Received September 28, 2018, accepted October 11, 2018, date of publication October 15, 2018, date of current

version November 8, 2018.

Digital Object Identifier 10.1109/ACCESS.2018.2876148

An Overview of OFDM-Based Visible Light

Communication Systems From the Perspective
of Energy Efficiency Versus Spectral Efficiency
YAQI SUN1 , FANG YANG 1,2 , (Senior Member, IEEE),
AND LING CHENG 3 , (Senior Member, IEEE)
1 Department of Electronic Engineering, Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing 100084, China
2 Key Laboratory of Digital TV System of Guangdong Province and Shenzhen City, Research Institute of Tsinghua University in Shenzhen, Shenzhen 518057,
China
3 School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg 2000, South Africa

Corresponding author: Fang Yang ([email protected])

This work was supported in part by the National Key Research and Development Program of China under Grant 2017YFE0113300 and in
part by the South African National Research Foundation under Grant 114626.

ABSTRACT In this paper, the energy efficiency and the spectral efficiency for orthogonal frequency division
multiplexing (OFDM)-based visible light communication schemes are studied, which is crucial for practical
application with limited energy resources. The conventional schemes, including asymmetrically clipped
optical OFDM (ACO-OFDM), pulse-amplitude-modulated discrete multitone, and direct current biased
optical OFDM, are compared in terms of energy efficiency and spectral efficiency relationship. The influence
of power allocation for asymmetrically clipped dc biased optical OFDM and hybrid ACO-OFDM is also
investigated. The energy efficiency and spectral efficiency of layered ACO-OFDM with a variable layer
number are calculated and their relationship is also formulated. These conventional and hybrid modulation
schemes are analyzed and compared through computer simulations, which should be considered in practice
according to the requirements of illumination and transmission.

INDEX TERMS Energy efficiency, spectral efficiency, orthogonal frequency division multiplexing (OFDM),
visible light communication (VLC).

I. INTRODUCTION inverse fast Fourier transform (IFFT), cannot be directly uti-

The data traffic of conventional radio frequency communi- lized for VLC. Therefore, appropriate conversions have to
cation is growing exponentially [1]. Recently, visible light be developed for optical OFDM systems. The real-valued
communication (VLC) has drawn increasing attention as a output of the IFFT is realized by imposing a Hermi-
potential complementary technology owing to its unlicensed tian symmetry constraint to the frequency-domain signal.
spectrum resource [2], [3]. Single subcarrier schemes, such Apart from that, several schemes have been proposed to guar-
as on-off keying (OOK) and pulse position modulation (PPM) antee the unipolarity such as asymmetrically clipped opti-
may suffer from inter-symbol interference (ISI) for high- cal OFDM (ACO-OFDM) [6], pulse-amplitude-modulated
data-rate transmission. Thus, orthogonal frequency division discrete multitone (PAM-DMT) [7], and direct current
multiplexing (OFDM) is considered as a promising strat- biased optical OFDM (DCO-OFDM) [8]. ACO-OFDM and
egy to transmit high data rates [4]. The data transmission PAM-DMT only utilize the odd subcarriers and the imag-
in VLC is usually realized by intensity modulation and inary part of subcarriers, respectively, resulting in loss
direct detection (IM/DD) [5], where the electrical signal of spectral efficiency. Some strategies can help amelio-
has to be transformed into a real-valued and nonnegative rate the situation. In literature [9], asymmetrically clipped
waveform for driving light emitting diodes (LEDs). The DC biased optical OFDM (ADO-OFDM) is introduced,
conventional OFDM for radio frequency (RF) communi- which is a combination of ACO-OFDM and DCO-OFDM.
cation, which generates complex and bipolar signals by A hybrid ACO-OFDM (HACO-OFDM) is also described

2169-3536 2018 IEEE. Translations and content mining are permitted for academic research only.
60824 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 6, 2018
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

