Asymmetrically Reconstructed Optical OFDM For Visible Light Communications

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2016.2520818, IEEE
Photonics Journal
IEEE Photonics Journal Asymmetrically Reconstructed Optical OFDM for VLC
Asymmetrically Reconstructed Optical OFDM for

Visible Light Communications
Jindan Xu, Wei Xu, Senior Member, IEEE , Hua Zhang, Member, IEEE ,
Xiaohu You, Fellow, IEEE
National Mobile Communications Research Laboratory, Southeast University,

Nanjing 210096, China
Manuscript received December 20, 2015; revised January 15, 2016; accepted January 19, 2016. This work was
supported by the 973 Program under 2013CB329203, and the National Natural Science Foundation of China
(NSFC) under Grant 61571118 and Grant 61471114. Corresponding author: W. Xu (e-mail: wxu@seu.edu.cn).
Abstract: In conventional optical orthogonal frequency division multiplexing (OOFDM) systems,

such as pulse amplitude modulated discrete multitone (PAM-DMT) and asymmetrically clipped optical
OFDM (ACO-OFDM), half of the time-domain transmitted signals carry meaningful information while
the other half are set at zeros after asymmetric clipping. By specifically exploiting the asymmetric
structure of OOFDM like PAM-DMT and ACO-OFDM, we present a new framework for implementing
OOFDM with enhanced performance in terms of both bit-error rate (BER) and peak-to-average power
ratio (PAPR). The proposed scheme introduces asymmetric signal reconstruction, instead of direct
clipping, in time domain before transmission. The designed asymmetric reconstruction refines the
time-domain statistics of the OOFDM signals by constructively exploiting zero-padded signal positions
in conventional OOFDMs. In this way, the PAPR of the reconstructed OOFDM signals is effectively
reduced. For receivers, in order to recover the received signals precisely, we further develop an optimal
maximum a posteriori (MAP) detection method, along with an efficient simplified one. Simulation
results validate that the proposed scheme enhances the BER performance compared to conventional
OOFDMs under both an ideal and a dispersive visible light communication (VLC) channel via our
prototype. Specifically, under an ideal AWGN channel, an approximately 4 dB gain is achieved at the
BER of 10−2 , along with a 2.5 dB lower PAPR constraint. While under the real dispersive channel,
our test shows about 6 dB gains in BER performance.
Index Terms: Visible light communications (VLC), orthogonal frequency division multiplexing (OFD-
M), pulse amplitude modulated discrete multitone (PAM-DMT), asymmetrically clipped optical OFDM
(ACO-OFDM), peak-to-average power ratio (PAPR) reduction.
1. Introduction
Recently, visible light communication (VLC), emerging as a promising technology, has gained
significant attention in the field of wireless transmission technologies [1]. Compared to conventional
radio frequency (RF) communication, VLC enjoys several unique advantages such as abundant
unlicensed optical spectrum and extremely high data rates. In VLC systems, intensity modulation
and direct detection (IM/DD) is commonly used where only the light intensity is modulated. The
transmitted signals therefore must be nonnegative as the light intensity can never be negative.
Traditional IM/DD schemes such as on-off keying (OOK) and pulse position modulation (PPM)
suffer the disadvantage of low power efficiency and low spectrum efficiency, respectively [2]. In
order to improve data rate and mitigate intersymbol interference (ISI) caused by a dispersive
channel, orthogonal frequency division multiplexing (OFDM) is employed in VLC [2], [3], which is
also known as discrete multitone (DMT) in wired communication like asymmetric digital subscriber
Vol. xx, No. xx, January 2016 Page 1
1943-0655 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2016.2520818, IEEE
Photonics Journal
line (ADSL). In OFDM IM/DD systems, a real signal is generated by forcing the input signals
Hermitian symmetric in frequency domain. In order to further make the signal nonnegative, a direct
current (DC) bias is added to make it unipolar, namely DC-biased optical OFDM (DCO-OFDM) [4].
Alternatively, another type of schemes, i.e., asymmetrically clipped optical OFDM (ACO-OFDM)
[5], [6] and pulse amplitude modulated discrete multitone (PAM-DMT) [7], directly clip the negative
values at zero. Although ACO-OFDM carries data on half subcarries of DCO-OFDM, it achieves a
higher power efficiency since no DC bias is required [8]. So does PAM-DMT. It has been revealed
that both PAM-DMT and ACO-OFDM achieve better performance than DCO-OFDM for a high
normalized bandwidth/bit-rate due to the absence of DC bias. Some modified schemes have been
proposed in [9], [10] to alleviate the negative effect of DC bias by combining DC in ACO-OFDM
schemes. In order to further enhance the spectral efficiency, several schemes were proposed to
utilize more frequency resource for data transmission. In [11], the entire bandwidth was divided
into several layers for simultaneous multi-stream transmission using ACO-OFDM strategy for data
modulation on each layer. A superimposed modulation technique was then introduced in [12] for
systems using PAM-DMT. The study in [13] further considered the incorporation of ACO-OFDM
and PAM-DMT under an additional concern of illumination.
However, few has explored the fact that there are several essential similarities between PAM-
DMT and ACO-OFDM [14]. Applying the properties of IFFT, both PAM-DMT and ACO-OFDM
generate the same magnitude values with opposite signs at a pair of time-domain positions. In
this way, the negative values are padded to zeros without any information loss. Although both
schemes achieve better power efficiency than DCO-OFDM, it is obvious that half of the channel
resource is unused. Moreover, all optical OFDM (OOFDM) schemes exhibit a very high peak-
to-average power ratio (PAPR) like most RF OFDM systems. Due to the nonlinearity of power
amplifier (PA), high PAPR could lead to signal distortion in RF systems. While in VLC systems,
the performance further degrades with a high PAPR due to the additional nonlinearity caused by
LED [15], [16].
In order to enhance the time efficiency of both ACO-OFDM and PAM-DMT, we present an
asymmetrically reconstructed OOFDM scheme with low PAPR and develop an efficient signal
detection method. Specifically, by taking the asymmetric structure of both OOFDM schemes, the
proposed scheme introduces a reconstruction operation, in the place of asymmetric clipping,
pairwisely on the optical signals before transmission. The additional operation refines the time-
domain statistics of the OOFDM signal. Besides validating the reduced PAPR statistics, we show
that this asymmetrically reconstructed scheme allows an efficient detection as well. We test the
proposed scheme under both an ideal and a measured dispersive VLC channel, and observe
noticeable BER gains compared to conventional schemes.
The rest of this paper is organized as follows. In Section 2, the concept of the conventional
OOFDM schemes, including PAM-DMT and ACO-OFDM, is introduced. An asymmetrically re-
constructed OOFDM scheme along with its implementation framework and detection method is
presented in Section 3. In Section 4 we compare both the PAPR and BER performance between
the proposed scheme and other various existing OOFDMs. Conclusions are drawn in Section 5.
2. Conventional asymmetrically clipped OOFDM schemes

