R-OFDM Transmission Scheme for Visible Light

Communication using RGBA-LED

Lei Kong, Wei Xu, Hua Zhang, and Chunming Zhao
National Mobile Communications Research Lab., Southeast University, Nanjing 210096, China
Email: {konglei, wxu, huazhang, cmzhao}@seu.edu.cn

Abstract—White light-emitting diode (LED) consisting of red, natural multiplexing system, which can support high data rate
green, blue, and amber chips (RGBA-LED) has recently been transmission [8], [9]. Specifically, a RGB-LED based VLC
adopted as transmitter in visible light communication (VLC) system with a data rate of 3.4 Gb/s was experimentally realized
systems. This paper proposed a reshaped orthogonal frequen-
cy division multiplexing (R-OFDM) scheme for RGBA-LED- by exploiting an optimized discrete multitone modulation in
based VLC systems. In the R-OFDM, the signal is adjusted [8]. Moreover, a RGB-LED based bi-directional VLC system
by separation and biasing after clipping (BAC) operations for was demonstrated by using DCO-OFDM in [9].
transmitting in RGBA-LED. Then, the biasing factor of BAC However, all the above multi-color-based VLC systems
is derived according to the color mix ratio (CMR) of RGBA-
directly employed the existing OFDM techniques without
LED. At the receiver, We develop a direct detection algorithm
to recover the original data, and analyze its the theoretical bit specifically considering the illumination requirement, where
error rate (BER). Furthermore, a lower bound of the BER is white light or colored light should be produced by mixing
analyzed under high clipping ratio (CR) by using SNR upper different colors together according to a certain color mix ratio
bound. Finally, simulation results confirm our theoretical analysis (CMR). Furthermore, the effect of multi-color interaction, also
and verify that R-OFDM outperforms the ACO-OFDM in term
referred to as cross-talk [10], was not considered. In this paper,
of BER.
considering the illumination requirement of RGBA-LED, a
Index Terms—RGBA light-emitting diode, OFDM, color mix reshaped OFDM (R-OFDM) which has the same bandwidth
ratio, bit error rate, visible light communication
efficiency as ACO-OFDM was proposed for the RGBA-LED-
based VLC system. In this scheme, signal separation and bias-
I. I NTRODUCTION
ing after clipping (BAC) are performed for signal transmission
Visible light communication (VLC) has emerged as a in RGBA-LED. Then, the biasing factor of BAC is derived
promising technique for indoor optical wireless communica- according to the CMR of RGBA-LED. Accordingly, a direct
tion [1], [2]. To meet the high data rate requirement of the detection is developed for recovering the original signal at the
VLC, orthogonal frequency division multiplexing (OFDM) has receiver. Furthermore, the effective signal-to-noise ratio (SNR)
been introduced due to its higher bandwidth efficiency and and its upper bound under high clipping level are conducted
immunity to inter-symbol interference (ISI) [3]–[6]. in order to derive the theoretical bit error rate (BER) and its
In the popular intensity modulation/direct detection lower bound for the R-OFDM, respectively. Finally, Monte-
(IM/DD) VLC system, the transmitted signal must be real Carlo simulation is presented for verifying the correctness of
and non-negative. Therefore, the traditional OFDM should be theoretical analysis and validating the BER performance of
adjusted for facilitating transmission over optical channel. In the R-OFDM. It is revealed that the R-OFDM outperforms
[4], direct current-offset OFDM (DCO-OFDM) was proposed ACO-OFDM in term of BER.
by adding a DC bias to the bipolar OFDM signal. However,
the DC bias reduces the power efficiency. To remedy this, II. S YSTEM D ESCRIPTION AND LED C HARACTERISTIC
asymmetrically clipped optical OFDM (ACO-OFDM) was
proposed by using odd subcarrier and clipping negative time- Here, we consider a VLC system that adopts optical OFDM
domain signals [5]. Moreover, a flip-OFDM was proposed for data transmission. In this VLC system, a RGBA-LED
by transmitting positive and flipped negative signals in two is employed as transmitter, while four photo-detectors (PDs)
consecutive OFDM symbols [6]. Then, the flipped concept was together with bandpass optical filters are used to receive the
extended to a 2 × 2 multiple-input multiple-output (MIMO)- optical signal. In this paper, the considered VLC system is
OFDM VLC system [7], where positive and flipped negative referred as RGBA-LED-based VLC system.
signals were transmitted in two independent LEDs.
A. LED characteristic
Recently, white LED consisting of multi-color chips, i.e.
red, green, and blue (RGB-LED) or red, green, blue, and In RGBA-LED-based VLC system, the RGBA-LED in-
amber (RGBA-LED), has also been employed as the trans- corporating with red, green, blue and amber LED chips is
mitter in the VLC systems due to their high modulation employed for both illumination and communication due to its
bandwidth and quick response time. Moreover, the nature flexibility for producing a wide range colored light and its
property of these multiple colors makes the system being a wide modulation bandwidth [11], [12]. After pre-distortion,

