IEEE Access

Empowering Non-Terrestrial Networks with Artificial

Journal: IEEE Access

Manuscript ID Access-2023-24685

Manuscript Type: Regular Manuscript

Date Submitted by the

Complete List of Authors: Iqbal, Amjad; Universiti Tunku Abdul Rahman Lee Kong Chian Fakulti
Kejuruteraan dan Sains; Carleton University Department of Systems and
Computer Engineering
Mau-Luen, Tham ; Universiti Tunku Abdul Rahman Lee Kong Chian
Fakulti Kejuruteraan dan Sains
WONG, YI JIE; Universiti Tunku Abdul Rahman Lee Kong Chian Fakulti
Kejuruteraan dan Sains
AL-HABASHNA, ALA’A; Carleton University Department of Systems and
Computer Engineering
Wainer, G.l A.; Carleton University Department of Systems and
Computer Engineering
Zhu, Yong Xu; University of Warwick School of Engineering
Dagiuklas, Tasos; London South Bank University Department of
Computer Science and Informatics

Subject Category<br>Please
select at least two subject Communications technology, Information theory, Vehicular and wireless
categories that best reflect technologies
the scope of your manuscript:

Keywords: <b>Please choose

Additional Manuscript
Keywords:

For Review Only

1
2
3
4
5
6
7
AUTHOR RESPONSES TO IEEE ACCESS
8 SUBMISSION QUESTIONS
9 Author chosen
10 Survey
manuscript type:
11
The fifth generation (5G) of wireless communication technology has
12
revolutionized the way we connect with each other; enabling faster
13
14
data transfer rates, lower latency, and higher reliability. However,
15 the demand for even more efficient and ubiquitous connectivity is
16 Author explanation ever-growing. As such, researchers have already been exploring
17 /justification for the potential of the sixth generation (6G) of wireless
18 choosing this communications. The 6G networks are expected to provide
19 manuscript type: unprecedented connectivity and reliability. An important component
20 of 6G networks is the use of non-terrestrial networks (NTNs) that
21 have the potential to extend the coverage of 6G networks to areas
22 where connectivity with traditional terrestrial networks is not feasible
23 or cost-effective.Â
24 Our survey provides a clear vision for the potential of AI-powered
25 Author description of NTNs for 6G networks and highlights the need for further research
26 how this manuscript fits and development in this field. We believe that our survey can serve
27 within the scope of IEEE as a valuable reference for industry and academic researchers
28 Access: interested in deploying and managing AI-powered NTNs for 6G
29
networks
30
We discuss the use of AI in NTNs for 6G communication networks.
31
32
The main focus of the survey is: 1. NTNs role in 6G networks and
Author description
33 unique security/privacy concerns. 2. Applicable AI approaches for
detailing the unique
34 NTN problems and proper technique selection. 3. AI-enabled NTN
contribution of the
35 research avenues and superiority over traditional methods. 4.
manuscript related to
36 AI-based NTN resource allocation case studies and research. 5.
existing literature:
37 Future open issues in AI for NTNs, considering constraints for its
38 maximum potential.
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 For Review Only
 IEEE Access Page 2 of 42

1
2 Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
3 Digital Object Identifier 10.1109/ACCESS.2022.Doi Number
4
5
6
7
8
Empowering Non-Terrestrial Networks with
9
10
Artificial Intelligence: A Survey
11
12 AMJAD IQBAL1,2, MAU-LUEN THAM1, YI JIE WONG1, ALA’A AL-HABASHNA2, GABRIEL
13 WAINER2, YONG XU ZHU3, AND TASOS DAGIUKLAS4
1
Department of Electrical and Electronic Engineering, Lee Kong Chian Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Sungai
14 Long Campus, Selangor 43000, Malaysia.
15 2
Department of Systems and Computer Engineering, Carleton University, 1125 Colonel By Dr., Ottawa, ON, K1S 5B6, Canada.
16
3
School of Engineering, Warwick University, UK.
4
Cognitive Systems Research Centre, School of Engineering/Computer Science, London South Bank University, 103 Borough Road, London SE1
17 0AA, UK.
18 Corresponding authors: Mau-Luen Tham (e-mail: thamml@utar.edu.my).
19 This work was supported in part by the Universiti Tunku Abdul Rahman (UTAR), Malaysia, under UTAR Research Fund (UTARRF)
(IPSR/RMC/UTARRF/2021C1/T05) and in part by the British Council under UK-ASEAN Institutional Links Early Career Researchers Scheme with project
20 number 913030644 and partially supported by NSERC Canada).”
21
22
23
ABSTRACT 6G networks can support global, ubiquitous and seamless connectivity through the
24
convergence of terrestrial and non-terrestrial networks (NTNs). Unlike terrestrial scenarios, NTNs pose
25
unique challenges including propagation characteristics, latency and mobility, owing to the operations in
26
spaceborne and airborne platforms. To overcome all these technical hurdles, this survey paper presents the
27
use of artificial intelligence (AI) techniques in learning and adapting to the complex NTN environments. We
28
begin by providing an overview of NTNs in the context of 6G, highlighting the potential security and privacy
29
issues. Next, we review the existing AI methods adopted for 6G NTN optimization, starting from machine
30
learning (ML), through deep learning (DL) to deep reinforcement learning (DRL). All these AI techniques
31
have paved the way towards more intelligent network planning, resource allocation (RA), and interference
32
management. Furthermore, we discuss the challenges and opportunities in AI-powered NTN for 6G networks.
33
Finally, we conclude by providing insights and recommendations on the key enabling technologies for future
34
AI-powered 6G NTNs.
35
36
37 INDEX TERMS Non-Terrestrial Networks (NTNs), Artificial Intelligence (AI), 5G/6G, Unmanned
38 Aircraft System (UAS), Resource Allocation (RA), Reinforcement Learning (RL), Deep Learning (DL)
39
40
41 I. INTRODUCTION (NTNs), which include Low/Medium/Geostationary (LEO,
42 The evolution of wireless communication technology has MEO, GEO) satellites, high altitude platform stations
43 been rapid in recent years, with mobile network operators (HAPS), unmanned aircraft systems (UASs), or a
44 deploying the fifth-generation (5G) technology worldwide. combination of these technologies [3]. NTNs can provide
45 The Third Generation Partnership Project (3GPP) has uninterrupted communication and high data transfer rates
46 begun the standardization of 5G-Advanced, which is in remote, disaster-stricken, and rural areas where
47 expected to offer higher data rates, lower latency, increased terrestrial networks are not available. Recent studies have
48 capacity, and more efficient spectrum utilization than any shown that NTNs can offer high availability and low
49 of its predecessors [1]-[2]. However, due to limited latency, making them an ideal candidate for 6G
50 coverage areas and geographic constraints, it is challenging communication systems [4]-[5].
51 to guarantee ubiquitous coverage everywhere with existing Despite these advantages, NTNs present significant
52 network infrastructures. In times of natural disasters, challenges related to their deployment and management.
53 connectivity outages are widespread, which can be One of the primary challenges is how to deploy and manage
54 detrimental to critical actions necessary for saving lives and efficiently, including the physical and ground-based
55 properties. To address these challenges, research on 5G- infrastructure, such as antennas, base stations (BSs), and
56 Advanced and sixth-generation (6G) communication backhaul networks. Furthermore, the dynamic nature of
57 networks are shifting towards non-terrestrial networks NTNs, with platforms moving at high speeds and in
58
59 VOLUME XX, 2017 For Review Only 1

60
Page 3 of 42 IEEE Access

1
2
3 different directions, introduces additional challenges overview of the vision, requirements, and challenges of 6G
4 related to signal propagation, interference, and handover wireless networks are explained in [18] and [19]. The current
5 management [6]-[7]. status and future directions of 6G wireless networks are
6 Researchers are turning to artificial intelligence (AI) as a reviewed in [20] and [21]. Furthermore, [22] and [23] discuss
7 promising solution to address these challenges. AI can help the key enabling technologies for 6G networks, such as
8 to optimize the performance of NTNs by analyzing data terahertz (THz) communication, visible light communication
9 from various sources, predicting network behavior, and (VLC), and wireless power transfer. The applications,
10 adapting to changing conditions [8]. For example, AI can technologies, and challenges of 6G wireless networks are
11 enable intelligent power management and beamforming to explored in [24] and [25]. The challenges and opportunities
12 maximize resource utilization while minimizing in 6G networks, including ultra-reliable, low-latency, and
13 interference and maintaining the quality of service (QoS) massive machine-type communication, are discussed in [26],
14 [9]. Also, AI can provide advanced analytics and automated [27], and [28]. A comprehensive 6G wireless communication
15 decision-making to enhance the efficiency and reliability of survey is presented in [29] and [30]. The concept and
16 NTNs. standardization of 6G networks, including new spectrum
17 The integration of AI in cellular networks is still in its early bands and wireless technologies, are explained in [31] and
18 stages compared to other fields, primarily due to wireless [32]. In [33] and [34], the authors explore the opportunities
19 networks’ complexity and time-varying nature [10]. The and challenges of 6G networks, including the use of satellite
20 multi-dimensional topology of the next-generation wireless communication and energy-efficient design. A
21 networks adds an additional layer of complexity to the comprehensive survey of 6G networks, including new radio
22 existing communication networks, making it challenging to access technologies and security challenges, is presented in
23 solve problems that arise in real networks. Nevertheless, AI [35] and [36]. In [37], the authors present a comprehensive
24 techniques can be exploited to overcome these challenges survey of 6G wireless networks, including new antenna
25 and provide efficient solutions. NTNs, being an integral technologies and network slicing. Similarly, [38] explains
26 part of next-generation wireless networks, holds great the five facets of the new wireless generation, along with its
27 potential for the application of AI. However, implementing research challenges and different opportunities for the new
28 these algorithms in real-world environments while ensuring wireless generation.
29 reliable vertical connectivity between ground and space The aforementioned studies do not incorporate the use of AI
30 networks can present practical challenges. Proper AI in their analysis. In contrast, this survey paper analyzes AI's
31 solutions must complement theoretical advancements in utilization in NTNs to enhance 6G wireless communications
32 communication systems design to achieve optimal networks, as summarized in Table 1. Firstly, we discuss the
33 performance in future networks. potential benefits and challenges of integrating AI in NTNs,
34 including its impact on network performance, reliability, and
35 A. RELATED WORKS AND PAPER CONTRIBUTION security. Secondly, we show different AI techniques that can
36 Several studies and surveys have been conducted to explore be applied to NTNs, such as machine learning (ML), deep
37 the possibility of NTNs for 6G wireless communications. learning (DL), and deep reinforcement learning (DRL).
38 However, it is very rare to explore the concept of AI for Finally, this survey paper discusses the potential applications
39 NTNs. Instead, numerous research has focused on network of AI-powered NTNs in various industries, including
40 architectures, standards, regulations, and use cases. For healthcare, transportation, and smart cities. We analyze the
41 instance, [11] and [12] give an overview of different NTN use cases for AI-powered NTNs in these industries,
42 use, including satellite communication, aerial drones, and including how they can improve efficiency and reduce costs.
43 terrestrial devices. In [13], challenges associated with The main contributions of this survey paper can be
44 satellite communication in 6G are examined, including summarized as follows:
45 power constraints, latency issues, and frequency allocation. 1) We provide an overview of NTNs, including their
46 The works in [14] and [15] explore the potential impact of introduction in the context of 6G networks, their role in
47 6G networks on various industries, including agriculture, enhancing network performance, and their key features
48 transportation, and healthcare. Unlike [14] and [15], the and requirements. Additionally, we analyze the unique
49 works in [16] and [17] focus on the potential security threats security and privacy concerns associated with NTNs,
50 of NTNs by discussing the technical aspects of NTNs, such providing valuable insights into their intrinsic nature.
51 as the different radio access technologies available and the 2) We discuss the AI approaches by explaining the
52 requirements and challenges of integrating 6G with existing fundamental aspects of AI techniques used in the NTNs.
53 technologies. Overall, these works provide valuable insights As a result, we can select appropriate AI approaches for
54 into the potential of NTNs for 6G and the various challenges dealing with various NTN issues.
55 to consider for their successful implementation. The
56
57
58
59 VOLUME XX, 2017 For Review Only 1

