is expressed as [45, 46, 61, 92]hdiff(t) =ηdiffτdiffe−tτdiff u(t) (4.41)where u(t) is the unit step function, ηdiff is the power efficiency of the diffusesignal, τdiff is the exponential decay time, both of which are given byηdiff =ArAroom〈ρ〉1− 〈ρ〉sin2(FOV) (4.42)τdiff = −〈t〉ln〈ρ〉, (4.43)〈ρ〉 is the average reflectivity of the room considering the individual reflectivityρi for each reflective surface R (walls, windows, desks, etc.) with area Ai〈ρ〉 =1AroomR∑iAiρi , (4.44)〈t〉 is the average transmission delay for one reflection in a room with dimensions(x, y, z), total surface area Aroom = 2(xy + xz + yz) and volume Vroom = xyz〈t〉 =4VroomcAroom. (4.45)464.2. VLC Channel0 2 4 6 8x 10−800.511.52x 10−7Time (s)Impulse response h(t)(a) Impulse response106 108−60−55−50−45Frequency (f)Magnitude response |H(f)| (dB)(b) Frequency responseFigure 4.12: Spherical modelFigure 4.12 shows an impulse response example for this model, where we canclearly see the slow-decaying tail of the exponential diffuse component that canspan multiple symbol periods (> 8 ns in the case of 100Base-TX). The frequencyresponse is shown in Figure 4.12(b), and has a noticeably lower 3 dB frequencyaround 9 MHz compared to the previous model.4.2.3 Hayasaka-Ito ModelThe Hayasaka-Ito model [45, 93] was adapted for VLC from a non-directedwireless IR indoor diffuse link. In this model, the link operates without LOS,and instead considers a ceiling bounce to be the primary path between thetransceivers as well as diffuse components. Both the transmitting and receivingdevice are positioned on the same horizontal plane with their normal perpen-dicular to the floor (i.e. pointed upwards towards the ceiling).The diffuse component of higher order reflections hdiff is as that of the spher-ical model from (4.41) above. The total impulse response is the summation ofthe primary reflection’s impulse response h1(t) (delayed by T1) and the diffusereflections’ impulse response hdiff(t) (delayed by ∆T )h(t) = h1(t− T1) + hdiff(t−∆T ) . (4.46)The main reflection follows a gamma probability density function Γ(α) and hasa normalized impulse response given byh1(t) =λ−αΓ(α)tα−1e−tλ (4.47)where the parameters α and λ are related to the physical properties of the474.2. VLC Channelchannel, and are obtained from the following relationships:α = 4(pathpeak − pathmirrorpathslow − pathfast)2, (4.48)λ =(pathslow − pathfast)24c(pathpeak − pathmirror)(4.49)where each pathx represents a separate path travelled by different rays (see [93]for more details).4.2.4 Ceiling Bounce ModelThis ceiling bounce model was introduced by J. Carruthers and J. Kahn in 1997also for IR channels [94]. Similar to the Hayasaka-Ito model, the transceiversare aimed upwards and assume no direct LOS. Only ceiling reflections areconsidered, but in this case, without higher-order diffuse reflections. Its impulseresponse is defined by [45, 46, 59, 94]h(t) = H(0)6a6(t+ a)7u(t) (4.50)where H(0) is the DC channel gain and a is the ceiling bounce parameterrelated to the Root Mean Square (RMS) delay spread Drms of the channel(a ≈ 12√1113Drms; or a =2zc with z being the ceiling height, and c the speed oflight).4.2.5 Gfeller & Bapst ModelF. Gfeller and U. Bapst pioneered one of the very first instances of wirelessIR communication, along with a functional channel model description in 1979.Though outdated, we include the following for completeness. The channel re-sponse is depicted as a diffuse link with single reflections [45, 95]:h(t) ={2t0t3 sin2(FOV) , t0 ≤ t ≤t0cos(FOV)0, otherwise(4.51)where t0 = Lc is the minimum delay, L is the distance from the photodiode andFOV is the receiver’s field of view.48Chapter 5Link AnalysisIn this chapter, we present a complete study of the EoL link in order to quantifyhow the signal is affected as it is transmitted starting from an Ethernetinterface and eventually received at a photodiode. This is followed by a noiseanalysis of the channel and a numerical analysis to determine expected SNRlevels in an EoL system deployment.5.1 Signal TransferGiven that nearly every component can be considered linear and that the linkis also time-invariant (the cable length is chosen at installation time and theVLC terminals are mostly stationary), the overall system is approximated to anLTI system. Based on our proposed system model (Figure 3.2), we designed alink model that represents the complete signal path that is shown in Figure 5.1.As previously mentioned, the channel is composed of a hybrid wired Cat 5 anda wireless VLC link, for which comprehensive models were developed in theprevious chapter.5.1.1 Ethernet TransmitterAt the transmitter side of the cabled link, a binary signal undergoes significantmodifications prior to being transmitted: (1) 4B/5B encoding, (2) scrambling,and (3) MLT-3 modulation (from Section 2.1.1). The input signal to the twistedpair is a differential MLT-3 voltage signal with a minimum allowed amplitudeof 950 mV and a maximum of 1050 mV [18, p28]. The PSD of the transmittedMLT-3 signal was previously depicted in Figure 2.4(b). The average electricalpower emitted over the cable was calculated asP cabletx =V 2rmsZ0(5.1)= 6.49 dBm (5.2)where Vrms ≈ 0.667 V is the RMS voltage of the transmitted signal, and Z0 =100 Ω is the cable impedance.5.1.2 Cat 5 Cable ChannelThe complete transfer function of the Cat 5 channel model was obtained inSection 4.1 and given in (4.28). The received signal voltage V cablerx at the termi-495.1. Signal TransferInputDataCat 5CableHcable(f,l)Amp LEDDriverDC BiasVLCChannelHvlc(f)TIAHPFHhp(f)Amp OutputDataVtxcable VrxcableIdcIled Pled Ppd Ip VtiaVLC Transmier VLC ReceiverVhp VoutFigure 5.1: EoL link modelnating end of the cable then isV cablerx (f) = Vcabletx (f)Hcable(f, l) [V] . (5.3)5.1.3 VLC Front-end (LED Driver)At the Ethernet receiver front-end, the differential signal extracted from thebalanced twisted pair cable is combined to obtain a single electrical signal. Thisvoltage signal is then amplified by a pre-amplifier that controls the maximalswing of the driving signal to the LED. The amount of gain needs to be regu-lated to operate the LED in its linear region, and as such will differ between LEDmanufacturers and models. The signal is then converted to a driving currentIled using an analog LED driver circuit. For our purposes, we assume that thebandwidth of the current driver circuitry is well above the maximum 100 MHzbandwidth of a Cat 5 cable and is essentially a non-factor as a potential bottle-neck. The LED driver is followed by a bias tee, which adds a DC current Idc.The DC bias, which is fixed for a given illumination level, shifts the transmitsignal to be strictly positive since the optical intensity emitted by the LED can-not be negative. The instantaneous input voltage to driving current is modeledbyIled(t) = %αVcablerx (t) + Idc [A] (5.4)where % (V V−1) is the voltage gain of the pre-amplifier unit, and α (A V−1) isthe conversion factor of the LED driver.5.1.4 LED TransmitterIf the LED is operating in the linear region, then the instantaneous opticalpower radiated by an LED Pled driven by a modulated forward current Iled canbe approximated as a linear relationship [96, 97]Pled(t) = βIled(t) [W] (5.5)subject to the constraint Imin ≤ Iled ≤ Imax, where Imin (A) is the minimumdriving current needed for forward operation of the LED and Imax (A) is themaximum current before saturation. For example, in [96], the LED transfer505.1. Signal Transfercharacteristic for an OSRAM LCW W5SM Golden Dragon white LED wasapproximated to [96]Pled(t) = 0.344Iled(t) + 0.056 [W], 0.1 A ≤ Iled ≤ 1 A . (5.6)While this is an affine model, the offset term is equivalent to an additional DCbias, and thus the data-signal transmission model remains linear. Otherwise, asecond-order polynomial is usually used to model the nonlinearity effects of theoptical power versus driving current curve as described in Section 2.2.3. How-ever, since we require our system to be LTI, we ignore the nonlinearity scenarioof the optical power curve and assume an ideal LED. Furthermore, LEDs alsofeature a first-order LPF response Hled(f) as introduced back in Section 2.2.2.Depending on its bandwidth capabilities, the system can be severely hamperedby the LED. We will determine the minimum LED bandwidth required forproper operation with an MLT-3 signal in Section VLC ChannelThe transfer function of the VLC channel is completely characterized by theproperties of the transceivers and the transmission environment. We discussedseveral models that account for this in Section 4.2. The LOS VLC channelresponse is non-frequency-selective and can be interpreted as simply the DCgain. The propagation path loss of the VLC channel, which is also independentof the wavelength employed [98, p18], can be derived from the luminous pathloss of the LOS in the photometric domain [98]:Lvlc ≈gs(φ)Ar cosψd2∫ θmax0 2pigs(θ) sin θdθ∣∣∣∣∣gs(φ)=cosm θ, θmax=pi/2(5.7)≈Ard2(m+ 1)2picosm φ cosψ (5.8)= Hlos(0) (5.9)where gs(φ) is the normalized spatial distribution function of the LED, and θmaxis the LED beam maximum half-angle. When gs(φ) = cosm θ (i.e. Lambertian)and θmax = pi/2, we notice that the propagation path loss (DC channel gain) isequivalent to the magnitude response of a pure LOS |Hlos(f)| = Lvlc in (4.34)(without the optical concentrator g(·) and filter gain Ts(·) terms). If the NLOScontributions are also considered, the transfer function becomes (4.39)Hvlc(f) = Hlos(f) +Hnlos(f) . (5.10)5.1.6 VLC Receiver Front-endPhotoreceiverThe VLC front-end is composed of a photosensitive device that converts theincident optical power back into an electrical current. In the time domain, the515.1. Signal Transferoutput current is derived from the baseband convolution of the transmittedpower Pled with the VLC channel’s impulse response hvlc as in (4.29). With thephotodetector also considered as a linear device [99], the instantaneous electricalcurrent produced by the photodiode isIp(t) = RPpd(t) = RPled(t) ∗ hvlc(t) [A] . (5.11)Although the spectral responsitivity curve of a photodiode is usually not flat,the responsitivity R (A W−1) is averaged as constant over the visible spectrumfor simplicity. In the case of a single LED transmitter, the received signal opticalpower at the photodiode Ppd in the frequency domain is [45, 84]Ppd(f) = Hvlc(f)Pled(f) (5.12)= (Hlos(f) +Hnlos(f))Pled(f) [W] . (5.13)With multiple LEDs, the individual contributions of each LED along with theirreflections are summed up [84, 100]:Ppd(f) =nled∑i(Hlosi(f) +Hnlosi(f))Pledi(f) (5.14)=nled∑i(Hlosi(f) +∫wallsdHnlosi(f))Pledi(f) . (5.15)Pre-amplifier and FilteringThe TIA is typically characterized as low-pass in nature, but for the purposeof this link analysis, we assume a very large TIA bandwidth that enables a flatgain well above the 100 MHz bandwidth of a standard Cat 5 cable. The TIAconverts this photocurrent Ip to an instantaneous output voltage VtiaVtia(t) = ζIp(t) [V] . (5.16)In a typical TIA configuration such as the one in Figure 3.4, the transimpedancegain ζ is controlled by the feedback resistor value Rf and the open-loop gainG (V A−1) of the op-amp ζ ≈ G1+GRf [101, p92]. Although the TIA providesnominal amplification, the resulting signal is still heavily attenuated, owing inmajority to the heavy attenuation loss of the VLC channel (upwards of −60 dB).At the receiver, an additional gain stage (an LNA, a VGA or an AGC) is requiredto compensate for this loss and boost the signal back to detectable levels. In ourlink budget, our pre-amplifier will provide a theoretical fixed gain of κ (V V−1)Vout(t) = κVtia(t) [V] . (5.17)To filter out DC components from the luminance associated with both the LEDand ambient light, a first-order HPF with unity gain is used [50]. The cut-offfrequency for the HPF needs to be carefully considered since the transmittedbaseband signal contains significant energy near DC as a result of the spectral525.2. Noise Analysisshaping property of MLT-3. The cut-off frequency fhp will be tuned in oursimulations to offer the best performance against ambient light. The frequencyresponse of the HPF is given byHhp(f) =ff − jfhp. (5.18)5.1.7 Received SignalWith the information gathered above, the overall link budget can finally becomputed. The complete transfer function of the link without noise and priorto signal processing can be described byY (f) = X(f)Heol(f) (5.19)where X(f) is the MLT-3 voltage input signal with corresponding PSD Sx(f),Y (f) is the output signal of the system with PSD Sy(f), and the overall transferfunction Heol(f) of the LTI system with a fixed cable length l isHeol(f) = Hcable(f, l)VLC transmitter︷ ︸︸ ︷%αβHled(f)Hvlc(f)VLC receiver︷ ︸︸ ︷Hhp(f)ζκ (5.20)where Hcable(f, l) (4.28) is the Cat 5 transfer function that includes all thecable distortions, % is the VLC transmitter pre-amplifier gain, α is the LEDdriver conversion factor (5.4), β is the LED model optical gain conversion (5.5),Hled(f) is the low pass characteristic of the LED (2.5), Hvlc(f) is the frequencyresponse of the VLC channel (4.39), Hhp(f) is the HPF used for rejecting DCcurrent from ambient light (5.18), ζ is the TIA gain (5.16) and κ is the VLCreceiver pre-amplifier gain (5.17). Therefore, the received signal isY (f) = X(f)Hcable(f, l)%αβHled(f)Hvlc(f)Hhp(f)ζκ . (5.21)And we can obtain the PSD at the output of this LTI system asSy(f) = Sx(f)|Heol(f)|2 (5.22)with average received powerPy =∫ ∞−∞Sy(f)df =∫ ∞−∞Sx(f)|Heol(f)|2df . (5.23)5.2 Noise AnalysisDetermining the amount of noise introduced in the overall link is essential inorder to analyze the minimum transmission quality of our EoL