978-1-5090-1701-0/16/$31.00 ©2016 IEEE

 LEDs
the output optical power of each LED chip in the RGBA- +
BAC D/A
P/S
LED is proportional to the input electrical current in a forward Data 1 Mod IFFT
+CP
Sep
-
current range [𝐼𝑚𝑖𝑛 , 𝐼𝑚𝑎𝑥 ], where 𝐼𝑚𝑖𝑛 and 𝐼𝑚𝑎𝑥 represent BAC D/A

the minimum and maximum input electrical current [13]. For +

BAC D/A
P/S
simplicity and without loss of generality, we assume that the Data 2 Mod IFFT
+CP
Sep
-
minimum input current equals to zero, i.e. 𝐼𝑚𝑖𝑛 = 0, and an BAC D/A
Transmitter
ideal intensity modulator is used.1 Noise PDs
-CP
In order to fulfill the requirement of illumination, the + FDE FFT
S/P
A/D +
Data 1 Demod +
optical power of each color in the RGBA-LED should be set - FDE FFT
-CP
A/D +
S/P
according to the CMR target for producing required illumi-
+ FDE -CP
nation light color. The CMR of the RGBA-LED is defined FFT A/D +
S/P
Data 2 Demod +
as 𝛾𝑅 : 𝛾𝐺 : 𝛾𝐵 : 𝛾𝐴 , where 𝛾𝑖 , 𝑖 ∈ {𝑅, 𝐺, 𝐵, 𝐴} denotes - FDE FFT
-CP
S/P
A/D +

the color ratio of the 𝑖th color in the RGBA-LED. Compared Optical
Receiver filter
with the RGB-LED, the RGBA-LED is more flexible for
producing the required light, and it also provides an additional Fig. 1. The block diagram of the R-OFDM scheme.
dimension for data transmission. However, the amber color
induces additional cross-talk at the receiver.
B. Optical Channel 𝑋[𝑚] = 𝑋 ∗ [𝑁 − 𝑚], 0 < 𝑚 < 𝑁2 , where ∗ is the complex
In this paper, the optical channel with line-of-sight (LoS) conjugation. While the two components 𝑋[0] and 𝑋[𝑁/2] are
path is considered [2], [14]. The channel gain of the LoS link set to zero, i.e. 𝑋[0] = 𝑋[𝑁/2] = 0, in order to avoid any DC
is expressed as: shift and residual complex component in time-domain signal.
{ Then an IFFT is performed to produce the time-domain 𝑥[𝑘].
(𝑚+1)𝐴 𝑚
ℎ= 2𝜋𝑑2 cos 𝜙 cos 𝜓, 0 ≤ 𝜓 ≤ 𝜓𝑐 , (1)
For large 𝑁 , the signal 𝑥[𝑘] in a symbol can be modeled as a
0, 𝜓 > 𝜓𝑐 , Gaussian random variable with zero mean and variance 𝜎 2 .The
probability density function (PDF) of the signal follows:
where 𝑚 = − ln 2/ ln(cos Φ1/2 ) is the order of the Lambertian ( )
emission with Φ1/2 being the semi-angle at half-power of the 1 𝑥2
𝑓𝑥 (𝑥) = √ exp − 2 . (2)
LED, 𝐴 is the PD effective area, 𝑑 is the distance between 2𝜋𝜎 2 2𝜎
the LED and the PD, 𝜙 and 𝜓 denote, respectively, the After a parallel to serial (P/S) operation, cyclic prefix (CP)
corresponding irradiance angle and incidence angle, and 𝜓𝑐 is added to the start of each symbol. In other words, a number
is the field of view (FOV) of the PD. of samples at the end of each symbol are appended to the start
In RGBA-LED-based VLC system, the optical channel for of the symbol, which yields:
each color can be treated as same since the space between
each LED chip can be ignored compared with the transmission x = (𝑥[𝑁 − 𝜈], ..., 𝑥[𝑁 − 1], 𝑥[0], 𝑥[1], 𝑥[2], ..., 𝑥[𝑁 − 1]) ,
distance. (3)
where 𝜈 is the length of the CP. Although the CP increases
III. P ROPOSED R-OFDM S CHEME the redundant of the signal, it eliminates both ISI and inter-
Now, we propose an optical R-OFDM scheme for reliable carrier interference (ICI), and simplifies the frequency-domain
data transmission in the RGBA-LED-based VLC system. The equalizer (FDE) at the receiver. For simplicity of notation, we
block diagram of the R-OFDM is shown in Fig. 1, where two omit the time-domain and frequency-domain indices [𝑘] or [𝑚]
independent data streams are transmitted simultaneously. In in the rest of this paper without causing any confusion.
the following, we will elaborate the transmitter and receiver Considering the non-negative requirement of the intensity
of the R-OFDM. modulation, the concept of flipping is first introduced to the R-
OFDM. The bipolar signal in (3) is separated into two unipolar
A. Transmitter
signals as follows:
For simplicity and without loss of generality, we assume that { {
two information data streams have the same transmit power. 𝑥, 𝑥 > 0, −𝑥, 𝑥 < 0,
𝑥𝑝𝑜 = 𝑥𝑛𝑒 = (4)
Now, we take one data stream as example to describe the 0, 𝑥 ≤ 0, 0, 𝑥 ≥ 0.
procedures of the transmitter in the R-OFDM.
At the transmitter, the modulated signals, i.e. quadrature To satisfy the illumination requirement, the separated signal
amplitude modulation (QAM), are mapped into all 𝑁 subcar- should be adjusted according to the CMR target of the RGBA-
riers with the property of Hermitian symmetric. It is given by: LED. Therefore, a BAC operation is further performed to each
separated signal as follows:
1 Note that, the minimum limitation can be realized by adding a DC bias
𝐼𝑚𝑖𝑛 to the input signal for satisfying the minimum input limitation of the ˜𝑝𝑜 = min(𝑥𝑝𝑜 , 𝐼𝑚𝑎𝑥 − 𝛽𝑝𝑜 ) + 𝛽𝑝𝑜 ,
𝑥
RGBA-LED. However, the DC bias 𝐼𝑚𝑖𝑛 carries no information, which only
affects the unused 0th subcarrier after the FFT [3], [5]. 𝑥
˜𝑛𝑒 = min(𝑥𝑛𝑒 , 𝐼𝑚𝑎𝑥 − 𝛽𝑛𝑒 ) + 𝛽𝑛𝑒 , (5)
where 𝛽𝑝𝑜 and 𝛽𝑛𝑒 represent the biases added to different will certainly decrease the received SNR, thus will degrade the
signals, which will be elaborated in the next section. Here, we performance of the R-OFDM system.
define the clipping ratio (CR) as [3]: Now, the optical power of the benchmark signal 𝑃¯𝑏𝑚 can
( ) be obtained from setting 𝛽𝑏𝑚 = 0 in (10). According to the
𝐼𝑚𝑎𝑥
𝐶𝑅 = 20 log10 . (6) CMR of the RGBA-LED, we have:
𝜎
𝑃¯𝑖 𝛾𝑖
After the BAC operation, four non-negative signals will ¯ = , (11)
be assigned to the specific color of the RGBA-LED. Note 𝑃𝑏𝑚 𝛾 𝑏𝑚

