Optics Communications 415 (2018) 177–185

Contents lists available at ScienceDirect

Optics Communications
journal homepage: www.elsevier.com/locate/optcom

Hermitian symmetry free optical-single-carrier frequency division multiple

access for visible light communication
Ali W. Azim a, *, Yannis Le Guennec b , Ghislaine Maury a
a
Université Grenoble Alpes, CNRS, Institute of Engineering, Grenoble INP, IMEP-LAHC, 38000 Grenoble, France
b
Université Grenoble Alpes, CNRS, Institute of Engineering, Grenoble INP, GIPSA-LAB, 38000 Grenoble, France

a r t i c l e i n f o a b s t r a c t

Keywords: Optical-orthogonal frequency division multiplexing (O-OFDM) is an effective scheme for visible light communi-
Intensity modulation-direct detection cations (VLC), offering a candid extension to multiple access (MA) scenarios, i.e., O-OFDMA. However, O-OFDMA
Optical-orthogonal frequency division exhibits high peak-to-average power ratio (PAPR), which exacerbates the non-linear distortions from the light
multiplexing emitting diode (LED). To overcome high PAPR while sustaining MA, optical-single-carrier frequency-division
Peak-to-average power ratio
multiple access (O-SCFDMA) is used. For both O-OFDMA and O-SCFDMA, Hermitian symmetry (HS) constraint
Optical-single-carrier frequency domain
multiple access
is imposed in frequency-domain (FD) to obtain a real-valued time-domain (TD) signal for intensity modulation-
direct detection (IM-DD) implementation of VLC. Howbeit, HS results in an increase of PAPR for O-SCFDMA. In
this regard, we propose HS free (HSF) O-SCFDMA (HSFO-SCFDMA). We compare HSFO-SCFDMA with several
approaches in key parameters, such as, bit error rate (BER), optical power penalty, PAPR, quantization, electrical
power efficiency and system complexity. BER performance and optical power penalty is evaluated considering
multipath VLC channel and taking into account the bandwidth limitation of LED in combination with its
optimized driver. It is illustrated that HSFO-SCFDMA outperforms other alternatives.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction real-valued and non-negative. Several O-OFDM schemes satisfying IM-

DD constraints, such as, direct-current (DC) O-OFDM (DCO-OFDM) [2],
Over the past couple of decades, there is an exponential surge in asymmetrically clipped (AC) O-OFDM (ACO-OFDM) [3], Flip-OFDM [6],
the deployment of radio frequency (RF) wireless systems, which has Hermitian symmetry (HS) free (HSF)-OFDM (HSF-OFDM) [7] etc., have
led to RF spectral congestion. Visible light communications (VLC) is been proposed.
perceived as a complementary technology to overcome this looming RF Despite advantages, one operational limitation of O-OFDM is the
spectral crisis. With onset of incoherent high power light emitting diode high peak-to-average-power ratio (PAPR), which makes it susceptible to
(LED) and sensitive photo-detector (PD), VLC has gained a substantial
non-linear distortions [8,9]. Besides, due to an increase in quantization
interest [1]. Along with some compelling advantages, such as, license
noise, limited bit resolution of digital-to-analog converter (DAC) and
free unlimited optical bandwidth, high-security and no electromagnetic
analog-to-digital converter (ADC) also bound the performance [10].
interference, VLC is appealing as it concurrently offers both lighting and
Various methods [8,9,11] and [12] are devised to counteract the high
communication.
For VLC, optical-orthogonal frequency division multiplexing (O- PAPR of O-OFDM. Though, an increased complexity overhead and/or
OFDM) is regarded as an effective approach as it allows high data- bandwidth inefficiency is associated with these techniques.
rate, a simple one tap equalization in the frequency-domain (FD) To realize a complete networking VLC system, multiple access (MA)
and shows an inherent resilience to combat inter-symbol-interference is essential [13,14]. O-OFDM provides a straightforward extension to
(ISI) resulting from dispersive multipath propagation [2–5]. In VLC, MA, i.e., O-OFDMA, where the subcarriers are allocated to different
O-OFDM operates using simple, low-cost intensity modulation-direct users based on a pre-defined resource allocation matrix. However, O-
detection (IM-DD), wherein, the intensity waveform modulates onto the OFDMA also manifests high PAPR, therefore, the non-linear distortions
brightness of the LED which is photo-detected at the receiver. For IM- are inevitable. To surmount the distortions and to accommodate MA,
DD implementation, the time-domain (TD) signal is necessitated to be optical-single carrier frequency division multiple access (O-SCFDMA)

* Corresponding author.
E-mail addresses: [email protected] (A.W. Azim), [email protected] (Y. Le Guennec), [email protected] (G. Maury).

https://doi.org/10.1016/j.optcom.2018.01.036
Received 1 November 2017; Received in revised form 17 January 2018; Accepted 18 January 2018
Available online 6 February 2018
0030-4018/© 2018 Elsevier B.V. All rights reserved.
A.W. Azim et al. Optics Communications 415 (2018) 177–185

