Content uploaded by Chen Chen
Author content
All content in this area was uploaded by Chen Chen on Mar 20, 2021
Content may be subject to copyright.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021 1693
NOMA for Energy-Efficient LiFi-Enabled
Bidirectional IoT Communication
Chen Chen ,Member, IEEE,ShuFu ,XinJian , Min Liu, Xiong Deng ,Member, IEEE,
and Zhiguo Ding ,Fellow, IEEE
Abstract— In this paper, we consider a light fidelity (LiFi)-
enabled bidirectional Internet of Things (IoT) communication
system, where visible light and infrared light are used in the
downlink and uplink, respectively. In order to efficiently improve
the energy efficiency (EE) of the bidirectional LiFi-IoT system,
non-orthogonal multiple access (NOMA) with a quality-of-service
(QoS)-guaranteed optimal power allocation (OPA) strategy is
applied to maximize the EE of both downlink and uplink chan-
nels. We derive closed-form OPA sets based on the identification
of the optimal decoding orders in both downlink and uplink
channels, which can enable low-complexity power allocation.
Moreover, we propose an adaptive channel and QoS-based user
pairing approach by jointly considering users’ channel gains and
QoS requirements. We further analyze the EE and the user
outage probability (UOP) performance of both downlink and
uplink channels in the bidirectional LiFi-IoT system. Extensive
analytical and simulation results demonstrate the superiority
of NOMA with OPA in comparison to orthogonal multiple
access (OMA) and NOMA with typical channel-based power
allocation strategies. It is also shown that the proposed adaptive
channel and QoS-based user pairing approach greatly outper-
forms individual channel/QoS-based approaches, especially when
users have diverse QoS requirements.
Index Terms—Non-orth ogonal multiple access (NOMA), light
fidelity (LiFi), Internet of Things (IoT), energy efficiency (EE),
user outage probability (UOP).
I. INTRODUCTION
WITH the explosive increase of smart devices in our
everyday life, the Internet of Things (IoT) has been
emerging as a promising solution to connect a large num-
ber of devices [1]. The IoT paradigm contains a variety
Manuscript received May 24, 2020; revised November 22, 2020; accepted
January 8, 2021. Date of publication January 18, 2021; date of current version
March 17, 2021. This work was supported in part by the National Natural
Science Foundation of China under Grant 61901065 and Grant 61701054, and
in part by the Fundamental Research Funds for the Central Universities under
Grant 2020CDJQY-A001 and Grant 2020CDJGFWDZ014. The associate
editor coordinating the review of this article and approving it for publication
was M. Safari. (Corresponding author: Chen Chen.)
Chen Chen and Shu Fu are with the School of Microelectronics and
Communication Engineering, Chongqing University, Chongqing 400044,
China, and also with the State Key Laboratory of Integrated Services Net-
works, Xidian University, Xi’an 710071, China (e-mail: c.chen@cqu.edu.cn;
shufu@cqu.edu.cn).
Xin Jian and Min Liu are with the School of Microelectronics and Com-
munication Engineering, Chongqing University, Chongqing 400044, China
(e-mail: jianxin@cqu.edu.cn; liumin@cqu.edu.cn).
Xiong Deng is with the Department of Electrical Engineering, Eindhoven
University of Technology (TU/e), 5600MB Eindhoven, The Netherlands
(e-mail: x.deng@tue.nl).
Zhiguo Ding is with the School of Electrical and Electronic Engineer-
ing, The University of Manchester, Manchester M13 9PL, U.K. (e-mail:
zhiguo.ding@manchester.ac.uk).
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCOMM.2021.3051912.
Digital Object Identifier 10.1109/TCOMM.2021.3051912
of devices such as electronic devices, mobile devices and
industrial devices, and different devices can have different
communication and computation capabilities and quality-
of-service (QoS) requirements [2]. As a key enabling tech-
nology of IoT, communication plays a vital role to connect all
the smart devices supported in the IoT networks. Many radio
frequency (RF)-based techniques have been considered for IoT
communication such as RFID, ZigBee, Bluetooth, WiFi and
5G [3]. Recently, Light Fidelity (LiFi) has been envisioned as
a promising IoT communication technology, which provides
many attractive features that the RF-based IoT networks might
struggle to offer, including accurate device positioning, energy
harvesting from light and inherent physical-layer security [4].
As a lightwave-based communication technology, LiFi aims
to realize a fully networked bidirectional wireless communi-
cation system by exploiting visible light in the downlink and
infrared light in the uplink [5]. In particular, visible light-based
LiFi downlink can be built upon the existing light emitting
diode (LED) fixture which is widely deployed for general
indoor lighting [6], [7].
A. Related Work and Motivation
Although LiFi reveals its potential for future IoT networks,
the research of LiFi-enabled IoT is still at the early stage.
In [8], a LiFi-based hierarchical IoT architecture was proposed
to analyze the collected data and build smart decisions. In [9]
and [10], the energy harvesting issues of LiFi-IoT were
investigated. Lately, a LiFi-IoT system vision was reported
in [4], where the conceptual architecture with four different
types of motes was presented.
Considering that pervasive IoT is usually required to con-
nect a huge number of IoT devices per unit area [4], the LiFi
access point (AP) of an optical attocell in LiFi-IoT networks
should be able to support multiple IoT devices. Therefore,
an efficient multiple access technique is of great significance to
successfully implement LiFi-IoT in practical scenarios. So far,
many multiple access techniques have been introduced for visi-
ble light-based downlink LiFi communication, i.e., visible light
communication (VLC), which can be mainly divided into two
categories: one is orthogonal multiple access (OMA) and the
other is non-orthogonal multiple access (NOMA). For OMA
schemes such as frequency division multiple access/orthogonal
frequency division multiple access (FDMA/OFDMA) and time
division multiple access (TDMA), users are allocated with
different orthogonal frequency or time resources [11]–[13].
Although OMA can eliminate mutual interference between
users, its resource utilization is inefficient. In contrast, NOMA
0090-6778 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1694 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
allows multiple users to simultaneously utilize all the fre-
quency and time resources through power domain superpo-
sition coding (SPC) and successive interference cancellation
(SIC) [14]. Due to its efficient resource utilization, NOMA
has been recognized as a promising multiple access technique
for multi-user VLC systems [15]–[20].
It has been shown that the performance gain of NOMA over
OMA is mainly determined by both the power allocation strat-
egy and the user pairing approach adopted in NOMA [21]. For
NOMA-based multi-user VLC systems, various channel-based
power allocation strategies have been proposed such as gain
ratio power allocation (GRPA) and normalized gain difference
power allocation (NGDPA) [14], [18]. Moreover, channel-
based user pairing approaches have also been proposed to
efficiently divide users into pairs [22], [23]. Nevertheless, all
the aforementioned works only focused on the application
of NOMA in visible light-based LiFi downlink channels.
To apply NOMA in LiFi-enabled bidirectional IoT commu-
nication, the following two important issues should be taken
into consideration:
1) Energy Consumption: In LiFi-enabled bidirectional IoT
communication, energy consumption originates from two
parts: one is the LiFi AP within each optical attocell and
the other is the connected IoT devices. Particularly, most IoT
devices rely on batteries and reducing the energy consumption
to extend their battery life is a top concern [2], [3]. Therefore,
it is of practical significance to design an energy-efficient
multiple access technique for bidirectional LiFi-IoT.
2) Diverse Device QoS Requirements: In practical LiFi-IoT
systems, the connected IoT devices can be generally divided
into two categories: one includes low-speed devices such as
environmental sensors and health monitors, and the other con-
sists of high-speed devices such as multimedia-capable mobile
phones [2], [24]. As a result, it is necessary to take the diverse
QoS requirements of IoT devices into account when designing
a multiple access technique for bidirectional LiFi-IoT.
B. Main Contributions
To address above-mentioned issues when applying NOMA,
in this paper, we propose an energy-efficient NOMA technique
for bidirectional LiFi-IoT communication. The main contribu-
tions of this work are summarized as follows:
•An energy-efficient NOMA technique is applied for
the bidirectional LiFi-IoT system, which adopts a
QoS-guaranteed optimal power allocation (OPA) strategy
to maximize the energy efficiency (EE) of both downlink
and uplink channels. The optimal decoding orders in
the downlink and uplink channels are first identified
and proved, and then the corresponding OPA sets are
obtained in a simple and closed form, which can enable
low-complexity power allocation.
•Three user pairing approaches are studied in the
NOMA-enabled bidirectional LiFi-IoT system, includ-
ing channel-based, QoS-based, and adaptive channel
and QoS-based approaches. The newly proposed adap-
tive channel and QoS-based user pairing approach can
dynamically select from the channel-based and the
QoS-based approaches to achieve a higher EE.
•Both EE and user outage probability (UOP) are ana-
lyzed in the bidirectional LiFi-IoT system using different
multiple access techniques. It is analytically proved that
NOMA with OPA always achieves higher EE than OMA
and NOMA with typical channel-based power allocation
strategies (such as GRPA and NGDPA [14], [18]) in
both downlink and uplink channels. The calculations of
downlink and uplink UOPs are also discussed.
•Extensive analytical and simulation results are presented
to evaluate the performance of different multiple access
techniques in a typical bidirectional LiFi-IoT system. The
obtained results demonstrate the superiority of NOMA
adopting OPA with adaptive channel and QoS-based
user pairing for energy-efficient bidirectional LiFi-IoT
systems.
The remainder of this paper is organized as follows.
Section II presents the system model. The principle of NOMA
and its application for energy-efficient bidirectional LiFi-IoT
communication are described in Section III. The EE and
UOP of the bidirectional LiFi-IoT system using different
multiple access techniques are analyzed in Section IV. Detailed
analytical and simulation results are presented in Section V.
Finally, Section VI concludes the paper.
II. SYSTEM MODEL
We present the basic model of the bidirectional LiFi-IoT
system in this section. The configuration of the bidirectional
LiFi-IoT system is first introduced, and then the light propa-
gation model and the noise model are further discussed.