by integrating ACO-OFDM and PAM-DMT [10]. Recently, are presented in Section IV, while this paper concludes in
a spectrally efficient strategy called layered ACO-OFDM Section V.
(LACO-OFDM) or enhanced ACO-OFDM (eACO-OFDM)
is proposed, which combines multilayer signals for simulta- II. SYSTEM MODEL
neous transmission [11]–[13]. A. MODULATION SCHEMES
With the enormous increase of communication devices, Fig. 1 illustrates which of the first 16 out of 64 subcarri-
the energy efficiency, which is defined as the transmitted ers are utilized for different modulations, including ACO-
data rate per power consumption, is widely considered in OFDM, PAM-DMT, DCO-OFDM, ADO-OFDM, HACO-
green communication [14]. Spectral efficiency, i.e., the trans- OFDM, and LACO-OFDM. Without loss of generality, the
mitted data rate per bandwidth, is a conventional metric for 3-Layer LACO-OFDM is presented as an example of
communication system, which can be improved by ampli- LACO-OFDM.
fying the power. However, the energy efficiency may suf-
fer a decrease with the increase of the spectral efficiency. 1) ACO-OFDM
It is a challenge to guarantee the quality of service (QoS) In ACO-OFDM, the data are only carried by odd sub-
with affordable energy for the application of the Internet of carriers. Considering the Hermitian symmetry, the N -
things (IoT). Therefore, a tradeoff is supposed to be made point frequency-domain signal can be denoted as X =
between the energy efficiency and the spectral efficiency. The [0, X1 , 0, X3 , . . . , XN /2−1 , 0, XN∗ /2−1 , . . . , X3∗ , 0, X1∗ ]. The
tradeoff between energy efficiency and spectral efficiency time-domain signal xn has the anti-symmetric property as
has been investigated for radio frequency communication. xn = −xn+N /2 , (0 ≤ n < N /2). The ACO-OFDM signal,
In literature [15], the relationship between the energy and xACO,n , is generated by clipping the negative part as
spectral efficiency is proved to be quasiconcave for down-
link OFDM access (OFDMA) network. A multi-objective (
xn , xn ≥ 0,
optimization approach is proposed for the tradeoff problem xACO,n = (1)
in [16]. For DCO-OFDM-based VLC systems, the relation- 0, xn < 0,
ship of energy efficiency and spectral efficiency is investi-
gated in [17], where the energy efficiency is defined as the where the data information is not lost. The transmitted data
data rate per power consumed by LED. can be demodulated correctly since the clipping noise only
In VLC, the performances of different modulation schemes falls on the even subcarriers.
are affected by some parameters. For DCO-OFDM, a low DC According to the central limit theorem, the unipolar sig-
bias will affect the channel capacity due to the remaining neg- nal obeys a clipped Gaussian distribution. σA2 refers to the
ative signals, while a high DC bias will make the signal higher variance of xn , then the mean value
 and mean square
√ value
than the upper bound of the dynamic range. Therefore, it is of hxACO,n ican be calculated as E xACO,n = σA / 2π , and
important to explore the influence of different DC biases on 2
E xACO,n = σA2 /2.
the energy-spectral efficiency. For HACO-OFDM and ADO-
OFDM, different power distributions result in different per- 2) PAM-DMT
formances, which should be taken into consideration. As for In PAM-DMT, real-valued signals drawn from PAM are
LACO-OFDM, more layers lead to higher spectral efficiency modulated onto the imaginary part of each subcarrier
since the number of modulated subcarriers increases, which except the 0-th and N /2-th subcarriers. The signal in
however causes larger power cost. the frequency domain can by represented as Y =
In this paper, the OFDM-based modulation schemes are [0, Y1 , Y2 , . . . , YN /2−1 , 0, YN∗ /2−1 , . . . , Y1∗ ], where Yk = ibk ,
comprehensively analyzed from the perspective of their bk (k = 1, 2, . . . , N /2 − 1) is the PAM signal and i2 = −1.
energy and spectral efficiencies, which is crucial for practical It has been proved that the time-domain signal yn follows the
application with limited energy resources in VLC. To opti- symmetry as yn = −yN −n , (0 ≤ n < N /2) [10]. Therefore,
mize the consumed energy per bit, the energy efficiency ver- PAM-DMT signal yPAM,n can also be clipped at zero, which
sus spectral efficiency curves for various modulation schemes is given by
are mathematically formulated and compared to investigate
the adaptive modulation according to different transmission
(
yn , yn ≥ 0,
and illumination requirements. The influence factors, such as yPAM,n = (2)
the DC bias, the power allocation and layer number, are varied 0, yn < 0.
to decide the optimal parameter setup for each modulation
The clipping noise only falls on the real part of each subcar-
scheme under various application environments.
rier, leaving the transmitted data easy for demodulation [10].
The rest of this paper is organized as follows. In Section II,
Similarly to ACO-OFDM, it is easy to calculate the expec-
√
the models of the OFDM-based systems and the optical 2 σP / 2π,

tations hof yPAM,n and y PAM,n as E yPAM,n =
channel are introduced. The energy efficiency and spec- i
tral efficiency for different OFDM-based VLC modulation and E y2PAM,n = σP2 /2, where σP2 is the variance
schemes are investigated in Section III. Simulation results of yn .

VOLUME 6, 2018 60825

 Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

FIGURE 1. The modulated subcarriers for different modulation schemes. (a) ACO-OFDM. (b) PAM-DMT or DCO-OFDM. (c) ADO-OFDM or
HACO-OFDM. (d) 3-Layer LACO-OFDM.

3) DCO-OFDM transmission. The obtained time-domain HACO-OFDM sig-

In DCO-OFDM, the complex-valued frequency-domain sig- nal, sHACO,n , is denoted as
nal is Z = [0, Z1 , Z2 , . . . , ZN /2−1 , 0, ZN∗ /2−1 , . . . , Z1∗ ]. A DC sHACO,n = xACO,n + yPAM,n . (7)
bias, zb , is added to the time-domain signal, zn , so that most
of the signal is positive. The remaining negative signal will 5) ADO-OFDM
be clipped at zero, leading to a clipping noise, which depends
Similarly to HACO-OFDM, ADO-OFDM combines ACO-
on the DC bias.
OFDM with DCO-OFDM. The ACO-OFDM signal is
The mean value and mean square value of zDCO,n can be
obtained by using the conventional method. The other stream
calculated as [25]
of signal only utilized even subcarriers with a DC bias. Then,
    
zb zb the transmitted ADO-OFDM signal, sADO,n , is obtained by
E zDCO,n = σD G , (3)
 
+ zb 1 − Q
σD σD adding them together as
  
zb sADO,n = xACO,n + zDCO,n .
h i  
E z2DCO,n = σD2 + z2b 1 − Q (8)
σD
 
zb 6) LACO-OFDM
+ zb σD G , (4)
σD The LACO-OFDM scheme consists of multiple layers of
ACO-OFDM signals. For clarity, L-Layer LACO-OFDM
where σD2 is the variance of zn . G(·) and Q(·) represent the denotes a LACO-OFDM with L layers, while l-th
standard Gaussian distribution and the tail probability of the ACO-OFDM refers to the l-th layer in LACO-OFDM.
standard normal distribution, respectively, which are given by In the l-th ACO-OFDM, only the 2l−1 (2k + 1)-th (k =
0, 1, . . . , N /2l − 1) subcarriers are modulated, denoted as
ω2
 