that, two independent data streams are separated into four where 𝛾𝑏𝑚 is the ratio of the benchmark color, 𝑃¯𝑖 and 𝛾𝑖
unipolar signals and transmitted by different color components denote the optical power and the ratio of the 𝑖th separated
of the RGBA-LED in the R-OFDM, hence it achieves same signal, respectively. By substituting (10) into (11), the expres-
bandwidth efficiency compared with ACO-OFDM. sion can be written in the function 𝛽𝑖 . However, it is difficult
to calculate 𝛽𝑖 in closed form, we resort to computing 𝛽𝑖
B. CMR-targeted Signal Adjustment by numerical methods. In practice, once the RGBA-LED is
chosen and the signal transmit power is set, the bias of each
To determine the biases added to different signals, we first
signal can be off-line calculated and stored in a look-up table.
characterize the statistical property of each separated signal.
For large 𝑁 , the separated signal, e.g. 𝑥𝑖 , 𝑖 ∈ {𝑅, 𝐺, 𝐵, 𝐴}, C. Receiver
follows a rectified Gaussian distribution [15], [16], whose PDF
After passing through the optical channel, the received
is: ( )
𝛿(𝑥) 𝑢(𝑥) 𝑥2 sample is expressed as:
𝑓𝑥𝑖 (𝑥) = +√ exp − 2 , (7)
2 2𝜋𝜎 2 2𝜎 y = Gx̃ + n, (12)
where 𝛿(𝑥) is the Dirac delta function and 𝑢(𝑥) is the 𝑇
where y = (𝑦𝑅 , 𝑦𝐺 , 𝑦𝐵 , 𝑦𝐴 ) is the received signal with 𝑇
Heaviside step function. Note that, four separated signals have
being the transpose operation, x̃ = (ℎ𝑅 ⊗ 𝑥˜ 𝑅 , ℎ𝐺 ⊗ 𝑥˜ 𝐺 , ℎ𝐵 ⊗
the identical optical power since two data streams use the same
˜𝐴 )𝑇 with ⊗ being the convolution operator and ℎ𝑖
˜ 𝐵 , ℎ𝐴 ⊗ 𝑥
𝑥
transmit power. According to (7), the optical and electrical
denoting the channel response of the 𝑖th optical channel, and
power of the 𝑖th signal are calculated as:
∫ ∞ n = (𝑛𝑅 , 𝑛𝐺 , 𝑛𝐵 , 𝑛𝐴 )𝑇 denotes the summation of the thermal
𝜎 noise and shot noise, which can be approximated as a Gaussian
𝑃𝑖 = 𝔼[𝑥𝑖 ] = 𝑥𝑓𝑥𝑖 (𝑥)𝑑𝑥 = √ , (8)
−∞ 2𝜋 process with zero mean and variance 𝜎𝑛2 𝑖 for the 𝑖th color.
Finally, 4×4 matrix G denotes the combined cross-talk matrix
and ∫ ∞
𝜎2 caused by the electrical interference at the transmitter, PD
𝐸𝑖 = 𝔼[𝑥2𝑖 ] = 𝑥2 𝑓𝑥𝑖 (𝑥)𝑑𝑥 = . (9) responsivity, and imperfection of the optical filter at receiver.
−∞ 2
Its element is represented by 𝑔𝑖,𝑗 , which denotes the optical
In the RGBA-LED, the CMR equals to the ratio of the opti- front-end gain between the transmit LED color 𝑗 to the receive
cal power for each reshaped signal. After the BAC operation, PD color 𝑖. A typical model of the cross-talk matrix is given
the optical power of the 𝑖th signal is calculated by using (5) as [17]:
and (7), which yields: ⎛ ⎞
∫ 𝜆𝑖 ( ) 1−𝜉 𝜉 0 0
1 𝑥2 ⎜ 𝜉 1 − 2𝜉 𝜉 0 ⎟
𝑃¯𝑖 = √ 𝑥 exp − 2 𝑑𝑥 G=⎜ ⎟, (13)
2𝜋𝜎 2 0 2𝜎 ⎝ 0 𝜉 1 − 2𝜉 𝜉 ⎠
∫ ∞ ( )
𝜆𝑖 𝑥2 0 0 𝜉 1−𝜉
+√ exp − 2 𝑑𝑥 + 𝛽𝑖
2𝜋𝜎 2 𝜆𝑖 2𝜎 where 𝜉 ∈ [0, 0.5) characterizes the interference ratio. To be
[ ( )] ( )
𝜎 𝜆2𝑖 𝜆𝑖 specific, the received signal of the 𝑖th color is rewritten as:
=√ 1 − exp − 2 + 𝜆𝑖 𝑄 + 𝛽𝑖 , (10) ∑
2𝜋 2𝜎 𝜎
( 2) 𝑦𝑖 = 𝑔𝑖,𝑖 ℎ𝑖 ⊗ 𝑥
˜𝑖 + 𝑔𝑖,𝑗 ℎ𝑗 ⊗ 𝑥
˜𝑗 + 𝑛𝑖
∫∞
where 𝜆𝑖 = 𝐼𝑚𝑎𝑥 − 𝛽𝑖 and 𝑄(𝑥) = √12𝜋 𝑥 exp − 𝑡2 𝑑𝑡 is 𝑗∕=𝑖

