Received 1 August 2022; accepted 6 August 2022. Date of publication 10 August 2022; date of current version 22 August 2022.

Digital Object Identifier 10.1109/OJCOMS.2022.3197811

Survey on Digital Twin Edge Networks (DITEN)

Toward 6G
FENGXIAO TANG1 (Member, IEEE), XUEHAN CHEN 1 (Member, IEEE),
TIAGO KOKETSU RODRIGUES 2 (Member, IEEE), MING ZHAO 1 (Member, IEEE),
AND NEI KATO 2 (Fellow, IEEE)

(Invited Paper)
1 School of Computer Science and Engineering, Central South University, Changsha 410083, China

2 Graduate School of Information Sciences, Tohoku University, Sendai 9840014, Japan

CORRESPONDING AUTHOR: X. CHEN (e-mail: xhss1991@163.com)

This work was supported in part by the National Natural Science Foundation of China under Grant 6200239; in part by the Key Research and
Development Program of Xinjiang Autonomous Region under Grant 2021B01002; and in part by the Scientific and Technological Innovation
2030 of China under Grant 2020AAA0109600.

ABSTRACT The next generation (6G) wireless systems aim to cater to the Internet of Everything (IoE)
and revolutionize customer services and applications to a fully intelligent and autonomous system. To
achieve this, the digital twin edge network (DITEN) is proposed to combine mobile/multi-access edge
computing (MEC) and digital twin (DT), thereby improving the network performance such as throughput
and security, and reducing the cost of communication, computation, and caching. In DITENs, the network
status can be continuously monitored, and based on the obtained network states, the networking schemes,
such as routing and resource management, can be studied in the established DITENs from a centralized
perspective. In this survey, we present a comprehensive overview of DITEN for 6G. First, we present the
fundamental aspects of DITEN, including concept, framework, and potential. Second, a comprehensive
design of DITEN is devised, including the DT modeling/updating, DT deployment, key issues, and enabling
technologies. Then, the typical applications of DITEN towards 6G are provided, including the Internet
of Things (IoT), vehicular network, space-air-ground integrated network (SAGIN), healthcare, wireless
systems, and other applications, along with the design of DITEN in each application, such as DT modeling,
DT association, incentive mechanisms, and so on. Finally, challenges and open issues are discussed.

INDEX TERMS Digital twin edge networks (DITEN), 6G, DT modeling, DT deployment, applications,
challenges.

I. INTRODUCTION critical to integrate some advanced technologies to design

T HE SIXTH-GENERATION (6G) wireless systems [1]

are designed to cater to the emerging Internet of
Everything (IoE) applications, such as brain-computer
the 6G wireless system, such as applying Terahertz com-
munications [4] and intelligent reflect surface (IRS) [5]
to improve the efficiency of communications [6], [7], and
interaction (BCI), multisensory extended reality (XR), and utilizing multi-access/mobile edging computing (MEC) [8]
autonomous systems, to achieve a fully intelligent and and artificial intelligence (AI) [9] to improve the efficiency
autonomous system for customer services and applications of computations [10], [11], as well as employing security-
in the future [2], as well as to provide users with an immer- related technologies such as blockchain to avoid the security
sive experience in a virtual space such as Metaverse [3]. risk [12]. Furthermore, 6G should also be designed as a
To achieve these, there are some strict requirements on self-sustaining wireless system to minimize the interven-
6G, such as ultra-high throughput and reliability, ultra-low tion from the network operators/end-users [13], and to be
latency, seamless connectivity, and so on. To this end, it is developed as a proactive-online-learning-enabled system to
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

1360 VOLUME 3, 2022

achieve efficient resource management by performing intel- Specifically, a DITEN improves the system performance,
ligent analysis such as proactive learning before the service such as latency and reliability, by accurate DT modeling
request [14]. and intelligent decisions (e.g., DT association, intelligent
The advanced wireless communication and computation task offloading, and resource allocation), applying various
technologies drive the applications of the digital twin (DT) technologies, i.e., optimization theory, game theory, ML,
technology [15]. DT can accurately replicate the physical and blockchain. Although DITEN has some potential on
object in the digital domain, and the physical object inter- improving system performance, it still faces some challenges
acts constantly and evolutes synchronously with its virtual considering the limited wireless resource and security con-
twin throughout its life cycle [16], [17]. With DT, the con- cerns [32]. First, how to ensure enough data samples for
nection between the physical components and the digital the edge servers for DT modeling is challenging. The lim-
space can be achieved, and the physical system can be sim- ited wireless resource and traffic congestion in a network
ulated, predicted, analyzed, and optimized, by using various hinder the real-time data collection from the physical com-
technologies in terms of sensing, communication, computa- ponents. Second, how to motivate the resource rich end
tion, and analysis [18]. There are three types of DTs based devices and edge/cloud servers to contribute their resource
on their functions, namely monitoring DT, simulation DT, to DT modeling and updating is a key issue in DITEN.
and operational DT [19]. Specifically, the monitoring DT Third, how to ensure the data privacy and security in DITEN
is used for monitoring a physical system’s status. While, is a challenge, considering the large-scale users and data
the simulation DT aims at predicting the dynamics of a transmission, and the untrusted and nontransparent wireless
physical system by utilizing advanced simulation tools and network environment.
machine learning (ML) technologies. Meanwhile, the opera-
tional DT achieves the interactions between system operators A. EXISTING SURVEYS AND TUTORIALS
and a cyber-physical system, and performs various oper- There are a few works that overview the DT technology
ations, such as prediction and optimization. DT has been in various applications such as IoT, industrial applications,
widely applied in different applications, such as the Internet wireless communications, etc. The key concepts, applica-
of Things (IoT) [20], and the emerging 6G system [21]. tions, and design implications of DT are presented in [33].
The realization of DT involves many processes ranging The enabling technologies, challenges, and open researches
from data collection to visualization and service feedback, of DT are surveyed in [34]. For IoT applications, in [35],
which requires intensive resource in terms of communica- the DT technologies for smart cities including their val-
tion, computation, and storage, so that some resource-limited ues, challenges, and enablers are illustrated. In [14], the
end devices are hindered to construct and maintain DTs in authors survey DTs in the context of IoT, including tech-
local. To address this issue, cloud/edge computing can be nical characteristics, scenarios, and architectural elements.
used to assist with the implementation of DTs. In cloud com- In [36], the key features and definitions of digital twin
puting based DT modeling, a cloud server collects real-time network (DTN) are presented, along with the key enabling
information from physical objects, and constructs the phys- technologies, applications, and open issues. In [37], [38], the
ical objects’ DT based on the collected information. Thus, authors focus on the DT in smart cities. In [39], the authors
high communication overhead incurs since the long trans- study the characteristics of DT, communication technologies
mission distance. In this context, a new paradigm, named and tools for DT modeling, reference models, and standards,
digital twin edge networks (DITEN) is emerged [22], [23]. and discuss the challenges and open issues. In [40], the exist-
By DITEN, the MEC and DT are integrated to provide ser- ing studies on the cognitive digital twin (CDT) concept are
vice to various applications. In DITENs, edge nodes such as reviewed, along with its definitions and key features. In [41],
base stations (BSs) and access points (APs) collect real-time the authors review the research on DTs, including the key
operation information of physical objects, and their behavior technologies to apply DT, industrial applications, and future
model can be built along with the dynamical environment directions. In [42], the concept, characteristics, key technolo-
based on the collected information. Furthermore, the edge gies, application, and future directions of digital twin city
nodes interact with the physical components continuously to (DTC) are illustrated. In [18], the architectures, case study,
keep the consistency between the physical components and and challenges for a mobility digital twin (MDT) framework
their DTs. Thus, by designing and optimizing networking are presented. Although the above surveys and tutorials pro-
schemes directly in the constructed DITEN, the networking vide a guideline to apply DT technology, they focus on the
schemes’ efficiency can be improved and the cost of com- application of DT and the integration of DT and MEC is
munications and computations can be reduced [24], [25]. missing. The comprehensive survey on the works on DITEN
In the following, DT is used to represent the digital twin is still a vacant, and need to be filled up.
technology or a digital twin of a physical object.
Recently, the design of a DITEN has been studied in B. CONTRIBUTIONS OF THIS WORK
various scenarios, such as IoT [26], vehicular network [27], In this paper, we survey the existing works about the DITEN
space-air-ground integrated network (SAGIN) [28], 6G wire- system. The contributions of this work can be summarized
less system [29], healthcare [30], and other applications [31]. as follows,

VOLUME 3, 2022 1361

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

• Monitoring DT [44]. In this case, DT is used to mirror

or virtually represent a physical object. The physical
object does not interact with its virtual model, and the
changes of the physical object have no effect on its
virtual model after it is established, and vice versa.
• Simulation DT [45]. In this case, DT is defined as a
simulator or software. By simulating with computers,
the physical objects can be understood, predicted, and
optimized, so that their performance can be improved.
The virtual model evolutes with the physical object, but
the physical object will not change with the dynamics
of the virtual model.
• Operation DT [46], [47], [48]. In this case, the data
transmission between physical objects and their twins is
bidirectional. Specifically, the physical objects transmit
FIGURE 1. The concept of digital twin (DT) and its evolution.
their status information to their twins, based on that
information, the twins are constructed and updated, and
TABLE 1. The three types concept of DT. the conditions of the physical object can be predicted.
Meanwhile, the twin will feed back information, such as
optimal solutions to a problem, to guide the operations
of physical objects.
More recently, based on the above definitions, DT is
usually described as “an intelligent and evolving system
that accurately digitally replicates a physical object across
• We present the concept, framework, and potential of
multiple granularity levels, and monitors, controls, and opti-
DITEN for 6G. (Section II)
mizes the physical objects during its life cycle” [36]. As
• A comprehensive discussion of the design of DITEN
shown in Fig. 1, DT consists of three parts: i) physical
is presented, including the DT modeling/updating, DT
objects, which can be a robot, a car, a complex system, a
deployment, key issues, and enabling technologies.
person, and so on; ii) virtual twins of physical objects; and
(Section III)
iii) connection between the physical objects and their virtual
• The typical applications of DITEN in 6G are pro-
twins. With the connection, the physical objects send their
vided, including the IoT, vehicular network, SAGIN,
states and generated sensing information to the virtual twins
healthcare, wireless systems, and other applications. In
for DT modeling and updating, and the virtual twin will
addition, the design issues of DITEN in each application
also feed back to the physical objects. As such, the dynamic
are discussed, such as DT modeling, DT association,
interactions and synchronous evolution between the physical
incentive mechanisms, and so on. (Section IV)
object and the virtual twin can be achieved. To this end, DT
• Several challenges and open issues are discussed.
is an efficient way to simulate, analyze, predict, and optimize
(Section V)
the physical system throughout its life cycle.
II. DITEN: CONCEPT, FRAMEWORK, AND POTENTIAL
In this section, we present the fundamental aspects of DITEN 2) MOBILE/MULTI-ACCESS EDGE COMPUTING (MEC)
in 6G, including its concept, framework, and potential.
MEC is an emerging paradigm proposed by the European
A. CONCEPT Telecommunications Standards Institute (ETSI) to reduce
Before giving the concept of DITEN, we first present some property constraints on the existing cloud infrastruc-
information on DT and MEC. tures [49], [50]. MEC as a kind of distributed cloud
architecture forms the backbone technology of the upcom-
1) DIGITAL TWIN (DT) ing 6G wireless systems, which transforms the traditional
DT is an efficient way to bridge the physical space and dig- mobile BSs by enabling IT and cloud computing capabili-
ital space [43]. Owing to the advanced communication and ties at the edge of radio access network (RAN) closing to
computation technologies, the DT has gone through four end users, thereby decreasing delay and relieving the burden
development processes, i.e., information mirroring, digital on the communications [51], and improving the utility of the
simulations, values chain interaction, and connected opera- mobile backhaul and core networks [52]. By applying MEC,
tions and industrial services, as shown in Fig. 1. To this end, the tasks such as computation and caching can be pushed to
there are three categories of DT definition summarized in the network edges, such as BSs and roadside units (RSUs),
Table 1, and the detailed descriptions are as follows, then be processed locally close to the users [53]. To this end,