1 (l)
G (ω) = √ exp − , (5) XACO,k . For each layer, the negative part of the time-domain
2π 2 signal can be clipped without loss of any information, result-
λ
Z ∞  2 (l)
1 ing in xACO,n . Then the several streams of non-negative sig-
Q(ω) = √ exp − dλ, (6)
2π ω 2 nals are superposed together and transmitted simultaneously
as
where λ represents the integral variable. Usually, zb is relative
L
toq
the electrical power of the signal zn , and can be set to zb = X (l)
sL,n = xACO,n . (9)
µ E z2n = µσD , where µ is a proportional constant.

l=1

4) HACO-OFDM B. CHANNEL MODEL

HACO-OFDM is a combination of ACO-OFDM and PAM- The channel for VLC can be modeled as a linear
DMT, where the information is carried by the odd subcarriers time-invariant channel with the additive white Gaussian noise
and the imaginary part of the even subcarriers. The negative (AWGN), where the channel frequency response can be
part is clipped individually for the two varieties of signals. considered to be flat near DC [20]. The transmitted signal
Then the two signals are added together for simultaneous generated by different schemes, including xACO,n , yPAM,n ,

60826 VOLUME 6, 2018

Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

zDCO,n , sHACO,n , sADO,n and sL,n , are generalized as sn . Then,

the received signal, rn , can be represented as
rn = H0 Rsn + vn , (10)
where H0 denotes the channel DC gain, vn is referred to
as AWGN, and R is the combined coefficient including the
voltage-current transfer, the LED responsivity, and the detec-
tor responsivity.

FIGURE 3. The current-voltage relationship for LED.

The turn-on voltage (TOV) of LED is the minimum

threshold value for the input that can generate current. The
nonlinearity of LED can be mitigated by the methods pro-
posed in [24]. Thus the current-voltage relationship can be
quasi-linear in a limited range, as shown in Fig. 3, where
UL denotes the LED TOV, UH and IH refer to the maximum
allowable voltage and current, respectively. The reciprocal of
FIGURE 2. The geometry for VLC systems.
the slope can be calculated as κ = (UH − UL )/IH .
A proper DC bias has to be added to the OFDM-based VLC
Fig. 2 demonstrates a geometry for VLC systems, which signal in consideration of the TOV. Since the OFDM-based
is considered in this paper, where 81/2 is the transmitter VLC signal is non-negative, the DC bias is supposed to be
semiangle, the light radiance and incidence angles are φ and equal to UL . Thus, the input voltage can be given by
ψ, respectively. h and w represent the vertical and horizontal
distances between √ the transmitter and the receiver, respec- Un = κsn + UL . (15)
tively. And d = h2 + w2 is the direct distance. In line-of-
sight links, the DC gain can be modeled with a generalized III. SPECTRAL EFFICIENCY AND ENERGY EFFICIENCY
Lambertian radiant intensity as [21]–[23] For VLC systems, the spectral efficiency can be defined as
(m + 1)A the channel capacity per unit of bandwidth, which is given by
H0 = cosm (φ)T (ψ)g(ψ) cos(ψ), (11)
2π d 2 C
where m = −1/ log2 (cos(81/2 )) is the Lambertian order ηSE = , (16)
W
and A denotes the detector physical area. T (ψ) is the filter
gain which can be set to 1 for simplification. g(ψ) is the where C and W denote the channel capacity and the whole
concentrator gain, which is given by bandwidth, respectively. The energy efficiency is defined
as the ratio of the channel capacity over the mean power
 ν
 2
consumption [17], which is given by
, 0 ≤ ψ ≤ 9c ,
g(ψ) = sin2 9c (12)
C
ψ > 9c , ηEE = ,

0, (17)
P
where 9c refers to the field of view (FOV) of the concentrator
where P denotes the LED power cost. It is worth mention-
and ν is the internal refractive index. As shown in Fig. 2,
ing that the power is also cost by the conversion of base-
the relationship between the angle and the distance can be
band electrical signals into respective OFDM/DMT signals,
given as
which, however, is too complicated to analyze. Thus, this
h paper only focuses on the power consumption by the LED.
cos(ψ) = cos(φ) = √ . (13)
h2 + w2 In this section, the spectral efficiency and energy efficiency
for OFDM-based VLC modulation schemes are investigated.
Thus, for 0 ≤ ψ ≤ 9c , the DC gain can be derived as
(m + 1)Aν 2 hm+1 A. ACO-OFDM
H0 = √ m+3 . (14)
2π sin 9c
2
2
h +w 2 Since the amplitude of the clipped signal in the frequency
domain is halved, the signal-to-noise ratio (SNR) for each

