2230 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO.
4, FOURTH QUARTER 2022
Digital Twin of Wireless Systems: Overview,
Taxonomy, Challenges, and Opportunities
Latif U. Khan , Zhu Han , Fellow, IEEE, Walid Saad , Fellow, IEEE, Ekram Hossain , Fellow, IEEE,
Mohsen Guizani , Fellow, IEEE, and Choong Seon Hong , Senior Member, IEEE
Abstract—Future wireless services will focus on improving
the quality of life by enabling various applications, such as
extended reality, brain-computer interaction, and healthcare.
These applications will have diverse performance requirements
(e.g., user-defined quality of experience metrics, latency, and
reliability) which will be challenging to be fulfilled by exist-
ing wireless systems. To meet the diverse requirements of the
emerging applications, the concept of digital twins has been
recently proposed. A digital twin uses a virtual representation
along with security-related technologies (e.g., blockchain), com-
munication technologies (e.g., 6G), computing technologies (e.g.,
edge computing), and machine learning, so as to enable the
smart applications. In this tutorial, we present a comprehensive
overview on digital twins for wireless systems. First, we present
the fundamental concepts (i.e., design aspects, high-level archi-
tecture, and frameworks) of digital twins for wireless systems.
Second, a comprehensive taxonomy is devised for two aspects,
namely, twins for wireless and wireless for twins. For the twins
for wireless aspect, we consider issues related to design of twin
objects, physical devices, and interface, as well as prototyping,
deployment trends, incentive mechanism, isolation of twins, and
decoupling. For the wireless for twins aspect, we consider issues
related to accessing twin objects, security and privacy, and air
interface design are considered. Finally, open research challenges
and opportunities are discussed.
Manuscript received 4 February 2022; revised 18 May 2022; accepted
8 July 2022. Date of publication 11 August 2022; date of current version
22 November 2022. This work was supported in part by the National Research
Foundation of Korea (NRF) Grant funded by the Korea Government (Ministry
of Science and ICT: MSIT) under Grant 2020R1A4A1018607; in part by
the Institute of Information and Communications Technology Planning and
Evaluation (IITP) Grant funded by the Korea Government (MSIT, Evolvable
Deep Learning Model Generation Platform for Edge Computing) under Grant
2019-0-01287; and in part by IITP Grant funded by the Korea Government
(MSIT, Artificial Intelligence Innovation Hub) under Grant 2021-0-02068.
(Corresponding author: Choong Seon Hong.)
Latif U. Khan and Choong Seon Hong are with the Department of Computer
Science and Engineering, Kyung Hee University, Yongin 17104, South Korea
(e-mail: cshong@khu.ac.kr).
Zhu Han is with the Electrical and Computer Engineering Department and
the Computer Science Department, University of Houston, Houston, TX 77004
USA, and also with the Department of Computer Science and Engineering,
Kyung Hee University, Seoul 02447, South Korea.
Walid Saad is with the Wireless@VT, Bradley Department of Electrical
and Computer Engineering, Virginia Tech, Blacksburg, VA 24061 USA, and
also with the Department of Computer Science and Engineering, Kyung Hee
University, Seoul 02447, South Korea.
Ekram Hossain is with the Department of Electrical and Computer
Engineering, University of Manitoba, Winnipeg, MB R3T 5V6, Canada.
Mohsen Guizani is with the Machine Learning Department, Mohamed
Bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE (e-mail:
mguizani@ieee.org).
Digital Object Identifier 10.1109/COMST.2022.3198273
1553-877X 
c 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
Index Terms—Digital twin, wireless system, machine learning,
federated learning, virtual modeling.
I. I NTRODUCTION
MERGING Internet of Everything (IoE) applications,
E such as haptics, brain-computer interaction, flying vehi-
cles, and extended reality (XR), among others, will enable a
merger of digital and physical worlds [1], [2], [3], [4]. These
IoE applications have widely diverse requirements (e.g., user
experience, reliability, and latency). To meet these diverse
requirements, there is a need to assist the wireless systems
by novel technologies. These new technologies will enable
the wireless systems to meet the diverse requirements via
enabling two main trends: self-configuring wireless systems
and proactive-online-learning-enabled systems [5]. Generally,
wireless systems rely on intelligent, seamless, and ubiquitous
connectivity for meeting the diverse requirements of end-users.
To enable wireless systems with these features, there is a
need for self-configuring wireless systems. Self-configuring
wireless systems will offer efficient management of wire-
less systems with minimum possible intervention from the
network operators/end-users. On the other hand, IoE services
have highly dynamic requirements in terms of user-defined
metrics, latency, data rate, and reliability, among others. To
meet these highly dynamic requirements, there is a need to
efficiently enable interaction among various entities of wire-
less systems and technologies which include wireless chan-
nel resources, security-related technologies (e.g., blockchain),
computing technologies (e.g., edge computing). Therefore,
upon request from the end-users, e.g., for latency-constrained
services (e.g., extended reality), intelligent analytics may need
to be performed for efficient resource management. Therefore,
proactive-online learning-based wireless systems will need to
be developed.
To design a self-configuring and proactive-online-learning-
enabled wireless system, we can create a digital twin to
represent the wireless system [5]. A digital twin will use a
virtual representation of the physical system to enable IoE
applications [5], [6], [7]. A digital twin takes inputs from
a real world and makes predictions and/or control decisions.
For this, in addition to the virtual representation of a wireless
system, a digital twin-based wireless system will use tools
from optimization theory, game theory, and machine learning.
Additionally, to enable transparent and immutable handling of
data, a digital twin-based system will use blockchain. Note that
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
Fig. 1. Conceptual overview of a digital twin for a wireless system.
a digital twin is different from traditional simulation tools that
enable us to test systems, processes, and concepts and effects
of different system variables using computer-based models
and/or mathematical concepts. A digital twin can use real-
time (i.e., actual) data generated at the sensors attached to the
physical objects and run simulations and analysis for online
control of wireless systems. The two-way flow of information
between the twin and the sensors can increase the accuracy of
predictive analytical models. A digital twin provides a more
general platform for simulations and decision making.
An overview of digital twin-based wireless systems is given
in Fig. 1. We can divide the twin-based wireless systems archi-
tecture into the physical interaction layer and the twin layer.
The physical interaction layer covers all the physical objects
(i.e., end-devices, edge/cloud servers, base stations, and core
network elements) necessary for a given wireless system appli-
cation. On the other hand, the twin layer is a logical layer
that contains logical twin objects. A more detailed discus-
sion about the creation of twin objects will be provided in
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2231
Sections II-C and III-C). Depending the latency constraints, a
digital twin can be deployed either at the network edge or at
the network core. A twin at the network edge can be prefer-
ably used for latency-constrained applications, whereas a twin
deployed at the network core can be used for applications that
requires more computing power.
A. Market Statistics and Research Trends
The Internet of Things (IoT) market will grow at a
Compound Annual Growth Rate (CAGR) of 26.9% during
the period of 2017-2022 [8]. The market share will grow
from 170.6 Billion USD in 2017 to 561.0 Billion USD in
2022. The main causes of an increase in the IoT market are
smart buildings, smart grids, smart industries, and intelligent
transportation. The key players in the IoT market are General
Electric (U.S.), Bosch Software Innovation GmbH (Stuttgart,
Germany), Amazon Web Services Inc. (U.S.), Hewlett-Packard
Enterprise (U.S.), Google Inc. (U.S.), PTC Inc. (U.S.),
2232
S UMMARY OF E XISTING S URVEYS AND T UTORIALS AND T HEIR P RIMARY F OCUS
Fig. 2. Publications count for (a) digital twin and (b) IoT.
International Business Machine (IBM) Corporation (U.S.),
Oracle Corporation (U.S.), Microsoft Corporation (U.S.),
Cisco Systems, Inc. (U.S.), SAP SE (Walldorf, Germany),
and Intel Corporation (U.S.). Among different regions (i.e.,
Latin America, MEA, APAC, Europe, and North America)
it is expected that China will lead the IoT market in APAC
region [9]. Globally, the U.S. will have the highest share in
the IoT market by 2029 [10] and will be followed by China,
Japan, and Germany.
According to statistics of Markets and Markets, the market
of digital twins will grow at a CAGR rate of 58% during the
period of 2020-2026 [11]. The market value of digital twin
in 2020 was 3.1 Billion USD and it will reach 48.2 Billion
USD by 2026. The key factors in the rise of the digital twin
market include the rise in the demand for manufacturing mon-
itoring assets, intelligent analytics-based healthcare systems,
smart warehouses, and intelligent transportation, among oth-
ers [11], [12]. Due to the increasing importance of digital
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
TABLE I
twins in the development of smart applications, various coun-
tries, such as Brazil, Norway, Mexico, China, Japan, South
Korea, and Singapore, are trying to implement twin-based
systems. The key market players in the digital twin mar-
ket are SWIM.AI (USA), Robert Bosch (Germany), ANSYS
(USA), Siemens AG (USA), Oracle (USA), SAP (Germany),
Microsoft Corporation (USA), PTC (USA), IBM (USA), and
General Electric (USA). Additionally, digital twins such as
product digital twin, process digital twin, and system digital
twin, will have the highest market share by 2026. On the other
hand, the number of publications for digital twin and IoT is
shown in Fig. 2. Fig. 2 clearly shows a significant increase in
both domains.
B. Existing Surveys and Tutorials
Few surveys and tutorials have reviewed digital
twins [5], [13], [14], [15], [16], [17]. The authors in [13]
focused on digital twins in the context of IoT. They discussed
the digital twin concept with architectural elements as well as
key enablers. Another work [14] surveyed the key technolo-
gies along with the use cases of digital twins toward enabling
IoT applications. Barricelli et al. in [15] presented the key
concepts, applications, and design implications. Furthermore,
they presented few open research challenges. On the other
hand, the works in [16] and [17] focused mainly on the role
of blockchain in enabling digital twins. The authors in [16]
presented the key benefits of using blockchain for digital
twins. Additionally, they devised taxonomy and presented
a few open challenges. Suhail et al. in [17] systematically
reviewed the role of blockchain in enabling digital twins.
The work in [5] presented the role of the digital twin toward
enabling of 6G wireless system. Additionally, the authors
provided architectural trends for twin-based wireless systems.
Different from the existing surveys and tutorials on digital
twins [5], [13], [14], [15], [16], [17], the goal of our survey
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
Fig. 3. Organization of the tutorial.
is to comprehensively discuss the fellowship of digital twin
and wireless systems, as given in Table I. We present the
fundamentals of digital twins and a general definition in the
context of wireless systems. A general architecture along with
design aspects is presented. We also present a comprehensive
taxonomy of digital twins-based wireless systems and open
research challenges.
C. Our Tutorial
Our tutorial (see organization in Fig. 3) aims to answer the
following questions:
• What is a digital twin in the context of wireless systems?
What are the main design aspects for digital twins of
wireless systems?
• How to use a digital twin for enabling wireless systems?
What are the challenges for digital twin signaling over a
wireless channel?
• How to classify the research areas for digital twins of
wireless systems?
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2233
• What are the open research challenges and the potential
solution approaches in enabling digital twins of wireless
systems?
The tutorial will focus on two main aspects: digital twins for
wireless and wireless for twins. The former deals with the role
of a digital twin in enabling wireless systems. The latter deals
with efficient utilization of wireless resources for enabling
effective twin signaling over a wireless link. Different from
the existing works [5], [13], [14], [15], [16], [17], we present
the concept, key design aspects, and high-level architecture
for digital twins of wireless systems. We also present a com-
prehensive taxonomy that covers both twins for wireless and
wireless for twins aspects. Furthermore, several open research
challenges are discussed. Our contributions are summarized as
follows.
• We present the key concept, main design aspects, and
high-level architecture for digital twins of wireless
systems.
• A comprehensive taxonomy considering both twins for
wireless and wireless for twins, is provided. We con-
sider issues related to designing twin objects, prototyping
twin objects, interface design, as well as twin objects
deployment trends, incentive mechanism, twin objects
access aspects, isolation and decoupling of twins, security
and privacy, and air interface design for communication
between a twin and the physical objects.
• Several open challenges are presented. Moreover, promis-
ing solution approaches are discussed.
II. D IGITAL T WINS : C ONCEPT, K EY D ESIGN
A SPECTS , AND F RAMEWORKS
A. Concept
A digital twin is a virtual representation of the physi-
cal system serving as a digital counterpart [18], [19]. The
main purpose of a digital twin is to jointly optimize the cost
and performance of a system or overall process using vari-
ous emerging technologies (e.g., virtual modeling, simulation
technologies, blockchain, edge computing, cloud comput-
ing) and optimization tools. Digital twin provides proactive
analysis of the physical process using various simulation
tools (e.g., AMEsim, SimScale, Simulink [20], [21]), arti-
ficial intelligence, mathematical optimization, game theory,
and graph theory, among others [5]. Digital twin technol-
ogy has gained significant interest since 2002 when the
firm, namely, Challenge Advisory hosted a presentation at
University of Michigan [22]. They discussed the fundamental
elements of a digital twin, such as virtual space, real space,
and the information flow between them. Prior to the event
at the University of Michigan, the U.S. space agency NASA
proposed the use of a digital twin around 1960’s for analyzing
the space systems at the ground.