1362 VOLUME 3, 2022

 through twin-to-twin communications. In DITEN, the real-
time network states can be acquired and are used to obtain the
optimal decisions in the network directly from a centralized
perspective. To this end, the constructed DITEN can be uti-
lized to design and optimize the network schemes directly,
such as task offloading, resource allocation, caching, etc,
and the networking schemes’ efficiency can be improved
and the cost can be reduced. In practice, in the case without
DITEN, to perform task offloading and resource allocation
decisions, constant communications between edge servers
and end users within their coverage are required to obtain
real-time information. Therefore, the built of DTs assist
with obtaining optimal solutions to resource allocation while
minimizing the communication cost. DITEN has the poten-
tial to support some computation-intensive services, such as
Metaverse and autonomous driving.
FIGURE 2. The concept of digital twin edge network (DITEN). The integration of MEC and DT has been studied in sev-
eral existing works. Lu et al. in [58] propose a DTWN, in
the mobile device with the limited resource can also pro- which the data process tasks are offloaded to the edge servers
cess intensive computation tasks while satisfying ultra-low with the assistance of DT. Sun et al. in [22] first consider
latency requirement [54]. a DITEN, where DTs estimate edge servers’ states and pro-
MEC enhances the existing applications and has great vide training data to DRL agents to obtain the optimal task
potential to develop a wide variety of emerging innovative offloading decisions. Moreover, the deviations between true
applications by authorizing the third parties with computation values and the value of DTs are taken into consideration, and
and storage capabilities at the RAN edge. In addition, thanks the impacts of such deviations on offloading decisions are
to the seamless connectivity provided by cellular networks, also explored. In [28], a DT empowered air-ground network
MEC becomes a key enabling technology to support mature is proposed to represent the network environment virtu-
machine-to-machine (M2M) and IoT services to create ver- ally for scheme optimization. In [59], DT is leveraged to
tical services [55]. For example, MEC has been applied deal with the resource allocation problem in industrial edge
in vehicular networks (vehicular edge computing), SAGIN environments.
(Aerial edge computing), to empower transport, UAV, and
etc. Moreover, by integrating AI in MEC powerfully, edge B. FRAMEWORK
intelligence can be achieved to further improve the system Fig. 3 presents a framework of the DITEN, which can be
performance [56]. divided into three layers, namely the physical layer, vir-
tual layer, and application layer. The physical layer contains
3) DIGITAL TWIN EDGE NETWORKS (DITEN) all the physical devices like end devices and edge/cloud
Recently, a new paradigm named DITEN is proposed to servers, and the wireless communication environment in a
bridge the physical MEC system and digital systems [57]. In MEC system. The end devices have limited computation and
DITEN, DT models are built and maintained with the assis- storage resources, and need to offload tasks to one or more
tance of network edge [32]. As shown in Fig. 2, the DITEN edge servers for cooperative processing. The network phys-
consists of three parts, namely the user/physical layer, edge ical entities and the wireless communication environment
layer, and DTN. In the user/physical layer, local computa- in the physical layer have their twin mapping in the virtual
tion and communications are processed, and the real-time layer, such as twin end devices, twin edge/cloud servers, and
state and model parameters information of the user/physical twin communication environment. Data such as the real-time
are transmitted to the edge layer via the physical-to-physical states of the physical entities are collected in the physical
communication. In the edge layer, real-time DT modeling layer then be transmitted to the twins in the virtual layer via
and updates are processed. Specifically, the edge nodes such the physical-to-twin interfaces [60].
as BSs can collect user/physical components’ running sta- The virtual layer contains the DTs that fully copy the phys-
tus information, and evolute their behavior model based on ical objects, and manages and controls the physical layer with
the collected information along with the time-varying envi- simulation, prediction, and optimization [60]. The DT system
ronments. In addition, the edge nodes continuously monitor saves the raw data of physical entities, including hardware
physical components’ states to maintain consistency with configuration, user information such as location and avail-
their twins. DT can be placed in any edge server, and the able resource, historical data, and real-time operation states
edge layer is transparent for users. Meanwhile, the edge of the MEC system, and also keeps monitoring the dynamics
resources are shared among all users. In DTN, DTs are vir- of the system. This data is essential to constructing DT mod-
tually connected to form a network with the shared resources els of the physical entities and communication environment

VOLUME 3, 2022 1363

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

FIGURE 3. Framework of the digital twin edge network (DITEN).

in the virtual layer with high precision. However, in practice, and healthcare, request a service from the DITEN system,
it is challenging to model a physical entity/phenomenon vir- and the virtual layer feeds back the optimal decisions to the
tually exactly similar to the physical one. Besides, there are application layer.
some key issues to be addressed, such as efficiency, fault-
tolerance, low latency, and security. To solve these issues, C. POTENTIALS OF DITEN TOWARDS 6G
various advanced enabling technologies in terms of com- The DITEN has some potential to support various applica-
munications, data processing, incentive mechanisms, ML, tions towards 6G [61], such as communication, computation,
and blockchain, can be applied to DT modeling and deci- and simulation. The details are given in the following.
sion making while protecting the privacy of users. Based
on the various formed models and big data, DTs can assist 1) COMMUNICATIONS
with obtaining intelligent solutions to the problems in the Communications are fundamental to a network, and directly
application layer, such as task offloading and resource alloca- impact the performance of the network, including the
tion in IoT, vehicular network, SAGIN, healthcare, wireless throughput, transmission latency, security, etc. Usually, the
system, etc. communication resource in a network is limited. It is infeasi-
Furthermore, effective interfaces between different layers ble to attempt different options in a real network. Fortunately,
of the DITEN system are required to bridge the physical with DITEN, we can process various communication opera-
objects, twins, and applications [61]. Specifically, through tions in a virtual edge network, and then obtain the optimal
the twin-to-physical object interface, the real-time interaction operation parameters to feedback to the real network to gain
between the DT layer and the physical layer can be achieved. the optimal network performance, such as communication
The global information of the system can be obtained rate, bandwidth usage, security, etc. Several researches are
through the communications between twins with the twin- focusing on applying DT to improve the communication
to-twin interfaces. Moreover, through twin-to-application quality in a network. In [62], DITEN is applied to obtain the
interfaces, the applications such as IoT, vehicular networks, optimal communication resource allocation to alleviate the

1364 VOLUME 3, 2022

traffic congestion. In [63], the authors apply DT to connect Based on the above potentials, DITEN can be applied to
the physical and network layer in the optical communication assist in the deployment and operation of a network for var-
system, to improve the communication reliability and effi- ious applications. A network can be deployed, tested, and
ciency, and make the network management scheme better verified with the DTs of physical components in the digital
match the real physical situation. In [64], DT is employed domain using AI. In addition, after the network deployment,
to design an intelligent clock synchronization method in an the operational parameters of a network can be optimized
industrial IoT scenario. By predicting the clock behaviors to in the digital domain based on the real-time information
reduce time stamp exchange related to the synchronization, and prior knowledge stored in the DITEN. The system
thereby improving the clock accuracy and decreasing the performance can be optimized by checking all possible sce-
cost of communication. narios and conditions in the digital domain. Meanwhile, with
DITEN, a network’s disruption can be predicted, such as the
2) COMPUTATION failure of components in a network, then protect them before
Although MEC can be used to process some computa- happening, and respond to them proactively.
tion tasks and relieve stress on computation resources in
a network, the computation resources in the network are III. THE DESIGN OF DITEN
limited and still hard to handle intensive tasks to sup- In this section, we discuss the basic aspects of the design of
port the emerging applications such as VR and Metaverse. DITEN. We first present the DT modeling/updating. After
Recently, by integrating AI with MEC, the intelligent edge that, the deployment of the DT in the DITEN is illustrated.
has emerged, where different AI models are trained using Then, the key issues to be addressed during the design of
various ML methods, such as deep learning, and transfer a DITEN is discussed. Finally, we introduce the enabling
learning, which requires many computation resources to train technologies to overcome those issues.
and update the model. DITEN can be utilized to improve
the computation efficiency [27], [65]. In particular, by con- A. DT MODELING/UPDATING
structing a DITEN for a network, the physical object and AI DT modeling and updating are the essential parts to design
models are digitized in the virtual domain to process task a DITEN. Based on the real-time status of the physical
computation and model training, then the results will be objects, the DT of the physical object can be constructed
fed back to the corresponding physical objects. To this end, in the resource-rich end devices, and edge/cloud servers.
with DITEN, resource-limited devices and servers can also Meanwhile, the DTs will be updated synchronously by
process intensive computation tasks and AI model training. interacting with the physical objects, and feeding back
information to the physical objects, thereby achieving the
3) SIMULATION synchronization between the physical objects and DTs.
DITEN can be applied to simulate various tasks in different The construction of the DT model has been widely studied
scenarios, especially for the areas involving a large workforce in different applications, such as manufacturing [69], product
or intensive resource consumption, such as anomaly detec- quality monitoring [70], etc. The pilot methods and tech-
tion and autonomous driving. As mentioned before, DITEN niques to build DT models are presented in [69], [71]. Digital
has some potential for reliable and efficient communication simulation models of different service levels can be built in
and computation to achieve accurate and real-time data trans- the DT space. Moreover, model methods include mathemat-
mission and results feedback. Thus, DITEN can be applied ical model, 3D model, and data-driven model [30], [38].
to simulate the physical environment, obtain the optimal Using a mathematical model to represent a physical object
decisions, and predict the development trend of the system. requires several assumptions [72], and the model accuracy
Employed DT to simulate a system has attracted much atten- is limited, so as to 3D models. To this end, physical objects
tion in industry and academia. In [66], the DT constructs a can be efficiently modeled using ML based on data-driven
cellular automata-based road traffic simulator and an LTE-V models [73], [74]. While, it is a challenge to choose an
MEC network simulator, which is applied to evaluate and effective ML model, which consumes a significant amount
verify the obtained strategy to guide the lane-changing of of time [75], [76].
the automated vehicles. In [67], DT is utilized to develop a To construct a DT model, large-scale data should be trans-
security framework, which can guarantee the security of the mitted from physical objects to DTs, which poses a high
physical system in the smart grid intelligently. The proposed burden on communications and incurs data leakage in many
framework evaluates the system behavior and simulates the applications and also increases communication risks for the
complex attacks using the real-time collected information. physical object. To this end, recently, federated learning (FL)
In [68], the authors utilize DT to simulate the healthcare has been utilized for DT modeling, which consists of three
era, such as body physical indicator detection and disease steps. First, each physical object trains a local model based
prevention. By constructing a DT of a human body, the on its local database. Then, the physical object transmits its
medication can be simulated and the body reaction can be model parameters to the server. After that, the edge server
monitored, which can promote the area of drug research, yields a global model based on the received local model
epidemic treatment, disease prediction, etc. parameters, and then shares the global model parameters

VOLUME 3, 2022 1365

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

to physical objects for local model updating. The above

processes will continue until convergence performance is
achieved.

B. DT DEPLOYMENT
In DITEN, DTs can be deployed on various locations includ-
ing cloud, edge, and physical entity [61], depending on
the distinct requirements for different applications, such
as latency, physical experience quality, computing resource
requirement, and reliability, etc.
FIGURE 4. Digital twin migration in DITEN.
1) DEPLOYMENT ON THE CLOUD [79], [80]
The first way to deploy DTs is on the cloud, which has
4) HYBRID DEPLOYMENT [26], [27]
been applied in most cases. Cloud has large coverage and
rich available resource, it can ensure the stability of the The above three DT deployment methods have different char-
data transmission between the physical entities and DTs, acteristics. To obtain the optimal system performance, the
and satisfy the intensive computation requirements for DT above three DT deployment methods can be integrated to
modeling and updating. As such, deploying DTs on the cloud yield a hybrid DT deployment, such as deploying DTs both
suits the case where the physical entities move constantly, in edge and cloud. For example, in the DT based infotain-
and the DTs are called by multiple devices, such as vehi- ment system for autonomous cars, DTs are deployed both at
cles, wearable devices, etc. However, although the cloud has the edge and cloud for efficient caching. The DTs deployed
rich computing resources than the edge, it will lead to high at the edge make caching decisions for the users with crit-
latency and low context awareness. ical latency requirements. In addition, the cloud is used to
cache the information with relatively low usage frequency
2) DEPLOYMENT ON THE EDGE [81], [82] to address the limited caching capacity at the edge. Cloud
The DT system can be also deployed at the network edge. DTs are used to manage the caching in the cloud.
Since the limited coverage of the edge servers, this DT
deployment way more suits the case where the physical enti- 5) DT MIGRATION
ties are relatively fixed and have the requirement of latency, DT migration has the potential to provide constant services
such as some large-scale types of equipment. Moreover, for end devices by deploying DTs on the edge. Due to the
deploying a DT system at the edge, the performance of AI mobility of end devices, a device connected to a small BS
models for data processing and analysis is improved with may move to the coverage area of another BS, where the BS
the localized edge environment, and the security and syn- is installed with an edge server running DTs. The mobile
chronization of data transmission can be guaranteed with device can connect with the original edge servers via the
the fast and inexpensive edge connections [38]. Low latency core network with the assistance of a newly associated BS.
and more context awareness such as location and mobility of But it may lead to high inherent latency, and is not applicable
end devices can be achieved by deploying DT at the network for many applications with strict latency constraints. In addi-
edge compared with deploying DT at the cloud. Although tion, constructing a new DT system on the edge server on the
edge-based twins have many advantages to enable a variety newly associated small BS is time-consuming and resource
of applications, they are constrained by limited computing cost. To this end, it is promising to migrate the DTs to the
resources. newly associated small BS. Based on the location of the
mobile device, DTs should be migrated dynamically [82].
3) DEPLOYMENT ON THE PHYSICAL ENTITY [83] As shown in Fig. 4, the physical object moves from loca-
The third way is to deploy the DT system on the phys- tion A to C through B, the DT of the physical object is
ical entities directly. This way does not require the data migrated from edge server ES1, ES2, and ES3 accordingly.
exchange between the edge/cloud servers and physical enti- The DTs migration can be achieved efficiently by explor-
ties, thereby relieving the burden on communications, and ing ML schemes. However, the heterogeneity of edge servers
improving security by protecting the physical entities from makes the model migration challenging. The interoperability
malicious communications. In this case, external communi- among different edge/cloud servers may hinder the migration
cation access to the inter-twin communications through DTs, of DTs from one edge/cloud server to another. To simplify
and private intra-twin communications are used to forward the migration of DTs to address end device mobility issues,
the verified and secured data from the DT to the entities’ interoperable must be considered in the design of the servers.
execution units. While, the physical entities always have Easier migration can be achieved utilizing common archi-
limited computing capability and storage space, which may tecture, such as a unified cloud interface/cloud proxy, and
hinder the DT modeling and updating. open cloud computing interface [83].

