1

A Novel Visible Light Communication SystemBased on a SiPM Receiver

Zhenzhou Deng, Anyi Li

Abstract—Si-Photomultiplier (SiPM) has several attractingfeatures, which can be helpful in the communication field,such as high photon detection efficiency, fast transient response,excellent timing resolution, and wide spectral range. In thispaper, we compare SiPM with photodiode (PD), Avalanchephotodiodes (APD), and Photomultiplier tube (PMT) in terms ofthe transformation related performance, and then implement Si-Photomultiplier (SiPM) based visible light communication (VLC)system driven with visible light light-emitting diode (LED). Thesystem throughout ability, transmission rate, data reconstruction,and required AD sample rate are evaluated. The encouragingresults suggest that the SiPM receiver has great applicationpotentials, such as optical wireless communication systems andlight fidelity, in which a wide bandwidth of the sensor responseis important to enhance the transfer rate.

Index Terms—Silicon-Photomultiplier, visible Light Communi-cation, mean Square Error, sample rate.

I. INTRODUCTION

Self-sustainable green-smart houses [1], optical wirelesscommunication (OWC) [2] and light fidelity (Li-Fi) [3] haveattracted considerable attention, in which users always de-mand large data capacity and multi-functional lighting controlduring daily life. In contrast to traditional Wi-Fi and fiber-optic communications, the visible light communication (VLC)system [4] is a complementary system with advantages ofcustomizable space, license free, electromagnetic immunity,communication safety and so on. The VLC has found itssuitable application scenarios in mobile connection [5], indoorpositioning [6], vehicle transportation [7], targeted commu-nication [8], underwater resource exploration [9] and hospi-tal/healthcare applications [10], [11], [12], [13], [14].

At this stage, some of the conceptual VLC prototypes havealready been implemented in practice for commercial appli-cations. Several worldwide companies including Bytelight,Target, Emart, and Royal Philips NV have successfully guidedshoppers to goods based on their position in retail stores,in which light-emitting diodes (LEDs) [15] based VLC areemployed to communicate with the camera of smartphones.Other than these applications, intelligent medical care system[16] using VLC has been established and working in severalhospitals through the cooperation with the Industrial Tech-nology Research Institute (ITRI), which stably and preciselyrealizes the wireless streaming of therapeutic data and thepositioning sensing of medical personnel. According to the

Zhenzhou Deng is from School of Life Science and technology, HuazhongUniversity of Science and Technology, Wuhan, China.

Anyi Li, Zhitao Liu are from School of Qianhu, Nanchang University,Nanchang, China.

report from Grand View Research, all the expected revenue ofthe global VLC market will reach up to USD 101.30 billionby 2024 [17], [18], [19], [20], [21].

Fig. 1. Four different photon sensors: PD, APD, PMT, SiPM, from left toright. We can see the PMT has much more bulky size than other photo sensors.SiPM use surface mount package which will be feasible in more applications.

Almost all these VLC applications employ photodiode(PD) to achieve photo-electronic transition and receive data.However, PD still has many shortcomings, such as smallarea, no internal gain, much lower overall sensitivity (theuse of photon counting is limited, usually used for coolingwith special electronic circuit photodiodes). Silicon photo-multipliers (SiPMs) [22], [23], also identified as multi-pixelphoton counters shown in Fig. 2, are a favorable class ofsemiconductor-based photodetectors addressing the challengeof detecting, timing and quantifying low-light optical signalsdown to the single-photon counting level. SiPMs offer a highlyattractive alternative that extremely increases the detectionsensitivity of photo-electronic transition while providing allthe benefits of PD [24], [25], [22], [26], [27].

In this paper, we will focus on a new particular applicationof SiPM to the visible light communication field. In the nextsections, we will discuss the choice cause of SiPM for VLCamong different sensors in Fig. 1. Next, we develop a SiPMbased VLC prototype link. Then the related performances wereinvestigated in Section V. At the end, we provide a summaryfor the VLC application of SiPM.

The following abbreviations used in this manuscript are listin Table. II.

II. COMPARISON WITH OTHER DETECTORTECHNOLOGIES

Silicon photomultipliers (SiPMs, also called SPM, GAPD,MPPC), are a favorable class of semiconductor-based pho-todetectors addressing the challenge of detecting, timing andquantifying low-light optical signals down to the single-photoncounting level. SiPMs offer a highly attractive alternativethat keeps the low-light detection capabilities of traditional

arX

iv:1

909.