60
 IEEE Access Page 4 of 42

1
2
3 Table 1. Summary of Related Works on 6G
4 Ref Year Contribution and Main Focus
5 [22] 2019
A vision for 6G networks in 2030 for superior performance and enabling emerging services and applications is discussed. The focus is to
propose a large-dimensional, autonomous network architecture integrating various networks and advanced technologies.
6
The security and privacy issues associated with 6G networks are being explored as next-generation solutions due to the limitations of 5G
7 [17] 2020
networks. The main focus is to discuss four key aspects of 6G networks and their associated security and privacy issues.
8 [19] 2020
Vision, technology trends, and challenges for 6G are discussed. The main focus is providing the key enabler of a ubiquitous intelligent
9 mobile society and suggesting a roadmap for the 6G standards.
Recent advances in 6G wireless systems are discussed. The main focus is to present a taxonomy of key technologies and open research
10 [21] 2020
challenges and propose practical guidelines such as neural networks and blockchain-based secure business models.
11 A discussion of 6G wireless communication technologies is provided, emphasizing fundamental breakthroughs at the physical layer. The
12 [31] 2020 main focus is to provide an overview of these technologies, including holographic radio, terahertz communication, large, intelligent
13 surface, and orbital angular momentum.
The limitations of 5G and the need to develop the 6G wireless system are discussed. The main focus is to provide the vision for 6G and
14 [32] 2020
outline a research agenda for enabling the new services and technologies required.
15 The potential of UAVs in beyond 5G and 6G wireless networks are discussed. The main focus is to highlight the use of cellular networks,
[13] 2021
16 advanced technologies, machine learning, and non-terrestrial networks to support UAVs in 6G.
A fiber-wireless network architecture is presented based on full spectrum, fully adaptive, and coordinated radio access networks (RANs).
17 [23] 2021 The main focus is to offer promising scenarios such as NR-free space optical backhauling and indoor systems via visible light
18 communication for high-speed data link and VLC-aided positioning systems.
19 [24] 2021
The need for 6G to overcome the limitations of current cellular networks and support high-bandwidth applications are discussed. The main
20 focus is to provide an overview of system requirements, potential technologies, and recent research progress.
The overviews of reconfigurable intelligent surfaces (RISs) for 6G wireless networks are explained. The main focus is to provide the use
21 [25] 2021
case of RISs to create a favorable propagation channel and improve performance gains.
22 A comparison of 5G and 6G technologies, including terahertz communication, RIS, and blockchain, are presented. The main focus is to
[26] 2021
23 illustrate how IRS can enhance signal quality by controlling passive reflecting elements and how blockchain can enhance system security.
24 The possibility of integrating terrestrial and NTNs is discussed as a means of improving user experience and connecting unconnected
[16] 2022 devices. The main focus is identifying the opportunities and challenges for defining and orchestrating a new 3D wireless network
25 architecture.
26 The challenges of 5G technology and the potential benefits of 6G technology for edge networks in processing real-time applications are
[28] 2022
27 examined. The focus is on integrating ultra-reliable 6G technology into edge computing networks.
The potential of IoT devices and energy-efficient 6G wireless communication in transforming smart cities into super-smart cities is
28 [33] 2022 conferred. The main focus is to review key technologies and applications, including quantum communication, blockchain, and VLC and
29 identifies promising trends for using 6G through IoT devices in smart cities.
30 [39] 2022
An overview of 6G mobile networks, including motivations, use cases, requirements, and research projects, are reviewed. The main focus
31 is on the transition from 5G to B5G and on the advanced features that will be required for 6G.
The role of wireless backhauls in 5G networks and its integration with new technologies like UAV, HAPS, mmWave, mMIMO, and
32 [40] 2022 beamforming are presented. This article focuses mainly on rural connectivity, mobile edge computing, and security issues related to wireless
33 backhaul in 5G and B5G.
34 [34] 2023
The design of an energy-efficient resource allocation system for NTNs is explained. The main focus is to maximize system energy
efficiency by collaboratively optimizing user equipment association, power control, and UAV deployment.
35
Researchers are exploring five Facets of 6G to develop next-generation solutions, i.e., next-generation architectures, networking, IoT,
36 [38] 2023 wireless positioning and sensing, and deep learning applications. The main focus is to review promising techniques and architectures,
37 address vulnerabilities, and advocate for multi-component Pareto optimization for optimal solutions.
38 [41] 2023
Distributed edge learning (EL) techniques and their integration with advanced communication optimization designs for B5G wireless
networks are explored. The main focus is to present the open problems and emerging application opportunities for the B5G network.
39 We discuss the use of AI in NTNs for 6G communication networks. The main focus of the survey is:
40 1. NTNs role in 6G networks and unique security/privacy concerns.
41 This work
2. Applicable AI approaches for NTN problems and proper technique selection.
42 3. AI-enabled NTN research avenues and superiority over traditional methods.
4. AI-based NTN resource allocation case studies and research.
43 5. Future open issues in AI for NTNs, considering constraints for its maximum potential.
44
45 3) We provide a holistic overview of AI-enabled NTN such as network optimization and resource allocation.
46 research and a motivating argument for their
47 implementation. We explore the challenges of NTNs, B. PAPER ORGANIZATION
48 outline the issue associated with traditional methods, and This paper is organized as follows. Section II provides an
49 provide ideas for the superiority of AI techniques. overview of NTN and its potential role in 6G networks.
50 4) We summarize relevant case studies and existing Different security and privacy concerns of NTNs are
51 research used to solve the resource allocation problem discussed in Section III. In Section IV, we introduce an
52 associated with AI in NTNs. overview of AI techniques that can be used to optimize the
53 5) We summarize the main challenges and opportunities of performance of NTNs. Section V discusses the recent
54 using AI for NTNs. In addition, we provide a set of research in AI-powered NTN for 6G networks. Section VI
55 recommendations for future research directions in this identifies the key research directions and challenges in this
56 area. We also identify potential AI applications in areas field, and Section VII concludes the paper. The structure of
57 the paper is shown in Fig. 1.
58 3
59 For Review Only
60
Page 5 of 42 IEEE Access

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Figure 1. Article Structure
22
23
To facilitate the readability of this paper, Table 2 provides III. Geostationary Orbit (GEO) Satellites: GEO satellites are
24
the list of abbreviations with the terms associated. positioned at approximately 35,786 kilometers above the
25
equator. These satellites have an orbital period that matches
26
II. NON-TERRESTRIAL NETWORKS the Earth's rotation, allowing them to remain fixed relative
27 NTNs have emerged as a promising solution for extending to a specific point on the Earth's surface. This characteristic
28 coverage and improving connectivity in 6G networks. NTNs makes them ideal for applications that require continuous
29 are a class of wireless networks that operate using non-earth- coverage of a specific geographic area, such as television
30 based platforms to provide connectivity [3],[10],[11]. Unlike broadcasting and weather monitoring.
31 terrestrial networks, NTNs can cover large areas, operate IV. High Altitude Platform Stations (HAPS): HAPS are
32 independently of terrestrial infrastructure, and be quickly unmanned platforms that operate in the stratosphere at
33 deployed in areas without adequate terrestrial infrastructure altitudes between 17 to 22 kilometers. They have the
34 [8].
potential to provide connectivity to areas that are difficult to
35
access or where terrestrial infrastructure is not feasible.
36 A. OVERVIEW OF NTNs TECHNOLOGIES
HAPS can be used for a range of applications, including
37 I. Low Earth Orbit (LEO) Satellites: LEO satellites are
communication, surveillance, and environmental
38 satellites that orbit the Earth at altitudes between 200 and
monitoring. HAPS includes balloons and airships. Some
39 2000 kilometers. They have emerged as a promising
specific examples might include Google’s Project
40 platform for providing connectivity to remote and rural
Loon, Facebook’s Aquila, and the Stratobus airship
41 areas. Recent developments in satellite technology have
[44]-[45].
42 reduced the cost of launching and operating LEO satellites,
V. Unmanned Aircraft Systems (UASs): UASs operate
43 making it a viable option for commercial use. Some
without a pilot. UASs can be quickly deployed to provide
44 specific technologies for LEO include the International
connectivity in areas affected by natural disasters or
45 Space Station (ISS), the Hubble Space Telescope, and
emergencies. They can be used for a variety of applications,
46 the Global Positioning System (GPS) [42]-[43].
including communication, surveillance, and delivery. Some
47 II. Medium Earth Orbit (MEO) Satellites: MEO satellites are
specific examples that use the UASs are DJI Mavic, the
48 positioned in orbits between approximately 2,000 and
Parrot Bebop 2, and Lockheed Martin Indago3 [46],
49 35,786 kilometers above the Earth's surface. They offer a
[47],[48],[49].
50 compromise between the coverage area and the signal
51 delay. MEO satellites are commonly used for navigation
B. ROLE OF NTNs IN 6G NETWORKS
52 and communication purposes. MEO allows longer
6G wireless communications networks will require a variety
53 communication windows than LEO satellites, resulting in
of new technologies to meet the high data rate, low latency,
54 reduced handovers and a more stable connection for users.
and mobility requirements of the future network. One
55 MEO is often used for global navigation satellite systems
possible solution is using NTNs that offer a viable option to
56 (GNSS) like GPS, GLONASS, and Galileo, as well as for
provide these services cost-effectively and efficiently. They
57 remote sensing and communication purposes.
58 4
59 For Review Only
60
 IEEE Access Page 6 of 42

1
2
3 Table 2. List of Abbreviations impractical. Moreover, NTNs can also serve as a backup or
4 Abbreviation Definition redundant network in case of network failures or disasters,
5 3GPP 3rd generation partnership project ensuring high reliability and availability of communication
6 5G Fifth generation services.
7 6G Sixth generation To date, several research efforts have been undertaken to
8 AI Artificial intelligence explore the potential of NTNs for 6G networks. For instance,
9 AP Access point the integration of LEO satellites into 6G communications
10 B5G Beyond fifth generation networks has been studied extensively, with a focus on
11 BSs Base stations optimizing the satellite constellation design, routing
12 CNN Convolutional neural networks
algorithms, and interference management techniques [50].
13 DBN Deep belief network
Similarly, the use of HAPs in 6G networks has been
14 investigated, focusing on developing efficient
DL Deep learning
15 communication protocols, beamforming techniques, and
DRL Deep reinforcement learning
16 energy-efficient power management schemes [51]. The use
EL Edge learning
17 of UASs in 6G networks has also been explored, with a focus
FL Federated learning
18 on developing aerial base stations and efficient trajectory
GA Genetic algorithms
19 planning algorithms [52].
GPS Global positioning system
20 Overall, the role of NTNs in 6G networks is crucial for
HAPs High-altitude platforms providing ubiquitous, reliable, and high-capacity wireless
21 IDS Intrusion detection systems
22 communication services. The unique characteristics of
IoT Internet of things NTNs, such as high altitude, 3D mobility support, and broad
23
IPS Intrusion prevention system coverage, make them a promising solution for meeting the
24
ISS International space station requirements of future wireless networks. Ongoing research
25
26
LEO Low earth orbit efforts are expected to further improve the performance and
27 MARL Multi-agent reinforcement learning efficiency of NTNs in 6G networks and pave the way for the
28 MIMO Multiple input multiple output realization of the full potential of these networks.
29 mMIMO Massive multiple input multiple output
30 ML Machine learning C. KEY NTNs FEATURES AND REQUIREMENTS
31 NFV Network function virtualization NTNs are critical for ensuring their success in 6G
32 NTNs Non-terrestrial networks communication systems. NTNs architecture must be
33 ORAN Open radio access network designed to address key challenges such as 3D mobility link
34 PCA Principal component analysis reliability, latency, energy efficiency, and capacity. The
35 QoS Quality of service
NTNs system must be able to support a large number of
36 RA Resource allocation
devices with high data rates and low latency while also being
37 scalable, flexible, and cost-effective.
RAN Radio access network
38 One key feature of NTNs is the use of advanced antenna
RIS Reconfigurable intelligent surfaces
39 technologies, such as beamforming, which can enhance
RL Reinforcement learning
40 signal strength and reduce interference. Other key features
SDN Software-defined networks
41 include the use of multi-frequency bands, efficient power
SL Supervised learning
42 management techniques, and advanced modulation schemes.
SVMs Support vector machines
43 NTNs must have strong authentication, encryption, and
UASs Unmanned aircraft systems access control mechanisms to ensure security and privacy.
44 UAV Unmanned aerial vehicle
45 NTN requirements for 6G communication systems include
UL Unsupervised learning their ability to provide ubiquitous connectivity, reliability,
46 VLC Visible light communication
47 and high data rates with low latency. The system must
VNFs Virtual network functions support seamless integration with terrestrial networks and
48
are expected to play a vital role in enabling the full potential enable global coverage with minimal delay. Additionally,
49
of 6G networks. NTNs must be designed to meet the specific needs of various
50
The terrestrial network infrastructure, including 5G applications, such as the Internet of Things (IoT), smart
51
networks, has limitations in terms of coverage (e.g., cities, and connected vehicles. It must be able to support
52
remote/rural areas, sea/air, etc.). One of the key roles of high-altitude platforms, LEO satellites, and geostationary
53
NTNs in 6G networks is to complement the terrestrial orbit satellites. The system should also be resilient to natural
54
network infrastructure and overcome its limitations. NTNs disasters and cyber-attacks.
55
can provide broader coverage, higher capacity, and mobility Fig. 2 summarizes the key features and requirements of
56
support in areas where terrestrial networks are unavailable or NTNs for 6G communication systems.
57
58 5
59 For Review Only
60
Page 7 of 42 IEEE Access

1
2
3 legitimate users [53].
4 Another important aspect of authentication and authorization
5 in 6G networks is the regular updating of policies. As
6 security threats evolve over time, it is crucial to keep
7 authentication and authorization policies up to date to
8 address new vulnerabilities and prevent potential security
9 breaches. Regular updates help ensure that the network
10 remains secure against emerging threats and that any new
11 authentication methods or standards are implemented
12 effectively [54].
13 Privacy and Confidentiality: Privacy and confidentiality
14 are crucial considerations for the 6G network due to the vast
15 amount of data generated and the increased use of AI-
16 powered devices. With the proliferation of IoT devices and
17 the integration of AI technologies, 6G networks are expected
18 to handle a massive volume of data, including personal,
19 sensitive, and confidential information.
20 One of the main challenges for privacy and confidentiality in
21 6G networks is ensuring that this data is protected from
unauthorized access as it traverses across various network
22
segments. Data transmitted over the network may pass
23
through multiple nodes, edge computing systems, and cloud
24 Figure 2. Basic Features of NTN
infrastructures, increasing the risk of interception or
25 unauthorized access at any point in the network. Therefore,
26 robust security measures must be implemented to safeguard
III. SECURITY AND PRIVACY IN NTN
27 data privacy and maintain confidentiality [55].
Security and privacy are critical considerations in designing
28 Network Congestion: The 6G network’s high-speed
and implementing NTNs for 6G. The increasing reliance on
29 capabilities can cause network congestion, leading to a
advanced technologies in the telecommunications industry
30 higher risk of network attacks. With more devices connected
means that network security threats are becoming more
31 to the network, the chances of network congestion increase,
sophisticated and complex. As the next generation of
32 creating a larger attack surface [56].
wireless networks, 6G networks are expected to bring a
33 Malware and Hacking: As 6G networks become more
revolutionary transformation to the world of communication.
34 complex, the risk of malware and hacking increases, putting
In the context of 6G, security and privacy concerns are
35 the security and privacy of the network at risk. It is essential
magnified due to the large-scale deployment of networked
36 to have robust, secure technologies in place, such as
devices and the increasing complexity of network
37 encryption, authentication, and firewalls, to protect the
architectures. These concerns include data leakage, identity
38 network. Additionally, users should follow best practices for
theft, and unauthorized access to sensitive information.
39 online security, such as using strong passwords and being
Developing secure and privacy-preserving technologies is
40 aware of potential phishing scams [36].
essential to ensure the secure operation of 6G networks. The
41 Resource Allocation: 6G technology is expected to have
following is a list of security and privacy concerns for the
42 enhanced security features to ensure data is protected from
NTNs in 6G networks.
43 unauthorized access. Additionally, the network must be able
Authentication and Authorization: Authentication and
44 to allocate resources to meet users' demands efficiently.
authorization refer to verifying the identity of devices or
45 Finally, the network must be able to adapt to changing user
users and determining their level of access privileges within
46 needs and respond quickly to changing conditions [57].
the network. These mechanisms are crucial for ensuring the
47 Fault Tolerance: Identifying and mitigating faults in the
security and privacy of the network, as they prevent
48 network is essential for maintaining the network’s security
unauthorized devices or malicious actors from gaining
49 and privacy. This is done by detecting errors, isolating them,
access and potentially causing harm.
50 and recovering from them without impacting the system.
Advanced technologies and protocols will be required to
51 Fault tolerance is a key element in network security and helps
achieve strong authentication in 6G networks. One potential
52 prevent unauthorized access or misuse of the network. It also
approach is biometric authentication, which relies on
53 helps to protect against data loss [58].
individuals' unique physical or behavioral traits, such as
54 Trust Management: Trust management is vital to the 6G
fingerprints or facial recognition. These biometric
55 network security, enabling secure interactions and
characteristics can serve as strong authentication factors,
56 collaborations between different devices, services, and users.
making it difficult for unauthorized entities to impersonate
57
58 6
59 For Review Only
60
 IEEE Access Page 8 of 42

