2021 Hybrid LiFi and WiFi Networks A Survey

1398 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 23, NO.
2, SECOND QUARTER 2021
Hybrid LiFi and WiFi Networks: A Survey

Xiping Wu , Member, IEEE, Mohammad Dehghani Soltani ,
Lai Zhou , Majid Safari , Senior Member, IEEE, and Harald Haas , Fellow, IEEE
Abstract—In order to tackle the rapidly growing number

of mobile devices and their expanding demands for Internet
services, network convergence is envisaged to integrate differ-
ent technology domains. For indoor wireless communications,
one promising approach is to coordinate light fidelity (LiFi) and
wireless fidelity (WiFi), namely hybrid LiFi and WiFi networks
(HLWNets). This hybrid network combines the high-speed data
transmission of LiFi and the ubiquitous coverage of WiFi. In
this article, we present a survey-style introduction to HLWNets,
starting with a framework of system design in the aspects of
network architectures, cell deployments, multiple access and
modulation schemes, illumination requirements and backhaul.
Key performance metrics and recent achievements are then
reviewed to demonstrate the superiority of HLWNets against
stand-alone networks. Further, the unique challenges facing
HLWNets are elaborated on key research topics including user
behavior modeling, interference management, handover and load
balancing. Moreover, the potential of HLWNets in the application
areas is presented, exemplified by indoor positioning and phys- Fig. 1. Vision of network convergence in 5G and beyond.
ical layer security. Finally, the challenges and future research
directions are discussed. will result in intense competitions for available channels. This
Index Terms—Light fidelity (LiFi), wireless fidelity (WiFi), vis- challenges the RF system to meet the exponentially increasing
ible light communication (VLC), radio frequency (RF), optical demand for mobile data traffic, which will increase seven-fold
wireless communication (OWC), hybrid network, heterogeneous between 2017 and 2022 and reach 77.5 exabytes per month
network, network convergence, handover, load balancing.
by the end of 2022.
I. I NTRODUCTION In order to tackle the looming spectrum shortage in RF,
wireless communication technologies employing extremely
HE RECENT visual networking index published by Cisco
T Systems predicts that by 2022, mobile data traffic will
account for 71 percent of Internet protocol traffic, and more
high frequencies have drawn significant attentions. Among
these technologies is light fidelity (LiFi) [2]. Using light
wave as signal bearers, this relatively new technology is able
than 80% of mobile data traffic will occur indoors [1]. This to exploit the vast optical spectrum, nearly 300 THz. LiFi
drives short-range wireless communication technologies such access points (APs) can be integrated into the existing lighting
as wireless fidelity (WiFi) to become a key component in the infrastructure, realizing a dual purpose system which provides
fifth generation (5G) and beyond era. Globally, there will be illumination and communication at the same time. Recent
nearly 549 million public WiFi hotspots by 2022, up from 124 research demonstrates that with a single light-emitting diode
million hotspots in 2017 [1]. Due to the limited spectrum of (LED), LiFi is capable of achieving peak data rates above
radio frequency (RF), the dense deployment of WiFi hotspots 10 Gbps [3]. LiFi offers many other advantages over its
Manuscript received September 3, 2020; revised January 1, 2021; accepted RF counterpart, including: i) a licence-free optical spectrum;
February 4, 2021. Date of publication February 9, 2021; date of current version ii) the ability to be used in RF-restricted areas such as hospi-
May 21, 2021. This work was supported by the Engineering and Physical tals and underwater; and iii) the capability of providing secure
Sciences Research Council (EPSRC) under Grant EP/L020009/1. The work
of Harald Haas was supported by the EPSRC through Established Career wireless communications, as light does not penetrate opaque
Fellowship under Grant EP/R007101/1. (Corresponding author: Xiping Wu.) objects. LiFi also has some limitations as it: i) covers a rela-
Xiping Wu is with the School of Electrical and Electronic Engineering, tively short range, usually a few meters with a single AP; and
University College Dublin, Dublin 4, D04 V1W8, Ireland (e-mail: xip-
ing.wu@ucd.ie). ii) is susceptible to connectivity loss due to obstructions.
Mohammad Dehghani Soltani and Majid Safari are with the Institute for Combining the high-speed data transmission of LiFi and the
Digital Communications, University of Edinburgh, Edinburgh EH9 3JL, U.K. ubiquitous coverage of WiFi, the concept of hybrid LiFi and
(e-mail: m.dehghani@ed.ac.uk; majid.safari@ed.ac.uk).
Lai Zhou is with the Department of Quantum Materials and Devices, the WiFi network (HLWNet) was first mentioned by Rahaim et al.
Beijing Academy of Quantum Information Sciences, Beijing 100193, China in 2011 [15]. Soon later, Stefan and Haas [16] extended the
(e-mail: zhoulai@baqis.ac.cn). research to the integration of LiFi and femtocells. These hybrid
Harald Haas is with the LiFi Research and Development Center, University
of Strathclyde, Glasgow G1 1XQ, U.K. (e-mail: harald.haas@strath.ac.uk) networks are proven to achieve a better network performance
Digital Object Identifier 10.1109/COMST.2021.3058296 than a stand-alone LiFi or RF system [17]. Fig. 1 presents
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
WU et al.: HYBRID LiFi AND WiFi NETWORKS: SURVEY 1399
TABLE I
A C OMPREHENSIVE L IST OF THE E XISTING S URVEY PAPERS ON THE T OPIC OF O PTICAL W IRELESS N ETWORKS √ . N OTATIONS : : S CATTERED
D ISCUSSION ( I . E ., THE C ORRESPONDING C ONTENT I S M ENTIONED B UT N OT IN A D EDICATED S UBSECTION
√√ ). : PARTIAL D ISCUSSION ( I . E ., T HERE I S
AT L EAST O NE D EDICATED S UBSECTION B UT L ACKS AN I N -D EPTH R EVIEW ). : D ETAILED D ISCUSSION
TABLE II
A C LOSE L OOK I NTO THE S URVEY PAPERS ON THE T OPIC OF O PTICAL W IRELESS H YBRID N ETWORKS
a vision of integrating the mainstream wireless networks and topics such as interference management, handover and load
LiFi in 5G and beyond environments. In outdoor scenarios, balancing.
mobile users can be served by satellites, macro/micro cells This article is focused on reviewing the state-of-the-art
or LiFi-enabled street lamps. When moving indoor, they are research in the field of HLWNets, addressing the unique chal-
shifted to a HLWNet for higher quality of service. lenges and discussing the key research directions. The main
So far a number of survey papers have been reported on contributions are:
the topic of optical wireless communications (OWC) and the • providing a framework of system design, which covers
relevant networks. A comprehensive list of these papers is network architectures, cell deployments, multiple access
presented in Table I, in a comparison with our paper. Among and modulation schemes, illumination requirements and
these papers, only a few are related to optical wireless hybrid backhaul.
networks. In [5], RF/FSO (free-space optical communication) • summarizing key performance metrics and reviewing the
hybrid systems were briefly discussed, focusing on optimal reported performance of HLWNets to highlight their
signaling and routing. The authors in [8] summarized the advantages.
opportunities and challenges with respect to the coexistence • introducing the user behavior modeling and its impact on
of LiFi and WiFi, particularly in terms of medium access and the performance of HLWNets, which is underexplored in
modulation schemes. Sarigiannidis et al. [9] focused on the the existing literature.
network level of hybrid networks and gave an overview of the • reviewing and classifying the existing studies on three key
enabling technologies including network function virtualiza- research topics in HLWNets: interference management,
tion (NFV), software-defined ratio (SDR) and software-defined handover and load balancing.
networking (SDN). The authors in [11] discussed load balanc- • studying the benefit of HLWNets to application services
ing and handover in visible light communication (VLC)-aided including the Internet of Things (IoT), indoor positioning
hybrid networks. Obeed et al. [13] elaborated the topic of opti- and physical layer security.
mizing hybrid VLC/RF networks, with the impact of field of • discussing the trends, challenges and research directions
view emphasized. In [14], an overview of hybrid RF/OWC towards practical implementation and future prosperity of
systems was provided, addressing the issues of vertical han- HLWNets.
dover and load balancing. The noted surveys are summarized The remainder of this article is organized as follows. The
in Table II. However, these surveys lack a comprehensive framework of system design for HLWNets is introduced in
overview of optical wireless hybrid networks and do not pro- Section II, and key performance metrics are summarized in
vide in-depth classifications of research work on the significant Section III. User behavior modeling is studied in Section IV.
1400 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 23, NO. 2, SECOND QUARTER 2021
TABLE III
L IST OF ACRONYMS
Fig. 2. Schematic diagram of an SDN-enabled HLWNet.
first approach is to extend the current autonomous network

structure of WiFi to LiFi. The user can freely choose an AP
from either network domain, and the AP can employ any
unoccupied channel. While this approach offers low com-
plexity of network management, the network performance is
compromised. Alternatively, it is feasible to manage the LiFi
and WiFi APs that belong to the same owner via a cen-
tral control unit. Optimal routing and resource allocation can
be achieved in the network level. This architecture is based
on SDN, which decouples the control plane from the data
plane of forwarding devices. The schematic diagram of imple-
menting HLWNets on an SDN platform is demonstrated in
Fig. 2. An SDN-enabled switch connects LiFi and WiFi APs
and extracts key performance indicator information from these
APs through SDN agents. This information is then sent to
an SDN controller, which makes decisions on the routing of
each incoming data packet. Currently, the experimental devel-
opment of HLWNets is still in its infancy. Relevant research
projects and their status on the implementation of HLWNets
are summarized in Table IV.
B. Cell Deployments
The present research related to interference management for While WiFi APs can reach up to 50 m indoors, LiFi APs
LiFi is reviewed in Section V. Handover and load balancing usually cover a relatively small area, only a few meters in
in HLWNets are elaborately discussed in Sections VI and VII, diameter. A proper placement of the LiFi APs, which are
respectively. The advancements of HLWNets in application normally integrated into the ceiling lamps, is important for
services are investigated in Section VIII. The challenges and achieving high-quality network performance. In practice, the
future research directions are addressed in Section IX. Finally, cell deployment is subject to environmental constraints, e.g.,
conclusions are drawn in Section X. The acronyms used in the room shapes. In the current literature, three cell deployment
paper are listed in Table III. models are usually considered: hexagon, matrix and Poisson
point process (PPP).
• The hexagon deployment is an ideal structure of cellular
II. F RAMEWORK OF S YSTEM D ESIGN
networks. It is proven to provide the highest signal-to-
In this section, a framework of HLWNets is introduced in interference-plus-noise ratio (SINR) coverage probability
five aspects: network architectures, cell deployments, multiple in LiFi [18]. Although it is not common to find such a
access and modulation schemes, illumination requirements and structure of lamps in daily life, the hexagon deployment
backhaul. The aim of this section is to provide guidelines for offers an upper bound analysis of performance.
designing HLWNet systems. • The matrix deployment of lamps is widely used due to the
simplicity of installation. This deployment is considered
A. Network Architectures in most studies related to VLC and HLWNets. It is able
In general, LiFi can be incorporated into the existing WiFi to obtain an SINR coverage probability very close to the
system in two basic ways: autonomous and centralized. The hexagon deployment [18].
TABLE IV
HLWN ET- RELATED R ESEARCH P ROJECTS
TABLE V
S TANDARDS FOR W I F I AND L I F I
schemes are summarized from the perspective of standard-

ization in Table V. With respect to multiple access, carrier
sense multiple access/collision avoidance (CSMA/CA) and
time-division multiple access (TDMA) are considered for the
current LiFi standards. Allowing only one link to be active
at a time, CSMA/CA can reduce inter-cell interference (ICI)
to a negligible level. However, the access process for users
is random and not always fair, especially in dense deploy-
ments. Also, unlike WiFi using time division duplex, LiFi
usually adopts visible light for downlink and infrared for
uplink. This might cause overwhelming collisions when the
existing CSMA/CA is used for LiFi. Broadcasting a chan-
Fig. 3. Optimized placements of LiFi APs. nel busy tone was suggested in [22], which can largely
reduce the collision probability. Compared with CSMA/CA,
TDMA is superior in terms of power consumption and band-
• The PPP deployment is employed to simulate randomly width utilization but relies on synchronization and interference
located APs. It is however difficult to achieve uniform management. Orthogonal frequency-division multiple access
illumination with this deployment. Apart from that, the (OFDMA) and non-orthogonal multiple access (NOMA) have
PPP offers the worst SINR coverage probability among also been extensively studied for LiFi. These schemes carry
the three deployments [18]. out a tight coordination of resource assignment in the entire
Given a deployment topology, the distance between APs network and thus require relatively high system complexity. In
renders a trade-off between the handover rate and coverage OFDMA, time-frequency resource blocks are allocated among
probability. The impact of AP separation was studied for the users to enable concurrent transmissions. In NOMA, grouped
matrix deployment in [19], showing the optimal separation as users are served at the same time and frequency but at differ-
a function of the handover overhead and the average speed ent power levels depending upon the channel condition. The
of users. It is concluded that a 3 meter separation is optimal performance gain of NOMA over OFDMA increases when
in most scenarios. In [20], the gradient projection method is the difference in channel conditions is large. Yin et al. [23]
used to find the optimal placement of APs. The regular and found that the LED semi-angle also has a significant impact
optimal placements for the matrix and hexagon deployments on the performance of NOMA in LiFi. It is proven that the
are presented in Fig. 3. For both deployments, the optimal performance gain of NOMA over orthogonal multiple access
placement shifts the APs outwards modestly in comparison to can be further increased by pairing users with distinctive
the regular placement. This optimal placement can improve channel conditions.
the system throughput by up to 70% [20]. In addition, the In regard to modulation, IEEE 802.15.7 adopts variable
locations of WiFi APs affect the distribution of LiFi users and pulse-position modulation (VPMM) and on-off keying (OOK)
thus influence the network performance of HLWNets. Recent in physical (PHY) I and II, and color shift keying (CSK)
research shows that compared with a random deployment, in PHY III. As single-carrier modulation schemes, OOK
a regular matrix deployment of WiFi APs can increase the and VPMM have relatively low complexity and can support
system throughput by up to 20% [21]. low/medium data rates, from ∼10 kbps to ∼100 Mbps. CSK
is similar to frequency shift keying but uses multiple optical
C. Multiple Access and Modulation Schemes
1 The 802.11 standards prior to 1999 adopt direct sequence spread spectrum
The multiple access and modulation schemes related to LiFi (DSSS), and those after 2009 employ MIMO-OFDM.
and WiFi have been broadly discussed in [8]. Here these 2 The PHY layer of G.vlc is still under discussion.
TABLE VI
S UMMARY OF BACKHAULING T ECHNOLOGIES
number of factors related to LEDs, including their loca-