system. The Cat5 cable’s noise environment is relatively well-behaved in a fixed transmissionmedium, while the VLC channel can experience many uncontrollable externalfactors. In this section, we perform an analysis using the noise model depictedin Figure 5.2.535.2. Noise AnalysisPhotodiodeShotNoiseTIAThermalNoiseLNAThermalNoiseNoiselessInputCat 5CableVLCChannelHigh PassFilterNoisyOutputCableBackgroundNoiseCircuitThermalNoiseFigure 5.2: EoL noise modelTx RxRx TxNEXT FEXTFigure 5.3: NEXT and FEXT in twisted pairs5.2.1 Noise in the Cat 5 ChannelThe ANSI/TIA-568-C.2 cabling standard [21] specifies the noise requirementsin addition to the performance characteristics for 100 Ω Cat 5 UTP cables. In100Base-TX transmissions, the main noise contributor is attributed to crosstalk.Other ambient noise sources can come from electromagnetic background whitenoise, impulse noise from nearby power lines or interfering RF signals, whichare all described as common-mode noise.CrosstalkCrosstalk defines the amount of signal energy induced by the electromagneticfield generated from a transmitting pair into adjacent copper twisted pairswithin the same cable. This unwanted signal coupling causes interference andundesirable effects on the transmission of signals in the cable. In 100Base-TX,multiple types of crosstalk are defined when full duplex transmission is employedover at least two separate twisted pairs (see [21] for details):1. Near-End Crosstalk (NEXT) measures the noise induced by the transmit-ter to neighboring pairs at the transmitting/near end (see Figure 5.3).2. Power Sum Near-End Crosstalk (PSNEXT) sums the NEXT contributionsfrom all adjacent twisted pairs.3. Far-End Crosstalk (FEXT) measures crosstalk at the receiving/far end ofthe cable.545.2. Noise Analysis4. Equal Level Far-End Crosstalk (ELFEXT) is the FEXT compensated bythe attenuation loss (ELFEXT = FEXT - Attenuation).5. Power Sum Equal Level Far-End Crosstalk (PSELFEXT) sums all of theELFEXT interference between all twisted pairs.However, in our case, uplink is currently not implemented and no signal isbeing transmitted on the received twisted pair. Therefore, NEXT/FEXT effectscan be safely ignored since the link is strictly unidirectional. In the future, witha functional uplink, crosstalk will have significant effects on the transmissionperformance of the cabled link, and would then need to be revisited.Common-mode NoiseNearby sources radiating EMI (e.g. power lines) can induce some noise on thecable that manifests simultaneously on both wires in the twisted pair, and isreferred to as common-mode noise. Due to the proximity of the two copper wiresand the twisting action keeping the pair as close together as possible, the noiseinjected on one line is duplicated in the other with equal energy and polarity.The ability to reject this common-mode noise comes with having well-balancedconnectors, cable shielding quality, the degree of balance and the twist property(ratio) of the cable. This balanced twisted pair design used in combination withdifferential signalling allows the receiver to reject almost all external common-mode noise [102].Ideally, perfect balance is required to negate all common-mode noise at thereceiver because subtracting one pulse from the other using a differential re-ceiver should produce a noiseless signal with high fidelity. However, manu-facturing imperfections or certain installation configurations (i.e. kinks) canaffect the balance of the twisted pair and cause common-mode noise to differslightly in phase and/or magnitude, and thus incur a certain level of noise tothe link. The parameter used to describe this degree of imbalance in a cableis defined as Longitudinal Conversion Loss (LCL) (or Transverse ConversionLoss (TCL)). This parameter measures the ratio between common-mode volt-age and the differential-mode voltage in dB. Unfortunately, LCL is only specifiedfor Cat 6 cables, and not for Cat 5 [21, p31]. Without loss of generality, wetreat the cable to be nearly perfectly balanced, capable of eliminating almostall common-mode induced noise. A residual background noise characterizedas Gaussian remains and is incorporated into the thermal noise of the VLCtransmitter detailed below in Section Noise in the VLC ChannelAs indicated previously, the VLC channel is modeled by a flat, baseband, AWGNchannel. The noise term n(t) in (4.29) arises from (1) shot noise generated atthe photodiode by ambient light and (2) thermal noise generated by the resistiveelectrical components of the transmitter front-end. In a VLC link, noise is the555.2. Noise AnalysisWavelength (nm)Normalized Power per Unit WavelengthPhotodiode Responsitivity (A/W)     500    600     700    800     900   1000    1100  1200   1300Sun IncandescentFluorescentPhotodiodeFigure 5.4: Responsitivity of a Thorlabs PDA10A Si-PIN photodiode comparedto the spectral power distribution of ambient light sourcesprimary limiting factor for achieving high BER. Both effects are well modeledas Gaussian [59, 84, 103, 104].Ambient Light Noise (Photodiode Shot Noise)In a typical indoor VLC setup, the room may also be illuminated by otherartificial lighting lamps (incandescent or fluorescent) or natural sunlight com-ing through the windows. The photodetector is capable of capturing any ofthese ambient radiation emitted in the visible spectrum within its responsitiv-ity spectrum (see Figure 5.4). These background radiations, which are typicallystationary or very slow time varying, generate an unwanted DC photocurrent.Ambient light manifests at the photodiode as shot noise, due to the quantummechanical nature of light, which produces a low-level background current out-put. The total amount of background current induced can depend on the season,weather, time of day, location of other lamps, position and orientation of thereceiver. Shot noise can usually be minimized by optimizing the receiver’s con-figuration, such as changing its directionality away from ambient sources, orusing optical concentrators, lenses and filters.Under direct sunlight, it is possible for the shot noise to completely saturatethe photodiode [45, 62] whereas scattered sunlight can generate a constant shotnoise current measured in the order of mA [62]. On the other hand, the lightemission from artificial illumination can produce a photocurrent of a few µA,but may also present an additional source of interference. Fluorescent lampsdriven by conventional ballasts have a natural switching frequency of a few tensof Hz and harmonics in the kHz, while those driven by electronic ballasts havefrequency components from tens of kHz and sometimes into the MHz range565.2. Noise Analysis[45, 61]. Incandescent lamps have strong frequency components typically in the50 − 60 Hz range with some harmonics [105]. These transient components cantypically be diminished by an electrical HPF [106].In OWC, shot noise is modeled as a stationary random process and approx-imated as Gaussian distributed [35, 45, 61]. For PIN photodiodes, shot noiseconsiders the light emitted by the LEDs, producing a photocurrent Ip, from theinformation signal, and that of the background radiation Ib, from both naturaland artificial sources, which is averaged by the mean background optical powerlevel emitted. The shot noise variance is given byσ2shot = 2qIp︷ ︸︸ ︷RPpdB + 2qIbI2B= 2q(RPpd + IbI2)B [A2] (5.24)where q = 1.6× 10−19 C is the electron charge, Ip = RPpd is the photocurrentgenerated by the received signal like in (5.11), I2 is the noise bandwidth factorand B is assumed to be the modulation bandwidth of the photodiode. In thecase of APDs, the shot noise variance is modified to [61, p78]σ2shot = 2qM2F (M)IbB [A2] (5.25)where M is the internal multiplication factor of the APD, F (M) is the excessnoise factor of the APD, and Ib and B are as before. We note that if we setM = 1 and F (M) = 1, the shot noise variance essentially corresponds to thatof PIN photodiodes in (5.24).VLC Receiver Thermal NoiseThe electronic noise at the receiver is independent of the received signal intensityand is usually dominated by the pre-amplifier front-end [98, p39]. Likewise, thethermal noise from the TIA is modeled as a stationary, zero-mean Gaussianrandom process. The amount of thermal noise present usually depends on thedesign of the TIA unit. For a BJT-based implementation of a TIA, the currentvariance is dominated by the feedback resistance Rf of the TIA op-amp setupand is given by [42, 61, 96]σ2thermal2 ≈Nthermal2︷ ︸︸ ︷4kBTRfB [A2] . (5.26)Conversely, the thermal noise variance for a FET-based TIA is typically lowerand is attributed to the thermal noise of (1) the feedback resistor and (2) theFET channel resistance. It has a current variance of [84]σ2thermal2 =(1)︷ ︸︸ ︷8piTGηArI2B2 +(2)︷ ︸︸ ︷16pi2kBTΓgmη2A2r I3B3 [A2] (5.27)575.2. Noise Analysiswhere kB = 1.381× 10−23 J K−1 is the Boltzmann constant, T (K) is the am-bient room temperature, η (F/cm2) is the photodetector capacitance per unitarea, Ar (m2) is the detector area, G is the gain of the TIA, I2, I3 are noisebackground factors, Γ is the FET channel noise factor and gm (S) is the FETtransconductance.Other VLC Noise TypesOther types of noise that may be considered in a VLC link can be:1. Photodiode dark current noise: shot noise generated when no photons aredetected by the photodiode.2. Photodiode excess noise: when the photodiode has an internal gain factor,usually present in APDs.3. Optical excess noise: usually restricted to laser communications and canbe ignored in VLC.4. Gate leakage: leakage current of the detector.5. 1/f noise: low frequency flicker noise in semiconductor components witha power density spectrum falling at 1/f .6. f2 noise: due to the detector capacitance and input impedance of theamplifier [107].Since our indoor channel is continuously lit by ceiling lamps during operationand the received signal current is much larger than the dark current, this canbe safely ignored. Additionally, the noise variance is typically dominated bythe photodiode shot noise and electronics thermal noise in VLC links and othernoise sources can usually be neglected.Total VLC NoiseThe total VLC noise variance is the sum of photodiode shot noise (5.24) andthermal noise (5.27), both Gaussian processes [84]:σ2total = σ2shot + σ2thermal2 (5.28)= 2q(RPr + IbI2)B +8piTGηArI2B2 +16pi2kBTΓgmη2A2I3B3 [A2] . (5.29)5.2.3 Other Electronic Thermal NoiseTransmitter Circuitry Thermal NoiseThe random agitation of electrons in conductors and resistive components pro-duces unwanted thermal noise from both the cable and the analog circuitrycomponents of the Ethernet receiver/VLC transmitter. This thermal noise can585.2. Noise Analysisbe modeled as a noiseless resistor and a voltage source with a noise varianceσ2thermal and is proportional to the ambient temperature and equivalent resis-tance Req of the noise-creating source.In our case, the Ethernet cable has an impedance of Z0 = 100 Ω seen fromthe receiver and the receiver impedance is matched at ZL = 100 Ω, then 50 %of the background noise power generated by the cable is taken into account.Since the receiver electronics also causes such noise with an equivalent resistanceReq = ZL, then 50 % of the noise is also transferred into the receiver. Therefore,this electronics thermal noise, which, in our case, includes the cable backgroundwhite noise, is naturally modeled as Gaussian distributed, with voltage variance[108]σ2thermal1 = 2∣∣∣∣ZLZL + Z0∣∣∣∣24kBTReqB (5.30)= 2∣∣∣∣100100 + 100∣∣∣∣24kBTReqB (5.31)=Nthermal1︷ ︸︸ ︷2kBTReqB [V2] (5.32)where B is the noise bandwidth as before, and Nthermal1 (V2/Hz) is the single-sided noise power spectral density.Receiver LNA Thermal NoiseAn LNA is employed post-filtering (the last gain stage in the receiver in Fig-ure 5.1) and introduces some additional electrical thermal noise that is enhancedat the front-end output. Given an LNA with noise figure Fn, then the varianceof the thermal noise voltage is [109, p247] [110]σ2thermal3 =Nthermal3︷ ︸︸ ︷4kBTRLFnB [V2] (5.33)where RL is the load resistance, and the noise figure Fn is the ratio of inputand output SNR, which is usually in the order of a few dB and can be obtaineddirectly from manufacturer datasheets.5.2.4 Total NoiseThe residual background noise and the thermal noise generated at the VLCtransmitter Nthermal1 (5.32) becomes coloured noise as it is amplified and at-tenuated along the link. At the receiver, the photodiode and receiver front endadds a shot noise Nshot (5.24) and thermal noise components Nthermal2 (5.27),Nthermal3 (5.33). The total noise power spectral density at the LNA output,prior to equalization is then given asSn(f) = Nthermal1 |%αβHled(f)Hvlc(f)Hhp(f)ζκ|2+ (Nshot +Nthermal2) |Hhp(f)ζκ|2 + σ2thermal3κ2 .