1366 VOLUME 3, 2022

C. KEY ISSUES • Computation efficient: The computation efficiency can
There are some key issues to be addressed during the be improved from the following two aspects: i) Select
design of the DITEN, including efficiency, fault-tolerance, the nodes with relatively high computation speed for
low latency, and security. We will discuss these four issues DT modeling to improve the update efficiency; ii)
in detail in the following. Optimize computation resource allocation to improve
the computation efficiency.
1) EFFICIENCY • Incentive mechanism: Efficient incentive mechanisms
The first key issue is efficiency, including communications are also another way to improve efficiency. By moti-
efficiency and computation efficiency. In terms of commu- vating the resource-rich end devices and edge servers to
nications, as mentioned before, is fundamental to design the contribute their resource to model training, the model
DITEN. Large-scale data generated from the physical layer accuracy can be also improved.
need to be transmitted to the virtual layer for DT modeling
and updating, which increases the burden on the wireless 2) FAULT-TOLERANCE [22], [86]
communication link and the risk of congestion and data In practice, the DT model may not perfectly reflect the
leakage, thereby affecting the efficiency of the communica- physical objects, so the DT mappings have a bias from the
tions. Meanwhile, processing such large-scale raw data and actual values, and the performance of the service provided
constructing and updating DT models requires intensive com- by DITEN, such as optimization, prediction, and analysis,
putation resources. To solve this issue, FL can be applied to can be affected by the bias. In this case, we can improve
improve the efficiency of the design of DITEN [23], where the DT modeling method and increase the update frequency
end devices train their model based on the local data set, to minimize the bias. However, it is difficult to completely
and only need to transmit the model parameters to servers. eliminate the bias due to the constraints on technologies and
To this end, the data transfer volume can be greatly reduced, the limited available resource in terms of communication,
compared with transmitting all raw data. computation and caching. To this end, to relieve the impact
There are three steps for the FL-enabled DITEN design. of bias on the system performance, another way is to take
The first step is computation, in which each node makes the the bias into consideration during the design of the system
local training and builds model. The second step is com- to enable that the system is fault tolerant.
munications, where each node transmits model parameters In [84], a DITEN with fault tolerance is studied, where
to the server. The third step is aggregations, e.g., the edge different reputation values are endowed for devices based
server aggregates all parameters to yield a global model. on the bias of their DTs from the true values. The indus-
Taking constructing DT model at the edge server with FL trial devices’ DTs are constructed and maintained at the
as an example [23]. The main purpose of FL is to build edge servers, and the DT of device i can be denoted as
an ML model that evolutes with the physical devices’ rules DTi (t) = {F(ωit ), fi (t) + 
fi (t), Ei (t)}, where ωit , F(ωit ), fi (t)
and states. Let f (ω) be the loss function to quantify the dis- and Ei (t) are the model parameter, training state, computa-
tance between the estimated value by DT and the true value. tional capability and energy consumption of device i within
Users participating in FL train their model locally with the time t, respectively. It is worth noting that  fi (t) is the mapping
objective of minimizing the local loss. The BS collects and bias in terms of CPU frequency. In practice, different devices
aggregates local model parameters to yield a global model have different mapping biases, and the resource-constrained
with the objective of minimizing the global loss. industrial devices are collaborative to complete a complex
There are two processes for users participating in FL for task, such as applying FL. In this case, by setting beliefs
model training, namely local training and parameters trans- for different devices based on their mapping bias, the effect
mission. The execution time varies for different users due of the bias can be relieved. Specifically, the belief value is
to their heterogeneous resources in terms of communication negatively related to DT error. A lower belief is set to a
and computation. To aggregate the local models of all users, device with greater DT error, and a higher weight is set to
a BS with the MEC server acting as an aggregator, needs the parameters uploaded by a device with lower mapping
to wait for all the users including the slowest one to finish bias during the aggregation process.
their execution, which slows down the whole synchronous
scheme. With regard to this, we can improve the FL effi- 3) REAL-TIME [27], [59]
ciency through communication, computation, and incentive In DITEN, DT models are constructed and updated at edge
mechanism, i.e., servers in many cases. The data exchange and processing are
• Communication efficient: To improve the communica- time- and resource-consuming, especially for the large-scale
tion efficiency, we can further reduce the size of model data for constantly synchronization between the physical and
parameters to be transmitted to the aggregator to relieve virtual. Due to the heterogeneity of the end devices and edge
the communication burden and to improve the update servers, their available resource for communication, compu-
efficiency of FL. In addition, we can improve commu- tation, and caching are various. To this end, it is necessary
nication efficiency by optimizing resource allocation in to optimize the usage of the limited resources in the wire-
terms of communications. less network to improve the resource utility and decrease

VOLUME 3, 2022 1367

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

TABLE 2. Requirements of different communications.

the latency for DT modeling and updating. To this end, it is machine learning, blockchain, and incentive mechanisms,
necessary to optimize the usage of the limited resources in which are discussed in the following.
the wireless network.
An efficient way to improve the resource utilization is 1) COMMUNICATIONS
to optimize the association between end devices and edge Communications are fundamental to realize a DITEN,
servers based on their computation capabilities and commu- where large-scale data exchange is required to achieve
nication channel state. By establishing and updating DTs at DT modeling/updating and the synchronization between
the associated edge servers, the data and tasks for model the physical objects and their virtual twin in the entire
training can be adaptively distributed to different edge DITEN system. There are three kinds of communications in
servers. The edge association jointly considers end devices’ DITEN, namely physical-to-physical (P2P) communication,
datasets, the computation capability of edge servers, and physical-to-twin (P2T) communication, and twin-to-twin
the communication states between end devices and edge (T2T) communication, which are summarized in Table 2.
servers. Meanwhile, the learning accuracy can be improved The P2P communications achieve the interaction and shar-
by increasing the training batch size of DTs. However, the ing of information among physical objects, which include
large-scale training batch size requires more time to process the communications among end devices, communications
more computation tasks, thereby resulting in long latency. among edge/cloud servers, and communications between end
Moreover, the optimal resource allocation, such as band- devices and edge/cloud servers. The P2P communications
width, should also be studied to improve the efficiency of have strict requirements for their performance of reliability,
communications. In [58], the edge association problem is latency, and capacity. The P2T communications refer to the
researched to balance the learning accuracy and the learning information transmission between physical objects and DTs
time cost, where the associations between DTs and edge through wireless communication technologies. With P2T
servers, the training data batch size, and the bandwidth communications, the physical objects transmit their real-time
allocation are optimized. information to DTN for DT model construction and updat-
ing, and the DTN will feed back the results and instructions
4) SECURITY to the physical object. In P2T communications, a physical
Security is also a key issue in DITEN, including the security object first accesses the network through a BS, then con-
and privacy of the physical objects and data privacy preser- nects to the DT on the Internet. To achieve the real-time
vation in DT. With the large scale of physical objects and data exchange between the physical object and the DTs,
their distribution, how to resist unauthorized access to the there are some strict requirements for P2T communications,
physical objects, servers, DTs, and so on is one of the key including robust communication infrastructure, high reliabil-
problems in the design of DITEN. Moreover, how to guar- ity, low latency, high capacity, and guaranteed transmission
antee the security of interfaces between different layers of privacy and security. The T2T communications achieve the
the DITEN is also a key issue to be addressed, since large- data transmissions between DTs of physical objects in DTN,
scale data transmission is required for the DT modeling and aiming to mirror the communication behavior in the practical
updating, and physical-virtual synchronization. physical communication system. The T2T communications
Besides, security and privacy issues during data transmis- rely on the DT servers’ computation ability to model the data
sion for model training are also needed to be considered. In transmission behavior and do not need to consume any com-
the distributed model training case, some malicious physical munication resources such as spectrum and radio power. In
objects may transmit error information to mislead the behav- addition, compared with the P2P communications that need
ior of the edge server for global model training. Conversely, to consume a certain amount of time, the T2T communica-
some malicious edge servers may transmit mistake global tions only require a very short time to complete the same
model parameters to mislead the updating of the local model transmission behavior. To this end, T2T communications can
in each physical object. To this end, privacy leakage will be be utilized to imitate the long-term transmission process in
caused and need to be considered during the design of the a relatively short period of time, and the results will be sent
DITEN [85], [86], [87]. to the physical communications to guide the schedule of the
practical applications.
Due to the different communication environments, the
D. ENABLE TECHNOLOGIES above three communications have different requirements for
In this subsection, we present some technologies to enable their performance, such as reliability, latency, capacity, and
the DITEN, including communications, data processing, connectivity, which are summarized in Table 2. First, for

1368 VOLUME 3, 2022

reliability, since the real-time status information from the
physical object to its DTs, and the results feedback from
DTs to the physical object, are the essential information
to realize and maintain the DITEN, which poses strict
requirement for reliability for P2P and P2T communica-
tions. However, the data and model sharing among DTs
are not the key information for DT modeling and results
feedback, so the requirement for reliability in T2T com-
munications is relatively low. Second, for the transmission
latency, the DT modeling and the synchronization between
physical objects and DTs all require real-time data transmis-
sion, there are strict requirements on latency for P2P and
P2T communications. As mentioned before, T2T communi-
cation can complete a long period of transmission behavior
in a short time period, so the latency requirement on T2T FIGURE 5. Machine learning in digital twin edge network (DITEN).
communications is low. Third, for the capacity, the large-
scale data transmission is required in DITEN, which poses
strict requirements on the capacity for the three communica- technologies, such as deep reinforcement learning (DRL),
tions. Finally, for the connectivity requirement, considering FL, and transfer learning (TL), have been used for DT
the requirements for the quickly model updates, real-time construction/updating, DT deployment, physical-virtual syn-
predictions, and feedback, the P2T communications have a chronization, and decision making in DITEN, as shown in
high requirement for connectivity. Since the P2P and T2T Fig. 5.
communications are mainly used to transmit collaborative DRL: Deep learning (DL) algorithms are able to obtain
information, which poses relatively low requirements on the the near-optimal solutions in real-time as the problem scale
network connectivity although a large amount of information. increases [93]. DL algorithms require to acquire optimal
solutions from a simplified model first, which may be
2) DATA PROCESSING impracticable in a practical system. To this end, some
In DITEN, the raw data from the physical objects are technologies, such as DRL, are applied to enable DL algo-
large-scale, multi-source, and heterogeneous. Meanwhile, the rithms [94]. DRL learns from the feedback of the real
amount of data will increase at an exponential rate, as the environment, does not need any labeled training samples, and
expansion of the scale of DITEN. Those data are transmitted has been applied to solve a problem modeled as a Markov
to edge/cloud servers for DT modeling/updating and deci- Decision Processes (MDP). DRL can also be used to solve
sion making, and it is necessary to fully use those data to the problems with the characteristic of time-varying [95],
improve the performance of the DITEN and to reduce the [96]. DRL has been widely applied to optimize the task
cost of communication and computation. Moreover, the con- offloading problem in a wireless network. Specifically,
gestion will cause if a large amount of data is transmitted in [97], the authors employ a deep-Q network (DQN) to
directly through the wireless communication system, espe- propose a task offloading scheme in a vehicular network to
cially for the high volume of data. To address the above optimize the edge server selection and transmission model,
problems, and fully use the data, it is necessary to process aiming at achieving the maximum task offloading utility.
the data efficiently, such as data cleaning and data fusion, In [98], a double DQN is applied to obtain the optimal task
to handle data redundancy, data missing, data conflicts, etc. offloading scheme to minimize the energy consumption for
Specifically, since the data always has spatiotemporal corre- communication. With the same subjective, in [99], a deep
lation, which causes data redundancy. As such, processing deterministic policy gradient (DDPG) is utilized for task
data fusion to aggregate the redundant data is necessary to offloading and resource allocation.
reduce the scale of the transmitted data and to improve the DRL has been also applied in DITEN for DT
robustness and reliability of the data. modeling/updating, and to find the optimal solutions, such
as edge association, task offloading, and resource alloca-
3) MACHINE LEARNING tion. In [84], a DQN-based adaptive global aggregation
ML is one of the key enabling technologies of the frequency calibration method is designed to increase the
DITEN. Recent advances in ML enable the capabilities learning accuracy of FL with a limited resource budget and
to achieve environment sensing, make decisions, and deal to achieve the dynamic balance between the energy con-
with unsolved ever-complex problems in various applica- sumption on computation and communication in a dynamic
tions, such as vehicular networks [88], SAGIN [89], [90], wireless communication environment. In [93], a DRL in
[91], and others [11], [92]. ML makes the DITEN intelli- terms of asynchronous actor-critic (AAC) based algorithm
gent for information collection, decision making, and smart is developed to obtain the optimal decisions on task offload-
and efficient system management. Specifically, various ML ing. In [26], a DRL in terms of proximal policy optimization