A. Bidirectional LiFi-IoT System Configuration
In this work, we consider a single-cell bidirectional LiFi-IoT
system,1where visible light is used in the downlink for
simultaneous illumination and communication, while infrared
light is adopted for uplink communication. Fig. 1 illustrates the
geometric configuration of the bidirectional LiFi-IoT system
with one LiFi AP and totally Kusers, i.e., IoT devices. As we
can see, the LiFi AP consists of a visible light LED transmitter
and an infrared light photodiode (PD) receiver, while each user
is equipped with a visible light PD receiver and an infrared
light LED transmitter. In the downlink, the visible light LED
of the LiFi AP radiates white light to provide lighting within
its coverage and broadcast downlink data to all the users at the
same time. Each user detects the broadcasted data by using the
equipped visible light PD. In the uplink, each user employs
the equipped infrared light LED to transmit its uplink data and
the LiFi AP utilizes the infrared light PD to collect the data
from all the users. Therefore, the bidirectional communication
between the LiFi AP and all the users in the LiFi-IoT system
can be established.
Without loss of generality, we assume that both the visible
and infrared light LEDs are point sources and they operate
within their linear dynamic range. We further assume that
1Although a simple single-cell scenario is considered here, the obtained
results are applicable to general multi-cell scenarios since inter-cell interfer-
ence can be efficiently mitigated by applying spectrum partitioning, i.e., allo-
cating adjacent LiFi APs with different subbands/subcarriers in the frequency
domain [25], [26].
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1695
Fig. 1. Geometric configuration of the bidirectional LiFi-IoT system with
one LiFi AP and Kusers, i.e., IoT devices.
the overall system has a flat frequency response.2Moreover,
we also assume that the visible light LED and the infrared
light PD of the LiFi AP are oriented vertically downwards,
while the visible light PD and the infrared light LED of each
user are oriented vertically upwards. For simplicity, we assume
that the visible/infrared light LEDs have the same con-
version efficiency and semi-angle, while the visible/infrared
light PDs have the same responsivity and active area. Under
the above assumptions, the models for visible light down-
link and infrared light uplink channels become exactly the
same [30].
B. Light Propagation Model
In practical LiFi-IoT systems, the visible light PD in the
downlink and the infrared light PD in the uplink can receive
both line-of-sight (LOS) and non-LOS components of the
corresponding transmitted optical signals. Nevertheless, since
the non-LOS component usually has much lower electrical
power than that of the LOS component, it is generally reason-
able to neglect the non-LOS component during most channel
conditions [31]. For simplicity, we only consider the LOS
component in the following channel model.
For the visible and infrared light LEDs with a Lambertian
emission pattern, the LOS direct current (DC) channel gain
between the LiFi AP and the k-th (k=1,2,··· ,K)userfor
both downlink and uplink channels can be calculated by [6],
[32], [33]
hk=⎧
⎨
⎩
(m+1)βρA
2πd2
k
cosm(ψk)gfgccos(φk),0≤φk≤Φ
0,φ
k>Φ
,
(1)
where m=−ln2/ln(cos(Ψ)) denotes the Lambertian emis-
sion order with Ψbeing the semi-angle of the visible/infrared
light LED; βand ρrepresent the current-to-light conversion
efficiency of the visible/infrared light LED and the respon-
sivity of the visible/infrared light PD, respectively; Ais the
2Note that this assumption can be easily ensured by applying efficient
frequency-domain equalization techniques [27]–[29].
active area of the visible/infrared light PD; dkis the distance
between the LiFi AP and the k-th user; ψkand φkdenote the
corresponding emission angle and incident angle, respectively;
gfand gcrepresent the gains of the optical filter and the optical
concentrator, respectively. The gain of the optical concentrator
can be calculated by gc=n2
sin2Φ,wherenis the refractive
index of the optical concentrator and Φis the half-angle field-
of-view (FOV) of the receiver.
Considering the fact that different users might have different
heights in practical LiFi-IoT systems, the locations of users
should be within a three-dimensional (3D) space. For a better
description of the 3D location of a user, the polar coordinate
system is adopted as shown in Fig. 1. In the polar coordinate
system, the 3D location of the k-th user can be represented
by (lk,r
k,θ
k),wherelkand rkrespectively denote its vertical
and horizontal distances from the LiFi AP, and θkdenotes its
polar angle from the reference axis.
As shown in Fig. 1, due to the assumption that the LiFi AP
is oriented vertically downwards while each user is oriented
vertically upwards, the emission angle and the incident angle
corresponding to the LiFi AP and the k-th user become the
same, i.e., ψk=φk. Hence, ψkand φkcan be represented by
ψk=φk=arctan
rk
lk. Moreover, dkcan be expressed by
dk=l2
k+r2
k. Therefore, hkcan be rewritten as follows:
hk=⎧
⎨
⎩
C
l2
k+r2
k
cosm+1 arctan rk
lk,0≤rk
lk≤tanΦ
0,rk
lk>tanΦ
,
(2)
where C=(m+1)βρAgfgc
2π. It can be found from (2) that hkis
dependent on both the vertical distance lkand the horizontal
distance rk, which is not affected by the polar angle θk.
C. Noise Model
The additive noises in both downlink and uplink chan-
nels consist of thermal and shot noises, which are generally
modeled as real-valued zero-mean additive white Gaussian
noises. For simplicity, it is assumed that the additive noises
in both downlink and uplink channels have the same constant
noise power spectral density (PSD) N0. For an overall system
bandwidth Bs, the noise powers in both downlink and uplink
channels can be obtained as Pz=N0Bs.
III. NOMA FOR BIDIRECTIONAL LIFI-IOT
A. Principle of NOMA
Fig. 2 illustrates the conceptual diagrams of
FDMA/OFDMA, TDMA and NOMA with two users, and
the QoS requirements for user 1and user 2are represented
by “QoS 1” and “QoS 2”, respectively. For the case of
FDMA/OFDMA, the subbands/subcarriers are orthogonal to
each other, and therefore the QoS requirement of a specific
user can be ensured by allocating it with a proper number
of subbands/subcarriers, i.e., a proper bandwidth. As shown
in Fig. 2(a), user 1and user 2are allocated with bandwidths
B1and B2to satisfy their QoS requirements, respectively.
Due to the orthogonality of all the subbands/subcarriers, there
is no mutual interference between the transmitted data of
two users. Nevertheless, interference-free multiple access is
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1696 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
Fig. 2. Conceptual diagram of (a) FDMA/OFDMA, (b) TDMA and
(c) NOMA.
achieved by splitting the overall bandwidth into two parts.
For the case of TDMA, as shown in Fig. 2(b), time slots T1
and T2are allocated for user 1and user 2to meet their QoS
requirements, respectively. Since time slots T1and T2are
independent of each other, the transmitted data of two users
are not mutually interfered. Hence, the mutual interference
is eliminated by splitting the overall time resource into
two parts. It can be concluded that the frequency or time
resources must be split and shared so as to support multiple
users without mutual interference for OMA schemes such as
FDMA/OFDMA and TDMA.
In contrast to OMA, NOMA allows both users to utilize
all the frequency and time resources. It can be viewed from
Fig. 2(c) that the transmitted data of user 1and user 2are
superposed in the power domain and there inevitably exists
mutual interference. To ensure their QoS requirements, user
1and user 2are allocated with powers P1and P2, respectively.
It has been well shown that the performance of NOMA
is largely dependent on the adopted power allocation strat-
egy [21]. In conventional NOMA-based systems, the power
allocation strategy is designed to maximize the sum rate of
the system under a total transmit power constraint. However,
when applying NOMA for energy-sensitive IoT applications,
the power allocation strategy should be designed from the
energy consumption perspective.
When there are more than two users in the system,
they are generally divided into multiple pairs and a hybrid
NOMA/OMA scheme can be adopted to support multiple user
pairs [22], [23]. Specifically, the two users within each user
pair are multiplexed in the power domain via NOMA while
different user pairs are multiplexed in the frequency/time
domain via OMA. When applying FDMA/OFDMA to
multiplex different user pairs, each user pair is allocated with
an independent portion of the overall system bandwidth (i.e.,
orthogonal subbands/subcarriers). Hence, there is no interfer-
ence between different user pairs owing to the orthogonality in
the frequency domain. Moreover, due to the assumption that
the overall system has a flat frequency response, the power
requirements of each user pair are not affected by the low-pass
frequency response of the LED transmitters, which can be
individually and independently obtained according to its
specific QoS requirements and channel conditions. Therefore,
the power requirements of all the users in both downlink and
uplink channels of the NOMA-enabled bidirectional LiFi-IoT
system are derived pairwisely in the following two subsections.
B. NOMA for Downlink LiFi-IoT Using Visible Light
In this subsection, NOMA is introduced for downlink
LiFi-IoT communication using visible light. Without loss of
generality, we assume that the bidirectional LiFi-IoT system
serves K=2Nusers,3which are divided into Nuser pairs.
Fig. 3(a) shows the schematic diagram of NOMA-enabled
LiFi-IoT downlink. Let sd
i,f and sd
i,n denote the modulated
message signals intended for the far and near users in the i-th
user pair, respectively. Here, the superscript “d” denotes the
LiFi-IoT downlink, while the subscripts “f”and“n”stand
for the far and near users, respectively. For the two users
within each user pair, intra-pair power domain superposition
is performed. Hence, the superposed electrical signal of the
i-th user pair can be expressed by
xd
i=pd
i,f sd
i,f +pd
i,nsd
i,n,(3)
where pd
i,f and pd
i,n are the downlink electrical transmit
powers allocated to the far and near users in the i-th user
pair, respectively. Hence, the total downlink electrical trans-
mit power allocated to all Npairs of users is obtained by
Pd
elec =N
i=1 pd
i,f +pd
i,n. After intra-pair power domain
superposition, OFDMA-based inter-pair bandwidth allocation
is further performed in the frequency domain, where the
i-th user pair is assumed to be allocated with a bandwidth
Bd
iand thus the required overall system bandwidth is given
by Bd
elec =N
i=1 Bd
i. Subsequently, a DC bias current Id
DC is
added to the resultant signal so as to simultaneously ensure the
non-negativity of the driving signal of the visible light LED
and guarantee sufficient and stable illumination.