Digital twins are categorized in many ways, as given
in Table II. According to Siemens, a digital twin can be
divided into product digital twins, production digital twins,
and performance digital twins [23]. Furthermore, the twins
can be categorized into status twin, simulation twin, and
2234 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022

TABLE II
D IGITAL T WIN C ATEGORIES : S OURCE , T YPE , AND E XPLANATION

operational twin [24]. The classification of twins by dif-

ferent sources is based on their objective and coverage
(i.e., single entity or entire system). From all the types of
twins [23], [24], [25], [26], we generalize the definition and
types of a twin for a wireless system, as shown in Fig. 4.
Digital twin for 6G can be intended for a single entity (i.e.,
edge server management and IoE end-device management),
an end-to-end service (i.e., resource management, new service
design, and network planning), and multi-services. A digi-
tal twin for 6G can enable analysis, design, and real-time
monitoring with control of devices to enable IoE services for
cost-efficient and resource optimized operation.
The implementation process of a digital twin starts with the
analysis of a physical system, as shown in Fig. 5. Such analysis
gives us insights into the system specifications, inputs, outputs,
and environment dynamics. Next to physical system analysis,
one can design a virtual system that represents the physical
system. Modeling a virtual system requires careful design and Fig. 4. Conceptual overview of a digital twin for a wireless system.
requires overcoming many challenges. The challenges include
an accurate representation of a system using mathematical conditions) are very difficult to model accurately due to their
equations. Additionally, some of the real-time system param- uncertain nature with time. Therefore, these parameters need to
eters (e.g., IoE device operating frequency, wireless channel be accurately estimated while performing the simulation of the

Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
Fig. 5. Digital twin design steps.
Fig. 6. Digital twin and wireless system: design aspects and challenges.
virtual model. After virtual modeling, the virtual design needs
to be verified by simulating the virtual model and comparing
the performance with the physical system. In this phase, the
virtual model may need o be modified to make it as close as
possible to the physical system. The final phase is the opera-
tion phase that involves controlling the real-time system using
a digital twin.
B. Design Aspects
We consider both of the following aspects: (a) twins for
wireless and (b) wireless for twins, as shown in Fig. 6.
The former deals with the design of twins to enable vari-
ous network functions/ applications. On the other hand, the
latter deals with efficient communication methods to enable
signaling for implementation of twins. A twin-enabled wireless
network will have a variety of players to enable the success-
ful operation of IoE services. These players are blockchain
networks, edge/cloud servers, data decoupling interfaces, func-
tion decoupling interfaces, and physical devices. Furthermore,
the requirements of various IoE applications can be signifi-
cantly diverse.
Wireless resources can be used in twinning mainly for
two operations: (a) training of twin objects and (b) sig-
naling for twin operation. Twins training will use wireless
resources to transfer data and learning updates. For central-
ized learning, a wireless channel is used to migrate the data
from end-devices to the shared storage (i.e., servers installed
at the edge/cloud). However, some practical scenarios (e.g.,
autonomous driving cars) frequently generate a significant
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2235
amount of data. Transferring such an enormous amount of data
to the shared storage will use a significant amount of commu-
nication resources. To address this issue, distributed learning
(e.g., federated learning [FL]) can be used. FL is based on
sending only learning model updates rather than the whole
data, and thus consumes fewer communication resources com-
pared to centralized learning. To train digital twin models
using distributed learning over wireless networks, a signifi-
cant amount of resources will be required if the number of
global rounds between devices and the global server is high.
Meanwhile, the wireless channel will degrade the performance
of distributed learning-based twin model during the transfer of
learning model updates between end-devices and aggregation
server [27]. Additionally, variable latencies will be induced
for transferring learning model updates between devices and
twin model servers. For a twin model update of size u and
channel throughput Γ, the transmission latency can be given
by d = Γ u . The transmission latency can be minimized, for
example, by reducing the size of twin model updates, and
improving the throughput, e.g., by performing efficient asso-
ciation of devices with edge/cloud server and efficient wireless
resource allocation [28], [29].
C. High-Level Architecture
A high-level architecture of a digital twin for wire-
less systems based on logical twin objects is shown in
Fig. 7 [5], [14], [30]. The architecture can be divided into
three layers, such as the physical devices interaction layer, the
twin objects layer, and the services layer. The services layer
contains interfaces for applications. One can request a service
from a twin-based wireless system. In response to the user
request, semantic reasoning schemes are used to translate the
request, which is then passed to the twin layer. The twin layer
contains logical twin objects which use a virtual representation
of the physical object/ phenomenon. For a digital twin, there
can be different models for the physical objects/ phenomenon,
such as mathematical model, 3D model, and data-driven
model [31], [32]. Representing a physical object using a math-
ematical model needs several assumptions [33]. Another way
can be to use 3D modeling. However, both mathematical and
3D models (e.g., 3D content in AR) may not accurately model
the physical model/ phenomenon. To address these limitations,
data-driven models based on machine learning can be used to
effectively model the physical objects [34], [35], [36], [37].
Note that the twin layer objects can be implemented using con-
tainer/s or virtual machine/s [5], [38]. Also, twin objects can
be deployed either at the network edge or at the remote cloud.
Implementing a twin object at the remote cloud can offer more
computing power but at the cost of high latency [39], [40]. On
the other hand, twin objects deployed at the network edge
can have low computing power [41], [42]. The last layer
is the devices’ physical interaction layer, which contains all
the physical devices, such as end-devices, edge/cloud servers,
base stations, miners, and core network switches, among oth-
ers. Effective interfaces are required between different layers
of in a twin-based wireless system. These interfaces can be
2236
Fig. 7. High-level architecture of a digital twin enabled wireless system.
twin-to-physical object interface, twin-to-twin interface, and
twin-to-service interface.
For a digital-twin-enabled wireless system, there are two
aspects of reliability: twin reliability and twin-based service
reliability [43]. Twin reliability refers to the operation of a
twin with minimum possible interruption due to the failure
of the edge/cloud server running the twin object. An IoT
service based on a single twin deployed at the cloud has
lower reliability than an IoT service based on multiple twin
objects deployed at edge servers. However, the management of
multiple twins for a certain service will incur more complexity.
Additionally, to ensure a reliable twin signaling in presence
of wireless channel fading and interference, channel coding
schemes (e.g., URLLC codes) and other techniques such as
multi-connectivity, packet duplication may need to be used.
On the other hand, twin service reliability mainly depends on
wireless channel reliability and reliability of edge/cloud com-
puting. Similar to twin signaling, channel coding schemes can
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
be used for this [44], [45], [46]. A digital twin can be used
for predictive maintenance of 6G systems to avoid system
malfunctions and cyber-attacks through artificial intelligence
analytics and simulation. The summary of the steps for a
twin-based wireless system operation is as follows.
• First of all, the end-user will request a service. This
request will be translated using semantic reasoning
schemes to make it compatible with the twin object-based
architecture.
• Next, the twin object will be created to serve the end-user.
• Finally, the twin object uses mathematical optimization
and machine learning schemes in addition to virtual rep-
resentation to enable efficient resource optimization for
various services. The twin objects will use pre-trained
twin model to serve the requesting user. Meanwhile,
the newly generated data may be used by twin object
to further train the pre-trained model for performance
enhancement. After training, the twin model will be
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
stored using a blockchain network. Finally, after serv-
ing the requesting user, the twin objects based on virtual
machines/ containers are released.
D. Frameworks
In this section, we discuss various frameworks designed for
implementing digital twins. Moreover, we critically analyze
them and discuss their advantages and limitations.
1) Eclipse Ditto: Eclipse Ditto is an open-source frame-
work for creating digital IoT twins [47]. The framework
consists of Ditto services (components), external dependen-
cies (MongoDB and nginx), and application programming
interfaces (APIs). Ditto consists of microservices, each with
its own data store where reading and writing take place. Every
microservice has a set of APIs (events, command responses,
commands). Moreover, a microservice can communicate with
other microservices using defined signals. Within a Ditto clus-
ter, all microservices can communicate asynchronously using
an open-source toolkit, namely, AKKA remoting. Therefore,
each service can act as a client, enabling a TCP end-
point, and a server. The messages between microservices
are serializable (i.e., from Java objects to JavaScript Object
Notation(JSON)) and deserializable (i.e., from JSON to Java
objects).
2) Model Conductor-eXtended Framework: Model
Conductor-eXtended (MCX) is an open-source framework
for digital twins experimentation [48], [49]. The framework
enables us to co-execute the digital systems and physical
systems as well as asynchronous communication. Furthermore,
support for machine learning models, customized models, and
running FMUs (simulation models packaged according to the
FMI specification) is also provided. The MCX uses standard
data transmission protocol, time-synchronous implementation
of time-consuming simulation models, and decoupling of
the queue and the model computation module, for offering
scalability which is one of the important design aspects of
IoT systems.
3) Mago3D: Mago3D is an open-source digital twin plat-
form developed by Gaia3D Inc. [50]. The purpose is to model
real-world objects, phenomenon, and processes on Web envi-
ronment [51]. The platform consists of a geospatial data server,
data conversion server, platform core server, and Web server,
for realizing the digital twin-based architecture. Mago3D has
been applied in various sectors, such as Korean national
defense, indoor data management, shipbuilding, and urban
management, and has shown good results.
All of the aforementioned digital twin platforms (i.e.,
Eclipse Ditto, Model Conductor-eXtended Framework, and
Mago3D) are promising for the realization of digital twin-
based systems. However, none of them effectively considers
the effects of wireless channels on the performance of digi-
tal twins. In wireless systems, wireless channel uncertainties
will significantly affect the performance of IoT applica-
tions. Therefore, these wireless channel uncertainties need
to be modeled effectively for digital twinning of wireless
systems.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2237
E. Example Scenario
Here, we present a example scenario that will show how
do we use digital twin to enable IoE application. We con-
sider an example of XR that is enabled by digital twin [5].
Consider an XR device that requests a service from the
access point. First of all, authentication will be performed
by the digital twin-based wireless system at the service
layer (shown in Fig. 7) in response to the end-user request.
Next to authentication, a translation scheme will translate
the user request into the form understandable by the twin-
based wireless system. This translation, which is performed
using a semantic reasoning scheme, will enable understand-
ing of end user request by twin objects deployed at network
edge/cloud. The next step will be instantiation of the twin
object/s. The deployment of twin object/s will be either at
the network edge or the cloud depending on the latency con-
straints. These twin objects should be associated with the
blockchain so that twin objects can reuse the pretrained mod-
els stored in blockchain. Additionally, the blockchain stores
the data (e.g., training data for proactive analytics) required
for efficient analysis. To serve the requesting user, twin
object/s will use pretrained models in addition to mathemat-
ical optimization to perform efficient resource management
(i.e., computing and communication resources) and other
functions (e.g., caching decisions for frequently requested
contents). Also, twin control signaling must be efficiently
performed.
F. Summary and Lessons Learned
We have discussed the fundamentals of digital twins in
the context of wireless systems and presented the key design
aspects related to “twins for wireless” and “wireless for twins”.
Several available frameworks for digital twins experimentation
have also been discussed. The key lessons learned here are as
follows:
• A digital twin should be designed in a generalized way
so that it can be easily reusable for future services.
Designing a twin takes significant effort and time. Twins
based on machine learning should be trained using more
data to make them generalized so as to use them for
multiple scenarios. Such kinds of twins will minimize the
design time of the service based on twins. Furthermore,
this twin will reduce the design cost.
• The current digital twin frameworks are designed with-
out effectively taking into account the wireless chan-
nel impairments. Novel frameworks will need to be
designed for digital twinning over wireless systems. Such
a framework should exploit effective methods for wire-
less resource allocation and association of user devices
to twin objects.
• Digital twins for a wireless network can be used for
effective use of computing and communication resources
and serving users with different requirements (“twin for
wireless”). On the other hand, efficient wireless resource
management methods will be required for twin signaling
(“wireless for twins”).
2238 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022

Fig. 8. Taxonomy of digital twin and wireless systems.

III. TAXONOMY: T WINS FOR W IRELESS players that act in a complex environment. To classify the
For twin-based wireless systems, both aspects (i.e., “twins areas for study related to both aspects, in Fig. 8, we devise a
for wireless” and “wireless for twins”) involve a variety of taxonomy. The taxonomy is divided into two layers, namely,

Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
the physical interaction layer and the twin layer. The main
aspects related to designing twins for wireless systems are
identified to be: twin isolation, incentive design, twin object
design, twin object prototyping, twin object deployment, phys-
ical end-device design, decoupling, and interface design. A
digital twin-based wireless system will use the shared physical
resources (e.g., edge servers, end-devices) for various services.