tions, orientations, field of view, emission pattern, output
power, bandwidth, color temperature, etc. In [26], a power
allocation scheme was proposed to maximize the multi-user
sum rate under lighting constraints. It is found that higher
Fig. 4. Example of illuminance distribution in a room. data rates can be achieved with higher color temperatures.
Alternatively, the process can be formulated as a multi-
objective optimization problem, which obtains the Pareto front
sources with different wavelengths. This modulation method of the spectral efficiency-illumination region. The authors
is specially tailored for LiFi, allowing PHY III to operate in [27] discovered that giving a large weight to the spectral
between 12 and 96 Mbps. The three above modulation tech- efficiency maximization causes the photometric flickering to
niques can be directly used for LiFi, as they fit the real increase. This signifies that it might be necessary to consider
and non-negative optical signals. On the contrary, orthogo- a control mechanism in the system design, to keep flickering
nal frequency-division multiplexing (OFDM) yields complex within a permissible level for eye-safety and productivity.
and bipolar signals. Real OFDM signals can be built by con-
straining the input vector of the inverse fast Fourier transform E. Backhaul
to have Hermitian symmetry. This process is termed optical
OFDM (O-OFDM). ITU-T G.vlc (i.e., G.9991) is now con- Backhaul is necessary for connecting APs to the core
sidering two forms of O-OFDM: direct current (DC)-biased network. The backhaul for HLWNets is challenging due to
optical OFDM (DCO-OFDM) and asymmetrically clipped three main factors: i) there are a relatively large number
optical OFDM (ACO-OFDM). In DCO-OFDM, signals are of APs; ii) the heterogeneous structure of the network; and
made positive by adding a DC bias. In ACO-OFDM, signals iii) considerable network capacity, where a single AP is capa-
are clipped at zero and only the positive parts are transmit- ble of providing link data rates in the range of Gbps. A number
ted. ACO-OFDM is more power efficient than DCO-OFDM, of technologies have been proposed as a backhaul solution for
except for large constellations such as 1024 quadrature ampli- indoor wireless networks, including power-line communica-
tude modulation (QAM) [24]. With respect to IEEE 802.11bb, tions (PLC), power over Ethernet (PoE), plastic optical fiber
there are two main proposals for PHY: i) to shift the cen- (POF), millimeter wave (mmWave), infrared (IR) and VLC.
tral frequency of the output signals of existing IEEE 802.11 These technologies, which are summarized in Table VI, can
chipsets, or ii) to use the PHY layer from G.vlc. The key be classified into two categories: wired and wireless backhaul.
advantage of using the existing PHY is that it requires the The concept of using the existing electrical wires within
least amount of change to the existing WiFi silicon. The hope buildings as backhaul for VLC was initially proposed by
is that this will greatly reduce any barriers to entry. Komine and Nakagawa in 2003 [28]. A recent work in [29]
introduced a hybrid PLC-VLC architecture to support multi-
user downlink communications, with the backbone capable of
D. Illumination Requirements handling data rates up to 1 Gbps. PoE is another approach to
The system design of LiFi must comply with illumination provide data transmission and power supply at the same time.
requirements. International organization for standardization Using a cascaded system of PoE and VLC, a dual-hop relaying
(ISO) on light and lighting specifies illuminance of 300 to transmission was proposed in [30], also with a 1 Gbps back-
1500 lx for office work. Komine and Nakagawa [25] mathe- baul. The third option of wired backhaul is POF, offering a
matically derived the illuminance distribution in a room. An data rate of several Gbps. In [31], a wide-band signals distribu-
example is demonstrated in Fig. 4, where the locations and tion network was reported by combining POF and LED-based
output power of LiFi APs are designed in a way so that VLC, where frequency division multiplexing is used on a 50 m
the minimum requirement of 300 lx is satisfied at the cor- POF to transmit signals to different APs.
ners, while the peak illuminance does not exceed 1500 lx. As Compared with wired backhaul techniques, wireless solu-
shown, the illuminance peaks in the center of the room but is tions provide a more flexible installation at a higher cost
significantly low at the corners. of hardware. The authors in [32] demonstrated a multi-Gbps
Optimizing LEDs for better network performance under point-to-point backhaul connectivity on the basis of millime-
illumination requirements is tricky. This process involves a ter wave (mmWave). In [33], a VLC-based backhaul solution
was proposed for LiFi, using the in-band full-duplex technique a higher spectral efficiency. With this approach, HLWNets
for the access and backhaul links. In this work, relaying pro- can effectively improve the spectral efficiency than operating
tocols such as amplify-and-forward and decode-and-forward the two networks in a stand-alone fashion, with an increase
are used to realize dual-hop transmission. In [34], both visible between 10% and 30% [17].
light and infrared bands are employed to compose the back-
haul solution, with the interference between inter-backhaul C. Area Spectral Efficiency
and backhaul-to-access network characterized. With respect In contrast to WiFi, LiFi can highly reuse the spectral
to HLWNets, resource allocation and network optimization resource in space, since a single LiFi AP only covers a con-
across LiFi and WiFi is feasible when they share the same fined area with a 2-3 meters diameter. To fairly compare the
backhaul. For instance, given a power budget, the power allo- spectral efficiency of LiFi and its RF counterparts, area spec-
cation can be optimized between LiFi and WiFi to enhance tral efficiency (ASE) is defined, which measures the sum of
the network performance [35]. the maximum average data rates per unit bandwidth per unit
area. Stefan et al. [42] showed that LiFi is able to provide an
III. P ERFORMANCE M ETRICS ASE at least 10 times higher than the RF femtocell system. As
A number of metrics are used to evaluate the performance for the hybrid LiFi and RF femtocell system reported in [16],
of wireless networks, including SINR coverage probability, it can increase the ASE by at least two orders of magnitude
spectral efficiency, area spectral efficiency, energy efficiency, over the stand-alone RF network. Considering the impact of
network capacity, quality of service, and user fairness. These light-path blockages, Wang et al. [43] analyzed the ASE of
metrics and relevant studies with respect to HLWNets are HLWNets. It is surprisingly found that modest blockages may
reviewed in this section. have a beneficial effect on the ASE, because the obstacles can
block more interference than the desired optical signals.
A. SINR Coverage Probability
The SINR coverage probability, i.e., the probability that the D. Energy Efficiency
user’s SINR is above a certain threshold, is crucial for pro- While increasing the deployment density of APs could
viding stable connectivity. In regard to the RF system using improve the ASE, the cost of energy consumption also rises
omnidirectional antennas, the received signal power is depen- rapidly. Energy efficiency has thus become a focal point in
dent on the link distance and shadowing. As for LiFi, the ultra-dense networks. There exists a trade-off exists between
user’s orientation also plays a decisive role. Specifically, nor- energy efficiency and spectral efficiency, as a higher spectral
mal incidence gives a peak received signal power, while no efficiency requires more energy per bit. The authors in [44]
signal will be received for incident directions beyond the field analyzed the energy efficiency of OFDM-based VLC systems.
of view (FoV) of photodiodes (PDs). For this reason, changes It is showed that ACO-OFDM is more energy efficient than
in the user’s orientation can significantly reshape the cover- DCO-OFDM, when the spectral efficiency is low, e.g., below
age areas of LiFi APs [36]. While the PD with a large FoV 2 bit/s/Hz. Due to the feasibility of switching users between
widens the range of receiving orientations, it results in receiv- different networks, HLWNets have the potential to improve
ing more interference. Alternative to using a single PD, the energy efficiency. This can be formulated as an optimization
angle diversity receiver (ADR) comprised of multiple narrow- problem of bandwidth and power allocation to maximize
FoV PDs is able to greatly improve the user’s SINR, by 20 dB the energy efficiency of HLWNets. Using this approach, the
∼ 50 dB depending on the combination scheme chosen [37]. authors in [45] demonstrated the superior performance of a
Research has been also conducted to analyze the SINR cover- hybrid RF/VLC network in comparison to an RF-only system,
age probability of HLWNets. In [38], it shows that HLWNets with an improvement up to 75%. In [46], end-to-end energy
can effectively improve the SINR coverage probability over efficiency for a heterogeneous LiFi and RF network was
stand-alone LiFi or WiFi networks, especially for single-PD analysed. This work shows that deploying LiFi attocell APs
receivers with a half angle of view below 45◦ . can reduce the overall power consumption by almost 10%
compared to the mmWave indoor wireless technology.
B. Spectral Efficiency
Spectral efficiency measures how efficiently a certain E. Network Capacity
amount of frequency spectrum is used. An experimental VLC Regarding wireless networks, network capacity measures
system was reported in [39], achieving a spectral efficiency the maximum achievable sum data rate that a network can
of 4.85 bit/s/Hz. This system is based on carrierless ampli- handle under certain constraints, usually a requirement on bit
tude and phase modulation (CAP), a variant of QAM. In [40], error ratio. This metric is paramount for guaranteeing decent
generalized spatial modulation with dimming control is used network performance. Thanks to the integration of wireless
for VLC to obtain a spectral efficiency above 10 bit/s/Hz. technologies of different spectra, HLWNets are capable of
To further enhance the spectral efficiency, the authors in [41] boosting the network capacity, especially in a scenario where
combined DCO-OFDM with adaptive bit loading and experi- WiFi APs are densely deployed. In addition, the existence of
mentally demonstrated a link data rate of 15.73 Gbps. As for WiFi can relieve the capacity degradation of LiFi caused by
HLWNets, it is feasible to transfer the users from one network light-path blockages, as demonstrated in [43]. Maximizing the
to another for a higher value of SINR, and consequently network capacity of HLWNets has attracted a massive amount
Fig. 5. Taxonomy of user behavior modeling in LiFi-related networks.
of research attention in recent years. Due to the coverage requirements on throughput, latency, user experience, etc.
overlap between LiFi and WiFi, This optimization process Hence, it is necessary to ensure that each user receives a fair
essentially involves load balancing, which is elaborated in share of system resources. Typical measures include Jain’s fair-
Section VII. ness index, max-min fairness and quality of experience (QoE)
fairness. Among them Jain’s fairness index, which assesses the
F. Quality of Service throughput fairness among users, is widely used in the cur-
rent literature. Maximizing network capacity only will result
While the metrics in the PHY layer are focused on bits, in
in a resource allocation preference for users with sound chan-
the network layer packets are the unit of data. Several aspects
nel quality. This unfairness becomes particularly pronounced
of packets, including throughput, packet loss ratio, latency
in HLWNets when a large number of users are competing
and jitter, are usually considered for measuring the quality
for limited WiFi resources. In order to enhance the user fair-
of service (QoS). Future indoor wireless network is expected
ness, proportional-fairness schemes are usually considered for
to support applications with diverse QoS requirements. For
allocating the resources in HLWNets. Detailed discussions are
instance, holographic 3D display requires very high data rates
given in Section VII.
above 10 Gbps [47], whereas automated guided vehicle (AGV)
positioning needs ultra low latency below 1 ms [48]. In the
meantime, the prevalence of IoT is significantly increasing the IV. U SER B EHAVIOR M ODELING
number of devices. The more dense the population of devices, As mentioned earlier, the system performance of LiFi and
the higher the average latency and jitter, limiting the types LiFi-involved hybrid networks is substantially affected by
of applications that can be supported by high-density WiFi.1 user-related factors including user mobility, device orientation
The research work in [49] shows that with the participation and light-path blockage. These factors are collectively referred
of LiFi, HLWNets can greatly improve the maximum packet to as user behavior. A taxonomy of these factors and their
arrival rate as well as decreasing the latency. modelling methods is presented in Fig. 5. In this section, the
models used for characterizing user behavior in LiFi-related
G. User Fairness networks are summarized.
The above metrics are all concentrated on the overall
network performance. In practice, users may have different A. User Mobility
1 A location can be classified as high density if more than 30 users are Mobility models have been well studied for examining the
connecting to an AP. features of wireless ad hoc networks. Depending on whether
the movement has a memory and/or restriction, these models

can classified into three categories: random models, models
with dependency, and models with geographic restrictions.
Compared to outdoor, indoor mobility is more arbitrary and
variable. In [50], advanced mobility models were proposed
for indoor scenarios, e.g., a rule-based model which mim-
ics realistic indoor maneuvers inside a building with several
rooms, where users move along specific paths from one room
to another. However, these models consider specific environ-
ments, making it difficult to evaluate the general performance
of a wireless network. Alternatively, random models have
been widely used in the current literature to measure the
network performance of HLWNets. The random waypoint
Fig. 6. Axial rotations of a mobile device.
(RWP) model was initially introduced in [51] to model human
movements in a random manner. The user moves along a
zigzag line from one waypoint to another one, with the data measurements. It is discovered that the coherence time of
waypoints randomly distributed. Between two consecutive the random orientation is in the order of hundreds of millisec-
waypoints, the user move forward in straight line with a con- onds. Indoor optical wireless channels, of which the typical
stant speed. With the original RWP model, the user wanders delay spread is in the order of nanoseconds, can thus be treated
within a large outdoor area and changes its speed when arriv- as slowly-varying channels. Combining the device orientation
ing at each waypoint. The distance between two waypoints model with the RWP mobility model, an orientation-aware
in an indoor scenario is however relatively short. A modified RWP model was first introduced in [56] to provide a real-
RWP model is feasible through keeping the speed constant istic and accurate framework for analyzing the performance
for a short period of time [52]. Research results demonstrate of LiFi. It is demonstrated that this model pronounces the
that the user’s speed has a great impact on access point selec- issue of frequent handovers in HLWNets, in comparison to the
tion in HLWNets [53]. In general, fast-moving users prefer conventional RWP model [59]. The orientation-aware RWP
WiFi, whereas slow-moving users can be served by LiFi. model was later applied for HLWNets in [60] to support
The movement path also plays a vital role in the handover dynamic load balancing and real-time resource allocation for
process [54]. Details are discussed in Section VI. mobile users.
B. Device Orientation
Photodiodes have a limited FoV, restricting the angles at C. Light-Path Blockage
which the device can receive optical signals. Within the range Like millimeter-wave and Terahertz communications, LiFi
of reception angles, the received optical intensity depends on is susceptible to channel blockages, which are caused by
the direction of incident light. This makes the device ori- opaque obstacles such as walls, furniture, human bodies, etc.
entation an important factor that can significantly affect the Researchers are particularly interested in the factor of human
link performance of LiFi. This issue however was not well bodies, since this is closely related to the use of mobile
addressed in the early research on LiFi and HLWNet. In devices. In this case, the light path of a device can be blocked
the existing literature, a fixed device orientation was mostly by the person using the device and other persons around it.
assumed due to the lack of a valid model. A few studies have The human body is usually modeled as a cylinder object or a
been carried out to evaluate the performance of HLWNets with rectangular one [61]. This blockage model can be combined
randomly oriented devices, e.g., [55], based on the Euler’s with the orientation-aware RWP model, establishing a joint
rotation theorem. Specifically, any form of rotation in the R3 model to comprehensively analyze the impact of user behav-
space can be uniquely interpreted by composing three axial ior. A statistical model of blockage is also available [62]. This
rotations in a three-dimensional Cartesian coordinate system, model characterizes blockage with occurrence rate and occu-
as illustrated in Fig. 6. pation rate, which measure how often the blockage occurs and
The first empirical model of the device orientation was how long the blockage lasts, respectively. Several methods are
reported in [56]. The authors managed to acquire real-time feasible to alleviate the performance degradation due to block-
values of axial rotations from smartphones, attributed to the age. In [63], the LED with a wider half-intensity angle is used
embedded gyroscope. The polar angle (i.e., the angle between to enlarge the coverage area. However, this method introduces
the Z-axis and the normal vector of the device) exhibits a more interference. Alternatively, an omni-directional receiver
Laplace distribution for sitting and a Gaussian distribution for that employs PDs on each side of the handset can make it
walking. Experimental measurements of the device orientation robust against blockage [64]. As for HLWNets, the user can
for uncontrolled activities were presented in [57]. It is found be transferred to the WiFi system when experiencing a severe
that in this case, the polar angle of mobile devices better fits a light-path blockage and shifted back once the LiFi connectiv-
Laplace distribution than a Gaussian distribution. In [58], the ity is restored. This process involves vertical handover, which
changes in the device orientation were studied based on the is discussed in Section VI.
Fig. 7. Taxonomy of interference management in LiFi.
V. I NTERFERENCE M ANAGEMENT transmission. MUD aims to cancel interference among co-