In OOFDM IM/DD systems, transmitted signals in time domain are nonnegative. Generally, Her-
mitian symmetric frequency-domain data is used to provide real signals after IFFT. Two OOFDM
schemes have been popular, namely ACO-OFDM [5] and PAM-DMT [7]. Both schemes generate
pairs of time-domain signals with the same magnitude and opposite polarity. As a result, clipping
away the negative signals will not lose any useful information. We first briefly review the design
philosophy of both schemes.
Assume that there are totally N subcarriers. According to Hermitian symmetry, the data vector is
modulated on subcarriers in the form of X = [0, X1 , X2 , ..., X N −1 , 0, X ∗N −1 , ..., X2∗ , X1∗ ], where the
2 2
upperscript (·)∗ represents conjugate of a complex value. The time-domain signal x is generated
Photonics Journal
by taking the IFFT of X. It yields:
1 ∑
N −1 ( n)
x(k) = √ Xn exp j2πk , k = 0, 1, ..., N − 1 (1)
N n=0 N
where x(k)(k = 0, 1, ..., N − 1) is the kth component of x. Xn (n = 0, 1, ..., N − 1), i.e., the nth
component of X, is a complex-valued symbol, and x(k) is real since X is Hermitian symmetric.
The difference between ACO-OFDM and PAM-DMT lies in the data mapping on subcarriers.
In PAM-DMT, complex signals with only imaginary part are mapped on all subcarriers, while in
ACO-OFDM we only modulate on odd subcarriers but with complex signals having both imaginary
and real components.
In particular, PAM-DMT utilizes N2 − 1 subcarriers to carry useful but imaginary data. Firstly,
2 − 1 symbols, denoted as bn (n = 1, 2, ..., 2 − 1), drawn from the real one-dimensional con-
N N
stellation PAM, are used to modulate the imaginary data on each subcarrier. Therefore, the data
vector modulated on subcarriers is X = [0, jb1 , jb2 , ..., jb N −1 , 0, −jb N −1 , ..., −jb2 , −jb1 ]. After the
2 2
operation of IFFT like in (1), x(k) satisfies:
N
x(k) = −x(N − k), k = 1, 2, ..., − 1. (2)
2
Let us denote x b(k) = x(N − k). Hence, we get each pair of signals [x(k), x
b(k)] with the same
magnitude but opposite polarity.
For ACO-OFDM, it utilizes only the odd subcarriers to transmit symbols. As a result, on-
ly N4 subcarriers carry useful data. At first, N4 complex symbols are generated from a com-
plex two-dimensional constellation such as QAM: Sn (n = 0, 1, 2, ..., N4 − 1). It generates: X =
[0, X1 , 0, X3 , 0, ..., X N −1 , 0, X ∗N −1 , ..., 0, X3∗ , 0, X1∗ ], where
2 2
 N

 S n−1 , n is odd, and n <

 2 2
Xn = ∗ N (3)

 S N −n−1 , n is odd, and n >

 2 2
0, otherwise.
Applying IFFT to X, we have
( )
N N
x(k) = −x k + , k = 0, 1, 2, ..., −1 (4)
2 2
b(k)] has the same magnitude but opposite
which also indicates that each pair of signals [x(k), x
b(k) = x(k + N2 ), in the ACO-OFDM.
signs, that is x
Given (2) and (4), for both PAM-DMT and ACO-OFDM, the asymmetric clipping is taken as
follows: {
x(k), x(k) ≥ 0
xc (k) = (5)
0, otherwise
where xc (k) denotes the signal after asymmetric clipping. Now, we get each final signal pair
bc (k)] with one nonnegative value and the other zero.
[xc (k), x
3. Asymmetrically reconstructed OOFDM

In both PAM-DMT and ACO-OFDM systems, signal pairs, i.e., [xc (k), x bc (k)], are obtained after
IFFT and asymmetric clipping. In each pair, only one position carries meaningful data while
the other is zero. Obviously, half of the time-domain resource bears no data. In other words,
the asymmetric clipping wastes half the channel resource in order to make transmitted signals
nonnegative. We present a new framework of generating positive OOFDM signals with some
Photonics Journal
X0 x(0)
X1 x(1) x xr
Inputbit Hermitian Signal D/
Modulation S/P IFFT P/S CP LED
stream symmetry
reconstruction A
X N 1 x(N 1) Optical
channel
Y0 y (0)
Remove Y1 y (1) y yr
Outputbit Signal Remove A/
stream Demodulation P/S symmetrical FFT S/P PD
conjugate recovery CP D
YN 1 y (N 1)
Fig. 1. System block diagram (S/P: serial to parallel; P/S: parallel to serial; CP: cyclic prefix; D/A:
digital to analog; A/D: analog to digital; LED: light emitting diode; PD: photodiode).
asymmetric reconstructions, instead of a direct asymmetric clipping, in order to better utilize the
time-domain resource.
As mentioned in the above, in each signal pair [x(k), x b(k)], the two signals bear the same
magnitude information with opposite signs. It is natural to propose a reconstruction operation
in order to generate a new signal pair bearing the same information but with better statistical
characteristics for transmission. In this way, the new signals can better use the channel responses
for achieving diversity gain as well as low PAPR. The challenge of the proposed design is to admit
precise signal recovery at the receiver. We show that a proper asymmetric reconstruction design
achieves enhanced performance in terms of both PAPR and BER.
3.1. Framework of asymmetric reconstruction