Therefore, there must be twin isolation to enable the seamless
operation of twins for various services without interrupting
the performance of each other. Next, twin object design helps
in the on-demand creation of twins using existing computing
hardware for various applications, whereas twin object proto-
typing helps in the creation of virtual models of the wireless
systems. Twin object deployment is about the placement (e.g.,
cloud or edge) of twin objects. Decoupling allows the seamless
operation of the twin-based services with minimum possible
dependency on the underlying hardware. On the other hand,
interfaces are used for various types of communications, such
as those among twin objects, and those between twin objects
and physical devices. To motivate the participation of different
devices in twinning, incentive mechanisms will be required.
A. Design of Twin Objects
Twin objects are instantiated upon request from the end-
users. Therefore, there must be an efficient way to deploy on-
demand twin objects at the edge/cloud. There are two ways to
instantiate the twin objects, such as virtual machine-based twin
objects and container-based twin objects [38], [41]. A virtual
machine can be defined as the architecture that is independent
of the underlying hardware. Virtual machines can be mainly of
two types, such as system virtual machine and process virtual
machine [52]. In the context of a twin-based wireless system,
the system virtual machine can be used to model a complete
IoT service (e.g., AR-based healthcare system), whereas the
process virtual machine can be used to model the particular
portion (e.g., edge caching [53] module for smart infotainment
system) of digital twin-based system. Modeling a complete
system may be easier than modeling a particular part.
Note that the virtual machine is different from the oper-
ating system. In an operating system, language-independent
extensions of hardware are created, whereas a virtual
machine creates a machine-independent instruction set. Virtual
machine-based virtualization can be seen as an infrastructure-
as-a-service (IaaS) that can enable the same hardware via
virtualization for running multiple operating systems (e.g.,
twins). This operational approach has the main drawback of
high management complexity. To address this issue, container-
based twin objects can be used. Containers can help minimize
management complexity by running multiple twins on the
same operating system [54], [55]. However, container-based
twin objects may be more prone to security attacks than the
virtual machine-based design [56], [57].
B. Prototyping of Twin Objects
Prototyping twin objects involves estimating the parame-
ters of virtual twin models for the physical system coun-
terparts. True prototyping of twin objects is one of the
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2239
primary challenges in digital twinning over wireless networks.
Physical objects are characterized by a set of attributes (e.g.,
shape, mass, energy). These attributes are difficult to model
exactly [58]. In experimental modeling, a series of experiments
are performed to find out the various parameters of a physical
wireless system [59]. Various entities of a wireless system
can be modeled using experimental data. For instance, the
work in [59] modeled the free-space path-loss for intelligent
reflecting surfaces using a series of experiments. Alternatively,
mathematical modeling can be used that is based on a math-
ematical representation of a wireless phenomenon. Typical
mathematical models are based on various assumptions that
need to be valid for a real-life scenario so that a more practical
model of the digital twin can be developed.
There are some scenarios which are difficult to model math-
ematically [33]. In these scenarios, data-driven modeling can
be used. Wireless system applications (e.g., XR) can generate
a significant amount of data that can be used in modeling their
behavior [38]. One can use the data to train machine learning
models. However, training of a machine learning model within
a certain period of time for a certain data set may require
a significant amount of computing resource (CPU-cycle/sec).
An increase in computing resources will minimize the training
time but at the cost of an increased energy consumption.
There are two types of machine learning models, namely,
centralized machine and distributed machine learning models.
Centralized machine learning relies on centralized training at
the remote cloud or edge server. With centralized machine
learning, there is an issue of privacy leakage because of trans-
ferring end-devices’ data to the centralized cloud/edge server.
To mitigate this problem, distributed machine learning can be
used, in which case the end-devices iteratively train the global
model using their local data sets. The local model updates are
then sent to the edge/cloud server for aggregation to yield a
global model.
C. Deployment Approaches for Twins
Depending on the application requirements, twins can be
mainly deployed either at the network edge or cloud. Every
application has distinct requirements in terms of latency,
quality of physical experience, computing resource require-
ment, and reliability, among others. Twins deployed at the
network edge can enable services with low latency compared
to those deployed at the cloud. Also, the twins at the edge can
have more context awareness (e.g., locations of end-devices,
mobility-awareness). However, the edge-based twins have lim-
itations in terms of low computing resources. When deployed
in the cloud, there could be more computing resources.
To benefit from both edge-based twins and cloud-based
twins, a hybrid approach can be used in which twins are
deployed at both network edge and cloud. For instance,
consider the digital-twin-enabled infotainment system for
autonomous cars. One can use caching assisted by a hybrid
twin that consists of two twins deployed at both the network
edge and the cloud. The edge twin will make the caching
decisions for infotainment users where latency is stringent.
Meanwhile, due to limited storage capacity at the edge, one
2240
C OMPARISON A MONG T WIN O BJECT D EPLOYMENT M ETHODS [5]
Fig. 9. Deployment trends for twin objects [5].
can use the cloud to cache the information that is needed
less frequently compared to cached information at the edge.
To control caching in the cloud, cloud twins can be used.
Comparisons of various twins depending on the deployment
trends are given in Table III [5]. Fig. 9 shows the differ-
ent ways in which twins can be deployed. Table III shows
that edge-based twins (i.e., for multiple twin objects) have the
highest robustness to failures compared to cloud-based twins
and hybrid twins. The is due to the fact that even if any of
the twin objects malfunctions due to physical damage or a
security attack, the other twin objects will serve the end-user.
Similarly, the elasticity of edge-based twins is higher than a
cloud-based twin. Also, mobility support for users and con-
text awareness could be higher for edge-based twins due to
the proximity of the end-devices.
D. Design of Physical Devices
To efficiently operate twin-based wireless systems, there is
a need to effectively design the end-devices (e.g., IoT sensors,
edge/cloud servers). These end-devices will be responsible
for taking actions (e.g., optimal transmit power, association,
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
TABLE III
and resource allocation) and/or training local models for dis-
tributed learning [5]. For distributed learning case, the devices
will require enough high computing power to learn the model
parameters using their local data. Training a twin model using
a set of devices depends on local learning model architecture,
local data set, and local iterations [60]. The energy consump-
tion at the devices will depend on their design [61]. The design
of end-devices can be mainly based on either hardware-based
design or software-based design [62].
A programmable hardware-based design is preferable which
is able to handle data with high dimensionality for various
tasks in a digital twin-enabled wireless system [5]. For the
hardware design, application-specific many core processors
can be used [63]. Kim et al. in [64] presented a machine
learning-based design for many core systems. On the other
hand, with a fixed hardware design, multiple software designs
can be used that can efficiently leverage the given hardware for
performing distributed learning tasks. To do so, neural archi-
tecture search (NAS) [65], [66], [67], [68] can be used. In
NAS, various possible neural architectures are checked for
certain hardware and the efficient one is selected.
E. Design of Interfaces
In a digital twin-based system, there will be many interfaces,
such as user-to-twin system interface, twin-to-object interface,
and twin object-to-twin object interface. Detailed discussions
about air interfaces that can be used for multiple purposes
will be given in Section IV-C. Here, we discuss the twin-
to-physical objects interface, user-to-twin system interface,
and twin object-to-twin object interface. The twin to physi-
cal object interface can be either wireless or wired. Similarly,
twin-to-twin interface can also be wired and wireless. For dis-
tributed twins deployed at multiple edge servers [5], there
is a need to enable wireless communication among them.
Another way of communication can be through the core
network [38]. On the other hand, there is a need for an effi-
cient and easy-to-use interface for communication between
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES 2241

Fig. 10. Overview of incentive mechanisms for digital twin-based wireless systems.

end-users and the system. The interface can be of various
types, such as voice interface (e.g., Amazon Echo), touch
interface (e.g., smartphone), and physical button interface [69].
These interfaces enable users to interact with the twin-based
system. For instance, smartphones use iOS, Samsung One UI,
OxygenOS, Android One, and Indus OS [70]. For a smart-
phone, one can use any of the aforementioned interfaces of
operating systems. However, for applications based on dig-
ital twins, new interfaces may be required because of their
different architecture from the existing ones.
F. Incentive Mechanism
To motivate the participation of a variety of players in
enabling digital twin-based wireless systems, there must be
attractive incentive mechanisms. Here, we consider the high-
level architecture of digital twin-enabled wireless systems
(Fig. 7) presented in Section II-C. The different types of incen-
tives required in digital twin-enabled wireless systems are
presented in Fig. 10. As described in Section II-C, a digital
twin-based wireless system can use distributed learning to train
twin models. These pre-trained twin models are stored using a
blockchain network in an immutable and transparent manner,
and can be used in the future to serve the requesting users.
Upon request from end-users, the twin-based system serves
the end-users by controlling the network entities for efficient
management of resources. To do so, there must be incentives
for various entities. Incentive mechanism design for digital
twin-based wireless systems will include the following cases:
incentive mechanism design for pre-training of twin models,
incentive mechanism design for blockchain mining, and incen-
tive mechanism design for SDN switches operators, as shown
in Fig. 10. In these cases, the players involved are end-devices,
aggregation servers, edge/cloud servers running twin objects,
and blockchain miners used for storing pre-trained models and
training data.
When using distributed learning for digital twins, end-
devices participate in learning local models [74]. Moreover,
edge/cloud servers run twin objects and also assist (i.e., aggre-
gate local models) distributed learning. Therefore, there is a
need to give end-devices rewards for performing their jobs.
For distributed learning-based twins, the edge/cloud servers
will interact with the end-devices for learning a global model.
End-devices perform model updates locally and the edge/cloud
server performs aggregation of the end-devices’ local model
updates to yield a global model. The global model is then

Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2242 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
TABLE IV
I NCENTIVE M ECHANISMS FOR F EDERATED L EARNING [62]
shared with the end-devices. Such an iterative process takes
place until convergence is achieved. One can design an incen-
tive mechanism for such a scenario by defining a utility at
the edge/cloud server. The utility of edge/cloud servers can
be to maximize the global model accuracy. Meanwhile, the
end-devices (i.e., used for running the local learning model)
with high local accuracy can be given more monetary rewards.
Various works proposed incentive mechanism design for dis-
tributed learning using Stackelberg game, auction theory, and
contract theory [62]. An overview of these incentive mech-
anisms is given in Table IV. Additionally, details about the
incentive model approach, motivation, device utilities, and
aggregation server utilities are given.
In digital twin-based architecture, pre-trained twin models
must be stored using blockchain for future use [5]. Miners
used for performing blockchain consensus algorithms and stor-
age must be given attractive incentives [75]. In [75], a Fee
and Waiting Tax scheme was proposed to provide miners with
incentives for mining and storage of the generated blockchain
blocks. With the existing blockchain protocols, selecting a
transaction by a certain miner will impose storage cost on
other miners and a user’s high rate of transaction generation
may increase the waiting time for other users. The Fee and
Waiting Tax scheme offers a set of fee-per-byte choices for
users while ensuring the miners get sufficient fees for their
work. Also, it charges a waiting tax on users proportional to
their negative impact on other users. This incentivizes the users
to become more conservative while generating transactions.
G. Isolation of Twins
Twin isolation deals with the operation of a twin-
based application (e.g., extended reality) without affect-
ing the performance of other twin-based applications (e.g.,
brain-computer interaction). Note that twinning for various
applications can be performed simultaneously using shared
physical resources. Therefore, the twins for different applica-
tions should be designed (i.e., isolated) such that they meet the
quality of service without interrupting the operation of each
other. The operation of various twins for different applications
requires computing, storage, and communication resources.
The computing resources are used for performing blockchain
mining as mentioned in architecture (for more details, please
refer to Fig. 7) in Section II-C. Meanwhile, there is a need
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
for storage to store pre-trained twin models for various appli-
cations. Additionally, sufficient communication resources are
required for the signaling of twin instructions. There can be
mainly two ways of enabling twins: using dedicated hardware,
and using shared hardware. Isolated operation of twin-based
applications is required for shared hardware-based operations.
Isolation in digital twin-enabled wireless systems can be
divided into twin object isolation, core network isolation, and
access network isolation [81], [82]. Radio access network
isolation refers to using the same access network for vari-
ous twins without affecting the performance of other twins.
Access network isolation can be achieved using radio resource
virtualization (i.e., slicing) [83], [84], [85]. In resource vir-
tualization, a mobile virtual network operator (MVNO) will
buy radio resources from multiple network operators and
then sell the resources to various twins based on their
requirements. A twin can also buy resources from multiple
MVNOs. The interactions between the MVNOs and the
twins can be modeled, analyzed, and optimized based on
heuristic schemes [76], deep reinforcement learning-based
schemes [77], game-theoretic schemes including matching
game-based schemes [78], [79], [80].
