Towards 10 Gb/s orthogonal frequency division
multiplexing-based visible light communication
using a GaN violet micro-LED
MOHAMED SUFYAN ISLIM,
1,
*
,†
RICARDO X. FERREIRA,
2,†
XIANGYU HE,
2,†
ENYUAN XIE,
2
STEFAN VIDEV,
3
SHAUN VIOLA,
4
SCOTT WATSON,
4
NIKOLAOS BAMIEDAKIS,
5
RICHARD V. P ENTY,
5
IAN H. WHITE,
5
ANTHONY E. KELLY,
4
ERDAN GU,
2
HARALD HAAS,
3
AND MARTIN D. DAWSON
2
1
Li–Fi R&D Centre, the University of Edinburgh, Institute for Digital Communications, King’s Buildings, Mayfield Road, Edinburgh EH9 3JL, UK
2
Institute of Photonics, Department of Physics, University of Strathclyde, Glasgow G1 1RD, UK
3
Institute for Digital Communications, Li–Fi R&D Centre, the University of Edinburgh, King’s Buildings, Mayfield Road, Edinburgh EH9 3JL, UK
4
School of Engineering, University of Glasgow, Glasgow G12 8LT, UK
5
Centre for Advanced Photonics and Electronics, Electrical Engineering Division, Department of Engineering, University of Cambridge, Cambridge
CB3 0FA, UK
*Corresponding author: m.islim@ed.ac.uk
Received 28 November 2016; revised 9 February 2017; accepted 9 February 2017; posted 10 February 2017 (Doc. ID 280671);
published 28 March 2017
Visible light communication (VLC) is a promising solution to the increasing demands for wireless connectivity.
Gallium nitride micro-sized light emitting diodes (micro-LEDs) are strong candidates for VLC due to their high
bandwidths. Segmented violet micro-LEDs are reported in this work with electrical-to-optical bandwidths up to
655 MHz. An orthogonal frequency division multiplexing-based VLC system with adaptive bit and energy loading
is demonstrated, and a data transmission rate of 11.95 Gb/s is achieved with a violet micro-LED, when the non-
linear distortion of the micro-LED is the domi nant noise source of the VLC system. A record 7.91 Gb/s data
transmission rate is reported below the forward error correction threshold using a single pixel of the segmented
array when all the noise sources of the VLC system are present.
© 2017 Chinese Laser Press
OCIS codes: (060.4510) Optical communications; (060.2605) Free-space optical communication; (230.3670) Light-emitting diodes;
(230.3990) Micro-optical devices.
https://doi.org/10.1364/PRJ.5.000A35
1. INTRODUCTION
The increasing demands of communication services are chal-
lenging radio frequency (RF) wireless communications technol-
ogies. The overall number of networked devices is expected to
reach 26.3 billion in 2020 [1]. Visible light communication
(VLC) is a promising solution to the limited availability of the
RF spectrum as the visible light spectrum offers abundant
bandwidth that is unlicensed and free to use. VLC improves
the spectral efficiency per unit area, which enhances the quality
of service in crowded environments and allows for secure and
localized services to be provided.
General lighting is under a rapid transformation to become
semiconductor based due to huge energy savings. This trans-
formation has already enabled applications such as active energy
consumption control and color tuning. Solid state lighting
devices such as gallium nitride (GaN)-based inorganic light
emitting diodes (LEDs) are ubiquitous power-efficient devices
to enable illumination and communications. Commercially
available LEDs have a limited frequency response due to the
yellow phosphor coating on top of the blue LED chips. How-
ever, the slow response of the yellow phosphor can be filtered
out using a blue filter in front of the receiver. Recent results
for VLC using a phosphorescent white LED with adaptive bit
and energy loading were reported at 2.32 Gb/s aided by a two-
staged linear software equalizer [2].