is used.1 O-SCFDMA is the discrete Fourier transform (DFT)-precoded respectively, where (𝑛1 , 𝑛2 ) ∈ 0, 1, … , 𝐿 − 1, and 𝑗 2 = −1. Furthermore,
variant of O-OFDMA and its foremost advantage is the lower PAPR [15]. C𝐿 , R𝐿 , and R𝐿
+ represent 𝐿-order complex vector, 𝐿-order real-valued
In the literature, O-OFDMA inspired O-SCFDMA schemes, e.g., ACO- and vector, and 𝐿-order positive real-valued vector, respectively. Further-
DCO-SCFDMA [15] have been introduced. Both ACO- and DCO-SCFDMA more, we use 𝐸(⋅), (⋅)H , |⋅|, (⋅)T , and ⊗ to represent the ensemble average,
manifest lower PAPR relative to O-OFDMA counterparts. However, both Hermitian conjugate, absolute, transpose and convolution operators,
are constrained to satisfy the HS constraint. Nevertheless, Wu et al. [16] respectively. ℜ[⋅] and ℑ[⋅] are used to obtain the real and imaginary
identified that in O-SCFDMA, only half of the TD symbols exploit single- components of a vector, respectively.
carrier (SC) like benefits because of HS, hence, DFT-precoding results in
a less compelling PAPR reduction. Consequently, the precoded schemes 2. HSFO-SCFDMA
which preclude HS, surpass others (which use HS) in their efficacy for
PAPR reduction. Recently, SC optical frequency-division multiplexing 2.1. Modulation concept
(SCO-FDM) has been investigated in [17]. SCO-FDM exhibits lower
PAPR compared to ACO-OFDMA/SCFDMA and demonstrates same bit The foremost advantage of precluding HS in O-SCFDMA is the low
error rate (BER) in an additive white Gaussian noise (AWGN) channel. PAPR; which is accomplished because all the TD symbols enjoy SC ben-
However, the PAPR is not reduced to the same level as in interleaved efits. Contrary to that, if HS is implemented, only half of the TD symbols
(I)-SCFDMA for RF systems because of insertion of zeros in the TD manifest SC like behavior, resulting in a less significant PAPR reduction,
signal [16]. Low complexity pulse-amplitude modulation (PAM) based as in ACO- and DCO-SCFDMA. Moreover, it has been established for
discrete Hartley transform (DHT)-precoded ACO-OFDM (DHTS-ACO- both RF [20] and VLC [16] that I-SCFDMA culminates the lowest PAPR.
OFDM) is proposed by Zhou and Qiao in [18] and [19], which also Therefore, in HSFO-SCFDMA, we couple the technique proposed by F.
averts HS. DHTS-ACO-OFDM features lower PAPR compared to ACO- Barrami et al. [7] with I-SCFDMA, to realize an O-SCFDMA approach
OFDMA/SCFDMA, however, with an increase in size of modulation which is capable of exhibiting low PAPR, whilst complying with IM-DD
alphabet, PAPR reduction becomes less efficient. constraints. Furthermore, in HSFO-SCFDMA, a real-valued TD signal is
Against what has been reported in the literature, we propose an O- obtained in a straightforward manner because by imitating I-SCFDMA,
SCFDMA approach capable of attaining superior performance compared only the even subchannels are modulated, which implicitly results in a
to other alternatives. The main contributions of this work are: half-wave symmetric TD signal [20], from which the real and imaginary
components can be separated in a forthright manner.
1. An O-SCFDMA approach, HS free (HSF) O-SCFDMA (HSFO- In the sequel, HSFO-SCFDMA is presented considering a simplified
SCFDMA) is introduced. HSFO-SCFDMA yields several concrete scenario (with only one user). Firstly, the transmitter of HSFO-SCFDMA
advantages over other alternatives, such as, lower PAPR, power is presented. Subsequently, the details of the dispersive channel are
and cost efficiency. presented and finally, the receiver of HSFO-SCFDMA is described.
2. For comparison, together with HSFO-SCFDMA, we adopt ACO-
SCFDMA, DHTS-ACO-OFDM, SCO-FDM, and DCO-SCFDMA. We 2.2. Transmitter
evaluate key performance parameters, such as, PAPR, BER, op-
tical power penalty, system complexity, quantization, electrical The block diagram of HSFO-SCFDMA transmitter is illustrated in
power and spectral efficiency. For BER evaluation, we consider a Fig. 1. The incoming bit stream is parsed into 𝑁∕2 parallel channels
dispersive channel where the dispersions are characterized by the using a serial-to-parallel (S/P) converter which are modulated to Gray-
multipath VLC impulse response and the bandwidth limitation of mapped complex 𝑀-ary quadrature-amplitude modulation (QAM) al-
LED/LED driver combination. phabets resulting in a TD symbol vector, 𝐬 ∈ C𝑁∕2 . Subsequently, 𝐬, is
3. Optical power penalty of the aforementioned approaches is transformed to a FD symbol vector, 𝐒, via 𝑁∕2-order DFT as
analyzed in a dispersive channel. We shall demonstrate that
HSFO-SCFDMA suffers the least optical power penalty. 𝐒 = DFT (𝐬) = 𝐅𝑁∕2 ⋅ 𝐬 ∈ C𝑁∕2 , (1)
4. We evaluate closed-form PAPR expressions as a function of
which represents DFT-precoding. Afterwards, subchannel mapping is
modulation alphabets for the approaches which exhibit constant
performed by assigning 𝑁∕2 DFT-precoded symbols from 𝐒 to the even
PAPR. ̃ 𝐒̃ is explicitly given as
subchannels of 𝑁-order FD symbol vector, 𝐒.
[ ]
̃𝐒 = S(0) , 0, S(1) , 0, … , S(𝑁∕2−2) , 0, S(𝑁∕2−1) , 0 T ∈ C𝑁 . The subchannels in
The rest of the article is organized as follows. In Section 2, we
present HSFO-SCFDMA. Additionally, a brief description of the disper- 𝐒̃ are mapped in an interleaved manner, because it is established that
sive channel is provided. In Section 3, we submit a succinct analysis interleaved subchannel mapping leads to better PAPR performance [16].
of the statistical characterization of HSFO-SCFDMA, where the average After subchannel mapping, 𝐒̃ is transformed to a DFT-precoded TD
symbol vector by applying 𝑁-order IDFT as
electrical and the average optical powers are evaluated. Simulation
results are provided in Section 4. In Section 5, we describe MA for HSFO- ( )
𝐬̃ = IDFT 𝐒̃ = 𝐅H ̃ 𝑁
𝑁 ⋅𝐒∈C . (2)
SCFDMA. Based on obtained results, conclusions are drawn in Section 6.
Owing to the frame structure of 𝐒, ̃ 𝐬̃ features the following properties:
1.1. Notation (P1) 𝐬 is a scaled and repeated version of 𝐬 [17,20]. (P2) 𝐬̃ features half-
̃
wave symmetry, i.e., s̃(𝑛) = s̃(𝑛+𝑁∕2) ∀ 𝑛 = 0, 1, … , 𝑁∕2−1 [20,21], which
Unless otherwise mentioned, lowercase boldface letters denote TD is the inherent outcome if only the even subchannels are modulated. It
symbol vectors, e.g., 𝐬. Moreover, lowercase letters with super-scripted may be noticed that (P1) implies that the 𝑁∕2-order DFT and 𝑁-order
index, like s(𝑛) represent the 𝑛th sample of the TD symbol vector. FD IDFT at the transmitter might be redundant. However, we accentuate
symbol vectors are represented by uppercase, boldface letters, e.g., 𝐒, that both processes are imperative for (frequency division) MA (see
whereas uppercase letters with super-scripted index, e.g., S(𝑛) denote Section 5). By exploiting (P2), the real and the imaginary sub-blocks
the symbol on 𝑛th subchannel. 𝐅𝐿 and 𝐅H 𝐿
, respectively, represent DFT of the half-wave symmetric signal, 𝐬̃ , are obtained as
and IDFT matrices of dimensions 𝐿 × 𝐿, for which the (𝑛1 , 𝑛2 )th element [ ] [ ]
is obtained as 𝐅(𝑛1 ,𝑛2 ) = 𝐿−1∕2 𝑒−𝑗2𝜋𝑛1 𝑛2 ∕𝐿 and 𝐅H = 𝐿−1∕2 𝑒𝑗2𝜋𝑛1 𝑛2 ∕𝐿 , 𝐱ℜ = ℜ s̃(𝑛) , 𝐱ℑ = ℑ s̃(𝑛+𝑁∕2) ∀ 𝑛 = 0, 1, … , 𝑁∕2 − 1, (3)
(𝑛 ,𝑛 )
1 2
with no loss of useful information, where {𝐱ℜ , 𝐱ℑ } ∈ √R𝑁∕2 . From (3), it is
1 In the sequel, we use the subcarrier and subchannel for O-OFDMA and O-SCFDMA, inferred that 𝑀-QAM (square) constellation is split into two 𝑀-PAM
respectively, to distinguish between the two systems. constellations.

178
A.W. Azim et al. Optics Communications 415 (2018) 177–185

Fig. 1. Block diagram of HSFO-SCFDMA transmitter.

Fig. 2. Block diagram of HSFO-SCFDMA receiver.