Increasing the density of APs is an important aspect of channel users within the same AP. Using block diagonal-
network densification, a key approach for wireless evolution ization, Hong et al. [67] showed that an SINR value of
over the next decade, while interference management is of vital 20 dB can be achieved for two VLC users in the major-
importance. Operating at different spectra, LiFi and WiFi do ity of the indoor region, when single LED’s power is 10
not interfere with each other. Also, the CSMA/CA adopted mW. At the same power level, the zero forcing-based VLC
by WiFi can suppress co-channel interference to a negligi- system in [68] can provide an SINR value of 30 dB, with
ble level. Hence, in this section we focus on discussing the more densely deployed APs. The work in [69] showed that
interference management for LiFi. These techniques can be with zero forcing, ACO-OFDM outperforms DCO-OFDM for
classified into two basic categories: interference cancellation low optical power, the same trend as no precoding used [24].
and interference avoidance. A detailed taxonomy is presented In [70], optical adaptive precoding was studied, which only
in Fig. 7, and a summary table of the current literature is given nulls destructive interference. With reduced dependence on
in Table VII. CSI, this method is more robust to imperfect CSI than channel
inversion precoding. With the aim of eliminating ICI, CoMP
requires coordination among APs to exchange the channel
state information (CSI) knowledge. Relevant research was car-
ried out for LiFi in [71]–[73], with interests in user grouping
A. Interference Cancellation and coordinated beamforming.
Andrews [65] defines interference cancellation as the class Precoding methods rely on channel state information at
of techniques that decode desired information and utilize the transmitter (CSIT) of all co-channel users. However, the
this information along with channel estimates to eliminate or uplink of LiFi usually employs infrared when lighting is not
reduce received interference from the received signal. This needed [74], composing a frequency division duplex system.
type of technique works at the receiver end, i.e., after the As a result, using precoding techniques for LiFi comes at
interference-affected signal is received. a cost of hefty feedback. This issue has not yet been well
1) Precoding: Precoding techniques are widely used to addressed in the current literature. In addition, inaccurate
eliminate interfering signals in downlinks. The basic principle CSI will impair the performance of precoding. This problem
is to artificially create orthogonal channels through singular- becomes more pronounced in LiFi, since rapid changes in the
value decomposition. Due to the non-negativity of optical device orientation can cause fast-varying channels. The prac-
signals, traditional precoding techniques need to be modified ticality of adopting precoding techniques for LiFi is yet to be
to suit LiFi, e.g., adding a DC biasing vector [66]. Precoding validated.
techniques can be divided into two subcategories: multi- 2) Blind Interference Alignment: When exact CSI is not
user detection (MUD) and coordinated multipoint (CoMP) available at the transmitter, blind interference alignment (BIA)
TABLE VII
S UMMARY OF I NTERFERENCE M ANAGEMENT T ECHNIQUES IN L I F I
can be achieved by pairing a time-selective user with a (PIC) detects all users simultaneously and can reduce latency
frequency-selective user. The concept is to maximize the at the cost of increased complexity. In regard to LiFi, SIC
degree of freedom for co-channel users through masking trans- is preferable to PIC, since each LiFi AP covers a relatively
mitted signals on the basis of channel coherence. Unlike small area and is likely to serve only a few users. Using SIC
precoding, BIA can only reduce interference to some extent. to realize multiple access forms the concept of NOMA [23].
Since the condition of channel coherence for BIA may not Note that SIC-based methods rely on an appropriate pair of
always be met, channel manipulation is required. In RF the co-channel users, which might not always be satisfied in
systems this is realized by reconfigurable antennas, which ultra-dense networks due to the sparsity of users in a single
enable the receiver to switch among different antennas. As cell.
for LiFi, it employs PDs as the receiver antennas. A few 4) Specially Tailored Methods: There are two interference
studies were conducted to utilize BIA in LiFi. Equipping cancellation methods specially tailored for LiFi: angle diver-
each user with one PD and multiple optical filters, the BIA sity receiver (ADR) and polarization techniques. The ADR
scheme in [75] requires 3-5 dB less optical transmit power than uses multiple narrow-FoV PDs instead of a single wide-FoV
TDMA. The performance of BIA in an HLWNet was studied PD, in order to reduce interference at each PD. Chen et al. [37]
in [76], showing that shifting LiFi cell-edge users to WiFi can analyzed the performance of different signal combining
allow BIA to obtain a greater gain over TDMA. However, BIA schemes for ADR, including select best combining, equal gain
only outperforms TDMA in the range of high optical transmit combining, maximum ratio combining, and optimum combin-
power, e.g., above 50 dBm in [75]. Such high levels of LED ing. This work shows that with optimum combining, ADR
power are not suitable for illumination, restricting the use of is able to achieve an SINR performance close to that of
BIA for LiFi. interference-free systems. In [78], zero-forcing precoding was
3) Successive Interference Cancellation: Successive combined with ADR. This approach can noticeably improve
interference cancellation (SIC) can detect co-channel signals the SINR performance, especially for LiFi cell-edge users.
by distinguishing their different power levels. It is worth In [79], the optimal structure of ADR was studied, depending
noting that when power control is implemented in LiFi, on the number of PDs and the LED layout. It is shown that
illumination requirements must be satisfied. Since only the choosing an appropriate tilt angle of side PDs can greatly
alternating current (AC) component of optical signals is affect SINR, with a fluctuation range of 20 dB. While capa-
converted to the effective electrical signal, it is feasible to ble of rejecting interference, narrow-FoV PDs are susceptible
adjust the amount of the AC component while keeping the to changes in the device orientation. So far the performance
same level of the average optical transmit power. In [77], validation of ADR in a realistic mobile environment has not
power control was studied when the user is served by multiple been addressed in the existing literature.
APs, with each AP consisting of multiple narrow-FoV LEDs. The polarization property of light can also be exploited
This method is able to obtain an SINR 2-5 dB higher than to realize differential detection for interference cancellation.
TDMA. As SIC detects one user per stage, the computational Specifically, two polarized optical signals with perpendicu-
complexity and latency are proportional to the number of co- lar directions do not interfere with each other, constructing
channel users. Alternatively, parallel interference cancellation an orthogonal division multiplexing. In [80], at the receiver
two PDs with different polarization filters are used to can- for LiFi. Centralized systems also facilitate spectrum schedul-
cel interference. A similar method was proposed in [81] ing in FR, which can exploit the wide spectrum available
to resist unpolarized optical interference. These approaches in LiFi. However, an over-complicate structure of FR would
do not require extra feedback to exchange CSI. However, worsen the problem of resource allocation.
they rely on a perfect alignment of polarization direc- 2) Dense AP Deployment: The LiFi APs are usually densely
tions between the transmitter and receiver. This is feasible deployed, with a separation of 2-3 meters. This makes ADR a
in laboratory experiments but much difficult to implement promising approach for rejecting interference. ADR can also
in practice. improve the received signal strength, as the PD’s sensitivity is
dependent on the incident direction. In contrast, a single wide-
B. Interference Avoidance FoV PD does not perform well when it is significantly tilted.
The interference avoidance refers to the techniques that This drives a momentum in using multiple PDs to construct an
work at the transmitter end to avoid yielding interference. omnidirectional receiver [64]. The dense deployment of LiFi
Among these techniques are orthogonal division schemes also boosts the area spectral efficiency when FR is used. In the
including TDMA [86], OFDMA [87] and space-division meantime, there are sparse users in a typical indoor scenario
multiple access (SDMA) [88]. Some studies, e.g., [89], list such as office and home. For this reason, adjustable spectrum
power control as an interference avoidance method. This allocation can make FR more efficient for LiFi.
type of method is unable to work without SIC, and hence 3) Opportunities in HLWNets: Though WiFi and LiFi do
we classify it as SIC-based interference cancellation. Other not interfere with each other, it does not mean WiFi has
interference avoidance techniques include frequency reuse and no impact on the interference management for LiFi. On the
frequency/time hopping. opposite, a delicate user association in HLWNets can help
1) Frequency Reuse: Frequency reuse (FR) is widely used mitigating interference in LiFi, e.g., based on a conflict graph
to avoid ICI among neighboring cells, where frequencies in [90]. The AP selection between LiFi and WiFi is a key
are reused in a regular pattern. A few studies have been issue in HLWNets and will be discussed in the following
carried out to apply FR to LiFi [82]–[84]. In [82], experi- sections. After user association is determined, an appropri-
mental work was carried out to demonstrate the use of FR ate resource allocation in terms of sub-channel or time slot
in LiFi among three APs, achieving a sum data rate of can also meliorate the interference situation [91]. In summary,
0.5 Mbps. Chen et al. [83] analyzed fractional frequency the interference management for LiFi should be in line with
reuse (FFR) for LiFi, including strict FFR and soft frequency user association and resource allocation in a full picture of
reuse (SFR). The former scheme partitions the cell area into HLWNets.
three equal sectors, while the latter one provides a two-tier
cellular structure. Compared with strict FFR, SFR is more
flexible and thus able to achieve a higher reuse ratio with the
same capability of suppressing interference. Considering dif- VI. H ANDOVER
ferent AP densities, a dynamic SFR scheme was proposed In general, the handover process in a hybrid network falls
in [84], using an adjustable spectrum allocation. This into two categories: horizontal handover (HHO) and vertical
scheme in essence creates a cellular structure more flexible handover (VHO). A HHO takes place within the domain of
than SFR. a single wireless access technology, whereas a VHO occurs
2) Frequency/Time Hopping: Hopping techniques rapidly between different technologies. With a VHO, the air interface
switch a carrier among many frequency channels or time slots, is changed, but the route to the destination remains the same.
using a pseudo-random sequence known to both the transmitter In some literature, e.g., [92], a third category named diago-
and receiver. A time hopping method for LiFi was reported nal handover is introduced, with the air interface and route to
in [85], where the period and duty cycle of the optical carrier the destination both changed. A significant body of research
are varied in a pseudo-random manner. While this type of was conducted on the topic of handover for heterogeneous
method can reduce the probability of two users occupying the networks (HetNets). A relevant survey was carried out in [93],
same time-frequency block, it requires a strict synchronization summarizing different types of handover schemes including
between the transmitter and receiver. received signal strength (RSS)-based, load balancing-related,
and energy-saving. Handling the handover process is more
challenging in HLWNets than HetNets, due to the small
C. Summary and Lessons Learnt
coverage areas especially of LiFi APs. In this section we
1) Centralized LiFi System: The LiFi APs can be read- review the current literature related to: i) HHO in LiFi,
ily managed in a centralized manner, as they are located in ii) VHO between LiFi and WiFi, and iii) the selection between
the same compartment. This renders LiFi more opportunities HHO and VHO. A detailed taxonomy is given in Fig. 8,
in interference management, while WiFi has to rely on sens- and the relevant studies are summarized in Table VIII. The
ing carries and avoiding collision. With dedicated backhaul, considerations and guidelines for implementing handovers in
the centralized LiFi system can ease the implementation of HLWNets are also discussed.
precoding. But LiFi channels could vary rapidly, compromis-
ing the performance of precoding. As a result, channel vari-
ance must be considered when developing precoding schemes 4 The ratio between the number of LiFi APs and the number of WiFi APs.
Fig. 8. Taxonomy of Handover in LiFi and HLWNets.
TABLE VIII
S UMMARY OF HANDOVER SCHEMES IN L I F I AND HLWN ETS
A. Horizontal Handover with both the movement and rotation of user equipment con-
sidered. It is found that the handover rate peaks when the user
LiFi has a relatively small coverage range with a single AP,
device is tilted between 60◦ and 80◦ .
usually 2-3 meters in diameter. The ultra small cell makes
Although the optimal placement of APs can relieve the
LiFi encounter considerably frequent handovers, even when
detriment of handovers to some extent, the degradation in
the user moves at a moderate speed. Also, the LiFi chan-
throughput is still outstanding for fast-moving users. To further
nel is related to the PD’s receiving orientation, of which the
reduce the handover rate, the concept of handover skip-
change could be very rapid and sudden. This might lead to
ping (HS) was introduced in [96], which enables the user
frequent and unexpected handovers. Therefore, the handover
to be transferred between non-adjacent APs. In this work,
cost becomes a critical factor to consider in LiFi. Taking a topology-aware HS scheme was proposed to let the user
handovers into account, the separation distance between APs skip the APs of which the chord length is below a predefined
affects network throughput in two aspects. On the one hand, threshold. A similar method was reported in [54], with the
a smaller separation provides a higher area spectral efficiency. research scope extended to multi-AP association. This type
On the other hand, a larger separation reduces the handover of approach relies on knowledge of the user’s trajectory and
rate. Motivated by this, the authors in [19] studied the optimal network topology. However, the equivalent network topology
placement of LiFi APs and concluded that the ideal cover- of LiFi is dynamic and user-dependant, due to the impact of
age area of a LiFi AP is 2 to 8 m2 , depending on the user the PD’s receiving orientation. Also, positioning techniques
density and handover overhead. While the coverage areas of are needed to acquire knowledge of the user’s trajectory, and
different APs usually overlap each other, the authors in [94] feedback is necessary for sending this information to APs. To
also investigated the handover procedure for non-overlapping circumvent the above stringent requirements, an RSS-based
coverage. This study suggests a soft handover for APs with HS approach was developed in [95] by exploiting the rate of
non-overlapping coverage and otherwise a hard handover. The change in RSS to indicate whether the user is moving towards
above two papers only consider user mobility with a fixed a certain AP. Using a weighted average of RSS and its rate
receiver orientation. In [36], the handover rate was analyzed of change to make handover decisions, this method does not
require extra feedback, since RSS is commonly used in the Unfortunately, this method does not address the issue of chan-
current handover schemes. More importantly, it does not rely nel blockages in LiFi. Exploiting the statistical information
on knowledge of the network topology. It is shown that the on light-path blockages, the handover process was formu-
RSS-based HS method can improve the network throughput lated as an optimization problem in [102], which maximizes
by up to about 70% over the handover scheme employed in throughput over a period of time. Such a method however
long-term evolution (LTE) and 30% over the trajectory-based requires a relatively high computational complexity. In [21],
HS method, respectively. the concept of handover skipping in LiFi [95] was extended
to HLWNets. Specifically, a dynamic network preference that
B. Vertical Handover adapts to the user’s speed is introduced to adjust the coverage
areas of different networks. This approach can reduce 40% of
The user usually requires a VHO from LiFi to WiFi when
handovers in a walking speed and 70% in a running speed.
losing LiFi connectivity. The loss of LiFi connectivity might
In [103], the network preference is trained through artificial
be caused by two reasons: i) the light-path is blocked by
neural network (ANN), considering the user’s speed as well
opaque objects, such as human bodies and furniture; and ii) the
as the network deployment. This method is able to further
PD’s receiving orientation is significantly deviated from the
improve the network throughput, with a gain of 50% higher
LoS path. The authors in [97] analyzed the probability of
than the trajectory-based method.
VHO, showing that a trade-off exists between the number of
handovers and their delay. This signifies the importance of
an appropriate level of hysteresis in the handover process. A D. Summary and Lessons Learnt
number of studies were carried out to develop VHO schemes The selection between HHO and VHO can be considered
for LiFi-involved hybrid networks [98]–[100]. In [98], a VHO in two parts. The first part is to select an AP without taking
scheme based on the Markov decision process was proposed. into account the handover cost. This is a typical AP selec-
This method determines whether to perform a VHO on the tion process which needs to measure channel quality as well
basis of the queue length for WiFi and the channel condition of as resource availability. A high channel quality means a high
LiFi. Another VHO scheme was proposed for hybrid LiFi and spectral efficiency. However, an AP that provides a high chan-
LTE networks in [99], which predicts the system state in terms nel quality does not necessarily render a high data rate since
of interruption duration, message sizes and access delays. the resource might be fully occupied. Thus, resource availabil-
These parameters, which are recorded by the user equipment ity must be considered in conjunction with channel quality.
in real time, can be used to make handover decisions. A sim- The second part is to evaluate the handover cost. Cell dwell
ilar approach was developed for hybrid LiFi and femtocell time (CDT), which is defined as the time that a user stays
networks in [100], considering multiple attributes including with an AP without being disconnected, is a key metric for the
dynamic network parameters (e.g., delay, queue length and handover process, no matter caused by user mobility or chan-
data rate) and actual traffic preferences. The above methods nel blockages. In summary, the HHO/VHO selection needs
have one common point: they adjust the network preference to jointly consider channel quality, resource availability and
based on channel and traffic conditions. However, they do CDT through optimization or decision-making methods. The
not consider the handover overhead and user mobility, which handover decision can be made in the interest of: i) a single
impose different impacts on different types of handovers. user or ii) overall network performance. In addition, the user
Without weighing the advantages and disadvantages of VHO can be served by multiple APs simultaneously, e.g., CoMP. In
and HHO, it is difficult to implement an effective handover this situation, the handover occurs in the form of a group of
process for HLWNets. APs. The six factors that need to be considered when design-
ing the handover scheme for HLWNets form a hexagram, as
C. Selection Between HHO and VHO shown in Fig. 9. The issue of choosing between HHO and
VHO essentially involves load balancing, which is elaborated
Due to the change of air interfaces, a VHO usually needs
in the next section.
a much longer processing time than a HHO [104]. Also, the
WiFi system has a lower system capacity than LiFi, and an
excessive number of WiFi users would cause a substantial VII. L OAD BALANCING
decrease in throughput. Thus, the choice between HHO and In the area of wireless networks, load balancing (LB) refers
VHO is critical to HLWNets. Specifically, not all of the users to the techniques that distribute user sessions across the APs
that lose LiFi connectivity should be switched to WiFi, e.g., the with overlapping coverage areas. The aims of LB are to
users encountering a transient light-path blockage. Apart from optimize resource utilization, to maximize throughput, to min-
that, the user’s velocity is also an important factor in deciding imize response time, and to reduce network congestion. In
whether a user should be served by LiFi or WiFi. In general, homogeneous networks, the coverage overlap among APs is
fast-moving users prefer WiFi, since they would experience restricted to mitigate ICI. As a result, LB only applies to
frequent HHOs in LiFi. To solve the complicate problem of cell-edge users when they impose unbalanced traffic loads
choosing between HHO and VHO, Wang et al. [101] proposed to different APs. In other words, LB is not needed when
a handover scheme based on fuzzy logic. This method makes the users’ demands for data rates are uniformly distributed
handover decisions by measuring parameters including not in geography. The authors in [105] classified WiFi-related
only CSI but also the user’s speed and data rate requirement. LB techniques into two categories: user-based and AP-based.
Fig. 10. Taxonomy of load balancing in HLWNets.
shown in Fig. 10, and the current literature is summarized in