(5.34)595.3. EoL SNR5.3 EoL SNRWe define the EoL SNR as the ratio between the received signal PSD Sy(f)(5.22) and the overall noise PSD Sn(f) (5.34)SNR(f) =Sy(f)Sn(f)(5.35)=Sx(f)|Heol(f)|2Nthermal1 |%αβHled(f)Hvlc(f)Hhp(f)ζκ|2+ (Nshot +Nthermal2) |Hhp(f)ζκ|2 + σ2thermal3κ2. (5.36)The contributions of the cable background white noise and transmitter thermalnoise (Nthermal1) was determined to be negligible compared to the total noisepower at the receiver output due to the fact that the average electrical SNR ofthe cabled link is > 90 dB if crosstalk and the majority of common-mode noiseare neglected. As a result, we can consider the receiver noise to be dominatedby the VLC shot noise Nshot, the TIA thermal noise Nthermal2 and the LNAthermal noise Nthermal3. The SNR can be approximated toSNR(f) ≈Sx(f)|Heol(f)|2(Nshot +Nthermal2) |Hhp(f)ζκ|2 + σ2thermal3κ2. (5.37)The average SNR is then obtained asSNR =PyPn=∫∞−∞ Sy(f)df∫∞−∞ Sn(f)df(5.38)=∫∞−∞ Sx(f)|Heol(f)|2df∫∞−∞ ((Nshot +Nthermal2) |Hhp(f)ζκ|2) df + σ2thermal3κ2. (5.39)5.4 Link Budget Numerical ExampleWe developed the following numerical example in order to assess the averageSNR levels expected in a typical EoL deployment in an indoor setup. Here, aPSE is provisioning a single array of ceiling-mounted LEDs using a regular Cat5 UTPs of length 10 m. The VLC link for this scenario is characterized as ashort distance, Single-Input Single-Output (SISO) LOS link without reflections.The photodiode is positioned directly below the LED array a few meters apart.Throughout this section, the LED model used is an OSRAM LCW W5SMGolden Dragon white LED [96, 111] and the photodiode used is a ThorlabsPDA10A [98, 112]. The devices’ parameters are listed in Table 5.1, and arepulled from their datasheets. The LED has a bandwidth of about 20 MHz.The PDA10A photodiode possesses a large enough bandwidth at 150 MHz suchthat it would not be the limiting factor in the overall system. This specificphotodiode also includes an integrated TIA of fixed gain.To simplify calculations, the following assumptions were made. In the Ether-net link, the source and load impedances (ZS, ZL)were perfectly matched with605.4. Link Budget Numerical Examplethe characteristic impedance of a UTP loop Z0 = 100 Ω. Accordingly, thevoltage reflection coefficients ΓS = ΓL = 0 and, the return loss factors canbe neglected HrlS = HrlL = 1. In the VLC link, the LEDs are operating atIdc = 350 mA and do not swing beyond the linear region, neglecting any non-linearity effects. The LED array is approximated to a point source for fastersimulations, and provides a minimum amount of luminosity required for officeillumination (300− 1000 lx recommended by the International Organization forStandardization (ISO) standard [103]). The NLOS diffuse contributions of re-flections were ignored for the path loss as the link is purely LOS. The VLC linkis subject to very low ambient light noise, induced by nearby fluorescent lamps.Numerical values for this example scenario are listed under Table 5.1 and 5.2.Description ValueEthernet TransmitterSource impedance ZS = 100 ΩTransmit voltage V cabletx = 2 Vp−pTotal transmitted elec-trical powerP cabletx = 6.49 dBmCat 5 CableCharacteristicimpedanceZ0 = 100 ΩCable length l = 10 mInsertion loss Lins = −0.52 dBFlat loss Lflat = −1.55 dBReturn loss LrlS = LrlL = 0 dB (perfect impedance matching)Total received electri-cal powerP cablerx = 4.43 dBmVLC TransmitterLoad impedance ZL = 100 ΩAmplifier gain % = 1 V V−1LED driver voltage-current conversion fac-torα = 0.350 A V−1LED driver DC bias Idc = 350 mALEDModel OSRAM LCW W5SM Golden Dragon white LEDModulation bandwidth f3dB = 20 MHzHalf-power angle Φ1/2 = 60◦ (Lambertian order m = 1)Luminous flux φv = 75 lm @ 350 mACenter luminous inten-sityI0 = 27.3 cdCurrent-optical powerconversion factorβ = 0.344 W A−1Continued on next page615.4. Link Budget Numerical ExampleTable 5.1 – Continued from previous pageDescription ValuePhotodiodeModel (type) Thorlabs PDA10A (Si-PIN)Collection area Ar = 0.8 mm2Responsitivity R = 0.45 A W−1FOV Ψc = 60◦Refractive index n = 1.5 (acrylic)Optical concentratorgaing(ψ) = 3Optical filter gain Ts(ψ) = 1VLC ChannelVLC channel profile Indoor SISO LOS short distanceNumber of LEDs nled = 600Angle of irradiance φ = 0◦Angle of incidence ψ = 0◦Distance d = 2.15 mVLC propagation lossLvlc =Ard2(m+ 1)2picosm φTs(ψ)g(ψ) cosψ= 1.65× 10−7= −67.8 dBTotal transmitted opti-cal powerPled = 50.2 dBmTotal received opticalpowerPpd = −17.6 dBmVLC ReceiverTIA fixed gain ζ = 5× 103 V A−1LNA gain κ = 12.2 V V−1HPF 3 dB frequency fhp = 100 kHzTotal Received PowerTotal output electricalpowerPy = 4.0 dBmTable 5.1: Sample power transfer analysisDescription ValueCat 5 Cable NoiseCable bandwidth B = 100 MHzVLC Transmitter Thermal NoiseAmbient temperature T = 290 KReceiver equivalent in-put resistanceReq = 100 ΩContinued on next page625.4. Link Budget Numerical ExampleTable 5.2 – Continued from previous pageDescription ValueElectronics thermalnoiseσ2thermal1 = 8.01× 10−11 V2Thermal noise totalpowerPthermal1 = −71.0 dBmPhotodiode Shot NoisePhotodiode bandwidth B2 = 150 MHzPhotocurrent Ip = 412 nABackground radiationcurrentIb = 2µA (low noise from Table 5.3)Noise bandwidth fac-torI2 = 0.562Photodiode shot noise σ2shot = 2.43× 10−16 A2FET TIA Thermal NoiseTIA open-loop voltagegainG = 10Fixed capacitance ofphotodiode per unitareaη = 1.12× 10−6 F/m2FET channel noise fac-torΓ = 1.5Noise bandwidth fac-torI3 = 0.0868FET transconductance gm = 30× 10−3 STIA thermal noise σ2thermal2 = 1.21× 10−16 A2Total VLC NoiseVLC noise σ2total = 3.65× 10−16 A2VLC noise total power Ptotal = −124 dBmLNA Thermal NoiseLoad resistance RL = 1000 ΩAmplifier Noise Figure Fn = 3 dBLNA thermal noise σ2thermal3 = 4.79× 10−9 V2LNA thermal noise to-tal powerPthermal3 = −53.2 dBmTotal Received NoiseTotal noise power Pn = −45.6 dBmTable 5.2: Sample noise analysis635.4. Link Budget Numerical Example0 0.05 0.1 0.15 0.2−100−80−60−40−2002040Frequency (GHz)Power/frequency (dB/Hz)(a) Desired signal0 0.05 0.1 0.15 0.2−100−90−80−70−60−50−40−30Frequency (GHz)Power/frequency (dB/Hz)(b) NoiseFigure 5.5: Received PSDsFrom the received signal PSD plot in Figure 5.5(a), we can observe that the100Base-TX signal is able to retain the general shape of the original signal PSD(Figure 2.4(b)). This is expected because the majority of the signal energy iscontained below the fundamental frequency (31.25 MHz) of the MLT-3 signal,at around 16 MHz, which is also below the modulation bandwidth of the LEDat 20 MHz, and well below the insertion loss 3 dB frequency of a 10 m Cat 5UTP cable around 51.8 MHz (see Figure 4.4). The average electrical powerlost in the cabled portion of the EoL link as a result of cable attenuation isLins + Lflat = −2.1 dB at 10 m. Although, this can rise upwards of −10.4 dBif a 100 m cable is used. The majority of the system loss is attributed to thepropagation path loss of the VLC link at 67.8 dB, and is largely due to the smallsize of the photodetector area compared to the separation distance betweenthe transmitter and receiver. The 3 dB frequency of the receiver HPF wasset to 100 kHz to offer the best compromise between an initial slow decay andpreserving signal power near DC. The total received noise PSD is shown inFigure 5.5(b), which we can observe is mostly white. We also observed thatthe electromagnetic background thermal noise of the Ethernet link was too lowto have a significant contribution in the total noise. The coloured thermalnoise from the cabled link and transmitter circuitry is heavily attenuated whentransmitted over the VLC channel and consequently can mostly be ignored.From this specific low ambient light scenario, we determined that the totalnoise at the receiver is dominated by the LNA thermal noise. However, this isonly the case for low shot noise at the photodiode (Ib = 2µA). The total noiselevel is usually highly contingent on the amount of ambient light present. Using(5.39), the average electrical SNR for this simple short-distance LOS EoL link645.4. Link Budget Numerical ExampleNoise level Description Ib (µA)Low noise Fluorescent lamp with filter 2 [113]Medium noise Indirect sunlight without filter 740 [113]High noise Direct sunlight without filter 5100 [113–115]Table 5.3: Background current values from ambient sourcesis calculated asSNR =PyPn(5.40)= 4.0− (−45.6) = 49.6 dB . (5.41)To achieve minimal BER transmission in this EoL scenario, an experimentaltarget SNR above 30 dB (see Figure 6.6(a)) was determined to work with mostpost-equalization setups (this will be expanded in Chapter 6). Then, the SNRlink margin was obtained to be approximately 19.6 dB, and the received powertarget was −15.6 dB for this setup. For ways to improve the link SNR margin,we can explore multiple options that do not involve changing the room’s config-uration. At the receiver side, the usage of an APD over a PIN photodiode cantypically increase the sensitivity of the photodiode by a few dB. Alternatively,increasing the surface area of the photodiode entails a larger photon collectionarea, also increasing the received optical power, although at the expense of in-creased shot noise. The usage of better optical filters or electrical filters canusually reduce the amount of ambient light absorbed, and hence the amount ofshot noise generated. At the transmitter side, we can choose to either deploymore LEDs per fixture or opt for an LED model with higher optical output toincrease the transmitted optical signal power. The overall quality and perfor-mance of the VLC link depends heavily on the capabilities of the transmitterand receiver pairing. The total link budget is shown as a graphical plot inFigure Estimated SNR Values in Typical EoLConfigurationsThe numerical example above used an optimistically low amount of shot ambientnoise in the VLC link, attributed strictly to some nearby fluorescent lamps.Under different conditions, the background photocurrent can reach up to a fewmA depending on the radiation source and the physical property of the indoorroom (e.g. presence of large windows). Table 5.3 lists some typical values forthe shot noise photocurrent Ib found in literature.We also developed a total of three VLC channel profiles that will be usedwith our EoL link to determine the SNR levels in different scenarios with thesimulation parameters listed under Table 5.4. Profile A is a simple pure LOSlink without reflections using a single LED transmitter, and follows the same pa-rameters from the numerical example developed for the link budget in Table 5.1.655.4. Link Budget Numerical Example−60−40−200204060Ptxcable=6.46dBmPrxcable=4.4dBmPled=50.2dBmPpd=−17.6dBmPy=4dBmPn=−45.6dBmTarget Py=−15.6dBmLins=−0.518dBLflat=−1.55dBVLC TX gainLvlc=−67.8dBVLC RX gain SNR margin=19.6dBTarget SNR=30dBPower (dB)Figure 5.6: Link budgetProfile B uses the spherical VLC channel model that includes a slow-decayingdiffuse reflection component (from Section 4.2.2). This link also employs a singletransmitter located on the ceiling in a typical 5 × 5 × 3 m3 office room config-uration, and a reflection time delay of ∆T = 1× 10−8. Profile C depicts ascenario more representative of a real world scenario since indoor environmentsare typically illuminated with multiple arrays of LEDs. This scenario employsfour LED luminaires distributed around a room to provide uniform illumina-tion. The VLC channel model considers the LOS and the first reflection of thewall reflection model of Section 4.2.1. An increased number of high power emit-ters results in a stronger LOS component, while also contributing additionalreflections. In all three scenarios, the photodetector is located at desk height(0.85 m) in the middle of the room. The same LED and photodiode componentsare reused from the numerical example in Table 5.1. Using these three profiles,we computed average EoL SNR values that would be expected in a typical EoLdeployment under different noise and VLC channel conditions. The backgroundradiation is assumed to be uniformly distributed across the room and the resultsare obtained in Table 5.5.For Profile C, we also simulated the SNR distribution across the room atlow noise. The location of the receiver is varied along the xy-plane of the roomx ∈ [−2.5 m, 2.5 m], y ∈ [−2.5 m, 2.5 m] at a fixed height. The SNR distributionin the room as a function of the receiver’s location is presented in Figure 5.7. Asanticipated, the SNR is distributed unevenly around the room with the highestSNR values observed directly underneath the luminaires at an electrical SNR665.4. Link Budget Numerical ExampleProfile A Profile B Profile CRoom size 5× 5× 3 m3Surface reflectivity ρwall = 0.8, ρceiling = 0.4, ρfloor = 0.6 (Figure 4.8)Receiver location (0,0,0.85)Number of lamps 1 1 4Number of LEDsper lamp600 600 256Transmitter (0,0,3) (0,0,3) (-1.25,-1.25,3),location(s) (-1.25,1.25,3),(1.25,-1.25,3),(1.25,1.25,3)VLC channel model LOS only LOS spheri-cal modelLOS with 1 re-flectionReceived photocur-rent Ip7.87× 10−6 A 9.42× 10−6 A 5.05× 10−6 ATable 5.4: SNR simulation parametersProfile A Profile B Profile CLow noise 49.6 dB 52.1 dB 47.6 dBMedium noise 35.2 dB 36.7 dB 31.4 dBHigh noise 26.9 dB 28.4 dB 23.1 dBTable 5.5: Typical average SNR levels obtainedof 50.6 dB, and the worst case SNR at the corners of the room with values of38.8 dB. An average SNR of 49.0 dB is perceived in the room for this particularsetup. Because a minimum level of illumination must be maintained indoors,the sum of all contributions from the LOS and NLOS reflections of a distributedLED configuration usually result in a very high received power [11, 24, 28, 100].This is ideal for our situation, since our system can make use of spectrallyefficient multi-level PAM schemes (like MLT-3) that usually require high SNRlevels to operate at adequate BER.675.4. Link Budget Numerical Example−202−2023540455055X location (m)Y location (m)SNR (dB)Figure 5.7: SNR distribution in a room68Chapter 6EqualizationISI is a common impairment to practical communication channels which isdescribed as successive symbol transmissions interfering with one another.This is due to the linear amplitude and phase distortions from transmittinga signal through a bandlimited and dispersive channel, such as our EoL link.Since it is not possible to modify either the coding and modulation scheme inour EoL amplify-and-forward approach, we are limited to utilizing equalizationstrategies to compensate for the deficiencies observed