1
2
3 As 6G networks become more complex, with a diverse range activities. Organizations can significantly reduce the risk of
4 of devices, services, and stakeholders, managing trust successful attacks and mitigate potential damage by acting
5 becomes increasingly challenging. Trust-related attacks, preemptively.
6 such as malicious entities masquerading as trusted devices or Identity Management: AI can manage identities and access
7 services, pose a significant risk. These attacks can lead to control more effectively, ensuring that only authorized
8 unauthorized access, data breaches, or disruptions in network devices and users can access the network resources. AI can
9 operations. facilitate advanced authentication mechanisms such as
10 To address the trust management challenges in 6G networks, biometric recognition, behavior analysis, and contextual
11 several approaches can be considered. One such approach is information to verify the identity of users and devices
12 the use of trust models and frameworks that assess and accurately. By leveraging AI-driven identity management
13 quantify the trustworthiness of devices, services, or entities solutions, 6G networks can enhance security, prevent
14 based on their behavior, reputation, or credentials. These unauthorized access, and mitigate privacy risks.
15 models can help establish a trust hierarchy and determine the Encryption: AI can contribute to enhancing encryption
16 level of trustworthiness associated with different entities techniques by improving encryption algorithms, key
17 within the network. management, and overall cryptographic processes.
18 Additionally, secure authentication protocols, such as public Leveraging AI algorithms can strengthen encryption
19 key infrastructure or certificate-based authentication, can be mechanisms to withstand increasingly sophisticated attacks
20 employed to verify the identities of devices or entities and and cryptographic vulnerabilities. AI can assist in
21 ensure that only trusted entities are granted access to the developing more robust encryption algorithms, optimizing
22 network [59]. key generation and distribution, and detecting potential
23 AI can play a crucial role in mitigating these security and weaknesses or patterns that attackers may exploit. This helps
24 privacy concerns. AI can be leveraged to provide enhanced create a more secure data transmission and storage
25 security and privacy features that go beyond traditional environment in 6G networks [60].
26 security measures. The following are some of the ways AI Behavioral Analysis: AI algorithms can analyze user and
27 can help resolve these concerns. device behavior patterns, allowing the network to detect and
28 Threat Detection and Prevention: Threat detection and prevent abnormal activities that could signify a potential
29 prevention play a crucial role in ensuring network security. security threat. By continuously monitoring behavior in real-
30 As the complexity of the 6G network increases; therefore, time, AI-based systems can identify suspicious actions, such
31 advanced mechanisms are required to detect and mitigate as unauthorized access attempts, unusual data transfers, or
32 potential threats in real-time. anomalous user behaviors. This proactive approach enables
33 AI can be a powerful threat detection and prevention tool in the network to respond swiftly and mitigate potential threats
34 6G networks. AI algorithms can analyze large volumes of before they can cause harm [17].
35 network data and identify patterns or anomalies that indicate In summary, the 6G network’s security and privacy concerns
36 potential security threats. By leveraging machine learning are significant and must be adequately addressed to ensure
37 techniques, AI systems can continuously learn and adapt to the network’s safety and secure operation. AI can provide
38 new attack vectors and evolving threats, effectively detecting enhanced security and privacy features, enabling the network
39 known and unknown threats. to mitigate these concerns.
40 Anomaly Detection: AI can be used to identify and flag
unusual behavior on the network, which may be an indicator IV. AI TECHNIQUES FOR NTNs OPTIMIZATION
41
of a potential security breach. The optimization process plays an important role in
42
Predictive Analysis: Predictive analysis uses AI techniques improving the performance and efficiency of wireless
43
to forecast potential security threats and take proactive networks deployed in non-terrestrial environments such as
44
measures to prevent them before they materialize. AI- satellites, drones, and balloons. Optimization can be used to
45
powered predictive analysis can identify indicators that may identify the best settings for the network, such as the power
46
lead to future security incidents or attacks by analyzing level, frequency, and type of antenna, as well as the optimal
47
historical data, network patterns, and security trends. placement of devices. This can lead to improved coverage,
48
Using ML algorithms, predictive analysis models can learn increased bandwidth, and reduced latency. Additionally,
49
from historical data to recognize patterns and correlations optimization can save energy by reducing the amount of
50
associated with security threats. These models can then power required to operate the network.
51
analyze real-time data and identify early warning signs or Numerous approaches were defined for NTN optimization.
52
anomalies that could indicate an imminent security breach. These traditional methods are based on mathematical models
53
The proactive measures taken based on predictive analysis and simulations that use prior knowledge of the network
54
can include strengthening network defenses, implementing parameters and environment to make predictions and
55
additional security controls, or raising alerts to security decisions. For example, network planning involves
56
teams to investigate potential vulnerabilities or suspicious designing the network architecture, coverage area, and
57
58 7
59 For Review Only
60
Page 9 of 42 IEEE Access

1
2
3 capacity based on the expected traffic and user requirements SL involves training a model using labelled data, where the
4 [61]. Link budget analysis involves calculating the power input-output pairs are known in advance. The model can then
5 budget and signal-to-noise ratio for each link in the network be used to make predictions on new inputs. In NTN
6 to ensure reliable and high-quality communication [62]. optimization, SL can be used for various tasks, such as
7 Antenna design involves selecting the antenna’s type, size, channel estimation [70], interference mitigation [71], and
8 and orientation based on the frequency, gain, and radiation resource allocation [72]. For example, a deep neural network
9 pattern [63]-[64]. Signal processing involves filtering, (DNN) can be trained to predict the best channel and power
10 equalizing, and modulating the signal to optimize allocation for a given set of users and resources. The
11 transmission and reception [65]. advantage of supervised learning is that it can produce
12 These traditional approaches have been used for a long time, accurate results with high precision. SL requires labelled
13 but they have some limitations when it comes to NTN data, where the input-output pairs are known in advance.
14 optimization. First, they require prior knowledge of the However, in NTN optimization, it may be difficult to obtain
15 network parameters and environment, which may not be labelled data, as the network operates in dynamic and
16 accurate or up-to-date, especially in dynamic and uncertain uncertain conditions, and the ground truth may not be
17 environments. Second, they are not adaptive enough to available. In addition, the labelled data may not be
18 handle the changing conditions and demands of the network, representative of the entire network, leading to biased or
19 as they rely on fixed and predefined models and rules. Third, inaccurate models. SL is shown in Fig. 4.
20 they may not be able to handle the heterogeneity and UL involves training a model using unlabeled data, where
21 complexity of the network, as they assume a uniform and the input-output pairs are not known in advance. The model
22 idealized network topology and behavior. Therefore, can then discover patterns and structures in the data and
23 exploring new approaches based on AI techniques can group them into clusters or categories. In NTN optimization,
24 overcome these limitations and provide more accurate and unsupervised learning can be used for various tasks, such as
25 adaptive solutions for NTN optimization [66]. These AI anomaly detection [73], network clustering [74], and traffic
26 techniques can learn from the data and adapt to changing analysis [75]. For example, a self-organizing map can be
27 environments. The AI techniques can be broadly classified used to cluster the satellites or drones based on their location
28 into three categories: supervised learning (SL) [67], and connectivity [76]. The advantage of unsupervised
29 unsupervised learning (UL) [68], and RL [69]. The basic learning is that it can find hidden patterns in data that may
30 architecture of AI can be shown in Fig. 3. not be easily visible. Unsupervised learning does not require
31 labelled data, but it relies on the assumption that the data has
32 some inherent structure or pattern that can be discovered.
33 However, in NTN optimization, the network may be too
34 complex or heterogeneous, and the data may not have clear
35 patterns or clusters that can be easily identified. In addition,
36 unsupervised learning may suffer from the problem of
37 overfitting, where the model memorizes the data instead of
38 learning the underlying structure. The basic structure of UL
39 is seen in Fig. 5.
40
41
42
43
44
45
46
47
48
Figure 4. Supervised Learning
49
50
51
52
53
54
55
56
Figure 5. Unsupervised Learning
57 Figure 3. Artificial Intelligence Architecture
58 8
59 For Review Only
60
 IEEE Access Page 10 of 42

1
2
3 RL involves training an agent to interact with an functions based on the new observations, which can result in
4 environment and learn an optimal policy to maximize a more robust and efficient operations.
5 cumulative reward [77]. In NTN optimization, RL can be As summarized in Table 3. SL can address some issues of
6 used for various tasks, such as beamforming, routing, and NTN optimization, such as anomaly detection, channel
scheduling. For example, a Q-learning algorithm can be used estimation, interference mitigation, and resource allocation.
7
to find the optimal beamforming angles and power levels for UL can address issues like network clustering and traffic
8
a given set of antennas and users [78]. analysis. However, neither of them can address the dynamic
9
One of the advantages of RL is that it can handle dynamic and uncertain environment of NTN optimization, which is a
10 critical factor for NTNs.
11 environments with large state and action spaces. However, it
is important to note that RL is typically used for discrete state DRL, on the other hand, can address all of the above issues,
12 as it can learn the optimal policy for complex and dynamic
13 and action spaces and is not well-suited for problems with
continuous state and action spaces. Additionally, while environments where the state and action spaces are high-
14 dimensional and continuous. DRL can be used for
15 RL can handle dynamic environments but may struggle with
large and complex ones. Overall, while RL has shown beamforming, routing, and scheduling tasks requiring real-
16 time decision-making and optimization. Therefore, RL,
17 promising results in a range of domains, it is crucial to
consider its applicability to the specific problem. In the especially DRL, is preferred for NTN optimization in 6G
18 networks, where the goal is to provide reliable and high-
19 context of NTN optimization, RL can be a powerful tool, but
it may not always be the best choice depending on the speed connectivity to remote and underserved areas.
20
21 complexity and nature of the problem. A basic representation
of RL is shown in Fig. 6. A. POSSIBLE APPLICATIONS OF AI IN NTNs
22 AI has the potential to revolutionize NTN optimization in 6G
23 DRL is an extension of RL that has been shown to be more
effective in handling complex and high-dimensional networks by enabling faster and more efficient
24 communication. There are several possible applications of
25 problems. In RL, the policy and value functions are
represented using simple linear or nonlinear models, AI in NTNs for 6G adoption, including Terahertz
26 Communications, Optical Wireless Communication, Remote
27 whereas, in DRL, deep neural networks are used to represent
these functions. This allows the DRL models to learn more Sensing, and Industrial Internet of Things (IIoT).
28 One possible application is satellite communication. AI can
29 complex and abstract representations of the state and action
spaces, which can result in better performance and more be used to optimize communication between satellites,
30 ground stations, and users. It can also help reduce latency
31 efficient exploration of the environment.
DRL has been able to achieve state-of-the-art performance and interference, thereby improving the system's overall
32 performance. For instance, AI can be used to optimize the
33 in various domains, such as robotics, games, wireless
communication, and natural language processing. These scheduling and routing of satellite links and predict and
34 mitigate the effects of atmospheric attenuation and weather
35 successes have motivated researchers to explore the use of
DRL for NTN optimization, where the state and action conditions.
36 Another application is drone-based connectivity. AI can be
37 spaces are often high-dimensional and continuous. By using
DRL, it is possible to learn an optimal policy for resource used to optimize the coverage and capacity of drone-based
38 networks, which can be used for various applications, such
39 allocation [79], [80], [81],[82], routing [83], and scheduling
[84] that operate in dynamic and uncertain conditions in as search and rescue, environmental monitoring, and
40 precision agriculture. For example, AI can be used to
41 NTN.
42 Another advantage of DRL is that it can handle the
heterogeneity and complexity of the network, which is Table 3. Dynamics of SL, UL and RL
43 Issues SL UL RL
44 difficult to model using traditional approaches. For example,
in a satellite network, the number and position of the Dynamic and uncertain environment x x ✓
45
satellites may change over time, the users may move in Heterogeneity and complexity of the network x ✓ ✓
46
different directions, and the signal quality may vary High-dimensional and continuous state and x x ✓
47
depending on the atmospheric conditions. DRL can adapt to action spaces
48
these changes by continuously updating the policy and value Anomaly detection ✓ ✓ x
49
Channel estimation ✓ x x
50
51 Interference mitigation ✓ x x
52 Resource allocation ✓ x x
53 Network clustering x ✓ x
54 Traffic analysis x ✓ x
55 Beamforming x x ✓
56 Routing x x ✓
Figure 6. Reinforcement Learning ✓
57 Scheduling x x
58 9
59 For Review Only
60
Page 11 of 42 IEEE Access