Table IX.
A. Stationary Channel Load Balancing

The wireless channel can be assumed stationary within the
coherence time. With stationary channels, the LB problem
needs to trade off channel quality with resource availability.
In [107], an LB method was proposed to achieve proportional
fairness (PF) among users, in forms of both centralized and
distributed resource-allocation algorithms. To improve qual-
ity of service, the LB issue is formulated as a mixed-integer
non-linear programming problem in [108], which considers
different data rate requirements among users. The two above
methods both construct an NP-hard problem, and solving the
problem requires an excessive amount of computational com-
plexity that exponentially increases with the number of APs.
To reduce the processing power, an iterative algorithm based
Fig. 9. Factors for consideration in the handover process for HLWNets. on evolutionary game theory was reported in [110], with
multiple fairness functions (MFF) considered. In this work,
light-path blockages, arbitrary receiver orientations and data
With user-based methods, each user selects the APs according rate requirements are characterized to model a practical com-
to its own interest, and the optimal network performance is munication scenario. The authors in [111] also introduced an
hard to achieve. On the contrary, AP-based methods imple- iterative algorithm but focused on power allocation. This algo-
ment a network-wide LB, which requires a central unit to rithm consists of two states: i) finding the optimal power
coordinate APs. allocation of each AP to maximize its throughput; and ii) seek-
LB becomes essential and challenging in hybrid networks ing another AP for the user with the minimum data rate to
due to two main factors: i) the coverage areas of LiFi and increase the overall throughput and to enhance the system
WiFi overlap each other; and ii) WiFi APs have a larger cov- fairness. The above iterative algorithms can be deemed as
erage area but a lower system capacity than LiFi APs [18]. autonomous optimization, which is carried out individually at
This makes WiFi susceptible to traffic overload even if the each AP.
users’ demands for data rates are uniformly distributed in The centralized optimization needs to solve an NP-hard
geography. A large body of research was carried out to study problem, whereas the autonomous optimization requires a
LB approaches in the HetNet, including relaxed optimization, quantity of iterations to reach a steady state. They both need
Markov decision process, game theory and cell range expan- a substantial amount of processing time. In HLWNets, CSI
sion [106]. Though these methods are applicable to HLWNets, could rapidly vary for mobile users with an even modest
they face a critical issue of user mobility due to the short speed. This restricts the processing time and thus challenges
coverage range of a single LiFi AP. According to the algo- the practicability of the above methods. Alternatively, direct
rithm type, the LB algorithms that have been developed decision-making methods are applicable, which provide a sig-
for HLWNets fall into two categories: i) optimization and nificantly reduced amount of processing time. Such an LB
ii) decision-making. In this section we classify these methods method was reported in [114], which splits the process into
into i) stationary-channel and ii) mobility-aware, depending on two stages: i) determine the users that should be served by
whether user mobility is considered. A detailed taxonomy is WiFi and ii) assign the remaining users as if in a stand-alone
TABLE IX
S UMMARY OF L OAD BALANCING A PPROACHES IN HLWN ETS
LiFi network. Relying on statistical knowledge of data rate the process of formulating the CDT-based LB problem. This
requirements and CSI, this fuzzy logic-based method is able modification can effectively reduce the negative impact caused
to achieve near-optimal performance in terms of throughput by intermittent light-path blockages.
and user fairness, while reducing the processing time by over To provide low computational complexity for prac-
10 orders of magnitude. tical implementations, decision-making methods have
also been investigated for mobility-aware load balancing.
Wang et al. [101] developed a fuzzy logic system to balance
B. Mobility-Aware Load Balancing the traffic loads between LiFi and WiFi, with multiple input
The noted LB methods all rely on CSI knowledge, which parameters including SINR, the user’s speed and data rate
varies due to user movements and environmental changes. requirements. Adding light-path blockage as an extra input,
Accordingly, these methods have to calculate their solutions another fuzzy logic-based LB approach was proposed in [62]
periodically. When the new solutions make a change to the to handle the situation of unbalanced traffic loads caused by
user association, the impact caused by handovers must be blockage. While fuzzy logic can be readily implemented,
taken into account. For instance, with stationary-channel LB the logic rules are pre-defined and lack flexibility. Instead,
methods, users will be transferred between LiFi and WiFi it is feasible to use machine learning to cope with the
repeatedly when moving across the LiFi APs, leading to uncertainties in network deployment, user distribution, traffic
frequent and unnecessary handovers. situations, etc. A LB method based on reinforcement learning
In order to tackle the above issue, user mobility has to was introduced in [115]. It is shown that this method can
be considered in conjunction with LB. This is referred to as outperform iterative algorithms in most scenarios.
mobility-aware load balancing. In [112], a method based on
the college admission model was proposed. Specifically, the
achievable data rate and the user’s moving direction are used C. Summary and Lessons Learnt
to measure the user’s preference, while the sum data rate is Channel variance due to user mobility is not negligible
used to compute the AP’s preference. These two preferences when solving the LB issue in HLWNets. This essentially
are then iteratively calculated to reach a steady solution. This renders a trade-off between the instantaneous data rate and
method however requires to know the user’s trajectory, which the handover rate when maximizing the average throughput.
is not ready to acquire in practice. A dynamic LB scheme Other QoS metrics such as the packet loss ratio, delay and
was proposed in [53], which also performs an iterative algo- jitter should also be considered, forming a multi-objective
rithm. In each iteration, AP assignment and resource allocation optimization problem. To solve the LB problem, the algorithms
are sequentially implemented to improve the effective data must compromise optimality for computational complexity.
rates which excludes handover overheads. In [113], AP assign- While optimization methods can provide optimal solutions,
ment and resource allocation are jointly implemented. The they need an excessive amount of computational complexity.
joint implementation can achieve a network throughput 50% In contrast, decision-making methods can significantly reduce
higher than the separate implementation, at the cost of a higher computational complexity, but the optimality is compromised.
computational complexity by 3 orders of magnitude. In [109], In general, it is possible to realize low-complexity LB in two
a globally-optimized LB method is realized by using CDT ways. One approach is to exploit the status information on
to measure the handover cost. This approach does not rely users and APs (e.g., the user’s speed and the AP’s queu-
on CSI and thus suits the scenario of fast-varying channels. ing length) through intelligent control methods such as fuzzy
Accordingly, it provides a sub-optimal network performance. logic, game theory and machine learning. The other way is to
In [102], light-path blockage is characterized and included in construct a decision flowchart with a number of pre-defined
Fig. 11. Taxonomy of applications benefiting from HLWNets.
thresholds. Many of the decision-making methods are exam- in indoor scenarios, as the transmitted signals are degraded
ined in specific environments, while the variance of network and interrupted by obstructions, especially ceilings and walls.
deployment has not been well studied in the current literature. Alternatively, IPS can be developed upon short-range wire-
It is meaningful to develop LB methods with low complexity less communication technologies, e.g., WiFi, LiFi, Bluetooth,
while being adaptive. radio frequency identification and ZigBee. Multiple surveys
have been carried out to summarize LiFi-based positioning
techniques, e.g., [118], [119]. In this subsection, we briefly
VIII. A PPLICATIONS
introduce the classification of IPS techniques and focus on
The HLWNet can fuel a wide range of applications and discussing the relevant development in HLWNets.
services, which can be classified into three major categories:
communication-based, functional and advanced applications. 1) Classification of IPS Techniques: IPS methods can be
A detailed taxonomy is given in Fig. 11. As mentioned, classified from two angles: mathematical algorithm and used
HLWNets are a promising approach to support high-speed information. The main algorithms used for IPS are triangu-
wireless communications, e.g., 4K/8K video streaming, VR, lation, proximity and fingerprint. The triangulation method
holographic display, etc. An integration system of mmWave exploits the geometric properties of triangles by measuring
and LiFi for the transmission of holographic 3D display data the distance or angle between the device and multiple fixed
was reported in [47]. Also, HLWNets can relieve the resource points, i.e., beacons. This method offers high accuracy at the
shortage in high-density scenarios such as airports, stadiums cost of a sophisticated system structure. With a single receiver,
and conference venues. Typical functional applications are the triangulation method needs at least three beacons for 2-D
indoor positioning system (IPS) and physical layer security positioning and four beacons for 3-D. The LiFi beacon can be
(PLS). These functions underlie the advanced applications in an LED-based AP or a modulated retro-reflector (MRR) [120],
IoT, ranging from consumer IoT to industrial IoT. A hybrid which avoids proactively emitting light. Proximity is the sim-
VLC/RF indoor IoT system with solar energy harvesting was plest algorithm, which links the device’s location to the AP’s
introduced in [116]. In this section we focus on discussing IPS coverage area. Specifically, when a device is recognized by
and PLS, while highlighting the benefits of HLWNets to these multiple APs, it is roughly located within the overlapping
applications. coverage area of these APs. Due to the dense deployment,
LiFi is naturally suitable for using this algorithm. Fingerprint
employs location-dependant information such as RSS and
A. Indoor Positioning System requires off-line radio maps. The optimal location is obtained
Positioning is an essential tool for providing location-based by minimizing the Euclidean distance between the radio map
services such as navigation, creating maps, tracking objects, and the real-time measurement. As a result, the positioning
etc. As a mainstream positioning technology at present, the accuracy of fingerprint is dependent on the accuracy of the
global positioning system (GPS) is a satellite-based radio- radio map [121].
navigation system which provides geolocation information to According to the used information, IPS techniques fall into
a GPS receiver. With the latest accuracy enhancement using four categories: RSS, time of arrival (TOA), time difference
the L5 band, the accuracy of GPS can be improved from of arrival (TDOA) and angle of arrival (AOA). RSS-based
5 m to 30 cm [117]. However, GPS becomes less accurate methods exploit channel attenuation to estimate the distance
between the device and the beacon. Among all the signal char- named Bob and Eve. To enhance the security of wireless com-
acteristics, RSS can be readily acquired. The accuracy of such munications, PLS has drawn a significant amount of research
a method depends on a reliable path-loss model and might attention. There are two basic categories: secure key gener-
suffer from uncontrollable errors caused by multipath propa- ation and secure data transmission. The former exploits the
gations [122]. TOA-based methods also compute the distance inherent randomness of wireless channels, e.g., RSS and phase
but use the travel time of the signal. This type of method information [133], to ensure the security of keys. As for secure
needs rigid time synchronization between the device and the data transmission, the aim is to enlarge the SINR difference
beacon [123]. In order to circumvent this requirement, TDOA- between the links of Bob and Eve. In general, Eve’s SINR
based methods employ multiple transmitters or receivers to peformance can be weakened in two ways: i) reducing RSS
obtain the time difference between the received signals [124]. and ii) increasing noise or interference [134]. The first way
However, time synchronization is still required between the is focused on optimizing the transmission scheme for Bob
beacons. AOA-based methods measure the angle between the through techniques such as beamforming, resource allocation,
transmitted signal and the normal angle of the beacon. In RF, interference alignment, etc. The transmission power is reduced
AOA is usually obtained by detecting the phase difference for Bob and so as for Eve. The second way is to inject artificial
between antennas [125]. However, AOA cannot be measured noise, which can be generated in the null-subspace of Bob’s
directly in LiFi due to the lack of phase information in channel so that only Eve’s SINR performance is impaired
intensity modulation/direct detection (IM/DD). Instead, AOA by the noise. In addition, when Eve’s channel is worse than
can be acquired through two approaches. One approach is Bob’s on average, secure channel coding such as low-density
called image transformation, which calculates AOA through parity-check can effectively increase secrecy [135].
the trigonometric relationship between the light beacons’ coor- In comparison with WiFi, LiFi has a number of intrinsic
dinates and their imaging locations on a photo [126]. The other advantages in terms of security. First, since light does not
one is modeling, which exploits the angular pattern of RSS at penetrate opaque objects, LiFi can be securely used in a com-
the PD [127]. partment space such as conference rooms. Second, LiFi covers
2) IPS in HLWNets: The key metric for positioning is accu- a smaller area than WiFi with a single AP. Thus, Eve has
racy. The intrinsic shorter coverage range of LiFi leads to to move closer to intercept signals. Third, the LoS path nor-
a smaller positioning error (0.1−0.35 m) than WiFi (1−7 mally contributes over 80% of the received signal power [18].