in our channel. In thischapter, we explore select equalization techniques as the primary defence againstISI and noise in the context of our EoL system. First, we investigate the primarylocations where ISI is introduced in the link. Then, we examine two types ofsub-optimal equalization: linear equalization and decision feedback equalization.We also discuss adaptive equalization, and employ these tools to mitigate ISIand improve the performance of the system.6.1 ISI in EoLThe EoL system is composed of two separate bandlimited links: a wired channeland a wireless (visible light) channel that also introduce unavoidable distortionsand noise impairments. The principal causes of LTI distortions in EoL areattributed to (1) the insertion loss of the twisted pair link, (2) the bandwidthlimitations of the LED and (3) the multipath propagation property of the VLCchannel. These effects were previously modeled in the overall transfer functionthat encompasses the whole system in (5.20) which allowed us to view thishybrid system equivalent to a single transmitter source, communication channeland receiver model.Firstly, the twisted pair medium introduces attenuation and temporal dis-persion depending on the length of the cable used. ISI is caused by frequency-selective insertion loss as the signal travels along the twisted pair at high fre-quency (Section 4.1.4). A common method to visualize the effects of ISI is byexamining the eye diagram of the received signal using an oscilloscope. Aneye diagram is a useful tool for qualitatively measuring the amount of symbolspreading of adjacent pulses, and can also be used towards determining the per-formance of the communication link in regards to noise immunity. The largerthe eye opening, the better the signal performs subject to noise and producesfewer decoding bit errors. ISI causes a narrower eye opening, which in turn,leads to a lower noise tolerance. Figures 6.1(a), 6.1(b), 6.1(c), 6.1(d) depict the696.1. ISI in EoLTime (s)Amplitude (AU)0 0.5 1 1.5 2x 10−8−1.5−1−0.500.511.5(a) InputTime (s)Amplitude (AU)0 0.5 1 1.5 2x 10−8−1.5−1−0.500.511.5(b) 10 m outputTime (s)Amplitude (AU)0 0.5 1 1.5 2x 10−8−1.5−1−0.500.511.5(c) 50 m outputTime (s)Amplitude (AU)0 0.5 1 1.5 2x 10−8−1.5−1−0.500.511.5(d) 100 m outputFigure 6.1: 100Base-TX eye diagramseye diagram of an input MLT-3 signal compared to the output signal at 10, 50and 100 m of the cable. The eye pattern is shown to contract with increasingcable length. At a maximum cable length of 100 m the amount of distortionsessentially completely closes the eye openings. In PAM schemes, the requiredSNR depends on the number of modulation levels, since a ternary-PAM necessi-tates higher SNR requirements than an equivalent binary PAM to realize similarBERs. Although neither the ANSI standard nor Clause 25 specify a minimumBER for 100Base-TX Ethernet over Cat 5 UTP, it is commonly assumed inliterature that a target BER of 1× 10−10 is considered reasonable [116, p12][70, p201] [104, p53] [20].In VLC, a major source of distortion is attributed to the LED itself. The datarate of 100Base-TX far exceeds the capabilities of conventional phosphor LEDs.706.2. Equalization MethodsDFEMMSEZFE MLSEEqualizerLinear Non-LinearML SymbolDetectorFigure 6.2: Family of equalizersSpecifically, the rise and fall time of such an LED does not couple properlywith those of the MLT-3 signal. The slower transient response of the LEDprevents the system to switch at a required time, causing the pulse to spreadover multiple symbol intervals, and thus symbol overlap. Until LED technologyachieves switching rates fast enough for high-speed transmission, we have todeal with the shortcomings associated with today’s LEDs. Again, we reiteratethe selection of the model and type of LED is an important consideration in thesystem’s performance, as its bandwidth can greatly bottleneck the VLC link.In addition to the LED, the multipath propagation nature of the VLC chan-nel lends additional ISI. Photons emitted by an LED transmitter may reflectmultiple times from the diffuse reflections on the walls, floor, ceiling and/orother objects. The delay propagation from these differing optical paths arriveat the receiver at a slightly delayed time following the LOS component. With adistributed LED transmitter setup, like the one in Figure 4.10, an optical pulsetransmitted by a more distant luminaire would also reach the photodetector ata later time compared to closer luminaires. These diffuse components will in-evitably cause interference with the primary LOS pulse of future symbols. Thismultipath dispersion leads to a transmitted pulse to spread to adjacent symbols,and a power penalty on the system driven by ISI. Although in this case, thedistortions associated with multipath propagation of the channel carries a farlesser effect on the channel compared to those from the LED’s bandwidth [25].6.2 Equalization MethodsWe previously established that our EoL link not only suffers from various ef-fects that distort the transmitted signal, but also from the influence of noise(described in Section 5.2). The idea then is to reconstruct the original signalby compensating for channel distortion, which is deterministic in nature, whilealso minimizing the impact of random and unpredictable noise through the useof equalization.We focus primarily on two popular types of sub-optimal equalizers: LinearEqualizers (LEs), and Decision Feedback Equalizers (DFEs) (see Figure 6.2).The equalizers we consider are implemented digitally using Finite Impulse Re-sponse (FIR) filters as these are more robust to quantization errors, more sta-716.2. Equalization MethodsT T Tw0k w1k wNkykyk-1 yk-NΣrkDigital InputEqualized Outputdk-k0^Figure 6.3: FIR LE block diagramble, and also easier to use with adaptive equalization than Infinite Impulse Re-sponse (IIR) filters in practice. The disadvantage is that finite length equalizersdo not completely eliminate ISI as a residual ISI can remain. This unequalizedISI can be minimized by increasing the number of taps, though at the expenseof increasing the complexity of the equalizer’s implementation.6.2.1 Linear EqualizersLEs are sub-optimal receivers that are highly effective at eliminating less severeISI. Figure 6.3 shows a linear filter implemented as an FIR equalizer with afinite number of weighted tap delay lines. In a digital implementation, currentvalues yk and past values yk−1, . . . , yk−N (sampled at symbol duration T ) aremultiplied with the N + 1 filter coefficients w = [w0, . . . , wN ] and summed toproduce an equalized outputrk =N∑n=0wnkyk−n , (6.1)based on which we decide on the data symbol dˆk−k0 , where k0 ≥ 0 is theoptimal equalizer decision delay. The filter coefficients are commonly optimizedaccording to the Zero Forcing (ZF) or Minimum Mean-Square Error (MMSE)criterion, though, compared to the ZF-LE, the MMSE criterion offers a greaterbalance between reduction in ISI and noise enhancements. The drawbacks ofeither linear strategies are the possibility of noise enhancement and the drop inperformance with a deeper fading frequency selective channel. In the case wherethe channel and the stochastic properties of the transmitted signal and noise areunknown or unavailable, the use of an adaptive algorithm based on the receivedsignal sequence, for instance Least-Mean Square (LMS) (Section 6.2.3), wouldbe used instead.6.2.2 Decision Feedback EqualizersDFEs are usually the more popular option for digital equalization, in part dueto their superior performance in dealing with severe ISI and ease of design in726.2. Equalization MethodsFilterFilterQuantizer Decision∑∑yk dk-k rkFeedforwardFeedback++--Detectorek Error signalDigital Input Equalized Output0Figure 6.4: DFE modelhardware compared to more complex solutions. A DFE enhances the equaliza-tion potential of an LE by inserting a supplementary causal linear filter in thefeedback path. A conventional digital DFE, depicted in Figure 6.4, is composedof two FIR filters, a Feedforward Filter (FFF) and a Feedback Filter (FBF),along with a quantizer/decision slicer unit. The FFF acts similarly to an LEfilter, which tries to shape the channel output signal into a causal minimum-phase signal and the FBF is driven by the slicer’s output symbol to assist inestimating the current symbol. The decision slicer unit makes a symbol decisionand feeds the output back to be filtered by the FBF. While the FFF compen-sates for pre-cursor ISI and, to a certain extent, some post-cursor ISI, the FBFcan only deal with post-cursor ISI. The FBF’s role is to cancel any trailing ISIcaused by previously received symbols from the current detected symbol. Usingthis non-linear setup, ISI induced on future symbols can be estimated and sub-tracted out before the detection of subsequent symbols. A major advantage ofthe DFE is that it does not suffer from noise enhancement problems observedin LEs since the slicer effectively eliminates noise prior to the feedback loop. Ingeneral, this translates to DFEs being able to provide significantly lower BERcompared to LEs. We note that a DFE is only optimal when correct decisionsare made. Decision errors in the feedback loop can cause severe error propaga-tion in a number of future decisions, which can result in nominal performanceloss compared to error-free feedback.A typical digital FIR DFE structure is shown in Figure 6.5, where c =[c0 . . . cNFF ] are the (NFF + 1) coefficients of the FFF and F = [F1 . . . FNFB ] arethe NFB coefficients of the FBF. While the FFF can be fractionally spaced, theFBF must have a tap spacing equal to the symbol interval. The input to thedecision slicer rk is weighted by the filter coefficients and multiplied with thecurrent values yk and the past decisions dˆk [84]rk =NFF∑n=0cnyk−n −NFB∑m=1Fmdˆk−k0−m . (6.2)In order to determine the tap weights, the DFE can also make use of the ZF or736.2. Equalization Methodsrkdk-1dk-NFBFNFBT T Tc0 c1 cNFFykyk-1 yk-NFFΣ dk-k0Digital InputEqualized OutputT T TF1Figure 6.5: FIR DFE block diagramMMSE criterion in conjunction with an adaptive algorithm. The error betweenthe slicer output and the actual output can be used to adjust the tap weightsof the FFF and FBF (as seen in Figure 6.4). In a typical adaptive MMSE-DFE setup, the weights are modified to the channel characteristics through thefeedback loop, and retroactively tuned through the use of an LMS algorithm.The filter weights are usually estimated using a training sequence with the goalof mapping the corresponding system transfer function. When the weights haveconverged, the output of the DFE should be close to the hard decision of outputs.6.2.3 Adaptive EqualizersIn a practical communication system, the channel dynamics are usually notaccurately known a priori by the receiver, or may also be slow time-varying.Hence, there is the need to estimate the proper channel characteristics anddesign adaptive equalizers that can automatically adjust to differing channelconditions. Adaptation refers to the adjustment of filter coefficients, and can beperformed either once at startup during an acquisition phase, or continuouslyto changing channel conditions. In the former, the initial coefficients are ob-tained via a training process using a known signal vector; this is also referredto as preset equalization. In the latter, adaptation can be accomplished duringdata transmission using periodic training signals, decision-directed adaptationor blind adaptation. Once adapted to the channel response, an equalizer willbe capable of tracking minute changes to the channel response [117].Various algorithms can be utilized to find an optimal set of filter coefficient,for example LMS or Recursive Least Squares (RLS), both well-known and ro-bust. Although the RLS algorithm offers faster convergence compared to LMS,it is computationally more complex. The LMS algorithm, which will be usedsubsequently, is based on the concept of steepest descent, and uses the errorsignal (the difference between the decision output and actual input) to indicatethe direction that the coefficients should move towards. The stepsize µ is theonly parameter that controls the adaptation rate and determines the magnitude746.2. Equalization MethodsFFF taps FBF taps LMS Stepsize Source5 20 2e-4 [20]8 12 - [71, 119]12 50 - [118, p243]13 15 2e-7 [70, 120]Table 6.1: 100Base-TX DFE propertiesof changes in coefficients. It needs to be large enough such that the equalizercoefficients converge in a reasonable amount of time but also small enough toensure that the coefficients are accurate. In both the LE and the DFE, thecoefficients wk (ck and Fk for DFE) are updated using the error signal and thedesired symbol vector xk asek = rk − dˆk−k0 , (6.3)wk+1 = wk − µekxk . (6.4)6.2.4 Equalization for EoLIn 100Base-TX Ethernet, equalization is used to mitigate the frequency-dependentattenuation and phase distortion caused by the limited bandwidth of the Cat 5cable. This bandwidth is inversely proportional to the cable length; the longerthe cable, the more compensation is necessary. Because Ethernet is required tosupport cable lengths up to 100 m, we have to be careful not to over-compensatefor short cables and under-compensate for longer cables. Most commonly, digitaladaptive DFEs are the preferred equalization option for commercial 100Base-TX receivers. Table 6.1 shows some DFE implementation found in literature.100Base-TX receivers typically use a fewer number of FFF taps over FBF tapsprovided that there is less pre-cursor ISI to compensate but more post-cursorISI. Since there is no pre-defined training sequence in 