Similar to radio access network isolation, in theory, core
network isolation can also be achieved through service
differentiation or end-to-end resource reservation schemes.
However, with the current IP technology, it would be more
challenging to achieve efficient isolation. On the other hand,
twin object isolation can be performed using shared computing
hardware (e.g., cloud) for multiple virtual machines running
twin objects. Isolation of such twin objects is necessary for
improving performance and reducing the chance of security
attacks [86], [87]. For instance, if the twin objects are not
well isolated, then the security attack instantiated within a twin
object can easily affect other twin objects.
H. Decoupling
Twinning of wireless networks for various applications
requires successful and efficient interaction between the twin
layer and physical interaction layer. For such an efficient
interaction, decoupling is necessary. Decoupling in digital
twin-enabled wireless systems can be of two types, such as
function decoupling and data decoupling [5]. Data decoupling
can be performed using data homogenization techniques. This
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
TABLE V
V IRTUALIZATION S CHEMES FOR R ADIO ACCESS N ETWORKS
will enable the twin-based system design to be independent of
the underlying network smart devices. The layered architec-
ture of twin-based wireless system is discussed in Section II-C.
Fig. 7 shows that there will be a variety of smart devices (e.g.,
temperature sensors, image sensors) at the physical interaction
layer. To transform this data into homogenized data, an effi-
cient data homogenization technique will need to be used. On
the other hand, functions decoupling enables the twin-based
system to decouple the management functions from the physi-
cal interaction layer to the twin layer. Function decoupling can
be implemented by network slicing that is based on software-
defined networking (SDN) and network function virtualization
(NFV) [88], [89], [90], [91], [92], [93]. NFV is based on
implementing various network functions using generic hard-
ware. To implement the physical devices using the NFV-based
approach, there may be few limitations, such as through-
put and latency deterioration [91]. The per-instance capacity
of the NFV-based device may not be similar to the actual
physical device. Other challenges faced by NFV-based design
are reliability and security issues. Multiple NFVs for various
network functions running on the same hardware might result
in failure due to a physical damage/ security attack [94]. To
address this issue, the NFV can be migrated to a new device.
However, migration may introduce a delay that is undesirable
in many IoE/IoT applications [95], [96], [97]. Additionally,
mobility of the end-devices also requires the migration of
NFV-based functions to the new network edge. To enable
such migration, we can use machine learning-based mobility
prediction schemes to proactively determine the next loca-
tion [38]. On the other hand, security must be ensured for
the NFV-based functions running on the cloud. In a typical
SDN control plane, a single centralized controller is used
to control multiple switches for performing various func-
tions (e.g., routing) [98], [99], [100], [101]. These switches
intended to control wireless system devices can be installed on
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2243
dedicated hardware for better performance. Controlling SDN-
based devices using a single centralized controller may suffer
from low scalability and low reliability. An increase in the
number of devices will increase the signaling between SDN-
based devices and the centralized controller. To address this
limitation, a distributed control plane, that consists of multiple
SDN controllers [102], [103] can be used. However, using
multiple controllers will lead to an increase in control plane
management complexity. To reap the advantages of both cen-
tralized control and distributed control planes, a hybrid control
plane approach can be used, which will use both centralized
and distributed controllers [104], [105]. In the hybrid control
plane, the main controller controls the local controllers. Each
local controller controls its small set of devices. All the func-
tions that are local to the devices are handled by the local
controller, whereas the functions that require global network
knowledge are controlled by the main controller. An overview
of the centralized, distributed, and hybrid control planes is
given in Fig. 11. Meanwhile, a tabular comparison is given in
Table VI.
I. Lessons Learned and Recommendations
We have devised a taxonomy of “twins for wireless”, which
considers twin isolation, incentive design, twin object design,
twin object prototyping, twin object deployment options, phys-
ical design of end-devices, decoupling, and interface design.
The lessons learned are as follows.
• There is a need to consider hardware and software
co-design for the end-devices. For example, hardware
designs for different neural network architectures can be
considered along with the learning algorithm.
• Efficient deployment of twin objects (at the network edge
and/or cloud) is critical. Analysis and optimization of
2244
Fig. 11. Twin function decoupling using (a) centralized controller, (b) distributed controllers, and (c) hybrid controllers.
TABLE VI
C OMPARISON A MONG T WIN F UNCTION D ECOUPLING M ETHODS
cost and performance tradeoffs for different deployment
options is an important research problem.
• Considerable care must be taken while virtual modeling
of the physical system/entity. Mathematical models are
based on assumptions that may not accurately follow the
real scenarios. On the other hand, experimental modeling
can be used which is based on extensive experimentation.
However, this approach can result in a long design time.
Another approach would be data driven modeling that is
based on training a machine learning model.
• Design of the devices in the physical interaction space
needs to be cost-efficient for which hardware-software
co-design approach can be used.
IV. TAXONOMY: W IRELESS FOR T WINS
Under “wireless for twins”, we deal with efficient twin
signaling for enabling various services in twin-based wire-
less systems. The related important considerations are air
interface design, accessing twin objects, and security and pri-
vacy (Fig. 8). The air interface will be used for twin signaling.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
Accessing the twin objects relates to association between twins
and physical devices. Finally, there must be effective security
and privacy mechanism for twin-based wireless systems.
A. Accessing Twin Objects
There are two main aspects related to accessing the twin
objects, namely, (i) association between twin objects and
physical devices and (ii) allocation of computing and com-
munication resources. Twin objects can be deployed either at
the cloud or at the edge. The association problem between
edge-based twins and physical objects will be a combinatorial
problem. This is due to the presence of binary variables (e.g.,
xn,m ) that represent the association between physical objects
and edge servers (e.g., xn,m, = 1 if the physical object n
is associated with server m and vice versa). Such a problem
can be solved by using various techniques, such as relaxation-
based schemes, heuristic schemes, and matching theory-based
schemes [28], [106], [107], [108], [109], [110], [111], [112].
The relaxation-based solution first transforms the binary asso-
ciation variable into a continuous variable which can be further
solved using various schemes (e.g., convex optimization solver
if the relaxed problem is convex). A relaxation-based solu-
tion can provide a low complexity solution but at the cost
of approximation error (i.e., transforming the binary variables
into continuous variables for getting a solution and then con-
tinuous variables into binary variables after solution). To avoid
the approximation error in relaxation-based schemes, one can
use heuristic schemes [28]. Matching theory-based solutions
can offer effective association between twins and the physical
objects with reasonable complexity [111] and [112].
Wireless resources need to be allocated to physical devices
for communication with twin objects deployed at the network
edge or cloud. Communication resources are required for
signaling and training of distributed twin models. For dis-
tributed training, wireless resources are required for sharing
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
Fig. 12. Security attacks in digital twin-based wireless systems.
the learning model updates between the end-devices and
the aggregation server. Wireless resource allocation prob-
lems can be solved by using heuristic algorithms [28],
relaxation-based schemes [113], and matching theory-based
schemes [114], [115], [116]. For twin signaling, wireless
resources are required for the transfer of control signals (e.g.,
signaling to/from SDN controllers). Also, blockchain min-
ers will use wireless resources for sharing blocks during the
consensus algorithm. Additionally, computing resources are
required for twin processing tasks, e.g., for running machine
learning models at edge/cloud, blockchain consensus algo-
rithms, and analysis of the virtual twin model prior to applying
for real-time applications.
B. Security and Privacy
An overview of various security attacks along with their
possible solutions for digital twin-based wireless systems is
given in Fig. 12. The security attacks in digital twins-based
wireless systems can be categorized into two types, namely,
physical device security [38] and interface security. Effective
authentication schemes are required to avoid unauthorized
access to devices, edge/cloud servers, and blockchains. In an
SDN-based architecture, the security threats include vulner-
ability of network controllers, forged control packet injec-
tion, misconfigured policy enforcement, and weak network
devices authentication [117]. Misconfigured interfaces (e.g.,
twin-to-twin interface) and protocols (twin packet routing pro-
tocols) result in various security vulnerabilities [118]. Weak or
improper authentication schemes and plain text channels may
lead to security attacks.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2245
In a centralized learning scenario, there may be a loss of
privacy during training of the twin models. To mitigate this
problem, twin models can be trained in a distributed man-
ner, which is based on-device training. Although end-devices
do not send their data to the remote cloud/edge server for
training, they still require privacy preservation techniques. A
malicious aggregation server can infer the end-devices’ sensi-
tive information using their local learning model updates, and
thus result in privacy leakage [62], [74], [119]. To address this
issue, schemes based on differential privacy and homomorphic
encryption can be used. In differential privacy, noise is added
to the local learning model updates prior to sending them to the
aggregation server. Although differential privacy can enhance
privacy preservation, it can slow down the convergence rate of
global learning [60]. To avoid this, homomorphic encryption
can be used, which is based on encrypting the local learning
models prior to sending them to the aggregation server. Similar
to differential privacy, homomorphic encryption works at the
cost of communication overhead [74]. Other than differential
privacy and homomorphic encryption, few works [120], [121],
[122] proposed over-the-air computation. In over-the-air com-
putation, the channel noise is considered a differential privacy
noise for preserving the distributed learning privacy. On the
other hand, the work in [74] proposed a dispersed federated
learning scheme that is based on computing sub-global models
within groups, as shown in Fig. 13. The computation of the
sub-global model is performed iteratively. Then, the sub-global
models are shared among each other, and finally, the global
model is computed. Note that dispersed federated learning
offers enhanced privacy protection compared to traditional fed-
erated learning. In traditional federated learning, a malicious
2246
Fig. 13. Dispersed federated learning.
Fig. 14. Collaborative federated learning.
aggregation server can infer some of the devices’ sensitive
information from their learning model updates. Dispersed
federated learning can offer better privacy preservation than
traditional federated learning. Another work [123] presented
a collaborative federated learning scheme that is based on
local aggregation at end-devices, as shown in Fig. 14. Such
local aggregation is performed due to communication resource
constraints. The global aggregation can infer the devices’ sen-
sitive information from the learning model updates (excluding
locally aggregated learning models). However, it is very dif-
ficult for the aggregation server to infer the devices’ local
information from the locally aggregated learning models.
Therefore, collaborative federated learning can offer bet-
ter privacy preservation compared to traditional federated
learning [124].
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
C. Air Interface Design
The air interface will be used for communication between
the digital twin and the physical objects (e.g., sensors) and
wireless users. There are two main aspects related to the
design of wireless interface for digital twinning, namely, the
frequency band for wireless access and the access scheme.
The wireless access can be based on orthogonal frequency-
division multiple access (OFDMA), time-division multiple
access (TDMA), code-division multiple access (CDMA), non-
orthogonal multiple access (NOMA) methods [125]. OFDMA
uses orthogonal resource blocks for communication, which
may result in spectral inefficiency. With NOMA, the entire
available bandwidth can be used by all users simultaneously
and decoding at the receiver is performed based on received
power levels. It can be more spectrum-efficient, but at the
cost of more complexity at the receiver. Energy efficiency is
another important consideration in the air interface design. A
comparison among NOMA, OFDMA, CDMA, and TDMA is
given in Table VII [126].
A large transmission bandwidth (and hence transmission
rate) will result in high throughput, and thus low latency
communication. Higher frequency bands such as millimeter-
wave and terahertz bands can offer large transmission band-
width. However, transmissions in these bands suffer high
attenuation. To mitigate this issue, reconfigurable intelligent
surfaces (RIS) [127] can be used. An RIS consists of an
array of antenna elements which can change either polar-
ization, frequency, amplitude or phase of the impinging sig-
nals [127], [128] to enable efficient communication between
a transmitter and a receiver that do not have a line-of-sight
path. The challenges in RIS-aided communications include
channel estimation [129], [130], [131], joint active and passive
beamforming, among others [132]. In [129], the authors con-
sidered channel estimation for an RIS-enabled millimeter-wave
communication system and proposed a compressed sensing-
based estimation scheme. The authors also extended their work
to a multi-antenna system. The schemes proposed in [129]
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
TABLE VII
C OMPARISON A MONG NOMA, OFDMA, CDMA, AND TDMA [126]
and [130] may not perform very well in a dynamic propaga-
tion scenario and they require retraining. To address this issue,
the work in [131] proposed a hybrid passive/active RIS archi-
tecture and a deep denoising neural network-based compressed
sensing broadband channel estimation method. The focus of
the authors was on a millimeter wave system and minimized
training overhead.
D. Lessons Learned and Recommendations
A taxonomy of “wireless for twins” has been devised. We
have considered the aspects related to accessing the twin
objects, air interface design, and security and privacy. The
lessons learned are as follows.
• Effective security schemes must be used for digital twin-
based wireless systems. Prior to developing the security
schemes, forensic schemes are required to investigate the
attacks in digital twin-based systems. Lightweight and
effective authentication schemes are required for devices.