In the following, we present the proposed asymmetrically reconstructed OOFDM in VLC systems,
as well as a corresponding efficient detection approach for the scheme. Fig. 1 shows the block
diagram of the transmitter and receiver for the proposed scheme. Compared with conventional
OOFDM schemes, such as PAM-DMT and ACO-OFDM, we replace the “Asymmetric clipping”
block with a special “Signal reconstruction” block at the transmitter. Correspondingly, a procedure
named “Signal recovery” is added at the receiver side. Next, we elaborate the two new blocks in
detail.
3.1.1. Signal reconstruction at transmitter

Under the proposed framework, the transmitted signal pairs are reconstructed from [x(k), x b(k)]
for k = 0, 1, 2, ..., N2 − 1 as shown in Fig. 2(a). In such a signal pair, two signals carry the same
meaningful information with opposite signs. For PAM-DMT, there is x b(k) = x(N −k), while for ACO-
OFDM, x b(k) = x(k + N2 ). Without loss of generality, we elaborate the reconstruction operation on
a specific signal pair, namely [x, x b], and the index k is omitted for brevity. In order to make full
use of the two signal positions, we try to take some kind of effective operation on the positive
signal and replace the negative one with a nonnegative value bearing the same information. Let
O(·) and F(·) denote two reconstruction operations. The pairwise reconstruction on [x, x b] can be
expressed as follows:
br ] = [O(x), F(x)]
[xr , x (6)
br ] denotes the pair of signal after reconstruction, as shown in Fig. 2(b). Without loss
where [xr , x
of generality, we assume x ≥ 0 while x b = −x ≤ 0. Specifically, as Fig. 2(c), asymmetric clipping in
conventional ACO-OFDM and PAM-DMT can be considered as a particular case of reconstruction,
where
O(x) = x (7)
Photonics Journal
x̂
(a)
Asymmetric Asymmetric
reconstruction direct clipping
xr = O( x) xˆr = F ( x) xc = x
xˆc = 0
(b) (c)
Fig. 2. (a) Original transmitted signal pair after IFFT. (b) Transmitted signal pair after asymmetric
direct clipping (ACO-OFDM/PAM-DMT). (c) Transmitted signal pair after asymmetric reconstruction
(proposed scheme).
and
F(x) = 0. (8)
It is obvious that this kind of reconstruction is inefficient since half of the signals are simply set at
zeros. In this study, we design more efficient operations O(·) and F(·) to reconstruct the signals
so long as the same information is kept in both positions.
Since F(·) bears meaningful information, some amount of diversity gain can be achieved at the
receiver. On the other hand, high PAPR is also a challenge in OFDM, especially in VLC because
of the nonlinearity of both PA and LED. In RF OFDM, there are multiple ways of alleviating this
issue. Nevertheless, these schemes generally reduce PAPR at the cost of clipped information loss
or additional overhead [17]. For OOFDM, the reconstruction process can fortunately be used to
refine the statistics of signals.
Now, the essential challenge lies in the design of reconstruction functions O(·) and F(·), which
should as well admit efficient signal recovery for detection. In order to enhance performance, we
take adaptive but efficient operations on [x, x b], which implies O(·) and F(·) could be piecewise
functions. As mentioned above, in order to achieve diversity gain, both O(x) and F(b x) are expected
to be a scaled version of x instead of zeros. Hence, we choose operations O(·) and F(·) as scaling
functions of the absolute value of x. Note that we use x to represent the absolute value in the
b] for notational brevity. A threshold ηc is defined in order to define the reconstruction
signal pair [x, x
operations, like in [18]. When x ≤ ηc , asymmetric clipping is taken to reconstruct the signals.
1
Without loss of generality, we define the common clipping ratio (CR) as 20log(ηc /(E{|xc (k)|2 }) 2 ),
where E{·} denotes the expectation. In summary, the reconstructed functions O(·) and F(·) can
be mathematically defined as:


x, 0 ≤ x ≤ ηc
O(x) = α1 x, ηc < x ≤ αηc1 (9)

 ηc
ηc , x > α1
Photonics Journal
and 

0, 0 ≤ x ≤ ηc
F(x) = α2 x, ηc < x ≤ αηc1 (10)

 α2
α1 ηc , x > αηc1
where α1 and α2 denote two scaling factors.
Obviously, this new operation could make it complicate for signal detection at the receiver for
arbitrary α1 and α2 . The two received signals in one pair both carry nonnegative values. As a
result, it is difficult to find out the original positive or negative positions. Therefore, we should
restrict α1 ̸= α2 for diversity gain in detection. It is safe to assume 0 < α2 < α1 < 1. The
values of α1 and α2 have an important effect on the performance of this scheme. Smaller the
values are, lower the average power will be, which results in higher PAPR. On the other hand,
the detection performance at receiver is also affected by these two values. In order to obtain the
best performance, we can solve this optimization problem via heuristic search. However, favorable
performance can be achieved by simply setting α1 = 12 and α2 = 14 . As it is assumed x ≥ 0 in the
signal pair [x, xb], the reconstructed OOFDM signal becomes


 0 ≤ x ≤ ηc
[x, 0] ,
br ] = [O(x), F(x)] = [α
[xr , x [ 1 x, α2 x]] , ηc < x ≤ αηc1 (11)