1
2
3 optimize the trajectory and positioning of drones and manage planning and design, where AI can be used to optimize the
4 the handover between drones and ground stations. placement and configuration of network elements such as
5 AI can also be used to optimize the altitude, trajectory, and antennas, base stations, and routers. This can help to improve
6 communication of balloon-based networks. AI can help coverage, capacity, and QoS while minimizing costs and
7 optimize the positioning and deployment of balloons and energy consumption.
8 predict and mitigate the effects of atmospheric conditions Another area is resource allocation, where AI can optimize
9 and other environmental factors. network resource allocation, such as bandwidth, power, and
10 Finally, AI can be used to optimize energy consumption, data spectrum. This can help to improve network efficiency and
11 processing, and transmission of IoT and remote sensing capacity while minimizing interference and congestion.
12 devices. This can be particularly important in limited energy AI can also be used for fault detection and diagnosis,
13 applications, such as remote and off-grid locations. AI can automatically identifying and diagnosing network faults and
14 be used to optimize the routing and processing of data and anomalies and providing recommendations for remedial
15 predict and mitigate the effects of interference and other action. This can help to reduce downtime and improve
16 environmental factors. network availability and reliability.
17 Table 4 summarizes the possible applications of AI in NTNs Finally, AI can be used for network security, automatically
18 for 6G networks. detecting and preventing security threats such as intrusion,
19 malware, and denial-of-service attacks. This can help to
20 B. AI-POWERED NETWORK OPTIMIZATION AND protect network assets and data and maintain user privacy
MANAGEMENT and trust.
21
AI-powered network optimization and management involves In summary, AI-powered network optimization and
22
the application of AI techniques to improve the performance, management have the potential to transform the way we
23
efficiency, and reliability of 6G communication networks. design, operate, and maintain communication networks. By
24
These techniques include ML, DL, and RL, which can be providing automated monitoring, analysis, and decision-
25
used to learn from network data, predict network behavior, making tools, AI can help improve network performance,
26
and make automated decisions. efficiency, and reliability and enable new applications and
27
One of the key benefits of AI-powered network optimization services.
28
is its ability to handle the complexity and heterogeneity of
29
modern communication networks. With the rise of 5G and C. AI-POWERED RECONFIGURABLE INTELLIGENT
30
the advent of 6G, networks are becoming more diverse and SURFACE WITH NTNs
31 Reconfigurable intelligent surface (RIS) is a new technology
dynamic, with a wide range of devices, protocols, and
32 that can manipulate electromagnetic waves by dynamically
services. AI can help manage this complexity by providing
33 changing their surface properties [85]. They consist of a
automated monitoring, analysis, and optimization tools.
34 planar array of small, passive, and tunable elements that can
There are several areas in which AI-powered network
35 manipulate electromagnetic waves to improve the
optimization can be applied. One of these is network
36 performance of wireless networks. RIS technology can
37 Table 4. AI Possible NTN Applications enhance wireless networks' coverage, capacity, and QoS by
38 Application Description adjusting the phase, amplitude, and direction of
39 Optimize communication between satellites, ground electromagnetic waves.
40 Satellite
stations, and users; reduce latency and interference; RIS can be used to optimize the signal strength and quality
41 communication
optimize scheduling and routing of satellite links. of NTN, improving its coverage and capacity. When RIS
42 technology is combined with NTNs, the resulting system can
Optimize coverage and capacity of drone-based
43 improve wireless communication performance even more.
networks; manage handover between drones and
44 Drone-based
This performance improvement is especially beneficial in
ground stations; enable search and rescue,
45 connectivity
areas with poor network coverage. RIS is also more cost-
environmental monitoring, and precision
46 effective than other methods of improving NTN signal
agriculture.
47 strength, making it a great solution for both small and large
Optimize altitude, trajectory, and communication of
48 networks [86]. Recent research has shown that integrating AI
balloon-based networks; provide connectivity to
49 Balloon-based
with RIS technology can improve the performance and
remote and underserved areas; predict and mitigate
50 networks
reliability of NTNs, such as those used for satellite
effects of atmospheric conditions and environmental
51 communications, deep-space exploration, and interplanetary
factors.
52 networks. These networks face unique challenges, such as
Optimize energy consumption, data processing, and
53 limited bandwidth, high latency, and harsh environmental
transmission of IoT and remote sensing devices;
54 IoT and remote
conditions that can impede data transmission and
optimize data routing and processing; predict and
55 sensing
connectivity [87]-[88].
mitigate effects of interference and environmental
56
factors.
57
58 10
59 For Review Only
60
 IEEE Access Page 12 of 42

1
2
3 AI-powered RIS can help to overcome some of these Crowdsensing, a collective sensing paradigm, taps into the
4 challenges by optimizing the wireless channel in real-time. power of connected individuals and their mobile devices to
5 By leveraging AI algorithms, RIS can adapt to changing gather data about the environment [96]. Through mobile
6 network conditions and dynamically adjust its configuration apps or wearable devices, individuals contribute data on
7 to improve signal quality, increase network capacity, and various aspects like traffic conditions, air quality, noise
8 reduce interference [89]-[90]. For example, AI-powered RIS levels, and social behaviors. This distributed data collection
9 can intelligently direct signals to avoid obstacles and approach provides extensive coverage and delivers real-time
10 optimize coverage, which can be especially useful in deep- insights into the physical world, empowering communities
11 space communications where distance can severely attenuate with a deeper understanding of their surroundings.
12 signal strength [91]. Moreover, AI-powered RIS can enable The relationship between edge computing, crowdsensing,
13 more efficient use of the available spectrum, which is an NTN, and AI is pivotal in advancing the capabilities of
14 important resource in NTNs [92]. By intelligently managing modern systems. In [97], a Stackelberg game-based
15 signal strength and direction, RIS can increase the capacity computation offloading method can be integrated into edge
16 of the existing spectrum, reduce the risk of interference, and computing and crowdsensing systems to optimize data
17 enable more reliable data transmission. processing at the network edge. Building upon this, the [98]
18 One of the key benefits of AI-powered RIS is its ability to proposes a Stackelberg game approach with the assistance of
19 create a more dynamic and adaptive wireless environment. UAVs, enhancing the offloading process in mobile edge
20 By constantly analyzing and adjusting the wireless channel, computing networks. The proposed approaches optimize RA
21 RIS can ensure that the network remains resilient and and enhance system performance by leveraging the insights
22 reliable, even in the face of equipment failure or gained from crowdsensing data and utilizing AI techniques,
23 environmental disturbances. Additionally, RIS can be used such as ML. Furthermore, [99] focuses on cost-
24 to create more secure and resilient NTNs by controlling and minimization-oriented computation offloading and service
25 manipulating the propagation of radio waves to prevent caching, employing an advanced ML-based approach.
26 unauthorized access and mitigate the risk of cyber-attacks Integrating such methodology with edge computing,
27 [93]. AI-powered RIS also has the potential to revolutionize crowdsensing, and NTN enables efficient resource
28 the field of space exploration. By enabling more advanced management, reduced operational costs, and intelligent
29 and sophisticated communications systems, RIS can improve decision-making based on data collected from NTN-enabled
30 the accuracy and reliability of spacecraft navigation systems, devices and sensors. Collectively, [97]-[99] contribute to
31 enabling more precise targeting and maneuvering. integrating edge computing, crowdsensing, NTN, and AI,
32 Furthermore, RIS can create more efficient and robust data fostering advancements in various domains, including smart
33 transmission systems for deep-space exploration, enabling cities, industrial IoT, and remote sensing applications.
34 more rapid and reliable transmission of scientific data and When edge computing, crowdsensing, NTNs, and AI come
35 images [94]. together, the possibilities for transformative applications
36 Although the integration of AI and RIS technology has great across domains become endless. In the realm of smart cities,
37 promise for enhancing the performance and reliability of this integration enables edge devices connected through
38 NTNs for future advancements in space exploration, satellite NTNs to monitor traffic flow, analyze real-time air quality
39 communications, and interplanetary networks, however, the data, and optimize energy consumption based on demand
40 technical challenges still need to be addressed; ongoing and availability. Advanced AI algorithms identify congestion
41 research and development in this field suggests that AI- hotspots, predict air pollution levels, and recommend
42 powered RIS will play an increasingly important role in efficient routes or dynamically adjust energy usage, fostering
43 shaping the future of wireless technology in non-terrestrial sustainable urban environments.
44 environments. Furthermore, this integration profoundly impacts public
45 safety and emergency response. In the event of natural
46 D. RELATIONSHIP OF EDGE COMPUTING AND disasters or emergencies, the combination of edge computing
47
CROWDSENSING WITH NTN AND AI and crowdsensing allows for rapid data collection and
With its decentralized approach, edge computing analysis. By harnessing data from mobile devices, sensors,
48
revolutionizes data processing and analysis by bringing and surveillance systems, AI algorithms can detect critical
49
computational power closer to the edge of the network. This situations in real-time, such as identifying earthquake-prone
50
proximity to data sources enables real-time insights and areas or predicting the spread of wildfires. This timely
51
analysis, minimizing latency and bandwidth usage and information enables emergency responders to allocate
52
making it ideal for time-sensitive applications. By leveraging resources effectively, coordinate evacuation plans, and save
53
a network of devices such as edge servers, gateways, and lives. The healthcare sector benefits tremendously from this
54
edge sensors, edge computing infrastructure collaboratively synergy. Crowdsensing facilitates remote patient monitoring,
55
processes and filters data at the edge, enhancing efficiency collecting vital signs, medication adherence, and activity
56
and scalability [95]. levels through wearable devices. Edge computing and AI
57
58 11
59 For Review Only
60
Page 13 of 42 IEEE Access

1
2
3 process this data in real-time, enabling healthcare providers predictive maintenance. In the past few years, researchers
4 to monitor patients' health conditions, detect anomalies, and have explored various AI-based approaches, such as DL, RL,
5 intervene promptly. Moreover, AI-powered algorithms and DRL to improve network performance [66],[100]. These
6 analyze population-level health data to identify patterns, approaches have been used for various applications such as
7 predict disease outbreaks, and allocate resources efficiently intelligent network slicing, load balancing, and resource
8 during public health emergencies, ultimately saving lives. allocation, enabling the network to operate optimally in
9 The symbiotic relationship between edge computing, dynamic and unpredictable environments.
10 crowdsensing, NTNs, and AI also extends to agriculture, The integration of AI into NTNs has also led to the
11 where edge devices combined with NTNs monitor soil development of new and innovative network architectures
12 moisture levels, temperature, and crop health. AI algorithms that can efficiently handle the vast amounts of data generated
13 analyze this data to optimize irrigation, predict crop yields, by 6G networks. One such architecture is the AI-powered
14 and identify disease or pest outbreaks, empowering farmers Open Radio Access Network (ORAN) [101], which utilizes
15 to make informed decisions, increase productivity, and AI to optimize network coverage and capacity and minimize
16 reduce resource waste. interference and latency. The AI-powered ORAN also
17 Edge computing enables real-time equipment performance allows for dynamic network configuration, enabling the
18 and predictive maintenance monitoring in industrial settings, network to adapt to changing network conditions in real-
19 ensuring optimal productivity and minimizing downtime. time. Furthermore, researchers have also explored the use of
20 Coupled with crowdsensing and AI, this convergence AI for network security and privacy in 6G networks. AI-
21 enhances worker safety by detecting hazards in real-time and based security solutions such as Intrusion Detection Systems
22 providing immediate alerts, mitigating accidents, and (IDS) and Intrusion Prevention Systems (IPS) can identify
23 improving overall workplace security. and prevent security threats in real-time, improving the
24 However, as these technologies advance, addressing network's overall security [102]. AI-powered privacy
25 important considerations such as data privacy and security is solutions such as differential privacy can also protect user
26 crucial. Robust security measures must be implemented to data while enabling network operators to collect valuable
27 protect sensitive information as data is collected and data for network optimization and management. In addition,
28 processed at the edge. Additionally, establishing data AI is being used to address energy efficiency challenges in
29 governance frameworks and ethical guidelines is imperative 6G networks [66]. With the increasing demand for high-
30 to ensure the responsible and transparent use of data, speed connectivity, the energy consumption of 6G networks
31 building trust, and fostering the long-term sustainability of is expected to increase significantly. However, AI-based
32 these technologies. approaches such as energy-efficient resource allocation,
33 In summary, the relationship between edge computing, dynamic sleep mode management, and intelligent power
34 crowdsensing, NTNs, and AI represents a powerful force control can help to reduce energy consumption and improve
35 that has the potential to revolutionize various industries and the overall energy efficiency of the network.
36 significantly improve the quality of life. By bringing
37 computational power closer to the source of data generation, A. NETWORK PLANNING AND OPTIMIZATION
38 leveraging the crowd's collective intelligence, extending The use of AI algorithms in network planning and
39 connectivity beyond traditional networks, and harnessing the optimization offers numerous benefits, including efficient
40 analytical capabilities of AI, this integration unlocks network management, automatic optimization, and
41 unprecedented opportunities for real-time insights, intelligent decision-making. AI-driven network planning can
42 intelligent decision-making, and enhanced efficiency. As help ensure optimal resource allocation, reduce the cost of
43 these technologies continue to advance and intertwine, their network deployment, and enhance network performance.
44 intricate synergy will shape the future of connectivity, data With the ability to analyze vast amounts of data from
45 analytics and redefine the way we interact with the digital multiple sources, such as user behavior, network traffic, and
46 world, paving the way for a smarter, more sustainable future. resource utilization, AI can enable efficient network
47 This transformative potential is not only limited to large- management and improve users' QoS.
48 scale industries but also has the ability to empower ML algorithms, such as Support Vector Machines (SVMs)
49 individuals and local communities, democratizing access to and Random forests, have been utilized to predict network
50 information and fostering innovation at every level. traffic, optimize routing, and allocate resources effectively.
51 SVMs can be used to classify network traffic based on
V. RECENT ADVANCES IN AI-POWER NTN FOR 6G various criteria, such as the source of the traffic, the type of
52
NETWORK data being transmitted, and the destination of the traffic. This
53 In recent years, significant progress has been made in
54 enables service providers to allocate network resources more
developing AI-powered NTNs. The use of AI in NTNs efficiently and reduce congestion. On the other hand,
55 provides numerous benefits, such as intelligent network
56 Random Forests can be used to predict network traffic based
management, automatic network optimization, and on historical data and network behavior, enabling service
57
58 12
59 For Review Only
60
 IEEE Access Page 14 of 42