For wireless transfer of data and control signals, effi-
cient and effective encryption schemes need to be used.
Twin-based wireless systems have a layered architec-
ture that communicate with each other using interfaces
(e.g., twin object-to-twin object interface and twin object-
to-physical object interface). There may be configured
interfaces for twinning which are vulnerable to security
attacks. Additionally, forged twinning control instruc-
tions will significantly degrade the performance of digital
twin-based systems.
• The computing resource at the network edge/cloud need
to be efficiently managed while deploying multiple twin
objects for various services. Different twins (e.g., XR
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2247
TABLE VIII
S UMMARY OF O PEN R ESEARCH C HALLENGES IN
E XISTING S URVEYS AND T UTORIALS
twin, healthcare twin) have different computing resource
requirements, therefore, deploying twins with heteroge-
neous requirements needs careful design considerations.
Computing resources need to be allocated effectively
among various twins. Additionally, there may be priori-
ties among twins. For instance, healthcare twins might
have more priority compared to infotainment twins.
Therefore, computing resource allocation will need to
take into account the priority of various twins.
• Energy-efficient algorithms are required for twinning over
wireless networks. A digital-twin-based system has a
wide variety of players (i.e., end-devices, edge/cloud
servers, SDN switches). Association of physical end-
devices with the edge server running twin objects should
also consider energy-efficiency of devices and servers.
V. O PEN C HALLENGES
Previous surveys and tutorials have discussed open chal-
lenges, such as standardization issues, security and privacy,
government regulations for medical twins, accurate repre-
sentation of digital twins, technical limitations, barriers to
blockchain applications in digital twins, and data related
issues. We list some novel challenges as in Table VIII.
A. Dynamic Twins
How to reuse twins for controlling various physical devices?
Re-usability of twin objects is one of the main features that can
make digital twins promising for wireless systems. Designing
twin objects requires significant efforts to make them an exact
replication of physical object/ phenomenon with energy- and
computational efficiency. Therefore, it will be highly desir-
able to make the twins reusable for various services. For
2248
this, twin machine learning models can be trained using a
more general data set (e.g., ImageNet data set) to enable its
applicability to different services/applications (e.g., classifica-
tion task for mini-ImageNet data set). Also, the concept of
meta-learning can be used to train them efficiently. Such a
general dynamic twin model can be based on either central-
ized machine learning or distributed machine learning (e.g.,
federated learning). However, training a twin machine learn-
ing model for general data might not perform well for some
scenarios [133], [134].
B. Interoperability for Migration of Twin Objects
How to enable a seamless operation of end-devices served
by edge-based twin objects? Mostly, end-devices in wire-
less systems are mobile. For instance, one device connected
to a small cell base station (equipped with an edge server
running twin objects) may move to the coverage area of
another base station. There can be two ways to serve the
mobile device. One can be to connect to the existing edge
server through the core network via the newly associated
base station. However, this approach will suffer from latency
that might not be desirable due to strict latency constraints
of some of the IoE applications. To address this issue, the
twin object can be migrated to the newly associated small
cell base station. Twin objects based on virtual machines
can be migrated dynamically depending on the location of
mobile devices [135]. Note that, transferring a virtual machine
from one edge/cloud server to another may face interoperabil-
ity issues. Two edge/cloud servers must be interoperable to
ease migration of virtual machines running twin objects to
tackle the mobility of end-devices. To enable easier migra-
tion, a common architecture can be used such as unified
cloud interface/cloud broker, enterprise cloud Orchestration
platform/orchestration layer, and open cloud computing
interface [136].
C. Prototyping of Physical Objects
How to prototype the physical object attributes (e.g., fea-
tures, data, actions, and events) into twin objects for various
applications? For twin modeling, it is necessary to estimate
the measurable aspects of the physical objects. However, it
is challenging to accurately measure the aspects of a physi-
cal system. During modeling of a physical object, to reduce
the complexity, we can focus on a limited set of parame-
ters. Various modeling schemes are experimental modeling,
three-dimensional modeling, and data-driven modeling [31].
Experimental modeling involves full-scale experimentation of
a physical phenomenon. In three-dimensional modeling, the
goal is to develop mathematical models of physical objects
using various techniques (e.g., 3D scanning). However, exact
representation using a mathematical model is challenging. On
the other hand, data-driven modeling uses data for deriving the
functional form of physical objects. Machine learning can be
used to model physical objects using data. However, proper
selection and training of a machine learning model can be
challenging.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
D. Incentive Mechanisms for Twinning
How to motivate various players of a twin-based wireless
system for successful and effective operation? In digital twin-
ning, a variety of players, such as edge/cloud servers, miners,
end-devices, and network operators, interact with each other to
enable wireless services. These players interact to perform var-
ious tasks, such as pretraining of twin models, twin operation,
and mining for management of twin pre-trained models. For
pretraining of twin models using distributed machine learn-
ing, there will be two main players, namely, end-devices and
edge/cloud servers. End-devices compute their local model and
expect incentives for their contributions toward learning of a
global model. To model and design an incentive for such a
scenario, for example, a Stackelberg game model with a sin-
gle leader and multiple followers can be used. The edge/cloud
server will act as a leader and devices will follow it. The
utility of the edge/cloud server can be a function of global
accuracy of the machine learning model and the end-devices’
utility can be the difference between cost (i.e., communica-
tion and computation) and reward [71]. Furthermore, contract
theory and auction theory can be used to design incentives for
pretraining of twin models [72], [73].
E. Twinning Forensics and Security
How to investigate and take measures to counter the secu-
rity attacks in a digital twin-enabled wireless system? Digital
twin-based wireless systems will involve a wide variety of
entities, such as edge/cloud computing servers, end-devices,
various interfaces (e.g., twin to device and twin to twin), and
twin objects, among others. They are prone to various kinds of
new attacks (e.g., twin object attacks and twin to twin interface
attacks) in addition to existing attacks (e.g., a man in the mid-
dle attack). Therefore, effective forensic schemes need to be
used to enable the successful operation of twin-based wireless
systems. Forensic techniques for a digital twin-based system
can be blockchain-based forensics schemes [137] and video-
based evidence analysis [138]. On the other hand, there may
be challenges in implementing forensics schemes due to the
mobility of nodes. To address this issue, mobility-aware foren-
sics schemes can be used. Security schemes for the edge/cloud
servers running the twin objects could be different from those
(e.g., encryption/decryption schemes) used for twin signaling.
F. Efficient Chaining of Twin Objects
How to chain various twins to enable efficient operation of
wireless system functions? Generally, designing a twin object
requires extensive efforts and time. For instance, twin-based
AR service may require multiple twin objects deployed at
network edge. It may be more efficient to reuse the existing
twins to enable a complete twin-based AR service. However,
combining multiple twins for enabling a service may incur
communication and computing cost. Similar twin objects run-
ning at edge and cloud server can have different latencies. The
available computing power can also be different depending on
the location of twin objects. These factors need to be taken
into consideration while chaining twin objects for a certain
service/function.
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES 2249

TABLE IX
S UMMARY OF THE O PEN R ESEARCH C HALLENGES AND S OLUTION G UIDELINES

G. Energy-Efficient Twinning edge/cloud servers and end-devices decreases with increasing

How to enable energy-efficient twinning over wireless
networks without compromising QoS? Twinning over a wire-
less network involves computation tasks at the end-devices
and at the edge/cloud, and transmission of data (i.e., learn-
ing updates and control signals) between end-devices and
edge/cloud servers running twin objects. Energy consump-
tion of the edge/cloud servers and the end-devices mainly
depends on their operating frequencies. The computing time of
operating frequencies. However, energy consumption increases
with increasing operating frequency [119]. On the other hand,
energy-efficient transmission of data between twin objects
deployed at edge/cloud and the physical devices depends
on wireless propagation characteristics, resource allocation,
object association. Resource allocation and association can
be efficiently performed based on matching theory [139].
Power allocation can be performed based on block successive

Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2250 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
upper bound minimization (BSUM) or deep reinforcement
learning [74], [140], [141].
H. Hardware-Software Co-Design of Devices
For a digital twin of a wireless system, how to efficiently
design devices of physical interaction space? Designing of
physical interaction space has two main aspects, namely, hard-
ware design and software design. For improved performance,
there is a need to consider hardware-software co-design. Such
a design can be obtained by using methods such as vir-
tual prototyping, co-verification-based embedded systems, and
high-level synthesis [38]. A virtual prototyping-based design
uses computing automated design, computer-aided design, and
computer-aided engineering before implementing the devices.
A high-level synthesis design uses algorithmic description to
create automated digital hardware. With co-verification-based
design, testing and debugging of both hardware and software
are performed at the same time during the design.
VI. C ONCLUSION
We have presented a comprehensive tutorial on digital twins
and wireless systems. We have presented the key design
aspects and a high-level framework for digital-twin-based
wireless systems. Also, we have outlined currently available
digital twinning frameworks. Additionally, a comprehensive
taxonomy has been devised using various parameters. Finally,
we have presented several open research challenges and pos-
sible solutions. The integration of digital twins and wireless
systems will enable advanced IoE applications. Digital twin-
ning will enable wireless systems to proactively manage
the network resources for strict latency applications (e.g.,
XR). Furthermore, the reusability of generalized digital twins
will make them attractive for use in many new emerging
applications.
Digital twins are expected to be one of the most promising
technology for 6G and beyond wireless systems. 6G services
will have diverse requirements for novel applications (e.g.,
human-computer interaction, XR). These requirements include
user-defined quality of physical experiences, extremely low
latency, and ultra-reliability, among others. Enabling a wireless
system to meet the aforementioned requirements needs proac-
tive analysis and machine learning-based schemes. The digital
twin will enable us to proactively analyze the system and train
effective machine learning models. The pre-trained machine
learning models at the digital twin will enable the wireless
system to make on-demand decisions for different wireless
services and applications. Furthermore, the pre-trained twin
models can be further trained to more effectively incorporate
the newly generated data.
R EFERENCES
[1] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and
M. Ayyash, “Internet of Things: A survey on enabling technologies,
protocols, and applications,” IEEE Commun. Surveys Tuts., vol. 17,
no. 4, pp. 2347–2376, 4th Quart., 2015.
[2] L. Atzori, A. Iera, and G. Morabito, “The Internet of Things: A survey,”
Comput. Netw., vol. 54, no. 15, pp. 2787–2805, Oct. 2010.
[3] S. Li, L. Da Xu, and S. Zhao, “5G Internet of Things: A survey,” J.
Ind. Inf. Integr., vol. 10, pp. 1–9, Jun. 2018.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
[4] M. Ge, H. Bangui, and B. Buhnova, “Big data for Internet of Things:
A survey,” Future Gener. Comput. Syst., vol. 87, pp. 601–614, Oct.
2018.
[5] L. U. Khan, W. Saad, D. Niyato, Z. Han, and C. S. Hong, “Digital-
twin-enabled 6G: Vision, architectural trends, and future directions,”
IEEE Commun. Mag., vol. 60, no. 1, pp. 74–80, Jan. 2022.
[6] F. Tao, H. Zhang, A. Liu, and A. Y. C. Nee, “Digital twin in indus-
try: State-of-the-art,” IEEE Trans. Ind. Informat., vol. 15, no. 4,
pp. 2405–2415, Apr. 2019.
[7] S. Haag and R. Anderl, “Digital twin—Proof of concept,” Manuf. Lett.,
vol. 15, pp. 64–66, Jan. 2018.
[8] “Internet of Things Market.” Markets and Markets. Sep. 2021.
[Online]. Available: https://www.marketsandmarkets.com/Market-
Reports/internet-of-things-market-573.html
[9] “Spending on Internet of Things (IoT) in Selected Countries of
the Asia-Pacific Region in 2019.” Statistica. Sep. 2021. [Online].
Available: https://www.statista.com/statistics/1037118/apac-internet-of-
things-spending-by-country/
[10] “IoT Spending Worldwide 2019, by Country.” Sep. 2021. [Online].
Available: https://www.statista.com/statistics/1118256/iot-spending-
worldwide-by-country/
[11] “Digital TWin Market.” Markets and Markets. Sep. 2021.
[Online]. Available: https://www.marketsandmarkets.com/Market-
Reports/digital-twin-market-225269522.html?
[12] “Global Digital Twin Market.” Sep. 2021. [Online]. Available:
https://www.marketsandmarkets.com/Market-Reports/digital-twin-
market-225269522.html?
[13] R. Minerva, G. M. Lee, and N. Crespi, “Digital twin in the IoT context:
A survey on technical features, scenarios, and architectural models,”
Proc. IEEE, vol. 108, no. 10, pp. 1785–1824, Oct. 2020.
[14] Y. Wu, K. Zhang, and Y. Zhang, “Digital twin networks: A survey,”
IEEE Internet Things J., vol. 8, no. 18, pp. 13789–13804, Sep. 2021.