Micro-LEDs are promising candidates in enabling lighting
as a service (LaaS) and Internet of things (IoT). The introduc-
tion of micro-LEDs has enabled high-performance value-added
lighting functions such as VLC and indoor positioning and
tracking [3]. Micro-LEDs are known for their small active areas
enabling high current density injection, which drives the
modulation bandwidth to hundreds of megahertz [4,5]. At
450 nm, micro-LEDs have set the standard for high-speed
VLC. A 60 μm diameter pixel has achieved 3 Gb/s [6], and
more recently a single pixel of a new segmented array has
demonstrated 5 Gb/s [7]. The novel micro-LEDs emitting at
Research Article
Vol. 5, No. 2 / April 2017 / Photonics Rese arch A35
2327-9125/17/020A35-09 Journal © 2017 Chinese Laser Press
400 nm featured in the current work offer a number of advan-
tages over the 450 nm devices previously reported [7]. From
typical trends concerning the internal quantum efficiency
(IQE) of indium GaN-based active regions, comparable IQEs
are expected at 400 and 450 nm, whereas the IQE decreases
steeply at shorter emission wavelengths [8]. For generation of
white light for illumination, the use of violet-emitting LEDs
exciting tricolor (red, green, and blue) phosphors also offers
advantages over the widely used method of combining blue
direct LED emission with a yellow-emitting phosphor. These
include much superior color rendering indices [9,10] and the
absence of a direct blue component, which has proven to be
disruptive to the human circadian rhythm [11]. The micro-
LED die shapes employed in this work are also expected to
be advantageous for efficient light extraction, by analogy with
previous designs employing non-circular emitting areas [12].
VLC is enabled by incoherent illumination from the light
sources. Therefore, only real and positive modulating wave-
forms can be realized. Single carrier modulation schemes such
as on–off keying (OOK), pulse amplitude modulation (PAM),
and pulse width modulation (PWM) are straightforward to
implement. However, the performance of these modulation
schemes degrades as the transmission speed increases due to
the increased inter-symbol interference (ISI). Equalization
techniques can be used to improve the system performance
at significant computation cost [13]. Multi-carrier modulation
techniques such as orthogonal frequency division multiplexing
(OFDM) are promising candidates for VLC. Computationally
efficient single-tap equalizers are straightforward to realize in
OFDM. Adaptive bit and energy loading in OFDM allows
the channel utilization to approach the information capacity
limit. In addition, multiple access can be easily supported in
OFDM by assigning groups of subcarriers to multiple users,
which is known as orthogonal frequency division multiple
access (OFDMA).
Previously, a 40 μm diameter micro-LED at 405 nm
achieved a data rate of 3.32 Gb/s at an optical power of
2.5 mW with electrical–optical bandwidth up to 307 MHz
[14]. In this paper, we present a high bandwidth VLC link at
400 nm. The emitter consists of a single pixel of the segmented
micro-LED array design introduced in Ref. [7 ]. This device
achieves 2.3 mW of optical output power while maintaining
an electrical-to-optical (E-O) bandwidth of 655 MHz.
A VLC system is realized with a modulation bandwidth of
1.81 GHz, evaluated beyond the 3 dB bandwidth of the sys-
tem. A transmission rate of 11.95 Gb/s is presented, when the
nonlinear distortion noise of the micro-LED is the major
source of noise in the system. A record transmission rate at
7.91 Gb/s is presented when all the noise sources of the
VLC system are considered.
2. VIOLET MICRO-LED
A. Design and Fabrication
The design of standard GaN LEDs is based on a large-area chip
assembled on a package that maximizes heat extraction through
an n-pad at the bottom for a flip-chip configuration. This cre-
ates two limitations: a large capacitance due to the package con-
tact area and an upper limit on the current density due to the
rapid self-heating of a large-area chip. The design and fabrica-
tion process of the micro-LED array used in this work is as
reported in our previous work [7]. It consists of two circular
micro-LED arrays, an inner and an outer, containing 5 and
10 pixels, respectively. Originally designed to match the geom-
etry of plastic optical fiber, the inner and outer pixels have
active areas of 435 and 465 μm
2
, respectively. This compares
with the 1256 μ m
2
active area for the 405 nm device in Ref. [14].