After removing the DC term, the received downlink signals
of the far and near users in the i-th user pair can be given by
⎧
⎨
⎩
yd
i,f =hd
i,f (pd
i,f sd
i,f +pd
i,nsd
i,n)+zd
i,f
yd
i,n =hd
i,n(pd
i,f sd
i,f +pd
i,nsd
i,n)+zd
i,n
,(4)
where hd
i,f and hd
i,n denote the downlink channel gains of the
far and near users in the i-th user pair, respectively, and hd
i,f ≤
hd
i,n;zd
i,f and zd
i,n are the corresponding additive noises.
To decode the intended message signals for the far and near
users in the i-th user pair, the decoding order should be first
obtained. Here, we assume that the two users are sorted as a
high priority user and a low priority user based on a specific
sorting criteria, i.e., the decoding order can be generally
given by Od
i,high ≥Od
i,low. The determination of the optimal
decoding order will be discussed in Section III.D. Moreover,
different users might have different QoS requirements in
practical LiFi-IoT systems. Generally, we can define the QoS
requirement of a specific user as its required achievable rate
per bandwidth, i.e., spectral efficiency [15]. Let
Rd
i,high and
3Although an even number of users is considered here, NOMA is generally
applicable to an arbitrary number of users. For an odd number of users, they
are first sorted and then divided into pairs, and the remaining unpaired user
can be allocated with separate power and bandwidth resources [23].
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1697
Fig. 3. Schematic diagram of NOMA-enabled LiFi-IoT (a) downlink and
(b) uplink.
Rd
i,low denote the rate requirements of the high and low priority
users in the i-th user pair in the downlink, respectively. The
corresponding power requirements for both users in the i-th
user pair in the downlink to meet their QoS requirements are
given by the following theorem.
Theorem 1: For the high and low priority users in the i-th
user pair with arbitrary QoS requirements in the downlink of
the bidirectional LiFi-IoT system, the power requirements to
satisfy their QoS requirements are given by
⎧
⎨
⎩
pd
i,high ≥Rd
i,high Rd
i,low
Pz
(hd
i,low)2+Pz
(hd
i,min)2
pd
i,low ≥Rd
i,low
Pz
(hd
i,low)2
,(5)
where Rd
i,high =2
2Rd
i,high −1,Rd
i,low =2
2Rd
i,low −1,hd
i,min =
min{hd
i,high,h
d
i,low}and Pzis the noise power defined in
Section II.C.
Proof: Please refer to the appendix.
Theorem 1 demonstrates that the QoS requirements of the
high and low priority users in the i-thuserpairinthedownlink
can be guaranteed under the impact of mutual interference by
allocating them with proper powers.
C. NOMA for Uplink LiFi-IoT Using Infrared Light
Besides the downlink LiFi-IoT communication as discussed
above, NOMA can also be applied for uplink LiFi-IoT com-
munication using infrared light.4The schematic diagram of
4Although RF technologies can also be applied in the uplink to build a
hybrid system [34]–[37], the use of infrared light in the uplink has attractive
advantages such as no electromagnetic interference radiation and potentially
high data rate [4]. Moreover, the eye safety issue can be addressed by carefully
setting the emit power and the FOV of the infrared light, and the blockage
issue can be tackled by adopting hybrid WiFi/LiFi transmission [5], [36].
NOMA-enabled LiFi-IoT uplink is plotted in Fig. 3(b). Let su
i,f
and su
i,n be the modulated uplink message signals intended for
the LiFi AP from the far and near users in the i-th user pair,
respectively, with the superscript “u” denoting the LiFi-IoT
uplink. The electrical signals to be transmitted by the infrared
light LEDs of the far and near users in the i-th user pair can
be expressed by
xu
i,f =pu
i,f su
i,f +Iu
DC,i,f
xu
i,n =pu
i,nsu
i,n +Iu
DC,i,n
,(6)
where pu
i,f and pu
i,n are the uplink electrical transmit powers of
the far and near users in the i-th user pair, respectively; Iu
DC,i,f
and Iu
DC,i,n are the DC bias currents added to guarantee the
non-negativity of the driving signals of the infrared LEDs.
The total uplink electrical transmit power of all Npairs of
users is given by Pu
elec =N
i=1 pu
i,f +pu
i,n.FortheNpairs
of users in the uplink, due to the asynchronous transmis-
sion nature of LiFi uplink channels, FDMA-based inter-pair
bandwidth allocation is carried out in the frequency domain.
Assuming the i-th user pair is allocated with a bandwidth
Bu
i, the required overall system bandwidth can be obtained by
Bu
elec =N
i=1 Bu
i.
At the LiFi AP, the received uplink signal of the i-th user
pair after removing the DC term can be expressed by
yu
i=hu
i,f pu
i,f su
i,f +hu
i,npu
i,nsu
i,n +zu
i,(7)
where hu
i,f and hu
i,n are the uplink channel gains of the far and
near users in the i-th user pair, respectively, and hu
i,f ≤hu
i,n;
zu
iis the corresponding additive noise.
Similarly, we assume that the far and near users in the
i-th user pair in the uplink are sorted as a high priority user
and a low priority user with the decoding order Ou
i,high ≥
Ou
i,low. In addition, the QoS requirements of the high and low
priority users in the i-thuserpairaregivenby
Ru
i,high and
Ru
i,low, respectively. The following theorem gives the power
requirements for the high and low priority users in the i-th
user pair in the uplink to meet their QoS requirements.
Theorem 2: For the high and low priority users in the
i-th user pair with arbitrary QoS requirements in the uplink
of the bidirectional LiFi-IoT system, the required powers to
meet their QoS requirements are given by
⎧
⎨
⎩
pu
i,high ≥Ru
i,high(Ru
i,low +1) Pz
(hu
i,high)2
pu
i,low ≥Ru
i,low
Pz
(hu
i,low)2
,(8)
where Ru
i,high =2
2Ru
i,high −1,Ru
i,low =2
2Ru
i,low −1and Pzis
the corresponding noise power.
Proof: Please refer to the appendix.
It is demonstrated by Theorem 2 that, by adopting a
proper power allocation strategy, the QoS requirements of the
mutually interfered high and low priority users in the i-th user
pair in the uplink can be successfully guaranteed.
D. QoS-Guaranteed Optimal Power Allocation (OPA)
For the efficient implementation of NOMA in bidirectional
LiFi-IoT systems, a QoS-guaranteed OPA strategy is derived
in this subsection. In practical LiFi-IoT systems, each user
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1698 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
has its own QoS requirements, i.e., rate requirements, in both
downlink and uplink channels. As a result, a bidirectional
LiFi-IoT system is considered working properly, as long as
the QoS requirements of all the users are satisfied. This is
quite different from conventional NOMA-based systems, since
their common goal is to maximize the achievable sum rate
of all the users under a total transmit power constraint [38].
As energy consumption is a very important factor that needs
to be considered when designing an IoT system, it is of
practical significance to reduce the energy consumption of
the system without compromising its working performance.
From the energy consumption perspective, the goal to design
NOMA-enabled bidirectional LiFi-IoT systems is to maxi-
mize the EE in both downlink and uplink channels via a
QoS-guaranteed OPA strategy. Generally, the EE (ηb) with unit
bits/J/Hz in the downlink/uplink channel can be defined as the
ratio of the achievable sum rate (Rb) to the total electrical
power consumption (Pb
elec):
ηb=Rb
Pb
elec
,(9)
where Pb
elec =N
i=1 pb
i,f +pb
i,n,andb∈{d,u}with “d” and
“u” denoting the downlink and uplink channels, respectively.
For the NOMA-enabled bidirectional LiFi-IoT system with
Npairs of users, the target sum rate of the downlink/uplink
channel can be calculated by Rb=N
i=1
Rb
i,f +
Rb
i,n with b∈
{d,u},where
Rb
i,f and
Rb
i,n denote the rate requirements of the
far and near users in the i-th user pair in the downlink/uplink
channel, respectively.
1) Optimal Decoding Order: To obtain the desired OPA
strategy, the optimal decoding orders for the far and near
users in the i-th user pair in both downlink and uplink
channels should be first identified. Based on the derived power
requirements in (5) and (8), the optimal decoding orders are
given by the following proposition.
Proposition 1: The optimal decoding orders for the far and
near users in the i-th user pair in the downlink and uplink
channels of the bidirectional LiFi-IoT system are given by
Od
i,f ≥Od
i,n and Ou
i,f <Ou
i,n, respectively.
Proof: Please refer to the appendix.
According to Proposition 1, the power requirements for the
far and near users in the i-th user pair in the downlink channel
of the bidirectional LiFi-IoT system can be obtained by
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
pd
i,f ≥Rd
i,f Rd
i,n
Pz
(hd
i,n)2+Pz
(hd
i,f )2
pd
i,n ≥Rd
i,n
Pz
(hd
i,n)2
,(10)
and the power requirements for the far and near users in the
i-th user pair in the uplink channel is further given by
⎧
⎪
⎪
⎨
⎪
⎪
⎩
pu
i,f ≥Ru
i,f
Pz
(hu
i,f )2
pu
i,n ≥Ru
i,n(Ru
i,f +1) Pz
(hu
i,n)2.
(11)
2) Problem Formulation: Let Pb
i={pb
i,f ,p
b
i,n}with
b∈{d,u}denote the power allocation sets for the far and
near users in the i-th user pair in the downlink/uplink channel.