100Base-TX operation,the equalizers instead adapt the coefficients during IDLE time using blind equal-ization [118, p249] when there is no data being transmitted. These filter tapsare sometimes pre-loaded with common values appropriate for Cat 5 to ensureconvergence to desired final values.To compensate LED-induced ISI, pre-equalization [39, 63, 98, 121–123] orpost-equalization [43, 49, 122] strategies have been explored. The idea behindboth techniques is to equalize for the low-pass frequency response of the LED byutilizing the inverse transfer function of the LED (Heq(s) = 1Hled(s) ). Practicallyspeaking, this method is not necessarily desirable because the filter would haveto be adapted to accommodate for all types and models of LEDs availableon the market. Multipath ISI from the VLC channel delay dispersion is alsomitigated via digital equalization, usually also using DFEs, that offer significantimprovement over passive analog equalizers [84, 117].In the context of EoL, the configuration of the Ethernet and VLC channel,and the properties of the VLC transmitter-receiver pair require the equalizer to756.3. Simulation Resultsbe able to work in various conditions. Even though our EoL channel is consid-ered static and time-invariant, a fixed equalizer is inflexible in accommodatingdifferent types of installation setups such as wire lengths or VLC configurations(i.e. components, room layout). In addition, since the VLC channel can be sus-ceptible to slow change at time scales of about tens to hundreds of milliseconds(e.g. mobile receiver), the usage of adaptive equalizers will be able to adjustaccordingly, and therefore is a more attractive option. The objective is then toadapt these filters tap coefficients to the EoL channel. If the impulse responseis known a priori at the receiver (e.g. from channel measurements), the adap-tive equalizer can appropriately adapt the coefficients once during link setupto compensate both for channel distortions and for noise. In the case wherethe impulse response is unknown, training can be performed during an initialacquisition phase, and then periodically during IDLE periods. A prefix trainingsequence of known symbols can be used in conjunction with adaptive filters tolearn the characteristics of the channel. Again, due to the fact that the cablelengths are fixed at the time of installation, channel estimation need only beperformed periodically to accommodate the possibility of a slow varying VLCchannel. Training symbols can then be transmitted infrequently to adapt tochanging channel conditions.6.3 Simulation ResultsWe performed a variety of simulations using the sub-optimal equalization meth-ods described previously for our EoL link. In total, we designed three equalizersand ran simulations in MATLAB to evaluate their effectiveness in canceling ISI:1. Non-adaptive FIR DFE with known overall channel impulse response2. Adaptive FIR LE with training3. Adaptive FIR DFE with trainingIn the first case, the non-adaptive equalizer assumes a known channel impulseresponse (i.e. through channel estimation) as to to pre-compute the DFE’stap coefficients. The other two adaptive equalizers use the LMS algorithm fortraining prior to data transmission in order to learn the channel properties.Given that the overall EoL channel suffers from greater ISI than a pure wiredEthernet link, a higher number of equalizer taps is required in general. Thenumber of filter coefficients for each FIR equalizer will be optimized for bestperformance. For all cases, the optimal equalizer decision delays k0 was obtainedby looping over all possible delays. Finally, BER was used as the primaryevaluation criterion for all the simulations performed below.The following assumptions and simplifications were made. The backgroundambient noise in the Ethernet link, and the thermal noise in the VLC trans-mitter (Section 5.2.3) are both neglected, as the noise in the VLC channel isconsidered to be dominant (previously established in Section 5.3). If more thanone LED composes a fixture, the LED array is again approximated to a point766.3. Simulation ResultsParameter ValueTraining symbols 20 000 bitsUDP packet size 1500 bytes (12 000 bits)Total packets transmitted 100Cable length 1− 100 mTable 6.2: BER simulation parameterssource. Perfect synchronization between LED luminaires is assumed, or alterna-tively, the lengths of all Cat 5 cables per fixture are identical. This is to ensurethat all LEDs transmit simultaneously and avoid unnecessary delays and ISI.Additionally, the VLC channel is completely confined from other possible nearbyVLC transmitters that may transmit different data and cause interference. TheEthernet simulation parameters are listed in Table Effect of Cable Length and SNRIn this first set of simulations, equalization is performed over an EoL channelwith different Cat 5 cable lengths, and a range of pre-defined EoL SNRs valuesthat correspond to different amounts of ambient light present (see Table 5.5).These parameters are two of the major factors that can affect the overall linkquality. The maximum LED bandwidth is fixed at 20 MHz for the OSRAMLED just as before in Section 5.4. We show the performance by plotting theBER versus the cable length and EoL SNR; the complete simulation results areshown at the end of this chapter (Figures 6.9, 6.10 and 6.11).Non-adaptive DFEThe non-adaptive DFE assumes the frequency response corresponding to theoverall discrete-time impulse response is known a priori. As such, the filtercoefficients c and F can be calculated at the start using the algorithm developedin [124, p222] (the derivation can be found in the same reference). The optimalnumber of taps NFF and NFB were based on the numbers used in 100Base-TX receivers (Table 6.1). Although, the number of FFF taps is greater fromthe prominence of precursor ISI in the EoL link. From the simulation results(blue plots in Figures 6.6(a) and 6.6(b)), we observe that the BER worsenswith increasing cable length for each VLC channel profile as expected. Thelonger the cable, the more the ISI from insertion loss becomes significant. Theequalizer’s ability to mitigate ISI is thus deteriorated. The transmission of anMLT-3 modulation scheme in an EoL system requires an SNR approximatelyabove 25 dB for a 10 m cable length and 40 dB for a 100 m. The sphericalVLC channel model of Profile B (blue square markers) introduces the most ISIcompared to LOS-only and the wall reflection model due to the long-decayingtail of the diffuse component and results in the poorest performance. We usethe results obtained in this section to be the baseline as the channel conditions776.3. Simulation Resultsare perfectly known at the receiver, allowing for optimal performance.Adaptive LEThe linear equalizer is implemented in MATLAB using the lineareq moduleand equalize function from the Communications System Toolbox. The equal-izer is preceded by an anti-alias filter (ideal LPF) and sampler. The linearequalizer is also fractionally-spaced at an oversampled factor of l = 2 times thesymbol rate. The LPF has a cutoff frequency of fc = l2T , where T = 8 ns is thesymbol period. To replicate the training phase during IDLE periods, adaptivelearning is performed before signal transmission to adjust tap weights using aknown symbol vector. The adaptive equalizer uses the LMS algorithm witha stepsize of µ = 2× 10−2 during the training stage. The number of optimalnumber of taps N was obtained to be 100 (equivalent to 50 symbol-spaced taps).The simulation results are again depicted in Figures 6.6(a) and 6.6(b) as the redplots. The performance is noticeably poorer compared to the non-adaptive DFEcase with the BER curves dropping at a slower rate. Across all VLC channelprofiles, the equalizer is only able to resolve symbols at a BER of < 1× 10−5with a minimum SNR of at least 30 dB at a short cable length of 10 m, and50 dB at the worst case 100 m length.Adaptive DFEThe DFE employed is a symbol-spaced equalizer implemented with MATLAB’sdfe module. Just as the case of the LE, this adaptive equalizer uses the LMSalgorithm with a stepsize of µ = 2× 10−2 to estimate the FFF and FBF co-efficients during the training phase. We used the optimal number of taps de-termined in the non-adaptive DFE case as a guideline for the adaptive DFE.Once again, in Figures 6.6(a) and 6.6(b), the results of the adaptive DFE agreewith the general trend of the previous equalizer implementations, with a dete-riorating BER at a longer cable length and at a lower SNR. Since Profile Bsuffers from the most post-cursor ISI and a greater number of feedback tapsare required. The same applies to Profile C, though the post-cursor ISI is lessprominent. Generally, at an SNR above 40 dB, the DFE will be able to decodesymbols properly and yield error-free performance. The performance is compa-rable, and in some instances, better than the LE, though it is worse than theDFE with known impulse response in general.Performance DiscussionAt low SNR values, below 5 dB at 10 m and below 15 dB at 100 m, the perfor-mance of each equalizer system is nearly the same. At a shorter cable length, allequalizers are able to reach a lower BER faster at a lower SNR, compared to thelonger cable. For instance, a BER of 1× 10−4 requires an increase of approx-imately 10 dB SNR from 10 m to 100 m. Overall, the non-adaptive DFE withperfect channel knowledge yielded better results and is able to outperform at a786.3. Simulation Results0 5 10 15 20 25 3010−610−510−410−310−210−1100SNR (dB)BER  Non−adaptive DFE (Profile A)Adaptive DFE (Profile A)Adaptive LE (Profile A)Non−adaptive DFE (Profile B)Adaptive DFE (Profile B)Adaptive LE (Profile B)Non−adaptive DFE (Profile C)Adaptive DFE (Profile C)Adaptive LE (Profile C)(a) 10 m Cat 510 15 20 25 30 35 40 4510−610−510−410−310−210−1100SNR (dB)BER  Non−adaptive DFE (Profile A)Adaptive DFE (Profile A)Adaptive LE (Profile A)Non−adaptive DFE (Profile B)Adaptive DFE (Profile B)Adaptive LE (Profile B)Non−adaptive DFE (Profile C)Adaptive DFE (Profile C)Adaptive LE (Profile C)(b) 100 m Cat 5Figure 6.6: BER vs SNR796.3. Simulation ResultsProfile A Profile B Profile CLow noise All All AllMedium noise All All Non-adaptiveDFEHigh noise Non-adaptiveDFENon-adaptiveDFENon-adaptiveDFETable 6.3: Optimal equalizer performance at 10 mProfile A Profile B Profile CLow noise All All Adaptive ornon-adaptiveDFEMedium noise Non-adaptiveDFENon-adaptiveDFENon-adaptiveDFEHigh noise Non-adaptiveDFENon-adaptiveDFENon-adaptiveDFETable 6.4: Optimal equalizer performance at 100 mconsistent rate for each channel profile. Its BER curves drop at a faster rate,and present a better BER at comparable SNRs in most cases. If we comparethe performance of the adaptive DFE with the adaptive LE, the DFE requiresa lower SNR to overtake the adaptive LE between medium and high SNR. Ingeneral, the performance of the adaptive DFE starts to pass that of the LE atabout 30 dB SNR and then drops quickly.Using the results obtained in this section and the corresponding SNR valuesfrom Table 5.5, we can evaluate the performance of our equalizers and choosethe optimal type subjected to different noise conditions, VLC channel profilesor cable lengths. The results are tabulated under Table 6.3 for a 10 m cable,and Table 6.4 for 100 m. At a short cable length, all equalizers can be employedin most cases of low and medium noise. At high noise (i.e. 23.1− 28.4 dB SNR)the non-adaptive shows significant improvement across all VLC profiles over theadaptive counterparts.6.3.2 Effect of LED Modulation Bandwidth and SNRAs seen in previous sections, the link bottleneck can be attributed to either thecable (its length) or the LED (its 3 dB frequency). Accordingly, we evaluate theperformance of the EoL link subject to different LED modulation bandwidthsat a fixed cable length. In this case, we used simulation Profile C strictly806.3. Simulation Resultsto determine the minimum modulation bandwidth supported that still enablesthe link to work at optimal BERs. The following results were obtained usingthe adaptive DFE with LMS algorithm. The 3 dB frequency of the LED isincreased from 1 to 40 MHz, while the EoL SNR is varied from 1 to 60 dB. Theperformance is compared using a cable length of 10 m and 100 m again. FromFigure 6.7(a), at 10 m, the minimum modulation bandwidth required is about10 MHz. The short cable length translates to a lesser amount of ISI in thewired channel and, as a result, the integrity of the MLT-3 signal is maintained.An LED with a lower bandwidth can be combined with our adaptive DFE toobtain reasonable performance at SNRs above 30 dB. In a worst case of a 100 m,much of the original signal power is lost from cable insertion loss. The link isclose to unusable at any LED modulation bandwidths below 15 MHz as seenin Figure 6.7(b). A higher modulation above 20 MHz is regarded as adequate,in addition to a higher SNR level as determined above. At a nominal SNR of50 dB, the benefits of the maximum modulation bandwidth starts to diminishbeyond a cutoff frequency of 25 MHz. This is expected, since the majorityof the signal energy of an MLT-3 signal is below the fundamental frequency of31.25 MHz and is collected at around 16 MHz (see Figure 2.4(b)). Consequently,an LED modulation bandwidth of 20 MHz is demonstrated to be sufficient forFast Ethernet over VLC.6.3.3 Effect of Receiver LocationIn the following, we simulated the BER distribution at different locations withina typical office space, subject to the diffuse environment of simulation ProfileC ). The amount of noise follows the SNR distribution previously