0064

1v1

[ph

ysic

s.in

s-de

t] 2

Sep

201

9

2

TABLE IFEATURE COMPARISON WITH OTHER DETECTOR TECHNOLOGIES

Features PD/PIN APD PMT SiPMGain 1 102 106 106

Operational Bias Low High High LowTemp.Sensitivity Low High Low Low

Package Size Compact Compact Bulky CompactMechanical Robustness Heigh Medium Low High

Ambient light exposure? OK OK NO OKSpectral range Red Red Blue/UV Blue

Readout/ Electronics Complex Complex Simple SimpleLarge area available? No No Yes Yes

Sensitive to magnetism? No No Yes NoNoise Low Medium Low High

photomultiplier tubes while providing all the benefits of othersolid-state devices.[28], [29], [30], [31], [32] During the last

Cathode(k)

Rq Rq Rq Rq

Anode(A）

GM-APD

GM-APD

GM-APD

GM-APD

Rq

CdVbd

Rs

GM-APD

Fig. 2. The inside structure of SiPM. Schematics of a modern siliconphotomultiplier (SiPM) is composed of an array of single photon avalanchediode microcells with passive quenching. It makes SiPM output recovers inshort time, and has a greater slew rate on the pulse leading edge.

decade, the SiPM has become more and more popular asphoton sensor, especially for applications where the need fora large sensitive area is not an issue but the photon countingcapability, pulse bandwidth, and time resolution are important.VLC is only one of the potential applications, in which highpulse bandwidth guarantees the high transfer rate in the lightinformation system.

Such a photodetector probabilily paves the way to compactphotonic systems which can be easily embedded into smallerand smaller frameworks. Innumerable applications are under-way or foreseen which make use of SiPM, as for instance innuclear physics, nuclear medicine, biochemistry, 3D imaging,automotive and several other fields.

Low-light photon detectors constitute the technology ofmany rapidly growing, such as radiation detection, nuclearmedicine imaging, spectroscopy, fluorescence analysis, qualitycontrol, etc. all require detectors for quantifying and/or time-marking optical signals with any place from 1 to 20,000 pho-tons per signal pulse. The ideal detector provides a responseproportional to the incident photon flow and uses an internalgain mechanism to generate a signal large enough to be easilyprocessed. It has sub-nanosecond response time and broadspectral sensitivity, be rugged, easy to operate, and produceonly manageable noise or dark counting rates.

Up to now, the Photomultiplier tube (PMT), a well-established and widely available vacuum tube device, has

become the detector of choice for such applications. Thesemi-transparent photocathode deposited inside the entrancewindow inherently limits the Photon Detection Efficiency(PDE) that they can achieve, with typical PMTs having about20% at 420nm. A gain of (1 − 10) × 106 is achieved at thecost of a high bias voltage of 1-2kV, which requires the use ofcostly high-voltage power supplies. PMT is generally stableand low noise, but it is bulky and small due to its vacuumtube structure. Magnetic films can also adversely affect them,limiting their suitability for certain applications. The fragileand bulky feature of PMT extremely limited the applicationof VLC.

Semiconductor devices have many practical advantages overthe PMT, and this led to the Positive-Intrinsic-Negative(PIN)diode being used in applications where PMTs are too bulkyor delicate, or where high voltages are not possible. However,PIN diodes are severely limited by their complete lack ofinternal gain. Avalanche photodiodes (APDs) are a relativelynew technology for the extension of simple PIN diodes. Herethe reverse bias increases to a point where impact ionizationallows for some internal multiplication but is below the break-down bias where the Geiger mode would take over. Thus, at abias of 100-200V, a gain of about 100 can be obtained. Withspecial manufacture, it is possible for gains of several thousandto be reached using a High voltage bias of more than 1500V.Whilst the gain may be lower than that of a PMT, APDshave the advantage of a PDE which can be >65% and also acompact size, ruggedness, and insensitivity to magnetic fields.Their main drawbacks are their excess noise (associated withthe stochastic APD multiplication process) and in an importanttrade-off: The capacitance increases with the device area anddecreasing thickness, whereas the transit times of the chargecarriers increase with the thickness, implying a performancetrade-off between noise and timing. Their size is limited to 10mm diameter.

The SiPM has high gain and enough PDE (more than 20%),very similar to the PMT, but has the physical benefits of com-pactness, ruggedness and magnetic insensitivity in commonwith the PD and APD. In addition, the SiPM achieves its highgain (1 × 106) with very low bias voltages ( 30V) and thenoise is almost entirely at the single photon level. Because thehigh degree of uniformity between the microcells the SiPM iscapable of discriminating the precise number of photoelectronsdetected as distinct, discrete levels at the output node. Theability to measure a well-resolved photoelectron spectrum isa feature of the SiPM which is generally not possible withPMTs owing to the variability in the gain or excess noise.

To summarize, among all the photosensors in Table. I, theextremely remarkable performance achieved by SiPM sensorsin terms of high photon detection efficiency, fast transientresponse, excellent timing resolution, and wide spectral range,has made considerable research activities and technologicaldevelopment to be constantly devoted to SiPM devices withinthe scientific community involved in medical imaging, high-energy physics, and astrophysics. Here, we aim at the realiza-tion of specifically designed LED driven VLC architecture toencode, transfer and decode the light or electrical signals con-taining information and data. Because the external electronics

3

2. AWG Modulates data into electrical

signals

1. Computer generates the data

to transfer

3. LED converts the electrical signals to

optical signals

5. OSC converts electrical signals

into digital signals

6. Computer restore the data from digital

signals

4. SiPM converts the optical signals to electrical signals

7. Data Comparison and

Efficiency Evaluation

Fig. 3. Operating procedure. In the experiment, there were 7 stages: Computer generates the data to transfer, arbitrary waveform generator(AWG) Modulatesdata into electrical signals, LED converts the electrical signals to optical signals, SiPM converts the optical signals to electrical signals, Oscilloscope(OSC)converts electrical signals into digital signals ,Computer restore the data from digital signals, Data Comparison and Efficiency Evaluation.

need to be close to the detector. SiPM with SensL, having anoperational bias of 30V, meet the requirements of the ExtraLow Voltage directive.