A cyclic prefix (CP) with a length (𝑁CP ) equal to the number of as Lambertian [25]. VLC channel response comprises both the line-of-
channel taps is affixed to both sub-blocks. For conciseness in notation, sight (LOS) and reflected diffuse components. Herewith, we obtain the
after the addition of CP, the sub-blocks are still expressed as 𝐱ℜ and multipath channel coefficients, ℎchan (𝑡), using the ray tracing recursive
[ ]T
𝐱ℑ . 𝐱 = 𝐱ℜ 𝐱ℑ ∈ R𝑁+2𝑁CP is obtained by concatenating 𝐱ℜ and 𝐱ℑ , algorithm for the indoor VLC studied by Lee et al. [26].
which is impinged on a DAC to obtain an analog electrical waveform, Sampling time (𝑇s ) of 1 ns and up to 4 diffused reflections are
𝐱(𝑡), which, however, is bipolar. Therefore, a bias, 𝛽dc > 0, is introduced, considered. It is assumed that the transmitter and the receiver are
such that perfectly synchronized, such that the channel is tapped from the time
of arrival of LOS signal [27]. Readers are directed to [26] for a detailed
𝐱𝛽 (𝑡) = 𝐱(𝑡) + 𝛽dc ∈ R+ , (4) analysis on VLC channel modeling.
where 𝛽dc is prescribed as
2.4. Receiver
√
𝛽dc = 𝛼 𝐸[𝐱2 (𝑡)], 𝛼 > 0. (5)
The receiver of HSFO-SCFDMA is illustrated in Fig. 2. At the receiver,
The bias-index on the decibel (dB) scale is defined as 10 log10 (𝛼 2 + 1)
dB. the light intensity is photo-detected using a PD, and the bias is elimi-
We follow sufficient biasing [2] for HSFO-SCFDMA, where 𝛼 is adjusted, nated using a DC blocking capacitor. The received intensity waveforms,
such that 𝛽dc is equal to the absolute value of the negative peak of 𝐱(𝑡). 𝐲(𝑡) is electronically amplified using a trans-impedance amplifier (TIA),
[ ]T
𝐱𝛽 (𝑡) is transmitted through an optical channel by the LED. and then fed to an ADC to yield 𝐲 = 𝐲ℜ 𝐲ℑ , from which the real and
In what follows, we assume perfect synchronization [8,9,22]. A the imaginary sub-blocks are separated and are given by
linear response of the LED is considered in system evaluation [8,9,22], 𝐲ℜ = 𝐇𝐱ℜ + 𝐰ℜ ∈ R𝑁∕2 ,
since, nonlinearity of the LED can be mitigated using digital pre- (8)
distortion [23]. 𝐲ℑ = 𝐇𝐱ℑ + 𝐰ℑ ∈ R𝑁∕2 ,
respectively, where 𝐰ℜ and 𝐰ℑ are modeled as AWGN. 𝐇 is 𝑁∕2 ×
2.3. Channel modeling 𝑁∕2 circulant convolution matrix with the first column equal to 𝐡 =
[ ]
ℎ0 , ℎ1 , … , ℎ𝐿−1 , 0, … , 0 for 𝐿 channel coefficients and ℎ𝑛 = ℎ(𝑛𝑇s ) with
The impulse response of the dispersive channel culminating the 𝑛 ∈ [0, 𝐿 − 1]. We assume a stationary channel, i.e., 𝐇 = 𝐇ℜ = 𝐇ℑ .
impact of bandwidth limitation of the LED/LED driver combination and Moreover, the circulant nature of 𝐇, allows to diagonalize it using the
multipath VLC channel is given as DFT matrix and write

ℎ(𝑡) = ℎLED (𝑡) ⊗ ℎchan (𝑡), (6) 𝐇 = 𝐅H 𝜦 𝐅

𝑁∕2 𝑁∕2 𝑁∕2
, (9)

where 𝜦 is a diagonal matrix with the channel frequency response on

where ℎLED (𝑡) and ℎchan (𝑡) are the impulse response of the LED and the [ ]T
multipath VLC channel, respectively. ℎLED (𝑡) is modeled as a Gaussian its diagonal and having eigen values 𝛬0 , 𝛬1 , … , 𝛬𝑁∕2 = 𝐅𝑁∕2 𝐡.
low-pass filter with 3 dB optical cut-off frequency, 𝑓3dB , having a transfer The sub-blocks, 𝐲ℜ and 𝐲ℑ , are transformed to FD by 𝑁∕2-order DFT
function of [24] to yield
( )
( )2 𝐘ℜ = DFT 𝐲ℜ = 𝐅𝑁∕2 ⋅ 𝐲ℜ ∈ C𝑁∕2 ,
𝑓
− ln(2) 𝑓
𝐻LED (𝑓 ) = 𝑒 3dB . (7) ( ) (10)
𝐘ℑ = DFT 𝐲ℑ = 𝐅𝑁∕2 ⋅ 𝐲ℑ ∈ C𝑁∕2 .
(7) implies that even if an ideal channel exists between the LED and
A FD minimum mean squared error (MMSE) equalizer with the knowl-
the PD, the received signal would be a filtered version of the transmitted
edge of channel impulse response is applied to 𝐘ℜ and 𝐘ℑ to yield
signal. The dispersive attributes of multipath VLC channel are taken into
equalized FD sub-blocks as
account by considering indoor environments where the received optical ( )
signal endures time dispersion from the objects inside the room. If the 𝐘̃ ℜ = 𝜦H 𝜦 + (𝐸b(elec) ∕𝑁0 )−1 𝐈 𝜦H 𝐘ℜ ∈ C𝑁∕2 ,
indoor fixtures of the room are deemed purely diffusive, the reflections ( ) (11)
from different objects are also diffusive and can be adequately modeled 𝐘̃ ℑ = 𝜦H 𝜦 + (𝐸b(elec) ∕𝑁0 )−1 𝐈 𝜦H 𝐘ℑ ∈ C𝑁∕2 ,

179
A.W. Azim et al. Optics Communications 415 (2018) 177–185

where 𝐸b(elec) ∕𝑁0 is the electrical signal-to-noise ratio (SNR) per bit, and
𝐈 denotes the identity matrix. Afterwards, 𝐘̃ ℜ and 𝐘̃ ℑ are transformed
to TD counterparts by using 𝑁∕2-order IDFT as
( )
𝐲̃ ℜ = IDFT 𝐘̃ ℜ = 𝐅𝑁∕2 ⋅ 𝐘̃ ℜ ∈ R𝑁∕2 ,
( ) (12)
𝐲̃ ℑ = IDFT 𝐘̃ ℑ = 𝐅𝑁∕2 ⋅ 𝐘̃ ℑ ∈ R𝑁∕2 .
Lastly, 𝐲̃ ℜ and 𝐲̃ ℑ are combined as
√ ( )
𝐲= 𝜅 𝐲̃ ℜ + 𝑗 ⋅ 𝐲̃ ℑ ∈ C𝑁∕2 , (13)
where 𝜅 = 2, if only one user is considered. 𝑀-ary QAM demapping is
performed on the serial symbols to obtain the output bits.