VOLUME 3, 2022 1369

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

(PPO) is applied to obtain the optimal task offloading and from others performing similar or related tasks. TL mini-
resource allocation decision in a distributed blockchain envi- mizes the statistical distance between the source and target
ronment. In [27], DRL is applied to solve the task offloading domains in the feature space, by sharing model parame-
problem in a DT-enabled vehicular edge computing (VEC) ters between the source and target domains and considering
network. In [79], DQN with experience replay and target the transferability of the learned parameters. In [107], deep
network is employed to solve the service offloading problem transfer learning is applied in DT-assisted fault diagnosis.
in a DT-enabled Internet of Vehicles (IoV) network. In [80] To achieve this, a virtual model is built for the system first,
Double Deep Q-learning Network (DDQN) is utilized to and then when the built virtual model reaches acceptable
solve the resource allocation problem at the cooperative edge performance, based on the real-time states of the physi-
servers. cal object and knowledge learned from the virtual model, a
FL: The centralized ML algorithms have a high risk of diagnostic model is constructed and updated applying deep
data leakage since it requires data exchange between the transfer learning.
edge servers and end devices. In addition, in various appli- In [29], TL is exploited to address the DT migration
cation scenarios, synchronizing all raw data to DTs will problem. Considering the mobility of devices in DITEN,
result in over-communication load and data leakage, as such the efficiency of service provided from edge servers to end
the end users may have communication risk. To this end, devices is reduced, so it is challenging to construct and
FL as a promising decentralized ML framework is applied maintain DTs. When end users move from one location to
to strengthen data security and protect user privacy [32]. another location, synchronization of the original DT model
FL does not require the end devices to transmit their raw deployed in the edge server will incur high resource con-
data to the server. In FL, the end devices train their ML sumption, while communication over long distances will
model locally based on their database and only need to lead to high latency. To this end, it is necessary to transfer
transmit their local model parameters to the server, thereby the DT model from the original edge server to the newly
alleviating users’ privacy concerns and providing an ML selected edge server after users’ mobility, like DT mobility
framework for distributed users [100]. FL is a key tech- in the Metaverse. By migrating DTs among servers based
nology to enable the development of IoT [101], [102] and on the location of end devices applying TL, guaranteed and
6G [103] due to its distributed and privacy-preserving fea- high-quality services are provided to users while preventing
tures. Recently, in [23], FL is applied to promote privacy resource consumption and latency for new model training.
and security in DT modeling. Clients train their model and Applying TL for DT migration, the knowledge in original
transmit their trained model to the BS. After that, the BS servers can be reused to construct a DT of users in a target
builds DTs of client devices. server immediately.
However, during DT modeling, a BS that has DTs of
all physical devices may mislead users’ behavior models. In 4) BLOCKCHAIN
addition, malicious physical devices may affect DT modeling Blockchain can be considered as a tamper-proof distributed
by propagating fake data or low-quality models to the BS. database, which guarantees the security of a system in
Moreover, signal distortion and aggregation errors occur a distributed way by using cryptography. Blockchain has
since the constrained resource of the edge devices and the the advantage of decentralization, tamper-proof, anonymous,
random wireless communication environment, which affects publicly verifiable, and traceable. When distributed learning
the convergence speed and prediction accuracy of FL. To is utilized to train DT models, blockchain has the potential
this end, in [104], the authors minimize the loss function of to store the pre-trained models in an immutable and trans-
ML by selecting the users participating in FL and allocat- parent manner, so that those pre-trained models can be used
ing resources optimally. In [105], to motivate more users to to serve to request users in the future. Thus, blockchain
join in FL, an incentive mechanism is proposed based on can be integrated with DITEN to achieve guaranteed secu-
an auction game. In [28], the dynamic DT and FL in an rity, immutability trust, accessibility, and traceability of
air-ground network is studied. An incentive scheme based transactions [108].
on Stackelberg game is developed to motivate customers to There are several works that apply blockchain in DT.
join in model training collaboratively. In [106], an incentive In [109], blockchain is utilized for DT construction, where
scheme combining reputation and contract theory is proposed the transactions initiated by participants involved in DT con-
to motivate mobile users with high reputations to join in FL. struction are managed and tracked with smart contracts.
In [32], cooperative FL is designed to build DTs for resource- In [110], a reference model is proposed to authorize the
limited devices at the edge of the network, and consortium owners of home appliances’ DT located on fog to its
blockchain is employed to secure model updates both in owner, and apply blockchain and clouds to facilitate human-
global and local. Moreover, an iterative double auction is centric services and applications. In [108], blockchain is
developed to motivate APs to train local models and verify applied in DTs’ reshape and transformation process, aiming
updates. at achieving secure manufacturing with traceability guar-
TL: TL is a promising ML approach in which an agent antees, authenticity, compliance, quality, and safety. It is
solves a problem with the help of prior knowledge gained difficult to sustain a conventional blockchain in MEC since

1370 VOLUME 3, 2022

 In DITEN, the servers cooperate with end devices to con-
struct DTs applying distributed learning. Specifically, the end
devices train models locally, and then transmit their model
parameters to the servers for aggregation and to obtain a
global model. Then, models in end devices will be updated
based on the global model. Finally, continue the above pro-
cess until reaching a convergence. In this case, incentive
mechanisms can be designed with the Stackelberg game fol-
lowing a leader-follower manner. Specifically, the servers act
as leaders aiming at maximizing the global accuracy, and the
end devices act as followers aiming at maximizing the reward
in terms of money while minimizing the cost of communi-
FIGURE 6. Transfer learning for digital twin migration. cation and computation [111]. The more monetary reward
should be provided to the device with high local accuracy,
considering that more computation resources are consumed
the traditional blockchain schemes with the characters of to obtain better accuracy for the local model. In addition,
high latency and resource cost. contract theory and auction theory can be also used to design
Currently, blockchain has been applied to DITEN to incentive mechanisms to pre-train DT models [106], [112].
achieve a security system. In [32], blockchain is exploited to Besides, incentive mechanisms for network operators need
achieve a new DITEN framework to support secure DT estab- to be designed with game theory and auction theory. The goal
lishment. In [77], the permissioned blockchain is applied in of the network operators is to serve more users to maximize
an industrial IoT scenario to configure decentralized, trace- their profits, the end users aim to optimize their performance
able, and secure intelligent manufacturing systems (IMS). such as throughput. For twining operations, incentives for the
In [26], blockchain is introduced to manage edge nodes edge/cloud servers running DTs are essential. In addition,
and enable real-time computation with security and relia- blockchain is applied to store the pre-trained twin models
bility. In [58], a blockchain empowered FL framework is for future use in the DT-based architecture [113]. To this
presented to relieve the burden on real-time data process end, attractive incentives need to be provided for miners to
and security on the edge. In [60], blockchain is utilized to perform blockchain consensus algorithms and storage [114].
select the mobile edge servers. In [57], a light blockchain Moreover, attractive incentives need to be also offered to the
scheme is proposed for DITENs, which is further improved software-defined network (SDN) for efficient twinning [115].
to integrate with the FL process efficiently. It is known from
the above mentioned that the blockchain has the potential IV. THE TYPICAL APPLICATIONS OF DITEN
to conquer the current shortcomings in DITEN. However, In this section, we first present the typical services pro-
applying blockchain in DITEN is still needed to be studied. vided by DITEN, such as edge association, task offloading,
Researchers are working to achieve scalable and deploy- resource allocation, and security. Then, the typical appli-
able blockchain-based DITEN platforms. There are still cations of DITEN are discussed, including IoT, vehicular
some challenges to be addressed, like distributed consensus network, SAGIN, healthcare, wireless system, and other
methods and data analysis with privacy protection. applications.

A. TYPICAL SERVICES OF DITEN

5) INCENTIVE MECHANISMS
As mentioned in the previous sections, we know that
In DITEN, incentive mechanisms are required for various the DITEN can mirror the physical MEC system, monitor the
tasks, such as pre-training of DT models, physical object network parameters and models, such as the variety of the
control, and blockchain mining. Various physical entities act- available resources, and task arriving process, the DITEN
ing as players, such as network operators, end devices, and can also make predictions, simulate the physical system,
servers, are motivated to contribute their available resources and contributes optimization and analysis. In DITEN, the
to drive different services. historical information and the current system status can be
As mentioned before, distributed learning has been applied obtained. Based on this information, the optimal decisions in
for DTs, where end devices participate in local models train- a system can be made from a centralized viewpoint. Hence,
ing, and edge/cloud servers run DTs and aggregate local the networking schemes, such as DT modeling/updating,
models for global model training [86]. To this end, it is edge association, task offloading, resource allocation, secu-
necessary to reward them for their contributions to model rity mechanism, etc, can be directly designed and optimized,
training, so that to motivate more devices and edge/cloud thereby promoting network performance and reducing cost.
servers to participate in distributed learning to improve the • DT Modeling/Updating [32], [119], which mainly aims
learning accuracy. Note that specific incentive mechanisms to improve DT modeling/update performance, such as
should be designed for different players for various tasks. the model accuracy, latency, and cost.

VOLUME 3, 2022 1371

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

TABLE 3. DITEN applied in IoT scenarios.

• Edge Association [60], [94], [120], which includes the unified time for cooperative FL and local model update
association between end devices/DTs and edge servers. verification.
• Task Offloading [60], [94], [120], which aims to decide In [119], the authors focus on DT construction and
whether to offload a task and the proportion, thereby DT-assisted resource allocation. To balance the delay and
improving the task processing efficiency. accuracy performance, the iteration latency and DT’s loss
• Resource Allocation [94], [119], [120], which includes function are jointly optimized. In addition, abnormal model
the allocation of resources in terms of communication, recognition (AMR) is leveraged at the servers to protect the
computation, caching, and so on. edge aggregation from malicious devices, thereby improving
• Security Mechanism [32], [60], [119], [121], [122], the security of DT. A two-stage algorithm is proposed based
which aims to guarantee the security while satisfying on DQN, which learns the optimal computational resource
the performance requirements. allocation, device access management, and power control
scheme, by interacting with the environment constantly.
Recently, some existing works are focusing on designing Furthermore, DT can improve the learning performance by
and optimizing the network schemes. In [120], the authors predicting some key state information for intelligent DQN-
apply DITEN to study the computing tasks in the indus- based resource scheduling, such as the available resource of
trial automation scenario. By jointly optimizing the task devices and edge servers, and communication status.
offloading and resource allocation, the end-to-end latency is In [60], DT is applied to assist the mobile users in edge
minimized. In [121], the authors aim to promote data security server selection and task offloading. An edge cooperative
in industrial applications by utilizing DT and the correspond- node selection scheme is proposed to guarantee data secu-
ing trusted architecture. In [122], the security system and rity and provide high-quality communication links so that
mechanism are designed based on DT to guarantee the profit to settle the security problem of cooperative edge servers
of the edging computing, such as low latency, flexibility, and selection in DITEN. The security of the edge servers is
self-organization. verified using the blockchained data stored in DTs. The
In [32], the authors exploit blockchain to develop a reliable edge cooperative nodes can be the ones that passed
new DITEN framework to enable flexible and secure DTs data consistency authentication, or the ones with superior
establishment. First, to construct the DTs of the resource- communication link quality, which can be measured by the
constrained devices at the network edges, a cooperative real-time communication environment data in DT. Moreover,
FL method via AP is developed. Then, a model update DT can obtain the real-time status of the network, so that
chain is proposed by applying a directed acyclic graph the mobile users can offload the task to the selected edge
(DAG) blockchain to ensure the security of local/global server with low power expenditure and latency. In addition,
model updates. In addition, a joint cooperative FL and the decision tree algorithm (DTA) and DDQN are applied
local model update verification scheme is proposed based to solve the formulated mathematical optimization model to
on the iterative double auction, aiming at motivating optimize the energy consumption and network delay.
APs to contribute their resource to local model training In [94], the authors propose a DL framework to mini-
and updating verification for resource-constrained devices. mize the energy consumption on per bit data transmission
Meanwhile, social welfare is maximized by optimizing the of URLLC and delay tolerant services in a MEC system,

1372 VOLUME 3, 2022

by jointly optimizing the task offloading and resource allo- communication in an ever-changing communication environ-
cation. An optimal user association scheme is proposed ment. Moreover, an asynchronous FL framework is proposed
based on DNN, which instructs the physical entity after it is to adapt to the heterogeneous industrial IoT environment.
trained at a central server off-line. When the user association The proposed framework uses clustering nodes to over-
scheme is known, each AP optimizes the resource allocation come the straggler effect and employs a time-weighted
and task offloading policies. Moreover, to explore labeled inter-cluster aggregation scheme to improve the learning
training samples, a DT of the MEC system is constructed efficiency.
to mirror the real system, including the network topology, In [93], a DTN is proposed to map the industrial IoT and
wireless communication environment, and resource states. its digital system. Based on the real-time state information
Various user association strategies can be evaluated in the of IoT devices and BSs, the virtual model of IoT entities
DITEN to obtain the corresponding network performance, can be constructed, which contains the network topology
and the scheme with the optimal performance is stored and random task arrival process. The edge association, and
in the memory as a labeled training sample. In addition, resource allocation in terms of bandwidth, transmit power,
a global and low complexity resource management algo- and computation resource are jointly optimized to maximize
rithm is proposed along with the user association policy. the finished computation tasks with the minimum energy
The proposed resource management algorithm is also used consumption. The optimal solutions to edge association and
for performance evaluation in the DT. resource allocation are acquired by employing the Lyapunov
In the following, we present the services of DITEN in its optimization technique and DRL-based algorithm in terms
typical applications. of AAC.
In addition, blockchain technology is integrated into
DITEN to address the security issue in IoT networks. In [26],
B. IOT the authors employ blockchain technology to develop a hier-
Recently, DITEN is leveraged to tackle various problems archical digital twin IoT system, where DT and MEC are
in IoT networks, including task offloading and resource combined, and blockchain technology is introduced to man-
allocation, DT modeling, security, and network deployment. age edge servers. The proposed system aims to address the
In [123], an end-to-end DT conceptual model is proposed problems in IoT applications including security, accuracy,
to represent the physical objects from the ground to the air, reliability, and sustainability. A data- and knowledge-enabled
which includes multiple layers from down to up, such as learning method is proposed to achieve efficient optimization
the physical layer, communication layer, virtual layer, etc. and real-time interaction among the physical and the digi-
The characters of each layer are presented, and the hard- tal worlds. Meanwhile, the computation and communication
ware and software technologies used for model construction allocation policy is designed based on PPO to minimize the
are discussed. In [47], the authors present the overarching delay and energy consumption of communication and com-
framework for developing DTs combined with industrial IoT putation, and improving the reliability of the system and
technologies to advance the autonomy of aerospace plat- the learning accuracy in the system. In [77], permissioned
forms. In addition, the function of data fusion in the DT blockchain is integrated into the IoT scenario and a novel
framework in predictive maintenance for aircraft is also manufacturing blockchain of things (MBCoT) architecture is
discussed and identified. designed to configure a traceable, secure, and decentralized
In [57], an architecture of DITEN is proposed to optimize intelligent manufacturing systems (IMS). A data-driven and
the industrial IoT network efficiently and appropriately, by knowledge-driven DT of the system is constructed to guide
integrating DT with MEC. The DT models of the indus- the operation of the IMS. Finally, an MBCoT prototype
trial IoT devices are constructed based on their running system is implemented.
data exploiting FL. Furthermore, an asynchronous model Besides, in [116], a realistic setup for a DT of IoT deploy-
update policy is proposed to improve communication effi- ment is designed to measure the impact of IoT networks and
ciency during the DT model’s construction. In [84], the compute slices on the physical resources of edge comput-
estimation deviations of DTs from their true values are ing hosts. The proposed DT setup can emulate the broker’s
seriously considered. The authors first introduce DTs to behavior, and massive real and emulated devices, thereby
map the physical objects to the digital world in real- measuring the network resource utilization and its varia-
time, including their operation states and behaviors. Then, tion over time. In [117], a semantic DT-based IoT model
in the trust-weighted aggregation strategy, the estimation is proposed, which mimics the IoT devices and enhances
deviations are considered to quantify the contribution of the interoperability of the heterogeneous IoT devices with the
each device to the global aggregation of FL, so that to support at the edge with container technology. With DT, the
promote the reliability and accuracy of training models. processing power and storage capacity at the edges through
Furthermore, an adaptive global aggregation frequency cal- docker images of connected IoT devices can be enhanced.
ibration is designed based on DQN, which minimizes the Meanwhile, the interoperability among devices and appli-
loss function of FL with constraints on resource budget, cations can be improved. The semantic DT model, along
and balances the energy consumption on computation and with microservices, can simplify the orchestration of various