determined inFigure 5.7. An Ethernet cable of maximum length was used and the location ofthe photodiode is varied on the xy-plane at a fixed height of 0.85 m. The numberof bit errors was measured for a transmission of 100 Ethernet frames as before.In Figure 6.8, an absence of data points indicates that no error were detectedfor the total amount of bits transmitted. The results show that, in general,the performance is the poorest at the corners of the room, where the BER fallsbetween 8.947× 10−5 and 1.445× 10−5. As long as the user is located awayfrom the absolute corners of the room, the equalization will offer reasonableBER performance. We also performed the same simulation with the adaptiveLE, but found that the LE was unable to deal effectively with the multipathdispersion of VLC channel and resulted in significantly higher BERs.816.3. Simulation Results024x 107020406010−5100fled (Hz)SNR (dB)BER(a) 10 m Cat 5 - NFF = 40, NFB = 20024x 107020406010−610−410−2100fled (Hz)SNR (dB)BER(b) 100 m Cat 5 - NFF = 34, NFB = 27Figure 6.7: BER vs LED modulation bandwidth vs SNR826.3. Simulation Results−202−20210−710−610−510−4X location (m)Y location (m)BERFigure 6.8: BER vs photodiode x-y location - NFF = 30, NFB = 20836.3. Simulation Results0204060 05010010−5100Cable length (m)SNR (dB)BER(a) Profile A - NFF = 40 and NFB = 100204060 05010010−5100Cable length (m)SNR (dB)BER(b) Profile B - NFF = 40 and NFB = 400204060 05010010−5100Cable length (m)SNR (dB)BER(c) Profile C - NFF = 40 and NFB = 25Figure 6.9: BER vs cable length vs SNR (Non-adaptive DFE)846.3. Simulation Results0204060 05010010−5100Cable length (m)SNR (dB)BER(a) Profile A - n = 1000204060 05010010−5100Cable length (m)SNR (dB)BER(b) Profile B - n = 1000204060 05010010−5100Cable length (m)SNR (dB)BER(c) Profile C - n = 100Figure 6.10: BER vs cable length vs SNR (Adaptive LE)856.3. Simulation Results0204060 05010010−5100Cable length (m)SNR (dB)BER(a) Profile A - NFF = 48, NFB = 200204060 05010010−5100Cable length (m)SNR (dB)BER(b) Profile B - NFF = 47, NFB = 430204060 05010010−5100Cable length (m)SNR (dB)BER(c) Profile C - NFF = 34, NFB = 27Figure 6.11: BER vs cable length vs SNR (Adaptive DFE)86Chapter 7Demonstrator DesignA prototype of the proposed EoL system was designed by fellow M. Eng.student Blake Stacy using commercially available off-the-shelf parts. Thedesign details were included in this thesis for a more comprehensive coverageof our EoL system proposal. The goal of the demonstrator was to investigatethe feasibility and challenges associated with building a real prototype com-posed of discrete components. A secondary objective was to provide an insightinto the costs associated with deploying such a system on a small scale. Thesimplicity of the EoL concept facilitated a relatively straightforward hardwareimplementation. Two prototype boards, the VLC transmitter circuitry and theaccompanying VLC receiver, were designed using the major components listedin Table 7.1. Although, at the time of writing of this thesis, the demonstratorwas not yet fully operational and was still being worked on.Description Component CostCat 5 cable 2× Generic brand ∼ $10PoE switch Cisco Catalyst 4500E PoE switch [125] > $10 000PoE controller Texas Instruments TPS23756EVM Evalua-tion board [126]$491LED OSRAM LW W5SM-JYKY-JKQL GoldenDragon White LED [127]$4.642Photodiode OSRAM SFH 2701 PIN photodiode [128] $0.893Transistors 1× STM 2N2222A NPN BJT ∼ $1.004Op-amps 1× Texas Instruments OPA3690 $3.1851× Texas Instruments THS3202 $3.4763× Texas Instruments OPA656 3× $7.5071× Texas Instruments OPA847 (TIA) $2.178Table 7.1: List of components used877.1. Transmitter Circuit DesignFigure 7.1: VLC transmitter implementation circuit diagram7.1 Transmitter Circuit DesignThe design of the transmitter is two parts: the PD front-end and the LED drivecircuitry (refer back to Section 3.2.1). For the PoE portion, a Texas Instru-ments PoE controller was used to simplify integration of the transmitter. Thecontroller received power from the twisted pair and converted it into a +5 V DCto be used by the rest of the circuit components of the transmitter. Althoughit would be possible to build the PoE front-end using analog components ex-clusively, a pre-built PoE controller was adopted to save time and effort. Thecontroller offers additional flexibility with built-in features, like selecting thedesired output DC voltage and providing overvoltage protection for the rest ofthe transmitter circuitry components.As previously stated, the implementation of the LED driver depends entirelyupon the LED type and model, since the amplification and DC bias amount needto be controlled for the LED’s linear operating range. The design of the VLCtransmitter follows that of the system model previously featured on the right-hand side of Figure 3.2, and the circuit schematic is presented in Figure 7.1.The +5 V DC supplied by the PoE controller (VS1) is used to power the LED(LED1) and the three op-amps (U1, U2, U3) in single supply operation. The1http://www.ti.com/tool/tps23756evm2http://ca.mouser.com/ProductDetail/OSRAM-Opto-Semiconductors/LW-W5SM-JYKY-JKQL-1/?qs=pCZPOPZMYPhtP5sxEioDGg==3http://www.digikey.com/product-search/en?site=us&lang=en&enterprise=36&mpart=SFH+27014http://octopart.com/2n2222a-stmicroelectronics-1365http://www.ti.com/product/opa36906http://www.ti.com/product/ths32027http://www.ti.com/product/opa6568http://www.ti.com/product/opa847887.2. Receiver Circuit DesignFigure 7.2: VLC receiver implementation circuit diagramanalog 100Base-TX signal transmitted over the twisted pair is extracted from themagnetic modules and is depicted as a differential signal (V TX+ and V TX-).This signal is combined into a single voltage signal using the differential amplifierconfiguration (U1, R1, R2, R3), by subtracting one signal from the other. Thecombined signal becomes 4 Vp−p, before being halved by the differential op-amp setup to 2 Vp−p. At the same time, common-mode noise is eliminatedfrom the signal in this operation. The DC/DC converter stage (U2, R4, R5 andC1) takes the 5 V supplied by the controller and converts it to a 2.5 V DC biasvoltage. This DC offset is added to the voltage signal using resistors R8 and R9.The final gain stage (U3, R10, R12) acts as a voltage-to-current converter thatregulates the driving current flowing through the LED. The amplitude of thesignal is controlled such that it falls within the proper linear range of operationof the LED. A fluctuating voltage at the base of the NPN-BJT (T1) controlsthe amount of current passing through the LED (LED1) in this open-collectorsetup, and so proportionally controlling the output optical power.7.2 Receiver Circuit DesignIncident light is captured using a typical SI-PIN photodiode at the VLC re-ceiver. The receiver circuit design is based off of the block diagram of Fig-ure 3.2. In this implementation, the receiver is composed of two separate stages:a bootstrapped-TIA that converts the photocurrent back to a voltage signal,and a subsequent LNA gain stage that amplifies the signal to detectable levels.The signal is fed back to a computer via a regular Ethernet Network InterfaceCard (NIC) for signal processing.In Figure 7.2, the photodiode is modeled by current source (IG1), junctioncapacitance (C3) and shunt resistance (R4). The OSRAM photodiode was cho-sen for its excellent performance in VLC applications and is characterized witha switching frequency of ∼ 250 MHz. A bootstrap circuit (U3, R3, R5, C4) isconnected prior to the TIA with the purpose of compensating for the effect of897.3. SimulationsFigure 7.3: Bode plot of bootstrapped-TIAspurious junction capacitance of the photodiode, while at the same time, en-hancing the bandwidth capabilities of the subsequent TIA stage. The TIA (U2,R1, C1, R2, C2), which is implemented as a simple amplifier with a feedbackloop, converts the photocurrent into an output voltage. Because the currentgenerated by the photodiode is in the order of µA, a large TIA gain is desir-able, and in this case, the combined bootstrapped-TIA stage provides a gain of61.5 dB (see Figure 7.3).Following the bootstrapped-TIA stage, an AC coupling capacitor (C6) isadded to the circuit path to remove the DC offset in the signal, induced by all DCsources from the LED and ambient light. To boost the received signal back tovoltage levels appropriate for detection (∼ 2 Vp−p), an additional amplificationstage is required. Considering that the demonstrator system operates over ashort distance with a non-mobile receiver, the gain of all amplification stagescan be made fixed. Otherwise, an AGC should be used to consider different EoLsetups, changing environments or a moving receiver. Here, the pre-amplifierstage (U1, R6, R7, R8, R9) provides another fixed gain to achieve a nominal2 Vp−p. The voltage signal is then split into two differential signals, in which theregular signal is obtained from the following gain stage (U4, R10, R11, R12),and the inverted signal is obtained from an inverting op-amp setup (U5, R13,R14, R15). This produces a differential voltage signal that is forwarded to acomputer via a short length Cat 5 UTP standard RJ-45 I/O port.7.3 SimulationsThe following simulations of this demonstrator prototype were performed usingTINA-TI as to evaluate the performance of the VLC transmitter and receiver cir-cuitry. At the transmitter side, the LED driving current was plotted comparedto the differential output voltage signal of a 20 m twisted pair cable in Figure 7.4.At the receiver, the differential output voltage was obtained post-amplificationfrom the photocurrent in Figure 7.5, and the results were comparable to ourMATLAB simulations.907.3. SimulationsFigure 7.4: Transmitter simulationFigure 7.5: Receiver simulation91Chapter 8ConclusionIn this thesis, we investigated the requirements for the implementation of aunidirectional broadcast Fast Ethernet link over VLC. We proposed anamplify-and-forward system in which s 100Base-TX electrical signal suppliedfrom a standard Cat 5 cable was used to drive an LED directly to be sent overVLC without demodulation or pre-equalization. This fluctuating optical signalwas successfully recovered at the receiving using a typical VLC front-end andusing post-equalization techniques.The EoL link presented unique characteristics and challenges that were dis-covered and analyzed. The first half of the link, the Ethernet channel, wasrelatively well-behaved and was mostly immune to noise impairments as a di-rect result of the twisted pair property. Although, this wired connection sufferedfrom prominent propagation loss from the skin effect at high frequency trans-missions. A critical factor to link performance was discovered to be the cablelength. The temporal dispersion becomes more prominent with increasing ca-ble length. The second half of the EoL channel was the VLC link which wassusceptible to ISI induced by the optical path differences in typical indoor en-vironments, and could lead to performance degradation of the EoL link. Inaddition, the LED was discovered to be a bottleneck element due to its limitedavailable bandwidth, a direct consequence of the current state of SSL technol-ogy. The LED’s minimum required switching frequency for optimal operationof our EoL system was determined to be about 20 MHz.The overall channel was accurately portrayed as an LTI system, which al-lowed us to analyze the complete system, component by component, and pro-duce a comprehensive link budget. In turn, this enabled us to estimate expectedSNR levels in different EoL configurations. We used existing equalization strate-gies as the primary means to combat dispersions introduced by the cable, LEDand VLC channel. We considered two cases. In the first case, the overall chan-nel impulse response was known to the transmitter, allowing an equalizer toaccurately compute the tap coefficients a priori. In the second case, trainingwas required to estimate the coefficients using an adaptive strategy. The perfor-mance of linear and non-linear equalizers (DFE) were evaluated and the optimaltap weights were determined for various channel configurations. Simulation re-sults showed that the performance of the channel relied heavily on the SNRlevels of the EoL channel, attributed to the illumination level and amount ofambient light present. We observed that a high SNR level was required themajority of the time to achieve BERs suitable for 100Base-TX. This is entirelyreasonable since indoor rooms usually need to meet a minimum level of illumi-928.1. Future Worknation resulting in very high SNR.Finally, as part of this effort and under the leadership of a fellow M.Eng.student, a proof-of-concept design was developed to complement the work pre-sented, and to showcase the potential and viability of Fast Ethernet over VLC.The design showed that the EoL system can be completely built using readilyavailable, off-the-shelf components.8.1 Future Work8.1.1 UplinkSince our EoL link is currently a strictly broadcast system, it would be beneficialto extend this system into a fully duplex 100Base-TX link by realizing an uplinkat the VLC receiver side. In doing so, we would enable packet acknowledgementand retransmissions requests to be sent back to the transmitter side, essentialfor establishing a TCP connection. An uplink also grants a feedback path tothe VLC transmitter that can be used to feedback link quality, or the channelimpulse response for example. The challenge lies in choosing the appropriatemedium for the uplink while providing a low complexity, seamless connectionbetween the network and