VOLUME 6, 2018 60827

 Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

subcarrier is B. PAM-DMT
The spectral efficiency and the energy efficiency for
H02 R2 14 εA H02 R2 εA PAM-DMT are similar to that for ACO-OFDM. The SNR for
ξACO = = , (18)
σn2 4σn2 each subcarrier is

where εA denotes the variance of the modulated symbol H02 R2 14 εP H02 R2 εP

ξPAM = = , (25)
X2k+1 (k = 0, 1, . . . N /4 − 1), σn2 = N0 W /N is the variance
2 σn
1 2 2σn2
of the AWGN noise. Considering that only the odd subcarri-
ers are utilized in ACO-OFDM, the relationship between εA where εP = E[Yk2 ] (k = 1, 2, . . . N /2 − 1). Since the 0-th
and σA2 can be derived as and the N /2-th subcarrier are set to zeros, the relationship
between εP and σP2 is given by
εA
σA2 = . (19)
2 (N − 2) εP
σP2 = . (26)
N
The total channel capacity for ACO-OFDM can be calcu-
lated as Considering that the data is only modulated onto the imag-
inary part of the subcarriers, the total channel capacity for
NW PAM-DMT is calculated as
CACO = log2 (1 + ξACO )
4 N ! 1N −2W
W H02 R2 εA CPAM = log2 (1 + ξPAM )
= log2 1 + 2 2 N
4 4σn2 !
(N − 2) W H02 R2 εP
! = log2 1 +
W H02 R2 σA2 4N 2σn2
= log2 1 + . (20)
4 2σn2 !
(N − 2) W NH02 R2 σP2
= log2 1 + . (27)
Thus, the spectral efficiency for ACO-OFDM is given by 4N 2 (N − 2) σn2

CACO Then, the spectral efficiency is derived as

ηSE,ACO =
W CPAM
! ηSE,PAM =
1 H02 R2 σA2 W
= log2 1 + . (21) !
4 2σn2 N −2 NH02 R2 σP2
= log2 1 + . (28)
4N 2 (N − 2) σn2
The power consumed by LED can be formulated as
The power consumed by LED for PAM-DMT is given by
PACO = E [Un In ] h i
PPAM = κE y2PAM,n + UL E yPAM,n
 
= E κxACO,n + UL xACO,n



σP2 σP
h i
= κE xACO,n
2
 
+ UL E xACO,n =κ + UL √ . (29)
2 2π
σ2 σA
= κ A + UL √ . (22) Thus, the energy efficiency can be represented as
2 2π
CPAM
Then the energy efficiency can be derived as ηEE,PAM =
PPAM  
CACO (N −2)W NH02 R2 σP2
ηEE,ACO = 4N log2 1 + 2(N −2)σn2
PACO
  = . (30)
H02 R2 σA2 σP2
W
4 log2 1 + 2σn2
κ 2 + UL √σP
2π
= . (23)
σ2
κ 2A + UL √σA The energy efficiency can be derived as a function of the
2π spectral efficiency as
Based on (21) and (23), the relationship between spectral W ηSE,PAM
efficiency and energy efficiency for ACO-OFDM can be ηEE,PAM = 4ηSE,PAM
! ,
deduced as αP κσn2 2 αP −1 q
4ηSE,PAM
αP UL σn
q
αP
+ π H0 R 2 −1
W ηSE,ACO H02 R2
ηEE,ACO =   . (24) (31)
κσn2 24ηSE,ACO −1
√UL σn
p
+ πH0 R
24ηSE,ACO −1
H02 R2 where αP = (N − 2)/N .

60828 VOLUME 6, 2018

Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

C. DCO-OFDM The power cost by LED is calculated as

For DCO-OFDM, the remaining negative signal after adding h i
PDCO = κE z2DCO,n + UL E zDCO,n
 
the DC bias will be clipped at zero. Thus, the transmitted    
zb
 
zb
DCO-OFDM signal can be modeled as zDCO,n = zclip,n + zb , = κ σD2 + z2b 1 − Q + zb σD G
where zclip,n is the signal clipped at a level of −zb , i.e.,    σD   σD
zb zb
+ UL σD G + zb 1 − Q . (40)
(
zn , zn > −zb , σD σD
zclip,n = (32)
−zb , zn ≤ −zb . Thus, the spectral efficiency and energy efficiency for
Based on Bussgang theorem, zclip,n can be modeled as DCO-OFDM can be formulated as
CDCO
zclip,n = czn + nDCO,n , (33) ηSE,DCO =
W !
N −2 H02 R2 c2 σD2
where nDCO,n is the clipping distortion and c is a constant, = log2 1 + 2 , (41)
which can be calculated as [25] 2N 2 + σ2
H0 R2 σclip n
CDCO
ηEE,DCO =
 
zb
c=1−Q , (34) PDCO
σD (N −2)W H 2 R2 c2 σ 2
log2 (1 + 2 02 2 D 2 )
2N H0 R σclip +σn
where σD2 = E[z2n ]. = h i . (42)
κE z2DCO,n + UL E zDCO,n

Based on (3) and (4), the mean value and mean square value
of zclip,n can be calculated as
    D. HACO-OFDM
E zclip,n = E zDCO,n − zb
    In HACO-OFDM systems, the ACO-OFDM signal occupies
zb zb odd subcarriers, whereas the PAM-DMT occupies (N /2 − 2)
= σD G − zb Q , (35)
σD σD subcarriers. Thus, σA2 and σP2 can be calculated as
h i h
2 i
E z2clip,n = E zDCO,n − zb εA
σA2 = , (43)
h i 2
= E z2DCO,n + z2b − 2zb E zDCO,n (N − 4) εP
 