VOLUME 3, 2022 1373

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

IoT devices and applications. In [118], the concept of edge real-time traffic data in IoV applications, aiming at optimiz-
computing based smart manufacturing with distributed con- ing network resource allocation and alleviating the potential
trol structure is demonstrated. The physical entities act as traffic congestion during peak periods. Applying DT to the
edge devices, the concepts of holons and digital agents are task offloading problem in IoV is also studied in [78], and
combined with local DTs to achieve local control, decision the optimal solutions are acquired by applying DRL. DT
making, and optimization. The global digital agent connected can be considered as a software agent located in the cloud,
to a global DT globally controls the network of manu- with which the global network information can be obtained
facturing nodes, which can be used for the optimization through the information exchange between DTs, and the
of the production order and logistics, local digital agents, historical states of a vehicle can be obtained through the
and DTs. communications within the twin. In [78], the system states
are captured and analyzed by employing DT and DRL. The
cost of renting cloud servers and bandwidth, time-varying
C. VEHICULAR NETWORK communication conditions, and the available computation
DITEN is a promising technology to achieve autonomous resources of MEC servers are jointly considered. A task
vehicles, which has attracted much attention from academia prediction module is designed to predict the task arriving
and industry. For task offloading, authors in [65] employ process, so that to reserve computation resources for the
DT technology and AI to design an vehicular MEC system. upcoming tasks, thereby solving the task queue overflow
Cooperative task offloading strategy for vehicles is proposed problem. The task offloading process is modeled as an MDP,
to minimize the offloading costs. An edge management where the state is obtained through the DT, and then as the
framework is developed to improve the learning efficiency of input of DRL. The subjective is to optimize the system utility
multi-agents with DT while enhancing the synchronization in terms of latency, energy consumption, and rent cost.
between DT and the physical system. Moreover, the potential Besides, in [124], DT is applied to obtain the optimal
cooperation among vehicles can be revealed by leveraging caching strategy in a VEN. By leveraging DT technology,
DT, and multi-agents learning groups can be formed adap- the edge caching system is mapped into virtual space, which
tively, thereby improving resource utilization and reducing facilitates the construction of social relation model to cater
learning complexity. In addition, a distributed strategy based to the complex and dynamic social characteristics of vehi-
on multi-agents learning is proposed to minimize the task cles. Based on the social relation model, a vehicular cache
offloading costs of vehicles under strict delay constraint cloud concept is proposed to incorporate the content cache
and adjusts the mapping mode of the DT network’s state dependencies among different vehicles in various traffic envi-
dynamically. ronments. Then, an optimal social-aware caching strategy is
In [27], the authors first introduce the framework of the proposed, which dynamically coordinates the caching capa-
VEC network and the key issues such as task offloading, bilities of RSUs and vehicles based on the similarity of
communication, and caching in the VEC network. Then, user preferences and service availability, and jointly consid-
an adaptive DT-based VEC network consisting of two AI- ers social model establishment, cache cloud formation, and
empowered closed loops is proposed for DT construction cache resource management.
and VEC network management. In addition, a DRL-based In [125], a DT framework is proposed for connected vehi-
intelligent method is designed to obtain the optimal VEC cles with vehicle-to-cloud (V2C) communication, to enable
offloading decisions to minimize total task execution time. the driver to control the vehicle more smartly. In [126], a sen-
In [79], the authors propose a DRL-based service offloading sor fusion method is proposed to prevent potential dangers,
method in a multiuser offloading system to minimize services which draws and matches the bounding box of the target
response time. The authors first analyze the response time vehicle by combing the camera images and DT knowledge.
in a multiuser offloading system, and then model the edge In [66], DT is employed in MEC framework designed to
devices as the agent. Finally, the optimal services offloading achieve safety and intelligent connections among vehicles,
strategy is obtained by applying a DQN to minimize the and the lane-changing of automated vehicles (CAV). The
response time. In [80], the latency minimization in a DT- DT of a MEC network enables the network information to
IoV framework is studied. The edge servers are modeled as be mapped to a coupled road traffic and wireless network
M/M/1/N queues, and the expression of the response time simulator. To this end, the CAVs can learn the optimal solu-
of offloading tasks is formulated mathematically. Then, the tions to lane-changing through a visionary smart method,
optimization model is constructed to minimize the response where self-interest and traffic flow efficiency improvement
time, and DDQN is employed to train the edge server to get are jointly taken into consideration. The established DT
the optimal solutions to the task offloading action. can be used to evaluate and verify the obtained strategy
The above works provide a variety of useful and promis- and then apply the strategy in the physical system for
ing task offloading solutions in the IoV era. However, CAVs to make decisions on lane-changing. In addition, the
some of the existing solutions consider a static environment authors present a case study, in which the DT is composed
and do not consider the task arrival in the coming time. of a cellular automata-based road traffic simulator and an
In [62], the authors apply DT technology to predict the LTE-V MEC network simulator. The lane-changing strategy

1374 VOLUME 3, 2022

TABLE 4. DITEN applied in vehicular network.

is trained with DRL based on the aggregated lane state for resource allocation, in which the leader is the DT of vehi-
information. cles or roadside units, and the followers are the RSUs, and
the objective is to jointly optimize the vehicles’ satisfaction
D. SAGIN
and roadside units’ overall energy efficiency. By establish-
In [81], the authors apply DT to catch the network dynam- ing the dynamic DT for an aerial-assisted IoV, the dynamics
ics in air-ground networks. The drones acting as aggregators of resources in terms of requirements and provide in the
collaboratively complete the model training with the ground network can be captured. The computation task of vehi-
clients such as vehicles and smartphones in FL. In addition, cles that out coverage of ground network can be offloaded
the stackelberg game based incentive strategy is designed for to the equipment like RSUs with rich resources, with the
FL, where the drone’s DT acting as a leader sets the pref- assistance of UAV. By establishing the DTs of vehicles and
erences of the client based on their reputation values, and RSUs at UAV, and keeping updating when the vehicles and
the clients acting as followers decide the training rounds RSUs are out of UAV’s coverage, the unified and efficient
for the global model in the game after trading off prof- resource management can be achieved in an aerial-assisted
its and costs. In addition, another incentive mechanism is IoV network.
proposed to adjust the optimal client selection and its par- In [127], the authors focus on the task scheduling strategy
ticipation degree in each global update, thereby adapting to in a UAV-assisted MEC system. The system dynamics and
the varies of air-ground networks. There are some advan- time-varying task arrivals due to the mobility of UAVs and
tages to the drone acting as the leader. First, since DTs hold mobile users (MUs) are jointly considered. The interactions
the clients’ state information such as available resources so among MUs are modeled as a stochastic game, and a proac-
that clients with high performance can be chosen to join tive DRL scheme is derived to obtain a nash equilibrium
the FL. Second, the direct communications between the (NE), which only uses the local information of each MU
clients and the drone and the corresponding communica- for local and remote computation scheduling. The homoge-
tion costs can be reduced by selecting the drone’s DT as the neous behaviors of MU enable to train the proposed scheme
leader of the Stackelberg game. Third, the state information offline using DT. In [128], the military large-scale UAVs
of each client is not required during the curator making are presented including their characteristics and advantages,
decisions, by deploying the client’s DT on the resource- and the basic problems currently encountered. Then, based
rich ground clients. Finally, the leader’s DT captures the on its design, manufacturing, and application, the urgent
real-time status and the training rounds of the participating to construct a cloud computing based DT framework for
clients, which can make sure the clients finish the training military large-scale UAVs is analyzed, which is further dis-
rounds. cussed from test cost, integrated sensing, centralized control,
In many existing works, it is assumed that the device business prediction, and mission planning.
mobility and service requests are predictable. As such,
In [28], DT is utilized in an aerial-assisted IoV network, E. HEALTHCARE
to catch the dynamics of the resource requirements. A DT is an important technology in the health domain. The
Stackelberg game based incentive mechanism is developed application of IoT makes it possible to monitor a patient’s

VOLUME 3, 2022 1375

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

TABLE 5. DITEN applied in SAGIN.

health status comprehensively by deploying massive intel- is integrated into a 6G wireless system, and a blockchain-
ligent human body monitoring equipment and environment enabled FL framework is proposed to relieve the burden of
sensors. A patient twin can be established to gather the edge servers on real-time data processing and privacy risk.
time-varying information of the patient, including the phys- In [29], a DT model for a 6G network is designed, where the
iological status and lifestyle, medication information, and DT is employed to ease the unreliable long-distance com-
emotion, to enable the doctors to provide comprehensive munication between the end devices and edge servers. The
medical care for patients and to predict the disease condition authors formulate an adaptive DT deployment problem to
changes to prevent disease worse in advance. For example, associate DTs with edge servers, the objective is to reduce the
by constructing a DT for a patient, a personalized recovery average latency and improve the user utility in DT-enabled
plan can be tailored based on the patient’s real-time physical edge networks. Then, a DRL-based algorithm is proposed
signs, which can improve the patient’s physique and faster to obtain the optimal solutions to the formulated problem,
the recovery time. where the DT placement scheme and the system delay are
Apart from detecting patient signs continuously, DT has jointly considered. Furthermore, a TL based DT migration
some potential for remote surgery. Experts can operate on method is proposed to adapt to the user mobility.
DTs of patients and control the surgery based on real- DITENs can be also applied to support interoperability
time body information. Meanwhile, based on the real-time in the network era, and reduce heterogeneity by designing
information, emergencies that may arise during operation an application drive layer on the top of physical equip-
can be predicted, and the corresponding optimal solutions ment. In [59], the Application-driven digital twin networking
can be obtained in advance. In this case, ultra-fast and ultra- (ADTN) middleware is designed to support the interac-
reliable communications are required to ensure information tions among the simplified, distributed, and heterogeneous
exchange. Furthermore, DT can be applied to develop twin industrial equipment, and to manage the network resource
organs with high precision and sensitivity. dynamically in a distributed industrial environment from
In [30], a DT framework is proposed for an intelligent an application point of view. In particular, the interactions
context-aware healthcare system, to improve the healthcare among various equipment are simplified by enabling DTs
process of patients and healthcare operations. The proposed to use IP protocols, and the network resource is man-
DT framework constructs a patient twin with IoT devices, aged dynamically by applying software defined network to
data analysis, and AI, and can collaborate healthcare profes- develop communication mechanisms that suit application
sionals effectively. Accordingly, an electrocardiogram (ECG) needs [132]. To this end, the management of the indus-
heart rhythms classifier model is proposed applying ML to trial network can be improved greatly, and then configure
process heart disease diagnosis and heart problems detection the topology structure faster and safer, so that the same
with high accuracy. In [129], a Cardio Twin architecture production site can be exploited for different industrial
runs on the Edge is designed to detect ischemic heart applications.
disease. In [130], the congestion control strategy is proposed for
DTENs applying Lyapunov optimization, which does not
require the prior information of the system, and converts
F. WIRELESS SYSTEMS the long-term congestion control problem to multiple online
DT technology has been applied to wireless systems to edge association problems in each time slot. Then, a long-
improve the network performance [22], [131]. In [113], term incentive mechanism is proposed based on contract
key requirements for implementing DT to design 6G are theory, which jointly considers the service delay, individual
presented firstly. Then, the authors introduce the architecture rationality (IR), and incentive compatibility (IC) of physi-
components and trends for different types of twins, such as cal entities. The optimal contracts are derived to maximize
edge-based twins, cloud-based-twins, etc. In addition, various the profit of the service provider. In [22], DT is applied
types of twins are compared, and providing some potential to estimate the edge server status and to provide training
directions and the corresponding guidelines. In [58], DT data for DRL agent, during the task offloading of mobile

1376 VOLUME 3, 2022

TABLE 6. DITEN applied in wireless systems.