III. SIPM BASED VLC PROTOTYPE IMPLEMENTATION ANDEXPERIMENT

Our research team has experience in the design and con-struction of dedicated positron emission tomography (PET)systems using SiPM, starting with the small animal PET calledTrans PET [33], [34], [35], [36], and then the human PET. Inorder to improve the performance of the photodetector, a sipmsarray capable of providing high intrinsic spatial resolution andgood timing response is designed, and at the same time, amagnetic film can be used. To demonstrate the advance ofSiPM detection in VLC, we implement the prototype utilizingSiPM and setup the following experiments.

A. Prototype VLC System Using SiPM

In the experiment, there were 7 stages as showed in Fig. 3.Computer generates the data to transfer, arbitrary waveformgenerator(AWG) Modulates data into electrical signals. LEDconverts the electrical signals to optical signals. SiPM convertsthe optical signals to electrical signals. Oscilloscope(OSC)converts electrical signals into digital signals. Computer re-stores the data from digital signals, Data Comparison, andEfficiency Evaluation.

As shown in Fig. 4, a source LED chip provided byNanchang University was driven by a current amplifier feed bykeysight arbitrary waveform generator(AWG) M8195S, which

can provide 65 GSamples/s and a 20GHz bandwidth, andeach 1-slot Axie module has 1, 2 or 4 channel configurations.And then an optical fiber, optically connected the LED toa MicroFB-30035 SMT Sensl SiPM, while the SiPM signalis output to the current amplifier. Despite the fact that theSiPM is sensitive to single photons, its dark count rate of100kHz/mm2 at room temperature renders it unsuitable foruse for applications at very low light levels. However, withthe application of cooling technology (such as with the SensLMiniSL), a two-order reduction in dark count rate is easilyachieved. Sensl has more efficient products now, which havemore PDE and faster output. We can expect more applicationwhen the employing SiPM has better performance. The SiPMis supported with an SMA package connector and its outputis directly connected to a keysight digital storage oscilloscopeDSOX6002A with a 50 Ohm termination. The oscilloscope isoperated with a 6 GHz bandwidth and a 10 GSps sampling rateper channel. At the LED-end, the data are wrapped as electricalpulses serious at different repeat rates: 2 4 5 8 10 20 40 50 80100 200 400 500MSps. The input data from AWG are encodedas different pulse amplitudes. At the SiPM-end, SiPM pulsescontaining data are digitized by the oscilloscope. Due to theevidential and fast rise of the pulse shape of SiPM, the sent-inbits can be easy discriminated after digitized. Since all pulseshave tails lasting more than 100 ns, a pulse reconstructionprocessing is required to obtain the transfer data, which canbe verified with the recorded original AWG data.

4

Amp

Data decodingDSA

7m

TO-canPackagedViolet LED

AWG

Amp

Dataencoding

Receiver

SMA Packaged

LED Shield for

Fast modulation

0.5m

Filter

LED SiPM OptionalCase

Receiver with Filter

Fig. 4. The proposed and implemented VLC system using SiPM. An source LED chip provided by Nanchang University was driven by a current amplifierfeed by keysight AWG M8195S.And then an optical fiber optically connected the LED to a MicroFB-30035 SMT Sensl SiPM, while the SiPM signal wereoutput to current amplifier.

50 100 150 200 250 300 350 400

Time (ns)

-15

-10

-5

0

Am

plitu

d (V

)

Fig. 5. Pulses pile up are common when the repeat frequency are higher than the reciprocal value of the pulse duration. SiPMs always have capacitivecharacteristic, then produce pulse tails. The SiPM was support with an SMA package connecter and its output was directly connected to a keysight digitalstorage oscilloscope DSOX6002A with a 50 Ohm termination.

B. Pulse Reconstruction

Caused by the capacitive characteristic, the transferringpulses always pileup with several adjacent ones as shown inFig. 5. Thus, amplitude of each pulse could be affected whenthe pileup effect is not considered. To guarantee the receivedthe data consistency and integrality, the pulse reconstructionmethod is employed. An amplitude modulated SiPM pulsecan be modeled as an impulse function fi(t) = Eiδ(t − ti),where δ(t− ti) is a shifted Dirac Delta function. For the pulsewith index i, its amplitude and arrival time are denoted asEi and ti = i∆T , respectively. Here, ∆T = 1/fr, wherefr is repeating rate, and ∆T is pulse interval. The overallinputs of the system are a series of photons pulse and canbe represented as f(t) =

∑fi(t). The output signals are

waveforms denoted as p(t). The VLC system is then expressedas a linear convolution equation

p(t) = f(t) ∗ ϕ(t) + n(t), (1)

where ϕ(t) is the system’s unit impulse response function,n(t) is the noise and ∗ is the convolution operator. If the

pulse interval ∆t of adjacent pulses is too close, there will bepileups in p(t). Otherwise, p(t) contains only single events. Toretrieve the transferred data Ei, we need to solve the inverseproblem of (1), and use the collected signal p(t) to computef(t).