σP2 = . (44)
  z 
b
 
zb 2N
= σD2 + z2b − σD2 Q − zb σD G , Thus, the channel capacity for HACO-OFDM can be formu-
σD σD
(36) lated as
CHACO
Then, the variance of the clipping distortion can be calculated W (N − 4) W
as [25] = log2 (1 + ξACO ) + log2 (1 + ξPAM )
4 ! 8N
H R εA
2 2
h i
σclip
2
= E z2clip,n − c2 σD2 − E zclip,n
 2 W
= log2 1 + 0 2
  z  4 4σn
b
= σD2 + z2b − σD2 Q
!
σD (N − 4) W H02 R2 εP
+ log2 1 +
8N 2σn2
zb 2
   
zb
− zb σD G − σD Q −
2 !
σD σD W H 2 R2 σ 2
= log2 1 + 0 2 A
  
zb
 2
zb 4 2σn
− σD G − zb Q . (37) !
σD σD (N − 4) W NH02 R2 σP2
+ log2 1 + . (45)
By taking clipping distortion into consideration, the signal- 8N (N − 4) σn2
to-noise-plus-distortion (SNDR) for each subcarrier can be The LED power consumption is given by
calculated as
PHACO
H02 R2 c2 σD2 = E hκ xACO,n +


+


+


ξDCO = . (38) y PAM,n U L xACO,n yPAM,n
H02 R2 σclip
2 + σ2 i h i
= κE xACO,n
2
+ κE y2PAM,n + 2κE xACO,n E yPAM,n
   
n
   
And the total channel capacity is given by + UL E xACO,n + UL ! E yPAM,n
N −2W σA2 σ2 σA σP σA + σP
CDCO = log2 (1 + ξDCO ) =κ + P + + UL √ . (46)
2 N ! 2 2 π 2π
(N − 2)W H02 R2 c2 σD2 Then, the spectral efficiency and energy efficiency can be
= log2 1 + 2 . (39)
2N H0 R2 σclip
2 + σ2
n derived according to (45) and (46).

VOLUME 6, 2018 60829

 Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

E. ADO-OFDM where
The odd subcarriers and even subcarriers (except 0-th 2L − 1


and N /2-th subcarriers) are utilized for ACO-OFDM and αL = .
2L+1
DCO-OFDM in ADO-OFDM systems. The channel capacity
The spectral efficiency is given by
and the power cost by LED can be calculated as
CLACO H 2 R2 εL
CADO ηSE,LACO = = αL log2 (1 + 0 2 ). (52)
W (N − 4) W W 4σn
= log2 (1 + ξACO ) + log2 (1 + ξDCO )
4 ! 4N For L-Layer LACO-OFDM systems, PLACO can be calcu-
W H 2 R2 σ 2 lated as
= log2 1 + 0 2 A h i
4 2σn PLACO = κE (xL−LACO )2 + UL E [xL−LACO ]. (53)
!
(N − 4) W H02 R2 c2 σD2 The expectation of sL,n can be derived as
+ log2 1 + 2 , (47)
4N 2 + σ2
H0 R2 σclip n " L # L
X (l) h i
(l)
X
PADO
 
E sL,n = E xACO = E xACO
κ xACO,n + zDCO,n + UL xACO,n + zDCO,n




=E l=1 l=1
h i h i PL PL √εL
= κE xACO,n
2
+ κE z2DCO,n + 2κE xACO,n E zDCO,n σ
   
l l=1 l/2
= √l=1 = √ 2
  
+ UL E xACO,n + UL E zDCO,n
 2π 2π
1 √
  
zb
  
zb 1 − 2L/2 εL √
= γL εL ,

= κ σD + zb 1 − Q
2 2
+ zb σD G = √ √ (54)
σD σD 2 − 1 2π
      
2κσA zb zb where
+ √ + UL σD G + zb 1 − Q
2π σD σ D 1
1 − L/2 1
σA2 σA γL = √ 2 √ .
+κ + UL √ . (48) 2 − 1 2π
2 2π Considering that the signals from different layers are inde-
Based on (47) and (48), ηSE,ADO and ηSE,ADO are calculable. pendent, the mean square value of sL,n can be formulated as
h
2 i
F. LACO-OFDM E sL,n
For L-Layer LACO-OFDM, assuming that the power is !2 

L
(l)
X
equally distributed to each subcarrier [26], the variance of the = E xACO 
(l)
modulated symbol XACO,k can be normalized as εL for each l=1
layer. Since only N /2l subcarriers are modulated for the l-th L  2  XL L h i
(l) (i) (j)
X X
layer, according to the Parseval’s theorem, the relationship = E xACO + E xACO xACO
between σl and εL can be derived as l=1 i=1 j=1,j6=i
εL L L L
σl2 = l . (49)
 2  X h i h i
(l) (i) (j)
X X
2 = E xACO + E xACO E xACO .
Then the signal-to-noise ratio (SNR) of each modulated l=1 i=1 j=1,j6=i
subcarrier is given by (55)
H02 R2 εL The first term in (55) can be simplified as
ξL = , (50)
4σn2 L L L εL
σl2 2L − 1
 2  X
(l) 2l
X X
where σn2 = N0 W /N is the variance of the AWGN noise. E xACO = = = L+1 εL . (56)
2 2 2
l=1 l=1 l=1
N0 is the noise power spectral density and W denotes the
whole bandwidth. Then the total channel capacity of L-Layer The simplification of the second term in (55) can be given by
LACO-OFDM can be calculated as L L h i h i
(i) (j)
X X
L E xACO E xACO
!
X W H02 R2 εL
CLACO = log2 1 + i=1 j=1,j6=i
2l+1 4σn2
l=1 L L
! X X σi σj
2L − 1 W H02 R2 εL