3. Statistical characterization

By adopting sufficient biasing, the average electrical and the average

optical powers for HSFO-SCFDMA are given by
[ ] [ ]
𝑃elec = 𝐸 𝐱𝛽2 (𝑡) = 𝐸 𝐱2 (𝑡) + 𝛽dc
2
, (14)
Fig. 3. PAPR comparison of different modulation schemes for 𝜂 = 1 bits/s/Hz.
and
[ ]
𝑃opt = 𝐸 𝐱𝛽 (𝑡) = 𝛽dc , (15)
respectively. The ratio of 𝑃elec and 𝑃opt can be expressed as
𝑃elec
𝛼oe = , (16)
𝑃opt
where 𝛼oe is a measure of electrical to optical power conversion
efficiency. For any value of 𝐸b(elec) ∕𝑁0 , the corresponding 𝐸b(opt) ∕𝑁0
is obtained using (16) and normalizing the optical power to unity
(i.e., 𝑃opt = 1) as
𝐸b(opt) ( )𝐸
1 b(elec)
= . (17)
𝑁0 𝛼oe 𝑁0

4. Performance evaluation

This section presents simulation and analytical results to validate

and compare the performance of HSFO-SCFDMA with other schemes.
Firstly, we examine the spectral efficiencies. Secondly, we study the
PAPR characteristics, where closed-form analytical PAPR expressions
Fig. 4. PAPR comparison of different modulation schemes for 𝜂 = 2 bits/s/Hz.
for the techniques having deterministic peak values of the TD signal
are evaluated. Thereafter, we compute Monte Carlo results for BER.
Subsequently, optical power penalty relative to on–off keying (OOK)
is analyzed. Further on, we compute the system complexities. Lastly, 4.2. PAPR analysis
we analyze quantization characteristics and electrical power efficiency
considering an AWGN channel.
PAPR is the measure of variation of the TD signal about its mean.
4.1. Spectral efficiency Mathematically, PAPR of an arbitrary 𝐿-order signal, 𝑧𝑛 is expressed as
( )
max0≤𝑛≤𝐿−1 ||𝑧𝑛 ||
2
Spectral efficiency is defined as the number of bits relative to
PAPR = 𝜉 ≜ ( ) . (18)
modulation index times the ratio of data-carrying subchannels to the 𝐸 ||𝑧𝑛 ||
2
total number of available subchannels. Here, we evaluate the spectral
efficiencies of different schemes relative to that of ACO- and DCO- PAPR of a signal is graphically illustrated using complementary
OFDMA. It is recalled that ACO-OFDMA have half the spectral efficiency cumulative distribution function (CCDF) which is the probability that
of DCO-OFDMA. PAPR of signal will exceed a given threshold, PAPR𝜖 , i.e., CCDF =
∙ Both ACO- and DCO-SCFDMA have the same spectral efficiency Prob(PAPR > PAPR𝜖 ).
as that of their O-OFDMA counterparts. CCDF curves are illustrated in Figs. 3 and 4 for spectral efficiencies
∙ DHTS-ACO-OFDM and SCO-FDM manifest same spectral effi- of 𝜂 = 1 bits/s/Hz and 𝜂 = 2 bits/s/Hz, respectively, which reveal that
ciency as ACO-OFDMA. HSFO-SCFDMA distinctly has the lowest PAPR. For 𝜂 = 1 bits/s/Hz,
∙ HSFO-SCFDMA exhibits the same spectral efficiency as DCO- HSFO-SCFDMA manifests approximately 5.4 dB, 8.2 dB, 8.7 dB and
OFDMA, therefore, it is spectrally efficient compared to ACO- 12 dB lesser PAPR compared to SCO-FDM, DCO-SCFDMA, DHTS-ACO-
SCFDMA, DHTS-ACO-OFDM, and SCO-FDM. OFDM and ACO-SCFDMA, respectively. Likewise, for 𝜂 = 2 bits/s/Hz,
It should be recognized that 𝑀 2 -QAM ACO-SCFDMA and SCO-FDM the PAPR of HSFO-SCFDMA is approximately 4.8 dB, 6.4 dB, 7.5 dB
would result in a same spectral efficiency as 𝑀-QAM DCO- and HSFO- and 9.4 dB less than SCO-FDM, DCO-SCFDMA, DHTS-ACO-OFDM and
SCFDMA. Moreover, 𝑀 ′ -PAM based DHTS-ACO-OFDM and 𝑀-QAM ACO-SCFDMA, respectively. The reduced PAPR of HSFO-SCFDMA is
based DCO- and HSFO-SCFDMA exhibit same spectral efficiency, with translated into a power gain, as higher modulation power can be
𝑀 ′ = 𝑀. attained after the DAC.

180
A.W. Azim et al. Optics Communications 415 (2018) 177–185

The closed-form PAPR expressions as a function of modulation index

can be readily evaluated for HSFO-SCFDMA, SCO-FDM and DHTS-ACO-
OFDM, because the peak values of the TD signals are deterministic. Be-
sides, the variance of the TD signal can be computed in a straightforward
manner. The analytical PAPR expressions are evaluated considering
𝑀 = {4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048} for HSFO-SCFDMA and
SCO-FDM, whereas, a generalized expression considering 𝑀 ′ -PAM is
obtained for DHTS-ACO-OFDM.
For HSFO-SCFDMA, the peak value, 𝜆peak,HSFO , and the variance,
2
𝜎HSFO , of the TD signal are given by
√
⎧ 𝑀 −1
⎪ √ , for 𝑀 = 4, 16, 64, 256, 1024
⎪ 2
⎪
⎪ log√
2 (𝑀)
, for 𝑀 = 8, 32
⎪
⎪ 2
Fig. 5. BER performance as a function of 𝐸b(elec) ∕𝑁0 in a dispersive channel considering a
⎪ 2 log2 (𝑀) − 3
𝜆peak,HSFO = ⎨ √ , for 𝑀 = 128 (19) spectral efficiency of 𝜂 = 1 bits/s/Hz. 𝑅b = 200 Mbps and 𝑓3dB = 150 MHz are considered.
⎪ 2 The bias-index for DCO-SCFDMA is 6 dB, while HSFO-SCFDMA is sufficiently biased. The
⎪ 2 log (𝑀) + 5 dashed curves represent the performance of different modulation schemes in an AWGN
⎪ 2
√ , for 𝑀 = 512 channel.
⎪ 2
⎪
⎪ 4 log2 (𝑀) + 3
⎪ √ , for 𝑀 = 2048,
⎩ 2
and
⎧𝑀 − 1, for 𝑀 = 4, 16, 64, 256, 1024
⎪ 6
2 ⎪ 2.5𝑀 − 2
𝜎HSFO =⎨ , for 𝑀 = 8 (20)
⎪ 12
⎪ 31𝑀 − 32 , for 𝑀 ≥ 32,
⎩ 192
respectively. Using (19) and (20), closed-form expressions for the PAPR
of HSFO-SCFDMA are evaluated as 𝜉HSFO = 𝜆2peak,HSFO ∕𝜎HSFO
2 . For SCO-
√
2 2
FDM, 𝜆peak,SCO = (1∕ 2)𝜆peak,HSFO and 𝜎SCO = (1∕4)𝜎HSFO . Hence, we
have