In practice, signals mentioned above are handled in theirdiscrete forms. We denote ~f = {fj}, ~p = {pl} and ~n = {nl}as the discrete forms of the input, output and noise signalsequences, respectively. The system impulse response functionis also expressed as a vector ~ϕ = {ϕr}, where ϕr = ϕ(r ·∆t)and ∆t is the sampling interval. r, j and l are elementindexes in the corresponding vectors. For the convenienceof computation, the convolution operation in (1) is rewrittenas matrix multiplication. A Toeplitz matrix H = T{~ϕ} isgenerated from ~ϕ and the convolution (1) is transferred to

~p = H ~f + ~n. (2)

The element of H is denoted as

hlj = ϕl−j , (3)

5

where ϕl−j is the (l− j)th element in vector ~ϕ. Obtaining ~ffrom (2) is an inverse problem which can be solved by manykinds of well established methods, including our proposedSuccessive Remove (SR) method, pulse model based iterativedeconvolution (PMID) method and Fourier deconvolution (FD)method.

In the numerical experiment, the reference method ispulse model based iterative deconvolution [37] and Fourierdeconvolution[38], [39]. PMID method employs MLEM it-eration to obtain the pulse amplitude. The general form ofMLEM algorithm that expresses the updating procedure at thekth iteration is

fkj =fk−1j∑l

hlj

∑l

hljpl∑

j

hljfk−1j

, (4)

where fkj denotes jth element of solution ~f at the kth iteration.With enough rounds of iterations, the solution will approachto the original impulse function ~f . PMID method can bedescribed as Algorithm 1:

Algorithm 1: Pulse Model based Iterative Deconvolution(PMID) Algorithm.

Data: Data Sample {S(t), t = t+ k4 t }, 4t is thesampling period, ∆T is pulse interval, t < N∆T

1 The start solution f0 = 1, which is a all-1 vector.2 Calculate

∑l

hlj .

3 do4 Forward Projection

∑j

hljfk−1j ;

5 Calculate pl∑jhljf

k−1j

;

6 Backward Projection∑l

hljpl∑

jhljf

k−1j

;

7 Calculatefk−1j∑l

hlj;

8 Calculate fkj =fk−1j∑l

hlj

∑l

hljpl∑

jhljf

k−1j

.

9 while r < M , where M is the loop times.;10 Output the Ek, k = 1, 2, 3..., N. from fM .

And Fourier Deconvolution employs regulated Fourier do-main division to calculate Wiener Filter as Eq. 5, whichexecutes an optimal tradeoff between inverse filtering andnoise smoothing. It eliminates additive noise while invertingblur. The Wiener filtering is optimal in terms of the MeanSquare Error(MSE). FD method can be described as Algorithm2 as: 2:

fFD =z{p}(|z{ϕ}|2)

z{ϕ}(|z{ϕ}|2 +K). (5)

where, z{·} denotes Fourier Transform, and K is a constant,which is defined by the noise property.

The VLC system employs an Successive Remove (SR)method. Successive Remove (SR) starts with a pulse withZERO height forward pulses {S{t}, 0 ≤ t < T}. Eachpulse height estimation follows the Remove operation, whichreduces the effect of forwards pulses. SR method can bedescribed as Algorithm 3:

Algorithm 2: Fourier Deconvolution (FD) Algorithm.Data: Data Sample {S(t), t = t+ k4 t }, 4t is the

sampling period1 Calculate the Fourier Transform of ϕ;2 Calculate the Fourier Transform of p;3 Zero padding z{ϕ} to the same length as z{p};4 Calculate fFD = z{p}(|z{ϕ}|2)

z{ϕ}(|z{ϕ}|2+K) ;5 Output the Ek, k = 1, 2, 3..., N. from fFD.

Algorithm 3: Successive Remove (SR) Algorithm.Data: Data Sample {S(t), t = t+ k4 t }, 4t is the

sampling period, ∆T is pulse interval1 The 1st SiPM pulse has no precomponent, then

E1 =∫4T

0S(t)dt.

2 do3 Estimation the precomponent Sk from previous

pulses to kth pulse;4 Calculate the integral I(k) between (k − 1)4 T and

k4 T ;5 Calculate Ek by removing the precomponent S(k),

Ek = I(k)− S(k)6 while t < N 4 T ;7 Output the Ek, k = 1, 2, 3..., N.