To obtain the desired QoS-guaranteed OPA strategy (i.e.,
optimal Pd,OPA
iwith i=1,2,··· ,N) in the downlink channel,
the EE maximization problem can be formulated as
max
{Pd,OPA
1,··· ,Pd,OPA
N}
ηd
s.t. C1: (10)
C2: Pd
elec ≤Pd
max,(12)
where constraint “C1” is to guarantee the power require-
ments of all the downlink users so as to meet their QoS
requirements and constraint “C2” is that the total downlink
electrical transmit power of the LiFi AP should not exceed its
maximum value Pd
max. Similarly, to obtain optimal Pu,OPA
iwith
i=1,2,··· ,N, the EE maximization problem in the uplink
channel can be formulated as
max
{Pu,OPA
1,··· ,Pu,OPA
N}
ηu
s.t. C3: (11)
C4: pu
i,f ≤pu
max
C5: pu
i,n ≤pu
max,(13)
where constraint “C3” is to ensure the power requirements
of all the uplink users, and constraints “C4” and “C5” are
imposed to ensure that the uplink electrical transmit power
does not exceed the maximum value pu
max.
Considering the fact that each user in the bidirectional
LiFi-IoT system normally has its fixed downlink/uplink QoS
requirement during a certain period of time, the target sum
rate Rbwith b∈{d,u}in the downlink/uplink channel
can be viewed as a fixed value during this time period.
Therefore, the EE maximization problems in (12) and (13)
can be respectively transformed into two power minimization
problems which are given as follows:
min
{Pd,OPA
1,··· ,Pd,OPA
N}
Pd
elec
s.t. C1: (10)
C2: Pd
elec ≤Pd
max,(14)
min
{Pu,OPA
1,··· ,Pu,OPA
N}
Pu
elec
s.t. C3: (11)
C4: pu
i,f ≤pu
max
C5: pu
i,n ≤pu
max.(15)
3) Optimal Solution: According to (10) and (11), given the
channel gains, QoS requirements and noise power, the power
requirements for the far and near users in the i-th user
pair in both downlink and uplink channels are independent
from each other. Hence, the optimal solutions for the above
power minimization problems are that the far and near users
in the i-th user pair in both downlink and uplink channels
are allocated with minimum powers to satisfy their QoS
requirements. Therefore, the closed-form OPA sets Pb,OPA
i=
{pb,OPA
i,f ,p
b,OPA
i,n }with b∈{d,u}can be obtained as follows:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
pd,OPA
i,f =Rd
i,f Rd
i,n
Pz
(hd
i,n)2+Pz
(hd
i,f )2
pd,OPA
i,n =Rd
i,n
Pz
(hd
i,n)2
,(16)
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1699
⎧
⎪
⎪
⎨
⎪
⎪
⎩
pu,OPA
i,f =Ru
i,f
Pz
(hu
i,f )2
pu,OPA
i,n =Ru
i,n(Ru
i,f +1) Pz
(hu
i,n)2
.(17)
Although the optimal solution is obtained for a fixed target
sum rate, it is generally applicable to dynamic LiFi-IoT
systems with variable user QoS requirements. In practical
applications, a central unit can be deployed to adaptively
update the optimal solution according to the change of the
channel information and QoS requirements of users.
According to (16) and (17), it is interesting to see that the
minimum total electrical transmit powers in both downlink and
uplink channels, i.e., Pb,OPA
elec,min with b∈{d,u}, can be expressed
by a unified formula which is given as follows:
Pb,OPA
elec,min =
N
i=1
Rb
i,f
Pz
(hb
i,f )2+Rb
i,n(Rb
i,f +1) Pz
(hb
i,n)2,(18)
and therefore, given the same rate requirements and channel
conditions of each user pair, the required minimum total
electrical transmit powers become exactly the same for both
downlink and uplink channels. It should be noted that the
QoS-guaranteed OPA strategy is obtained by assuming that
the 2Nusers are divided into Npairs. Hence, efficient
user pairing should be first performed before executing the
QoS-guaranteed OPA strategy within each user pair.
E. User Pairing
As a typical IoT system usually consists of many users,
i.e., IoT devices, it is efficient to divide these users into pairs5
and a hybrid NOMA/OMA scheme can be adopted to support
multiple user pairs. Specifically, NOMA is adopted for the
two users within each user pair, while OMA is applied for
different user pairs. Hence, user pairing plays an important role
in NOMA-based systems [39]–[41]. In this subsection, three
user pairing approaches are discussed to efficiently divide 2N
users into Nuser pairs.
1) Channel-Based User Pairing: Channel-based user pair-
ing is the most popular user pairing approach in NOMA-based
systems [15], [22], [23]. The key to implement channel-based
user pairing is to pair the two users which have more distinc-
tive channel conditions. For the channel-based user pairing,
the 2Nusers are sorted based on their channel gains in the
ascending order:
hb
1≤···≤ hb
k≤···≤ hb
2N,(19)
where hb
kwith b∈{d,u}is given in (2). After that, the sorted
2Nusers are divided into two groups: the first group Gb,c
1
contains the first half of the sorted users starting from user 1to
user N, with the superscript “c” standing for channel-based
user pairing; the second group Gb,c
2consists of the second
half starting from user N+1 to user 2N. Hence, user
pairing can be performed in the following manner: Ub,c
i=
{Gb,c
1(i),G
b,c
2(i)}, i.e., the i-th user pair Ub,c
icontains both the
i-th user in Gb,c
1and the i-th user in Gb,c
2with i=1,2,··· ,N.
5Considering the computational complexity and time delay at the receiver
side, it is generally assumed that only two users are multiplexed in the power
domain [21].
Algorithm 1 Adaptive channel and QoS-based user pairing
1: Input: hb
k,
Rb
k,Pz,b∈{d,u},k=1,2,··· ,2N
2: Output: optimal user pair Ub,opt
i,i=1,2,··· ,N
3: Step 1: channel-based user pairing
4: Sort {hb
k}k=1,2,··· ,2Nin ascending order
5: Divide the sorted users into Gb,c
1and Gb,c
2
6: Obtain Ub,c
i={Gb,c
1(i),G
b,c
2(i)},i=1,2,··· ,N
7: Calculate Pb,c
elec,min using Ub,c
iand (18)
8: Step 2: QoS-based user pairing
9: Sort {
Rb
k}k=1,2,··· ,2Nin descending order
10: Divide the sorted users into Gb,q
1and Gb,q
2
11: Obtain Ub,q
i={Gb,q
1(i),G
b,q
2(i)},i=1,2,··· ,N
12: Calculate Pb,q
elec,min using Ub,q
iand (18)
13: Step 3: adaptive selection
14: for i=1to Ndo
15: if Pb,c
elec,min ≤Pb,q
elec,min then
16: Ub,opt
i=Ub,c
i
17: else
18: Ub,opt
i=Ub,q
i
19: end if
20: end for
Nevertheless, the channel-based user pairing approach only
takes the channel conditions of different users into account,
while their specific QoS requirements are not considered.
2) QoS-Based User Pairing: In practical bidirectional
LiFi-IoT systems, the supported users, i.e., IoT devices, might
have their own distinctive QoS requirements in both downlink
and uplink channels. Hence, the impact of users’ QoS require-
ments should be considered when performing user pairing. For
the QoS-based user pairing, all the 2Nusers are sorted based
on their QoS requirements in the descending order:
Rb
1≥···≥
Rb
k≥···≥
Rb
2N,(20)
where
Rb
kwith b∈{d,u}denotes the QoS requirement,
i.e., rate requirement, of the k-th user. Similarly, the first
half and the second half of the sorted 2Nusers can be
divided into two groups which are respectively denoted as
Gb,q
1and Gb,q
2, with the superscript “q” standing for QoS-based
user pairing. Hence, QoS-based user pairing can be given as
follows: Ub,q
i={Gb,q
1(i),G
b,q
2(i)}with i=1,2,··· ,N.
3) Adaptive Channel and QoS-Based User Pairing: In both
the channel-based and the QoS-based user pairing approaches,
only a single factor (i.e., channel gain or QoS requirement) is
considered for all the users. However, due to the randomness
of channel gains caused by random users’ locations and the
randomness of users’ QoS requirements, both channel gains
and QoS requirements of users should be considered when
putting them into pairs.
In the following, an adaptive channel and QoS-based user
pairing approach is proposed, which takes both users’ channel
gains and QoS requirements into consideration. The detailed
procedures to perform adaptive channel and QoS-based user
pairing are summarized in Algorithm 1, which includes three
steps. In the first step, channel-based user pairing is performed
and the corresponding minimum power requirement Pb,c
elec,min
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1700 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
is calculated based on the obtained user pairs Ub,c
iand (18).
In the second step, QoS-based user pairing is conducted and
the corresponding minimum power requirement Pb,q
elec,min is
obtained by utilizing Ub,q
iand (18). In the third step, the user
pairing approach which requires a lower minimum power is
adaptively selected as the optimal one for the bidirectional
LiFi-IoT system. Considering the randomness of users’ loca-
tions and/or users’ QoS requirements, the proposed adaptive
user pairing approach is able to dynamically select the optimal
one from the channel-based and the QoS-based user pairing
approaches to achieve a lower minimum power consumption
and hence a higher EE.
IV. PERFORMANCE ANALYSI S
To fairly evaluate the performance of the bidirectional
LiFi-IoT system using various multiple access techniques,
we here adopt EE and UOP as two metrics for performance
evaluation, which are analyzed in the following.