the Gaussian Q-function. = 𝑔𝑖,𝑖 ℎ𝑖 ⊗ 𝑥

˜𝑖 + 𝑤𝑖 + 𝑛𝑖 , (14)
To facilitate the calculation of the bias for each signal, we ∑
where 𝑖, 𝑗 ∈ {𝑅, 𝐺, 𝐵, 𝐴} and 𝑤𝑖 = 𝑗∕=𝑖 𝑔𝑖,𝑗 ℎ𝑗 ⊗ 𝑥 ˜𝑗 denotes
take one signal as the benchmark and set the bias to zero, the interference caused by cross-talk.
i.e. 𝛽𝑏𝑚 = 0. In the biasing operation, the signal with the After the CP removal and S/P conversion, the time-domain
minimum color ratio is chosen as the benchmark. Under this signal in (14) is transformed to the frequency-domain signal by
situation, different DC biases are added to other signals, which performing FFT. Next, a single-tap equalizer is performed for
have no impacts on the effective signals since the DC bias eliminating ISI. For the 𝑖th signal, the output of the equalizer
only affects the unused 0th subcarrier after the FFT at the is:
receiver. To the contrary, if other signals are chosen as the 𝑌𝑖 ˜ 𝑖 + 𝑊𝑖 + 𝑁𝑖 ,
𝑌˜𝑖 = =𝑋 (15)
benchmark, parts of the effective signal will be clipped, which 𝑔𝑖,𝑖 𝐻𝑖 𝑔𝑖,𝑖 𝐻𝑖 𝑔𝑖,𝑖 𝐻𝑖
where 𝑌𝑖 and 𝑋 ˜ 𝑖 denote the frequency-domain signal of 𝑦𝑖 where 𝜅𝑖 is the attenuation factor given by:
and 𝑥˜𝑖 , 𝐻𝑖 is the frequency-domain channel response, which ¯𝑖 ]
𝔼[𝑥, 𝑥
is considered flat in each subcarrier, and 𝑊𝑖 and 𝑁𝑖 denote the 𝜅𝑖 =
𝔼[𝑥2 ]
frequency-domain representations of 𝑤𝑖 and 𝑛𝑖 , respectively. [∫ ∫ ∞ ]
𝜆𝑖
At the transmitter, the bipolar time-domain signal of each 1 2
= 2 𝑥 𝑓𝑥 (𝑥)𝑑𝑥 + 𝜆𝑖 𝑥𝑓𝑥 (𝑥)𝑑𝑥
data stream is separated into two unipolar signals and trans- 𝜎 0 𝜆𝑖
mitted over two different colors. Accordingly, two signals ( )
1 𝜆𝑖
separated from the same data stream should be reconstructed = −𝑄 . (19)
2 𝜎
before detection. At the receiver, the reconstruction operation
is performed after the FDE operation, which is equivalent to The variance of the clipping distortion is:
that in time domain due to the linearity property of FFT. 𝜎𝑑2𝑖 = 𝔼[¯
𝑥2𝑖 ] − 4𝜅2𝑖 𝐸𝑖 , (20)
Therefore, the reconstructed signal for one data stream is
obtained as: where
∫ ∞
𝑌 = 𝑌˜𝑖 − 𝑌˜𝑗 , (16) 𝑥2𝑖 ] =
𝔼[¯ 𝑥¯2𝑖 𝑓𝑥𝑖 (𝑥)𝑑𝑥
−∞
where 𝑖, 𝑗 ∈ {𝑅, 𝐺, 𝐵, 𝐴} denote the indices of two signals ( ) ( )
𝜎2 2 2 𝜆𝑖 𝜆𝑖 𝜎 𝜆2𝑖
separated from the data stream. Finally, the reconstructed = + [𝜆𝑖 − 𝜎 ]𝑄 − √ exp − 2 . (21)
2 𝜎 2𝜋 2𝜎
complex signal 𝑌 is detected for recovering the transmitted
data. Considering the interference, PD noise and clipping distor-
tion at the receiver, the received signal in (14) is rewritten
as:
IV. P ERFORMANCE A NALYSIS
𝑦𝑖 = 2𝜅𝑖 𝑔𝑖,𝑖 ℎ𝑖 ⊗ 𝑥𝑖 + 𝑔𝑖,𝑖 ℎ𝑖 ⊗ 𝑑𝑖 + 𝑤𝑖 + 𝑛𝑖 . (22)
In the R-OFDM, two independent information data streams