IV. SIMULATION RESULTS

A. Single Photon Level Detection

To overcome proportionality lack in APD, the SiPM in-tegrates a dense array of small, electrically and opticallyisolated Geiger-mode photodiodes. Each photodiode elementin the array is referred as a ”microcell”. Typically numberingbetween 100 and 1000 per mm2, each microcell has its ownquenching resistor. The signals of all microcells are thensummed to form the output of the SiPM. Each microcelldetects photons identically and independently. The sum ofthe discharge currents from each of these individual binarydetectors combines to form a summed photons output and isthus capable of giving information on the magnitude of anincident photon flux. A spectrum of the same pulse is shownin Fig. 7 , and the response to low-level light pulses is shownin Fig. 6.

B. Pulse shape and SiPM Band Width

In the VLC system, different optical pulses, are fed toSiPM, and generated by LED, and then produced the pulseschain. Transferred data are modulated at the height of opticalpulses. We find the allowing data rate without errors dependslargely on the single pulse response of the SiPM output.Sharper the SiPM response, more possible transfer rate can beafforded. To know the SiPM response bandwidth, we evaluatethe SiPM pulse and its frequency response. Fig. 8 shows theimpulse response of SiPM, which have a sharp pulse peak withfast leading and recovery edge. And we employ Fast FourierTransform to obtain its frequency response in Fig. 9. In Fig. 9,we also can see the cut-off frequency that exceeds 100 MHz

6

5 photons

Fig. 6. Responding intensity graph integrated by pulse shapes.Each photodi-ode element in the array is referred to as a ”microcell”. Typically numberingbetween 100 and 1000 per mm2, each microcell has its own quenchingresistor. The signals of all microcells are then summed to form the outputof the SiPM.

0 500 1000 1500Pulse Charge (fC)

0

1000

2000

3000

4000

5000

6000

Cou

nts

0 photon1 photon

2 photons

4 photons

3 photons

5 photons

Fig. 7. The single photons’ energy spectrum.To overcome proportionality lackin APD, the SiPM integrates a dense array of small, electrically and opticallyisolated Geiger-mode photodiodes.Each microcell detects photons identicallyand independently. The sum of the discharge currents from each of theseindividual binary detectors combines to form a summed photons output, andis thus capable of giving information on the magnitude of an incident photonflux.

in Fig. 9, which is significantly better than common photondevices.

C. Reconstruction Precision and Required Time

It also can be noted that recovery edge is followed bypositive tails in the SiPM pulses. Then pulse aliasing cannotbe avoided when the repetition rate is enough to keep the datatransfer rate. Thus we employ three different reconstructionmethods to inverse the data input: SR, PMID, and FD. Whencarrying out the calculation, 100 iterations PMID elapsed morethan 105 times which takes more computing time than SRmethod. And all the 1000 MByte pulse chain with 10 Gspssample rates is too high to encode by PMID method. To obtainthe reconstruction precision of the three methods, we calculatethe residual error for just 100 MByte pulse chains.

In Fig. 10, Residual Error of each method is evaluated.Residual Error here means the difference between the receivedsample and theoretical value calculated from reconstruct data,which is calculated by Eq. 6.

100 200 300 400 500 600 700 800 900 1000

Time (ns)

-0.1

-0.08

-0.06

-0.04

-0.02

0

Am

plitu

de (

mV

)

Fig. 8. Mean pulse of the SiPM pulse. Here, the mean pulse is obtainedfrom averaged pulses output by the SiPM on low event rate. Low event ratepulses avoid pileup case. The mean pulse has a sharp pulse peak with fastleading and recovery edge.

100 200 300 400 500 600 700 800 900 1000

Feq (MHz)

-2

0

2

4

6Angle

Modulus

Fig. 9. Frequency property of the SiPM pulse. The VLC system transferdata using the repeating pulses amplitude. So the frequency width is expectedhigh for more transfer data rate.

r = |−→p −H−→f∗|2, (6)

where f∗ is the vector containing reconstruct data. With theincrease of Repeat Rate, Residual Error of three algorithmskeeps rising. By contrast, SR and PMID algorithm tend to bestable, and is not affected by Repeat Rate too much. We cansee that SR and PMID produce one order of magnitude lowerresidual error than FD. Considering SR processed data freeof whole data iteration, we believe the SR is the appropriatemethod in the practical system. Since each PMID iterationneeds 2 convolution operators and 3 vector multiplicationoperators, it has large computing burden when the iterationnumber and sample rate increased. We also find the requiredreconstruction time is affected by the pulse repetition rate. InFig. 11, we investigate the reconstruction time of SR method.We find the required reconstruction time is decreased as therepetition rate increases. This can be illustrated as that fewersamples need processing when the repetition rate increases.

7

2 4 5 8 10 20 40 50 80 100 200 400 500Repeat Rate (MSps)

0

20

40

60

80

100

120

Res

idua

l Err

or (

mV2

/s)

SR PMID FD

Fig. 10. The residual errors produced by SR, PMID and FD method.Residual Error here means the difference between received sample and theoretical valuecalculated from reconstruct data ,is calculated by Eq. 6.