1
2
3 providers to allocate network resources more effectively. the effectiveness of AI-based approaches in 6G networks.
4 DRL and Genetic Algorithm (GA) have also been applied to One such study focused on using AI in network slicing,
5 improve the network architecture and adapt to changing which allows network operators to partition the network into
6 environmental conditions, enhancing the network's overall virtual slices with customized functionalities and resources.
7 performance. DRL is an AI technique that involves training The study demonstrated that the use of AI-based approaches
8 an agent to act in an environment to maximize a cumulative such as DL and CNNs could significantly improve network
9 reward. This approach has been used to optimize network slicing performance, enabling better resource utilization and
10 routing, enabling the network to adapt to changing traffic enhancing the QoS for end-users [103].
11 patterns and reduce latency. Conversely, GA has been used The authors in [110] propose a cooperative transmission
12 to optimize network parameters such as the number of BSs scheme between a satellite and an aerial BS based on two
13 and the frequency allocation, enabling the network to operate unsupervised ML algorithms, namely K-means and K-
14 optimally in dynamic and unpredictable environments. medoids. The simulation results show that the proposed
15 AI can analyze data from multiple sources, such as user approach demonstrates promising gains in terms of spectral
16 behavior, network traffic, and resource utilization to identify efficiency and system sum rate. In [111], the authors
17 areas where network resources are being underutilized or consider the issue of mobility management due to the
18 overutilized. This can help service providers optimize their movements of LEO satellites. They group user equipment in
19 network architecture and allocate resources more efficiently, different clusters based on K-means clustering algorithm and
20 reducing the overall cost of network deployment. decide handover process based on the distance from its cell
21 Furthermore, AI-driven network planning can ensure center.
22 efficient resource allocation and reduce the cost of network In [104], the authors focused on using RL to optimize
23 deployment, thereby enabling service providers to deliver network resource allocation in dynamic and unpredictable
24 high-quality services while reducing operational costs. environments. The study proposed a new approach to
25 The application of AI in network planning and optimization resource allocation, which combined RL with the GA to
26 is expected to become more widespread with the enable the network to adapt to changing environmental
27 advancement of technology and the increasing demand for conditions. The experimental results demonstrated that the
28 high-speed and reliable connectivity in 6G networks. proposed approach can significantly improve network
29 However, there are several challenges that must be addressed performance by optimizing the allocation of resources to
30 to ensure the effective implementation of AI in network different network functions.
31 planning and optimization. These challenges include the The work in [105] adopted similar MARL strategy for
32 need for standardized data formats, the integration of AI dynamic network slicing. However, the proposed approach
33 algorithms with existing network infrastructure, and the accelerated the policy deployment by integrating a transfer
34 development of ethical guidelines for the use of AI in learning method. The results showed that the proposed
35 network management. approach can achieve 27% better network performance and
36 In summary, AI-driven network planning and optimization utilization compared to traditional approaches.
37 have emerged as critical research areas in developing AI- In [106], an AI-based approach for energy-efficient resource
38 powered NTNs for 6G networks. AI algorithms offer allocation in 6G networks is proposed. The proposed
39 numerous benefits, such as efficient network management, approach utilized a GA to optimize the allocation of network
40 automatic optimization, and intelligent decision-making. ML resources, considering each network function's energy
41 algorithms such as SVMs and Random Forests have been consumption. The results demonstrated that the proposed
42 utilized to predict network traffic and allocate resources approach can significantly reduce the network's energy
43 effectively, while DRL and GA have been applied to consumption while maintaining the desired level of network
44 improve the network architecture and adapt to changing performance.
45 environmental conditions. AI-driven network planning can Another study by [107] focused on using AI in network
46 ensure efficient resource allocation and reduce the cost of function virtualization (NFV) to improve network efficiency
47 network deployment, thereby enabling service providers to and resource utilization. The study proposed an AI-based
48 deliver high-quality services while reducing operational approach for the placement of virtual network functions
49 costs. However, there are several challenges that must be (VNFs) in the network, which utilized a DRL algorithm. The
50 addressed to ensure the effective implementation of AI in results showed that the proposed approach can significantly
51 network planning and optimization. improve network performance and reduce the resource
52 utilization of the network.
53 B. CASE STUDIES AND EXPERIMENTAL RESULTS In a different study [108], the authors proposed an AI-based
54 With the rapid development of 6G networks, AI-powered approach for network traffic prediction in 6G networks. The
55 NTNs have emerged as a critical area of research to improve proposed approach utilized a two-dimensional CNN-based
56 network management, optimization, and performance. long short-term memory (LSTM) to predict network traffic,
57 Several recent case studies have been conducted to evaluate which can enable better resource allocation and network
58 13
59 For Review Only
60
Page 15 of 42 IEEE Access

1
2
3 management. The experimental results showed that the AI-powered NTNs in 6G networks.
4 proposed approach can achieve higher prediction accuracy Multi-objective optimization for AI-powered NTNs: 6G
5 compared to traditional approaches. networks are expected to support diverse applications with
6 In [109], the authors focused on the use of AI in network different quality-of-service requirements. Multi-objective
7 function placement for edge computing in 6G networks. The optimization can help balance conflicting objectives, such as
8 study proposed an AI-based approach for the placement of energy efficiency, spectral efficiency, and reliability, in AI-
9 network functions in the network, which utilized a Q- powered NTNs.
10 learning algorithm. The results showed that the proposed Intelligent network planning and deployment: AI-
11 approach could significantly improve network performance. powered NTNs can leverage intelligent algorithms to
12 The above-explained case studies are summarized in Table automate the network planning and deployment process.
13 5. This includes using ML to identify the optimal network
14 topology, antenna placement, and resource allocation to
VI. CHALLENGES AND OPPORTUNITIES IN AI- minimize interference and maximize coverage.
15
POWERED NTNs FOR 6G NETWORKS AI-driven resource allocation and management: The use
16 The development of the 6G networks is rapidly advancing,
17 of AI can help to efficiently allocate and manage network
and AI has emerged as a key technology in realizing the resources, including frequency bands, power levels, and
18 vision of intelligent and efficient 6G networks. AI-powered
19 computing resources. This can improve network
network management is expected to provide numerous performance and reduce energy consumption.
20 benefits, including improved network performance,
21 Federated learning for distributed NTNs: Federated
enhanced user experience, and reduced operational costs. learning is an ML technique allowing devices to
22 However, achieving these benefits comes with significant
23 collaboratively learn a shared model without sharing data.
challenges and requires overcoming various technical and This can be used to train AI models for NTNs in a privacy-
24 practical obstacles. Therefore, this survey paper discusses
25 preserving and energy-efficient manner.
the challenges and opportunities in AI-powered network Network slicing for AI-powered NTNs: Network slicing
26 management for 6G networks.
27 allows multiple logical networks to be created on top of a
28 shared physical infrastructure. This can help to meet the
A. KEY RESEARCH DIRECTIONS FOR AI-POWERED
29 diverse requirements of different applications in AI-powered
NTNs
30 NTNs, such as low latency, high bandwidth, and high
As AI technologies continue to evolve, they offer
31 reliability.
tremendous potential to revolutionize the way future 6G
32 Edge intelligence for AI-powered NTNs: Edge computing
networks are designed, deployed, and managed. AI-powered
33 can perform AI computations at the network edge, reducing
NTNs is an emerging research area that aims to leverage AI
34 latency and improving energy efficiency. This includes using
techniques to enhance the performance and efficiency of 6G
35 edge devices, such as BSs and user equipment, to perform AI
networks. However, realizing the full potential of AI-
36 computations and make real-time decisions.
powered NTNs requires identifying key research directions
37 Explainable AI for NTNs: The development of explainable
that can overcome the challenges posed by the highly
38 AI can enhance the transparency and interpretability of
dynamic and complex nature of 6G networks. This section
39 NTNs powered by AI. This includes developing algorithms
discusses some of the most promising research directions for
40 that can explain how AI models make decisions and identify
41 Table 5. Summary of AI Use Cases potential biases in the data.
42 AI Related Work(s) Relevance to NTNs
AI-driven security for NTNs: The use of AI can help to
43 [103] - LSTM Smart network slicing, node
improve the security of NTNs, including detecting and
44 SL [106] - CNN clustering or traffic prediction for
mitigating attacks, identifying vulnerabilities, and enhancing
45 [108] - CNN-LSTM increasing energy efficiency.
privacy and data protection.
46 AI for network optimization and self-healing: AI can be
Cooperative transmission
47 used to continuously monitor and optimize network
UL [110] – K-means between terrestrial networks and
48 performance and detect and repair faults automatically. This
NTNs.
49 includes using AI to predict and prevent network failures,
Handover of user equipment
50 optimize routing and traffic flow, and manage network
UL [111] – K-means between satellites and ground
51 congestion.
stations.
52 AI for user experience and behavior analysis: AI can be
[104] - DQN Smart resource management for
53 RL used to analyze user behavior and preferences and predict
[105] - MADRL reducing multi-cell interference.
54 and personalize services and content. This includes using AI
[107] - DQN Computation offloading via
55 to optimize user engagement, enhance content delivery, and
RL [109] - Soft Actor coverage deployment or multi-
56 improve overall user experience. The key research directions
Critic access edge computing (MEC).
57 for AI-Powered NTNs are summarized in Fig. 7.
58 14
59 For Review Only
60
 IEEE Access Page 16 of 42

1
2
3 human factors involved in network management, including
4 user behavior, network administrator skills, and
5 organizational structures. Organizations must ensure that AI-
6 powered networks are designed to meet the needs of the users
7 and the organization. This requires understanding the skills
8 and knowledge of the network administrators and users and
9 the organizational structure and culture.
10 Resource limitations: AI-powered networks may require
11 significant computational resources, which can be a
challenge in resource-constrained environments.
12
Organizations must ensure that AI-powered networks are
13
designed to operate within the resource constraints of the
14
environment. Additionally, organizations must consider AI-
15 powered networks' power consumption and environmental
16 impact.
17 Lack of standardization: There is a lack of standardization
18 in AI-powered network technologies and protocols, which
19 can hinder interoperability and scalability. Organizations
20 must work together to develop standards for AI-powered
21 networks that promote interoperability and scalability.
22 Bias and fairness: AI-powered networks must address bias
23 and fairness issues in decision-making, particularly in areas
24 Figure 7. Key Research Directions
such as resource allocation and network management.
25 Organizations must ensure that AI-powered networks are
26 designed to be fair and unbiased and that decisions made by
B. CHALLENGES AND OPEN ISSUES IN DEPLOYING
27 AI-POWERED NTNs the network are transparent and explainable.
28 While AI-powered NTNs have significant potential, there Ethical concerns: AI-powered networks must consider
29 are also several challenges and open issues that must be ethical concerns related to the use of AI, including
30 addressed. These include but are not limited: transparency, accountability, and potential unintended
31 Data privacy and security: AI techniques require large consequences. Organizations must ensure that AI-powered
32 amounts of data to train models, but ensuring data privacy networks are designed to be transparent and accountable and
33 and security is a significant concern. Organizations must that they do not have unintended negative consequences.
34 ensure that sensitive data is protected and not accessible to Regulatory and legal frameworks: AI-powered networks
35 unauthorized personnel. Additionally, organizations must be must comply with regulatory and legal frameworks related
36 transparent about the data they are collecting, how it will be to privacy, security, and other issues. Organizations must
37 used, and who has access to it. ensure that AI-powered networks are designed to meet
38 Trustworthiness: AI-powered networks must be regulatory and legal requirements and that they are
39 trustworthy and reliable, with clear accountability transparent and accountable to regulators and other
40 mechanisms in place to ensure the network functions as stakeholders.
41 intended. Organizations must ensure that AI-powered Lack of domain expertise: Developing AI-powered
42 networks are transparent and explainable so that users can networks requires domain expertise in networking and AI,
43 understand how the network is making decisions. which can be challenging to find in a single individual or
44 Additionally, organizations must be transparent about the organization. Organizations must ensure that they have
45 limitations of AI-powered networks so that users understand access to the necessary expertise to develop and deploy AI-
46 when the network may not be appropriate for specific use powered networks and that they are able to collaborate with
47 cases. other organizations to share expertise and resources. We
48 Integration and interoperability: AI-powered networks have summarized these challenges and open issues in Fig. 8
49 must be able to integrate and interoperate with existing for better and easier understanding.
50 networks and devices, including legacy systems.
51 Organizations must ensure that AI-powered networks are C. OPPORTUNITIES FOR FUTURE RESEARCH
52 designed to work with existing infrastructure and can operate Despite the challenges and open issues, AI-powered NTNs
53 within existing networks. Additionally, organizations must for 6G networks present numerous future research
54 ensure that AI-powered networks can communicate with opportunities. In the following, we list such opportunities:
55 other devices and networks using standard protocols. Development of new AI techniques: AI-powered NTNs for
56 Human factors: AI-powered networks must consider the 6G networks require the development of new AI techniques
57
58 15
59 For Review Only
60
Page 17 of 42 IEEE Access

1
2
3 For example, AI-powered network management can
4 optimize network performance and resource allocation for
5 virtual reality applications or enable intelligent traffic
6 management for connected vehicles.
7 Standardization: This is an important area for future
8 research in AI-powered NTNs for 6G networks.
9 Standardization involves the development of consistent and
10 uniform technology and protocols across the industry. This
11 is important to ensure the interoperability and scalability of
12 AI-powered network technologies. In other words,
13 standardization ensures that different AI-powered network
14 management systems can work together seamlessly and that
15 they can be easily scaled up or down depending on the
16 network size and complexity. To achieve standardization,
17 researchers can work on developing new standards and
18 protocols that can facilitate the adoption and deployment of
19 AI-powered NTNs for 6G networks. This can involve
20 creating standardized communication protocols, data
21 formats, and interfaces that can be used across different
22 network management systems. Researchers can also work
23 with industry stakeholders and regulatory bodies to create
24 Figure 8. Summary of Challenges and Open Issues industry-wide standards and guidelines that can promote the
25 development and adoption of AI-powered NTNs for 6G
that can address the unique challenges of 6G networks. These networks. By achieving standardization, AI-powered NTNs
26
challenges include the need for real-time processing, large- for 6G networks can become more efficient, effective, and
27
scale data analysis, and dynamic network management. reliable, ultimately leading to a better user experience and
28
Research in this area can lead to the development of new improved network performance.
29
algorithms and models that can provide better performance, Sustainability: AI-powered NTNs for 6G networks must
30
accuracy, and efficiency in network management. also consider the sustainability of network management,
31
Integration with emerging technologies: The integration including the use of renewable energy sources, the reduction
32
of AI-powered network management with emerging of energy consumption, and the optimization of network
33
technologies such as edge computing, blockchain, and the resources. Researchers can work on developing new
34
IoT can create intelligent and efficient networks. Research in techniques and protocols that can ensure the sustainability of
35
this area can lead to the development of new architectures AI-powered network management.
36
and protocols that can leverage the benefits of these Explainability and transparency: AI-powered network
37
emerging technologies. For example, edge computing can management should be explainable and transparent to ensure
38
reduce latency and improve network response time, while trust and accountability. Researchers can work on
39
blockchain can provide a secure and decentralized network developing new techniques and models that can provide
40
management. insights into how AI-powered network management makes
41
Adoption of new network architecture: AI-powered NTNs decisions and provide explanations for these decisions.
42
for 6G networks requires adopting new network Resilience: AI-powered NTNs for 6G networks must be
43
architectures optimized for AI-powered network resilient to cyber-attacks and other security threats.
44
management. Research in this area can lead to the Researchers can work on developing new techniques and
45
development of new architectures and protocols that can protocols that can ensure the resilience of AI-powered
46
provide better scalability, reliability, and performance in network management and protect against these threats.
47
network management. For example, software-defined Collaboration and partnerships: AI-powered NTNs for
48
networks (SDNs) and NFV can provide greater flexibility 6G networks require collaboration and partnerships between
49
and programmability in network management, which AI- academia, industry, and government to address the complex
50
powered techniques can leverage. challenges involved in network management. Research in
51
Exploration of new applications: AI-powered NTNs for this area can lead to the development of new collaborative
52
6G networks can enable a wide range of new applications, models and frameworks that can facilitate sharing of
53
such as immersive virtual reality, connected vehicles, and knowledge, resources, and expertise.
54
smart cities. Research in this area can lead to the Sustainability: AI-powered NTNs for 6G networks must
55
development of new use cases and applications that can also consider the sustainability of network management,
56
leverage the benefits of AI-powered network management. including the use of renewable energy sources, the reduction
57
58 16
59 For Review Only
60
 IEEE Access Page 18 of 42