[15] B. R. Barricelli, E. Casiraghi, and D. Fogli, “A survey on digital twin:
Definitions, characteristics, applications, and design implications,”
IEEE Access, vol. 7, pp. 167653–167671, 2019.
[16] I. Yaqoob, K. Salah, M. Uddin, R. Jayaraman, M. Omar, and M. Imran,
“Blockchain for digital twins: Recent advances and future research
challenges,” IEEE Netw., vol. 34, no. 5, pp. 290–298, Sep./Oct. 2020.
[17] S. Suhail et al., “Blockchain-based digital twins: Research trends,
issues, and future challenges,” ACM Comput. Surveys, to be published.
[18] “Digital Twins are Mission Critical.” [Online]. Available: https://www.
ge.com/digital/applications/digital-twin (Accessed: May 16, 2021).
[19] “What is Digital Twin?” 2021. [Online]. Available: https://www.twi-
global.com/technical-knowledge/faqs/what-is-digital-twin
[20] “Compare Simulation & CAE Software.” 2021. [Online]. Available:
https://www.g2.com/categories/simulation-cae
[21] “Best Digital Twin Software.” Jun. 2021. [Online]. Available:
https://www.g2.com/categories/digital-twin
[22] “The History and Creation of the Digital Twin Concept.” Jun.
2021. [Online]. Available: https://www.challenge.org/insights/digital-
twin-history/
[23] “Digital Twins.” Siemens. May 2021. [Online]. Available: https://www.
plm.automation.siemens.com/global/en/our-story/glossary/digital-twin/
24465#
[24] “Digital Twins: The Ultimate Guide.” XMPRO. May 2021. [Online].
Available: https://xmpro.com/digital-twins-the-ultimate-guide//
[25] “How to Make a Digital Twin: The Options, Types and Outputs.”
Vercator. 2021. [Online]. Available: https://info.vercator.com/blog/how-
to-make-a-digital-twin-the-options-types-and-outputs
[26] “The Digital Twin—An Introduction.” Vercator. 2021. [Online].
Available: https://www.tributech.io/blog/the-digital-twin
[27] M. Chen, Z. Yang, W. Saad, C. Yin, H. V. Poor, and S. Cui, “A joint
learning and communications framework for federated learning over
wireless networks,” IEEE Trans. Wireless Commun., vol. 20, no. 1,
pp. 269–283, Jan. 2021.
[28] L. U. Khan, M. Alsenwi, Z. Han, and C. S. Hong, “Self organizing
federated learning over wireless networks: A socially aware clustering
approach,” in Proc. Int. Conf. Inf. Netw. (ICOIN), 2020, pp. 453–458.
[29] Z. Han, D. Niyato, W. Saad, T. Başar, and A. Hjørungnes, Game
Theory in Wireless and Communication Networks: Theory, Models, and
Applications. Cambridge, U.K.: Cambridge Univ. Press, 2012.
[30] F. Pires, A. Cachada, J. Barbosa, A. P. Moreira, and P. Leitão, “Digital
twin in industry 4.0: Technologies, applications and challenges,” in
Proc. IEEE 17th Int. Conf. Ind. Inform. (INDIN), vol. 1, Jan. 2019,
pp. 721–726.
[31] A. Rasheed, O. San, and T. Kvamsdal, “Digital twin: Values, challenges
and enablers,” 2019, arXiv:1910.01719.
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
[32] H. Elayan, M. Aloqaily, and M. Guizani, “Digital twin for intelligent
context-aware IoT Healthcare systems,” IEEE Internet Things J., vol. 8,
no. 23, pp. 16749–16757, Dec. 2021.
[33] S. Ali et al., “6G white paper on machine learning in wireless
communication networks,” 2020, arXiv:2004.13875.
[34] D. P. Kumar, T. Amgoth, and C. S. R. Annavarapu, “Machine learning
algorithms for wireless sensor networks: A survey,” Inf. Fusion, vol. 49,
pp. 1–25, Sep. 2019.
[35] J. Jagannath, N. Polosky, A. Jagannath, F. Restuccia, and T. Melodia,
“Machine learning for wireless communications in the Internet of
Things: A comprehensive survey,” Ad Hoc Netw., vol. 93, Oct. 2019,
Art. no. 101913.
[36] M. S. Mahdavinejad, M. Rezvan, M. Barekatain, P. Adibi, P. Barnaghi,
and A. P. Sheth, “Machine learning for Internet of Things data analysis:
A survey,” Digit. Commun. Netw., vol. 4, no. 3, pp. 161–175, Aug.
2018.
[37] B. Sudharsan and P. Patel, “Machine learning meets Internet of Things:
From theory to practice,” in Proc. Eur. Conf. Mach. Learn. Principles
Pract. Knowl. Disc. Databases, 2021, pp. 1–3.
[38] L. U. Khan, I. Yaqoob, N. H. Tran, S. A. Kazmi, T. N. Dang, and
C. S. Hong, “Edge-computing-enabled smart cities: A comprehensive
survey,” IEEE Internet Things J., vol. 7, no. 10, pp. 10200–10232, Oct.
2020.
[39] N. Fernando, S. W. Loke, and W. Rahayu, “Mobile cloud computing:
A survey,” Future Gener. Comput. Syst., vol. 29, no. 1, pp. 84–106,
Jan. 2013.
[40] E. F. Coutinho, F. R. de Carvalho Sousa, P. A. L. Rego, D. G. Gomes,
and J. N. de Souza, “Elasticity in cloud computing: A survey,” Ann.
Telecommun., vol. 70, no. 7, pp. 289–309, 2015.
[41] W. Z. Khan, E. Ahmed, S. Hakak, I. Yaqoob, and A. Ahmed,
“Edge computing: A survey,” Future Gener. Comput. Syst., vol. 97,
pp. 219–235, Aug. 2019.
[42] N. Abbas, Y. Zhang, A. Taherkordi, and T. Skeie, “Mobile edge com-
puting: A survey,” IEEE Internet Things J., vol. 5, no. 1, pp. 450–465,
Feb. 2018.
[43] “Digital Twins for Aerospace: Better Fleet Reliability and
Performance.” 2021. [Online]. Available: https://www.altoros.
com/blog/digital-twins-for-aerospace-better-fleet-reliability-and-
performance/
[44] D. S. Lun, M. Médard, R. Koetter, and M. Effros, “On coding for
reliable communication over packet networks,” Phys. Commun., vol. 1,
no. 1, pp. 3–20, 2008.
[45] Z. Guo, J. Huang, B. Wang, S. Zhou, J.-H. Cui, and P. Willett, “A
practical joint network-channel coding scheme for reliable communi-
cation in wireless networks,” IEEE Trans. Wireless Commun., vol. 11,
no. 6, pp. 2084–2094, Jun. 2012.
[46] C. Hausl, “Joint network-channel coding for the multiple-access relay
channel based on turbo codes,” Eur. Trans. Telecommun., vol. 20, no. 2,
pp. 175–181, 2009.
[47] E. Ditto. “Digital Twins: The Ultimate Guide.” 2021. [Online].
Available: https://www.eclipse.org/ditto/
[48] “COMEA-TUAS/MCX-Public.” 2021. [Online]. Available:
https://github.com/COMEA-TUAS/mcx-public
[49] “MCX—An Open-Source Framework for Digital Twins.” 2021.
[Online]. Available: https://research.utu.fi/converis/portal/detail/
Publication/66484819?
[50] “Experience the Future with Mago3D.” Jul. 2021. [Online]. Available:
http://mago3d.com/en/index.html
[51] “State of mago3D, An Open Source Based Digital Twin Platform.”
Jul. 2021. [Online]. Available: https://callforpapers.2021.foss4g.org/
foss4g2021/talk/MJPNQC/
[52] “Lecture #19: Virtual Machine Design & Implementation.” Oct. 2021.
[Online]. Available: http://course.ece.cmu.edu/ ece600/fall17/lectures/
lecture19.pdf
[53] R. W. L. Coutinho and A. Boukerche, “Modeling and analysis of
a shared edge caching system for connected cars and industrial
IoT-based applications,” IEEE Trans. Ind. Informat., vol. 16, no. 3,
pp. 2003–2012, Mar. 2020.
[54] “Containers vs. Virtual Machines (VMs): What’s the Difference?”
NetApp. Oct. 2021. [Online]. Available: https://www.netapp.com/blog/
containers-vs-vms/
[55] “To Containerize or Not to Containerize, That is the Question,
or Containers vs VMs: The Eternal Debate.” Oct. 2021. [Online].
Available: https://www.mirantis.com/blog/containers-vs-vms-eternal-
debate/
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2251
[56] “What is Container Security and How to Secure Containers?”
Nov. 2022. [Online]. Available: https://www.decipherzone.com/blog-
detail/what-why-how-container-security
[57] “Containers vs. Virtual Machines.” Nov. 2022. [Online]. Available:
https://www.atlassian.com/microservices/cloud-computing/containers-
vs-vms#
[58] A. S. Khan and F. U. Khan, “A survey of wearable energy harvesting
systems,” Int. J. Energy Res., vol. 46, no. 3, pp. 2277–2329, 2022.
[59] W. Tang et al., “Wireless communications with reconfigurable intelli-
gent surface: Path loss modeling and experimental measurement,” IEEE
Trans. Wireless Commun., vol. 20, no. 1, pp. 421–439, Jan. 2021.
[60] L. U. Khan et al., “Federated learning for edge networks: Resource
optimization and incentive mechanism,” IEEE Commun. Mag., vol. 58,
no. 10, pp. 88–93, Oct. 2020.
[61] “Designing for Extreme Low Power.” Oct. 2021. [Online]. Available:
https://semiengineering.com/designing-for-extreme-low-power/
[62] “Federated learning for Internet of Things: Recent advances, taxonomy,
and open challenges,” IEEE Commun. Surveys Tuts., vol. 23, no. 3,
pp. 1759–1799, 3rd Quart., 2021.
[63] L. Ceze, M. D. Hill, and T. F. Wenisch, “Arch2030: A vision
of computer architecture research over the next 15 years,” 2016,
arXiv:1612.03182.
[64] R. G. Kim, J. R. Doppa, and P. P. Pande, “Machine learning for design
space exploration and optimization of manycore systems,” in Proc.
IEEE/ACM Int. Conf. Comput.-Aided Design (ICCAD), 2018, pp. 1–6.
[65] C. Liu et al., “Progressive neural architecture search,” in Proc. Eur.
Conf. Comput. Vis. (ECCV), 2018, pp. 19–34.
[66] Y. Zhou et al., “EPNAS: Efficient progressive neural architecture
search,” 2019, arXiv:1907.04648.
[67] T. Elsken, J. H. Metzen, and F. Hutter, “Neural architecture search:
A survey,” J. Mach. Learn. Res., vol. 20, no. 1, pp. 1997–2017, Mar.
2019.
[68] H. Pham, M. Guan, B. Zoph, Q. Le, and J. Dean, “Efficient neural
architecture search via parameters sharing,” in Proc. Int. Conf. Mach.
Learn., 2018, pp. 4095–4104.
[69] “How IoT Works—Part 4: User Interface.” 2021. [Online]. Available:
https://trackinno.com/iot/how-iot-works-part-4-user-interface/
[70] “How a VM is Really Isolated?” 2021. [Online]. Available:
https://www.business-standard.com/
[71] S. R. Pandey, N. H. Tran, M. Bennis, Y. K. Tun, A. Manzoor,
and C. S. Hong, “A crowdsourcing framework for on-device fed-
erated learning,” IEEE Trans. Wireless Commun., vol. 19, no. 5,
pp. 3241–3256, May 2020.
[72] T. H. T. Le, N. H. Tran, Y. K. Tun, Z. Han, and C. S. Hong, “Auction
based incentive design for efficient federated learning in cellular wire-
less networks,” in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC),
2020, pp. 1–6.
[73] J. Kang, Z. Xiong, D. Niyato, S. Xie, and J. Zhang, “Incentive mech-
anism for reliable federated learning: A joint optimization approach
to combining reputation and contract theory,” IEEE Internet Things J.,
vol. 6, no. 6, pp. 10700–10714, Dec. 2019.
[74] L. U. Khan, Y. K. Tun, M. Alsenwi, M. Imran, Z. Han, and C. S. Hong,
“A dispersed federated learning framework for 6G-enabled autonomous
driving cars,” 2021, arXiv:2105.09641.
[75] Y. Liu, Z. Fang, M. H. Cheung, W. Cai, and J. Huang, “An
incentive mechanism for sustainable blockchain storage,” 2021,
arXiv:2103.05866v3.
[76] K. Ludwig, A. Fendt, and B. Bauer, “An efficient online heuristic for
mobile network slice embedding,” in Proc. 23rd Conf. Innov. Clouds
Internet Netw. Workshops (ICIN), 2020, pp. 139–143.