are transmitted by the RGBA-LED at the instantaneous time. Note that the bias component is ignored in (22) since it only
For one stream, the BER of M-QAM can be approximately affects the unused 0th subcarrier after FFT. After passing
evaluated by using the expression [18]: through the FFT and FDE, the signal is reconstructed as:
√ (√ )
4( 𝑀 − 1) 3 log2 (𝑀 ) 𝑌 = 2𝜅𝑖 𝑋𝑖 − 2𝜅𝑗 𝑋𝑗 + 𝐷𝑖 − 𝐷𝑗
𝑃𝑏 = √ 𝑄 Γ 𝑊𝑖 𝑊𝑗 𝑁𝑖 𝑁𝑗
𝑀 log2 𝑀 𝑀 −1 + + + + , (23)
√ ( √ ) 𝑔𝑖𝑖 𝐻𝑖 𝑔𝑗𝑗 𝐻𝑗 𝑔𝑖𝑖 𝐻𝑖 𝑔𝑗𝑗 𝐻𝑗
4( 𝑀 − 2) 3 log2 (𝑀 )
+√ 𝑄 3 Γ , (17) where the 𝑖th and 𝑗th signal are the two signals separated from
𝑀 log2 𝑀 𝑀 −1 one data stream, and 𝐷𝑖 and 𝐷𝑗 denote the frequency-domain
where Γ denotes the effective receive SNR. Accordingly, representation of the clipping distortion of two signals.
the total BER of M-QAM for the R-OFDM can be readily The effective receive SNR is defined as the ratio of effective
calculated by averaging the BER of two data streams. signal power to the summation power of clipping distortion,
noise and interference. It can be expressed as:
A. Effective SNR 4𝜅2𝑖 𝐸𝑖 + 4𝜅2𝑗 𝐸𝑗
Γ= 2
𝜎𝑛
,
𝜎𝑛2
𝔼[𝑤𝑖2 ] 𝔼[𝑤𝑗2 ]
In the R-OFDM, the effective SNR depends on the effective 𝜎𝑑2𝑖 + 𝜎𝑑2𝑗 + 𝑖
∣𝑔𝑖𝑖 𝐻𝑖 ∣2 + ∣𝑔𝑗𝑗 𝐻𝑗
𝑗
∣2 + ∣𝑔𝑖𝑖 𝐻𝑖 ∣2 + ∣𝑔𝑗𝑗 𝐻𝑗 ∣2
transmit power, the interference caused by cross-talk, the noise (24)
caused by PD and ambient light, and the clipping distortion
where
induced by BAC at the transmitter. First, we evaluate the ∑
clipping distortion at the transmitter. 𝔼[𝑤𝑖2 ] = ˜ 𝑘 ∣2 ],
∣𝑔𝑖𝑘 𝐻𝑘 ∣2 𝔼[∣𝑋 (25)
For satisfying the illumination requirement, different biases 𝑘∕=𝑖,𝑘∈{𝑅,𝐺,𝐵,𝐴}
are added to different colors according to the CMR target. For ˜ 𝑘 being the frequency-domain representation of 𝑥
with 𝑋 ˜𝑘 , and
color 𝑖 ∈ 𝑅, 𝐺, 𝐵, 𝐴, the separation operation together with ˜
𝔼[∣𝑋𝑘 ∣2 ] is calculated as:
BAC is equivalent to the BAC operation in current optical ( )
∫ 𝜆𝑖
OFDM, where the bottom clipping level equals to zero and ˜ 𝑘 ∣2 ] = √ 1 𝑥2
the top clipping level is 𝜆𝑖 . By assuming the Gaussianity of 𝔼[∣𝑋 𝑥2 exp − 2 𝑑𝑥
2𝜋𝜎 2 (0 2𝜎
𝑥, the clipping distortion can be evaluated by applying the ∫ ∞ )
𝜆2𝑖 𝑥2
Bussgang’s Theorem [19], [20]. The signal after clipping 𝑥 ¯𝑖 +√ exp − 2 𝑑𝑥 + 𝛽𝑖2
2𝜋𝜎 2 𝜆𝑖 2𝜎
can be decomposed into an attenuated version of 𝑥 plus a ( ) ( )
clipping distortion, 𝑑𝑖 , which is independent to 𝑥. It can be 𝜎2 𝜆 𝑖 𝜆𝑖 𝜎 𝜆2
= + (𝜆2𝑖 − 𝜎 2 )𝑄 − √ exp − 𝑖2 + 𝛽𝑖2 .
expressed as: 2 𝜎 2𝜋 2𝜎
¯𝑖 = 2𝜅𝑖 𝑥 + 𝑑𝑖 ,
𝑥 (18) (26)
 The expression of effective SNR in (24) seems a little
cumbersome to give direct insight. Here, we define the signal- 4QAM, N = 1024