𝜉SCO = 2𝜉HSFO . (21)

where, 𝜉SCO is the PAPR exhibited by SCO-FDM. Moreover, for DHTS- Fig. 6. BER performance as a function of 𝐸b(elec) ∕𝑁0 in a dispersive channel considering a
2
ACO-OFDM, 𝜆peak,DHTS−ACO ≈ 𝑀 ′ − 1, and 𝜎DHTS−ACO = (𝑀 ′2 − 1)∕12, spectral efficiency of 𝜂 = 2 bits/s/Hz. 𝑅b = 200 Mbps and 𝑓3dB = 150 MHz are considered.
The bias-index of DCO-SCFDMA is 10 dB, while HSFO-SCFDMA is sufficiently biased.
with 𝑀 ′ being the PAM alphabet size. So, using 𝜆peak,DHTS−ACO and
2
𝜎peak,DHTS−ACO , the PAPR manifested by DHTS-ACO-OFDM, 𝜉DHTS−ACO ,
is given as
averaged over 2000 Monte Carlo runs with the number of subchannels,
𝜆2peak,DHTS−ACO 12(𝑀 ′ − 1)
𝜉DHTS−ACO = = . (22) 𝑁 = 512.
2
𝜎DHTS−ACO 𝑀′ + 1 The results for BER versus 𝐸b(elec) ∕𝑁0 for 𝜂 = 1 bits/s/Hz and 𝜂 =
2 bits/s/Hz are presented in Figs. 5 and 6, respectively, from which we
For brevity, we have skipped the evaluation of PAPR expressions
discern the following
for the remaining approaches. However, interested readers are referred
to [28] for a comprehensive analysis of PAPR of ACO- and DCO-OFDM, ∙ The 𝐸b(elec) ∕𝑁0 required to achieve BER of 10−3 for ACO-
which may be extended to O-SCFDMA counterparts. SCFDMA, DHTS-ACO-OFDM, SCO-FDM, and DCO-SCFDMA is al-
most the same for 𝜂 = 1 bits/s/Hz. However, for 𝜂 = 2 bits/s/Hz,
4.3. Bit error rate performance ACO-SCFDMA, DHTS-ACO-OFDM, and SCO-FDM become less
efficient compared to DCO-SCFDMA, as higher order modulation
Here, we investigate BER performance of HSFO-SCFDMA and com- alphabets are required.
pare it with that of ACO-SCFDMA, DHTS-ACO-OFDM, SCO-FDM, and
∙ For both 𝜂 = {1, 2} bits/s/Hz, HSFO-SCFDMA signifies a superior
DCO-SCFDMA considering 𝜂 = {1, 2} bits/s/Hz. We consider a dispersive
performance compared to other alternatives. It may be noticed
channel with 𝑓3dB set at 150 MHz. Noise is modeled as AWGN. For DCO- that for 𝜂 = 1 bits/s/Hz at BER of 10−3 , 𝐸b(elec) ∕𝑁0 gain of
SCFDMA, bias-index of 6 dB and 10 dB is used for 𝜂 = 1 bits/s/Hz and approximately 2 dB over ACO-SCFDMA and DHTS-ACO-OFDM,
𝜂 = 2 bits/s/Hz, respectively, whereas, HSFO-SCFDMA is sufficiently and around 3 dB over DCO-SCFDMA and SCO-FDM is observed
biased. Data-rate, 𝑅b equal to 200 Mbps is considered. Thus, for 𝜂 = 1 for HSFO-SCFDMA. In case of 𝜂 = 2 bits/s/Hz, for BER of 10−3 ,
bits/s/Hz, the bandwidth (BW) of the transmitted signal is 200 MHz a gain of almost 5 dB over DCO-SCFDMA, and nearly 6 dB over
culminating in a prominent impact of the bandwidth limitation of the ACO-SCFDMA, DHTS-ACO-OFDM, and SCO-FDM is achieved for
LED/LED driver combination. Whereas, for 𝜂 = 2 bits/s/Hz, BW of the HSFO-SCFDMA.
transmitted signal is 100 MHz, therefore, evading the bandwidth limita-
tion of the LED/LED driver combination. As a reference, BER of HSFO-, ∙ Comparing HSFO-SCFDMA with DCO-SCFDMA, the 𝐸b(elec) ∕𝑁0
gain is correlated to the difference in bias required to achieve
DCO- (with bias-index of 6 dB) and ACO-SCFDMA in AWGN without
non-negativity for both approaches. Note that, a lower bias is
considering bandwidth limitation of LED/LED driver combination for
required for HSFO-SCFDMA because of lower PAPR.
𝜂 = 1 bits/s/Hz is presented. Unless otherwise specified, all results are

181
A.W. Azim et al. Optics Communications 415 (2018) 177–185

∙ The 𝐸b(elec) ∕𝑁0 gain for HSFO-SCFDMA over other approaches

(apart from DCO-SCFDMA) is realized because lower order con-
stellations are needed to achieve a given spectral efficiency.
∙ If 𝑅b = 200 Mbps is considered for 𝜂 = 2 bits/s/Hz, the BER
performance of all the schemes approaches to their performance
in AWGN because the dispersive impact of the VLC multipath
channel and the bandwidth limitation of the LED/LED driver
combination is averted due to the low bandwidth of the trans-
mitted signal.

4.4. Optical power penalty

In this section, the optical power penalty for different modulation

approaches incurred due to the dispersive channel is analyzed. For a
given BER, 𝑃b , the optical power penalty is obtained by normalizing Fig. 7. Optical power penalty as a function of 𝑅b ∕𝑓3dB in a dispersive channel. Spectral
the required optical power by the average optical power required for efficiency of 𝜂 = 2 bits/s/Hz and 𝑓3dB = 150 MHz are considered. The bias-index for
OOK in an AWGN channel with no bandwidth limitation expressed as DCO-SCFDMA is 10 dB, while HSFO-SCFDMA is sufficiently biased.
𝐸bOOK ∕𝑁0 . 𝐸bOOK ∕𝑁0 to achieve 𝑃b is given as
(opt) (opt)
Table 1
( ) Computational complexity of different modulation techniques.
𝐸bOOK ∕𝑁0 = erf c−2 2𝑃b , (23)
(opt)
Modulation scheme Complexity
where erf c(𝜙) is√ the complementary error function expressed as HSFO-SCFDMA 4𝑁 log2 (𝑁) + 10𝑁 log2 (𝑁∕2)
𝜙
erf c(𝜙) = 1 − 2∕ 𝜋 ∫0 exp(−𝑡2 )d𝑡. The results are obtained by setting ACO-SCFDMA 8𝑁 log2 (𝑁) + 2𝑁 log2 (𝑁∕4)
𝑃b = 10−3 , 𝜂 = 2 bits/s/Hz and varying the ratio of the data-rate to the DHTS-ACO-OFDM 4𝑁log2 (𝑁) + 2𝑁 log2 (𝑁∕2)
3 dB optical cut-off frequency, i.e., 𝑅b ∕𝑓3dB . The result is illustrated in SCO-FDM 4𝑁 log2 (𝑁) + 6𝑁 log2 (𝑁∕4)
DCO-SCFDMA 8𝑁 log2 (𝑁) + (4𝑁 − 8) log2 (𝑁∕2 − 1)
Fig. 7, from which following observations are drawn