m) [128]. Also, LiFi can provide more dense beacons than As a result, the information leakage to Eve from scattered
WiFi. Due to the existing and ubiquitous lighting infrastruc- optical signals would be very limited. A quantity of research
tures, the installation cost and energy consumption will be has been conducted to analyze the secrecy performance of
relatively low for LiFi beacons. Further, LiFi-based IPS can LiFi. Chen and Haas [136] demonstrated that the hexagonal
readily detect the device’s orientation via ADRs [129], while deployment provides the highest secrecy capacity, whereas the
it is difficult for WiFi-based IPS to achieve. The challenges
matrix deployment performs marginally worse. Ayman and
facing LiFi-based IPS and WiFi-based IPS are quite different.
Lampe [137] studied the secrecy capacity of LiFi under ampli-
WiFi signals may experience severely rich multipath fading,
tude constraints of LEDs, while using beamforming to hinder
especially in some special environments such as factories,
eavesdroppers in specified areas. In addition, quantum key
underground mines and tunnels. As for LiFi, signals might
distribution (QKD) is a specially-tailored approach for OWC-
confront light-path blockage, resulting in a complete loss of
connectivity. aided PLS [138]. A photon can be readily encoded as a
A hybrid IPS using both LiFi and WiFi (or other RF tech- zero/one state, e.g., using horizontal and vertical polarizations.
nologies) is envisaged to improve the accuracy of indoor Based on quantum mechanics such as the quantum no-cloning
positioning. In [130], the proximity positioning concept was principle, QKD is able to generate and distribute the quan-
applied in a hybrid environment of LiFi and Zigbee. This tum random key among two parties. A handheld QKD system
method however has a relatively low accuracy (∼130 cm). was demonstrated in [139], to achieve a secret key rate above
A two-stage positioning system was proposed in [131]. It first 30kb/s in the free-space link over a distance of 0.5m.
determines a possible area via a LiFi-based proximity method Studies have been carried out on the topic of PLS in
and then locates the specific position in that area by using HLWNets. In [140], the secrecy outage performance of an
the RSS of the RF signals. Such a system is able to keep the RF uplink was analyzed with solar energy harvesting in
positioning error within 20 cm. Another two-stage positioning a LiFi downlink. In [141], the power consumption of the
system was developed in [132]. In the first stage, RF is used HLWNet was minimized under the secrecy rate constraint.
to detect which room the device is currently located in. In the It is shown that to achieve the same secrecy rate, HLWNets
second stage, LiFi is employed to detect the specific position consume a power level 10 dB less than stand-alone networks.
of the device. The estimation error was reported to be only In [142], Ucar et al. proposed a HLWNet-based security proto-
5.8 cm. col for vehicular platoon communications, where LiFi provides
resilience to security attacks and WiFi offers redundancy for
link reliability. The authors in [143] analyzed the secrecy
B. Physical Layer Security performance for dual-hop HLWNets, where the energy har-
Wireless signals are broadcast in the open air and can be vested from LiFi signals is used to relay data through RF.
received by the intended user as well as an eavesdropper, Similar work was reported in [144], with the aim of finding
the minimum transmission power that can achieve a certain introduce more noise. It is a significant research direction to
amount of secrecy capacity. investigate the hardware integration of LiFi and WiFi to satisfy
their respective requirements.
C. Summary and Lessons Learnt 2) Modulation Suitability: While OFDM is the accepted
Due to the complementarity between LiFi and WiFi, the modulation technique for WiFi, now there are a number of
HLWNet can not only boost the network capacity but also candidates for the modulation in LiFi, including O-OFDM,
benefit application services such as IPS and PLS. In regard to OOK and pulse modulation.2 Among them, OOK and pulse
IPS, the nature of LiFi allows it to offer a much higher accu- modulation enable moderate data rates from 1 Mbps to some
racy than WiFi for slow-moving users. A typical use case is to 100 Mbps. These single-carrier modulation techniques have a
navigate the AGV (of which the speed is usually limited to 0.5 relatively low peak-to-average power ratio (PAPR). In contrast,
m/s) in a factory, where RF-based IPS is significantly inaccu- O-OFDM with adaptive bit loading, which allows the use of
rate due to severe multipath fading. As for fast-moving users, bandwidth beyond -3dB, can achieve much higher data rates
outdated information could degrade the accuracy of LiFi-based but requires a high PAPR. The modulation suitability relies on
IPS when the LiFi channel varies rapidly. How to employ the LEDs and PDs used. In practice, commercial LEDs have
HLWNets to improve the positioning accuracy in this case a limited linear zone of the I-V curve with a restricted band-
is still an open issue. With respect to PLS, the impenetrabil- width. This renders a trade-off between the spectral efficiency
ity of light has two sides. On the one hand, it offers LiFi a and energy efficiency. Meanwhile, the PDs affect sensitivity
robust security performance in the physical layer as it makes and thus the link budget. It is necessary to study the mod-
eavesdropping more difficult. On the other hand, the transmis- ulation suitability for realistic LiFi front-ends. Further, the
sion link is susceptible to loss of connectivity due to intended hardware integration will also have a impact on choosing the
or unintended activities. Therefore, it is important to exploit suitable modulation techniques.
HLWNets to provide secure and reliable communication links.
For example, QKD is able to provide an absolute security for B. Network Layer and Network Management
sending keys at a low data rate, while the encrypted data can 1) Parallel Transmission: In the existing literature on the
be transmitted through a high data rate link on WiFi. Further topic of HLWNets, it is commonly considered to serve the user
research is still required to understand the use of HLWNets by a single AP at a time. This is subject to the conventional
for these applications. transmission control protocol (TCP), which does not support
the packets sent from different APs to be reordered at the desti-
IX. C HALLENGES AND R ESEARCH D IRECTIONS nation. Since 2013, the Internet Engineering Task Force (IETF)
The key challenges facing HLWNets are integrating the has been working on multipath transmission control protocol
different wireless technologies in an efficient manner. In the (MPTCP) [146], which adds a subflow sequence number in
majority of current research, LiFi and WiFi are treated as two the packet overhead to solve the issue of packet reordering.
individual technologies, and the integration is carried out in Enabled by MPTCP, the users can be served by LiFi and WiFi
terms of network management. In order to lift the restrictions simultaneously. This emerging approach of network manage-
imposed by vertical handovers, it is important to enable LiFi ment has great benefit on HLWNets in two aspects. First, it
and WiFi to work at the same time, realizing a parallel trans- can completely avoid vertical handovers, which have been rec-
mission. To achieve this goal, the integration process must not ognized as a key issue in HLWNets. Second, it offers a more
be contained in the network layer. In this section, we will dis- flexible way of resource allocation than the single-AP associ-
cuss the challenges and future research directions of HLWNets ation, since the traffic load of one user can now be distributed
in different layers. among multiple APs. For this reason, parallel transmission
is a promising direction for improving the performance of
A. Physical Layer and Hardware Implementation HLWNets.
2) Cell Deployment Optimization: While using parallel
1) Cost-Effective Hardware Integration: In order to allow transmission in HLWNets can eliminate vertical handovers,
the operation of HLWNets, the frond-end circuits of LiFi and horizontal handovers still exist for mobile users. Offering a
WiFi need to be integrated on the same board. There are balance between the handover rate and SINR coverage prob-
a number of common electronic components in the signal ability, the cell deployment optimization is fundamental to
processing chains, e.g., power amplifiers, up/down converters the network convergence of HLWNets. This optimization pro-
and analogue-to-digital converters. It is feasible to share these cess involves a number of factors. The first factor includes
electronic components to provide a compact and economic the properties of LEDs in terms of their density, locations,
hardware implementation. The multi-standard RF front-end FoV, output optical power, etc. Second, the placement of WiFi
has been realized with a reconfigurable baseband filter, which has an impact on the coverage requirement for LiFi. Third, it
is compatible with WiFi and the cellular technologies [145]. is necessary to consider the influence of user behavior. For
The hardware integration of LiFi and WiFi is more challenging instance, the optimized cell deployment in [20] would not be
as they use different antenna components. To simultaneously ideal when users are seldom located in room corners. Finally
process the baseband signals of LiFi and WiFi in the same but importantly, the cell deployment must meet illumination
signal processing chain, they need to be converted to different
frequencies. However, this would reduce the system gain and 2 Pulse modulation is currently under discussion in IEEE P802.15.13.
constraints. Hence, the optimization on cell deployment is a applications have stringent requirements on security and pri-
complicate issue and solving it is challenging. An effective vacy, which can be enhanced by the HLWNet due to its nature
approach to this issue is yet to be developed. of high PLS. Last but not least, IoT requires low power con-
3) Intelligent Network Management: The core component sumption that enables devices to operate for many years on
of HLWNets is network management including handover, a single charge. Backscatter communications in both RF and
resource allocation and load balancing. The challenges lie in VLC have attracted massive research attention to reduce the
two main aspects: problem complexity and processing time. power consumption and cost of IoT devices. Combining with
As discussed, hybrid networks are much more complex than energy harvesting, it is feasible to develop self-powered IoT
homogeneous networks, in terms of both the density of APs devices, which completely avoid the hassle of charging and
and their network topology. This significantly increases the significantly reduce the maintenance cost.
complexity of network management. Meanwhile, the pro-
cessing time is restricted as the mobility of indoor users is
relatively high for ultra small cells. Particularly, a drastic X. C ONCLUSION
change in the device orientation can greatly affect the LiFi Along with the looming spectrum crunch in RF, LiFi has
channel conditions in a split second. Handling an adaptive emerged in recent years as a promising technology for indoor
network management within a limited amount of processing wireless communications. At the mean time, WiFi continues
time is one of the key research topics for HLWNets. One its wide deployment in daily life. The coexistence of LiFi and
potential approach is applying machine learning, which is WiFi is gaining momentum with the roll-out of LiFi commer-
capable of solving complex optimization problems where an cial products from companies such as pureLiFi and Signify.
explicit mathematical model is hard to establish. However, it Located in the same local area, LiFi and WiFi can be readily
is difficult to collect sample data in a real-time networking managed through a central control unit, forming the structure
system as the best solution is unknown. Using machining of HLWNets. Combining the high data rate of LiFi and the
learning in such an unsupervised scenario to realize intelligent ubiquitous coverage of WiFi, HLWNets are able to provide
network management is still an open issue. greater network performance than a single wireless technology.
Research on implementing HLWNets in realistic environments
and optimizing the network performance is underway.
This article introduced a framework of system design for
C. Application Layer
HLWNets, followed by an overview of key performance metrics
1) HLWNet-Facilitated WLAN: In the WLAN domain, and recent achievements, validating the superiority of HLWNets
there are many different use cases with a wide variety of against stand-alone networks. The modeling work on user
different requirements. For instance, the data rate of VoIP is behavior was summarized in terms of user movement, device
around 100 kbps within a 150 ms latency, whereas virtual orientation and light-path blockage, highlighting the impor-
reality consumes multi-Gbps with a latency requirement less tance of practical user behavior models to LiFi and HLWNets.
than 20 ms. This results in the demand for high adaptabil- Afterwards, the existing studies were classified and analyzed for
ity of the wireless communication system, which HLWNets three key technical topics: interference management, handover
can well support thanks to the high complementarity between and load balancing, with the unique challenges in supporting
LiFi and WiFi. When using HLWNets to provide service to user mobility identified. Further, we discussed the benefits of
these applications, it is necessary to consider their different HLWNets to application services, exemplified by indoor posi-
requirements in the network management. In other words, the tioning and physical layer security. Finally, the challenges and
network convergence of HLWNets is better to be carried out research directions for HLWNets were summarized in differ-
across the physical layer, network layer, transport layer and ent layers. It is concluded that parallel transmission has the
application layer, rather than being done in each layer sepa- potential to eliminate vertical handovers for HLWNets, while
rately. The cross-layer design for homogeneous networks has cross-layer design can further improve network performance.
been extensively investigated in the past decade, e.g., [147]. As The hope is that this overview paper will push forward both the-
for HLWNets, there are quite a few unsolved challenges, such oretical and experimental research towards the future success
as resource description, compatible MAC protocol, prioritized of HLWNets in the 6G era.
packet routing, etc.
2) HLWNet-Facilitated IoT: Besides WLAN, HLWNets can
also contribute to the prosperity of IoT. First, the small cell R EFERENCES
nature of LiFi allows it to readily support a very high density [1] “Cisco visual networking index: Forecast and trends, 2017-2022,”