1
2
of energy consumption, and the optimization of network [7] S. Kota and G. Giambene, “6G Integrated Non-Terrestrial Networks:
3 Emerging Technologies and Challenges,” 2021 IEEE Int. Conf.
4 resources. Research in this area can lead to the development Commun. Work. ICC Work. 2021 - Proc., 2021, doi:
5 of new techniques and protocols that can ensure the 10.1109/ICCWorkshops50388.2021.9473743.
sustainability of AI-powered network management. [8] M. Giordani and M. Zorzi, “Non-Terrestrial Networks in the 6G Era:
6 Challenges and Opportunities,” IEEE Netw., vol. 35, no. 2, pp. 244–
7 In summary, AI-powered NTNs for 6G networks present
251, 2021, doi: 10.1109/MNET.011.2000493.
8 significant challenges and opportunities. Addressing the [9] R. Giuliano and E. Innocenti, “Machine Learning Techniques for
challenges and leveraging the opportunities will require Non-Terrestrial Networks,” Electron., vol. 12, no. 3, 2023, doi:
9 10.3390/electronics12030652.
10 significant research and development efforts, but the
[10] C. X. Wang, M. Di Renzo, S. Stańczak, S. Wang, and E. G. Larsson,
11 potential benefits of intelligent and efficient networks are “Artificial Intelligence Enabled Wireless Networking for 5G and
12 worthy of these efforts. Beyond: Recent Advances and Future Challenges,” IEEE Wirel.
Commun., vol. 27, no. 1, pp. 16–23, 2020, doi:
13 10.1109/MWC.001.1900292.
VII. CONCLUSIONS
14 [11] O. T. H. Alzubaidi et al., “Interference Challenges and Management
In this paper, we comprehensively surveyed AI-powered in B5G Network Design: A Comprehensive Review,” Electron., vol.
15
NTNs for 6G wireless communication networks. We first 11, no. 18, 2022, doi: 10.3390/electronics11182842.
16 [12] A. Vanelli-Coralli, A. Guidotti, T. Foggi, G. Colavolpe, and G.
discussed the background and motivation for using NTNs in
17 Montorsi, “5G and beyond 5G Non-Terrestrial Networks: Trends and
6G networks and their deployment and management
18 research challenges,” 2020 IEEE 3rd 5G World Forum, 5GWF 2020
challenges. We then highlighted the key features and - Conf. Proc., pp. 163–169, 2020, doi:
19 10.1109/5GWF49715.2020.9221119.
requirements of NTNs, including the role of
20 [13] M. Mozaffari, X. Lin, and S. Hayes, “Toward 6G with Connected
LEO/MEO/GEO satellites, HAPS, and UAVs in 6G
21 Sky: UAVs and Beyond,” IEEE Commun. Mag., vol. 59, no. 12, pp.
networks. Then, we discussed various AI techniques for 74–80, 2021, doi: 10.1109/MCOM.005.2100142.
22 M. Banafaa et al., “6G Mobile Communication Technology:
optimizing NTNs, including ML, DL, RL, and DRL, and [14]
23 Requirements, Targets, Applications, Challenges, Advantages, and
explored the recent advances in AI-powered NTNs for 6G
24 Opportunities,” Alexandria Eng. J., vol. 64, pp. 245–274, 2023, doi:
networks. We also identified key research directions,
25 10.1016/j.aej.2022.08.017.
challenges, and opportunities in AI-powered NTNs for 6G [15] B. Barakat et al., “6G opportunities arising from internet of things use
26 cases: A review paper,” Futur. Internet, vol. 13, no. 6, pp. 1–29, 2021,
networks.
27 doi: 10.3390/fi13060159.
Our survey sheds the light that AI-assisted techniques are the
28 [16] G. Geraci, D. Lopez-Perez, M. Benzaghta, and S. Chatzinotas,
key ingredients for realizing 6G since AI facilitates NTNs to “Integrating Terrestrial and Non-terrestrial Networks: 3D
29 Opportunities and Challenges,” IEEE Commun. Mag., pp. 1–7, 2022,
operate optimally in dynamic and unpredictable
30 doi: 10.1109/MCOM.002.2200366.
environments. However, the reliability of AI models depends
31 [17] M. Wang, T. Zhu, T. Zhang, J. Zhang, S. Yu, and W. Zhou, “Security
on the quantity and quality of training dataset. How to create and privacy in 6G networks: New areas and new challenges,” Digit.
32 Commun. Networks, vol. 6, no. 3, pp. 281–291, 2020, doi:
a synthetic training dataset which mimics real-world 6G
33 10.1016/j.dcan.2020.07.003.
observations will be crucial. In this context, utilizing digital
34 [18] H. Tataria, M. Shafi, A. F. Molisch, M. Dohler, H. Sjoland, and F.
twin to generate virtual replicas of real-world 6G Tufvesson, “6G Wireless Systems: Vision, Requirements,
35
communications data could be an interesting avenue for Challenges, Insights, and Opportunities,” Proc. IEEE, vol. 109, no. 7,
36 pp. 1166–1199, 2021, doi: 10.1109/JPROC.2021.3061701.
future research.
37 [19] S. Chen, Y. C. Liang, S. Sun, S. Kang, W. Cheng, and M. Peng,
38 “Vision, Requirements, and Technology Trend of 6G: How to Tackle
REFERENCES the Challenges of System Coverage, Capacity, User Data-Rate and
39 Movement Speed,” IEEE Wirel. Commun., vol. 27, no. 2, pp. 218–
[1] X. Lin, “An Overview of 5G Advanced Evolution in 3GPP Release
40 18,” IEEE Commun. Stand. Mag., vol. 6, no. 3, pp. 77–83, 2022, doi: 228, 2020, doi: 10.1109/MWC.001.1900333.
41 10.1109/MCOMSTD.0001.2200001. [20] N. Promwongsa et al., “A Comprehensive Survey of the Tactile
Internet: State-of-the-Art and Research Directions,” IEEE Commun.
42 [2] I. Rahman, S. M. Razavi, O. Liberg, C. Hoymann, H. Wiemann, C.
Surv. Tutorials, vol. 23, no. 1, pp. 472–523, 2021, doi:
Tidestav, P. Schliwa-Bertling, P. Persson, and D. Gerstenberger, “5G
43 evolution toward 5G advanced: An overview of 3GPP releases 17 and 10.1109/COMST.2020.3025995.
44 18,” Ericsson Technol. Rev, vol. 2021, no. 14, pp. 2–12, 2021. [21] L. U. Khan, I. Yaqoob, M. Imran, Z. Han, and C. S. Hong, “6G
Wireless Systems: A Vision, Architectural Elements, and Future
45 [3] M. Mahdi Azari, S. Solanki, S. Chatzinotas, O. Kodheli, H. Sallouha,
Directions,” IEEE Access, vol. 8, pp. 147029–147044, 2020, doi:
A. Colpaert, J. Fabian Mendoza Montoya, S. Pollin, A. Haqiqatnejad,
46 A. Mostaani, E. Lagunas, and B. Ottersten, “Evolution of Non- 10.1109/ACCESS.2020.3015289.
47 Terrestrial Networks from 5G to 6G: A Survey,” IEEE Commun. [22] Z. Zhang et al., “6G Wireless Networks: Vision, Requirements,
48 Surv. Tutorials, vol. 24, no. 4, pp. 2633–2672, 2022, doi: Architecture, and Key Technologies,” IEEE Veh. Technol. Mag., vol.
14, no. 3, pp. 28–41, 2019, doi: 10.1109/MVT.2019.2921208.
49 10.1109/COMST.2022.3199901.
[23] Y. W. Chen, R. Zhang, C. W. Hsu, and G. K. Chang, “Key Enabling
[4] M. B. Janjua, A. E. Duranay, and H. Arslan, “Role of Wireless
50 Communication in Healthcare System to Cater Disaster Situations Technologies for the Post-5G Era: Fully Adaptive, All-Spectra
51 Under 6G Vision,” Front. Commun. Networks, vol. 1, no. December, Coordinated Radio Access Network with Function Decoupling,”
IEEE Commun. Mag., vol. 58, no. 9, pp. 60–66, 2020, doi:
52 pp. 1–10, 2020, doi: 10.3389/frcmn.2020.610879.
10.1109/MCOM.001.2000186.
[5] M. Vaezi et al., “Cellular, Wide-Area, and Non-Terrestrial IoT: A
53 Survey on 5G Advances and the Road Toward 6G,” IEEE Commun. [24] H. H. H. Mahmoud, A. A. Amer, and T. Ismail, “6G: A
54 Surv. Tutorials, vol. 24, no. 2, pp. 1117–1174, 2022, doi: comprehensive survey on technologies, applications, challenges, and
research problems,” Trans. Emerg. Telecommun. Technol., vol. 32,
55 10.1109/COMST.2022.3151028.
no. 4, pp. 1–14, 2021, doi: 10.1002/ett.4233.
[6] G. Araniti, A. Iera, S. Pizzi, and F. Rinaldi, “Toward 6G Non-
56 Terrestrial Networks,” IEEE Netw., vol. 36, no. 1, pp. 113–120, 2022, [25] S. Basharat, S. A. Hassan, H. Pervaiz, A. Mahmood, Z. Ding, and M.
57 doi: 10.1109/MNET.011.2100191. Gidlund, “Reconfigurable Intelligent Surfaces: Potentials,
58 17
59 For Review Only
60
Page 19 of 42 IEEE Access