∙ DCO-SCFDMA experiences the highest optical power penalty

because of the bias required to achieve non-negativity. ∙ For ACO-SCFDMA, 𝑁∕4-order FFT and 𝑁-order IFFT are re-
quired at the transmitter, whereas, at the receiver, FFT/IFFT are
∙ The optical power penalty for SCO-FDM is marginally higher
inverted, i.e., 𝑁-order FFT and 𝑁∕4-order IFFT are needed.
compared to ACO-SCFDMA and DHTS-ACO-OFDM.
∙ In DHTS-ACO-OFDM, real constellations along with DHT are
∙ DHTS-ACO-OFDM undergoes approximately the same optical
used. 𝑁-order DHT approximately requires 2𝑁log2 (𝑁) arith-
power penalty as that of ACO-SCFDMA for 𝑅b ∕𝑓3dB ≤ 1.90.
metic operations [31]. At the transmitter, 𝑁∕2-order DHT and
However, for 𝑅b ∕𝑓3dB > 1.90, the optical power penalty increases
𝑁-order IDHT are performed. Additionally, at the receiver, 𝑁-
substantially if high data-rates are used.
order DHT and 𝑁∕2-order IDHT are taken.
∙ HSFO-SCFDMA experiences minimum optical power penalty for
all the values of 𝑅b ∕𝑓3dB . For 𝑅b ∕𝑓3dB ≤ 1.90, DCO-SCFDMA and ∙ For SCO-FDM, at the transmitter, 𝑁∕4-order FFT and 𝑁-order
SCO-FDM, respectively, perform approximately 6 dB and 3.5 dB IFFT operations are required, whereas, at the receiver, four 𝑁∕4-
worse compared to HSFO-SCFDMA, whereas, ACO-SCFDMA and order FFT and one 𝑁∕4-order IFFT are needed.
DHTS-ACO-OFDM are penalized 3 dB more than HSFO-SCFDMA. ∙ For DCO-SCFDMA, 𝑁∕2 − 1-order FFT and 𝑁-order IFFT are
required at the transmitter, while 𝑁-order FFT and 𝑁∕2−1-order
IFFT are taken at the receiver.
4.5. Complexity analysis
The system complexities are compiled in Table 1 and plotted as a
In this section, we analyze the system complexities of various function of input block size, 𝑁 scaled by 𝜂, in Fig. 8. Scaling is performed
modulation schemes. The complexity analysis in terms of probability to ensure a fair comparison, e.g., for a given 𝑅b , if HSFO-SCFDMA
of convergence, theoretical run-time boundaries etc. is cumbersome. requires an input block size of 𝑁, ACO-SCFDMA (or a scheme with
Hence, system complexity is calculated in terms of a total number of similar spectral efficiency) would require a block size of 2𝑁. Fig. 8
required arithmetic operations at the transceiver. We consider that a reveals that DCO-SCFDMA and DHTS-ACO-OFDM manifest the least
block of 𝑁 data symbols is taken as an input by the transceivers and system complexity, whereas, HSFO-SCFDMA is less complex than ACO-
is equal to DFT/IDFT order. For an efficient implementation, DFT and SCFDMA and SCO-FDM. We highlight that even though the system
IDFT are implemented utilizing fast Fourier transform (FFT) and inverse complexity of HSFO-SCFDMA is marginally higher than DCO-SCFDMA
FFT (IFFT) algorithms, respectively. 𝑁-order FFT/IFFT approximately and DHTS-ACO-OFDM, yet, HSFO-SCFDMA demonstrates superior per-
requires 4𝑁log2 (𝑁) arithmetic operations [29]. For both O-OFDMA formance compared to DCO-SCFDMA and DHTS-ACO-OFDM in other
and O-SCFDMA systems, the complexity is due to the FFT/IFFT and performance parameters.
equalization. Due to the diagonal nature of 𝜦, equalization is realized
in (𝑁) operations [30], thus, in the sequel, we only consider the 4.6. Quantization characteristics
complexity incurred due to FFT/IFFT. The complexities of different
modulation approaches are: In this subsection, we investigate the impact of quantization by
evaluating the effective number of bits required to achieve a target BER.
∙ For HSFO-SCFDMA, 𝑁∕2-order FFT and 𝑁-order IFFT are per-
Following [32], the required electrical SNR in AWGN, 𝛤(elec) , for the
formed at the transmitter, whereas, at the receiver, two 𝑁∕2-
target BER, 𝑃b , is evaluated, using which the effective number of bits are
order FFT and two 𝑁∕2-order IFFT are required.
calculated. Target BER of 10−3 and uniform quantization are considered.

182
A.W. Azim et al. Optics Communications 415 (2018) 177–185

Fig. 8. Complexity comparison of different modulation schemes. The block size 𝑁 is Fig. 9. Effective number of bits required for quantization by the transmitter for different
normalized with the spectral efficiency, 𝜂 to ensure same data-rate. modulation schemes.

Table 2
Effective number of bits required at the transmitter and the receiver for different modu-
lation schemes. ⌈⋅⌉ represents ceiling operation to nearest integer. 𝛾HSFO and 𝛾SCO , respec-
tively, are the number of distinct levels of TD signal for HSFO-SCFDMA and SCO-FDM. 𝛼
is the bias coefficient for DCO-SCFDMA. 𝜎 2 represents the transmitted signal power for the
respective modulation scheme. 𝜆peak for ACO-SCFDMA has been evaluated after clipping.
Transmitter
Modulation scheme Effective number of bits
HSFO-SCFDMA ⌈log2 [𝛾HSFO ]⌉
⌈ [ 2 ]⌉
𝜆peak
ACO-SCFDMA 0.5 log2 6𝜎 2
𝛤(elec)
⌈ [ 2 𝑀 ′ −1 ]⌉
( )
DHTS-ACO-OFDM 0.5 log2 (𝑀 ′ +1) 𝛤(elec)

SCO-FDM ⌈log2 [𝛾SCO ]⌉

⌈ [ 2 2
]⌉
4𝜆peak +(𝜆peak +𝛼𝜎 )
DCO-SCFDMA 0.5 log2 12𝜎 2
𝛤(elec)

Receiver
Fig. 10. Effective number of bits required for quantization by the receiver for different
Modulation scheme Effective number of bits modulation schemes.
⌈ [ 2 ]⌉
𝜆peak
HSFO-SCFDMA 0.5 log2 3𝜎 2
𝛤(elec)
⌈ [ 2 ]⌉
𝜆peak
SCO-FDM 0.5 log2 12𝜎 2
𝛤(elec)
4.7. Power efficiency