2 4 5 8 10 20 40 50 80 100 200 400 500

Repeat Freqency (MHz)

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Rec

on T

ime

(s)

Fig. 11. Required time to restore the pulses chain containing 1 Mbytesusing SR method. These time values are calculated on computer workstation.The comparison of elapsed time was based on the hardware of Intel Core i3-3220@3.30GHz and 8GB 1600MHz Kingston DDR3 memory. The programwas implemented in MATLAB.

D. Required Sample Rate

The required sample rate of SiPM digitizer is another criticalaspect in VLC systems. When the sample rate of the digitizeris insufficient, the digitized signal sample cannot guarantee thecorrectness of pulse reconstruction. And when the sample rateof the digitizer is too high, the cost and the computation timewill be both unaffordable. The above results are obtained fromdata sampled at 10 GSps. The pre-stored pulses data with ahigh sampling rate allows the electronic designer to evaluatetheir DSP algorithms with good operability and flexibility.However, 10 Giga-Samples per second (GSps) is not necessaryfor the VLC system to process the pileups. We investigate thesampling rate’s effect on the SiPM VLC system. Data pointsare picked up at different intervals from the original sequenceto simulate different Analog-to-Digital Converter(ADC) sam-pling rates, ranging from 2 Mega-Samples per second (MSps)to 10 GSps. Thereafter, the re-sampled waveform is screenedand processed by the SR method. We investigate the minimumrequired sample rate of digitizer which all guarantees 1000Mbytes pulses amplitude totally and correctly restores usingre-sample data experiment as shown in Fig. 12. In Fig. 12,

when Repeat Rate is 10MSps, the sampling rate can reach100MSps, and considering the point Repeat Rate 100MSps,the sampling rate value of this point is about 104 MSps, whichproves that our system can achieve high data rate when samplerate is sufficient. Overall, the required sample rate increaseswith the pulse repetition rate. This result suggests that thesample rate of SiPM digitizer can be vital when the superiortransfer rate is expected.

101 102

Repeat Rate (MSps)

100

102

104

Sam

ple

Rat

e (M

Sps

)

Fig. 12. Required minimal sample rate of digitizer to guarantee 1000Mbytes all right restored pulses amplitude. We investigated the sampling rate’seffect on the SiPM VLC system. Data points were picked up at differentintervals from the original sequence to simulate different Analog-to-DigitalConverter(ADC) sampling rates, ranging from 2 Mega-Samples per second(MSps) to 10 GSps.

V. CONCLUSION

Based on the experience of SiPM application in the imagingsystem, we constructed a SiPM based VLC system and thustransferred the data transmission problem to a signal pro-cessing problem. Thereafter, we applied the SR algorithm tosolve this problem. Experiments with real signals showed thatthis method provided good signal pulse fitting at a differentrepetition rates. The pulse reconstruction performance of thismethod was also better than the previously used PMID andFD methods.

The proposed and implemented system has 4 Gbit/s transferrates under one 10 GSps DA/AD channels. This system does

8

TABLE IIABBREVIATIONS TABLE

Abbreviations Full NameVLC visible light communicationPD photo diode

LED light-emitting diodeSiPM Silicon photomultiplierPMT Photomultiplier tubePDE Photon Detection EfficiencyAPD Avalanche PhotodiodePIN Positive Intrinsic-NegativePET positron emission tomography

PMID pulse model based iterative deconvolutionFD Fourier deconvolutionSR Successive Remove

GSps Giga-Samples per secondMSps Mega-Samples per secondADC Analog-to-Digital Converter

not require the mean pulse of signals to be in a particu-lar shape, as long as it is fixed. Since different front-endelectronics in detection systems may produce different pulseshapes, the adaptability of the proposed method is appropriatein applications. We preliminarily conclude the high transferrate of SiPM based VLC system attributes the high bandwidthof SiPM signal pulse and single photon counting ability.

In the future work, we will apply the SiPM to differentcolor visible light and multiple voltage threshold digitizers,instead of a single color LED and the high-speed ADC.Since both Multi-Voltage threshold method and the SR methodrequire the prior knowledge of pulse shape, they could bewell-matched. Other iterative algorithms, such as maximum aposterior (MAP) or ordered subset expectation maximization(OSEM), can also be considered in the pulse recovery process.

REFERENCES

[1] N. Balta-Ozkan, R. Davidson, M. Bicket, and L. Whitmarsh, “Thedevelopment of smart homes market in the uk,” Energy, vol. 60, pp.361–372, 2013.

[2] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wireless communi-cation: potential and state-of-the-art,” IEEE Communications Magazine,vol. 49, no. 9, 2011.

[3] D. Tsonev, S. Videv, and H. Haas, “Light fidelity (li-fi): towards all-optical networking,” in Broadband Access Communication TechnologiesVIII, vol. 9007. International Society for Optics and Photonics, 2014,p. 900702.

[4] T. Komine and M. Nakagawa, “Fundamental analysis for visible-lightcommunication system using led lights,” IEEE transactions on Con-sumer Electronics, vol. 50, no. 1, pp. 100–107, 2004.