to-distortion ratio (SDR) as: 10

−2

4𝜅2𝑖 𝐸𝑖 + 4𝜅2𝑗 𝐸𝑗
Γ𝑆𝐷𝑅 = . (27)
𝜎𝑑2𝑖 + 𝜎𝑑2𝑗 10
−3

The receive signal-to-interference-noise ratio (SINR) is de-

BER
fined as: 10
−4

𝔼[¯𝑥2𝑖 ] + 𝑥2𝑗 ]
𝔼[¯
Γ𝑆𝐼𝑁 𝑅 = 2
𝜎𝑛
. (28)
𝜎𝑛2
𝔼[𝑤𝑖2 ] 𝔼[𝑤𝑗2 ] CR = 6 dB, Simulation
𝑖
∣𝑔𝑖𝑖 𝐻𝑖 ∣2 + 𝑗
∣𝑔𝑗𝑗 𝐻𝑗 ∣2 + ∣𝑔𝑖𝑖 𝐻𝑖 ∣2 + ∣𝑔𝑗𝑗 𝐻𝑗 ∣2
10
−5
CR = 8 dB, Simulation
CR = 10 dB, Simulation
Analytical, ξ = 0
From substituting (27) and (28) into (24), the effective Analytical, ξ = 0.03

receive SNR can be rewritten as: 10

−6

6 7 8 9 10 11 12 13 14 15 16

Γ𝑆𝐷𝑅 Γ𝑆𝐼𝑁 𝑅 ΓTx (dB)

Γ= . (29)
1 + Γ𝑆𝐷𝑅 + Γ𝑆𝐼𝑁 𝑅
Fig. 2. BER performance of the R-OFDM for 4QAM.
Note that, the BER of M-QAM for one data stream can be
calculated by substituting (29) into (17).
Remark: In the RGBA-LED-based VLC system, the inter-
N = 1024, CR = 10 dB
ference caused by cross-talk exists all the times. When CR
−1
10
becomes larger, the clipping distortion can be ignored, i.e.
𝜅𝑖 → 1, 𝜎𝑑2𝑖 → 0. Therefore, the interference and noise
will be the dominative components that affect the effective 10
−2

SNR. Under this situation, we have Γ𝑆𝐼𝑁 𝑅 ≪ Γ𝑆𝐷𝑅 , and the

effective SNR is upper bounded by Γ ≤ Γ𝑆𝐼𝑁 𝑅 . Accordingly,
BER

−3
10
the lower bound of the BER for M-QAM can be calculated
by substituting (28) in (17).
−4
10 16QAM, Simulation
V. S IMULATION R ESULTS 32QAM, Simulation
64QAM, Simulation