 η c , α2 η c , x > α1ηc
α1
which shows that both xr and x br carry useful information for a large signal x, where diversity gain
can be reaped. Meanwhile, the asymmetric reconstruction decreases the peak value of signals
effectively, resulting in a significant PAPR reduction.
It is interesting to note that the conventional ACO-OFDM [5] and PAM-DMT [7], and the pre-
viously developed RoC-ACO-OFDM [18], can be regarded as special cases of the proposed
scheme with direct clipping operation and additive clipping operation, respectively. The asymmetric
reconstruction with scaling functions carries two copies of meaningful information in each signal
pair, while additive clipping in RoC-ACO-OFDM carries only one. As a result, the proposed
scheme enjoys better BER performance. Although the conventional direct clipping operation looks
more simple than the proposed asymmetric reconstruction, the proposed scheme achieves better
performance, including a larger PAPR reduction and better BER performance with comparable
detection complexity, as validated by numerical simulation results in Section 4. Before that, we
develop the signal recovery and detection of the proposed asymmetrically reconstructed OOFDM
scheme in the following.
3.1.2. Signal recovery at receiver

For conventional OOFDM schemes, such as PAM-DMT and ACO-OFDM, the detection at receiver
is simple and efficient via FFT. Regarding signal detection of the proposed asymmetrically recon-
structed OOFDM, it follows a similar procedure, except for the “Signal recovery” part, as illustrated
in Fig. 1. This procedure tries to recover the reconstructed signal pairs, which technically includes
two steps named “mode detection” and “inverse operation”.
At the receiver, the received optical intensity is converted into a digital electrical signal r(k)
through an optical detector and an analog-to-digital (A/D) converter. Considering the shot noise
and thermal noise, which are usually modeled as additive white Gaussian noise (AWGN), we have
r(k) = hxr (k) + w(k) (12)
where h denotes the channel gain and w(k) is the discrete time version of AWGN. The index k
can be omitted for brevity. Let us assume [yr , ybr ] as the sample pair of received signals. From
Photonics Journal
(12), it follows: {
yr = xr + z
(13)
ybr = x
br + zb
where z = w/h, zb = w/h, b yr = r/h, ybr = rb/h given h is generally known via channel estimation.
In ACO-OFDM and PAM-DMT, the sampled time-domain signal pairs have the form of [xc , 0] or
bc ]. As a result, it is simple to distinguish between the positive and negative positions. In our
[0, x
proposed scheme, it appears nontrivial to detection the signals due to the reconstruction in (11).
To facilitate the OFDM demodulation, we first determine that the received signal pair belongs to
one of the three cases in (11), which is referred to as “mode detection” in the following. Given
the result of “mode detection”, the signal pair can be recovered via inverse operations according
to (11), called “inverse operation”. In the following, we first present an optimal mode detection
method in terms of maximum a posteriori (MAP) detection. Although complicated, the MAP
detection achieves the benchmark performance of the proposed OOFDM scheme. An efficient
mode detection method is latter developed, having comparable complexity with conventional
OOFDMs. We will show by numerical methods that the efficient detection approach achieves
almost the same BER performance as the optimal MAP one.
According to the philosophy of MAP detection, the major task of “mode detection” is to calculate
the probability of each hypothesis conditioned on received signal pairs. Specifically, consider the
reconstructions operation defined in (9) and (10). Since the receiver does not know which position
of the signal pair [x, x b] is positive, the corresponding transmitted signal belongs to either of the
following six hypotheses, defined as:
br ] = [x, 0],
H1 : [xr , x br ] = [0, x],
H2 : [xr , x 0 ≤ x ≤ ηc
ηc
br ] = [α1 x, α2 x],
H3 : [xr , x br ] = [α2 x, α1 x], ηc < x ≤
H4 : [xr , x
[ ] [ ] α 1 (14)
α2 α2 ηc
br ] = ηc , ηc ,
H5 : [xr , x br ] =
H6 : [xr , x ηc , ηc , x >
α1 α1 α1
where x denotes the nonnegative value in signal pair [x, x b].
Given the received signal pair [yr , ybr ], the mode detection in (14) can be solved by using a
typical MAP detection. The most possible transmitted hypothesis is estimated as:
b = arg
H max PH|[yr ,byr ] {Hi |(y1 , y2 )} (15)
Hi Hi ,i=1,2,...,6
where PH|[yr ,byr ] {·} denotes the probability of hypothesis H conditioned on the received signal
pair [yr , ybr ]. For brevity, the detailed calculations of H b are presented in Appendix. After obtaining
b
H, the received signal pair is recovered according to

 yr , b
H = H1



 b

 −byr , H = H2

 β y + 1−β yb , b
r r H = H3
y = α11−β α2 β b (16)

 − y − br ,
α1 y H = H4

 α2 r

 ηc b
 α1 ,

H = H5
 ηc b
− α1 , H = H6
and
yb = −y (17)
which is determined according to the asymmetric reconstruction defined in (9) and (10). The
coefficient β in (16) is generated from the maximum ratio combining (MRC) principle in order to
obtain the largest diversity gain [19]. Specifically, β is 0.8 when α1 = 21 and α2 = 14 .
Photonics Journal
(a)
(b)
Fig. 3. Decision areas for H b with CR=10 dB and SNR=20 dB. (a) Optimal MAP detection, (b)
Simplified linear detection.
3.2. Linear detection

Although the MAP detection method demonstrated in subsubsection 3.1.2 is optimal, it needs
involved calculations of conditioned probabilities. In order to reduce the complexity, we develop
a simplified detection method in this subsection. To be more specific, we consider the proposed
asymmetrically reconstructed scheme with scaling functions where α1 = 21 and α2 = 14 as in (14).
b with CR = 10 dB and SNR = 20 dB, which is obtained
Fig. 3(a) exemplifies decision areas for H
by the means of the above-mentioned optimal MAP detection. The average power of time-domain
signals after IFFT is normalized. It is obvious that the MAP classification region is nonlinear. In
order to simplify the detection method, we try to present linear classifications via some asymptotic
analysis of the detection problem.
Due to symmetry, the classification between hypotheses H1 , H3 , H5 and hypotheses H2 , H4 ,
H6 is directly obtained via the linear classifier y1 = y2 , as exemplified in Fig. 3(a). Let us assume
Photonics Journal
α1 = 21 and α2 = 14 . If it is H3 , the signal is recovered as y = 1.6yr + 0.8b yr while if it is H5 , the