A. Analysis of EE
1) EE Using NOMA With OPA: For the bidirectional
LiFi-IoT system using NOMA with OPA, the target sum
rate in the downlink/uplink channel is given by Rb=
N
i=1
Rb
i,f +
Rb
i,n with b∈{d,u}and the corresponding
minimum total electrical transmit power requirement Pb,OPA
elec,min
is derived in (18). Hence, according to (9), the EE of both
downlink and uplink channels using NOMA with OPA can be
obtained by the following unified formula:
ηb
OPA =N
i=1
Rb
i,f +
Rb
i,n
N
i=1 Rb
i,f
Pz
(hb
i,f )2+Rb
i,n(Rb
i,f +1) Pz
(hb
i,n)2
.(21)
2) EE Using NOMA With Channel-Based Power Allocation:
To compare the performance of NOMA with OPA and conven-
tional NOMA with channel-based power allocation, the fol-
lowing two typical channel-based power allocation strategies
are considered: (i) GRPA [14] and (ii) NGDPA [18]. Letting
αb
i=pb,NOMA
i,n
pb,NOMA
i,f
denote the power allocation ratio between the
near user and the far user in the i-th user pair using NOMA
with b∈{d,u}, the power allocation ratio using GRPA and
NGDPA can be expressed by
αb
i=⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
hb
i,f
hb
i,n 2
,GRPA
hb
i,n −hb
i,f
hb
i,n
,NGDPA
.(22)
To satisfy the QoS requirements of both the far and near
users in i-th user pair, the minimum required electrical transmit
powers using NOMA with GRPA and NGDPA can be given by
(pb,NOMA
i,f ,p
b,NOMA
i,n )=⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
pb,OPA
i,f ,α
b
ipb,OPA
i,f ,α
b
i≥pb,OPA
i,n
pb,OPA
i,f
pd,OPA
i,n
αb
i
,p
b,OPA
i,n ,α
b
i<pb,OPA
i,n
pb,OPA
i,f
.
(23)
Using (23), the minimum total electrical transmit power using
NOMA with GRPA and NGDPA can be obtained. It can be
clearly observed that the minimum required electrical transmit
power using NOMA with GRPA and NGDPA is always larger
or equal to that using NOMA with OPA. Therefore, the EE
using NOMA with GRPA and NGDPA is always lower or
equal to that using NOMA with OPA in both downlink and
uplink channel, which is omitted here for brevity.
3) EE Using OMA: For the bidirectional LiFi-IoT system
using OMA techniques such as FDMA/OFDMA and TDMA,
as shown in Fig. 2, the QoS requirements of different users are
satisfied by allocating them with different frequency or time
resources.
For the far and near users in the i-thuserpairinboth
downlink and uplink channels using NOMA, the required
rates to meet their QoS requirements are denoted by
Rb
i,f
and
Rb
i,n with b∈{d,u}, respectively. When OMA is
applied, to achieve the same individual rates as using NOMA,
the required rates for both the far and near users in the i-th
user pair is given by
Rb,OMA
i=
Rb
i,f +
Rb
i,n. Since the far and
near users in the i-th user pair are not mutually interfered,
their achievable rates can be calculated by
⎧
⎪
⎪
⎨
⎪
⎪
⎩
Rb,OMA
i,f =1
2log21+(hb
i,f )2pb,OMA
i,f
Pz
Rb,OMA
i,n =1
2log21+(hb
i,n)2pb,OMA
i,n
Pz,(24)
where pb,OMA
i,f and pb,OMA
i,n denote the electrical transmit powers
allocated to the far and near users in the i-th user pair,
respectively. By denoting Rb
i=2
2(Rb
i,f +Rb
i,n)−1with b∈
{d,u}, to ensure their QoS requirements, the following power
requirements should be satisfied:
⎧
⎪
⎪
⎨
⎪
⎪
⎩
pb,OMA
i,f ≥R
b
i
Pz
(hb
i,f )2
pb,OMA
i,n ≥R
b
i
Pz
(hb
i,n)2
,(25)
and the minimum required electrical transmit powers for the
far and near users in the i-th user pair in the downlink/uplink
channel can be given by
⎧
⎨
⎩
pb,OMA,min
i,f =Rb
i
Pz
(hb
i,f )2
pb,OMA,min
i,n =Rb
i
Pz
(hb
i,n)2
.(26)
As a result, the required minimum total electrical transmit
power in the downlink/uplink channel using OMA is obtained
by
Pb,OMA
elec,min =
N
i=1
Rb
i
Pz
(hb
i,f )2+Rb
i
Pz
(hb
i,n)2,(27)
and the EE of the downlink/uplink channel using OMA is
given as follows:
ηb
OMA =N
i=1
Rb
i,f +
Rb
i,n
N
i=1 Rb
i
Pz
(hb
i,f )2+Rb
i
Pz
(hb
i,n)2
.(28)
Based on the derived EE of the downlink/uplink channel
in the bidirectional LiFi-IoT system using NOMA with OPA
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1701
Algorithm 2 Calculation of downlink UOP Pd
out
1: Input: hd
k,
Rd
k,Pz,Pd
max,k=1,2,··· ,2N
2: Output: Pd
out
3: Initialize kd
out =0
4: Calculate pd,OPA
i,f and pd,OPA
i,n using (16), i=1,2,··· ,N
5: Sort {pd,OPA
i,f ,p
d,OPA
i,n }i=1,2,··· ,N in descending order
6: Obtain the sorted powers {pd
k}k=1,2,··· ,2N
7: for k=1to 2Ndo
8: Calculate Pd
k=2N
j=kpd
k
9: if Pd
k>P
d
max then
10: kd
out =kd
out +1
11: end if
12: end for
13: Calculate Pd
out =kd
out
2N
and OMA, as given by (21) and (28), we have the following
proposition.
Proposition 2: The EE of the downlink/uplink channel in
the bidirectional LiFi-IoT system using NOMA with OPA is
always larger or equal to that using OMA, i.e., ηb
OPA ≥ηb
OMA,
when
Rb
i,f ,
Rb
i,n,≥0 (bit/s/Hz).
Proof: Please refer to the appendix.
Proposition 2 demonstrates that NOMA with OPA is
more energy-efficient than OMA, which is very suitable for
energy-sensitive IoT applications.
B. Analysis of UOP
Considering the maximum total downlink electrical transmit
power constraint at the LiFi AP and the maximum uplink
electrical transmit power constraint at each user, user outage
might occur in both downlink and uplink channels of the bidi-
rectional LiFi-IoT system. To evaluate the outage performance
of the system, UOP is adopted as the metric in the following
analysis.
For the downlink channel, the minimum required total
electrical transmit power using NOMA with OPA is given by
Pd,OPA
elec,min =
N
i=1
Rd
i,f
Pz
(hd
i,f )2+Rd
i,n(Rd
i,f +1) Pz
(hd
i,n)2.(29)
Due to the maximum total downlink electrical transmit power
constraint at the LiFi AP, Pd,OPA
elec,min cannot exceed its maximum
value Pd
max.IfPd,OPA
elec,min exceeds Pd
max, the LiFi AP will fail to
support all the downlink users. Hence, the LiFi AP can only
connect with a selected subset of the downlink users so as to
meet the maximum total downlink electrical transmit power
constraint. With the goal to let the LiFi AP connect with more
downlink users, it is reasonable to exclude the users which
require the highest powers outside the subset.
Let kd
out denote the number of downlink users that cannot
connect with the LiFi AP, the downlink UOP can be calculated
by Pd
out =kd
out
2N. The detailed procedures to calculate Pd
out are
given in Algorithm 2. Due to the randomness of both users’
locations and QoS requirements, Pd
out is calculated for multiple
times so as to obtain a stable average value. For the uplink
channel, the electrical transmit power of each user should not
TAB LE I
SIMULATION PARAMETERS
exceed the maximum value pu
max. The calculation of the uplink
UOP Pu
out is hence straightforward. We only need to count the
number of uplink users which require a minimum electrical
transmit power higher than pu
max, i.e., ku
out,andPu
out =ku
out
2N.
Similarly, Pu
out is calculated for multiple times to yield a stable
average value.
Following the similar manner, both downlink and uplink
UOPs using NOMA with channel-based power allocation and
OMA can also be achieved, which are omitted here for brevity.
V. P ERFORMANCE EVALUATION AND COMPARISON
A. Simulation Setup
In order to substantiate our derived analytical results,
extensive Monte Carlo simulations are performed. If not
otherwise specified, the simulation parameters of the con-
sidered single-cell bidirectional LiFi-IoT system are listed
in Table I. For the purpose of performance comparison, four
multiple access techniques are considered including: (i) OMA,
(ii) NOMA with GRPA, (iii) NOMA with NGDPA and (iv)
NOMA with OPA. Moreover, three user pairing approaches
as discussed in Section III.E are evaluated which includes:
(i) channel-based user pairing, (ii) QoS-based user pairing
and (iii) adaptive channel and QoS-based user pairing. In the
following evaluation, we assume that perfect channel state
information (CSI) of all the users is available when perform-
ing user pairing and power allocation in the bidirectional
LiFi-IoT system. For simplicity and without loss of generality,
we assume that the overall system bandwidth for both down-
link and uplink channels are the same (i.e., Bd
elec =Bu
elec =
Bs) and equal bandwidth allocation6is considered for different
user pairs (i.e., Bd
i=Bu
i=Bs
Nfor i=1,2,··· ,N).
B. Results and Discussions
In the subsection, we present analytical and simulation
results to evaluate and compare the performance of the consid-
ered single-cell bidirectional LiFi-IoT system using different
multiple access techniques. Without loss of generality, we here
assume that the vertical and horizontal distances of all the
users are uniformly distributed between 1.5 to 2.5 m and 0to
6As a simple but effective bandwidth allocation approach, equal band-
width allocation has been considered in most hybrid NOMA/OMA
systems [22], [23].
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1702 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
Fig. 4. Average downlink/uplink (DL/UL) EE vs. the number of users using NOMA with OPA with different user pairing approaches for (a) R=1,
(b) R={1,2},(c)R={1,2,3}and (d) R={1,2,3,4}.
Fig. 5. Average downlink (DL) EE vs. the number of users using different multiple access techniques with adaptive channel and QoS-based user paring for
(a) R=1,(b)R={1,2},(c)R={1,2,3}and (d) R={1,2,3,4}, and average uplink (UL) EE vs. the number of users using different multiple access
techniques with adaptive channel and QoS-based user paring for (e) R=1, (f) R={1,2},(g)R={1,2,3}and (h) R={1,2,3,4}.