[77] S. De Bast, R. Torrea-Duran, A. Chiumento, S. Pollin, and H. Gacanin,
“Deep reinforcement learning for dynamic network slicing in IEEE
802.11 networks,” in Proc. IEEE INFOCOM Conf. Comput. Commun.
Workshops (INFOCOM WKSHPS), 2019, pp. 264–269.
[78] D. H. Kim, S. M. A. Kazmi, A. Ndikumana, A. Manzoor, W. Saad,
and C. S. Hong, “Distributed radio slice allocation in wireless network
virtualization: Matching theory meets auctions,” IEEE Access, vol. 8,
pp. 73494–73507, 2020.
[79] T. M. Ho, N. H. Tran, L. B. Le, Z. Han, S. A. Kazmi, and C. S. Hong,
“Network virtualization with energy efficiency optimization for wire-
less heterogeneous networks,” IEEE Trans. Mobile Comput., vol. 18,
no. 10, pp. 2386–2400, Oct. 2019.
[80] S. M. A. Kazmi and C. S. Hong, “A matching game approach for
resource allocation in wireless network virtualization,” in Proc. 11th
Int. Conf. Ubiquitous Inf. Manage. Commun., 2017, pp. 1–6.
2252 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
[81] S. A. Kazmi, L. U. Khan, N. H. Tran, and C. S. Hong, Network Slicing
for 5G and Beyond Networks, vol. 1. Cham, Switzerland: Springer,
2019.
[82] L. U. Khan, I. Yaqoob, N. H. Tran, Z. Han, and C. S. Hong, “Network
slicing: Recent advances, taxonomy, requirements, and open research
challenges,” IEEE Access, vol. 8, pp. 36009–36028, 2020.
[83] C.-Y. Chang, N. Nikaein, and T. Spyropoulos, “Radio access
network resource slicing for flexible service execution,” in Proc.
IEEE INFOCOM Conf. Comput. Commun. Workshops (INFOCOM
WKSHPS), 2018, pp. 668–673.
[84] E. Datsika, A. Antonopoulos, N. Zorba, and C. Verikoukis, “Matching
game based virtualization in shared LTE-A networks,” in Proc. IEEE
Global Commun. Conf. (GLOBECOM), 2016, pp. 1–6.
[85] S. A. Kazmi, N. H. Tran, T. M. Ho, and C. S. Hong, “Hierarchical
matching game for service selection and resource purchasing in wire-
less network virtualization,” IEEE Commun. Lett., vol. 22, no. 1,
pp. 121–124, Jan. 2018.
[86] “How a VM is Really Isolated?” 2021. [Online]. Available:
https://vinfrastructure.it/2017/04/vm-really-isolated/
[87] “VMware Cloud Director Documentation.” 2021. [Online]. Available:
https://docs.vmware.com/en/VMware-Cloud-Director/index.html
[88] S. Bera, S. Misra, and A. V. Vasilakos, “Software-defined networking
for Internet of Things: A survey,” IEEE Internet Things J., vol. 4, no. 6,
pp. 1994–2008, Dec. 2017.
[89] A. H. Mohammed, R. M. Khaleefah, M. K. Hussein, and
I. A. Abdulateef, “A review software defined networking for Internet of
Things,” in Proc. Int. Congr. Human-Comput. Interact. Optim. Robotic
Appl. (HORA), 2020, pp. 1–8.
[90] K. Kalkan and S. Zeadally, “Securing Internet of Things with software
defined networking,” IEEE Commun. Mag., vol. 56, no. 9, pp. 186–192,
Sep. 2018.
[91] B. Han, V. Gopalakrishnan, L. Ji, and S. Lee, “Network function
virtualization: Challenges and opportunities for innovations,” IEEE
Commun. Mag., vol. 53, no. 2, pp. 90–97, Feb. 2015.
[92] R. Mijumbi, J. Serrat, J.-L. Gorricho, N. Bouten, F. De Turck, and
R. Boutaba, “Network function virtualization: State-of-the-art and
research challenges,” IEEE Commun. Surveys Tuts., vol. 18, no. 1,
pp. 236–262, 1st Quart., 2016.
[93] K. Joshi and T. Benson, “Network function virtualization,” IEEE
Internet Comput., vol. 20, no. 6, pp. 7–9, Nov. 2016.
[94] S. Lal, T. Taleb, and A. Dutta, “NFV: Security threats and best
practices,” IEEE Commun. Mag., vol. 55, no. 8, pp. 211–217, Aug.
2017.
[95] S. Wang, J. Xu, N. Zhang, and Y. Liu, “A survey on service migration
in mobile edge computing,” IEEE Access, vol. 6, pp. 23511–23528,
2018.
[96] N. Hassan, K.-L. A. Yau, and C. Wu, “Edge computing in 5G: A
review,” IEEE Access, vol. 7, pp. 127276–127289, 2019.
[97] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, “Mobile
edge computing—A key technology towards 5G,” ETSI, Sophia
Antipolis, France, White Paper, 2015.
[98] M. Karakus and A. Durresi, “A survey: Control plane scalability
issues and approaches in software-defined networking (SDN),” Comput.
Netw., vol. 112, pp. 279–293, Jan. 2017.
[99] E. Amiri, E. Alizadeh, and K. Raeisi, “An efficient hierarchical dis-
tributed SDN controller model,” in Proc. 5th Conf. Knowl. Based Eng.
Innov. (KBEI), 2019, pp. 553–557.
[100] D. E. Sarmiento, A. Lebre, L. Nussbaum, and A. Chari, “Decentralized
SDN control plane for a distributed cloud-edge infrastructure: A sur-
vey,” IEEE Commun. Surveys Tuts., vol. 23, no. 1, pp. 256–281, 1st
Quart., 2021.
[101] F. Bannour, S. Souihi, and A. Mellouk, “Distributed SDN control:
Survey, taxonomy, and challenges,” IEEE Commun. Surveys Tuts.,
vol. 20, no. 1, pp. 333–354, 1st Quart., 2018.
[102] M. A. Togou, D. A. Chekired, L. Khoukhi, and G.-M. Muntean, “A
hierarchical distributed control plane for path computation scalability
in large scale software-defined networks,” IEEE Trans. Netw. Service
Manag., vol. 16, no. 3, pp. 1019–1031, Sep. 2019.
[103] S. Ahmad and A. H. Mir, “Scalability, consistency, reliability and secu-
rity in sdn controllers: A survey of diverse sdn controllers,” J. Netw.
Syst. Manage., vol. 29, no. 1, pp. 1–59, Nov. 2021.
[104] S. Huang, J. Zhao, and X. Wang, “HybridFlow: A lightweight control
plane for hybrid SDN in enterprise networks,” in Proc. IEEE/ACM 24th
Int. Symp. Qual. Service (IWQoS), Jun. 2016, pp. 1–2.
[105] Y. Fu et al., “A hybrid hierarchical control plane for flow-based large-
scale software-defined networks,” IEEE Trans. Netw. Service Manag.,
vol. 12, no. 2, pp. 117–131, Jun. 2015.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
[106] Z. Zhou, K. Ota, M. Dong, and C. Xu, “Energy-efficient matching for
resource allocation in D2D enabled cellular networks,” IEEE Trans.
Veh. Technol., vol. 66, no. 6, pp. 5256–5268, Jun. 2017.
[107] H. Zhang, Y. Xiao, S. Bu, D. Niyato, F. R. Yu, and Z. Han, “Computing
resource allocation in three-tier IoT fog networks: A joint optimization
approach combining Stackelberg game and matching,” IEEE Internet
Things J., vol. 4, no. 5, pp. 1204–1215, Oct. 2017.
[108] A. Ahmed, M. Awais, T. Akram, S. Kulac, M. Alhussein, and
K. Aurangzeb, “Joint placement and device association of UAV base
stations in IoT networks,” Sensors, vol. 19, no. 9, p. 2157, 2019.
[109] H. Zhang, S. Huang, C. Jiang, K. Long, V. C. Leung, and H. V. Poor,
“Energy efficient user association and power allocation in millimeter-
wave-based ultra dense networks with energy harvesting base stations,”
IEEE J. Sel. Areas Commun., vol. 35, no. 9, pp. 1936–1947, Sep. 2017.
[110] A. Azizi, N. Mokari, and M. R. Javan, “Joint radio resource alloca-
tion, 3D placement and user association of aerial base stations in IoT
networks,” 2017, arXiv:1710.05315.
[111] A. K. Bairagi, M. S. Munir, M. Alsenwi, N. H. Tran, and C. S. Hong,
“A matching based coexistence mechanism between eMBB and uRLLC
in 5G wireless networks,” in Proc. 34th ACM/SIGAPP Symp. Appl.
Comput., 2019, pp. 2377–2384.
[112] K. Kim and C. S. Hong, “Optimal task-UAV-edge matching for com-
putation offloading in UAV assisted mobile edge computing,” in Proc.
20th Asia-Pac. Netw. Oper. Manage. Symp. (APNOMS), 2019, pp. 1–4.
[113] R. O. Afolabi, A. Dadlani, and K. Kim, “Multicast scheduling and
resource allocation algorithms for OFDMA-based systems: A sur-
vey,” IEEE Commun. Surveys Tuts., vol. 15, no. 1, pp. 240–254, 1st
Quart., 2013.
[114] B. Di, S. Bayat, L. Song, and Y. Li, “Radio resource alloca-
tion for downlink non-orthogonal multiple access (NOMA) networks
using matching theory,” in Proc. IEEE Global Commun. Conf.
(GLOBECOM), 2015, pp. 1–6.
[115] T. LeAnh et al., “Matching theory for distributed user association and
resource allocation in cognitive femtocell networks,” IEEE Trans. Veh.
Technol., vol. 66, no. 9, pp. 8413–8428, Sep. 2017.
[116] A. Zakeri, M. Moltafet, and N. Mokari, “Joint radio resource alloca-
tion and SIC ordering in NOMA-based networks using submodularity
and matching theory,” IEEE Trans. Veh. Technol., vol. 68, no. 10,
pp. 9761–9773, Oct. 2019.
[117] J. C. C. Chica, J. C. Imbachi, and J. F. B. Vega, “Security in SDN: A
comprehensive survey,” J. Netw. Comput. Appl., vol. 159, Jun. 2020,
Art. no. 102595.
[118] K. K. R. Kendall, “A database of computer attacks for the evaluation
of intrusion detection systems,” Ph.D. dissertation, Dept. Electr. Eng.
Comput. Sci., Massachusetts Inst. Technol., Cambridge, MA, USA,
1999.
[119] L. U. Khan, Z. Han, D. Niyato, and C. S. Hong, “Socially-aware-
clustering-enabled federated learning for edge networks,” IEEE Trans.
Netw. Service Manag., vol. 18, no. 3, pp. 2641–2658, Sep. 2021.
[120] K. Yang, T. Jiang, Y. Shi, and Z. Ding, “Federated learning via over-
the-air computation,” IEEE Trans. Wireless Commun., vol. 19, no. 3,
pp. 2022–2035, Mar. 2020.
[121] W. Liu, X. Zang, Y. Li, and B. Vucetic, “Over-the-air computa-
tion systems: Optimization, analysis and scaling laws,” IEEE Trans.
Wireless Commun., vol. 19, no. 8, pp. 5488–5502, Aug. 2020.
[122] T. Jiang and Y. Shi, “Over-the-air computation via intelligent reflecting
surfaces,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), 2019,
pp. 1–6.
[123] M. Chen, H. V. Poor, W. Saad, and S. Cui, “Wireless communications
for collaborative federated learning,” IEEE Commun. Mag., vol. 58,
no. 12, pp. 48–54, Dec. 2020.
[124] U. Majeed, L. U. Khan, and C. S. Hong, “Cross-silo horizontal fed-
erated learning for flow-based time-related-features oriented traffic
classification,” in Proc. 21st Asia-Pac. Netw. Oper. Manage. Symp.
(APNOMS), 2020, pp. 389–392.
[125] L. U. Khan, Z. Sabir, S. A. Mahmud, and G. M. Khan, “Comparison
of three interpolation techniques in comb-type pilot-assisted channel
coded OFDM system,” in Proc. 27th Int. Conf. Adv. Inf. Netw. Appl.
Workshops, 2013, pp. 977–981.
[126] Z. Wei, J. Yuan, D. W. K. Ng, M. Elkashlan, and Z. Ding, “A
survey of downlink non-orthogonal multiple access for 5G wireless
communication networks,” 2016, arXiv:1609.01856.
[127] J. Zhao, “A survey of intelligent reflecting surfaces (IRSs): Towards
6G wireless communication networks with massive MIMO 2.0,” 2019,
arXiv:1907.04789.
KHAN et al.: DIGITAL TWIN OF WIRELESS SYSTEMS: OVERVIEW, TAXONOMY, CHALLENGES, AND OPPORTUNITIES
[128] E. Basar, M. D. Renzo, J. De Rosny, M. Debbah, M.-S. Alouini, and
R. Zhang, “Wireless communications through reconfigurable intelligent
surfaces,” IEEE Access, vol. 7, pp. 116753–116773, 2019.