=
= log2 1 + 2π
2L+1 4σn2 i=1 j=1,j6=i
 
L L
εL 
!
H 2 R2 εL =
X X 1 1 
= αL W log2 1 + 0 2 , (51) 2π 2i/2 2j/2
4σn i=1 j=1,j6=i

60830 VOLUME 6, 2018

Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

 
L L L
εL X X 1 1 X 1 1  TABLE 1. Simulation parameters.
= −
2π 2i/2 2j/2 2i/2 2i/2
i=1 j=1 i=1
   
L L L
εL X  1 X 1  X 1 
= −
2π 2i/2 2j/2 2i
i=1 j=1 i=1
 !2  
1
εL  1 − 2L/2

1
= √ − 1− L 
2π 2−1 2
√ √ !
√ 3 + 2 2 2 + 2 εL
= 1+ 2− + . (57)
2L/2 2L π
Therefore, equation (55) can be simplified based on
(56) and (57), which is given by
h
2 i
E sL,n = βL εL , (58)

where
√ √ !
2L − 1 √ 3+2 2 2+ 2 1
βL = L+1 + 1 + 2 − + .
2 2L/2 2L π
Thus the power consumed by LED in (53) can be derived as
√
PLACO = βL κεL + γL UL εL . (59)
Then, the energy efficiency is calculated by
 
H 2 R2 ε
αL W log2 1 + 04σ 2 L
n
ηEE,LACO = √ . (60)
βL κεL + γL UL εL
According to (52), εL can be formulated as a function of
ηSE,LACO , which is given by
4σn2 (2ηSE,LACO /αL − 1)
εL = . (61)
H02 R2 FIGURE 4. Spectral efficiency versus energy efficiency for ACO-OFDM,
PAM-DMT, and DCO-OFDM.
Thus, the relationship between energy efficiency and spectral
efficiency can be deduced as wasted in ACO-OFDM and PAM-DMT, which leads to a
W ηSE,LACO limited achievable spectral efficiency in consideration of the
ηEE,LACO = √ .
4βL κσn2 (2ηSE /αL −1) 2γL UL σn constrained clipping ratio. The DCO-OFDM utilizes all sub-
+ 2ηSE /αL − 1
H02 R2 H0 R carriers except 0-th and N /2-th subcarriers, thus the spectral
(62) efficiency can exceed 2 bits/s/Hz, as shown in Fig. 4. In
DCO-OFDM, a higher DC bias results in a lower energy
IV. SIMULATION RESULTS efficiency. Besides, the reachable ηSE is restricted for both
In this section, simulation results of the energy efficiency and small and large DC biases due to the clipping noise caused
spectral efficiency relationship are demonstrated. The simu- by the negative signal and the signal which is higher than IH .
lation parameters are listed in Table 1. For fair comparison, According to the simulation results, the DCO-OFDM with
the largest acceptable clipping ratio, which is the proportion µ = 2 is relatively satisfactory.
of the signal that is higher than IH , is set to 1%. The simulated ηEE and ηSE for HACO-OFDM with differ-
Fig. 4 demonstrates the spectral efficiency versus energy ent power allocations are presented in Fig. 5. As ACO-OFDM
efficiency for the conventional ACO-OFDM, PAM-DMT, and PAM-DMT have similar performances, the four curves
and DCO-OFDM. In DCO-OFDM, the parameter for the are pretty close. For a low spectral efficiency, the energy effi-
DC bias, µ, ranges from 1 to 4. Since no extra DC bias ciency increases if ACO-OFDM occupies more power. How-
is added to the ACO-OFDM and PAM-DMT signals, it can ever, the largest achievable spectral efficiency decreases with
be seen that these two clipping-based strategies can achieve the rising proportion of ACO-OFDM. All things considered,
higher energy efficiency than DCO-OFDM when ηSE is low. εA = 4εP would be a superior choice, which has a high energy
However, approximately half of the spectrum resources are efficiency and a wide spectral efficiency range.

VOLUME 6, 2018 60831

 Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

FIGURE 7. Spectral efficiency versus energy efficiency for LACO-OFDM.

FIGURE 5. Spectral efficiency versus energy efficiency for HACO-OFDM.

The simulations are also carried out for ADO-OFDM,

where the parameter µ is set to 2. The results are plot-
ted on Fig. 6. For the low spectral efficiency, the more
power ACO-OFDM possesses, the higher energy efficiency
ADO-OFDM can attain. As for the largest achievable ηSE ,
εA = 4εP is shown to be the optimal choice. On the one hand,
if the ACO-OFDM occupies too much power, the spectral
efficiency gain that DCO-OFDM produces is slight. On the
other hand, DCO-OFDM suffers from the clipping noise,
which means that more power allocated to DCO-OFDM can
degrade the largest spectral efficiency.

FIGURE 8. Spectral efficiency versus energy efficiency for ACO-OFDM for

HACO-OFDM, ADO-OFDM, and LACO-OFDM.

LACO-OFDM with more layers has superior performance.