3m, respectively. Moreover, we define
Ras the given QoS
set for both downlink and uplink channels, where the elements
of
Rhave a unit of bits/s/Hz. Both the downlink and uplink
QoS requirements of the users are randomly selected from
the given QoS set
R. In order to obtain stable performance
metrics, including EE and UOP, under random user locations
and QoS requirements, we calculate the average EE and UOP
over totally 10000 independent trials.
We first evaluate the average EE versus the number of
users using NOMA with OPA in the bidirectional LiFi-IoT
system. Since the EE of both downlink and uplink channels
using NOMA with OPA can be expressed by a unified
formula as in (21), the obtained average downlink EE and
uplink EE become nearly the same after 10000 independent
trials. Figs. 4(a)-(d) depict the average downlink/uplink EE
versus the number of users using NOMA with OPA with
different user pairing approaches under various given QoS
sets. When
R=1, as shown in Fig. 4(a), the average
EE decreases with the increase of users for all the three
user pairing approaches. Moreover, the channel-based user
paring approach and the adaptive channel and QoS-based
user paring approach obtain the same average EE, and both
greatly outperform the QoS-based user paring approach. This
is because the QoS requirements of all the user are the same,
and hence random user pairing is achieved when sorting the
users according to their QoS requirements. However, when
diverse QoS requirements are desired by the users, the adaptive
approach achieves higher average EE than the channel-based
approach, as can be seen from Figs. 4(b), (c) and (d). More
specifically, when
R={1,2,3,4}, the QoS-based user paring
approach outperforms the channel-based approach, especially
for a relatively large number of users. It can be found from
Figs. 4(a)-(d) that it is beneficial to consider the impact of
users’ QoS requirements when performing user pairing and
the proposed adaptive approach can be an efficient user pairing
approach for energy-efficient bidirectional LiFi-IoT systems.
It can also be seen from Fig. 4 that the simulation results agree
well with the analytical results.
In the next, we compare the average downlink EE of four
different multiple access techniques utilizing adaptive channel
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1703
Fig. 6. Average downlink (DL) UOP vs. the maximum total DL power using different multiple access techniques with adaptive channel and QoS-based user
paring for (a) R=1,(b)R={1,2},(c)R={1,2,3}and (d) R={1,2,3,4}, and average uplink (UL) UOP vs. the maximum UL power using different
multiple access techniques with adaptive channel and QoS-based user paring for (e) R=1, (f) R={1,2},(g)R={1,2,3}and (h) R={1,2,3,4}.
and QoS-based user paring under various given QoS sets,
as shwon in Figs. 5(a)-(d). As we can see, OMA always
achieves the lowest average downlink EE, while NOMA with
NGDPA outperforms OMA. NOMA with GRPA is shown to
be more energy-saving than NOMA with NGDPA, especially
when users’ downlink QoS requirements are less diverse.
Nevertheless, NOMA with OPA is proven to be the most
energy-efficient one among all the four techniques, which
achieves greatly improved average downlink EE than NOMA
with GRPA when users have diverse downlink QoS require-
ments. Figs. 5(e)-(h) show the average uplink EE of four
different multiple access techniques using adaptive channel
and QoS-based user paring under various given QoS sets.
It can be seen that the lowest average uplink EE is obtained by
OMA, while NOMA with NGDPA outperforms NOMA with
GRPA in the uplink. Furthermore, NOMA with OPA achieves
a significant average uplink EE improvement in comparison to
NOMA with NGDPA, regardless of the number of users and
the diversity of users’ QoS requirements.
Figs. 6(a)-(d) show the average downlink UOP versus the
maximum total downlink power using different multiple access
techniques with adaptive channel and QoS-based user paring
under various given QoS sets. It can be seen that OMA always
obtains the highest average downlink UOP among all the four
multiple access techniques. Moreover, NOMA with GRPA is
shown to be more superior than NOMA with NGDPA in terms
of average downlink UOP, especially for less diverse user QoS
requirements. Among them, NOMA with OPA achieves the
lowest average downlink UOP regardless of the maximum
total downlink power and the diversity of users’ QoS require-
ments. Figs. 6(e)-(h) depict the average uplink UOP versus
the maximum uplink power using NOMA with OPA using
different user pairing approaches under various given QoS sets.
It is shown that OMA performs the worst while NOMA with
OPA performs the best among all the four multiple access
techniques. Moreover, NOMA with GRPA slightly outper-
forms OMA, while NOMA with NGDPA greatly outperforms
NOMA with GRPA and it achieves comparable average uplink
UOP as NOMA with OPA for
R=1. When more diverse QoS
requirements are desired by users, NOMA with OPA gradually
outperforms NOMA with NGDPA. Fig. 6 demonstrates that
NOMA with OPA is more efficient than OMA and NOMA
with GRPA/NGDPA to satisfactorily support multiple users in
both downlink and uplink channels.
VI. CONCLUSION
In this paper, we have proposed an energy-efficient NOMA
technique for bidirectional LiFi-IoT communication, which
adopts a QoS-guaranteed OPA strategy to maximize the EE
of both downlink and uplink channels. We have identified
and proved the optimal decoding orders in both downlink and
uplink channels, and derived closed-form OPA sets. We have
further proposed an adaptive channel and QoS-based user
pairing approach which considers both users’ channel gains
and QoS requirements. The EE and UOP performance of the
bidirectional LiFi-IoT system using different multiple access
techniques has also been analyzed. Extensive analytical and
simulation results show that, compared with OMA and con-
ventional NOMA techniques, the proposed NOMA adopting
OPA with adaptive channel and QoS-based user pairing can
significantly improve both the EE and UOP performance of
the bidirectional LiFi-IoT system.
APPENDIX
A. Proof of Theorem 1
With the decoding order Od
i,high ≥Od
i,low, the high pri-
ority user decodes its message signal directly by treating
the intended message signal for the low priority user as
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1704 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
interference, while the low priority user needs to decode the
intended message signal for the high priority user and apply
SIC to decode its own message signal. Hence, the achievable
rates of the high and low priority users in the i-th user pair
can be given by
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
Rd
i,high =1
2log21+ (hd
i,high)2pd
i,high
(hd
i,high)2pd
i,low +Pz
Rd
i,low =1
2log21+(hd
i,low)2pd
i,low
Pz,(30)
where the scaling factor 1
2is due to the Hermitian sym-
metry [15]. Note that (30) is obtained under the condition
that perfect SIC can be performed for the low priority user.
As discussed in [15], perfect SIC can be achieved when the
low priority user can successfully decode the message intended
for the high priority user, i.e., the achievable rate for the low
priority user to decode the high priority user’s message signal
should be equal or higher than the rate requirement of the
high priority user. Accordingly, the achievable rate for the low
priority user to decode the high priority user’s message signal
is given by
Rd
i,low→high =1
2log21+ (hd
i,low)2pd
i,high
(hd
i,low)2pd
i,low +Pz.(31)
Hence, to meet the rate requirements of both the high
priority user and the low priority user with perfect SIC in the
i-th user pair, the following conditions need to be satisfied:
min{Rd
i,high,R
d
i,low→i,high}≥
Rd
i,high
Rd
i,low ≥
Rd
i,low
.(32)
By denoting Rd
i,high =2
2Rd
i,high −1,Rd
i,low =2
2Rd
i,low −1and
hd
i,min =min{hd
i,high,h
d
i,low}, according to (32), we can obtain
the power requirements of two users as follows:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
pd
i,high ≥Rd
i,high pd
i,low +Pz
(hd
i,min)2
pd
i,low ≥Rd
i,low
Pz
(hd
i,low)2
.(33)
By observing (33), we can easily rewrite it into (5). Therefore,
Theorem 1 is proved.
B. Proof of Theorem 2
Based on the decoding order Ou
i,high ≥Ou
i,low, (7) can be
rewritten as:
yu
i=hu
i,highpu
i,highsu
i,high +hu
i,lowpu
i,lowsu
i,low +zu
i.(34)
Hence, the LiFi AP first decodes the high priority user’s mes-
sage signal directly by treating the low priority user’s message
signal, and then decodes the low priority user’s message signal
after applying SIC to remove the high priority user’s message
signal. Hence, the achievable rates of the high and low priority
users in i-th user pair are obtained as follows:
⎧
⎪
⎪
⎨
⎪
⎪
⎩
Ru
i,high =1
2log21+ (hu
i,high)2pu
i,high
(hu
i,low)2pu
i,low +Pz
Ru
i,low =1
2log21+(hu
i,low)2pu
i,low
Pz.(35)
To meet the QoS requirements for the LiFi AP to success-
fully decode the intended message signals for both the high
and low priority users in i-th user pair, the following conditions
need to be satisfied:
Ru
i,high ≥
Ru
i,high
Ru
i,low ≥
Ru
i,low
.(36)
By denoting Ru
i,high =2
2Ru
i,high −1and Ru
i,low =2
2Ru
i,low −1,
according to (36), we can have the power requirements of the
high and low priority users in i-th user pair:
⎧
⎪
⎪
⎨
⎪
⎪
⎩
pu
i,high ≥Ru
i,high
(hu
i,low)2pu
i,low +Pz
(hu
i,high)2
pu
i,low ≥Ru
i,low
Pz
(hu
i,low)2
.(37)
Substituting (hu
i,low)2pu
i,low +Pz≥(Ru
i,low +1)Pzinto (37)
yields (8). Hence, the proof of Theorem 2 is completed.
C. Proof of Proposition 1
Since there are only two users in the i-th user pair,
the decoding orders for both downlink and uplink channels
can only have two options: i.e., Od
i,f ≥Od
i,n and Od
i,f <Od
i,n
for downlink, and Ou
i,f ≥Ou
i,n and Ou
i,f <Ou
i,n for uplink.