[5] R. Ling, The mobile connection: The cell phone’s impact on society.Elsevier, 2004.

[6] H. Liu, H. Darabi, P. Banerjee, and J. Liu, “Survey of wireless indoorpositioning techniques and systems,” IEEE Transactions on Systems,Man, and Cybernetics, Part C (Applications and Reviews), vol. 37, no. 6,pp. 1067–1080, 2007.

[7] N. R. C. U. T. R. B. C. for a Study of Transportation and S. Environment,Toward a sustainable future: addressing the long-term effects of motorvehicle transportation on climate and ecology. National AcademyPress, 1997, vol. 251.

[8] P. Manchanda, Y. Xie, and N. Youn, “The role of targeted communi-cation and contagion in product adoption,” Marketing Science, vol. 27,no. 6, pp. 961–976, 2008.

[9] I. F. Akyildiz, D. Pompili, and T. Melodia, “Underwater acoustic sensornetworks: research challenges,” Ad hoc networks, vol. 3, no. 3, pp. 257–279, 2005.

[10] M. Aminian and H. R. Naji, “A hospital healthcare monitoring systemusing wireless sensor networks,” J. Health Med. Inform, vol. 4, no. 02,p. 121, 2013.

[11] P. H. Pathak, X. Feng, P. Hu, and P. Mohapatra, “Visible light commu-nication, networking, and sensing: A survey, potential and challenges,”IEEE communications surveys &amp; tutorials, vol. 17, no. 4, pp. 2047–2077, 2015.

[12] C. Biton and S. Arnon, “Improved multiple access resource allocationin visible light communication systems,” Optics Communications, vol.424, pp. 98–102, 2018.

[13] M. S. Islim, R. X. Ferreira, X. He, E. Xie, S. Videv, S. Viola, S. Watson,N. Bamiedakis, R. V. Penty, I. H. White et al., “Towards 10 gb/s orthog-onal frequency division multiplexing-based visible light communicationusing a gan violet micro-led,” Photonics Research, vol. 5, no. 2, pp.A35–A43, 2017.

[14] H. Haas, “Visible light communication,” in Optical Fiber Communica-tion Conference. Optical Society of America, 2015, pp. Tu2G–5.

[15] E. F. Schubert, Light-emitting diodes. E. Fred Schubert, 2018.[16] W. A. Cahyadi, T.-I. Jeong, Y.-H. Kim, Y.-H. Chung, and T. Adiono,

“Patient monitoring using visible light uplink data transmission,” inIntelligent Signal Processing and Communication Systems (ISPACS),2015 International Symposium on. IEEE, 2015, pp. 431–434.

[17] Y.-C. Chi, Y.-F. Huang, T.-C. Wu, C.-T. Tsai, L.-Y. Chen, H.-C. Kuo,and G.-R. Lin, “Violet laser diode enables lighting communication,”Scientific Reports, vol. 7, no. 1, p. 10469, 2017.

[18] J. Wang, Y. Che, X. Wang, L. Guo, and J. Li, “Study of led layout inindoor visible light communication and performance analysis,” in AOPC2017: Fiber Optic Sensing and Optical Communications, vol. 10464.International Society for Optics and Photonics, 2017, p. 1046417.

[19] L. U. Khan, “Visible light communication: Applications, architecture,standardization and research challenges,” Digital Communications andNetworks, vol. 3, no. 2, pp. 78–88, 2017.

[20] Y.-C. Chi, D.-H. Hsieh, C.-T. Tsai, H.-Y. Chen, H.-C. Kuo, and G.-R. Lin, “450-nm gan laser diode enables high-speed visible lightcommunication with 9-gbps qam-ofdm,” Optics express, vol. 23, no. 10,pp. 13 051–13 059, 2015.

[21] B. Janjua, H. M. Oubei, J. R. D. Retamal, T. K. Ng, C.-T. Tsai, H.-Y. Wang, Y.-C. Chi, H.-C. Kuo, G.-R. Lin, J.-H. He et al., “Goingbeyond 4 gbps data rate by employing rgb laser diodes for visible lightcommunication,” Optics express, vol. 23, no. 14, pp. 18 746–18 753,2015.

[22] N. Dinu, R. Battiston, M. Boscardin, G. Collazuol, F. Corsi, G. F.Dalla Betta, A. Del Guerra, G. Llosa, M. Ionica, G. Levi et al.,“Development of the first prototypes of silicon photomultiplier (sipm) atitc-irst,” Nuclear Instruments and Methods in Physics Research SectionA: Accelerators, Spectrometers, Detectors and Associated Equipment,vol. 572, no. 1, pp. 422–426, 2007.

[23] F. Corsi, A. Dragone, C. Marzocca, A. Del Guerra, P. Delizia, N. Dinu,C. Piemonte, M. Boscardin, and G. F. Dalla Betta, “Modelling asilicon photomultiplier (sipm) as a signal source for optimum front-enddesign,” Nuclear Instruments and Methods in Physics Research SectionA: Accelerators, Spectrometers, Detectors and Associated Equipment,vol. 572, no. 1, pp. 416–418, 2007.