signal is recovered as y = 2ηc , according to (16). Since the transmitted signal pair is [ηc , 12 ηc ] for
H5 , we have 1.6yr + 0.8b yr ≈ 2ηc in this condition. Thus, detection result shall not be significantly
affected if we merge hypothesis H5 and H3 . This is also true for merging hypothesis H6 and H4 .
In this way, we can now focus on further simplifying detection problem among the four hypothe-
ses namely H1 , H2 , H3 and H4 . Considering a relatively high SNR case, the received signal pair
[yr , ybr ] applies the linear classifier y2 = 0 and 0 ≤ y1 ≤ ηc when it is H1 , while it applies y2 = 12 y1
and 12 ηc < y1 ≤ ηc when it is H3 , according to (14). Due to symmetry, the linear classifier for H2
is y1 = 0 and 0 ≤ y2 ≤ ηc , and for H4 the classifier is y1 = 12 y2 and 21 ηc < y2 ≤ ηc . The decision
areas are these four line segments in this ideal case. By taking noise into account, the lines
turn into regions as shown in Fig. 3(b). In summary, we have four approximate linear classifiers
including y1 = y2 , y1 + y2 = 12 ηc , y2 = 14 y1 and y1 = 14 y2 .
Now that, given each pair of received signal [yr , ybr ], we can accordingly conduct the hypothesis
detection using the following linearly simple rules:
( )
1 1
H1 : yr ≥ ybr and yr + ybr < ηc or yr ≥ ybr ,
2 4
( )
1 1
H2 : yr < ybr and yr + ybr < ηc or yr < ybr ,
2 4
( ) (18)
1 1
H3 : yr ≥ ybr and yr + ybr ≥ ηc and yr < ybr ,
2 4
( )
1 1
H4 : yr < ybr and yr + ybr ≥ ηc and yr ≥ ybr .
2 4
According to the above rules, it shows that, for each signal pair, only two multiplications, one
addition and three comparison computations are needed, in order to determine which hypothesis
it is. This is the major difference at receiver between our proposed scheme and the conventional
ACO-OFDM and PAM-DMT. Therefore, we can conclude that our proposed scheme requires
comparable complexity with a marginal increment compared to the conventional ones, while it
achieves promising performance gains in terms of both PAPR and BER as will be verified in
Section 4.
4. Numerical Test Results

In this section, we compare the performance of the proposed asymmetrically reconstructed OOFD-
M to conventional OOFDM schemes. Not only the PAPR but also the BER performance are
compared with both PAM-DMT and ACO-OFDM.
As for the proposed asymmetrically reconstructed scheme, we set α1 = 1/2 and α2 = 1/4. Ideal
PAM-DMT and ideal ACO-OFDM are two schemes introduced in Section 2, which theoretically
generate an ideal lower bound to BER performance, respectively. The term “ideal” implies that
no signal distortion due to PA and LED nonlinearity is considered whatever the PAPR is. While
for “conventional” ACO-OFDM and PAM-DMT, we take the nonlinearity into account. Note that in
most applications, the common operation to deal with the nonlinearity is direct upper clipping. As
for detection, we use the theoretically optimal receiver, i.e. maximum likelihood detection [6], for
conventional schemes.
For simulation verification, we first compare the BER performance of the above schemes under
an ideal AWGN channel, where a commonly used Line-of-Sight (LoS) VLC channel is considered.
Subsequently, in order to verify the proposed scheme under more realistic scenarios, we further
present BER performance comparison under a dispersive VLC channel from real measurements.
We measure the channel responses of the VLC system using our constructed prototype and use
the data for simulation. The frequency response of this dispersive channel is also exemplified in
Fig. 4. Detailed simulation parameters are listed in Table I. We note that the measured channel
Photonics Journal
10
Amplitude (dB)
−5
−10
−15
−20
0 10 20 30 40 50 60 70 80 90 100
Frequency (MHz)
Fig. 4. Frequency response of a real VLC channel.
TABLE I
S YSTEM S IMULATION PARAMETERS
Parameters Values
Sample rate 200 MHz
IFFT size 256
CP length 24
Modulation 64-QAM
Transmission distance 3m
Exemplified in Fig. 4
Measured from our testbed with:
VLC channels LED: OSTAR Lighting Plus LE UM S2LN
PA: ZHL-3A+ from Mini-Circuits @25 dB
PD: C5331-11 from Hamamatsu
under test includes the effects of both channel dispersion, namely frequency selectiveness/fading,
and the distortions and nonlinearity properties in the electric domain. We believe that simulations
under the real measured VLC channel reflects multiple fading and distortion effects.
4.1. Comparison between the proposed scheme and PAM-DMT