Fig. 11 depicts how ⟨𝐸b(elec) ∕𝑁0 ⟩ (the required 𝐸b(elec) ∕𝑁0 to achieve
The quantization noise is modeled as an additive, uniformly distributed a BER of 10−3 ) varies with 𝜂 in an AWGN channel. For HSFO- and
white noise [32]. DCO-SCFDMA, 𝜂 between [1, 6.5] bits/s/Hz is considered, whereas, for
As 𝜆peak , for HSFO-SCFDMA, DHTS-ACO-OFDM, and SCO-FDM are ACO-SCFDMA, SCO-FDM, 𝜂 between [0.5, 3.25] bits/s/Hz is taken into
deterministic (see Section 4.2), calculating the effective number of bits account. Note that, 𝜂 = 3.25 bits/s/Hz (for ACO-SCFDMA and SCO-FDM)
is rather straightforward. Whereas, for remaining approaches, Monte corresponds to 8192-QAM, hence, evaluating the power efficiencies for
Carlo averaging is needed to obtain an estimate of 𝜆peak . At the transmit- 𝜂 > 3.25 bits/s/Hz might not be meaningful. For DHTS-ACO-OFDM, 𝜂
ter, for HSFO-SCFDMA and SCO-FDM, the TD signal manifests distinct between [0.5, 2.5] bits/s/Hz is investigated. Besides, for DCO-SCFDMA,
number of levels as a function of the modulation index, 𝑀 given as 𝛾HSFO the results are obtained by recognizing 6 dB bias-index for 𝜂 = 1
and 𝛾SCO . This simplifies the evaluation of the required effective number bits/s/Hz, 7.5 dB for 𝜂 = 1.5 bits/s/Hz and a bias-index of 10 dB for
of bits for the transmitter of HSFO-SCFDMA and SCO-FDM which are 𝜂 ≥ 2 bits/s/Hz.
evaluated using 𝛾HSFO and 𝛾SCO , therefore, does not depend on 𝛤(elec) . For It is demonstrated that HSFO-SCFDMA attains superior performance
simplicity, we assume same effective number of bits at the transmitter compared to other alternatives, e.g., for 𝜂 = 2 bits/s/Hz, a power gain of
and at the receiver [32]. However, it should be recognized that for almost 5.2 dB over DCO-SCFDMA, and approximately 8 dB over ACO-
HSFO-SCFDMA and SCO-FDM, the effective number of bits required at SCFDMA, DHTS-ACO-OFDM and SCO-FDM is observed. Moreover, for
the transmitter and at the receiver differ because the received signal smaller values of 𝜂, DCO-SCFDMA signifies the same power requirement
would be analog and suffer from the AWGN noise, consequently, the as that of ACO-SCFDMA, DHTS-ACO-OFDM, and SCO-FDM. However,
bits are evaluated using 𝛤(elec) at the receiver. for larger values of 𝜂, DCO-SCFDMA becomes more power efficient com-
Table 2 presents the analytical formulas to compute the effective pared to other approaches apart from HSFO-SCFDMA. Similar trends are
number of bits at the transmitter and receiver to achieve a BER of 10−3 . foreseen if optical power efficiency is computed.
Graphical illustrations are provided in Figs. 9 and 10 which show that
for a given 𝜂, the effective number of bits required by the transmitter or 5. Multiple access
the receiver of HSFO-SCFDMA are less compared to other counterparts.
From the results, we can perceive that the bit resolution requirement In this section, we explain the MA capability of HSFO-SCFDMA. LED
of both the ADC and DAC can be relaxed if HSFO-SCFDMA is adopted, lighting source comprising of LED array, 𝐾 users and 𝑁 subchannels
consequently, the overall cost of the system can be reduced. are considered. All the users use same 𝑀-ary QAM constellations and

183
A.W. Azim et al. Optics Communications 415 (2018) 177–185

Fig. 12. Illustration of subchannel mapping for two different users.

2. HSFO-SCFDMA manifests significantly less PAPR which results in

Fig. 11. Analysis of electrical power efficiency of different modulation techniques. The
power efficiency through a limited bit resolution DAC, and less
average 𝐸𝑏(elec) ∕𝑁0 for a BER of 10−3 represented as ⟨𝐸𝑏(elec) ∕𝑁0 ⟩ has been evaluated for
different spectral efficiencies, 𝜂.
sensitivity to any non-linear impairments. Moreover, the power
efficiency could be vital to enhance the battery life if considered
for user terminal.
3. As expected from first and second conclusion, it is demonstrated
have the feasibility to modulate 𝑁∕2𝐾 subchannels. For 𝑘th user, the that the bit resolution requirement of both DAC and ADC can be
𝐾
TD symbol vector is 𝐬𝑘 ∈ C𝑁∕2 , and the corresponding DFT-precoded relaxed for HSFO-SCFDMA, consequently, the overall cost of the
FD symbol vector is obtained by using 𝑁∕2𝐾 -order DFT as system can be reduced.
( ) 𝐾 4. HSFO-SCFDMA exhibits superior BER performance compared to
𝐒𝑘 = DFT 𝐬𝑘 = 𝐅𝑁∕2𝐾 ⋅ 𝐬𝑘 ∈ C𝑁∕2 . (24)
other alternatives considering a dispersive channel.
From (24), it is inferred that for 𝐾 users, 𝑁∕2𝐾 -order DFT-precoding 5. HSFO-SCFDMA undergoes the least optical power penalty and
is required. Subchannel mapping is performed such that each user is achieves the lowest optical power penalty floor.
allocated unique subchannels. For 𝑘th user, the FD precoded symbols 6. HSFO-SCFDMA is less complex than ACO-SCFDMA and SCO-
from 𝐒𝑘 are mapped to 𝐒̃ 𝑘 ∈ C𝑁 on indexes 2(𝑘 − 1) ∶ 2𝐾 − 1 ∶ 𝑁, with FDM. However, it is more complex compared to DCO-SCFDMA
2𝐾 − 1 zeros padded between the two adjacent modulated subchannels. and DHTS-ACO-OFDM, but surpasses them in other performance
The TD counterparts of 𝐒̃ 𝑘 ∀ 𝑘 = 1, … , 𝐾 are transmitted using different parameters.
LEDs in the array as depicted in Fig. 12 for a system with 𝐾 = 2. 𝑁-order 7. HSFO-SCFDMA is the most power efficient approach in terms of
IDFT is applied to 𝐒̃ 𝑘 to obtain electrical power dissipation to achieve BER of 10−3 .
( )
𝐬̃ 𝑘 = IDFT 𝐒̃ 𝑘 = 𝐅H ̃ 𝑁
𝑁 ⋅ 𝐒𝑘 ∈ C . (25) References
𝐬̃ 𝑘 exhibits a half-wave symmetry, from which, the real and imaginary
[1] S. Dimitrov, S. Sinanovic, H. Hass, Clipping noise in OFDM-based optical wireless
sub-blocks for 𝑘th user are obtained as communication systems, IEEE Trans. Commun. 60 (4) (2012) 1072–1081.
[ ] [ ] [2] J.B. Carruthers, J.M. Kahn, Multiple-subcarrier modulation for nondirected wireless
(𝑛+𝑁∕2)
𝐱𝑘,ℜ = ℜ s̃(𝑛)𝑘
, 𝐱𝑘,ℑ = ℑ s̃𝑘 , (26) infrared communication, IEEE J. Sel. Areas Commun. 14 (3) (1996) 538–546.
[3] J. Armstrong, A.J. Lower, Power efficient optical OFDM, Electron. Lett. 42 (6) (2006)
∀ 𝑛 = 0, 1, … , 𝑁∕2 − 1, respectively, where {𝐱𝑘,ℜ , 𝐱𝑘,ℑ } ∈ R𝑁∕2 ∀ 𝑘 = 370–372.
1, … , 𝐾. The IM consistent signal [𝐱𝑘,ℜ 𝐱𝑘,ℑ ]T is transmitted through [4] J. Armstrong, OFDM for optical communications, J. Lightwave Technol. 27 (3)
LEDs. (2009) 189–204.
The receiver is exactly the same as presented in Fig. 2. At the [5] H. Elgala, R. Mesleh, H. Haas, Indoor optical wireless communication: potential and
state-of-the-art, IEEE Commun. Mag. 49 (9) (2011).
receiver, a signal containing the sum of all the users is obtained. After
[6] N. Fernando, Y. Hong, E. Viterbo, Flip-OFDM for unipolar communication systems,
processing, the received symbols for different users are distinguished by IEEE Trans. Commun. 60 (12) (2012) 3726–3733.
choosing the appropriate subchannels. Note that, the received symbol [7] F. Barrami, Y. Le Guennec, E. Novakov, J.-M. Duchamp, and P. Busson. novel
vector would always be 𝑁∕2-order, from which the allocated subchan- FFT/IFFT size efficient technique to generate real time optical OFDM signals
nels for 𝑘th user are obtained as (𝑘 − 1) ∶ 𝐾 ∶ 𝑁∕2. Moreover, 𝜅 also compatible with IM/DD systems in: Euro. Microw. Conf., 2013, pp. 1247–1250.
depends on the number of users in the system, thus, if 𝐾 users are [8] A.W. Azim, Y. Le Guennec, G. Maury, Decision-directed iterative methods for PAPR
reduction in optical wireless OFDM systems, Opt. Commun. 389 (2017) 318–330.
considered, the value of 𝜅 would be equal to 2𝐾 .
[9] W.O. Popoola, Z. Ghassemlooy, B.G. Stewart, Pilot-assisted PAPR reduction tech-
From the presented analysis, it can be ascertained that HSFO- nique for optical OFDM communication systems, J. Lightwave Technol. 32 (7)
SCFDMA can be used in a VLC down-link scenario, for which each LED (2014) 1374–1382.
in the array transmits the data/information for a different user. [10] C.R. Berger, Y. Benlachtar, R.I. Killey, P.A. Milder, Theoretical and experimental
evaluation of clipping and quantization noise for optical OFDM, Opt. Express 19 (18)
6. Conclusions (2011) 17713–17728.
[11] T. Mao, Z. Wang, Q. Wang, L. Dai, Ellipse-based DCO-OFDM for visible light
communications, Opt. Commun. 360 (2016) 1–6.
In this article, we study O-SCFDMA as an alternate to O-OFDMA. [12] H. Zhang, Y. Yuan, W. Xu, PAPR reduction for DCO-OFDM visible light communi-
From the study of the proposed HSFO-SCFDMA, following conclusions cations via semidefinite relaxation, IEEE Photonics Technol. Lett. 26 (17) (2014)
are drawn: 1718–1721.
[13] H. Marshoud, S. Muhaidat, P.C. Sofotasios, S. Hussain, M.A. Imran, B.S. Sharif,
1. HSFO-SCFDMA possesses same spectral efficiency as DCO- Optical non-orthogonal multiple access for visible light communication. arXiv
SCFDMA, whereas, the spectral efficiencies of ACO-SCFDMA, preprint arXiv:1704.07621, 2017.
[14] D. Tsonev, S. Videv, H. Haas, Light fidelity (Li-Fi): towards all-optical networking,
DHTS-ACO-FDM and SCO-FDM are half of HSFO-SCFDMA. Ac-
in: SPIE OPTO, vol. 9007, 2013, p. 900702.
cordingly, same spectral efficiency is obtained by employing [15] R. Mesleh, H. Elgala, H. Haas, LED nonlinearity mitigation techniques in optical
higher modulation alphabets for ACO-SCFDMA, DHTS-ACO-FDM wireless OFDM communication systems, J. Opt. Commun. Netw. 4 (11) (2012) 865–
and SCO-FDM. 875.