Figure 1 shows optical images of this micro-LED array, together
with a schematic of the pixel layout.
The wafer used in this work is for a commercially available
GaN-based LED emitting at 400 nm. In order to fabricate
these arrays, micro-LEDs emitters are etched by inductively
coupled plasma to expose n-type GaN. An annealed Pd layer
is used as a metal contact to p-type GaN. Each emitter is iso-
lated by a layer of SiO
2
. The metallization on the n-type GaN is
formed by depositing a Ti/Au metal bilayer, which fills the area
between each micro-LED and enables an improved current
spreading. This bilayer connects each micro-LED emitter in
order to individually address them. The micro-LED array
allows increasing the total output power with minimal reduc-
tion in performance due to mutual heating between pixels. The
low optical power per pixel in micro-LEDs is a challenge when
combined illumination and communication is considered. This
problem can be addressed by using large arrays of pixels, where
a system capable of handing the communication link over
multiple pixels can be designed to reduce the duty cycle, reduce
the junction temperature on individ ual pixels, and maintain
high efficiency. These investigations are subject to future work.
B. Performance Measurements
The electrical performance of the micro-LED arrays was mea-
sured by a semiconductor analyzer (HP 4155). The optical
power of the arrays under direct current (DC) conditions was
measured using a Si detector placed in close proximity to the
polished sapphire substrate. A spectrometer and a charge
coupled device detection system were used for the collection
Fig. 1. Plan view micrographs of the segmented micro-LED arrays. The magnified micrographs on the right show the array configuration and
individual pixel design. A diagram is also included noting the inner and outer pixels (dimensions in micrometers).
A36 Vol. 5, No. 2 / April 2017 / Photonics Research
Research Article
of electroluminescence spectra. The small signal frequency
response was measured by a network analyzer with a 20 mV
alternating current (AC) frequency sweep signal combined in
a bias-tee with a DC-bias current ranging from 5 to 50 mA.
The optical response was collected by a lens system and focused
onto a fast photodiode (PD) and fed to the network analyzer.
All the measurements were performed at room temperature
with the device directly probed on chip with a high-speed probe
to guarantee minimal parasitic effects.
1. I–V and L–I Characteristics
Devices with linear luminescence–voltage (L–V) characteristics
and high optical power allow for a large dynamic range that can
accommodate large swings of modulating signals, and this sub-
sequently improves the signal-to-noise ratio (SNR) of the VLC
system. The current–voltage (I–V) and luminescence–current
(L–I) characterist ics for the micro-LED are presented in Fig. 2.
The pixels present a shunt resistance responsible for a sub-
threshold turn-on; this is attributed to damaged regions in the
junction and by surface imperfections. Differences between the
inner and the outer pixels are minimal in I–V with a series re-
sistance of 27 and 26 Ω and threshold turn-on voltage of 4.60
and 4.64 V for the inner and outer pixels, respectively. In terms
of optical power, at the roll-over point, the outer pixels achieve
a maximum of 2.79 mW, 17% higher than the inner pixel,
which is expected given the larger active area. This compares
to 2.5 mW from a pixel at 405 nm with a 2.88 times larger
active area [14]. The improvement in the optical power is
due to the improved Pd p-type contact, resulting in 50%
higher optical power compared to Ref. [14]. In addition,
the commercially supplied wafer for this micro-LED gives
better IQE.
2. Frequency Response
The frequency response from 100 kHz to 1.5 GHz for the low-
est and highest bias currents of the testing set are presented in
Fig. 3. At 1.5 GHz modulation, the pixels do not reach the
noise floor of the system, thus providing a large useful band-
width for data transmission. The calculated E-O bandwidth
against current density is shown in the inset of Fig. 3 for the
set of bias currents covering the full operating range. The cur-
rent densities for these pixels are in line with what was previ-
ously seen at 450 nm [7], meaning that the pixels achieve
similar bandwidth at the same current density. This is also com-
pared with 307 MHz bandwidth for the 405 nm device in
Ref. [14]. The improvement in bandwidth over [14] is due
to the smaller active area of the device that allowed higher cur-
rent den sity and shorter carrier lifetime. With a bandwidth of
655 MHz for the inner pixel, to the authors’ best knowledge,
this micro-LED has the highest bandwidth yet shown in the
violet wavelength band.