1
2
Applications, and Challenges for 6G Wireless Networks,” IEEE Earth Orbit Satellite Access,” IEEE Commun. Mag., vol. 59, no. 12,
3 Wirel. Commun., vol. 28, no. 6, pp. 184–191, 2021, doi: pp. 36–42, 2021, doi: 10.1109/MCOM.001.2100298.
4 10.1109/MWC.011.2100016. [44] D. Renga and M. Meo, “Can High Altitude Platform Stations Make
5 [26] B. Ji, Y. Han, S. Liu, F. Tao, G. Zhang, Z. Fu, and C. Li, “Several 6G Sustainable?,” IEEE Commun. Mag., vol. 60, no. 9, pp. 75–80,
Key Technologies for 6G: Challenges and Opportunities,” IEEE 2022, doi: 10.1109/MCOM.002.2101048.
6 Commun. Stand. Mag., vol. 5, no. 2, pp. 44–51, 2021, doi: [45] C. Palanci, F. Akleman, and G. K. Kurt, “High Altitude Platform
7 10.1109/MCOMSTD.001.2000038. Station (HAPS) to Satellite Channel Models for 6G Networks,” IEEE
8 [27] S. Mumtaz, V. G. Menon, and M. I. Ashraf, “Guest Editorial: Ultra- Aerosp. Conf. Proc., vol. 2022-March, 2022, doi:
Low-Latency and Reliable Communications for 6G Networks,” IEEE 10.1109/AERO53065.2022.9843327.
9 Commun. Stand. Mag., vol. 5, no. 2, pp. 10–11, 2021, doi: [46] B. Li, Z. Fei, and Y. Zhang, “UAV communications for 5G and
10 10.1109/MCOMSTD.2021.9464926. beyond: Recent advances and future trends,” IEEE Internet Things J.,
11 [28] M. Adhikari and A. Hazra, “6G-Enabled Ultra-Reliable Low-Latency vol. 6, no. 2, pp. 2241–2263, 2019, doi: 10.1109/JIOT.2018.2887086.
12 Communication in Edge Networks,” IEEE Commun. Stand. Mag., [47] A. Sharma et al., “Communication and networking technologies for
vol. 6, no. March, pp. 67–74, 2022. UAVs: A survey,” J. Netw. Comput. Appl., vol. 168, p. 102739, 2020,
13 [29] M. Alsabah, M. A. Naser, B. M. Mahmmod, S. H. Abdulhussain, and doi: 10.1016/j.jnca.2020.102739.
14 M. R. Eissa, “6G Wireless Communications Networks : A [48] R. Shrestha, R. Bajracharya, and S. Kim, “6G Enabled Unmanned
15 Comprehensive Survey,” IEEE Access, vol. 9, pp. 148191–148243, Aerial Vehicle Traffic Management: A Perspective,” IEEE Access,
2021, doi: 10.1109/ACCESS.2021.3124812. vol. 9, pp. 91119–91136, 2021, doi:
16 [30] W. B. H. M. A. H. H. D. Jiang, “The Road Towards 6G : A 10.1109/ACCESS.2021.3092039.
17 Comprehensive Survey,” IEEE Open J. Commun. Soc., vol. 2, pp. [49] X. JIANG, M. SHENG, N. ZHAO, C. XING, W. LU, and X. WANG,
18 334–366, 2021, doi: 10.1109/OJCOMS.2021.3057679. “Green UAV communications for 6G: A survey,” Chinese J.
[31] Y. Yuan, Y. Zhao, B. Zong, and S. Parolari, “Potential key Aeronaut., vol. 35, no. 9, pp. 19–34, 2022, doi:
19 technologies for 6G mobile communications,” Sci. China Inf. Sci., 10.1016/j.cja.2021.04.025.
20 vol. 63, no. 8, pp. 1–19, 2020, doi: 10.1007/s11432-019-2789-y. [50] X. Fang, W. Feng, T. Wei, Y. Chen, N. Ge, and C. X. Wang, “5G
21 [32] W. Saad, M. Bennis, and M. Chen, “A Vision of 6G Wireless Embraces Satellites for 6G Ubiquitous IoT: Basic Models for
Systems: Applications, Trends, Technologies, and Open Research Integrated Satellite Terrestrial Networks,” IEEE Internet Things J.,
22 Problems,” IEEE Netw., vol. 34, no. 3, pp. 134–142, 2020, doi: vol. 8, no. 18, pp. 14399–14417, 2021, doi:
23 10.1109/MNET.001.1900287. 10.1109/JIOT.2021.3068596.
24 [33] M. M. Kamruzzaman, “Key Technologies, Applications and Trends [51] G. K. Kurt et al., “A Vision and Framework for the High Altitude
25 of Internet of Things for Energy-Efficient 6G Wireless Platform Station (HAPS) Networks of the Future,” IEEE Commun.
Communication in Smart Cities,” Energies, vol. 15, no. 15, p. 5608, Surv. Tutorials, vol. 23, no. 2, pp. 729–779, 2021, doi:
26 2022. 10.1109/COMST.2021.3066905.
27 [34] S. S. Hassan, D. H. Kim, Y. K. Tun, N. H. Tran, W. Saad, and C. S. [52] S. A. Al-Ahmed, M. Z. Shakir, and S. A. R. Zaidi, “Optimal 3D UAV
28 Hong, “Seamless and Energy Efficient Maritime Coverage in base station placement by considering autonomous coverage hole
Coordinated 6G Space-Air-Sea Non-Terrestrial Networks,” IEEE detection, wireless backhaul and user demand,” J. Commun.
29 Internet Things J., vol. 10, no. 6, pp. 4749–4769, 2022, doi: Networks, vol. 22, no. 6, pp. 467–475, 2020, doi:
30 10.1109/JIOT.2022.3220631. 10.23919/JCN.2020.000034.
31 [35] P. Porambage, G. Gur, D. P. Moya Osorio, M. Livanage, and M. [53] I. Natgunanathan, A. Mehmood, Y. Xiang, G. Beliakov, and J.
Ylianttila, “6G security challenges and potential solutions,” 2021 Jt. Yearwood, “Protection of Privacy in Biometric Data,” IEEE Access,
32 Eur. Conf. Networks Commun. 6G Summit, EuCNC/6G Summit 2021, vol. 4, pp. 880–892, 2016, doi: 10.1109/ACCESS.2016.2535120.
33 pp. 622–627, 2021, doi: [54] N. H. Mahmood et al., “Machine type communications: key drivers
34 10.1109/EuCNC/6GSummit51104.2021.9482609. and enablers towards the 6G era,” Eurasip J. Wirel. Commun. Netw.,
35 [36] V. L. Nguyen, P. C. Lin, B. C. Cheng, R. H. Hwang, and Y. D. Lin, vol. 2021, no. 1, 2021, doi: 10.1186/s13638-021-02010-5.
“Security and Privacy for 6G: A Survey on Prospective Technologies [55] M. A. Lopez, G. N. N. Barbosa, and D. M. F. Mattos, “New Barriers
36 and Challenges,” IEEE Commun. Surv. Tutorials, vol. 23, no. 4, pp. on 6G Networking: An Exploratory Study on the Security, Privacy,
37 2384–2428, 2021, doi: 10.1109/COMST.2021.3108618. and Opportunities for Aerial Networks,” 2022 1st Int. Conf. 6G
38 [37] L. U. Khan, I. Yaqoob, N. H. Tran, Z. Han, and C. S. Hong, “Network Networking, 6GNet 2022, no. Pdpa, pp. 1–6, 2022, doi:
Slicing: Recent Advances, Taxonomy, Requirements, and Open 10.1109/6GNet54646.2022.9830402.
39 Research Challenges,” IEEE Access, vol. 8, pp. 36009–36028, 2020, [56] G. A. Akpakwu, B. J. Silva, G. P. Hancke, and A. M. Abu-Mahfouz,
40 doi: 10.1109/ACCESS.2020.2975072. “A Survey on 5G Networks for the Internet of Things:
41 [38] L. LI-HSIANG SHEN; KAI-TEN FENG; Hanzo, “Five Facets of Communication Technologies and Challenges,” IEEE Access, vol. 6,
6G : Research Challenges and Opportunities,” ACM Comput. Surv., pp. 3619–3647, 2017, doi: 10.1109/ACCESS.2017.2779844.
42 vol. 55, no. 11, pp. 1–39, 2023, doi: 10.1145/3571072. [57] N. Panwar and S. Sharma, “Security and Privacy Aspects in 5G
43 [39] S. A. Abdel Hakeem, H. H. Hussein, and H. W. Kim, “Vision and Networks,” 2020 IEEE 19th Int. Symp. Netw. Comput. Appl. NCA
44 research directions of 6G technologies and applications,” J. King 2020, 2020, doi: 10.1109/NCA51143.2020.9306740.
Saud Univ. - Comput. Inf. Sci., vol. 34, no. 6, pp. 2419–2442, 2022, [58] R. Banu and T. Vladimirova, “Fault-tolerant encryption for space
45 doi: 10.1016/j.jksuci.2022.03.019. applications,” IEEE Trans. Aerosp. Electron. Syst., vol. 45, no. 1, pp.
46 [40] B. Tezergil and E. Onur, “Wireless Backhaul in 5G and Beyond: 266–279, 2009, doi: 10.1109/TAES.2009.4805278.
47 Issues, Challenges and Opportunities,” IEEE Commun. Surv. [59] M. Ylianttila, R. Kantola, A. Gurtov, L. Mucchi, I. Oppermann, Z.
48 Tutorials, vol. 24, no. 4, pp. 2579–2632, 2022, doi: Yan, T. H. Nguyen, F. Liu, T. Hewa, and M. Liyanage, ‘‘6G white
10.1109/COMST.2022.3203578. paper: Research challenges for trust, security and privacy,’’ 2020,
49 [41] W. Xu, Z. Yang, D. W. K. Ng, M. Levorato, Y. C. Eldar, and M. arXiv:2004.11665.
50 Debbah, “Edge Learning for B5G Networks With Distributed Signal [60] I. Ahmad, J. Suomalainen, P. Porambage, A. Gurtov, J. Huusko, and
51 Processing: Semantic Communication, Edge Computing, and M. Hoyhtya, “Security of Satellite-Terrestrial Communications:
Wireless Sensing,” IEEE J. Sel. Top. Signal Process., vol. 17, no. 1, Challenges and Potential Solutions,” IEEE Access, vol. 10, no.
52 pp. 9–39, 2023, doi: 10.1109/JSTSP.2023.3239189. August, pp. 96038–96052, 2022, doi:
53 [42] E. Juan, M. Lauridsen, J. Wigard, and P. E. Mogensen, “5G New 10.1109/ACCESS.2022.3205426.
54 Radio Mobility Performance in LEO-based Non-Terrestrial [61] M. Ericson et al., “6G Architectural Trends and Enablers,” Proc. -
Networks,” 2020 IEEE Globecom Work. GC Wkshps 2020 - Proc., 2021 IEEE 4th 5G World Forum, 5GWF 2021, pp. 406–411, 2021,
55 pp. 1–6, 2020, doi: 10.1109/GCWkshps50303.2020.9367419. doi: 10.1109/5GWF52925.2021.00078.
56 [43] X. Lin, S. Cioni, G. Charbit, N. Chuberre, S. Hellsten, and J. F. [62] T. Patra, “Link Budget Analysis for 5G Communication in the
57 Boutillon, “On the Path to 6G: Embracing the Next Wave of Low Tropical Regions,” Wirel. Commun. Mob. Comput., vol. 2020, pp. 1–
58 18
59 For Review Only
60
 IEEE Access Page 20 of 42

1
2
9, 2020. UNIVERSITI TUNKU ABDUL RAHMAN, no. June. 2022.
3 [63] M. Ikram, K. Sultan, M. F. Lateef, and A. S. M. Alqadami, “A Road [81] A. Iqbal, M. L. Tham, and Y. C. Chang, “Convolutional neural
4 towards 6G Communication — A Review of 5G Antennas , Arrays , network-based deep Q-network (CNN-DQN) resource management
5 and Wearable Devices,” electronics, vol. 11, no. 169, pp. 1–29, 2022. in cloud radio access network,” China Commun., vol. 19, no. 10, pp.
[64] Z. Wang, S. Liu, and Y. Dong, “Compact Wideband Pattern 129–142, 2022, doi: 10.23919/JCC.2022.00.025.
6 Reconfigurable Antennas Inspired by End-Fire Structure for 5G [82] A. Iqbal, M. L. Tham, and Y. C. Chang, “Resource allocation for joint
7 Vehicular Communication,” IEEE Trans. Veh. Technol., vol. 71, no. energy and spectral efficiency in cloud radio access network based on
8 5, pp. 4655–4664, 2022. deep reinforcement learning,” Trans. Emerg. Telecommun. Technol.,
[65] Y. U. Jianjun and Y. Wei, “Digital Signal Processing for High-Speed vol. 33, no. 4, pp. 1–16, 2022, doi: 10.1002/ett.4417.
9 THz Communications,” Chinese J. Electron., vol. 31, no. 3, pp. 534– [83] R. A. Nazib and S. Moh, “Reinforcement Learning-Based Routing
10 546, 2022, doi: 10.1049/cje.2021.00.258. Protocols for Vehicular Ad Hoc Networks: A Comparative Survey,”
11 [66] B. Mao, F. Tang, Y. Kawamoto, and N. Kato, “AI Models for Green IEEE Access, vol. 9, pp. 27552–27587, 2021, doi:
12 Communications Towards 6G,” IEEE Commun. Surv. Tutorials, vol. 10.1109/ACCESS.2021.3058388.
24, no. 1, pp. 210–247, 2022, doi: 10.1109/COMST.2021.3130901. [84] R. F. Atallah, C. M. Assi, and J. Y. Yu, “A Reinforcement Learning
13 [67] V. Verdhan, “Introduction to Supervised Learning,” Univ. Technique for Optimizing Downlink Scheduling in an Energy-
14 Massachusetts, vol. 3, no. 1, pp. 1–46, 2014, doi: 10.1007/978-1- Limited Vehicular Network,” IEEE Trans. Veh. Technol., vol. 66, no.
15 4842-6156-9_1. 6, pp. 4592–4601, 2017, doi: 10.1109/TVT.2016.2622180.
[68] H. U. Dike, Y. Zhou, K. K. Deveerasetty, and Q. Wu, “Unsupervised [85] Y. Liu et al., “Reconfigurable Intelligent Surfaces: Principles and
16 Learning Based On Artificial Neural Network: A Review,” 2018 Opportunities,” IEEE Commun. Surv. Tutorials, vol. 23, no. 3, pp.
17 IEEE Int. Conf. Cyborg Bionic Syst. CBS 2018, pp. 322–327, 2019, 1546–1577, 2021, doi: 10.1109/COMST.2021.3077737.
18 doi: 10.1109/CBS.2018.8612259. [86] T. Kürsat, K. Kurt, R. Ali, and Y. Halim, “Reconfigurable Intelligent
[69] H. Zhang and T. Yu, “Taxonomy of Reinforcement Learning Surfaces in Action for Non-Terrestrial Networks,” IEEE Veh.
19 Algorithms,” Deep Reinforcement Learning, Springer, 2020, pp. Technol. Mag., vol. 17, no. 3, pp. 45–53, 2022.
20 125–33. [87] K. W. Park, H. M. Kim, and O. S. Shin, “A Survey on Intelligent-
21 [70] D. Dampahalage, K. B. S. Manosha, N. Rajatheva, and M. Latva- Reflecting-Surface-Assisted UAV Communications,” Energies, vol.
Aho, “Supervised Learning Based Sparse Channel Estimation for Ris 15, no. 14, pp. 1–20, 2022, doi: 10.3390/en15145143.
22 Aided Communications,” ICASSP, IEEE Int. Conf. Acoust. Speech [88] C. C. Gonzalez, S. Pizzi, M. Murroni, and G. Araniti, “Multicasting
23 Signal Process. - Proc., vol. 2022-May, pp. 8827–8831, 2022, doi: Over 6G Non-Terrestrial Networks: A Softwarization-Based
24 10.1109/ICASSP43922.2022.9746793. Approach,” IEEE Veh. Technol. Mag., vol. 18, no. 1, pp. 91–99, 2023,
25 [71] A. E. Perez, A. Estrada Moscoso, N. G. Gonzalez, and J. J. G. Torres, doi: 10.1109/MVT.2022.3232919.
“Training Approaches in Supervised Learning for ICI Mitigation in [89] S. K. Das et al., “Comprehensive review on ML-based RIS-enhanced
26 Gridless Nyquist-WDM,” 2020 IEEE Photonics Conf. IPC 2020 - IoT systems: basics, research progress and future challenges,”
27 Proc., no. 53, pp. 2–3, 2020, doi: 10.1109/IPC47351.2020.9252366. Comput. Networks, vol. 224, p. 109581, 2023, doi:
28 [72] J. Jang, J. H. Park, and H. Jong Yang, “Supervised-Learning-Based 10.1016/j.comnet.2023.109581.
Resource Allocation in Wireless Networks,” Int. Conf. ICT Converg., [90] X. Wang, X. Li, and V. C. M. Leung, “Artificial intelligence-based
29 vol. 2020-Octob, pp. 1022–1024, 2020, doi: techniques for emerging heterogeneous network: State of the arts,
30 10.1109/ICTC49870.2020.9289481. opportunities, and challenges,” IEEE Access, vol. 3, pp. 1379–1391,
31 [73] Y. Zhang, J. Wang, Y. Chen, H. Yu, and T. Qin, “Adaptive Memory 2015, doi: 10.1109/ACCESS.2015.2467174.
Networks with Self-supervised Learning for Unsupervised Anomaly [91] B. Siniarski et al., “Need of 6G for the Metaverse Realization,”
32 Detection,” IEEE Trans. Knowl. Data Eng., vol. 4347, no. 12, pp. 1– arXiv:2301.03386, vol. 7, pp. 1–31, 2022, [Online]. Available:
33 13, 2022, doi: 10.1109/TKDE.2021.3139916. http://arxiv.org/abs/2301.03386
34 [74] W. Zhang, L. Jiao, F. Liu, S. Yang, W. Song, and J. Liu, “Sparse [92] P. K. Thiruvasagam et al., “Open RAN: Evolution of Architecture,
35 Feature Clustering Network for Unsupervised SAR Image Change Deployment Aspects, and Future Directions,” arXiv:2301.06713, pp.
Detection,” IEEE Trans. Geosci. Remote Sens., vol. 60, no. 5226713, 1–28, 2023.
36 pp. 1–13, 2022, doi: 10.1109/TGRS.2022.3167745. [93] Y. O. Imam-Fulani et al., “5G Frequency Standardization,
37 [75] M. U. Khan, M. Azizi, A. Garcia-Armada, and J. J. Escudero-Garzas, Technologies, Channel Models, and Network Deployment:
38 “Unsupervised Clustering for 5G Network Planning Assisted by Real Advances, Challenges, and Future Directions,” Sustainability, vol.
Data,” IEEE Access, vol. 10, no. 2, pp. 39269–39281, 2022, doi: 15, no. 6, p. 5173, 2023, doi: 10.3390/su15065173.
39 10.1109/ACCESS.2022.3165799. [94] S. Coveney and K. Roberts, “Lightweight UAV digital elevation
40 [76] M. Nair, T. Cappello, S. Dang, V. Kalokidou, and M. A. Beach, “RF models and orthoimagery for environmental applications: data
41 Fingerprinting of LoRa Transmitters Using Machine Learning with accuracy evaluation and potential for river flood risk modelling,” Int.
Self-Organizing Maps for Cyber Intrusion Detection,” IEEE MTT-S J. Remote Sens., vol. 38, no. 8–10, pp. 3159–3180, 2017, doi:
42 Int. Microw. Symp. Dig., vol. 2022-June, no. 1, pp. 491–494, 2022, 10.1080/01431161.2017.1292074.
43 doi: 10.1109/IMS37962.2022.9865441. [95] A. Traspadini, M. Giordani, and M. Zorzi, “UAV/HAP-Assisted
44 [77] M. L. Tham, A. Iqbal, and Y. C. Chang, “Deep reinforcement Vehicular Edge Computing in 6G: Where and What to Offload?,”
learning for resource allocation in 5G communications,” 2019 Asia- 2022 Jt. Eur. Conf. Networks Commun. 6G Summit, EuCNC/6G
45 Pacific Signal Inf. Process. Assoc. Annu. Summit Conf. APSIPA ASC Summit 2022, pp. 178–183, 2022, doi:
46 2019, no. November, pp. 1852–1855, 2019, doi: 10.1109/EuCNC/6GSummit54941.2022.9815734.
47 10.1109/APSIPAASC47483.2019.9023112. [96] Z. Ma, Y. Wang, J. Li, and Y. Liu, “A Blockchain-Based Privacy-
48 [78] S. Tripathi, S. Member, O. J. Pandey, S. Member, and R. M. Hegde, Preserving Incentive Mechanism for Internet of Vehicles in Satellite-
“An Optimal Reflective Elements Grouping Model for RIS-Assisted Terrestrial Crowdsensing,” 2021 7th Int. Conf. Comput. Commun.
49 IoT Networks Using Q-Learning,” IEEE Trans. Circuits Syst. II ICCC 2021, pp. 2062–2067, 2021, doi:
50 Express Briefs, pp. 1–5, 2023, doi: 10.1109/TCSII.2023.3251373. 10.1109/ICCC54389.2021.9674460.
51 [79] A. Iqbal, M. L. Tham, and Y. C. Chang, “Double Deep Q-Network- [97] H. Zhou, Z. Wang, N. Cheng, D. Zeng, and P. Fan, “Stackelberg-
Based Energy-Efficient Resource Allocation in Cloud Radio Access Game-Based Computation Offloading Method in Cloud-Edge
52 Network,” IEEE Access, vol. 9, pp. 20440–20449, 2021, doi: Computing Networks,” IEEE Internet Things J., vol. 9, no. 17, pp.
53 10.1109/ACCESS.2021.3054909. 16510–16520, 2022, doi: 10.1109/JIOT.2022.3153089.
54 [80] A. Iqbal, "DEVELOPMENT OF DEEP REINFORCEMENT [98] H. Zhou, Z. Wang, G. Min, and H. Zhang, “UAV-Aided Computation
LEARNING BASED RESOURCE ALLOCATION TECHNIQUES Offloading in Mobile-Edge Computing Networks: A Stackelberg
55 IN CLOUD RADIO ACCESS NETWORK" AMJAD IQBAL Game Approach,” IEEE Internet Things J., vol. 10, no. 8, pp. 6622–
56 DOCTOR OF PHILOSOPHY ( ENGINEERING ) LEE KONG 6633, 2023, doi: 10.1109/JIOT.2022.3197155.
57 CHIAN FACULTY OF ENGINEERING AND SCIENCE [99] H. Zhou, Z. Wang, H. Zheng, S. He, and M. Dong, “Cost
58 19
59 For Review Only
60
Page 21 of 42 IEEE Access