Fig. 5 and Fig. 6 compare the performance of the proposed asymmetrically reconstructed scheme
and conventional PAM-DMT. We test two different cases. For example, in case 1, we choose CR
= 8 dB and 12 dB for the proposed scheme and conventional PAM-DMT, respectively. In order
to highlight the performance of the proposed scheme, we can see in this case the proposed
scheme exhibits lower PAPR as shown in Fig. 5, as well as better BER performance as observed
in Fig. 6, which is tested under an ideal LoS VLC channel. Similarly, the CRs of the proposed and
conventional schemes are set as CR = 10 dB and 13 dB in case 2. In each case, it also shows
that the proposed scheme outperforms the conventional PAM-DMT, in terms of both PAPR and
BER, as shown in Fig. 5 and Fig. 6. As for a particular scheme, a lower CR provides a higher
PAPR reduction at the cost of worse BER. More specifically, from the definition of CR, a lower
CR implies a smaller peak value of the signal, that is a lower PAPR given the average power of
the signal is fixed. On the other hand, in order to achieve a low CR, signal clipping is inevitable
Photonics Journal
which causes BER performance degradation. In other words, there is a tradeoff between PAPR
reduction and BER performance.
It is observed that when SNR is higher than 30 dB in Fig. 6, the BERs exhibit saturation trend.
In VLC systems, the system performance is dominated by the nonlinearity of both the power
amplifier and LEDs. Since the signal distortion becomes worse with an increasing SNR, the BER
performance saturates when SNR grows high or even without the presence of noise. This is
typically named as “error floor” of BER performance in a communications system. Moreover, it
is worth mentioning that the simplified efficient detection method achieves almost the same BER
performance as the optimal MAP one. In Fig. 7, similar observations can be concluded over a real
dispersive VLC channel. According to case 2 in Fig. 5-6, an approximately 4 dB gain is achieved
at the BER of 10−2 , along with a 2.5 dB lower PAPR constraint, under an ideal AWGN channel.
While under the real dispersive channel, Fig. 7 shows about 6 dB BER gains.
0
10
Ideal PAM−DMT
Conventional PAM−DMT (case 1)
Proposed scheme (case 1)
CCDF(Probability(PAPR>PAPR0))
−1
10 Proposed scheme (case 2)
−2
10
−3
10
−4
10
8 10 12 14 16 18
PAPR0(dB)
Fig. 5. PAPR comparison between asymmetrically reconstructed scheme and conventional PAM-DMT (64-QAM).
0
10
Ideal PAM−DMT
Conventional PAM−DMT
Proposed scheme w. simplified detection
Proposed scheme w. MAP detection
−1
10
BER
case 1
−2
10
case 2
−3
10
5 10 15 20 25 30 35
Eb,elec/N0(dB)
Fig. 6. BER comparison between asymmetrically reconstructed scheme and conventional PAM-DMT
under an ideal AWGN channel (64-QAM).
Photonics Journal
0
10
Ideal PAM−DMT
−1
10
BER
case 1
−2
10
case 2
−3
10
10 15 20 25 30 35 40 45
Eb,elec/N0(dB)
under a real dispersive VLC channel (64-QAM).
4.2. Comparison between the proposed scheme and ACO-OFDM

Comparison between the proposed asymmetrically reconstructed scheme and conventional ACO-
OFDM is displayed in Fig. 8 and Fig. 9. There are two cases. The CRs of the proposed and
conventional schemes are (5.5 dB, 10 dB) in case 3 and (7.5 dB, 11 dB) in case 4, respectively.
It is shown that in each case, the proposed scheme always outperforms conventional ACO-
OFDM, both in terms of PAPR reduction and BER performance with a high SNR under an ideal
AWGN channel. Similarly, there is a tradeoff between PAPR reduction and BER performance. The
simplified detection method achieves almost the same BER performance as the optimized MAP
one. Similar BER performance comparisons under a real dispersive VLC channel are obversed
in Fig. 10.
0
10
Ideal ACO−OFDM
Conventional ACO−OFDM (case 3)
−1
−2
10
−3
10
−4
10
5 10 15 20
PAPR0(dB)
Fig. 8. PAPR comparison between asymmetrically reconstructed scheme and conventional ACO-OFDM (64-QAM).
Photonics Journal
0
10
Ideal ACO−OFDM
Conventional ACO−OFDM
−1
10
case 3
BER
−2
10
case 4
−3
10
−4
10
5 10 15 20
Eb,elec/N0(dB)
Fig. 9. BER comparison between asymmetrically reconstructed scheme and conventional ACO-OFDM
0
10
−1
10
case 3
BER
−2
10
case 4
−3
10
Ideal ACO−OFDM
Conventional ACO−OFDM
−4
10
10 12 14 16 18 20 22 24 26 28
E /N (dB)
b,elec 0
Fig. 10. BER comparison between asymmetrically reconstructed scheme and conventional ACO-
OFDM under a real dispersive VLC channel (64-QAM).
4.3. Comparison results using low-order modulations

According to the above simulation results using 64-QAM, the SNR requirement becomes high
for the system to achieve satisfactory performance for a low PAPR constraint, or equivalently a
smaller CR setup. In fact, one would resort to using low-order modulations under these cases
to achieve satisfactory BER. Therefore, we here further present comparison results using lower
order modulations, like 16-QAM. Figs. 11-12 and 13-14 show the comparison results in PAM-DMT
and ACO-OFDM systems, respectively. Similar observations can be concluded as in the previous
subsection using 64-QAM, except for some trivial differences in, e.g., SNR and CR values under
test. The CRs of the proposed and conventional schemes are (8 dB, 12 dB) in case 5, (10 dB,
13 dB) in case 6, (5.5 dB, 8 dB) in case 7 and (7.5 dB, 9 dB) in case 8, respectively.
Finally it would be necessary to note that our proposed scheme is effective in reducing the PAPR
of an ACO-OFDM/PAM-DMT signal while independent of the specific behavior of the nonlinearity
Photonics Journal
function. Since the laser diode (LD) exhibits higher nonlinearity than the LED, it is believed that the
proposed scheme is still applicable in the LD based VLC system, such as the directly modulated
OFDM scheme in [20]. Specifically, we can refer to the performance comparison in Figs. 11-14
under smaller clipping ratio (CR) setups which correspond to severe clipping distortions like in LD
based VLC systems.
0
10
Ideal PAM−DMT
−1
−2
10
−3
10
−4
10
8 10 12 14 16 18 20
PAPR0(dB)
Fig. 11. PAPR comparison between asymmetrically reconstructed scheme and conventional PAM-DMT (16-QAM).
−1
10
−2 case 5
10
BER
−3 case 6
10
Ideal PAM−DMT
−4
10
12 14 16 18 20 22
Eb,elec/N0(dB)
5. Conclusion
In this paper, we propose an asymmetrically reconstructed OOFDM scheme for VLC. A reconstruc-
tion procedure is added at transmitter in the place of asymmetric clipping while a signal recovery
procedure is added at receiver. Negative signals generated from IFFT carry meaningful nonnega-
tive values after reconstruction instead of simply being padded at zeros. As a result, diversity gain
Photonics Journal
0
10
Ideal ACO−OFDM
−1
−2
10
−3
10
−4
10
6 8 10 12 14 16 18
PAPR0(dB)
Fig. 13. PAPR comparison between asymmetrically reconstructed scheme and conventional ACO-OFDM (16-QAM).
Ideal ACO−OFDM
−1 Conventional ACO−OFDM
10
−2
10 case 7
BER
−3
10
case 8
−4
10
6 8 10 12 14 16
E /N (dB)
b,elec 0
Fig. 14. BER comparison between asymmetrically reconstructed scheme and conventional ACO-
OFDM under an ideal AWGN channel (16-QAM).
is achieved, and the statistical property of time-domain signals is also refined. Furthermore, an
optimal MAP detection method, along with an efficient simplified one, is developed to recover the
received signals. Simulation results indicate that this reconstructed OOFDM achieves a significant
PAPR reduction and a noticeable BER performance gain compared to the conventional ones,
although their complexity are comparable. Specifically, an approximately 4 dB gain is achieved at
the BER of 10−2 , in addition to a reduction of 2.5 dB in PAPR, under an ideal AWGN channel.
While under the real dispersive channel, our test shows about 6 dB BER gain.
Appendix
In this appendix, we calculate the MAP mode detection of (15). In order to apply MAP detection, the
probabilities of hypothesis H conditioned on received signal pair [yr , ybr ], denoted as PH|[yr ,byr ] {·},
Photonics Journal
is needed. From the Bayes theorem, it follows:

PH {Hi } × f[yr ,byr ]|H (y1 , y2 |Hi )
PH|[yr ,byr ] {Hi |(y1 , y2 )} = (19)
f[yr ,byr ] (y1 , y2 )
where f[yr ,byr ] (·, ·) and f[yr ,byr ]|H (·, ·) denote the joint probability distribution function (PDF) of [yr , ybr ]
and the conditioned joint PDF on H, respectively. PH {·} denotes the probability of hypothesis
H. Obviously, the denominator of (19), f[yr ,byr ] (y1 , y2 ), is the same for all hypotheses Hi (i =
1, 2, 3, 4, 5, 6) given the received pair [yr , ybr ]. Consequently, the MAP detection in (15) is equivalent
to:
Hb = arg max PH {Hi } × f[yr ,byr ]|H (y1 , y2 |Hi ). (20)
Hi Hi ,i=1,2,...,6
The remaining task is to calculate PH {·} and f[yr ,byr ]|H (·, ·) for all Hi , i = 1, 2, ..., 6. From [18], for
a large number of subcarriers, the original transmitted signal x follows Gaussian distribution due
to the central limit theorem. It gives
−x2
1
fx (x) = √ e 2σx2 (21)
2πσx
where σx2 is the variance of x, given by
∫ ∞
σx2 = E{|x|2 } = x2 fx (x)dx (22)
−∞
where E{·} denotes the expectation and σx2 represents the transmit signal power.
Concerning the probability of [yr , ybr ] conditioned on H, we have
∫ ∞
f[yr ,byr ]|H (y1 , y2 |H1 ) = f[yr ,byr ]|H,x (y1 , y2 |H1 , x)fx|H (x|H1 )dx
−∞
∫ ηc
1
= f[yr ,byr ]|H,x (y1 , y2 |H1 , x)fx (x)dx
P (H )
∫ H∞ 1 0
−∞
∫ αηc (23)
1 1
PH (H3 ) ηc
∫ ∞
−∞
∫ ∞
1
PH (H5 ) αηc
1
where f[yr ,byr ]|H,x (·, ·) denotes the joint PDF of [yr , ybr ] conditioned on H and a given x, and fx|H (·)
denotes the PDF of x conditioned on H. The same probabilities conditioned on H2 , H4 and H6
can be similarly obtained via symmetry.
From (13) and (14), the received signal pair equal


 [x + z, zb] , x ≤ ηc


 [α x + z, α x + zb] , ηc < x ≤
ηc
1 2
[yr , ybr ] = [ ] α1 (24)



 α 2 η c
 ηc + z, ηc + zb , x>
α1 α1
Photonics Journal
where z is additive white Gaussian noise (AWGN) with zero mean and variance σz2 , that is
−z 2
1
fz (z) = √ e 2σz2 . (25)
2πσz
According to (24) and (25), it further gives
−y2 2 2
−(y1 −x)
1 1
f[yr ,byr ]|H,x (y1 , y2 |H1 , x) = √ e 2σz2 √ e 2σz2
2πσz 2πσz
−(y1 −α1 x)2 −(y2 −α2 x)2
1 1
f[yr ,byr ]|H,x (y1 , y2 |H3 , x) = √ e 2σz2
√ e 2σz2
(26)
2πσz 2πσz
α 2
−(y2 − 2 ηc )
−(y1 −ηc )2 α1
1 1
f[yr ,byr ]|H,x (y1 , y2 |H5 , x) = √ e 2σz2
√ e 2σz 2
.
2πσz 2πσz
Next, by substituting (21) and (26) into (23), the conditioned joint PDF of f[yr ,byr ]|H (·, ·) can be
finally calculated as
( 2 2 )
1 1 y1 σz + y22 (σx2 + σz2 )
f[yr ,byr ]|H (y1 , y2 |H1 ) = √ exp −
PH (H1 ) 2πσz σx2 +σz2 2σz2 (σx2 + σz2 )
  σ2 y   σ2 y

−σ2x+σ12 ηc − σ2x+σ12
× Q  x z  −Q  x z 
σ1 σ1
( 2 2 )
1 1 −σz (y1 +y22 )−σx2 (α2 y1 −α1 y2 )2
f[yr ,byr ]|H (y1 , y2 |H3 ) = √ exp
PH (H3 ) 2πσz (α21 +α22 )σx2 +σz2 2σz2 [(α12 +α22 )σx2 +σz2 ]
  2
σx (α1 y1+α2 y2 )
  2
σx (α1 y1+α2 y2 )