Here, we present the Monte-Carlo simulations for eval- −5

Analytical, ξ = 0
Analytical, ξ = 0.03
10
uating the performance of the R-OFDM scheme over the Lower bound

ideal optical channel, which is often modeled as additive 6 8 10 12 14 16 18

SNR (dB)
20 22 24 26 28 30

white Gaussian noise (AWGN) channel. In the simulations,

we perform 104 times Monte-Carlo simulations. In each time,
Fig. 3. BER performance of the R-OFDM for different modulation schemes.
2(𝑁/2−1) log2 (𝑀 ) random bits are generated and modulated
by rectangle M-QAM modulation scheme. The CMR of the
RGBA-LED is 𝛾𝑅 : 𝛾𝐺 : 𝛾𝐵 : 𝛾𝐴 = 1 : 1.2 : 1.5 : 1.59 at
a color temperature of 5500K. The cross-talk in the system of the analytical results consist with the simulation results.
depends on the parameter 𝜉 as shown in (13). On the other hand, the cross-talk severely deteriorates the
For simplicity and without loss of generality, we normalize BER performance of the R-OFDM because of the interference
the electrical transmit power for each symbol to unity, and we caused by the cross-talk between different colors.
assume that the variance of the receiver noise in each color is We further test the BER performance for different modu-
Gaussian and identical. In the results, transmit SNR denotes lation schemes in Fig. 3, where 𝐶𝑅 = 10 dB, 𝑁 = 1024.
the total electrical transmit power of each symbol to the power The BER lower bound is calculated by substituting (28) into
of the noise at the receiver. It is defined as: (17). It is observed from Fig. 3 that the analytical results
∑ are in accordance with the simulation results for different
𝔼[𝑥2 ]
Γ𝑇 𝑥 = ∑𝑖 2 𝑖 . (30) modulation schemes, and the lower bound is closed to the
𝑖 𝜎𝑛𝑖 simulation results. Furthermore, we can see that higher order
We first evaluate the BER performance of 4QAM for the modulation scheme is more sensitive to the cross-talk due to
R-OFDM in Fig. 2, where 𝑁 = 1024, the analytical results its high peak-to-average power ratio (PAPR). For example, the
are obtained from (17), the simulation results are obtained performance loss under cross-talk scenario compared with no
by performing the Monte-Carlo simulation. As we can see cross-talk scenario is about 1.2 dB for 16QAM, while the loss
from Fig. 2, the analytical results perfectly agree with the is about 2 dB for 32QAM at 10−4 BER.
simulation results under no cross-talk, i.e. 𝜉 = 0. While We now compare the BER performance of the R-OFDM and
there is an ignorable gap exist between analytical results and ACO-OFDM. In ACO-OFDM, four independent data streams
simulation results under cross-talk, i.e. 𝜉 = 0.03, but the trend can be transmitted simultaneously by the RGBA-LED. To
 0
10
Universities, and the Research Fund of National Mobile Com-
N = 1024, CR = 10 dB, ξ = 0.03
munications Research Laboratory, Southeast University under
−1
10
2016A02.
R EFERENCES
−2
10
[1] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wireless commu-
nication: Potential and state-of-the-art,” IEEE Commun. Mag., vol. 49,
BER