3. VLC SYSTEM
A. Optical OFDM
Multiple variants of OFDM have been proposed for VLC [15].
Conventional OFDM waveforms are both complex and bipo-
lar; however, Hermitian symmetry can be imposed on the
OFDM subcarrier frame to realize real-valued OFDM wave-
forms, X kX
N
FFT
− k, where N
FFT
is the OFDM frame
size, and k is the subcarrier index. In addition, subcarriers at
X 0 and X N
FFT
∕2 are set to zero. DC-biased optical OFDM
(DCO-OFDM) uses a DC bias to shift most of the negative
real-valued OFDM samples into positive. The block diagram
for OFDM is shown in Fig. 4. The generation of DCO-
OFDM in this VLC experiment starts with generating a
real-valued OFDM waveform in MATLAB. A pseudo-random
bit sequence (PRBS) is generated and then modulated using
quadrature amplitude modulation (QAM). Given the a priori
estimated SNR, the M
k
-QAM constellation size at subcarrier k
and its corresponding relative energy, ν
2
k
, are adaptively
allocated based on the probability of error target, P
T
e
.
The QAM symbols are loaded into orthogonal subcarriers
with subcarrier spacing equal to the symbol duration. The
OFDM frame size is set to N
FFT
1024 subcarriers. Smaller
sizes for the OFDM frame result in less statistical significance;
larger sizes result in an increased peak-to-average power ratio
(PAPR). The symbols can then be multiplexed into a serial time
Fig. 2. Combined current–voltage (I–V), left, and luminescence–
current (L–I), right, characteristics of both inner and outer pixels.
The inset shows the emission spectrum of an inner pixel at 50 mA.
)B
d
(
esno
p
s
e
r ycneuqerF
Fig. 3. Small signal frequency response for the inner pixel at 5 and
50 mA. The inset shows 6 dB E-O bandwidth at different values for
the current density J, corresponding to DC-bias values of 5–50 mA.
Research Article
Vol. 5, No. 2 / April 2017 / Photonics Research A37
domain output using an inverse fast Fourier transform (IFFT).
Cyclic prefix es (CPs) are inserted at the start of each OFDM
frame. Adequate length of the CPs, N
CP
,allowsforISItobe
eliminated by the computationally efficient single-tap equalizer.
AvalueofN
CP
5 is found to be sufficient for the ISI to
be removed at less than 0.97% loss in the spectral efficiency.
Root-raised cosine (RRC) pulse shaping filter is used to achieve
band limited communication since it allows a trade-off control
between pulse duration and bandwidth requirements [16].
OFDM time domain waveforms have high PAPR due to the
coincidence of multiple in-phase QAM symbols in the same
OFDM frame. Extreme values for the OFDM modulating signal
are clipped to minimize the effect of nonlinearity at acceptable
error margins. The upper and lower clipping values are set to
3σ
x
and −3.5σ
x
, respectively, where σ
x
is the standard
deviation of the OFDM waveform. Asymmetric values for the
clipping points are used since the upper clipping due to the sat-
uration of the micro-LED is higher than the lower clipping. The
received waveform is processed with matched filters, fast Fourier
transform (FFT) with CPs removal, single-tap equalizer using the
aprioriestimated channel, and demodulator. Bit error rate (BER)
is calculated based on the demodulated binary stream.