For the downlink with Od
i,f ≥Od
i,n, using (5), the power
requirements can be obtained by
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
pd
i,f ≥Rd
i,f Rd
i,n
Pz
(hd
i,n)2+Pz
(hd
i,min)2
pd
i,n ≥Rd
i,n
Pz
(hd
i,n)2
,(38)
where hd
i,min =hd
i,f . Hence, the total required power of both
the far and near users in the i-th user pair in the downlink
with Od
i,f ≥Od
i,n is obtained by
pd
i=pd
i,f +pd
i,n ≥Rd
i,f Rd
i,n
Pz
(hd
i,n)2
+Rd
i,f
Pz
(hd
i,f )2+Rd
i,n
Pz
(hd
i,n)2.(39)
Similarly, the total required power of both the far and near
users in the i-th user pair in the downlink with Od
i,f <Od
i,n
can also be achieved by
pd,†
i≥Rd
i,f Rd
i,n
Pz
(hd
i,f )2+Rd
i,f
Pz
(hd
i,f )2+Rd
i,n
Pz
(hd
i,f )2.(40)
Since hd
i,f ≤hd
i,n, by observing (39) and (40), we can find that
the minimum power requirement with Od
i,f <Od
i,n is larger
or equal to that with Od
i,f ≥Od
i,n. As a result, Od
i,f ≥Od
i,n is
the optimal decoding order for the downlink.
Next, for the uplink with Ou
i,f ≥Ou
i,n, using (8), the power
requirements can be achieved by
⎧
⎪
⎪
⎨
⎪
⎪
⎩
pu
i,f ≥Ru
i,f (Ru
i,n +1) Pz
(hu
i,f )2
pu
i,n ≥Ru
i,n
Pz
(hu
i,n)2
,(41)
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
CHEN et al.: NOMA FOR ENERGY-EFFICIENT LiFi-ENABLED BIDIRECTIONAL IoT COMMUNICATION 1705
Pb,OMA
elec,min −Pb,OPA
elec,min =
N
i=1
22Rb
i,f 22Rb
i,n −1Pz
(hb
i,f )2+22Rb
i,f −1Pz
(hb
i,n)2.(44)
and the total required power of both the far and near users in
the i-th user pair in the uplink with Ou
i,f ≥Ou
i,n is given by
pu
i=pu
i,f +pu
i,n ≥Ru
i,f Ru
i,n
Pz
(hu
i,f )2
+Ru
i,f
Pz
(hu
i,f )2+Ru
i,n
Pz
(hu
i,n)2.(42)
Similarly, we can also obtain the total power requirement of
both the far and near users in the i-thuserpairintheuplink
with Ou
i,f <Ou
i,n as follows:
pu,†
i≥Ru
i,f Ru
i,n
Pz
(hu
i,n)2+Ru
i,f
Pz
(hu
i,f )2+Ru
i,n
Pz
(hu
i,n)2.(43)
Using hu
i,f ≤hu
i,n and observing (42) and (43), it can be
seen that the minimum power requirement with Ou
i,f ≥Ou
i,n
is larger or equal to that with Ou
i,f <Ou
i,n. Hence, Ou
i,f <
Ou
i,n is the optimal decoding order for the uplink. Therefore,
Proposition 1 is proved.
D. Proof of Proposition 2
By observing (21) and (28), to prove that ηb
OPA ≥ηb
OMA,
we only need to prove Pb,OPA
elec,min ≤Pb,OMA
elec,min. Using (18) and (27),
the difference between Pb,OMA
elec,min and Pb,OPA
elec,min can be given
by (44), as shown on the top of the page.
Due to
Rb
i,f ,
Rb
i,n ≥0, we have 22Rb
i,f ,22Rb
i,n ≥1and
hence we can obtain Pb,OMA
elec,min −Pb,OPA
elec,min ≥0, i.e., Pb,OPA
elec,min ≤
Pb,OMA
elec,min. Therefore, the proof of Proposition 2 is
completed.
REFERENCES
[1] L. Atzori, A. Iera, and G. Morabito, “The Internet of Things: A survey,”
Comput. Netw., vol. 54, no. 15, pp. 2787–2805, Oct. 2010.
[2] L. Da Xu, W. He, and S. Li, “Internet of Things in industries: A survey,”
IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2233–2243, Nov. 2014.
[3] M. R. Palattella et al., “Internet of Things in the 5G era: Enablers,
architecture, and business models,” IEEE J. Sel. Areas Commun., vol. 34,
no. 3, pp. 510–527, Mar. 2016.
[4] I. Demirkol, D. Camps-Mur, J. Paradells, M. Combalia, W. Popoola,
and H. Haas, “Powering the Internet of Things through light communi-
cation,” IEEE Commun. Mag., vol. 57, no. 6, pp. 107–113, Jun. 2019.
[5] H. Haas, L. Yin, Y. Wang, and C. Chen, “What is LiFi?” J. Lightw.
Technol., vol. 34, no. 6, pp. 1533–1544, Mar. 15, 2016.
[6] T. Komine and M. Nakagawa, “Fundamental analysis for visible-light
communication system using LED lights,” IEEE Trans. Consum. Elec-
tron., vol. 50, no. 1, pp. 100–107, Feb. 2004.
[7] H. Burchardt, N. Serafimovski, D. Tsonev, S. Videv, and H. Haas, “VLC:
Beyond point-to-point communication,” IEEE Commun. Mag., vol. 52,
no. 7, pp. 98–105, Jul. 2014.
[8] L. I. Albraheem, L. H. Alhudaithy, A. A. Aljaser, M. R. Aldhafian,
and G. M. Bahliwah, “Toward designing a Li-Fi-based hierarchical IoT
architecture,” IEEE Access, vol. 6, pp. 40811–40825, 2018.
[9] P. D. Diamantoulakis and G. K. Karagiannidis, “Simultaneous lightwave
information and power transfer (SLIPT) for indoor IoT applications,”
in Proc. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2017,
pp. 1–6.
[10] P. K. Sharma, Y.-S. Jeong, and J. H. Park, “EH-HL: Effective
communication model by integrated EH-WSN and hybrid LiFi/WiFi
for IoT,” IEEE Internet Things J., vol. 5, no. 3, pp. 1719–1726,
Jun. 2018.
[11] A. M. Abdelhady, O. Amin, A. Chaaban, B. Shihada, and M.-S. Alouini,
“Downlink resource allocation for dynamic TDMA-based VLC sys-
tems,” IEEE Trans. Wireless Commun., vol. 18, no. 1, pp. 108–120,
Jan. 2019.
[12] J.-Y. Sung, C.-H. Yeh, C.-W. Chow, W.-F. Lin, and Y. Liu, “Orthog-
onal frequency-division multiplexing access (OFDMA) based wireless
visible light communication (VLC) system,” Opt. Commun., vol. 355,
pp. 261–268, Nov. 2015.
[13] M. Hammouda, A. M. Vegni, H. Haas, and J. Peissig, “Resource allo-
cation and interference management in OFDMA-based VLC networks,”
Phys. Commun., vol. 31, pp. 169–180, Dec. 2018.
[14] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat,
“Non-orthogonal multiple access for visible light communica-
tions,” IEEE Photon. Technol. Lett., vol. 28, no. 1, pp. 51–54,
Jan. 1, 2016.
[15] L. Yin, W. O. Popoola, X. Wu, and H. Haas, “Performance evaluation of
non-orthogonal multiple access in visible light communication,” IEEE
Trans. Commun., vol. 64, no. 12, pp. 5162–5175, Dec. 2016.
[16] H. Marshoud, P. C. Sofotasios, S. Muhaidat, G. K. Karagiannidis,
and B. S. Sharif, “On the performance of visible light communication
systems with non-orthogonal multiple access,” IEEE Trans. Wireless
Commun., vol. 16, no. 10, pp. 6350–6364, Oct. 2017.
[17] B. Lin, Z. Ghassemlooy, X. Tang, Y. Li, and M. Zhang, “Experimental
demonstration of optical MIMO NOMA-VLC with single carrier trans-
mission,” Opt. Commun., vol. 402, pp. 52–55, Nov. 2017.
[18] C. Chen, W.-D. Zhong, H. Yang, and P. Du, “On the performance
of MIMO-NOMA-based visible light communication systems,” IEEE
Photon. Technol. Lett., vol. 30, no. 4, pp. 307–310, Feb. 15, 2018.
[19] H. Ren et al., “Performance improvement of NOMA visible light com-
munication system by adjusting superposition constellation: A convex
optimization approach,” Opt. Exp., vol. 26, no. 23, pp. 29796–29806 ,
Nov. 2018.
[20] C. Chen, W.-D. Zhong, H. Yang, P. Du, and Y. Yang, “Flexible-rate
SIC-free NOMA for downlink VLC based on constellation partitioning
coding,” IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 568–571,
Apr. 2019.
[21] Z. Ding, X. Lei, G. K. Karagiannidis, R. Schober, J. Yuan, and
V. K. Bhargava, “A survey on non-orthogonal multiple access for 5G
networks: Research challenges and future trends,” IEEE J. Sel. Areas
Commun., vol. 35, no. 10, pp. 2181–2195, Oct. 2017.
[22] E. M. Almohimmah, M. T. Alresheedi, A. F. Abas, and J. Elmirghani,
“A simple user grouping and pairing scheme for non-orthogonal multiple
access in VLC system,” in Proc. 20th Int. Conf. Transparent Opt. Netw.
(ICTON), Jul. 2018, pp. 1–4, Paper Tu.B2.3.
[23] M. B. Janjua, D. B. da Costa, and H. Arslan, “User pairing and
power allocation strategies for 3D VLC-NOMA systems,” IEEE Wireless
Commun. Lett., vol. 9, no. 6, pp. 866–870, Jun. 2020.
[24] S. R. Teli, S. Zvanovec, and Z. Ghassemlooy, “Optical Internet of
Things within 5G: Applications and challenges,” in Proc. IEEE Int. Conf.