−3
10
no. 9, pp. 56–62, Sep. 2011.
[2] T. Komine and M. Nakagawa, “Fundamental analysis for visible-light
communication system using LED lights,” IEEE Trans. Consum. Elec-
−4
10
Reshaped−OFDM, 16QAM
tron., vol. 50, no. 1, pp. 100–107, Feb. 2004.
Reshaped−OFDM, 32QAM [3] J. Armstrong, “OFDM for optical communications,” J. Lightw. Technol.,
Reshaped−OFDM, 64QAM vol. 27, no. 3, pp. 189–204, Feb. 2009.
−5 ACO−OFDM, 16QAM
10
ACO−OFDM, 32QAM
[4] J. B. Carruthers and J. M. Kahn, “Multiple-subcarrier modulation for
ACO−OFDM, 64QAM non-directed wireless infrared communication,” IEEE J. Sel. Areas
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Commun., vol. 14, no. 3, pp. 538–546, Apr. 1996.
ΓTx (dB) [5] J. Armstrong and A. J. Lowery, “Power efficient optical OFDM,”
Electron. Lett., vol. 42, no. 6, pp. 370–371, Mar. 2006.
[6] N. Fernando, H. Y, and E. Viterbo, “Flip-OFDM for unipolar communi-
Fig. 4. BER performance of the R-OFDM compared with ACO-OFDM for cation systems,” IEEE Trans. Commun., vol. 60, no. 12, pp. 3726–3733,
different modulation schemes. Dec. 2012.
[7] L. Wu, Z. C. Zhang, and H. P. Liu, “MIMO-OFDM visible light
communications system with low complexity,” in Proc. IEEE ICC, Jun.
2013, pp. 3933–3937.
achieve fair comparison, we apply the BAC operation to ACO- [8] G. Cossu, A. M. Khalid, P. Choudhury, R. Corsini, and E. Ciaramella,
OFDM, and the total transmit power of ACO-OFDM is same “3.4 Gbit/s visible optical wireless transmission based on RGB LED,”
as that of the R-OFDM. Opt. Express, vol. 20, no. 26, pp. B501–B506, Dec. 2012.
[9] Y. Q. Wang and N. Chi, “A high-speed bi-directional visible light
The BER comparison of the R-OFDM and ACO-OFDM communication system based on RGB-LED,” China Commun., vol. 11,
for different modulation scheme is shown in Fig. 4, where no. 3, pp. 40–44, Mar. 2014.
𝑁 = 1024, 𝐶𝑅 = 10 dB, 𝜉 = 0.03 and {16, 32, 64}- [10] D. Bykhovsky and S. Arnon, “OFDM allocation optimization for
crosstalk mitigation in multiple free-space optical interconnection links,”
QAM are tested. It can be seen from Fig. 4 that the BER J. Lightw. Technol., vol. 33, no. 13, pp. 2777–2783, Jul. 2015.
performance of the R-OFDM outperforms ACO-OFDM for [11] R. Singh, T. O.Farrell, and J. P. R. David, “An enhanced color shift
different modulation schemes. For instance, the SNR gains of keying modulation scheme for high-speed wireless visible light com-
munications,” J. Lightw. Technol., vol. 32, no. 14, pp. 2582–2592, Jul.
the R-OFDM over ACO-OFDM at 10−4 BER are about 0.2 2014.
dB and 1.4 dB for 16QAM and 32QAM, respectively, and [12] L. Engin, “High luminous efficacy RGBA LED emitter LZC-03MA07.”
the SNR gain becomes large for 64QAM. This is because the [Online]. Available: http://www.mouser.com/ds/2/228/LZC-03MA07-
253130.pdf
negative clipper at the transmitter of ACO-OFDM decreases [13] Z. C. Wang, Q. Wang, S. Chen, and L. Hanzo, “An adaptive scaling and
the signal power by half, which makes ACO-OFDM more biasing scheme for OFDM-based visible light communication systems,”
sensitive to the cross-talk. Therefore, the R-OFDM is more Opt. Express, vol. 22, no. 10, pp. 12 707–12 715, May. 2014.
[14] J. Grubor, S. Randel, K.-D. Langer, and J. W. Walewski, “Broadband
reliable to implement in the RGBA-LED-based VLC system. information broadcasting using LED-based interior lighting,” J. Lightw.
Technol., vol. 26, no. 24, pp. 3883–3892, Dec. 2008.
VI. C ONCLUSION [15] K. Asadzadeh, A. Dabbo, and S. Hranilovic, “Receiver design for
asymmetrically clipped optical OFDM,” in Proc. IEEE GLOBECOM
This paper presents the R-OFDM scheme in the RGBA- Workshops, Dec. 2011, pp. 777–781.
LED-based VLC system. Consider the requirement of both [16] W. Xu, M. Wu, H. Zhang, X. H. You, and C. M. Zhao, “ACO-OFDM-
illumination and communication, the OFDM signal for each specified recoverable upper clipping with efficient detection for optical
wireless communications,” IEEE Photonics J., vol. 6, no. 5, pp. 1–17,
color is reshaped for transmitting through the RGBA-LED. At Oct. 2014.
the receiver, a direct detection is designed for recovering the [17] Q. Gao, C. Gong, R. Wang, Z. Y. Xu, and Y. B. Hua, “Constellation
information data. Furthermore, the effective SNR is conducted design for multi-color visible light communications,” arXiv preprint
arXiv, 2014.
for evaluating the BER performance of the R-OFDM. Simula- [18] S. Dimitrov and H. Haas, “Information rate of OFDM-based optical
tion results confirm the correctness of the theoretical analysis wireless communication systems with nonlinear distortion,” J. Lightw.
and verify that the R-OFDM is suitable for implementing in Technol., vol. 31, no. 6, pp. 918–929, Mar. 2013.
[19] J. J. Bussgang, “Cross correlation function of amplitude-distorted Gaus-
the RGBA-LED-based VLC system. In our further work, we sian signals,” Research Lab. Electron, M.I.T., Cambridge, Tech. Rep.
will focus on other schemes for satisfying the illumination 216, Mar. 1952.
requirement and design enhanced receiver for improving the [20] S. Dimitrov, S. Sinanovic, and H. Haas, “Clipping noise in OFDM-
based optical wireless communication systems,” IEEE Trans. Commun.,
BER performance of the R-OFDM. vol. 60, no. 4, pp. 1072–1081, Apr. 2012.

ACKNOWLEDGMENT
This work was supported by the National Basic Research
Program of China under 2013CB329203, the National Natural
Science Foundation of China under 61471114, 61571118 and
61223001, the Fundamental Research Funds for the Central