Before any data transmission, the channel is first estimated
by pilots composed of multiple OFDM frames. A conventional
mean estimator is used with random pilots that would take the
nonlinearity effects into account. Details about the used esti-
mation method can be found in [17]. An estim ation of the
SNR is also obtained using the same method. The received
OFDM waveform, yt, can be expressed as follows:
ythtzxt nt; (1)
where ht is the VLC system channel, nt is the additive
white Gaussian noise (AWGN) at the receiver with a variance
σ
2
n
, and z· is the nonlinear transformation of the micro-LED.
For Gaussian inputs such as the real-valued OFDM waveform,
the Bussgang theorem can be applied and the nonlinear trans-
formation can be expressed as [18]
zxt αxtdt; (2)
given that the processes xt and dt are uncorrelated
Extdt 0,whereE· is the statistical expectation and
dtis the distortion noise. The constant α can be calculated as [18]
α
Ezxt · xt
σ
2
x
: (3)
The distortion noise dt is a non-Gaussian noise. However,
its representation in the frequency domain Df follows a
Gaussian distribution with a DC mean and a variance σ
2
d
[18].
Detailed analysis of the nonlinear distortion noise effect on
DCO-OFDM can be found in Ref. [19]. The used arbitrary
waveform generator (AWG) has 10 bits resolution for the
digital-to-analog converter (DAC), and the oscilloscope used
has an effective number of bits of 5.5 for the analog-to-digital
converter (ADC). The nonlinearity effect from the amplifier is
minimal at the operational frequencies and at the injected
power levels. The harvested optical power at the photoreceiver
is well below the saturation level. Therefore, the micro-LED is
assumed to be the main source of nonlinearity in the overall
system due to the relatively limited dynamic range, compared
to other system components.
The estimated SNR is used to adaptively load the subcarriers
with variable constellation sizes at different energy levels
based on the Levin–Campello algorithm [20]. The algorithm
allows more energy to be allocated to the subcarriers, which
require minimal additional power to be elevated into larger
constellation sizes, while preserving the probability of error
target, P
T
e
. Assuming that N
FFT
> 64, the adaptive bit and
energy loading can be formulated in the following optimization
problem:
maximize
η
η
P
N
FFT
2
−1
k1
M
k
>0
log
2
M
k
N
FFT
N
CP
1 β
; (4a)
subject to BER
M
k
;
ν
2
k
α
2
E
bk
N
o
∕jHkj
2
σ
2
d
≤ P
T
e
; (4b)
X
N
FFT
2
−1
k1
M
k
>0
ν
2
k
N
FFT
2
− 1
1; (4c)
where β is the roll-off factor of the RRC filter, E
bk
is the energy
per bit at subcarrier k, N
o
is the double-sided power spectral
density (PSD) of the noise at the receiver, jHkj
2
is the chan-
nel gain at subcarrier k when a zero forcing (ZF) equalizer is
used, η is the spectral efficiency given in bits/s/Hz, and BER
(M
k
, γ
k
) is the theoretical BER equation of M
k
-QAM at sub-
carrier k and SNR per bit γ
k
, given in non-flat channels as [21]
BERM
k
; γ
k
≅
4
log
2
M
k
1 −
1
ffiffiffiffiffiffiffi
M
k
p
×
X
R
l1
X
N
FFT
k1
Q
2l − 1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 log
2
M
k
γ
k
2M
k
− 1
s
!
;
(5)
where Q· is the complementary cumulative distribution
function (CCDF) for the standard normal distribution, and
R min2;
ffiffiffiffiffiffiffi
M
d
p
.
The adaptive bit and energy loading on each subcarrier is
shown in Fig. 5 along with the channel capacity limit defined
by Shannon as [22]
C log
2
1
α
2
E
bk
N
0
∕jHkj
2
σ
2
d
: (6)
It is shown that the gap between the exact loading and the
capacity limit is already small; however, it can be closed when
Fig. 4. Block diagram for OFDM transmitter and receiver.
A38 Vol. 5, No. 2 / April 2017 / Photonics Research
Research Article
channel coding is employed. The cumulative distribution func-
tion (CDF) of the estimated SNR for multiple QAM constel-
lation sizes is presented in Fig. 6. The results show the
distribution of the SNR values required to achieve a BER below
the forward error correction (FEC) target based on the bit and
energy loading algorithm.