of IoT devices, while WiFi provides ubiquitous coverage to Cisco, San Jose, CA, USA, Rep., Nov. 2018.
guarantee connectivity. Second, in some scenarios such as a [2] 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.
factory, the radio propagation environment can be quite chal-
[3] M. S. Islim et al., “Towards 10 Gb/s orthogonal frequency division
lenging. This is due to a large number of metallic objects multiplexing-based visible light communication using a GaN violet
in the immediate surroundings of transmitter and receiver, as micro-LED,” Photon. Res., vol. 5, no. 2, pp. A35–A43, Apr. 2017.
well as potentially high interference caused by certain indus- [4] A. Sevincer, A. Bhattarai, M. Bilgi, M. Yuksel, and N. Pala,
“LIGHTNETs: Smart LIGHTing and mobile optical wireless
trial machines. In this case, LiFi can complement WiFi to NETworks—A survey,” IEEE Commun. Surveys Tuts., vol. 15, no. 4,
provide robust and reliable wireless links. Third, many IoT pp. 1620–1641, 4th Quart., 2013.
[5] M. A. Khalighi and M. Uysal, “Survey on free space optical communi- [28] T. Komine and M. Nakagawa, “Integrated system of white LED visible-
cation: A communication theory perspective,” IEEE Commun. Surveys light communication and power-line communication,” IEEE Trans.
Tuts., vol. 16, no. 4, pp. 2231–2258, 4th Quart., 2014. Consum. Electron., vol. 49, no. 1, pp. 71–79, Feb. 2003.
[6] A. G. Sarigiannidis et al., “Architectures and bandwidth allocation [29] H. Ma, L. Lampe, and S. Hranilovic, “Hybrid visible light and power
schemes for hybrid wireless-optical networks,” IEEE Commun. Surveys line communication for indoor multiuser downlink,” IEEE/OSA J. Opt.
Tuts., vol. 17, no. 1, pp. 427–468, 1st Quart., 2014. Commun. Netw., vol. 9, no. 8, pp. 635–647, Aug. 2017.
[7] X. Bao, G. Yu, J. Dai, and X. Zhu, “Li-Fi: Light fidelity—A survey,” [30] P. Mark, “Ethernet over light,” Ph.D. dissertation, Dept. Elect. Comput.
Wireless Netw., vol. 21, no. 6, pp. 1879–1889, 2015. Eng., Univ. Brit. Columbia, Vancouver, BC, Canada, 2014.
[8] M. Ayyash et al., “Coexistence of WiFi and LiFi toward 5G: Concepts, [31] G. Cossu, A. M. Khalid, P. Choudhury, R. Corsini, M. Presi, and
opportunities, and challenges,” IEEE Commun. Mag., vol. 54, no. 2, E. Ciaramella, “VLC-signals distribution over GI-POF for in-home
pp. 64–71, Feb. 2016. wireless networks,” in Proc. Int. Workshop Opt. Wireless Commun.
[9] P. Sarigiannidis, T. Lagkas, S. Bibi, and A. Ampatzoglou, “Hybrid 5G (IWOW), Pisa, Italy, 2012, pp. 1–3.
optical-wireless SDN-based networks, challenges and open issues,” IET [32] K. Chandra, R. V. Prasad, and I. Niemegeers, “An architectural
Netw., vol. 6, no. 6, pp. 141–148, Nov. 2017. framework for 5G indoor communications,” in Proc. Int. Wireless
[10] T. Koonen, “Indoor optical wireless systems: Technology, trends, and Commun. Mobile Comput. Conf. (IWCMC), Dubrovnik, Croatia, 2015,
applications,” J. Lightw. Technol., vol. 36, no. 8, pp. 1459–1467, pp. 1144–1149.
Apr. 15, 2018. [33] H. Kazemi, M. Safari, and H. Haas, “A wireless backhaul solution
[11] X. Li, R. Zhang, and L. Hanzo, “Optimization of visible-light optical using visible light communication for indoor Li-Fi attocell networks,”
wireless systems: Network-centric versus user-centric designs,” IEEE in Proc. IEEE Int. Conf. Commun. (ICC), Paris, France, 2017, pp. 1–7.
Commun. Surveys Tuts., vol. 20, no. 3, pp. 1878–1904, 3rd Quart., [34] H. Kazemi, M. Safari, and H. Haas, “A wireless optical backhaul solu-
2018. tion for optical attocell networks,” IEEE Trans. Wireless Commun.,
[12] M. Z. Chowdhury, M. T. Hossan, A. Islam, and Y. M. Jang, “A vol. 18, no. 2, pp. 807–823, Feb. 2019.
comparative survey of optical wireless technologies: Architectures and [35] V. K. Papanikolaou, P. P. Bamidis, P. D. Diamantoulakis, and
applications,” IEEE Access, vol. 6, pp. 9819–9840, 2018. G. K. Karagiannidis, “Li-Fi and Wi-Fi with common backhaul:
[13] M. Obeed, A. M. Salhab, M. Alouini, and S. A. Zummo, “On Coordination and resource allocation,” in Proc. IEEE Wireless
optimizing VLC networks for downlink multi-user transmission: A sur- Commun. Netw. Conf. (WCNC), Barcelona, Spain, 2018, pp. 1–6.
vey,” IEEE Commun. Surveys Tuts., vol. 21, no. 3, pp. 2947–2976, [36] M. D. Soltani, H. Kazemi, M. Safari, and H. Haas, “Handover modeling
3rd Quart., 2019. for indoor Li-Fi cellular networks: The effects of receiver mobility and
[14] M. Z. Chowdhury, M. K. Hasan, M. Shahjalal, M. T. Hossan, and rotation,” in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC), San
Y. M. Jang, “Optical wireless hybrid networks: Trends, opportunities, Francisco, CA, USA, 2017, pp. 1–6.
challenges, and research directions,” IEEE Commun. Surveys Tuts., [37] Z. Chen, D. A. Basnayaka, X. Wu, and H. Haas, “Interference mit-
vol. 22, no. 2, pp. 930–966, 2nd Quart., 2020. igation for indoor optical attocell networks using an angle diversity
[15] M. B. Rahaim, A. M. Vegni, and T. D. C. Little, “A hybrid radio receiver,” J. Lightw. Technol., vol. 36, no. 18, pp. 3866–3881, Sep. 15,
frequency and broadcast visible light communication system,” in Proc. 2018.
IEEE GLOBECOM Workshops (GC Wkshps), Houston, TX, USA, [38] H. Tabassum and E. Hossain, “Coverage and rate analysis for co-
2011, pp. 792–796. existing RF/VLC downlink cellular networks,” IEEE Trans. Wireless
[16] I. Stefan and H. Haas, “Hybrid visible light and radio frequency com- Commun., vol. 17, no. 4, pp. 2588–2601, Apr. 2018.
munication systems,” in Proc. IEEE 80th Veh. Technol. Conf. (VTC [39] P. A. Haigh et al., “A multi-CAP visible-light communications system
Fall), Vancouver, BC, Canada, 2014, pp. 1–5. with 4.85-b/s/Hz spectral efficiency,” IEEE J. Sel. Areas Commun.,
[17] D. A. Basnayaka and H. Haas, “Hybrid RF and VLC systems: vol. 33, no. 9, pp. 1771–1779, Sep. 2015.
Improving user data rate performance of VLC systems,” in Proc. IEEE [40] T. Wang, F. Yang, L. Cheng, and J. Song, “Spectral-efficient gener-
81st Veh. Technol. Conf. (VTC Spring), Glasgow, U.K., 2015, pp. 1–5. alized spatial modulation based hybrid dimming scheme with LACO-
[18] C. Chen, D. A. Basnayaka, and H. Haas, “Downlink performance OFDM in VLC,” IEEE Access, vol. 6, pp. 41153–41162, 2018.
of optical attocell networks,” J. Lightw. Technol., vol. 34, no. 1, [41] R. Bian, I. Tavakkolnia, and H. Haas, “15.73 Gb/s visible light commu-
pp. 137–156, Jan. 1, 2016. nication with off-the-shelf LEDs,” J. Lightw. Technol., vol. 37, no. 10,
[19] S. Pergoloni, M. Biagi, S. Colonnese; R. Cusani, and G. Scarano pp. 2418–2424, May 15, 2019.
“Optimized LEDs footprinting for indoor visible light communication [42] I. Stefan, H. Burchardt, and H. Haas, “Area spectral efficiency
networks,” IEEE Photon. Technol. Lett., vol. 28, no. 4, pp. 532–535, performance comparison between VLC and RF femtocell networks,”
Feb. 15, 2016. in Proc. IEEE Int. Conf. Commun. (ICC), Budapest, Hungary, 2013,
[20] M. A. Dastgheib, H. Beyranvand, and J. A. Salehi, “Optimal placement pp. 3825–3829.
of access points in cellular visible light communication networks: An [43] Y. Wang, X. Wu, and H. Haas, “Analysis of area data rate with shadow-
adaptive gradient projection method,” IEEE Trans. Wireless Commun., ing effects in Li-Fi and RF hybrid network,” in Proc. IEEE Int. Conf.
vol. 19, no. 10, pp. 6813–6825, Oct. 2020. Commun. (ICC), Kuala Lumpur, Malaysia, 2016, pp. 1–5.
[21] X. Wu, D. C. O’Brien, X. Deng, and J.-P. M. G. Linnartz, “Smart [44] Y. Sun, F. Yang, and L. Cheng, “An overview of OFDM-based visible
handover for hybrid LiFi and WiFi networks,” IEEE Trans. Wireless light communication systems from the perspective of energy efficiency
Commun., vol. 19, no. 12, pp. 8211–8219, Dec. 2020. versus spectral efficiency,” IEEE Access, vol. 6, pp. 60824–60833,
[22] M. D. Soltani, X. Wu, M. Safari, and H. Haas, “Bidirectional 2018.
user throughput maximization based on feedback reduction in LiFi [45] M. Kashef, M. Ismail, M. Abdallah, K. A. Qaraqe, and E. Serpedin,
networks,” IEEE Trans. Commun., vol. 66, no. 7, pp. 3172–3186, “Energy efficient resource allocation for mixed RF/VLC heteroge-
Jul. 2018. neous wireless networks,” IEEE J. Sel. Areas Commun., vol. 34, no. 4,
[23] L. Yin, W. O. Popoola, X. Wu, and H. Haas, “Performance evaluation of pp. 883–893, Apr. 2016.
non-orthogonal multiple access in visible light communication,” IEEE [46] Y. Fu et al., “End-to-end energy efficiency evaluation for B5G ultra
Trans. Commun., vol. 64, no. 12, pp. 5162–5175, Dec. 2016. dense networks,” in Proc. IEEE 91st Veh. Technol. Conf. (VTC Spring),
[24] S. D. Dissanayake and J. Armstrong, “Comparison of ACO-OFDM, Antwerp, Belgium, 2020, pp. 1–6.
DCO-OFDM and ADO-OFDM in IM/DD systems,” J. Lightw. [47] T. H. Loh et al., “Link performance evaluation of 5G mm-wave and
Technol., vol. 31, no. 7, pp. 1063–1072, Apr. 1, 2013. LiFi systems for the transmission of holographic 3D display data,” in
[25] T. Komine and M. Nakagawa, “Fundamental analysis for visible- Proc. 13th Eur. Conf. Antennas Propag. (EuCAP), Krakow, Poland,
light communication system using LED lights,” IEEE Trans. Consum. 2019, pp. 1–4.
Electron., vol. 50, no. 1, pp. 100–107, Feb. 2004. [48] V. Jungnickel et al., “Enhance lighting for the Internet of Things,” in
[26] R. Jiang, Z. Wang, Q. Wang, and L. Dai, “Multi-user sum-rate Proc. Global LIFI Congr. (GLC), Paris, France, 2019, pp. 1–6.
optimization for visible light communications with lighting con- [49] M. Hammouda, S. Akin, A. M. Vegni, H. Haas, and J. Peissig,
straints,” J. Lightw. Technol., vol. 34, no. 16, pp. 3943–3952, Aug. 15, “Link selection in hybrid RF/VLC systems under statistical queue-
2016. ing constraints,” IEEE Trans. Wireless Commun., vol. 17, no. 4,
[27] A. M. Abdelhady, O. Amin, A. Chaaban, B. Shihada, and pp. 2738–2754, Apr. 2018.
M.-S. Alouini, “Spectral-efficiency—Illumination Pareto front for [50] N. Aschenbruck, R. Ernst, and P. Martini, “Indoor mobility modelling,”
energy harvesting enabled VLC systems,” IEEE Trans. Commun., in Proc. IEEE GLOBECOM Workshops (GC Wkshps), Miami, FL,
vol. 67, no. 12, pp. 8557–8572, Dec. 2019. USA, 2010, pp. 1264–1269.
[51] J. Broch, D. A. Maltz, D. B. Johnson, Y.-C. Hu, and J. Jetcheva, “A [73] H. Ma, A. Mostafa, L. Lampe, and S. Hranilovic, “Coordinated beam-
performance comparison of multi-hop wireless ad hoc network routing forming for downlink visible light communication networks,” IEEE
protocols,” in Proc. 4th Annu. ACM/IEEE Int. Conf. Mobile Comput. Trans. Commun., vol. 66, no. 8, pp. 3571–3582, Aug. 2018.
Netw., Dallas, TX, USA, 1998, pp. 85–97. [74] H. Haas, “LiFi: Conceptions, misconceptions and opportunities,”
[52] X. Wu and H. Haas, “Mobility-aware load balancing for hybrid LiFi in Proc. IEEE Photon. Conf. (IPC), Waikoloa, HI, USA, 2016,
and WiFi networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 11, no. 12, pp. 680–681.
pp. 588–597, Dec. 2019. [75] L. Wu, Z. Zhang, J. Dang, and H. Liu, “Blind interference alignment for
[53] Y. Wang and H. Haas, “Dynamic load balancing with handover in multiuser MISO indoor visible light communications,” IEEE Commun.
hybrid Li-Fi and Wi-Fi networks,” J. Lightw. Technol., vol. 33, no. 22, Lett., vol. 21, no. 5, pp. 1039–1042, May 2017.
pp. 4671–4682, Nov. 15, 2015. [76] A. A. Qidan, M. Morales-Cespedes, and A. G. Armada, “The role of
[54] E. Demarchou, C. Psomas, and I. Krikidis, “Mobility management in WiFi in LiFi hybrid networks based on blind interference alignment,”
ultra-dense networks: Handover skipping techniques,” IEEE Access, in Proc. IEEE 87th Veh. Technol. Conf. (VTC Spring), Porto, Portugal,
vol. 6, pp. 11921–11930, 2018. 2018, pp. 1–5.
[55] M. D. Soltani, X. Wu, M. Safari, and H. Haas, “Access point selec- [77] Y. S. Eroǧlu, I. Güvenç, A. Sahin, Y. Yapici, N. Pala, and M. Yüksel
tion in Li-Fi cellular networks with arbitrary receiver orientation,” in “Multi-element VLC networks: LED assignment, power control, and
Proc. IEEE 27th Annu. Int. Symp. Pers. Indoor Mobile Radio Commun. optimum combining,” IEEE J. Sel. Areas Commun., vol. 36, no. 1,
(PIMRC), Valencia, Spain, 2016, pp. 1–6. pp. 121–135, Jan. 2018.
[78] H. B. Eldeeb, H. A. I. Selmy, H. M. Elsayed, and R. I. Badr,
[56] M. D. Soltani, A. A. Purwita, Z. Zeng, H. Haas, and M. Safari,
“Interference mitigation and capacity enhancement using constraint
“Modeling the random orientation of mobile devices: Measurement,
field of view ADR in downlink VLC channel,” IET Commun., vol. 12,
analysis and LiFi use case,” IEEE Trans. Commun., vol. 67, no. 3,
no. 16, pp. 1968–1978, Oct. 2018.
pp. 2157–2172, Mar. 2019.
[79] C. Chen, W.-D. Zhong, H. Yang, S. Zhang, and P. Du, “Reduction
[57] Z. Zeng, M. D. Soltani, H. Haas, and M. Safari, “Orientation model of of SINR fluctuation in indoor multi-cell VLC systems using opti-
mobile device for indoor VLC and millimetre wave systems,” in Proc. mized angle diversity receiver,” J. Lightw. Technol., vol. 36, no. 17,
IEEE 88th Veh. Technol. Conf. (VTC Fall), Chicago, IL, USA, 2018, pp. 3603–3610, Sep. 1, 2018.
pp. 1–6. [80] H.-N. Ryoo, D.-H. Kwon, S.-H. Yang, and S.-K. Han, “Differential
[58] A. A. Purwita, M. D. Soltani, M. Safari, and H. Haas, “Terminal orien- optical detection in VLC for inter-cell interference reduced flexible cell
tation in OFDM-based LiFi systems,” IEEE Trans. Wireless Commun., planning,” IEEE Photon. Technol. Lett., vol. 28, no. 23, pp. 2728–2731,
vol. 18, no. 8, pp. 4003–4016, Aug. 2019. Dec. 1, 2016.
[59] M. D. Soltani, A. A. Purwita, Z. Zeng, C. Chen, H. Haas, and M. Safari, [81] M. A. Atta and A. Bermak, “A polarization-based interference-tolerant
“An orientation-based random waypoint model for user mobility in VLC link for low data rate applications,” IEEE Photon. J., vol. 10,
wireless networks,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC no. 2, pp. 1–11, Apr. 2018.
Workshops), Dublin, Ireland, 2020, pp. 1–6. [82] H.-S. Kim, D.-R. Kim, S.-H. Yang, Y.-H. Son, and S.-K. Han,
[60] Z. Zeng, M. D. Soltani, Y. Wang, X. Wu, and H. Haas, “Realistic “Mitigation of inter-cell interference utilizing carrier allocation in visi-
indoor hybrid WiFi and OFDMA-based LiFi networks,” IEEE Trans. ble light communication system,” IEEE Commun. Lett., vol. 16, no. 4,
Commun., vol. 68, no. 5, pp. 2978–2991, May 2020. pp. 526–529, Apr. 2012.
[61] S. Jivkova and M. Kavehrad, “Shadowing and blockage in indoor [83] C. Chen, S. Videv, D. Tsonev, and H. Haas, “Fractional frequency reuse
optical wireless communications,” in Proc. IEEE Global Telecommun. in DCO-OFDM-based optical attocell networks,” J. Lightw. Technol.,
Conf. (GLOBECOM), vol. 6. San Francisco, CA, USA, 2003, vol. 33, no. 19, pp. 3986–4000, Oct. 1, 2015.
pp. 3269–3273. [84] K. Zhou, C. Gong, and Z. Xu, “Color planning and intercell
[62] X. Wu and H. Haas, “Load balancing for hybrid LiFi and WiFi interference coordination for multicolor visible light communication
networks: To tackle user mobility and light-path blockage,” IEEE networks,” J. Lightw. Technol., vol. 35, no. 22, pp. 4980–4993, Nov. 15,
Trans. Commun., vol. 68, no. 3, pp. 1675–1683, Mar. 2020. 2017.