the user. A number of avenues exists as it is possibleto either reuse the VLC medium again, or create a hybrid system (non-VLCuplink) with IR or RF.A VLC/IR uplink enables a purely optical duplex link that could use thesame amplify-and-forward concept adopted for the downlink. However, the mainissue is in the practicality associated with implementing LEDs at the receiver.In the context of eye safety, if the device is located at desk level and is pointedupwards towards the ceiling, the light emitted can cause undesirable discomfortto its users. This is magnified by the fact that high power LEDs would berequired in order to satisfy minimum SNR requirements and achieve acceptableBERs. We see RF, more specifically Wi-Fi, to be the best alternative for anuplink. Wi-Fi enjoys the same widespread usage in today’s wireless LAN marketas Ethernet, since nearly all mobile computing units (smartphones, tablets orlaptops) support both technologies. As the original purpose of EoL was totake some load off of the wireless RF downlink, we can make use of the Wi-Fiuplink of an indoor wireless LAN to create an asymmetric optical-RF hybridsystem. The idea is to reuse an existing RF infrastructure and piggyback theuplink onto the RF network, combining an EoL downlink with a Wi-Fi uplink.We envision devices already equipped with Wi-Fi could integrate with an EoLplug-and-play receiver to provide a straightforward duplex 100 Mbps solution.Such a VLC-Wi-Fi hybrid system has been explored in [129] with a preliminaryinvestigation. The authors proposed using VLC as a supplement to Wi-Fi toincrease the aggregate capacity by offloading users to the VLC channel.938.1. Future Work8.1.2 1000Base-T over VLCAfter 100Base-TX, the next step would be to explore the implementation of itssuccessor, gigabit Ethernet over VLC, increasing the data rate tenfold to 1000Mbps. Similar to Fast Ethernet, 1000Base-T is also compatible with PoE andcan reuse the same Cat 5 cable. In this case, all four twisted pairs are used simul-taneously for duplex transmission at 250 Mbps each, for a combined 1000 Mbps.If the same EoL amplify-and-forward approach was considered in a 1000Base-T-VLC system, the VLC link would require a MIMO setup to realize such asystem. For instance, the cable would be split into four individual twisted pairsto feed four separate LED luminaires, essentially forming four distinct SISOchannels. The VLC fixtures would then transmit the independent data streamssimultaneously. At the receiver side, there would be two possible implemen-tation options. The first would use four non-imaging receivers, separated ata small distance. This would result in the simplest hardware implementationwhich has been demonstrated to work in [28]. The second option would employimaging receivers that would allow for easier separation of the MIMO chan-nels [62, 130]. Naturally, there are many more hurdles to overcome in realizingsuch a 1000Base-T-VLC system. 1000Base-T uses a 4-dimensional 5-level PAMscheme to modulate at a 125 mega symbol rate (two bits per symbol). Thetransmission of a PAM-5 signal would be significantly more challenging thanMLT-3 in a similar amplify-and-forward strategy. PAM schemes suffer fromincreased power penalty with each increasing number of modulation levels asthey require even more stringent SNR margins. Though, 1000Base-T over VLCwas implemented recently in [23] and [131] using different approaches.94Bibliography[1] Cisco, “Cisco Visual Networking Index: Global Mobile Data Traffic Fore-cast Update , 2013 - 2018,” Cisco, Tech. Rep., 2014.[2] Red Mobile Consulting, “Study of Future Demand For Radio Spectrumin Canada 2011-2015,” Tech. Rep., 2011.[3] H. Falaki, D. Lymberopoulos, R. Mahajan, S. Kandula, and D. Estrin,“A first look at traffic on smartphones,” in Proceedings of the 10th annualconference on Internet measurement - IMC ’10. New York, New York,USA: ACM Press, 2010, p. 281.[4] H. Burchardt, N. Serafimovski, D. Tsonev, S. Videv, and H. Haas,“VLC: Beyond Point-to-Point Communication,” Communications Mag-azine, IEEE, vol. 52, no. 7, pp. 98–105, 2014.[5] Navigant, “Adoption of Light-Emitting Diodes in Common Lighting Ap-plications,” Tech. Rep., 2013.[6] T. Komine and M. Nakagawa, “Integrated system of white LED visible-light communication and power-line communication,” IEEE Transactionson Consumer Electronics, vol. 49, no. 1, pp. 71–79, 2003.[7] F. Delgado, I. Quintana, J. Rufo, J. Rabadan, C. Quintana, andR. Perez-Jimenez, “Design and Implementation of an Ethernet-VLCInterface for Broadcast Transmissions,” IEEE Communications Letters,vol. 14, no. 12, pp. 1089–1091, 2010.[8] A. Burton, E. Bentley, H. L. Minh, Z. Ghassemlooy, N. Aslam, and S.-K.Liaw, “Experimental demonstration of a 10BASE-T Ethernet visible lightcommunications system using white phosphor light-emitting diodes,”IET Circuits, Devices & Systems, pp. 1–9, 2014.[9] P. A. Haigh, T. T. Son, E. Bentley, Z. Ghassemlooy, H. Le Minh,and L. Chao, “Development of a Visible Light Communicationssystem for optical wireless local area networks,” in 2012 Computing,Communications and Applications Conference, 2012, pp. 351–355.[10] Z. Ghassemlooy, H. L. Minh, P. Haigh, and A. Burton, “Developmentof Visible Light Communications: Emerging Technology and IntegrationAspects,” in Optics and Photonics Taiwan International Conference(OPTIC) 2012, 2012, pp. 1–7.95Bibliography[11] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wirelesscommunication: potential and state-of-the-art,” CommunicationsMagazine, IEEE, vol. 49, no. 9, pp. 56–62, 2011.[12] C. Quintana, V. Guerra, J. Rufo, and J. Rabadan, “Ethernet to Visible-Light Communications Adapter for In-Flight Passenger Data Network-ing,” in IEEE International Conference on Wireless Communications andNetworks (CWCN2011), 2011, pp. 1–4.[13] C. Quintana, V. Guerra, J. Rufo, J. Rabadan, and R. Perez-Jimenez,“Reading lamp-based visible light communication system for in-flightentertainment,” IEEE Transactions on Consumer Electronics, vol. 59,no. 1, pp. 31–37, 2013.[14] X. Liu, S. Wang, L. C. Png, K. S. Yeo, S. Anandraj, B. J. H. Loke, andA. Y. R. Wong, “Visible Light Audio, Video, and Ethernet TransceiverDesigns for Desktop Applications,” 2014.[15] IEEE Computer Society, “IEEE Standard for Ethernet - Section One,”pp. 1–580, 2012.[16] ——, “IEEE Standard for Ethernet - Section Two,” pp. 1–780, 2012.[17] M. Mazzola, L. Cafiero, and M. DeNicolo, “Method and apparatus formultilevel encoding for a local area network,” pp. 1–6, 1994.[18] American National Standards Institute, “ANSI X3.263-1995,” pp. 1–68,1995.[19] G. Fairhurst, “The Ethernet Local Area Network,” pp. 1–39.[20] Y. Fan, J. Chu, W. Yan, and R. Junyan, “Equalizer for 10Base-T/100Base-TX Ethernet transceiver based on DSP structure,” in ASIC,2003. Proceedings. 5th International Conference on, 2003, pp. 874–876.[21] Telecommunications Industry Association, “ANSI/TIA-568-C.2-2009,”pp. 1–204, 2009.[22] G. Mendelson, “All You Need To Know About Power over Ethernet(PoE) and the IEEE 802.3 af Standard,” pp. 1–24, 2004.[23] L. Grobe, A. Paraskevopoulos, J. Hilt, D. Schulz, F. Lassak, F. Hartlieb,C. Kottke, V. Jungnickel, K.-D. Langer, and H. H. Institute, “High-SpeedVisible Light Communication Systems,” IEEE Communications Maga-zine, pp. 60–66, Dec. 2013.[24] J. Grubor, S. Randel, K.-D. Langer, and J. Walewski, “Bandwidth-efficient indoor optical wireless communications with white light-emittingdiodes,” in 2008 6th International Symposium on CommunicationSystems, Networks and Digital Signal Processing, 2008, pp. 165–169.96Bibliography[25] M. Biagi, T. Borogovac, and T. Little, “Adaptive Receiver for IndoorVisible Light Communications,” Journal of Lightwave Technology, pp.1–11, 2013.[26] D. C. O’Brien, G. Faulkner, H. L. Minh, O. Bouchet, M. E. Tabach,M. Wolf, J. W. Walewski, S. Randel, S. Nerreter, M. Franke,K.-D. Langer, J. Grubor, and T. Kamalakis, “Home access networksusing optical wireless transmission,” in Personal, Indoor and MobileRadio Communications, 2008. PIMRC 2008. IEEE 19th InternationalSymposium on, 2008, pp. 1–5.[27] J. Song, W. Ding, F. Yang, H. Yang, J. Wang, X. Wang, and . Zhang,“Indoor Hospital Communication Systems : An Integrated Solution BasedPower Line and Visible Light Communication,” in Faible Tension FaibleConsommation (FTFC), 2014 IEEE, 2014, pp. 1–6.[28] L. Zeng, D. C. O’Brien, H. L. Minh, G. E. Faulker, K. Lee, D. Jung,Y. Oh, and E. T. Won, “High data rate multiple input multiple output(MIMO) optical wireless communications using white LED lighting,”IEEE Journal on Selected Areas in Communications, vol. 27, no. 9, pp.1654–1662, 2009.[29] J. Grubor, O. Gaete Jamett, J. Walewski, S. Randel, and K.-D. Langer,“High-Speed Wireless Indoor Communication via Visible Light,” in ITGFachbericht 198, vol. 1, 2007, pp. 203–208.[30] Luxeon, “Luxeon III Star Techincal Datasheet DS46,” pp. 1–10, 2003.[31] Seoul Semiconductor, “F50360,” pp. 1–19, 2008.[32] J. Vucic, C. Kottke, K. Habel, and K.-d. Langer, “803 Mbit/s visiblelight WDM link based on DMT modulation of a single RGB LEDluminary,” in Optical Fiber Communication Conference and Exposition(OFC/NFOEC), 2011 and the National Fiber Optic Engineers Conference,2011, pp. 1–3.[33] G. Cossu, a. M. Khalid, P. Choudhury, R. Corsini, and E. Ciaramella,“3.4 Gbit/s visible optical wireless transmission based on RGB LED.”Optics express, vol. 20, no. 26, pp. B501–B506, 2012.[34] C. Kottke, J. Hilt, K. Habel, J. Vucic, and K.-D. Langer, “1.25 Gbit/svisible light WDM link based on DMT modulation of a single RGB LEDluminary,” in Optical Communications (ECOC), 2012 38th EuropeanConference and Exhibition on, 2012, pp. 1–3.[35] E. Monteiro, “Design and Implementation of Color-Shift Keying forVisible Light Communications,” Master’s thesis, McMaster University,2013.97Bibliography[36] Y. Zhang, X. Ke, M. Xie, and Y. Kang, “Multi-dimension PulsePosition Modulation for RGB LEDs,” in Proc. SPIE 8906, InternationalSymposium on Photoelectronic Detection and Imaging 2013: LaserCommunication Technologies and Systems, K. E. Wilson, J. Ma, L. Liu,H. Jiang, and X. Ke, Eds., 2013, pp. 1–8.[37] E. F. Schubert, Light-Emitting Diodes. Cambridge University Press,2006.[38] D. C. O’Brien, G. E. Faulkner, E. B. Zyambo, K. Jim, D. J. Edwards,P. Stavrinou, G. Parry, J. Bellon, M. J. Sibley, V. A. Lalithambika, V. M.Joyner, R. J. Samsudin, D. M. Holburn, and R. J. Mears, “Integratedtransceivers for optical wireless communications,” IEEE Journal ofSelected Topics in Quantum Electronics, vol. 11, no. 1, pp. 173–183,2005.[39] H. L. Minh, Z. Ghassemlooy, A. Burton, and P. A. Haigh, “Equalizationfor Organic Light Emitting Diodes in Visible Light Communications,” inGLOBECOM Workshops (GC Wkshps), 2011 IEEE, 2011, pp. 828–832.[40] H. Chun, C.-J. Chiang, A. Monkman, and D. O’Brien, “A Study ofIllumination and Communication using Organic Light Emitting Diodes,”Journal of Lightwave Technology, vol. 31, no. 22, pp. 3511–3517, 2013.[41] P. A. Haigh and Z. Ghassemlooy, “Visible light communications usingorganic light emitting diodes,” IEEE Communications Magazine, pp.148–154, 2013.[42] P. Butala, H. Elgala, and T. Little, “SVD-VLC: A novel capacity maxi-mizing VLC MIMO system architecture under illumination constraints,”in Globecom 2013 Workshop on Optical Wireless Communications, 2013,pp. 1–15.[43] L. Zeng, D. O’Brien, H. Le-Minh, K. Lee, D. Jung, and Y. Oh,“Improvement of Date Rate by using Equalization in an IndoorVisible Light Communication System,” in 2008 4th IEEE InternationalConference on Circuits and Systems for Communications, 2008, pp.678–682.[44] K. Cui and Z. Xu, “Experimental characterization of indoor visiblelight communication channels,” in 2012 8th International Symposiumon Communication Systems, Networks & Digital Signal Processing(CSNDSP), Jul. 2012, pp. 1–5.[45] Z. Ghassemlooy, W. Popoola, and S. Rajbhandari, “Channel Modelling,”in Optical wireless communications: system and channel modelling withMATLAB. CRC Press, 2013, pp. 77–159.98Bibliography[46] L. Grobe and K.-d. Langer, “Block-Based PAM with Frequency DomainEqualization in Visible Light Communications,” in 4th IEEE Workshopon Optical Wireless Communications, 2013, pp. 1075–1080.[47] K. Hild, T. E. Sale, S. J. Sweeney, M. Hirotani, Y. Mizuno, and T. Kato,“Modulation speed and leakage current in 650 nm resonant-cavity lightemitting diodes,” Optoelectronics, IEE Proceedings, vol. 151, no. 2, pp.94–97, 2004.[48] M. Guina, S. Orsila, M. Dumitrescu, M. Saarinen, P. Sipila,V. Vilokkinen, B. Roycroft, P. Uusimaa, M. Toivonen, and M. Pessa,“Light-emitting diode emitting at 650 nm with 200-MHz small-signalmodulation bandwidth,” IEEE Photonics Technology Letters, vol. 12,no. 7, pp. 786–788, 2000.[49] P. A. Haigh, Z. Ghassemlooy, H. Le Minh, S. Rajbhandari, F. Arca, S. F.Tedde, O. Hayden, and I. Papakonstantinou, “Exploiting EqualizationTechniques for Improving Data Rates in Organic Optoelectronic Devicesfor Visible Light Communications,” Journal of Lightwave Technology,vol. 30, no. 19, pp. 3081–3088, 2012.[50] T. Lueftner, C. Kroepl, M. Huemer, J. Hausner, R. Hagelauer, andR. Weigel, “Edge-position modulation for high-speed wireless infraredcommunications,” IEE Proceedings Optoelectronics, vol. 150, no. 5, pp.427–437, 2003.[51] M. Wolf, L. Grobe, M. R. Rieche, A. Koher, and J. Vucic, “Blocktransmission with linear frequency domain equalization for dispersiveoptical channels with direct detection,” in Transparent Optical Networks(ICTON), 2010 12th International Conference on, 2010, pp. 1–8.