B. Experimental Setup
The experimental setup, shown in Fig. 7, starts with a laptop
connected to a Tektronix AWG (AWG70001A) that has a
maximum sampling frequency of 50 GS/s with an ADC res-
olution of 10 bits per sample. Bipolar OFDM waveforms
are generated in MATLAB as detailed in Section 3.A and then
transmitted to the AWG. The maximum peak-to-peak voltage
(V
PP
) of the AWG is 0.5 V
PP
. The output of the AWG is
amplified with a broadband amplifier (SHF 100AP) that has
a maximum gain of 20 dB in the bandwidth range
(100 kHz–20 GHz). A 3 dB attenuator is used at the output
of the amp lifier to allow flexible control of the signal modula-
tion depth, V
PP
. The power budget of the system is adjusted to
allow complete utilization of the micro-LED dynamic range
shown in Fig. 2.
The amplified bipolar signal is DC-biased with a Bias-tee
(Mini-Circuits ZFBT-4R2GW+). Low values for the DC bias
result in high zero-level clipping of the OFDM waveform,
which degrades the SNR. High values for the DC bias result
in optical power saturation at the micro-LED, which also de-
grades the SNR. After extensive experiments, the DC bias is set
to I
DC
30 mA corresponding to a measured DC voltage of
V
DC
5.23 V. This value allows the OFDM bipolar signal to
swing in the linear region of the L–V characteristic of the
micro-LED. The biased signal is then fed to the micro-LED
via a high-speed probe. An optical plano-convex lens (Thorlabs
LA1116) is used to collimate most of the light into a dielectric
mirror (Thorlabs CM1-E02) with higher than 97% reflectance
in the desired wavelength region. The reflected light is then
focused onto the photoreceiver by a bi-convex lens (Thorlabs
LB4879) followed by a microscopic objective lens (NewPort
M–40 × ) with a numerical aperture (NA) of 0.65. A silicon
positive–intrinsic–negative (PIN) photoreceiver is used (Femto
HSPR-X-I-1G4-SI) with a 3 dB bandwidth of 1.4 GHz and a
responsivity of 0.135 A/W around 400 nm.
4. RESULTS AND DISCUSSION
The VLC data transmission experiment was only conducted on
the inner pixels due to their higher E–O bandwidth compared
to the outer pixels. The sampling frequency of the AWG is set
(a)
0 100 200 300 400 500
0 100 200 300 400 500
Bits
0
2
4
6
8
10
Channel Capacity
(b)
Sub-carrier index
0
0.5
1
1.5
Fig. 5. (a) Bit loading and channel capacity per subcarrier, both
given in bits per subcarrier. (b) Energy loading per subcarrier.
10 15 20 25
SNR (dB)
0
0.2
0.4
0.6
0.8
1
CDF
128-QAM
32-QAM
8-QAM
64-QAM
16-QAM
4-QAM
Fig. 6. Statistical CDF for different QAM constellation sizes real-
ized at BER 2.3 × 10
−3
, below the FEC target.
20 dB Amplifier
SHF 100AP
3 dB
Attenuator
(a)
16 cm
11.5 cm
Microscope objective lens
Newport (M-40x)
SI-PIN Photoreceiver
FEMTO (HSPR-X-I-1G4-SI)
Violet Micro LED
Bi-Convex lens
Thorlabs (LB4879)
Plano-Convex lens
Thorlabs (LA1116)
Mirror
Thorlabs (CM1-E02)
(b)
Fig. 7. Experimental setup. (a) Schematic setup of the experiment
showing the optical system, AWG, oscilloscope, amplifier, attenuator,
and Bias-tee. (b) Photograph of the optical system showing the micro-
LED, the optical lens system, and the photoreceiver.
Research Article
Vol. 5, No. 2 / April 2017 / Photonics Research A39