Nevertheless, the 4-Layer and 5-Layer cases have almost
the same highest ηSE , which is approximately 2.8 bits/s/Hz.
Therefore, the optimal layer number should be adap-
tively chosen according to the requirement of the spectral
efficiency.
Fig. 8 displays ηEE versus ηSE results for the hybrid modu-
lation schemes, including HACO-OFDM, ADO-OFDM, and
LACO-OFDM. For each scheme, two curves are selected
based on the results in Figs. 5-7. As the graph shows,
HACO-OFDM and 2-Layer LACO-OFDM have the supe-
FIGURE 6. Spectral efficiency versus energy efficiency for ADO-OFDM. rior performances for a relative low spectral efficiency. For
Fig. 7 compares the differences of ηEE − ηSE relationships 2 < ηSE < 3, 4-Layer LACO-OFDM is the most energy
with various layer numbers for LACO-OFDM. The optimal efficient scheme. Despite of relatively low energy efficiency,
layer number, which achieves the highest energy efficiency, ADO-OFDM can achieve a wider spectral efficient range
rises with the increase of spectral efficiency. For ηSE that is compared with the other hybrid methods. Therefore, the opti-
lower than 1 bits/s/Hz, L = 1 is shown to be the optimal mal modulation schemes should be adaptively selected based
choice, while 2-Layer LACO-OFDM outperforms the other on the application requirement.
schemes with the spectral efficiency ranging from 1 bits/s/Hz
to 1.8 bits/s/Hz. For ηSE that is higher than 2.5 bits/s/Hz, V. CONCLUSIONS
the simulation results for 4-Layer LACO-OFDM and 5-Layer The energy efficiency and spectral efficiency for the conven-
LACO-OFDM are quite close, which means the increase tional and hybrid OFDM-based VLC modulation schemes
of the layer number can hardly bring any gain. In the per- are formulated in this paper, which assists in ensuring the
spective of the largest achievable spectral efficiency, the QoS with affordable energy. The parameters such as the DC