[24] P. Eraerds, M. Legre, A. Rochas, H. Zbinden, and N. Gisin, “Sipmfor fast photon-counting and multiphoton detection,” Optics Express,vol. 15, no. 22, pp. 14 539–14 549, 2007.

[25] L. H. Braga, L. Gasparini, L. Grant, R. K. Henderson, N. Massari,M. Perenzoni, D. Stoppa, and R. Walker, “A fully digital 8\,\times\,16 sipm array for pet applications with per-pixel tdcs and real-timeenergy output,” IEEE Journal of Solid-State Circuits, vol. 49, no. 1, pp.301–314, 2014.

[26] F. Powolny, E. Auffray, S. Brunner, E. Garutti, M. Goettlich, H. Hille-manns, P. Jarron, P. Lecoq, T. Meyer, H. Schultz-Coulon et al., “Time-based readout of a silicon photomultiplier (sipm) for time of flightpositron emission tomography (tof-pet),” IEEE Transactions on NuclearScience, vol. 58, no. 3, pp. 597–604, 2011.

[27] G. Varkonyi, J. J. McKendry, N. McAlinden, M. D. Dawson, andK. Mathieson, “Data transmission for high-bandwidth neural interfacingusing visible light communication,” in Photonics Conference (IPC),2016 IEEE. IEEE, 2016, pp. 41–42.

[28] H. Yepes-Ramirez, J. F. Castano, R. Gutierrez, L. Castaneda, A. Her-nandez, V. Paschka, J. Calderon-Cordoba, N. Rios-De-Antonio, andC. Montano, “Development of a high performance characterizationsetup for sipms and mpgds towards their integration in mid-large scalesystems,” PoS, p. 1057, 2017.

[29] L. F. Castaneda, R. Gutierrez, H. Yepes-Ramirez, A. Hernandez, andJ. Castano, “High speed visible light comunication system based onsipm,” PoS, p. 842, 2016.

[30] G. Konstantinou, W. Ali, R. Chil, G. Cossu, E. Ciaramella, andJ. Vaquero, “Experimental demonstration of an optical wireless mri

9

compatible pet/spect insert front-end,” in Nuclear Science Symposium,Medical Imaging Conference and Room-Temperature SemiconductorDetector Workshop (NSS/MIC/RTSD), 2016. IEEE, 2016, pp. 1–4.

[31] P. Buzhan, B. Dolgoshein, L. Filatov, A. Ilyin, V. Kaplin, A. Karakash,S. Klemin, R. Mirzoyan, A. Otte, E. Popova et al., “Large area siliconphotomultipliers: Performance and applications,” Nuclear Instrumentsand Methods in Physics Research Section A: Accelerators, Spectrome-ters, Detectors and Associated Equipment, vol. 567, no. 1, pp. 78–82,2006.

[32] R. Bencardino and J. E. Eberhardt, “Development of a fast-neutrondetector with silicon photomultiplier readout,” IEEE Transactions onNuclear Science, vol. 56, no. 3, pp. 1129–1134, 2009.

[33] Z. Deng and Q. Xie, “Empirical bayesian energy estimation for multi-voltage threshold digitizer in PET,” in Nuclear Science SymposiumConference Record (NSS/MIC). IEEE, 2013, pp. 1–5.

[34] Q. Xie, C.-M. Kao, X. Wang, N. Guo, C. Zhu, H. Frisch, W. W. Moses,and C.-T. Chen, “Potentials of digitally sampling scintillation pulses intiming determination in PET,” IEEE Trans. Nucl. Sci., vol. 56, no. 5,pp. 2607–2613, Oct. 2009.

[35] Z. Deng and Q. Xie, “Quadratic programming time pickoff method formultivoltage threshold digitizer in PET,” Transaction on nuclear science,vol. 62, no. 3, pp. 805–813, 2015.

[36] Z. Deng, Y. Liu, Z. Wu, and Q. Xie, “Maximum likelihood solutionof arrival time for the event pulse modeled by filtered inhomogeneouspoisson process in PET detectors,” Journal of Nuclear Medicine, vol. 56,no. supplement 3, pp. 1842–1842, 2015.

[37] Z. Deng, Q. Xie, Z. Duan, and P. Xiao, “Scintillation event energymeasurement via a pulse model based iterative deconvolution method,”Phys. Med. Biol., vol. 58, no. 21, p. 7815, 2013.

[38] R. G. Brown, P. Y. Hwang et al., Introduction to random signals andapplied Kalman filtering. Wiley New York, 1992, vol. 3.

[39] R. Nagaoka, S. Yoshizawa, S. I. Umemura, and Y. Saijo, “Basic studyof improvement of axial resolution and suppression of time side lobe byphase-corrected wiener filtering in photoacoustic tomography,” JapaneseJournal of Applied Physics, vol. 57, no. 7S1, p. 07LD11, 2018.