184
A.W. Azim et al. Optics Communications 415 (2018) 177–185

[16] C. Wu, H. Zhang, W. Xu, On visible light communication using LED array with DFT- [24] M. Wolf, S.A. Cheema, M.A. Khalighi, S. Long, Transmission schemes for visible
spread OFDM, in: IEEE ICC, 2014, pp. 3325–3330. light communications in multipath environments, in: Intl. Conf. on Transparent Opt.
[17] O. Saied, Z. Ghassemlooy, R.C. Kizilirmak, X. Dai, C. Ribeiro, M. Zhang, S. Rajbhan- Networks, 2015, pp. 1–7.
dari, Single carrier optical FDM in visible light communication, in: IEEE CSNDSP, [25] F.R. Gfeller, U. Bapst, Wireless in-house data communication via diffuse infrared
2016, pp. 1–5. radiation, Proc. IEEE 67 (11) (1979) 1474–1486.
[18] J. Zhou, Y. Qiao, Low-PAPR asymmetrically clipped optical OFDM for intensity- [26] K. Lee, H. Park, J.R. Barry, Indoor channel characteristics for visible light commu-
modulation/direct-detection systems, IEEE Photonics J. 7 (3) (2015) 1–8. nications, IEEE Commun. Lett. 15 (2) (2011) 217–219.
[19] J. Zhou, Y. Qiao, Low-peak-to-average power ratio and low-complexity asymmetri- [27] A. Nuwanpriya, S.-W. Ho, J.A. Zhang, A.J. Grant, L. Luo, PAM-SCFDE for optical
cally clipped optical orthogonal frequency-division multiplexing uplink transmission wireless communications, J. Lightwave Technol. 33 (14) (2015) 2938–2949.
scheme for long-reach passive optical network, Opt. Lett. 40 (17) (2015) 4034–4037. [28] J. Wang, Y. Xu, X. Ling, R. Zhang, Z. Ding, C. Zhao, PAPR analysis for OFDM visible
[20] Y.S. Cho, J. Kim, W.Y. Yang, C.G. Kang, MIMO-OFDM Wireless Communications light communication, Opt. Express 24 (24) (2016) 27457–27474.
with MATLAB, John Wiley & Sons, 2010. [29] S.G. Johnson, M. Frigo, A modified split-radix FFT with fewer arithmetic operations,
[21] N. Wu, Y. Bar-Ness, A novel power-efficient scheme asymmetrically and symmetri- IEEE Trans. Signal Process. 55 (1) (2007) 111–119.
cally clipping optical (ASCO)-OFDM for IM/DD optical systems, EURASIP J. Adv. [30] P. Schniter, Low-complexity equalization of OFDM in doubly selective channels,
Sig. Process. 2015 (1) (2015) 3. IEEE Trans. Signal Process. 52 (4) (2004) 1002–1011.
[22] A. Weiss, A. Yeredor, M. Shtaif, Iterative symbol recovery for power-efficient dc- [31] J. Zhou, Y. Qiao, Z. Cai, Y. Ji, An improved scheme for flip-OFDM based on Hartley
biased optical OFDM systems, J. Lightwave Technol. 34 (9) (2016) 2331–2338. transform in short-range IM/DD systems, Opt. Express 22 (17) (2014) 20748–20756.
[23] H. Elgala, R. Mesleh, H. Haas, Practical considerations for indoor wireless optical [32] J.K. Perin, M. Sharif, J.M. Kahn, Modulation schemes for single-laser 100 Gb/s links:
system implementation using OFDM, in: IEEE ConTEL, 2009, pp. 25–29. Multicarrier, J. Lightwave Technol. 33 (24) (2015) 5122–5132.

185