[52] B. Inan and S. J. Lee, “Impact of LED nonlinearity on discrete multitonemodulation,” Optical Communications and Networking, IEEE/OSAJournal of, vol. 1, no. 5, pp. 439–451, 2009.[53] L. Peng, S. Haese, M. He´lard, and A. L. E. D. N. Phenomenon, “Fre-quency Domain LED Compensation for Nonlinearity Mitigation in DMTSystems,” IEEE Photonics Technology Letters, vol. 25, no. 20, pp. 1–4,2013.[54] I. Neokosmidis, T. Kamalakis, J. Walewski, B. Inan, and T. Sphicopoulos,“Impact of Nonlinear LED Transfer Function on Discrete MultitoneModulation: Analytical Approach,” Journal of Lightwave Technology,vol. 27, no. 22, pp. 4970–4978, 2009.[55] H. Elgala, R. Mesleh, and H. Haas, “Predistortion in Optical WirelessTransmission Using OFDM,” in 2009 Ninth International Conference onHybrid Intelligent Systems, 2009, pp. 184–189.99Bibliography[56] ——, “An LED Model for Intensity-Modulated Optical CommunicationSystems,” IEEE Photonics Technology Letters, vol. 22, no. 11, pp. 835–837, 2010.[57] T. Komine and M. Nakagawa, “Fundamental analysis for visible-lightcommunication system using LED lights,” IEEE Transactions onConsumer Electronics, vol. 50, no. 1, pp. 100–107, 2004.[58] IEEE Computer Society, “IEEE Standard for Local and metropolitanarea networks - Part 15.7: Short-Range Wireless Optical CommunicationUsing Visible Light,” pp. 1–286, 2011.[59] J. M. Kahn and J. R. Barry, “Wireless Infrared Communications,” Pro-ceedings of the IEEE, vol. 85, no. 2, pp. 265–298, 1997.[60] Z. Ghassemlooy, W. Popoola, and S. Rajbhandari, “Modulation Tech-niques,” in Optical wireless communications: system and channel mod-elling with MATLAB. CRC Press, 2013, pp. 161–265.[61] J. Vucic, “Adaptive Modulation Technique for Broadband Communica-tion in Indoor Optical Wireless Systems,” Ph.D. dissertation, TechnischenUniversitat Berlin, 2009.[62] S. Arnon, J. R. Barry, G. K. Karagiannidis, R. Schober, and M. Uysal, Ad-vanced Optical Wireless Communication Systems. Cambridge UniversityPress, 2006.[63] K. Cui, G. Chen, Q. He, and Z. Xu, “Indoor optical wirelesscommunication by ultraviolet and visible light,” in Proc. SPIE 7464,Free-Space Laser Communications IX, 2009, pp. 1–9.[64] J.-Y. Wang, J.-B. Wang, M. Chen, and X. Song, “Dimming schemeanalysis for pulse amplitude modulated visible light communications,”in Wireless Communications & Signal Processing (WCSP), 2013International Conference on, 2013, pp. 1–6.[65] R. Green, H. Joshi, M. Higgins, and M. Leeson, “Recent developments inindoor optical wireless systems,” IET communications, vol. 2, no. 1, pp.3–10, 2008.[66] Z. Tang, “A 100 MHz Gm-C analog equalizer for 100Base-TXapplication,” in Solid-State and Integrated-Circuit Technology, 2001.Proceedings. 6th International Conference on, 2001, pp. 240–242.[67] W. Y. Chen, Home Networking Basis: Transmission Environments andWired/Wireless Protocols. Prentice Hall PTR, 2004.[68] Fast Ethernet Consortium, “Clause 25 Physical Medium Dependent(PMD) Test Suite,” University of New Hampshire, Tech. Rep., 2011.100Bibliography[69] J. Everitt, J. Parker, P. Hurst, D. Nack, and K. R. Konda, “A CMOStransceiver for 10-Mb/s and 100-Mb/s Ethernet,” Solid-State Circuits,IEEE Journal of, vol. 33, no. 12, pp. 2169–2177, 1998.[70] J. Hong, J. Baek, and M. H. Sunwoo, “Novel Digital Signal ProcessingUnit Using New Digital Baseline Wander Corrector for Fast Ethernet,”Journal of Signal Processing Systems, vol. 61, no. 2, pp. 193–204, 2009.[71] S. Huss, M. Mullen, C. Gray, R. Smith, M. Summers, J. Shafer, P. Heron,T. Sawinska, and J. Medero, “A DSP based 10BaseT/100BaseTXEthernet transceiver in a 1.8 V, 0.18 µm CMOS technology,” in CustomIntegrated Circuits, 2001, IEEE Conference on, 2001, pp. 135–138.[72] O. Shoaei, A. Shoval, and R. Leonowich, “A 3V Low-Power 0.25um CMOS100MB/s Receiver for Fast Ethernet,” in 2000 IEEE International Solid-State Circuits Conference, 2000, pp. 4–6.[73] M. Kavehrad, J. F. Doherty, J.-h. Jeong, A. Roy, and G. Malhotra,“10Gbps Transmission over Standard Category-5 Copper Cable,” inGlobal Telecommunications Conference, 2003. GLOBECOM ’03. IEEE,vol. 7, 2003, pp. 4106–4110.[74] W. Chen, “A direct equalization method,” in Acoustics, Speech,and Signal Processing, 1997. ICASSP-97., 1997 IEEE InternationalConference on, vol. 3, 1997, pp. 2505–2508.[75] H. Dehghan and C. Weber, “Aided Decision Feedback Equalizationfor Wired Communication,” in Signals, Systems and Computers, 2005.Conference Record of the Thirty-Ninth Asilomar Conference on, 2005, pp.782–787.[76] N. Holte, “Transmission properties of pair cables,” pp. 1–37, 2002.[77] M. M. Al-Asadi, A. P. Duffy, K. G. Hodge, and A. J. Willis, “Twisted paircable design analysis and simulation,” in Proceedings of the InternationalWire and Cable Symposium, 2000, pp. 111–120.[78] Commercial Building Telecommunications Industry Association,“ANSI/TIA/EIA-568-B.2-2001,” pp. 1–139, 2001.[79] G. Im and J. Werner, “Bandwidth-efficient digital transmission overunshielded twisted-pair wiring,” Selected Areas in Communications,IEEE Journal on, vol. 13, no. 9, pp. 1643–1655, 1995.[80] D. Johns and D. Essig, “Integrated circuits for data transmission overtwisted-pair channels,” IEEE Journal of Solid-State Circuits, vol. 32,no. 3, pp. 398–406, 1997.[81] Quabbin Wire and Cable Co., “What is Return Loss? Why is itImportant?” 1999.101Bibliography[82] J. Rivernider, “Return Loss and Data Transmission.”[83] Z. Ghassemlooy, W. Popoola, and S. Rajbhandari, “Introduction,” in Op-tical wireless communications: system and channel modelling with MAT-LAB. CRC Press, 2013, pp. 1–33.[84] T. Komine, “Visible Light Wireless Communications and Its FundamentalStudy,” Ph.D. dissertation, Keio University, 2005.[85] N. Rafael and M. Lourenc¸o, “Communication Systems Using Visible Light:Emitter/Receiver,” Master’s thesis, University of Aveiro, 2009.[86] K. Lee, H. Park, and J. Barry, “Indoor channel characteristics for visiblelight communications,” Communications Letters, IEEE, vol. 15, no. 2,pp. 217–219, 2011.[87] J. Barry, J. Kahn, W. Krause, E. Lee, and D. Messerschmitt, “Simulationof multipath impulse response for indoor wireless optical channels,”IEEE Journal on Selected Areas in Communications, vol. 11, no. 3, pp.367–379, 1993.[88] T.-H. Do and M. Yoo, “TDOA-based indoor positioning using visiblelight,” Photonic Network Communications, vol. 27, no. 2, pp. 80–88,2014.[89] S. Rodriguez, R. Perez-Jimenez, F. J. Lopez-Hernandez, O. B. Gonzalez,and B. R. Mendoza, “Models and algorithm for the calculation of theimpulse response on IR-wireless indoor channels,” Proc. SPIE 5117, VLSICircuits and Systems, vol. 5117, pp. 199–208, 2003.[90] F. J. Lopez-Hernandez, R. Perez-Jimenez, and A. Santamaria, “Ray-tracing algorithms for fast calculation of the channel impulse response ondiffuse IR wireless indoor channels,” Optical engineering, vol. 39, no. 10,pp. 2775–2780, 2000.[91] S. P. Rodr´ıguez, R. P. Jime´nez, B. R. Mendoza, F. J. L. Herna´ndez, andA. J. A. Alfonso, “Simulation of impulse response for indoor visible lightcommunications using 3D CAD models,” EURASIP Journal on WirelessCommunications and Networking, vol. 2013, no. 1, p. 7, 2013.[92] V. Jungnickel, V. Pohl, S. Nonnig, and C. von Helmolt, “A Physical Modelof the Wireless Infrared Communication Channel,” IEEE Journal on Se-lected Areas in Communications, vol. 20, no. 3, pp. 631–640, 2002.[93] N. Hayasaka and T. Ito, “Channel modeling of nondirected wirelessinfrared indoor diffuse link,” Electronics and Communications in Japan(Part I: Communications), vol. 90, no. 6, pp. 9–19, 2007.[94] J. Carruthers and J. Kahn, “Modeling of nondirected wireless infraredchannels,” IEEE Transactions on Communications, vol. 45, no. 10, pp.1260–1268, 1997.102Bibliography[95] F. R. Gfeller and U. Bapst, “Wireless in-house data communication viadiffuse infrared radiation,” Proceedings of the IEEE, vol. 67, no. 11, pp.1474–1486, 1979.[96] I. Stefan, H. Elgala, and H. Haas, “Study of dimming and LEDnonlinearity for ACO-OFDM based VLC systems,” in 2012 IEEEWireless Communications and Networking Conference (WCNC), 2012,pp. 990–994.[97] M. S. A. Mossaad and S. Hranilovic, “Practical OFDM Signalling for Visi-ble Light Communications Using Spatial Summation,” in Communications(QBSC), 2014 27th Biennial Symposium on, 2014, pp. 5–9.[98] K. Cui, “Physical Layer Characteristics and Techniques for Visible LightCommunications,” Ph.D. dissertation, University of California, Riverside,2012.[99] H. Qian, S. Member, S. J. Yao, S. Z. Cai, and T. Zhou, “AdaptivePost-Distortion for Nonlinear LEDs in Visible Light Communications,”IEEE Photonics Journal, vol. 6, no. 4, pp. 1–9, 2014.[100] A. Burton, Z. Ghassemlooy, S. Rajbhandari, and S.-k. Liaw, “Design andanalysis of an angular-segmented full-mobility visible light communica-tions receiver,” in Transactions on Emerging Telecommunications Tech-nologies, 2014, pp. 1–9.[101] B. Razavi, “Transimpedance Amplifiers,” in Design of Integrated Circuitsfor Optical Communications, 2nd ed. John Wiley & Sons, Inc., 2012,ch. 4, pp. 62–122.[102] Extron Electronics, “UTP Technology,” Tech. Rep., 2001.[103] Z. Ghassemlooy, W. Popoola, and S. Rajbhandari, “Visible Light Com-munications,” in Optical wireless communications: system and channelmodelling with MATLAB. CRC Press, 2013, no. 1880, pp. 443–496.[104] W. Buchanan, “Gigabit Ethernet Over Category 5 U.T.P. Cabling,”Master’s thesis.[105] K. Dividis, “Design and Prototyping of a Visible Light Indoor PositioningSystem,” Philips Research, Tech. Rep., 2007.[106] A. Moreira, V. R.T., and A. d. O. Duarte, “Performance of infraredtransmission systems under ambient light interference,” Optoelectronics,IEE Proceedings, vol. 143, no. 6, pp. 339–346, 1996.[107] A. M. Street, P. N. Stavrinou, D. C. O’Brien, and D. J. Edwards, “Indooroptical wireless systems a review,” Optical and Quantum Electronics,vol. 29, no. 3, pp. 349–378, 1997.103Bibliography[108] M. Antoniali, A. M. Tonello, and F. Versolatto, “A Study on the OptimalReceiver Impedance for SNR Maximization in Broadband PLC,” Journalof Electrical and Computer Engineering, vol. 2013, pp. 1–11, 2013.[109] M. Cvijetic and I. B. Djordjevic, “Chapter 4,” in Advanced Optical Com-munication Systems and Networks, 2013, pp. 232–319.[110] S. Blair, “Noise in detection,” pp. 91–94, 2012.[111] OSRAM, “Golden DRAGON LCW W5SM,” pp. 1–23, 2010.[112] Thorlabs, “PDA10A Si Amplified Fixed Detector User Guide,” pp. 1–14,2014.[113] C. G. Lee, “Visible Light Communication,” in Advanced Trends in Wire-less Communications. InTech, 2004, pp. 327–339.[114] L. a. Azizan, M. S. Ab-Rahman, M. R. Hassan, a. A. a. Bakar,and R. Nordin, “Optimization of signal-to-noise ratio for wirelesslight-emitting diode communication in modern lighting layouts,” OpticalEngineering, vol. 53, no. 4, pp. 1–9, 2014.[115] N. Saha, R. K. Mondal, and Y. M. Jang, “Receiver performanceimprovement utilizing diversity in MIMO VLC,” in 2013 InternationalConference on ICT Convergence (ICTC), 2013, pp. 546–549.[116] P. Kish, “The Effect of Network Cabling on Bit Error Rate Performance,”pp. 1–12, 2000.[117] J. M. Kahn, J. R. Barry, M. D. Audeh, J. B. Carruthers, W. J. Krause, andG. W. Marsh, “Non-Directed Infrared Links for High-Capacity WirelessLANs,” IEEE Personal Communications, pp. 12–25, 1994.[118] W. Y. Chen, “Ethernet,” in Home Networking Basis: Transmission En-vironments and Wired/Wireless Protocols, 2004, pp. 189–276.[119] N. Sitthimahachaikul, J. P. Keane, and P. J. Hurst, “An Adaptive DFEusing an IIR Feedback Equalizer for 100BASE-TX Ethernet,” in Circuitsand Systems, 2004. NEWCAS 2004. The 2nd Annual IEEE NortheastWorkshop on, 2004, pp. 173–176.[120] J. Baek, J. Hong, and M. H. Sunwoo, “Novel digital signal processing unitfor Ethernet receiver,” in Circuits and Systems, 2005. ISCAS 2005. IEEEInternational Symposium on, 2005, pp. 4477–4480.[121] H. L. Minh, D. O’Brien, G. Faulkner, L. Zeng, K. Lee, D. Jung, andY. Oh, “80 Mbit/s visible light communications using pre-equalized whiteLED,” in Optical Communication, 2008. ECOC 2008. 34th EuropeanConference on, 2008, pp. 223–224.104Bibliography[122] D. O’Brien and G. Faulkner, “Equalisation for high-speed VisibleLight Communications using White-LEDs,” in 2008 6th InternationalSymposium on Communication Systems, Networks and Digital SignalProcessing, 2008, pp. 170–173.[123] H. L. Minh, D. O’Brien, G. Faulkner, L. Zeng, K. Lee, D. Jung, andY. Oh, “High-speed visible light communications using multiple-resonantequalization,” Photonics Technology Letters, IEEE, vol. 20, no. 14, pp.1243–1245, 2008.[124] J. M. Cioffi, “Equalization,” in Data Transmission Theory, ch. 3, pp. 147–291.[125] Cisco, “Cisco Catalyst 4500E Supervisor Engine 7-E : Leading BorderlessNetwork Access and Aggregation Supervisor Engine,” pp. 1–12, 2011.[126] Texas Instruments, “TPS23756EVM User’s Guide,” pp. 1–15, 2010.[127] OSRAM, “Golden DRAGON LCW W5SM-JYKY-JKQL-1,” pp. 1–20,2012.[128] ——, “SFH 2701 High Speed PIN Photodiode,” pp. 1–10, 2008.[129] M. Rahaim, A. M. Vegni, and T. D. Little, “A hybrid radio frequency andbroadcast visible light communication system,” in 2nd IEEE Workshopon Optical Wireless Communications, 2011, pp. 792–796.[130] T. Yamazato, I. Takai, H. Okada, T. Fujii, T. Yendo, S. Arai, M. Andoh,T. Harada, K. Yasutomi, K. Kagawa, and S. Kawahito, “Image-Sensor-Based Visible Light Communication for Automotive Applications,” Com-munications Magazine, IEEE, vol. 52, no. 7, pp. 88–97, 2014.[131] Y. Wang, N. Chi, Y. Wang, L. Tao, and J. Shi, “Network Architecture ofa High-speed Visible Light Communication Local Area Network,” IEEEPhotonics Technology Letters, vol. PP, no. 99, pp. 1–4, 2014.105


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items