ηc − (α 2+α2 )σ 2+σ 2
ηc
α1
− (α 2+α2 )σ 2+σ 2
× Q  1 2 x z  −Q  1 2 x z 
σ2 σ2
 ( )2  ( )
1 1  (y 1 − ηc )2
+ y 2 − α2
α1 c
η 
ηc
α1
f[yr ,byr ]|H (y1 , y2 |H5 ) = exp − Q
PH (H5 ) 2πσz2 2σz2 σx
(27)
√ √ ∫ ∞ 1 − x2
2 2 2 2 2 2
where σ1 = σx σz / σx + σz , σ2 = σx σz / (α1 + α2 )σx + σz , and Q(x) = x √ e 2 dx is the
2π
tail probability of the standard Gaussian distribution.
Thus far, by substituting (27) into (20), we can obtain the MAP detection result H. b Note that
the values of PH (Hi )(i = 1, 2, 3, 4, 5, 6) need not be calculated since they will be counteracted in
the final (20).
References
[1] T. Komine and M. Nakagawa, “Fundamental analysis for visible-light communication system using LED lights,” IEEE
Trans. Consumer Electron., vol. 50, no. 1, pp. 100–107, Feb. 2004.
[2] J. Armstrong, “OFDM for optical communications,” J. Lightwave Technol., vol. 27, no. 3, pp. 189–204, Feb. 2009.
[3] L. Chen, B. Krongold, and J. Evans, “Performance analysis for optical OFDM transmission in short-range IM/DD
systems,” J. Lightwave Technol., vol. 30, no. 7, pp. 974–983, Apr. 2012.
[4] M. Zhang and Z. Zhang, “An optimum DC-biasing for DCO-OFDM system,” IEEE Commun. Lett., vol. 18, no. 8, pp.
1351–1345, Aug. 2014.
[5] J. Armstrong and A. J. Lowery, “Power efficient optical OFDM,” Electron. Lett., vol. 42, no. 6, pp. 370–372, Mar. 2006.
[6] K. Asadzadeh, A. Dabbo, and S. Hranilovic, “Receiver design for asymmetrically clipped optical OFDM,” in Proc.
IEEE Globecom, Houston, USA, Dec. 2011, pp. 777–781.
[7] S. C. J. Lee, S. Randel, F. Breyer, and A. M. J. Koonen, “PAM-DMT for intensity-modulated and direct-detection
optical communication systems,” IEEE Photon. Technol. Lett., vol. 21, no. 23, pp. 1749–1751, Dec. 2009.
[8] J. Armstrong and B. J. C. Schmidt, “Comparison of asymmetrically clipped optical OFDM and DC-biased optical
OFDM in AWGN,” IEEE Commun. Lett., vol. 12, no. 5, pp. 343–345, May 2008.
[9] S. D. Dissanayake and J. Armstrong, “Comparison of ACO-OFDM, DCO-OFDM and ADO-OFDM in IM/DD systems,”
J. Lightwave Technol., vol. 31, no. 7, pp. 1063–1072, Apr. 2013.
Photonics Journal
[10] S. D. Dissanayake and J. Armstrong, “Novel techniques for combating DC offset in diversity combined ACO-OFDM,”
IEEE Commun. Lett., vol. 15, no. 11, pp. 1237–1239, Nov. 2011.
[11] Q. Wang, C. Qian, X. Guo, Z. Wang, D. G. Cunningham, and I. H. White, “Layered ACO-OFDM for intensity-modulated
direct-detection optical wireless transmission,” Opt. Exp., vol. 23, no. 9, pp. 12382–12393, May 2015.
[12] M. S. Islim, D. Tsonev and H. Haas, “Spectrally enhanced PAM-DMT for IM/DD optical wireless communications
systems,” in Proc. IEEE PIMRC, Hong Kong, China, Aug. 2015, pp. 877–882.
[13] Q. Wang, Z. Wang, and L. Dai, “Asymmetrical hybrid optical OFDM for visible light communications with dimming
control,” IEEE Photon. Technol. Lett., vol. 27, no. 9, pp. 974–977, May 2015.
[14] D. J. F. Barros, S. K. Wilson and J. M. Kahn, “Comparison of orthogonal frequency-division multiplexing and pulse-
amplitude modulation in indoor optical wireless links,” IEEE Trans. Commun., vol. 60, no. 1, pp. 153–163, Jan. 2012.
[15] W. O. Popoola, Z. Ghassemlooy, and B. G. Stewart, “Pilot-assisted PAPR reduction technique for optical OFDM
communication systems,” J. Lightwave Technol., vol. 32, no. 7, pp. 1374–1382, Apr. 2014.
[16] H. Qian, S. Cai, S. Yao, T. Zhou, Y. Yang, and X. Wang, “On the benefit of DMT modulation in nonlinear VLC systems,”
Opt. Exp., vol. 23, no. 3, pp. 2618–2632, Feb. 2015.
[17] S. Han and J. Lee, “An overview of peak-to-average power ratio reduction techniques for multicarrier transmission,”
IEEE Wireless Commun., vol. 12, no. 2, pp. 56–65, Apr. 2005.
[18] W. Xu, M. Wu, H. Zhang, X. You, and C. Zhao, “ACO-OFDM-specified recoverable upper clipping with effidient
detection for optical wireless communications,” IEEE Photon. J., vol. 6, no. 5, pp. 1–17, Oct. 2014.
[19] D. Tse and P. Viswanath. Fundamentals of Wireless Communications. New York: Cambridge University Press, 2005.
[20] Y. Chi, D. Hsieh, C. Tsai, H. Chen, H. Kuo, and G. Lin, “450-nm GaN laser diode enables high-speed visible light
communication with 9-Gbps QAM-OFDM,” Opt. Exp., vol. 23, no. 10, pp. 13051–13059, May 2015.

Asymmetrically Reconstructed Optical OFDM For Visible Light Communications

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Asymmetrically Reconstructed Optical OFDM For Visible Light Communications

Hochgeladen von

Copyright:

Verfügbare Formate

This article has been accepted for publication in a future issue of this journal, but has not been

IEEE Photonics Journal Asymmetrically Reconstructed Optical OFDM for VLC