users in DITEN. In addition, the impact of estimation devi- V. CHALLENGES AND OPEN ISSUES
ations of DTs from their true value on task offloading is In the following, the challenges and open issues are
explored. Then, the optimal solutions to the task offloading presented, including high-precision modeling, physical-
are obtained by applying Lyapunov optimization and DRL, virtual synchronization, DT migration for mobility, and
to minimize the offloading latency with the constraints on security and privacy.
the total migration cost.
A. HIGH-PRECISION MODELING
G. OTHERS APPLICATIONS In DITEN, the precision of the DT modeling directly decides
In [31], the authors apply the DT technology to a bat- the services provided by the DITEN, such as task offloading,
tery management system (BMS). First, a cloud collaborative resource allocation, network management, and so on. It is
architecture is proposed to break through the constraints challenging to model the physical objects in high precision
on computation capability and storage space in the conven- due to the communication and computation constraints. To
tional BMS, by exploring the massive computing and storage overcome this issue, a large scale of real-time information
capacity of cloud-based servers. Then, the DTs of the batter- on the objects should be provided and further processed,
ies are constructed in the cloud based on the real-time data and it is also required to design some efficient modeling
of the batteries, and the two-way dynamic mapping between methods, which poses some challenges to communications
the batteries and their DTs can be achieved. The refinement and computation. On one hand, it has strict requirements
of safety management of the battery life cycle can be real- for wireless communication systems in terms of ultra-high
ized, by online learning and model updating to overcome capacity and reliability, to support the large-scale data trans-
the shortage of using fixed model parameters in traditional mission in P2T communications. To do so, some advanced
BMS. In addition, based on the accumulated data of batter- technologies, such as Terahertz communication, communi-
ies and their DTs, the optimal system performance upgrade cation with an IRS, and AI, can be integrated into the design
route is constructed based on the intelligent on-the-air remote of the wireless communication system. On the other hand,
program upgrade technology. the computation capability is constrained by the available
In [133], an intelligent edge-based DT model is proposed resources. Thus, it is urgent to design some lightweight and
for Robotics, where robots offload computation and anal- tiny compute methods to process the computation task.
ysis models to the edge, and even towards the cloud. Recently, distributed learning such as FL-based DT
The proposed model requires the support of 5G connec- modeling has been widely investigated, in which the end
tivity and facilitates the enhanced automation and control devices train the model locally, and then send the local model
by a set of intelligent functions, such as task learn- parameters to the server for aggregation to train the global
ing, prediction, and optimization. In [134], a concept of model, thereby decreasing the scale of transmitted data. In
edge-based DT is presented to assess the ecological sus- this case, how to incentive the physical objects and servers
tainability of a cross-company production network. The to contribute their resource to DT modeling is a challenge.
sustainability indicators are computed locally, which can Another challenge is to balance the modeling precision and
reduce the data communications. To this end, the sustain- cost in terms of latency. Since the end devices have differ-
ability of a manufactured product can be traced and does ent communication and computation time, it will cost a long
not require suppliers to publish sensitive data or domain time on waiting for local model parameters of all physical
knowledge. Besides, in [135], DT is applied to proposed objects for global aggregation. A promising method is to
a cloud-edge collaborative architecture for FDM additive select a part of physical objects for model training. To this
manufacturing. The system development and architecture end, how to select the participating physical objects [104],
components of the system are elaborated, thus providing thereby balancing the modeling precision and waiting time
a guideline for other manufacturing resources in cloud is a challenge, which requires a lot of careful follow-ups
manufacturing. research work.

VOLUME 3, 2022 1377

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

B. PHYSICAL-VIRTUAL SYNCHRONIZATION it is challenging to design an efficient authentication scheme

In DITEN, the virtual twin receives data and controls its to resist unauthorized access to the physical objects, servers,
corresponding physical object. Since the real world environ- DTs, and so on. In addition, the DT modeling and updating,
ment is dynamic, the state of physical objects is always and the physical-virtual synchronization, require large-scale
time-varying. It is essential to synchronize the physical data exchange between the physical objects and DTs through
objects and their virtual twin to provide services in different the interfaces. How to guarantee the security of interfaces
applications. Real-time and reliable two-way communication between different layers of the DITEN is also a challenge.
for large-scale data is the foundation to achieve physical- Apart from the interfaces, there are still some other security
virtual synchronization. The latency during DT building and and privacy issues during the DT modeling. In the central-
updating includes three parts, namely data transmission over ized model training case, all physical objects transmit their
wireless links, data transmission between edge servers, and information to a centralized server for model training, which
DT model computation/training. To reduce the latency during will cause security and privacy issues during data transmis-
the transmission, apart from the advanced communication sion, and model training and integrating blockchain into the
technologies mentioned before, routing algorithm, conges- DITEN is a promising method. In the distributed model train-
tion control, and communication resources allocation are ing case, physical objects train their local model based on
also significant to improve the communication efficiency, their local information and then transmit their local model
and need to be further investigated to cater to the DITEN. parameters to a server for aggregation to train a global model.
Since the physical objects do not need to transmit their raw
C. DT MIGRATION FOR MOBILITY data to the server, the distributed method can improve the
Due to the mobility of the physical objects and the limited security and privacy problem to some extent compared with
coverage range of BSs/edge servers, when a physical object the centralized case. However, there may be some malicious
moves from one location to a new location, it will cause high physical objects that transmit error information to mislead
latency during the physical object communicates with the the behavior of the edge server for global model training.
original connected BS/edge server, or even cause connection Conversely, some malicious edge servers may transmit mis-
outage when out of the coverage range. In this case, some take global model parameters to mislead the updating of the
DTs are deployed on the edge server installed on the BS local models in each physical object. To this end, privacy
in DITEN, thus the mobility of the physical object will leakage will be caused and need to be considered during the
have impact on the real-time interaction between the physical design of the DITEN, which is challenging and need further
object and its DT. A feasible solution to this issue is to investigated in the further research.
place a DT of the physical object on a new BS/edge server
near the physical object. However, constructing a new DT VI. CONCLUSION
is time-consuming, the physical object may move to another This paper has comprehensively surveyed the research in the
new location when its DT is constructed. Meanwhile, DT area of DITEN. In particular, we first gave the basic aspect
modeling costs intensive communication and computation of DITEN towards 6G, including its concept, framework, and
resources, and frequent construction of new DTs will pose potential. Then, the aspects of the design of DITEN were
great pressure on wireless communication systems. Thereby, presented, such as DT modeling, DT deployment, key issues,
it is a challenge to migrate DTs to cater to the mobility of and enabling technologies. After that, we comprehensively
physical objects. discussed the typical applications of DITEN towards 6G,
Recently, TL is applied in DT migration to save time and including IoT, vehicular network, SAGIN, healthcare, wire-
network resources [29]. However, the heterogeneous hard- less systems, and other applications. Finally, some challenges
ware/software designs of different edge servers may hinder and open issues related to DITEN were elaborated.
the transfer learning designed in DITEN. Transferring an ML
model from one server to another may face interoperability REFERENCES
problems. Besides, the selection of a new BS/edge server [1] G. Gui, M. Liu, F. Tang, N. Kato, and F. Adachi, “6G: Opening new
is another issue to be addressed. To achieve the optimal horizons for integration of comfort, security, and intelligence,” IEEE
Wireless Commun., vol. 27, no. 5, pp. 126–132, Oct. 2020.
performance, such as the service time for the physical object, [2] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems:
the achieved profit, the complex system conditions, the Applications, trends, technologies, and open research problems,”
mobility and operation states, and the availability of various IEEE Netw., vol. 34, no. 3, pp. 134–142, May/Jun. 2020.
[3] F. Tang, X. Chen, M. Zhao, and N. Kato, “The roadmap
BSs/edge servers should be jointly taken into consideration, of communication and networking in 6G for the metaverse,”
which is an open issue need to be further investigated in the IEEE Wireless Commun., early access, Jun. 24, 2022, doi:
further researches. 10.1109/MWC.019.2100721.
[4] Z. Ding and H. V. Poor, “Design of THz-NOMA in the pres-
ence of beam misalignment,” IEEE Commun. Lett., vol. 26, no. 7,
D. SECURITY AND PRIVACY pp. 1678–1682, Jul. 2022.
In DITEN, it is challenging to guarantee the security and [5] Z. Ding et al., “A state-of-the-art survey on reconfig-
urable intelligent surface-assisted non-orthogonal multiple
privacy of physical objects and data privacy preservation in DT. access networks,” Proc. IEEE, early access, May 23, 2022,
Due to the large scale of physical objects and their distribution, doi: 10.1109/JPROC.2022.3174140.