60832 VOLUME 6, 2018

Y. Sun et al.: Overview of OFDM-Based VLC Systems From the Perspective of Energy Efficiency Versus Spectral Efficiency

bias in DCO-OFDM, the power allocation in HACO-OFDM [20] J. M. Kahn and J. R. Barry, ‘‘Wireless infrared communications,’’ Proc.
and ADO-OFDM, the layer number in LACO-OFDM have IEEE, vol. 85, no. 2, pp. 265–298, Feb. 1997.
[21] K. Lee, H. Park, and J. R. Barry, ‘‘Indoor channel characteristics for visible
various effects on the energy efficiency and spectral effi- light communications,’’ IEEE Commun. Lett., vol. 15, no. 2, pp. 217–219,
ciency. The comparisons of these schemes are explored Feb. 2011.
by theoretical analysis and computer simulations. Different [22] D. A. Basnayaka and H. Haas, ‘‘Hybrid RF and VLC systems: Improving
user data rate performance of VLC systems,’’ in Proc. IEEE 81st Veh.
schemes have varied performances in terms of energy effi- Technol. Conf. (VTC Spring), May 2015, pp. 1–5.
ciency whereas the achievable ranges of spectral efficiency [23] K. Cai, M. Jiang, and X. Ma, ‘‘Photodetector selection aided multiuser
are diverse. For practical application, the optimal scheme MIMO optical OFDM imaging visible light communication system,’’
IEEE Access, vol. 4, pp. 9870–9879, Dec. 2016.
should be adaptively chosen with comprehensive consider- [24] R. Mesleh, H. Elgala, and H. Haas, ‘‘LED nonlinearity mitigation tech-
ation of the requirements of transmission and illumination as niques in optical wireless OFDM communication systems,’’ IEEE/OSA
well as the tradeoff between energy and spectral efficiency. J. Opt. Commun. Netw., vol. 4, no. 11, pp. 865–875, Nov. 2012.
[25] X. Ling, J. Wang, X. Liang, Z. Ding, and C. Zhao, ‘‘Offset and power
optimization for DCO-OFDM in visible light communication systems,’’
REFERENCES IEEE Trans. Signal Process., vol. 64, no. 2, pp. 349–363, Jan. 2016.
[1] S. Dimitrov and H. Haas, ‘‘Information rate of OFDM-based optical [26] Y. Sun, F. Yang, and J. Gao, ‘‘Comparison of hybrid optical modulation
wireless communication systems with nonlinear distortion,’’ J. Lightw. schemes for visible light communication,’’ IEEE Photon. J., vol. 9, no. 3,
Technol., vol. 31, no. 6, pp. 918–929, Mar. 15, 2013. Jun. 2017, Art. no. 7904213.
[2] A. Jovicic, J. Li, and T. Richardson, ‘‘Visible light communication: Oppor-
tunities, challenges and the path to market,’’ IEEE Commun. Mag., vol. 51, YAQI SUN received the B.Eng. degree from the
no. 12, pp. 26–32, Dec. 2013. Department of Electronic Engineering, Tsinghua
[3] M. Z. Chowdhury, M. T. Hossan, A. Islam, and Y. M. Jang, ‘‘A comparative
University, Beijing, China, in 2017, where she
survey of optical wireless technologies: Architectures and applications,’’
is currently pursuing the M.Eng. degree with
IEEE Access, vol. 6, pp. 9819–9840, Jan. 2018.
[4] J. Armstrong, ‘‘OFDM for optical communications,’’ J. Lightw. Technol.,
the Department of Electronic Engineering. Her
vol. 27, no. 3, pp. 189–204, Feb. 1, 2009. research interests include visible light communi-
[5] J. Gancarz, H. Elgala, and T. D. C. Little, ‘‘Impact of lighting require- cation and modulation.
ments on VLC systems,’’ IEEE Commun. Mag., vol. 51, no. 12,
pp. 34–41, Dec. 2013.
[6] J. Armstrong and A. J. Lowery, ‘‘Power efficient optical OFDM,’’ Electron.
Lett., vol. 42, no. 6, pp. 370–372, Mar. 2006. FANG YANG (M’11–SM’13) received the B.S.E.
[7] S. C. J. Lee, S. Randel, F. Breyer, and A. M. J. Koonen, ‘‘PAM-DMT and Ph.D. degrees in electronic engineering from
for intensity-modulated and direct-detection optical communication sys- Tsinghua University, Beijing, China, in 2005 and
tems,’’ IEEE Photon. Technol. Lett., vol. 21, no. 23, pp. 1749–1751,
2009, respectively. He is currently an Associate
Dec. 1, 2009.
Professor with the Research Institute of Infor-
[8] J. B. Carruthers and J. M. Kahn, ‘‘Multiple-subcarrier modulation for
nondirected wireless infrared communication,’’ IEEE J. Sel. Areas Com-
mation Technology, Tsinghua University. He has
mun., vol. 14, no. 3, pp. 538–546, Apr. 1996. published over 120 peer-reviewed journal and
[9] S. D. Dissanayake, K. Panta, and J. Armstrong, ‘‘A novel technique to conference papers. He holds over 40 Chinese
simultaneously transmit ACO-OFDM and DCO-OFDM in IM/DD sys- patents and 2 PCT patents. His research interests
tems,’’ in Proc. IEEE GLOBECOM Workshops, Dec. 2011, pp. 782–786. lie in the fields of channel coding, channel estima-
[10] B. Ranjha and M. Kavehrad, ‘‘Hybrid asymmetrically clipped OFDM- tion, interference cancellation, and signal processing techniques for com-
based IM/DD optical wireless system,’’ IEEE/OSA J. Opt. Commun. Netw., munication system, especially in power line communication, visible light
vol. 6, no. 4, pp. 387–396, Apr. 2014. communication, and digital television terrestrial broadcasting. He received
[11] Q. Wang, C. Qian, X. Guo, Z. Wang, D. G. Cunningham, and I. H. White, the IEEE Scott Helt Memorial Award (best paper award in the IEEE
‘‘Layered ACO-OFDM for intensity-modulated direct-detection optical TRANSACTIONS ON BROADCASTING) in 2015. He is the Secretary General of
wireless transmission,’’ Opt. Express, vol. 23, no. 9, pp. 12382–12393, Sub-Committee 25 of the China National Information Technology Standard-
May 2015. ization (SAC/TC28/SC25). He currently serves as an Associate Editor for
[12] A. J. Lowery, ‘‘Enhanced asymmetrically clipped optical ODFM for high the IEEE ACCESS.
spectral efficiency and sensitivity,’’ in Proc. Opt. Fiber Commun. Conf.
Exhibit. (OFC), Mar. 2016, pp. 1–3.
[13] M. S. Islim, D. Tsonev, and H. Haas, ‘‘On the superposition modulation LING CHENG (M’10–SM’15) received the
for OFDM-based optical wireless communication,’’ in Proc. IEEE Global B.Eng. degree (cum laude) in electronics and
Conf. Signal Inf. Process. (GlobalSIP), Orlando, FL, USA, Dec. 2015, information from the Huazhong University of
pp. 1022–1026. Science and Technology in 1995, and the M.Ing.
[14] L. Deng, Y. Rui, P. Cheng, J. Zhang, Q. T. Zhang, and M. Li, ‘‘A unified degree (cum laude) in electrical and electronics
energy efficiency and spectral efficiency tradeoff metric in wireless net- engineering and the D.Ing. degree in electrical
works,’’ IEEE Commun. Lett., vol. 17, no. 1, pp. 55–58, Jan. 2013. and electronics engineering from the University
[15] C. Xiong, G. Y. Li, S. Zhang, Y. Chen, and S. Xu, ‘‘Energy- and spectral- of Johannesburg in 2005 and 2011, respectively.
efficiency tradeoff in downlink OFDMA networks,’’ IEEE Trans. Wireless In 2010, he joined the University of the Witwa-
Commun., vol. 10, no. 11, pp. 3874–3886, Nov. 2011. tersrand, where he was promoted to an Associate
[16] O. Amin, E. Bedeer, M. H. Ahmed, and O. A. Dobre, ‘‘Energy efficiency- Professor in 2015. He has been serving as the Secretary of the IEEE South
spectral efficiency tradeoff: A multiobjective optimization approach,’’
African Information Theory Chapter since 2012. He has been a visiting pro-
IEEE Trans. Veh. Technol., vol. 65, no. 4, pp. 1975–1981, Apr. 2016.
fessor at four universities and the principal advisor for over 30 full research
[17] E. Li, W. Zhang, J. Sun, C.-X. Wang, and X. Ge, ‘‘Energy-spectral
efficiency tradeoff of visible light communication systems,’’ in Proc. post-graduate students. His research interests are in digital communications
IEEE/CIC Int. Conf. Commun. China (ICCC), Jul. 2016, pp. 1–5. and information theory. He has published more than 70 research papers in
[18] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products. journals and conference proceedings. He was a recipient of the Chancellor’s
New York, NY, USA: Academic, 2014. Medal in 2005 and the National Research Foundation rating in 2014. The
[19] S. D. Dissanayake and J. Armstrong, ‘‘Comparison of ACO-OFDM, IEEE ISPLC 2015 Best Student Paper Award was made to his Ph.D. student
DCO-OFDM and ADO-OFDM in IM/DD systems,’’ J. Lightw. Technol., in Austin.
vol. 31, no. 7, pp. 1063–1072, Apr. 1, 2013.

VOLUME 6, 2018 60833