[63] M. Hammouda, S. Akln, A. M. Vegni, H. Haas, and J. Peissig, [85] C. Quintana, J. Rabadan, J. Rufo, F. Delgado, and R. Perez-
“Hybrid RF/LC systems under QoS constraints,” in Proc. 25th Int. Jimenez, “Time-hopping spread-spectrum system for wireless optical
Conf. Telecommun. (ICT), St. Malo, France, 2018, pp. 312–318. communications,” IEEE Trans. Consum. Electron., vol. 55, no. 3,
[64] C. Chen, M. D. Soltani, M. Safari, A. A. Purwita, X. Wu, and H. Haas, pp. 1083–1088, Aug. 2009.
“An omnidirectional user equipment configuration to support mobility [86] A. M. Abdelhady, O. Amin, A. Chaaban, B. Shihada, and
in LiFi networks,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC M.-S. Alouini, “Downlink resource allocation for dynamic TDMA-
Workshops), Shanghai, China, 2019, pp. 1–6. based VLC systems,” IEEE Trans. Wireless Commun., vol. 18, no. 1,
[65] J. G. Andrews, “Interference cancellation for cellular systems: A pp. 108–120, Jan. 2019.
contemporary overview,” IEEE Wireless Commun., vol. 12, no. 2, [87] B. Lin, X. Tang, Z. Ghassemlooy, C. Lin, and Y. Li, “Experimental
pp. 19–29, Apr. 2005. demonstration of an indoor VLC positioning system based on
[66] K. Ying, H. Qian, R. J. Baxley, and S. Yao, “Joint optimization of OFDMA,” IEEE Photon. J., vol. 9, no. 2, pp. 1–9, Apr. 2017.
precoder and equalizer in MIMO VLC systems,” IEEE J. Sel. Areas [88] Z. Chen and H. Haas, “Space division multiple access in visible light
Commun., vol. 33, no. 9, pp. 1949–1958, Sep. 2015. communications,” in Proc. IEEE Int. Conf. Commun. (ICC), London,
U.K., 2015, pp. 5115–5119.
[67] Y. Hong, J. Chen, Z. Wang, and C. Yu, “Performance of a precoding
[89] T. Zahir, K. Arshad, A. Nakata, and K. Moessner, “Interference man-
MIMO system for decentralized multiuser indoor visible light commu-
agement in femtocells,” IEEE Commun. Surveys Tuts., vol. 15, no. 1,
nications,” IEEE Photon. J., vol. 5, no. 4, Aug. 2013, Art. no. 7800211.
pp. 293–311, 1st Quart., 2013.
[68] Z.-G. Sun, H.-Y. Yu, W.-C. Li, Z.-J. Tian, and Y.-J. Zhu, “Power- [90] H. Liu, Z. Lin, Y. Chen, and P. Xin, “Elite user clustering-based
efficient linear precoding for MU-MISO VLC systems with channel indoor heterogeneous VLC interference management and sub-channel
uncertainty,” IEEE Photon. Technol. Lett., vol. 30, no. 7, pp. 626–629, allocation strategy,” IEEE Access, vol. 8, pp. 43582–43591, 2020.
Apr. 1, 2018. [91] H. Liu, X. Pu, Y. Chen, J. Yang, and J. Chen, “User-centric access
[69] Q. Wang, Z. Wang, and L. Dai, “Multiuser MIMO-OFDM for visi- scheme based on interference management for indoor VLC-WIFI het-
ble light communications,” IEEE Photon. J., vol. 7, no. 6, pp. 1–11, erogeneous networks,” IEEE Photon. J., vol. 12, no. 4, pp. 1–12,
Dec. 2015. Aug. 2020.
[70] H. Marshoud, P. C. Sofotasios, S. Muhaidat, B. S. Sharif, and [92] A. Ahmed, L. M. Boulahia, and D. Gaiti, “Enabling vertical han-
G. K. Karagiannidis, “Optical adaptive precoding for visible light dover decisions in heterogeneous wireless networks: A state-of-the-art
communications,” IEEE Access, vol. 6, pp. 22121–22130, 2018. and a classification,” IEEE Commun. Surveys Tuts., vol. 16, no. 2,
[71] L. Yin, X. Wu, H. Haas, and L. Hanzo, “Low-complexity SDMA pp. 776–811, 2nd Quart., 2014.
user-grouping for the CoMP-VLC downlink,” in Proc. IEEE Global [93] Y. Li, B. Cao, and C. Wang, “Handover schemes in heterogeneous
Commun. Conf. (GLOBECOM), San Diego, CA, USA, 2015, pp. 1–6. LTE networks: Challenges and opportunities,” IEEE Wireless Commun.,
[72] Y.-Y. Zhang, H.-Y. Yu, and J.-K. Zhang, “Block precoding for peak- vol. 23, no. 2, pp. 112–117, Apr. 2016.
limited MISO broadcast VLC: Constellation-optimal structure and [94] A. M. Vegni and T. D. C. Little, “Handover in VLC systems with coop-
addition-unique designs,” IEEE J. Sel. Areas Commun., vol. 36, no. 1, erating mobile devices,” in Proc. Int. Conf. Comput. Netw. Commun.
pp. 78–90, Jan. 2018. (ICNC), Maui, HI, USA, 2012, pp. 126–130.
[95] X. Wu and H. Haas, “Handover skipping for LiFi,” IEEE Access, vol. 7, [119] Y. Zhuang et al., “A survey of positioning systems using visible LED
pp. 38369–38378, 2019. lights,” IEEE Commun. Surveys Tuts., vol. 20, no. 3, pp. 1963–1988,
[96] R. Arshad, H. Elsawy, S. Sorour, T. Y. Al-Naffouri, and M.-S. Alouini, 3rd Quart., 2018.
“Handover management in 5G and beyond: A topology aware skipping [120] D. C. O’brien et al., “Design and implementation of optical wireless
approach,” IEEE Access, vol. 4, pp. 9073–9081, 2016. communications with optically powered smart dust motes,” IEEE J. Sel
[97] A. A. Purwita, M. D. Soltani, M. Safari, and H. Haas, “Handover Areas Commun., vol. 27, no. 9, pp. 1646–1653, Dec. 2009.
probability of hybrid LiFi/RF-based networks with randomly-oriented [121] J. Vongkulbhisal, B. Chantaramolee, Y. Zhao, and W. S. Mohammed,
devices,” in Proc. IEEE 87th Veh. Technol. Conf. (VTC Spring), Porto, “A fingerprinting-based indoor localization system using intensity mod-
Portugal, 2018, pp. 1–5. ulation of light emitting diodes,” Microw. Opt. Technol. Lett., vol. 54,
[98] F. Wang, Z. Wang, C. Qian, L. Dai, and Z. Yang, “Efficient vertical no. 5, pp. 1218–1227, 2012.
handover scheme for heterogeneous VLC-RF systems,” IEEE/OSA J. [122] W. Xu, J. Wang, H. Shen, H. Zhang, and X. You, “Indoor positioning
Opt. Commun. Netw., vol. 7, no. 12, pp. 1172–1180, Dec. 2015. for multiphotodiode device using visible-light communications,” IEEE
[99] S. Liang, H. Tian, B. Fan, and R. Bai, “A novel vertical handover Photon. J., vol. 8, no. 1, pp. 1–11, Feb. 2016.
algorithm in a hybrid visible light communication and LTE system,” in [123] T. Q. Wang, Y. A. Sekercioglu, A. Neild, and J. Armstrong, “Position
Proc. IEEE 82nd Veh. Technol. Conf. (VTC Fall), Boston, MA, USA, accuracy of time-of-arrival based ranging using visible light with appli-
2015, pp. 1–5. cation in indoor localization systems,” J. Lightw. Technol., vol. 31,
[100] S. Liang, Y. Zhang, B. Fan, and H. Tian, “Multi-attribute vertical han- no. 20, pp. 3302–3308, Oct. 15, 2013.
dover decision-making algorithm in a hybrid VLC-femto system,” IEEE [124] S.-Y. Jung, S. Hann, and C.-S. Park, “TDOA-based optical wireless
Commun. Lett., vol. 21, no. 7, pp. 1521–1524, Jul. 2017. indoor localization using LED ceiling lamps,” IEEE Trans. Consum.
[101] Y. Wang, X. Wu, and H. Haas, “Fuzzy logic based dynamic handover Electron., vol. 57, no. 4, pp. 1592–1597, Nov. 2011.
scheme for indoor Li-Fi and RF hybrid network,” in Proc. IEEE Int. [125] Y. Shen and M. Z. Win, “Fundamental limits of wideband
Conf. Commun. (ICC), Kuala Lumpur, Malaysia, 2016, pp. 1–6. localization—Part I: A general framework,” IEEE Trans. Inf. Theory,
[102] X. Wu and H. Haas, “Access point assignment in hybrid LiFi and WiFi vol. 56, no. 10, pp. 4956–4980, Oct. 2010.
networks in consideration of LiFi channel blockage,” in Proc. IEEE [126] M.-G. Moon, S.-I. Choi, J. Park, and J. Y. Kim, “Indoor positioning
18th Int. Workshop Signal Process. Adv. Wireless Commun. (SPAWC), system using LED lights and a dual image sensor,” J. Opt. Soc. Korea,
Sapporo, Japan, 2017, pp. 1–5. vol. 19, no. 6, pp. 586–591, Dec. 2015.
[103] X. Wu and D. C. O’Brien, “A novel machine learning-based handover [127] G. B. Prince and T. D. C. Little, “A two phase hybrid RSS/AoA algo-
scheme for hybrid LiFi and WiFi networks,” in Proc. IEEE Global rithm for indoor device localization using visible light,” in Proc. IEEE
Commun. Conf. Workshop (GC Wkshps), Taipei, Taiwan, 2020, pp. 1–5. Global Commun. Conf. (GLOBECOM), Anaheim, CA, USA, 2012,
[104] “Requirements for evolved UTRA (E-UTRA) and evolved UTRAN pp. 3347–3352.
(E-UTRAN) (release 9),” 3GPP, Valbonne, France, Rep. TR 25.913, [128] N. U. Hassan, A. Naeem, M. A. Pasha, T. Jadoon, and C. Yuen, “Indoor
Dec. 2009. positioning using visible LED lights: A survey,” ACM Comput. Surveys,
[105] L.-H. Yen, T.-T. Yeh, and K.-H. Chi, “Load balancing in IEEE vol. 48, no. 2, pp. 1–32, 2015.
802.11 networks,” IEEE Internet Comput., vol. 13, no. 1, pp. 56–64,
[129] P. F. Mmbaga, J. Thompson, and H. Haas, “Performance analysis of
Jan./Feb. 2009.
indoor diffuse VLC MIMO channels using angular diversity detectors,”
[106] J. G. Andrews, S. Singh, Q. Ye, X. Lin, and H. S. Dhillon, “An J. Lightw. Technol., vol. 34, no. 4, pp. 1254–1266, Feb. 15, 2016.
overview of load balancing in HetNets: Old myths and open problems,”
[130] Y. U. Lee and M. Kavehrad, “Two hybrid positioning system design
IEEE Wireless Commun., vol. 21, no. 2, pp. 18–25, Apr. 2014.
techniques with lighting LEDs and ad-hoc wireless network,” IEEE
[107] X. Li, R. Zhang, and L. Hanzo, “Cooperative load balancing in
Trans. Consum. Electron., vol. 58, no. 4, pp. 1176–1184, Nov. 2012.
hybrid visible light communications and WiFi,” IEEE Trans. Commun.,
vol. 63, no. 4, pp. 1319–1329, Apr. 2015. [131] J. Ziyan, “A visible light communication based hybrid positioning
method for wireless sensor networks,” in Proc. 2nd Int. Conf. Intell.
[108] W. Ma and L. Zhang, “QoE-driven optimized load balancing design for
Syst. Design Eng. Appl., Sanya, China, 2012, pp. 1367–1370.
hybrid LiFi and WiFi networks,” IEEE Commun. Lett., vol. 22, no. 11,
pp. 2354–2357, Nov. 2018. [132] D. Konings, B. Parr, C. Waddell, F. Alam, K. M. Arif, and
[109] X. Wu, M. Safari, and H. Haas, “Joint optimisation of load balanc- E. M.-K. Lai, “HVLP: Hybrid visible light positioning of a mobile
ing and handover for hybrid LiFi and WiFi networks,” in Proc. IEEE robot,” in Proc. 24th Int. Conf. Mechatronics Mach. Vis. Pract.
Wireless Commun. Netw. Conf. (WCNC), San Francisco, CA, USA, (M2VIP), Auckland, New Zealand, 2017, pp. 1–6.
2017, pp. 1–5. [133] K. Ren, H. Su, and Q. Wang, “Secret key generation exploiting channel
[110] Y. Wang, X. Wu, and H. Haas, “Load balancing game with shadow- characteristics in wireless communications,” IEEE Wireless Commun.,
ing effect for indoor hybrid LiFi/RF networks,” IEEE Trans. Wireless vol. 18, no. 4, pp. 6–12, Aug. 2011.
Commun., vol. 16, no. 4, pp. 2366–2378, Apr. 2017. [134] J. M. Hamamreh, H. M. Furqan, and H. Arslan, “Classifications and
[111] M. Obeed, A. M. Salhab, S. A. Zummo, and M.-S. Alouini, “Joint applications of physical layer security techniques for confidentiality: A
optimization of power allocation and load balancing for hybrid comprehensive survey,” IEEE Commun. Surveys Tuts., vol. 21, no. 2,
VLC/RF networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 10, no. 5, pp. 1773–1828, 2nd Quart., 2019.
pp. 553–562, May 2018. [135] A. Thangaraj, S. Dihidar, A. R. Calderbank, S. W. McLaughlin, and
[112] L. Li, Y. Zhang, B. Fan, and H. Tian, “Mobility-aware load balancing J.-M. Merolla, “Applications of LDPC codes to the wiretap channel,”
scheme in hybrid VLC-LTE networks,” IEEE Commun. Lett., vol. 20, IEEE Trans. Inf. Theory, vol. 53, no. 8, pp. 2933–2945, Aug. 2007.
no. 11, pp. 2276–2279, Nov. 2016. [136] Z. Chen and H. Haas, “Physical layer security for optical attocell
[113] Y. Wang, D. A. Basnayaka, X. Wu, and H. Haas, “Optimization of networks,” in Proc. IEEE Int. Conf. Commun. (ICC), Paris, France,
load balancing in hybrid LiFi/RF networks,” IEEE Trans. Commun., 2017, pp. 1–6.
vol. 65, no. 4, pp. 1708–1720, Apr. 2017. [137] A. Mostafa and L. Lampe, “Physical-layer security for MISO visible
[114] X. Wu, M. Safari, and H. Haas, “Access point selection for hybrid light communication channels,” IEEE J. Sel. Areas Commun., vol. 33,
Li-Fi and Wi-Fi networks,” IEEE Trans. Commun., vol. 65, no. 12, no. 9, pp. 1806–1818, Sep. 2015.
pp. 5375–5385, Dec. 2017. [138] T. S. Humble, “Quantum security for the physical layer,” IEEE
[115] R. Ahmad, M. D. Soltani, M. Safari, A. Srivastava, and A. Das, Commun. Mag., vol. 51, no. 8, pp. 56–62, Aug. 2013.
“Reinforcement learning based load balancing for hybrid LiFi WiFi [139] H. Chun et al., “Handheld free space quantum key distribution
networks,” IEEE Access, vol. 8, pp. 132273–132284, 2020. with dynamic motion compensation,” Opt. Exp., vol. 25, no. 6,
[116] G. Pan, H. Lei, Z. Ding, and Q. Ni, “3-D hybrid VLC-RF indoor IoT pp. 6784–6795, 2017.
systems with light energy harvesting,” IEEE Trans. Green Commun. [140] G. Pan, J. Ye, and Z. Ding, “Secure hybrid VLC-RF systems with
Netw., vol. 3, no. 3, pp. 853–865, Sep. 2019. light energy harvesting,” IEEE Trans. Commun., vol. 65, no. 10,
[117] S. K. Moore, “Superaccurate GPS chips coming to smartphones in pp. 4348–4359, Oct. 2017.
2018,” in IEEE Spectrum: Technology, Engineering, and Science News. [141] M. F. Marzban, M. Kashef, M. Abdallah, and M. Khairy,
Piscataway, NJ, USA, IEEE Spectr., Sep. 2017. “Beamforming and power allocation for physical-layer security in
[118] J. Luo, L. Fan, and H. Li, “Indoor positioning systems based on visible hybrid RF/VLC wireless networks,” in Proc. 13th Int. Wireless
light communication: State of the art,” IEEE Commun. Surveys Tuts., Commun. Mobile Comput. Conf. (IWCMC), Valencia, Spain, 2017,
vol. 19, no. 4, pp. 2871–2893, 4th Quart., 2017. pp. 258–263.
[142] S. Ucar, S. C. Ergen, and O. Ozkasap, “IEEE 802.11p and visible light Majid Safari (Senior Member, IEEE) received
hybrid communication based secure autonomous platoon,” IEEE Trans. the B.Sc. degree in electrical and computer engi-
Veh. Technol., vol. 67, no. 9, pp. 8667–8681, Sep. 2018. neering from the University of Tehran, Iran, in
[143] Z. Liao, L. Yang, J. Chen, H.-C. Yang, and M.-S. Alouini, “Physical 2003, the M.Sc. degree in electrical engineering
layer security for dual-hop VLC/RF communication systems,” IEEE from the Sharif University of Technology, Iran, in
Commun. Lett., vol. 22, no. 12, pp. 2603–2606, Dec. 2018. 2005, and the Ph.D. degree in electrical and com-
[144] J. Al-khori, G. Nauryzbayev, M. M. Abdallah, and M. Hamdi, “Secrecy puter engineering from the University of Waterloo,
performance of decode-and-forward based hybrid RF/VLC relaying Canada, in 2011. He is currently a Reader with
systems,” IEEE Access, vol. 7, pp. 10844–10856, 2019. the Institute for Digital Communications, University
[145] F. Agnelli et al., “Wireless multi-standard terminals: System analysis of Edinburgh. Before joining Edinburgh in 2013,
and design of a reconfigurable RF front-end,” IEEE Circuits Syst. Mag., he held a Postdoctoral Fellowship with McMaster
vol. 6, no. 1, pp. 38–59, 1st Quart., 2006. University, Canada. His main research interest are in the application of
[146] A. Ford, C. Raiciu, M. Handley, and O. Bonaventure, “TCP extensions information theory and signal processing in optical communications, includ-
for multipath operation with multiple addresses,” IETF Trust, Reston, ing fiber-optic communication, free-space optical communication, visible light
VA, USA, RFC 6824, Jan. 2013. communication, and quantum communication. He is currently an Associate
[147] S. Shakkottai, T. S. Rappaport, and P. C. Karlsson, “Cross-layer Editor of IEEE T RANSACTIONS ON C OMMUNICATIONS and he was the
design for wireless networks,” IEEE Commun. Mag., vol. 41, no. 10, TPC Co-Chair of the 4th International Workshop on Optical Wireless
pp. 74–80, Oct. 2003. Communication in 2015.
Xiping Wu (Member, IEEE) received the B.Sc.