[129] P. Wang, J. Fang, H. Duan, and H. Li, “Compressed channel estima-
tion for intelligent reflecting surface-assisted millimeter wave systems,”
IEEE Signal Process. Lett., vol. 27, pp. 905–909, May 2020.
[130] Q.-U.-A. Nadeem, A. Kammoun, A. Chaaban, M. Debbah, and
M.-S. Alouini, “Intelligent reflecting surface assisted wireless commu-
nication: Modeling and channel estimation,” 2019, arXiv:1906.02360.
[131] S. Liu, Z. Gao, J. Zhang, M. Di Renzo, and M.-S. Alouini, “Deep
denoising neural network assisted compressive channel estimation for
mmWave intelligent reflecting surfaces,” IEEE Trans. Veh. Technol.,
vol. 69, no. 8, pp. 9223–9228, Aug. 2020.
[132] S. Gong et al., “Toward smart wireless communications via intelligent
reflecting surfaces: A contemporary survey,” IEEE Commun. Surveys
Tuts., vol. 22, no. 4, pp. 2283–2314, 4th Quart., 2020.
[133] F. Emmert-Streib, Z. Yang, H. Feng, S. Tripathi, and M. Dehmer, “An
introductory review of deep learning for prediction models with big
data,” Front. Artif. Intell., vol. 3, p. 4, Feb. 2020.
[134] S. Pouyanfar et al., “A survey on deep learning: Algorithms, techniques,
and applications,” ACM Comput. Surveys, vol. 51, no. 5, pp. 1–36, Sep.
2019.
[135] M. Masdari, S. S. Nabavi, and V. Ahmadi, “An overview of virtual
machine placement schemes in cloud computing,” J. Netw. Comput.
Appl., vol. 66, pp. 106–127, May 2016.
[136] P. Thakur and D. K. Shrivastava, “Interoperability issues and standard
architecture for service delivery in federated cloud: A review,” in Proc.
Int. Conf. Comput. Intell. Commun. Netw., 2015, pp. 908–912.
[137] H. F. Atlam, A. Alenezi, M. O. Alassafi, and G. B. Wills, “Blockchain
with Internet of Things: Benefits, challenges, and future directions,”
Int. J. Intell. Syst. Appl., vol. 10, no. 6, pp. 40–48, 2018.
[138] M. Stoyanova, Y. Nikoloudakis, S. Panagiotakis, E. Pallis, and
E. K. Markakis, “A survey on the Internet of Things (IoT) foren-
sics: Challenges, approaches, and open issues,” IEEE Commun. Surveys
Tuts., vol. 22, no. 2, pp. 1191–1221, 2nd Quart., 2020.
[139] C. S. Hong, L. U. Khan, M. Chen, D. Chen, W. Saad, and Z. Han,
“Resource optimization for wireless federated learning,” in Federated
Learning for Wireless Networks. Singapore: Springer, 2021, pp. 27–69.
[140] N. C. Luong et al., “Applications of deep reinforcement learning in
communications and networking: A survey,” IEEE Commun. Surveys
Tuts., vol. 21, no. 4, pp. 3133–3174, 4th Quart., 2019.
[141] C. S. Hong, L. U. Khan, M. Chen, D. Chen, W. Saad, and
Z. Han, “Smart industries and intelligent reflecting surfaces,” in
Federated Learning for Wireless Networks. Singapore: Springer, 2021,
pp. 221–239.
Latif U. Khan received the M.S. degree in
electrical engineering with distinction from UET
Peshawar, Pakistan, in 2017, and the Ph.D. degree in
computer engineering from Kyung Hee University
(KHU), South Korea, in 2021. He worked as a
Leading Researcher with the intelligent Networking
Laboratory under a project jointly funded by the
prestigious Brain Korea 21st Century Plus and
Ministry of Science and ICT, South Korea. Prior to
joining the KHU, he has served as a Faculty Member
and a Research Associate with UET Peshawar. He
has published his works in highly reputable conferences and journals. He is the
author/coauthor of two conference best paper awards. He is also the author
of two books titled “Network Slicing for 5G and Beyond Networks” and
“Federated Learning for Wireless Networks”. His research interests include
analytical techniques of optimization and game theory to edge computing,
end-to-end network slicing, digital twins, and federated learning for wireless
networks. He is the recipient of KHU best thesis award.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
2253
Zhu Han (Fellow, IEEE) received the B.S. degree
in electronic engineering from Tsinghua University
in 1997, and the M.S. and Ph.D. degrees in electri-
cal and computer engineering from the University of
Maryland, College Park, in 1999 and 2003, respec-
tively. From 2000 to 2002, he was a Research and
Development Engineer with JDSU, Germantown,
MD, USA. From 2003 to 2006, he was a Research
Associate with the University of Maryland. From
2006 to 2008, he was an Assistant Professor with
Boise State University, Boise, ID, USA. He is
Currently a John and Rebecca Moores Professor with the Electrical and
Computer Engineering Department as well as Computer Science Department,
University of Houston, Houston, TX, USA. His research interests include
wireless resource allocation and management, wireless communications and
networking, game theory, big data analysis, security, and smart grid. He
received the NSF Career Award in 2010, the Fred W. Ellersick Prize of the
IEEE Communication Society in 2011, the EURASIP Best Paper Award for
the Journal on Advances in Signal Processing in 2015, the IEEE Leonard G.
Abraham Prize in the field of Communications Systems (Best Paper Award in
IEEE JSAC) in 2016, and several Best Paper Awards in IEEE conferences. He
is also the winner of 2021 IEEE Kiyo Tomiyasu Award, for outstanding early
to mid-career contributions to technologies holding the promise of innovative
applications, with the following citation: “for Contributions to Game Theory
and Distributed Management of Autonomous Communication Networks." He
has been a 1% Highly Cited Researcher since 2017 according to Web of
Science. He was an IEEE Communications Society Distinguished Lecturer
from 2015 to 2018. He has been an AAAS Fellow and an ACM Distinguished
Member since 2019.
Walid Saad (Fellow, IEEE) received the Ph.D.
degree from the University of Oslo, Oslo, Norway,
in 2010. He is currently a Professor with the
Department of Electrical and Computer Engineering,
Virginia Tech, Blacksburg, VA, USA, where he
leads the Network sciEnce, Wireless, and Security
Laboratory. He was named the Stephen O. Lane
Junior Faculty Fellow with Virginia Tech from 2015
to 2017. In 2017, he was named as the College of
Engineering Faculty Fellow. His research interests
include wireless networks, machine learning, game
theory, security, unmanned aerial vehicles, cyber-physical systems, and
network science. He was a recipient of the NSF CAREER Award in 2013,
the Air Force Office of Scientific Research Summer Faculty Fellowship in
2014, and the Young Investigator Award from the Office of Naval Research
in 2015. He was the author/coauthor of eight conference best paper awards
at WiOpt in 2009, the International Conference on Internet Monitoring and
Protection in 2010, the IEEE Wireless Communications and Networking
Conference in 2012, the IEEE International Symposium on Personal, Indoor
and Mobile Radio Communications in 2015, IEEE SmartGridComm in 2015,
EuCNC in 2017, IEEE GLOBECOM in 2018, and the International Federation
for Information Processing International Conference on New Technologies,
Mobility and Security in 2019. He was also a recipient of the 2015 Fred
W. Ellersick Prize from the IEEE Communications Society, the 2017 IEEE
ComSoc Best Young Professional in Academia award, the 2018 IEEE ComSoc
Radio Communications Committee Early Achievement Award, and the 2019
IEEE ComSoc Communication Theory Technical Committee. He received
the Dean’s Award for Research Excellence from Virginia Tech in 2019.
He also serves as an Editor for the IEEE T RANSACTIONS ON W IRELESS
C OMMUNICATIONS, the IEEE T RANSACTIONS ON M OBILE C OMPUTING,
and the IEEE T RANSACTIONS ON C OGNITIVE C OMMUNICATIONS AND
N ETWORKING. He is an Editor-at-Large for the IEEE T RANSACTIONS ON
C OMMUNICATIONS and an IEEE Distinguished Lecturer.
2254 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 24, NO. 4, FOURTH QUARTER 2022
Ekram Hossain (Fellow, IEEE) is a Professor
with the Department of Electrical and Computer
Engineering, University of Manitoba, Canada
(https://home.cc.umanitoba.ca/ hossaina/). He is a
member (Class of 2016) of the College of the
Royal Society of Canada, a Fellow of the Canadian
Academy of Engineering, and a Fellow of the
Engineering Institute of Canada. He was elevated
to an IEEE Fellow “for Contributions to Spectrum
Management and Resource Allocation in Cognitive
and Cellular Radio Networks”. He was listed as a
Clarivate Analytics Highly Cited Researcher in Computer Science in 2017,
2018, 2019, 2020, and 2021. He served as the Editor-in-Chief for the IEEE
C OMMUNICATIONS S URVEYS AND T UTORIALS from 2012 to 2016, and the
IEEE Press from 2018 to 2021.
Mohsen Guizani (Fellow, IEEE) received the B.S.
(with distinction) and M.S. degrees in electrical
engineering, and the M.S. and Ph.D. degrees in
computer engineering from Syracuse University,
New York, in 1984, 1986, 1987, and 1990,
respectively. He is currently a Professor with
the Machine Learning Department, Mohamed Bin
Zayed University of Artificial Intelligence, Abu
Dhabi, UAE. Previously, he served in different
academic and administrative positions with the
University of Idaho, Western Michigan University,
the University of West Florida, the University of Missouri-Kansas City, the
University of Colorado-Boulder, and Syracuse University. He is currently an
IEEE ComSoc Distinguished Lecturer. He is the author of nine books and
more than 500 publications in refereed journals and conferences. His research
interests include wireless communications and mobile computing, computer
networks, mobile cloud computing, security, and smart grid. He serves on
the Editorial Boards of several international technical journals, and is the
Founder and the Editor-in-Chief of the Wireless Communications and Mobile
Computing journal (Wiley). He has also served as a TPC member, the Chair,
and the General Chair of a number of international conferences. Throughout
his career, he received three teaching awards and four research awards. He also
received the 2017 IEEE Communications Society WTC Recognition Award
as well as the 2018 AdHoc Technical Committee Recognition Award for
his contribution to outstanding research in wireless communications and ad
hoc sensor networks. He served as a IEEE Computer Society Distinguished
Speaker. He was the Chair of the IEEE Communications Society Wireless
Technical Committee and the Chair of the TAOS Technical Committee. He
was the Editor-in-Chief of IEEE N ETWORK. He has guest edited a number
of Special Issues in IEEE journals and magazines. He is a Senior Member of
ACM.
Authorized licensed use limited to: KAUST. Downloaded on May 07,2023 at 18:12:47 UTC from IEEE Xplore. Restrictions apply.
Choong Seon Hong (Senior Member, IEEE)
received the B.S. and M.S. degrees in electronic
engineering from Kyung Hee University, Seoul,
South Korea, in 1983 and 1985, respectively, and the
Ph.D. degree from Keio University, Tokyo, Japan,
in 1997. In 1988, he joined with KT, Gyeonggi-do,
South Korea, where he was involved in broadband
networks as a member of the Technical Staff. Since
1993, he has been with Keio University. He was
with the Telecommunications Network Laboratory,
KT, as a Senior Member of Technical Staff and
as the Director of the Networking Research Team until 1999. Since 1999,
he has been a Professor with the Department of Computer Science and
Engineering, Kyung Hee University. His research interests include future
Internet, intelligent edge computing, network management, and network secu-
rity. He is a member of the Association for Computing Machinery, the Institute
of Electronics, Information and Communication Engineers, the Information
Processing Society of Japan, the Korean Institute of Information Scientists and
Engineers, the Korean Institute of Communications and Information Sciences,
the Korean Information Processing Society, and the Open Standards and ICT
Association. He has served as the General Chair, the TPC Chair/Member,
or an Organizing Committee Member of international conferences, such
as the Network Operations and Management Symposium, International
Symposium on Integrated Network Management, Asia-Pacific Network
Operations and Management Symposium , End-to-End Monitoring Techniques
and Services, IEEE Consumer Communications and Networking Conference,
Assurance in Distributed Systems and Networks, International Conference
on Parallel Processing, Data Integration and Mining, World Conference
on Information Security Applications, Broadband Convergence Network,
Telecommunication Information Networking Architecture, International
Symposium on Applications and the Internet, and International Conference
on Information Networking. He was an Associate Editor of the IEEE
T RANSACTIONS ON N ETWORK AND S ERVICE M ANAGEMENT and the IEEE
J OURNAL OF C OMMUNICATIONS AND N ETWORKS. He currently serves as
an Associate Editor for the International Journal of Network Management.