Internet Things Intell. Syst. (IOTAIS), Nov. 2018, pp. 40–45.
[25] H.-S. Kim, D.-R. Kim, S.-H. Yang, Y.-H. Son, and S.-K. Han, “Mit-
igation of inter-cell interference utilizing carrier allocation in visible
light communication system,” IEEE Commun. Lett., vol. 16, no. 4,
pp. 526–529, Apr. 2012.
[26] S.-Y. Jung, D.-H. Kwon, S.-H. Yang, and S.-K. Han, “Inter-cell inter-
ference mitigation in multi-cellular visible light communications,” Opt.
Exp., vol. 24, no. 8, p. 8512, Apr. 2016.
[27] H. Le Minh et al., “100-Mb/s NRZ visible light communications using a
postequalized white LED,” IEEE Photon. Technol. Lett., vol. 21, no. 15,
pp. 1063–1065, Aug. 1, 2009.
[28] Y.-F. Liu, Y.-C. Chang, C.-W. Chow, and C.-H. Yeh, “Equalization and
pre-distorted schemes for increasing data rate in in-door visible light
communication system,” in Proc. Opt. Fiber Commun. Conf. (OFC),
2011, pp. 1–3, Paper JWA083.
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
1706 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021
[29] C. Chen, W.-D. Zhong, and D. Wu, “Indoor OFDM visible light com-
munications employing adaptive digital pre-frequency domain equaliza-
tion,” in Proc. Conf. Lasers Electro-Opt. (CLEO), Jun. 2016, pp. 1–2,
Paper JTh2A.118.
[30] M. D. Soltani, X. Wu, M. Safari, and H. Haas, “Bidirectional user
throughput maximization based on feedback reduction in LiFi networks,”
IEEE Trans. Commun., vol. 66, no. 7, pp. 3172–3186, Jul. 2018.
[31] L. Zeng et al., “High data rate multiple input multiple output (MIMO)
optical wireless communications using white led lighting,” IEEE J. Sel.
Areas Commun., vol. 27, no. 9, pp. 1654–1662, Dec. 2009.
[32] J. M. Kahn and J. R. Barry, “Wireless infrared communications,” Proc.
IEEE, vol. 85, no. 2, pp. 265–298, Feb. 1997.
[33] A. Mostafa and L. Lampe, “Physical-layer security for MISO visible
light communication channels,” IEEE J. Sel. Areas Commun., vol. 33,
no. 9, pp. 1806–1818, Sep. 2015.
[34] H. Tabassum and E. Hossain, “Coverage and rate analysis for co-existing
RF/VLC downlink cellular networks,” IEEE Trans. Wireless Commun.,
vol. 17, no. 4, pp. 2588–2601, Apr. 2018.
[35] V. Papanikolaou, P. Diamantoulakis, Z. Ding, S. Muhaidat, and
G. Karagiannidis , “Hybrid VLC/RF networks with non-orthogonal mul-
tiple access,” in Proc. IEEE Global Commun. Conf. (GLOBECOM),
Dec. 2018, pp. 1–6.
[36] X. Wu, M. D. Soltani, L. Zhou, M. Safari, and H. Haas, “Hybrid LiFi and
WiFi networks: A survey,” 2020, arXiv:2001.04840. [Online]. Available:
http://arxiv.org/abs/2001.04840
[37] H. Abuella et al., “Hybrid RF/VLC systems: A comprehensive
survey on network topologies, performance analyses, applications,
and future directions,” 2020, arXiv:2007.02466. [Online]. Available:
http://arxiv.org/abs/2007.02466
[38] X. Zhang, Q. Gao, C. Gong, and Z. Xu, “User grouping and power
allocation for NOMA visible light communication multi-cell networks,”
IEEE Commun. Lett., vol. 21, no. 4, pp. 777–780, Apr. 2017.
[39] H. Tabassum, M. S. Ali, E. Hossain, M. J. Hossain, and D. I. Kim,
“Uplink vs. downlink NOMA in cellular networks: Challenges
and research directions,” in Proc. IEEE 85th Veh. Technol. Conf.
(VTC Spring), Jun. 2017, pp. 1–7.
[40] L. Zhu, J. Zhang, Z. Xiao, X. Cao, and D. O. Wu, “Optimal user pairing
for downlink non-orthogonal multiple access (NOMA),” IEEE Wireless
Commun. Lett., vol. 8, no. 2, pp. 328–331, Apr. 2019.
[41] V.-P. Bui, P. Nguyen, H. Nguyen, V.-D. Nguyen, and O.-S. Shin,
“Optimal user pairing for achieving rate fairness in downlink NOMA
networks,” in Proc. Int. Conf. Artif. Intell. Inf. Commun., Feb. 2019,
pp. 575–578.
Chen Chen (Member, IEEE) received the B.S. and
M.Eng. degrees from the University of Electronic
Science and Technology of China, Chengdu, China,
in 2010 and 2013, respectively, and the Ph.D. degree
from Nanyang Technological University, Singapore,
in 2017. He was a Post-Doctoral Researcher with
the School of Electrical and Electronic Engineering,
Nanyang Technological University, from 2017 to
2019. He is currently a Tenure-Track Assistant
Professor with the School of Microelectronics and
Communication Engineering, Chongqing University,
Chongqing, China. His research interests include optical wireless communi-
cation, optical access networks, Internet of Things, and machine learning.
Shu Fu received the Ph.D. degree in communication
and information system from the University of Elec-
tronic Science and Technology of China, Chengdu,
China, in June 2016, with a focus on cooperative
multi-point (CoMP) wireless network, QoS routing
in wavelength division multiplexing (WDM) net-
work, and cross-network energy efficiency. He is
currently an Associate Professor with the School
of Microelectronics and Communication Engineer-
ing, Chongqing University, Chongqing, China. His
research interests include the next generation of
wireless networks, integrated networks, and network virtualization.
Xin Jian received the B.Eng. and Ph.D. degrees
from Chongqing University, Chongqing, China,
in 2009 and 2014, respectively. He is currently an
Associate Professor with the School of Microelec-
tronics and Communication Engineering, Chongqing
University. His research interests include the Internet
of Things, next-generation mobile communication,
and wireless ad hoc networks.
Min Liu received the B.Eng. and M.Eng. degrees
from Chongqing University, Chongqing, China,
in 1997 and 2000, respectively, and the Ph.D. degree
from Nanyang Technological University, Singapore,
in 2004. She is currently a Professor and the Vice
Dean of the School of Microelectronics and Com-
munication Engineering, Chongqing University. Her
main research interests include optical fiber commu-
nication and photonic crystal fibers.
Xiong Deng (Member, IEEE) received the M.Eng.
degree in communication and information engineer-
ing from the University of Electronic Science and
Technology of China, in 2013, and the Ph.D. degree
in optical wireless communications from the Eind-
hoven University of Technology, Eindhoven, The
Netherlands, in 2018. In 2013, he was a Researcher
with the Terahertz Science and Technology Research
Center, China Academy of Engineering Physics,
where he was involved in the integrated terahertz
communication and imaging systems. He was a
Guest Researcher with Signify (Philips Lighting) Research, where he was
involved in light fidelity. He is currently a Post-Doctoral Researcher with
the Eindhoven University of Technology. He serves as a Reviewer for
multiple IEEE/OSA journals, including IEEE TRANSACTIONS ON INDUS-
TRIAL ELECTRONICS (TIE), I EEE JOURNAL OF EMERGING AND SELECTED
TOPICS IN POW ER ELECTRONICS (JESTPE), I EEE T RANSACTIONS ON
COMPUTERS (TOC), I EEE TRANSACTIONS ON VEHICULAR TECHNOL-
OGY (TVT), I EEE JOURNAL OF LIGHTWAVE TECHNOLOGY (JLT), IE EE
TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING
(TCCN), IEEE COMMUNICATIONS LETTERS (CL), and IEEE PHOTONICS
JOURNAL (PJ). His research interests include the system modeling, digital
signal processing, and circuits for intelligent lighting, millimeter wave, radio
over fiber, and optical wireless communications.
Zhiguo Ding (Fellow, IEEE) received the B.Eng.
degree in electrical engineering from the Bei-
jing University of Posts and Telecommunications,
in 2000, and the Ph.D. degree in electrical engineer-
ing from Imperial College London, in 2005.
From July 2005 to April 2018, he was worked with
Queen’s University Belfast, Imperial College, New-
castle University, and Lancaster University. Since
April 2018, he has been with The University of
Manchester, as a Professor in communications. Since
October 2012, he has also been an Academic Visitor
with Princeton University. His research interests include B5G networks,
machine learning, cooperative and energy harvesting networks, and statistical
signal processing. He is serving as an Area Editor for IEEE Open Journal
of the Communications Society, an Editor for I EEE TRANSACTIONS ON
VEHICULAR TECHNOLOGY, and IEEE OPEN JOURNAL OF THE SIGNAL
PROCESSING SOCIETY, and was an Editor for IE EE TRANSACTIONS ON
COMMUNICATIONS, IEEE WIRELESS COMMUNICATION LETTERS,and
IEEE COMMUNICATION LETTERS. He received the Best Paper Award in
IET ICWMC-2009 and IEEE WCSP-2014, the EU Marie Curie Fellowship
2012–2014, the Top IEEE TVT Editor 2017, the IEEE Heinrich Hertz Award
2018, the IEEE Jack Neubauer Memorial Award 2018, the IEEE Best Signal
Processing Letter Award 2018, and the Friedrich Wilhelm Bessel Research
Award 2020. He is a Distinguished Lecturer of IEEE ComSoc, and a Web of
Science Highly Cited Researcher in two categories 2020.
Authorized licensed use limited to: CHONGQING UNIVERSITY. Downloaded on March 18,2021 at 01:33:40 UTC from IEEE Xplore. Restrictions apply.