1378 VOLUME 3, 2022

[6] Z. Ding, J. Xu, O. A. Dobre, and H. V. Poor, “Joint power and time [28] W. Sun, P. Wang, N. Xu, G. Wang, and Y. Zhang, “Dynamic digital
allocation for NOMA–MEC offloading,” IEEE Trans. Veh. Technol., twin and distributed incentives for resource allocation in aerial-
vol. 68, no. 6, pp. 6207–6211, Jun. 2019. assisted Internet of Vehicles,” IEEE Internet Things J., vol. 9, no. 6,
[7] H. V. Nguyen et al., “On the spectral and energy efficiencies of pp. 5839–5852, Apr. 2022.
full-duplex cell-free massive MIMO,” IEEE J. Sel. Areas Commun., [29] Y. Lu, S. Maharjan, and Y. Zhang, “Adaptive edge association for
vol. 38, no. 8, pp. 1698–1718, Aug. 2020. wireless digital twin networks in 6G,” IEEE Internet Things J., vol. 8,
[8] Z. Ding, D. Xu, R. Schober, and H. V. Poor, “Hybrid NOMA offload- no. 22, pp. 16219–16230, Nov. 2021.
ing in multi-user MEC networks,” IEEE Trans. Wireless Commun., [30] H. Elayan, M. Aloqaily, and M. Guizani, “Digital twin for intelli-
vol. 21, no. 7, pp. 5377–5391, Jul. 2022. gent context-aware IoT healthcare systems,” IEEE Internet Things
[9] Z. Ding, R. Schober, and H. V. Poor, “No-pain no-gain: DRL J., vol. 8, no. 23, pp. 16749–16757, Dec. 2021.
assisted optimization in energy-constrained CR-NOMA networks,” [31] Y. Wang, R. Xu, C. Zhou, X. Kang, and Z. Chen, “Digital twin
IEEE Trans. Commun., vol. 69, no. 9, pp. 5917–5932, Sep. 2021. and cloud-side-end collaboration for intelligent battery management
[10] N. Kato, B. Mao, F. Tang, Y. Kawamoto, and J. Liu, “Ten chal- system,” J. Manuf. Syst., vol. 62, pp. 124–134, Jan. 2022.
lenges in advancing machine learning technologies toward 6G,” IEEE [32] L. Jiang, H. Zheng, H. Tian, S. Xie, and Y. Zhang, “Cooperative
Wireless Commun., vol. 27, no. 3, pp. 96–103, Jun. 2020. federated learning and model update verification in blockchain
[11] F. Tang, B. Mao, Y. Kawamoto, and N. Kato, “Survey on machine empowered digital twin edge networks,” IEEE Internet Things J.,
learning for intelligent end-to-end communication toward 6G: From vol. 9, no. 13, pp. 11154–11167, Jul. 2022.
network access, routing to traffic control and streaming adaption,” [33] B. R. Barricelli, E. Casiraghi, and D. Fogli, “A survey on digital twin:
IEEE Commun. Surveys Tuts., vol. 23, no. 3, pp. 1578–1598, 3rd Definitions, characteristics, applications, and design implications,”
Quart., 2021. IEEE Access, vol. 7, pp. 167653–167671, 2019.
[12] F. Tang, Y. Kawamoto, N. Kato, and J. Liu, “Future intelligent and [34] A. Fuller, Z. Fan, C. Day, and C. Barlow, “Digital twin: Enabling
secure vehicular network toward 6G: Machine-learning approaches,” technologies, challenges and open research,” IEEE Access, vol. 8,
Proc. IEEE, vol. 108, no. 2, pp. 292–307, Feb. 2020. pp. 108952–108971, 2020.
[13] B. Mao, F. Tang, Y. Kawamoto, and N. Kato, “AI mod- [35] M. Farsi, A. Daneshkhah, A. Hosseinian-Far, and H. Jahankhani,
els for green communications towards 6G,” in IEEE Commun. Digital Twin Technologies and Smart Cities. Heidelberg, Germany:
Surveys Tuts., vol. 24, no. 1, pp. 210–247, 1st Quart., 2022, Springer, 2020.
doi: 10.1109/COMST.2021.3130901. [36] Y. Wu, K. Zhang, and Y. Zhang, “Digital twin networks: A survey,”
[14] R. Minerva, G. M. Lee, and N. Crespi, “Digital twin in the IoT IEEE Internet Things J., vol. 8, no. 18, pp. 13789–13804, Sep. 2021.
context: A survey on technical features, scenarios, and architectural [37] J. M. Taylor and H. R. Sharif, “Leveraging digital twins to enhance
models,” Proc. IEEE, vol. 108, no. 10, pp. 1785–1824, Oct. 2020. performance of IoT in disadvantaged networks,” in Proc. IEEE Int.
[15] F. Tao, H. Zhang, A. Liu, and A. Y. Nee, “Digital twin in indus- Wireless Commun. Mobile Comput. (IWCMC), 2020, pp. 1303–1308.
try: State-of-the-art,” IEEE Trans. Ind. Informat., vol. 15, no. 4, [38] A. Rasheed, O. San, and T. Kvamsdal, “Digital twin: Values,
pp. 2405–2415, Apr. 2019. challenges and enablers,” 2019, arXiv:1910.01719.
[16] M. Singh, E. Fuenmayor, E. P. Hinchy, Y. Qiao, N. Murray, and [39] S. Khan, T. Arslan, and T. Ratnarajah, “Digital twin perspective of
D. Devine, “Digital twin: Origin to future,” Appl. Syst. Innov., vol. 4, fourth industrial and healthcare revolution,” IEEE Access, vol. 10,
no. 2, p. 36, 2021. pp. 25732–25754, 2022.
[17] K. Y. H. Lim, P. Zheng, and C.-H. Chen, “A state-of-the-art survey of [40] X. Zheng, J. Lu, and D. Kiritsis, “The emergence of cognitive digital
digital twin: Techniques, engineering product lifecycle management twin: Vision, challenges and opportunities,” Int. J. Prod. Res., to be
and business innovation perspectives,” J. Intell. Manuf., vol. 31, no. 6, publihsed.
pp. 1313–1337, 2020. [41] M. Liu, S. Fang, H. Dong, and C. Xu, “Review of digital twin about
[18] Z. Wang et al., “Mobility digital twin: Concept, architecture, case concepts, technologies, and industrial applications,” J. Manuf. Syst.,
study, and future challenges,” IEEE Internet Things J., early access, vol. 58, pp. 346–361, Jan. 2021.
Mar. 2, 2022, doi: 10.1109/JIOT.2022.3156028. [42] T. Deng, K. Zhang, and Z.-J. M. Shen, “A systematic review of a
[19] S. Boschert and R. Rosen, “Digital twin—The simulation aspect,” digital twin city: A new pattern of urban governance toward smart
in Mechatronic Futures. Heidelberg, Germany: Springer, 2016, cities,” J. Manag. Sci. Eng., vol. 6, no. 2, pp. 125–134, 2021.
pp. 59–74. [43] R. Saracco, “Digital twins: Bridging physical space and cyberspace,”
[20] O. E. Marai, T. Taleb, and J. Song, “Roads infrastructure digital Computer, vol. 52, no. 12, pp. 58–64, 2019.
twin: A step toward smarter cities realization,” IEEE Netw., vol. 35, [44] Y. Zheng, S. Yang, and H. Cheng, “An application framework of dig-
no. 2, pp. 136–143, Mar./Apr. 2020. ital twin and its case study,” J. Ambient Intell. Humanized Comput.,
[21] L. U. Khan, Z. Han, W. Saad, E. Hossain, M. Guizani, and vol. 10, no. 3, pp. 1141–1153, 2019.
C. S. Hong, “Digital twin of wireless systems: Overview, taxonomy, [45] R. Söderberg, K. Wärmefjord, J. S. Carlson, and L. Lindkvist,
challenges, and opportunities,” 2022, arXiv:2202.02559. “Toward a digital twin for real-time geometry assurance in indi-
[22] W. Sun, H. Zhang, R. Wang, and Y. Zhang, “Reducing offloading vidualized production,” CIRP Ann., vol. 66, no. 1, pp. 137–140,
latency for digital twin edge networks in 6G,” IEEE Trans. Veh. 2017.
Technol., vol. 69, no. 10, pp. 12240–12251, Oct. 2020. [46] F. Tao, J. Cheng, Q. Qi, M. Zhang, H. Zhang, and F. Sui, “Digital
[23] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, twin-driven product design, manufacturing and service with big data,”
“Communication-efficient federated learning for digital twin edge Int. J. Adv. Manuf. Technol., vol. 94, no. 9, pp. 3563–3576, 2018.
networks in industrial IoT,” IEEE Trans. Ind. Informat., vol. 17, [47] Z. Liu, N. Meyendorf, and N. Mrad, “The role of data fusion
no. 8, pp. 5709–5718, Aug. 2021. in predictive maintenance using digital twin,” in Proc. AIP Conf.,
[24] Y. He, J. Guo, and X. Zheng, “From surveillance to digital twin: vol. 1949, 2018, Art. no. 020023.
Challenges and recent advances of signal processing for Industrial [48] S. Boschert, C. Heinrich, and R. Rosen, “Next generation digital
Internet of Things,” IEEE Signal Process. Mag., vol. 35, no. 5, twin,” in Proc. TMCE, 2018, pp. 7–11.
pp. 120–129, Sep. 2018. [49] T. Taleb, K. Samdanis, B. Mada, H. Flinck, S. Dutta, and D. Sabella,
[25] M. Groshev, C. Guimarães, J. Martín-Pérez, and A. de la Oliva, “On multi-access edge computing: A survey of the emerging 5G
“Toward intelligent cyber-physical systems: Digital twin meets arti- network edge cloud architecture and orchestration,” IEEE Commun.
ficial intelligence,” IEEE Commun. Mag., vol. 59, no. 8, pp. 14–20, Surveys Tuts., vol. 19, no. 3, pp. 1657–1681, 3rd Quart., 2017.
Aug. 2021. [50] P. Cruz, N. Achir, and A. C. Viana, “On the edge of the deployment:
[26] D. Wang, B. Li, B. Song, Y. Liu, K. Muhammad, and X. Zhou, A survey on multi-access edge computing,” ACM Comput. Surveys,
“Dual-driven resource management for sustainable computing in the to be published.
blockchain-supported digital twin IoT,” IEEE Internet Things J., early [51] Multi-Access Edge Computing (MEC): Framework and Reference
access, Apr. 6, 2021, doi: 10.1109/JIOT.2022.3162714. Architecture, ETSI Standard GS MEC 003, 2019.
[27] Y. Dai and Y. Zhang, “Adaptive digital twin for vehicular edge [52] A. A. Al-Habob, A. Ibrahim, O. A. Dobre, and A. G. Armada,
computing and networks,” J. Commun. Inf. Netw., vol. 7, no. 1, “Collision-free sequential task offloading for mobile edge comput-
pp. 48–59, 2022. ing,” IEEE Commun. Lett., vol. 24, no. 1, pp. 71–75, Jan. 2020.

VOLUME 3, 2022 1379

TANG et al.: SURVEY ON DIGITAL TWIN EDGE NETWORKS (DITEN) TOWARD 6G

[53] A. A. Al-Habob, O. A. Dobre, A. G. Armada, and S. Muhaidat, [74] B. Sudharsan and P. Patel, “Machine learning meets Internet of
“Task scheduling for mobile edge computing using genetic algo- Things: From theory to practice,” in Proc. Eur. Conf. Mach. Learn.
rithm and conflict graphs,” IEEE Trans. Veh. Technol., vol. 69, no. 8, Principles Practice Knowl. Disc. Databases, 2021, pp. 1–9.
pp. 8805–8819, Aug. 2020. [75] M. S. Murshed, C. Murphy, D. Hou, N. Khan, G. Ananthanarayanan,
[54] M. Zeng, N.-P. Nguyen, O. A. Dobre, and H. V. Poor, “Delay and F. Hussain, “Machine learning at the network edge: A survey,”
minimization for NOMA-assisted MEC under power and energy con- ACM Comput. Surveys, vol. 54, no. 8, pp. 1–37, 2021.
straints,” IEEE Wireless Commun. Lett., vol. 8, no. 6, pp. 1657–1661, [76] Q. Min, Y. Lu, Z. Liu, C. Su, and B. Wang, “Machine learning based
Dec. 2019. digital twin framework for production optimization in petrochemical
[55] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, “Mobile industry,” Int. J. Inf. Manag., vol. 49, pp. 502–519, Dec. 2019.
edge computing—A key technology towards 5G,” White Paper, ETSI, [77] C. Zhang, G. Zhou, H. Li, and Y. Cao, “Manufacturing blockchain of
Sophia Antipolis, France, 2015. things for the configuration of a data-and knowledge-driven digital
[56] J. Moura and D. Hutchison, “Game theory for multi-access edge twin manufacturing cell,” IEEE Internet Things J., vol. 7, no. 12,
computing: Survey, use cases, and future trends,” IEEE Commun. pp. 11884–11894, Dec. 2020.
Surveys Tuts., vol. 21, no. 1, pp. 260–288, 1st Quart., 2018. [78] J. Zheng, T. H. Luan, L. Gao, Y. Zhang, and Y. Wu, “Learning based
[57] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, task offloading in digital twin empowered Internet of Vehicles,” 2021,
“Communication-efficient federated learning and permissioned arXiv:2201.09076.
blockchain for digital twin edge networks,” IEEE Internet Things [79] X. Xu et al., “Service offloading with deep Q-network for digital
J., vol. 8, no. 4, pp. 2276–2288, Jan. 2020. twinning-empowered Internet of Vehicles in edge computing,” IEEE
Trans. Ind. Informat., vol. 18, no. 2, pp. 1414–1423, Feb. 2022.
[58] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, “Low-
[80] T. Liu, L. Tang, W. Wang, X. He, and Q. Chen, “Resource allocation
latency federated learning and blockchain for edge association in
via edge cooperation in digital twin assisted Internet of Vehicle,” in
digital twin empowered 6G networks,” IEEE Trans. Ind. Informat.,
Proc. IEEE Global Commun. Conf. (GLOBECOM), 2021, pp. 1–6.
vol. 17, no. 7, pp. 5098–5107, Jul. 2021.
[81] W. Sun, N. Xu, L. Wang, H. Zhang, and Y. Zhang, “Dynamic
[59] P. Bellavista, C. Giannelli, M. Mamei, M. Mendula, and M. Picone, digital twin and federated learning with incentives for air-ground
“Application-driven network-aware digital twin management in networks,” IEEE Trans. Netw. Sci. Eng., vol. 9, no. 1, pp. 321–333,
industrial edge environments,” IEEE Trans. Ind. Informat., vol. 17, Jan./Feb. 2022.
no. 11, pp. 7791–7801, Nov. 2021. [82] M. Masdari, S. S. Nabavi, and V. Ahmadi, “An overview of virtual
[60] T. Liu, L. Tang, W. Wang, Q. Chen, and X. Zeng, “Digital- machine placement schemes in cloud computing,” J. Netw. Comput.
twin-assisted task offloading based on edge collaboration in the Appl., vol. 66, pp. 106–127, May 2016.
digital twin edge network,” IEEE Internet Things J., vol. 9, no. 2, [83] P. Thakur and D. K. Shrivastava, “Interoperability issues and standard
pp. 1427–1444, Jan. 2022. architecture for service delivery in federated cloud: A review,” in
[61] T. H. Luan, R. Liu, L. Gao, R. Li, and H. Zhou, “The paradigm of Proc. IEEE Int. Conf. Comput. Intell. Commun. Netw. (CICN), 2015,
digital twin communications,” 2021, arXiv:2105.07182. pp. 908–912.
[62] C. Hu et al., “Digital twin-assisted real-time traffic data prediction [84] W. Sun, S. Lei, L. Wang, Z. Liu, and Y. Zhang, “Adaptive federated
method for 5G-enabled Internet of Vehicles,” IEEE Trans. Ind. learning and digital twin for Industrial Internet of Things,” IEEE
Informat., vol. 18, no. 4, pp. 2811–2819, Apr. 2022. Trans. Ind. Informat., vol. 17, no. 8, pp. 5605–5614, Aug. 2021.
[63] D. Wang et al., “The role of digital twin in optical communica- [85] L. U. Khan, W. Saad, Z. Han, E. Hossain, and C. S. Hong, “Federated
tion: Fault management, hardware configuration, and transmission learning for Internet of Things: Recent advances, taxonomy, and
simulation,” IEEE Commun. Mag., vol. 59, no. 1, pp. 133–139, open challenges,” IEEE Commun. Surveys Tuts., vol. 23, no. 3,
Jan. 2021. pp. 1759–1799, 3rd Quart., 2021.
[64] P. Jia, X. Wang, and X. Shen, “Digital-twin-enabled intelligent [86] L. U. Khan, Y. K. Tun, M. Alsenwi, M. Imran, Z. Han, and
distributed clock synchronization in Industrial IoT systems,” IEEE C. S. Hong, “A dispersed federated learning framework for 6G-
Internet Things J., vol. 8, no. 6, pp. 4548–4559, Mar. 2021. enabled autonomous driving cars,” 2021, arXiv:2105.09641.
[65] K. Zhang, J. Cao, and Y. Zhang, “Adaptive digital twin and [87] L. U. Khan, Z. Han, D. Niyato, and C. S. Hong, “Socially-
multiagent deep reinforcement learning for vehicular edge com- aware-clustering-enabled federated learning for edge networks,”
puting and networks,” IEEE Trans. Ind. Informat., vol. 18, no. 2, IEEE Trans. Netw. Service Manag., vol. 18, no. 3, pp. 2641–2658,
pp. 1405–1413, Feb. 2022. Sep. 2021.
[66] B. Fan, Y. Wu, Z. He, Y. Chen, T. Q. Quek, and C.-Z. Xu, [88] F. Tang, B. Mao, N. Kato, and G. Gui, “Comprehensive survey
“Digital twin empowered mobile edge computing for intelligent on machine learning in vehicular network: Technology, applica-
vehicular lane-changing,” IEEE Netw., vol. 35, no. 6, pp. 194–201, tions and challenges,” IEEE Commun. Surveys Tuts., vol. 23, no. 3,
Nov./Dec. 2021. pp. 2027–2057, 3rd Quart., 2021.
[67] W. Danilczyk, Y. Sun, and H. He, “ANGEL: An intelligent digital [89] N. Kato et al., “Optimizing space-air-ground integrated networks
twin framework for microgrid security,” in Proc. IEEE North Amer. by artificial intelligence,” IEEE Wireless Commun., vol. 26, no. 4,
Power Symp. (NAPS), 2019, pp. 1–6. pp. 140–147, Aug. 2019.
[90] F. Tang, H. Hofner, N. Kato, K. Kaneko, Y. Yamashita, and
[68] Y. Xiaohu, C.-X. Wang, J. Huang, and X. Gao, “Towards 6G wireless
M. Hangai, “A deep reinforcement learning-based dynamic traf-
communication networks: Vision, enabling technologies, and new
fic offloading in space-air-ground integrated networks (SAGIN),”
paradigm shifts,” Sci. China Inf. Sci., vol. 64, no. 1, pp. 1–74, 2021.
IEEE J. Sel. Areas Commun., vol. 40, no. 1, pp. 276–289,
[69] T. Lechler et al., “Introduction of a comprehensive structure Jan. 2022.
model for the digital twin in manufacturing,” in Proc. 25th IEEE [91] B. Mao, F. Tang, Y. Kawamoto, and N. Kato, “Optimizing compu-
Int. Conf. Emerg. Technol. Factory Autom. (ETFA), vol. 1, 2020, tation offloading in satellite-UAV-served 6G IoT: A deep learning
pp. 1773–1780. approach,” IEEE Netw., vol. 35, no. 4, pp. 102–108, Jul./Aug. 2021.
[70] S. Zhang, C. Kang, Z. Liu, J. Wu, and C. Ma, “A product quality [92] F. Tang, Y. Zhou, and N. Kato, “Deep reinforcement learning
monitor model with the digital twin model and the stacked auto for dynamic uplink/downlink resource allocation in high mobil-
encoder,” IEEE Access, vol. 8, pp. 113826–113836, 2020. ity 5G HetNet,” IEEE J. Sel. Areas Commun., vol. 38, no. 12,
[71] J. Conde, A. Munoz-Arcentales, A. Alonso, S. Lopez-Pernas, and pp. 2773–2782, Dec. 2020.
J. Salvachua, “Modeling digital twin data and architecture: A building [93] Y. Dai, K. Zhang, S. Maharjan, and Y. Zhang, “Deep reinforce-
guide with FIWARE as enabling technology,” IEEE Internet Comput., ment learning for stochastic computation offloading in digital twin
vol. 26, no. 3, pp. 7–14, May/Jun. 2022. networks,” IEEE Trans. Ind. Informat., vol. 17, no. 7, pp. 4968–4977,
[72] S. Ali et al., “6G white paper on machine learning in wireless Jul. 2021.
communication networks,” 2020, arXiv:2004.13875. [94] R. Dong, C. She, W. Hardjawana, Y. Li, and B. Vucetic, “Deep
[73] D. P. Kumar, T. Amgoth, and C. S. R. Annavarapu, “Machine learn- learning for hybrid 5G services in mobile edge computing systems:
ing algorithms for wireless sensor networks: A survey,” Inf. Fusion, Learn from a digital twin,” IEEE Trans. Wireless Commun., vol. 18,
vol. 49, pp. 1–25, Sep. 2019. no. 10, pp. 4692–4707, Oct. 2019.