degree from Southeast University, China, in 2008,
and the M.Sc. and Ph.D. degrees from the University
of Edinburgh, U.K., in 2011 and 2015, respec-
tively. He is currently an Assistant Professor with
the School of Electrical and Electronic Engineering,
University College Dublin (UCD), Ireland. Prior to
joining UCD, he was a Research Fellow with the
University of Oxford, working on visible light com-
munications and the Internet of Things (IoT). He has
authored or coauthored over 50 journal and confer-
ence articles in these areas. His main research interests are in the areas of
optical wireless communications, hybrid network management and integration
of OWC and radio frequency technologies, and the relevant applications in
self-powered IoT systems.
Mohammad Dehghani Soltani received the

M.Sc. degree from the Department of Electrical
Engineering, Amirkabir University of Technology,
Tehran, Iran, in 2012, and the Ph.D. degree in elec-
trical engineering from the University of Edinburgh,
Edinburgh, U.K., in 2019. During his M.Sc., he stud-
ied wireless communications, MIMO coding, and
low complexity design of MIMO-OFDM systems.
He worked for two years in the telecommunica- Harald Haas (Fellow, IEEE) received the Ph.D.
tion industry in Iran. During his Ph.D., he studied degree from the University of Edinburgh in 2001.
the effect of device orientation, user mobility, and He is the Director of the LiFi Research and
link blockage on LiFi channel as well as handover management. He is cur- Development Center, University of Strathclyde. He
rently a Research Associate with the Institute for Digital Communications, is also the Initiator, the Co-Founder, and the Chief
University of Edinburgh, funded by the British Engineering and Physical Scientific Officer of pureLiFi Ltd. He has authored
Sciences Research Council. His current research interests include mobility 550 conference and journal papers, including papers
and handover management in wireless cellular networks, optical wireless in Science and Nature Communications. His main
communications, visible light communications, and LiFi. research interests are in optical wireless communi-
cations, hybrid optical wireless and RF communica-
tions, spatial modulation, and interference coordina-
tion in wireless networks. His team invented spatial modulation. He introduced
LiFi to the public at an invited TED Global talk in 2011. This talk on Wireless
Data from Every Light Bulb has been watched online over 2.72 million times.
He was listed among the 50 best inventions in TIME Magazine in 2011.
Lai Zhou received the B.S. and Ph.D. degrees He gave a second TED Global lecture in 2015 on the use of solar cells
from Tsinghua University, Beijing, China, in 2013 as LiFi data detectors and energy harvesters. This has been viewed online
and 2018, respectively. He is currently an Assistant over 2.75 million times. In 2016, he received the Outstanding Achievement
Researcher with the Beijing Academy of Quantum Award from the International Solid State Lighting Alliance. In 2019, he was
Information Sciences, China. Before that, he was a a recipient of the IEEE Vehicular Society James Evans Avant Garde Award.
Postdoctoral Research Assistant with the University In 2017, he received a Royal Society Wolfson Research Merit Award. He
of Oxford, working on quantum communications. was elected a Fellow of the Royal Society of Edinburgh in 2017. In 2018,
His research interests include quantum key dis- he received a three-year EPSRC Established Career Fellowship extension and
tribution, integrated photonic chip, and quantum was elected Fellow of the IET. He was elected Fellow of the Royal Academy
repeaters. of Engineering (FREng) in 2019.

2021 Hybrid LiFi and WiFi Networks A Survey

Uploaded by

Copyright:

Available Formats

2021 Hybrid LiFi and WiFi Networks A Survey

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2021 Hybrid LiFi and WiFi Networks A Survey

Uploaded by

Copyright:

Available Formats

1398 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 23, NO.

2, SECOND QUARTER 2021

Hybrid LiFi and WiFi Networks: A Survey

Abstract—In order to tackle the rapidly growing number

Fig. 2. Schematic diagram of an SDN-enabled HLWNet.

first approach is to extend the current autonomous network

schemes are summarized from the perspective of standard-

number of factors related to LEDs, including their loca-

Fig. 5. Taxonomy of user behavior modeling in LiFi-related networks.

the movement has a memory and/or restriction, these models

Fig. 7. Taxonomy of interference management in LiFi.

V. I NTERFERENCE M ANAGEMENT transmission. MUD aims to cancel interference among co-

Fig. 8. Taxonomy of Handover in LiFi and HLWNets.

Fig. 10. Taxonomy of load balancing in HLWNets.

shown in Fig. 10, and the current literature is summarized in

A. Stationary Channel Load Balancing

Fig. 11. Taxonomy of applications benefiting from HLWNets.

Xiping Wu (Member, IEEE) received the B.Sc.

Mohammad Dehghani Soltani received the

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.