1380 VOLUME 3, 2022

[95] Y. Dai, D. Xu, K. Zhang, S. Maharjan, and Y. Zhang, “Deep [115] L. Zhao, G. Han, Z. Li, and L. Shu, “Intelligent digital twin-based
reinforcement learning and Permissioned blockchain for content software-defined vehicular networks,” IEEE Netw., vol. 34, no. 5,
caching in vehicular edge computing and networks,” IEEE Trans. pp. 178–184, Sep./Oct. 2020.
Veh. Technol., vol. 69, no. 4, pp. 4312–4324, Apr. 2020. [116] F. Granelli, R. Capraro, M. Lorandi, and P. Casari, “Evaluating a
[96] K. I. Ahmed, H. Tabassum, and E. Hossain, “Deep learning for radio digital twin of an IoT resource slice: An emulation study using
resource allocation in multi-cell networks,” IEEE Netw., vol. 33, the ELIoT platform,” IEEE Netw. Lett., vol. 3, no. 3, pp. 147–151,
no. 6, pp. 188–195, Nov./Dec. 2019. Sep. 2021.
[97] K. Zhang, Y. Zhu, S. Leng, Y. He, S. Maharjan, and Y. Zhang, [117] S. Muralidharan, B. Yoo, and H. Ko, “Designing a semantic digital
“Deep learning empowered task offloading for mobile edge com- twin model for IoT,” in Proc. IEEE Int. Conf. Consum. Electron.
puting in urban informatics,” IEEE Internet Things J., vol. 6, no. 5, (ICCE), 2020, pp. 1–2.
pp. 7635–7647, Oct. 2019. [118] J. Protner, M. Pipan, H. Zupan, M. Resman, M. Simic, and N. Herakovic,
[98] Y. Xie, Z. Xu, J. Xu, S. Gong, and Y. Wang, “Backscatter-aided “Edge computing and digital twin based smart manufacturing,” IFAC-
hybrid data offloading for mobile edge computing via deep reinforce- PapersOnLine, vol. 54, no. 1, pp. 831–836, 2021.
ment learning,” in Proc. Int. Conf. Mach. Learn. Intell. Commun., [119] Z. Zhou et al., “Secure and latency-aware digital twin assisted resource
2019, pp. 525–537. scheduling for 5G edge computing-empowered distribution grids,”
[99] Y. Dai, K. Zhang, S. Maharjan, and Y. Zhang, “Edge intelligence IEEE Trans. Ind. Informat., vol. 18, no. 7, pp. 4933–4943, Jul. 2022.
for energy-efficient computation offloading and resource alloca- [120] T. Do-Duy, D. Van Huynh, O. A. Dobre, B. Canberk, and
tion in 5G beyond,” IEEE Trans. Veh. Technol., vol. 69, no. 10, T. Q. Duong, “Digital twin-aided intelligent offloading with edge
pp. 12175–12186, Oct. 2020. selection in mobile edge computing,” IEEE Wireless Commun. Lett.,
[100] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, vol. 11, no. 4, pp. 806–810, Apr. 2022.
“Communication-efficient learning of deep networks from decentral- [121] C. Gehrmann and M. Gunnarsson, “A digital twin based industrial
ized data,” in Proc. Artif. Intell. Stat., 2017, pp. 1273–1282. automation and control system security architecture,” IEEE Trans.
[101] D. Kwon, J. Jeon, S. Park, J. Kim, and S. Cho, “Multiagent Ind. Informat., vol. 16, no. 1, pp. 669–680, Jan. 2020.
DDPG-based deep learning for smart ocean federated learning IoT [122] Y. Sun, X. Xu, R. Qiang, and Q. Yuan, “Research on security
networks,” IEEE Internet Things J., vol. 7, no. 10, pp. 9895–9903, management and control of power grid digital twin based on edge
Oct. 2020. computing,” in Proc. IEEE 2nd Int. Seminar Artif. Intell. Netw. Inf.
[102] J. Kang, Z. Xiong, D. Niyato, Y. Zou, Y. Zhang, and M. Guizani, Technol. (AINIT), 2021, pp. 606–610.
“Reliable federated learning for mobile networks,” IEEE Wireless [123] A. Al-Ali, R. Gupta, T. Z. Batool, T. Landolsi, F. Aloul, and
Commun., vol. 27, no. 2, pp. 72–80, Apr. 2020. A. A. Nabulsi, “Digital twin conceptual model within the context of
[103] X. Huang, S. Leng, S. Maharjan, and Y. Zhang, “Multi-agent deep Internet of Things,” Future Internet, vol. 12, no. 10, p. 163, 2020.
reinforcement learning for computation offloading and interference [124] K. Zhang, J. Cao, S. Maharjan, and Y. Zhang, “Digital twin empow-
coordination in small cell networks,” IEEE Trans. Veh. Technol., ered content caching in social-aware vehicular edge networks,” IEEE
vol. 70, no. 9, pp. 9282–9293, Sep. 2021. Trans. Comput. Social Syst., vol. 9, no. 1, pp. 239–251, Feb. 2022.
[104] M. Chen, Z. Yang, W. Saad, C. Yin, H. V. Poor, and S. Cui, “A joint [125] Z. Wang et al., “A digital twin paradigm: Vehicle-to-cloud based
learning and communications framework for federated learning over advanced driver assistance systems,” in Proc. IEEE 91st Veh. Technol.
wireless networks,” IEEE Trans. Wireless Commun., vol. 20, no. 1, Conf. (VTC-Spring), 2020, pp. 1–6.
pp. 269–283, Jan. 2021. [126] Y. Liu, Z. Wang, K. Han, Z. Shou, P. Tiwari, and J. H. Hansen, “Sensor
[105] T. H. T. Le et al., “An incentive mechanism for federated learning fusion of camera and cloud digital twin information for intelligent
in wireless cellular networks: An auction approach,” IEEE Trans. vehicles,” in Proc. IEEE Intell. Veh. Symp. (IV), 2020, pp. 182–187.
Wireless Commun., vol. 20, no. 8, pp. 4874–4887, Aug. 2021. [127] X. Chen, T. Chen, Z. Zhao, H. Zhang, M. Bennis, and J. Yusheng,
[106] J. Kang, Z. Xiong, D. Niyato, S. Xie, and J. Zhang, “Incentive mech- “Resource awareness in unmanned aerial vehicle-assisted mobile-
anism for reliable federated learning: A joint optimization approach edge computing systems,” in Proc. IEEE 91st Veh. Technol. Conf.
to combining reputation and contract theory,” IEEE Internet Things (VTC-Spring), 2020, pp. 1–6.
J., vol. 6, no. 6, pp. 10700–10714, Dec. 2019. [128] Y.-C. Wang, N. Zhang, H. Li, and J. Cao, “Research on digital twin
[107] Y. Xu, Y. Sun, X. Liu, and Y. Zheng, “A digital-twin-assisted framework of military large-scale UAV based on cloud computing,”
fault diagnosis using deep transfer learning,” IEEE Access, vol. 7, J. Phys. Conf., vol. 1738, no. 1, 2021, Art. no. 012052.
pp. 19990–19999, 2019. [129] R. Martinez-Velazquez, R. Gamez, and A. El Saddik, “Cardio twin:
[108] I. Yaqoob, K. Salah, M. Uddin, R. Jayaraman, M. Omar, and A digital twin of the human heart running on the edge,” in Proc.
M. Imran, “Blockchain for digital twins: Recent advances and future IEEE Int. Symp. Med. Meas. Appl. (MeMeA), 2019, pp. 1–6.
research challenges,” IEEE Netw., vol. 34, no. 5, pp. 290–298, [130] X. Lin, J. Wu, J. Li, W. Yang, and M. Guizani, “Stochastic
Sep./Oct. 2020. digital-twin service demand with edge response: An incentive-based
[109] H. R. Hasan et al., “A blockchain-based approach for the creation congestion control approach,” IEEE Trans. Mobile Comput., early
of digital twins,” IEEE Access, vol. 8, pp. 34113–34126, 2020. access, Oct. 21, 2021, doi: 10.1109/TMC.2021.3122013.
[110] C. Altun, B. Tavli, and H. Yanikomeroglu, “Liberalization of digital [131] H. Ahmadi, A. Nag, Z. Khan, K. Sayrafian, and S. Rahadrja,
twins of IoT-enabled home appliances via blockchains and absolute “Networked twins and twins of networks: An overview on the
ownership rights,” IEEE Commun. Mag., vol. 57, no. 12, pp. 65–71, relationship between digital twins and 6G,” 2021, arXiv:2108.05781.
Dec. 2019. [132] B. Mao, F. Tang, Z. M. Fadlullah, and N. Kato, “An intelligent
[111] S. R. Pandey, N. H. Tran, M. Bennis, Y. K. Tun, A. Manzoor, route computation approach based on real-time deep learning strategy
and C. S. Hong, “A crowdsourcing framework for on-device fed- for software defined communication systems,” IEEE Trans. Emerg.
erated learning,” IEEE Trans. Wireless Commun., vol. 19, no. 5, Topics Comput., vol. 9, no. 3, pp. 1554–1565, Jul./Sep. 2021.
pp. 3241–3256, May 2020. [133] L. Girletti, M. Groshev, C. Guimarães, C. J. Bernardos, and
[112] T. H. T. Le, N. H. Tran, Y. K. Tun, Z. Han, and C. S. Hong, A. de la Oliva, “An intelligent edge-based digital twin for robotics,”
“Auction based incentive design for efficient federated learning in in Proc. IEEE Globecom Workshops (GC Wkshps, 2020, pp. 1–6.
cellular wireless networks,” in Proc. IEEE Wireless Commun. Netw. [134] M. Glatt, P. Kölsch, C. Siedler, P. Langlotz, S. Ehmsen, and
Conf. (WCNC), 2020, pp. 1–6. J. C. Aurich, “Edge-based digital twin to trace and ensure sustain-
[113] L. U. Khan, W. Saad, D. Niyato, Z. Han, and C. S. Hong, “Digital- ability in cross-company production networks,” in Proc. 28th CIRP
twin-enabled 6G: Vision, architectural trends, and future directions,” Conf. Life Cycle Eng., vol. 98, Mar. 2021, pp. 276–281.
IEEE Commun. Mag., vol. 60, no. 1, pp. 74–80, Jan. 2022. [135] L. Guo, Y. Cheng, Y. Zhang, Y. Liu, C. Wan, and J. Liang,
[114] Y. Liu, Z. Fang, M. H. Cheung, W. Cai, and J. Huang, “An incentive “Development of cloud-edge collaborative digital twin system for
mechanism for sustainable blockchain storage,” IEEE/ACM Trans. FDM additive manufacturing,” in Proc. IEEE 19th Int. Conf. Ind.
Netw., early access, May 5, 2022, doi: 10.1109/TNET.2022.3166459. Inform. (INDIN), 2021, pp. 1–6.

VOLUME 3, 2022 1381