1
2
Minimization-Oriented Computation Offloading and Service AMJAD IQBAL received the B.Sc. degree in
3 Caching in Mobile Cloud-Edge Computing: An A3C-Based Electrical Engineering from CECOS
4 Approach,” IEEE Trans. Netw. Sci. Eng., vol. 10, no. 3, pp. 1326– University, Peshawar, Pakistan, in 2014, and
5 1338, 2023, doi: 10.1109/TNSE.2023.3255544. the M.Sc. degree in Communication and
[100] A. Heidari, N. J. Navimipour, and M. Unal, “Applications of ML/DL Electronic engineering from the University of
6 in the management of smart cities and societies based on new trends Engineering and Technology (UET), Peshawar,
7 in information technologies: A systematic literature review,” Sustain. in 2017, and the Ph.D. degree in Wireless
8 Cities Soc., vol. 85, no. February, p. 104089, 2022, doi: Communication from University Tunku Abdul
10.1016/j.scs.2022.104089. Rahman (UTAR), Malaysia, in 2022. He is
9 [101] P. H. Masur, J. H. Reed, N. K. Tripathi, and V. Tech, “Regular currently a Postdoc research fellow in the
10 Article : Artificial Intelligence in Open-Radio Access Network,” Department of Systems and Computer
11 IEEE Aerosp. Electron. Syst. Mag., vol. 37, no. 9, pp. 6–15, 2022. Engineering at Carleton University, Canada.
12 [102] P. L. S. Jayalaxmi, R. Saha, G. Kumar, M. Conti, and T. H. Kim, His research interests include resource allocation optimization, machine
“Machine and Deep Learning Solutions for Intrusion Detection and learning, deep reinforcement learning, 5G/B5G cellular networks, and
13 Prevention in IoTs: A Survey,” IEEE Access, vol. 10, no. October, Reflecting Intelligent Surface (RIS).
14 pp. 121173–121192, 2022, doi: 10.1109/ACCESS.2022.3220622.
15 [103] X. Shen et al., “AI-Assisted Network-Slicing Based Next-Generation
Wireless Networks,” IEEE Open J. Veh. Technol., vol. 1, pp. 45–66,
16 2020.
17 [104] M. K. Tefera, “Deep Reinforcement Learning-Assisted Optimization
18 for Resource Allocation in Downlink OFDMA Cooperative
Systems,” Entropy, vol. 25, no. 3, p. 413, 2023.
19 [105] T. Hu, Q. Liao, Q. Liu, and G. Carle, “Network Slicing via Transfer
20 Learning aided Distributed Deep Reinforcement Learning,” in IEEE
21 Global Communications Conference: Next-Generation Networking MAU-LUEN THAM received the Bachelor of
and Internet, 2023, pp. 2909–2914. doi:
22 10.1109/globecom48099.2022.10000763.
Engineering and Doctor of Philosophy
degrees in telecommunication engineering
23 [106] A. Mukherjee, P. Goswami, M. A. Khan, L. Manman, L. Yang, and from University of Malaya. He is currently an
24 P. Pillai, “Energy-efficient resource allocation strategy in massive iot Assistant Professor with Universiti Tunku
25 for industrial 6G applications,” IEEE Internet Things J., vol. 8, no. 7, Abdul Rahman. He has been a principal
pp. 5194–5201, 2021, doi: 10.1109/JIOT.2020.3035608.
26 [107] C. Zhang, M. Dong, and K. Ota, “Fine-Grained Management in 5G:
investigator (PI) and a co-investigator of more
than 20 research and development projects.
27 DQL Based Intelligent Resource Allocation for Network Function This includes five international grants, two of
28 Virtualization in C-RAN,” IEEE Trans. Cogn. Commun. Netw., vol. which were simultaneously led by him as the
6, no. 2, pp. 428–435, 2020, doi: 10.1109/TCCN.2020.2982886.
29 [108] Y. Fu and X. Wang, “Traffic Prediction-Enabled Energy-Efficient
PI/Co-PI under the support of ASEAN IVO
and the British Council. He has published two IEEE Transactions papers as
30 Dynamic Computing Resource Allocation in CRAN Based on Deep a principal author. His research interests include the IoT, machine
31 Learning,” IEEE Open J. Commun. Soc., vol. 3, pp. 159–175, 2022, learning/deep learning/deep reinforcement learning, and beyond-5G
doi: 10.1109/OJCOMS.2022.3146886.
32 [109] C. Sun, X. Wu, X. Li, Q. Fan, J. Wen, and V. C. M. Leung,
communications.
33 “Cooperative Computation Offloading for Multi-Access Edge
34 Computing in 6G Mobile Networks via Soft Actor Critic,” IEEE
35 Trans. Netw. Sci. Eng., pp. 1–14, 2021, doi:
10.1109/TNSE.2021.3076795.
36 [110] M. Karavolos, L. Tsipi, P. S. Bithas, D. Vouyioukas and P. T.
37 Mathiopoulos, "Satellite Aerial Terrestrial Hybrid NOMA Scheme in
38 6G Networks: An Unsupervised Learning Approach," 2022 1st
International Conference on 6G Networking (6GNet), Paris, France,
39 2022, pp. 1-5, doi: 10.1109/6GNet54646.2022.9830368.
40 [111] M. K. Dahouda, S. Jin, and I. Joe, "Machine Learning-Based
41 Solutions for Handover Decisions in Non-Terrestrial Networks, "
Electronics, vol. 12, no. 8, 2023, doi:10.3390/electronics12081759
42 YI JIE WONG received the B.Eng. degree
(Hons.) in biomedical engineering from the
43 Universiti Tunku Abdul Rahman, Sungai
44 Long, Malaysia in 2022, where he is currently
45 pursuing the Ph.D. degree in digital
technology with specialization of
46 reinforcement learning-based federated
47 learning. His research interests include the
48 Internet of Things (IoT), machine learning,
49 federated learning, and deep reinforcement learning.
50
51
52
53
54
55
56
57
58 20
59 For Review Only
60
 IEEE Access Page 22 of 42

1
2
ALA’A AL-HABASHNA received his Master TASOS DAGIUKLAS received the
3 of Engineering degree from Memorial engineering degree from the University of
4 University of Newfoundland in 2010 and his Patras, Greece, in 1989, the M.Sc. degree from
5 Ph.D. in Electrical and Computer Engineering the University of Manchester, U.K., in 1991, and
from Carleton University in 2018. Currently, the Ph.D. degree from the University of Essex,
6 Dr. Al-Habashna is an Adjunct Research U.K., in 1995, all in electrical engineering. He is
7 Professor at Carleton University, Ottawa, a Leading Researcher and Expert in the field of
8 Canada. Dr. Al-Habashna is also a Senior smart internet technologies. He is the Leader of
Researcher at Statistics Canada, Ottawa, the Smart Internet Technologies (SuITE)
9 Canada. He has won multiple awards, including excellence and best-paper Research Group, London South Bank University, where he also acts as the
10 awards. He worked as a reviewer for many conferences and journals and as Head of the Cognitive Systems Research Centre. He has been a principal
11 a Technical Program Committee Member of multiple conferences. He has investigator, a co-investigator, the project and technical manager, a
12 also been the co-chair of the Communications and Networking Simulation coordinator, and a focal person of more than 25 internationally research and
(CNS) track of the Annual Modeling and Simulation Conference. His development and capacity training projects in the areas of fixed-mobile
13 current research interests include 5G wireless networks, machine learning, convergence, 4G/5G networking technologies, VoIP, and multimedia
14 computer vision, IoT applications, localization, multimedia communication networking. His research interests lie in the field of systems beyond 5G/6G
15 over wireless networks, signal detection and classification, cognitive radio networking technologies, programmable networks, UAVs, V2X
systems, and discrete-event modeling and simulation. communications, and cyber security for smart internet systems.
16
17
18
19
20
21
22
23
24 GABRIEL WAINER received the Ph.D. degree
25 (Highest Hons.) from UBA/Université d'Aix-
Marseille III, Marseille, France, in 1998. He held
26 visiting positions with the University of Arizona;
27 LSIS (CNRS), Université Paul Cézanne,
28 University of Nice, INRIA Sophia-Antipolis,
Université de Bordeaux (France); UCM, UPC
29 (Spain), University of Buenos Aires, National
30 University of Rosario (Argentina) and others. He
31 is currently a full professor with Carleton
University, Ottawa, ON, Canada, where he is also the head with the
32 Advanced Real-Time Simulation Laboratory, Centre for advanced
33 Simulation and Visualization (V-Sim). He is Editor in Chief of
34 SIMULATION (Sage) and a member of the editorial board of the IEEE
35 Computing in Science & Engineering, Wireless Networks (Elsevier), and
The Journal of Defense Modeling and Simulation (SCS). He is a fellow of
36 the Society for Modeling and Simulation International (SCS).
37
38
39
40
41
42
43
44 YONGXU ZHU received the Ph.D. degree in
45 electrical engineering from University College
46 London in 2017. From 2017 to 2019, she was a
47 Research Associate with Loughborough
University. She is currently a Senior Lecturer with
48 the Division of Computer Science and Informatics,
49 London South Bank University. Her research
50 interests include B5G/6G, heterogeneous
networks, UAV communications, and physical-layer security. She also
51 serves as an Editor for IEEE Wireless Communications Letters and IEEE
52 Transactions on Wireless Communications.
53
54
55
56
57
58 21
59 For Review Only
60
Page 23 of 42 IEEE Access

1
2 Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
3 Digital Object Identifier 10.1109/ACCESS.2022.Doi Number
4
5
6
7
8
Empowering Non-Terrestrial Networks with
9
10
Artificial Intelligence: A Survey
11
12 AMJAD IQBAL1,2, MAU-LUEN THAM1, YI JIE WONG1, ALA’A AL-HABASHNA2, GABRIEL
13 WAINER2, YONG XU ZHU3, AND TASOS DAGIUKLAS4
1
Department of Electrical and Electronic Engineering, Lee Kong Chian Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Sungai
14 Long Campus, Selangor 43000, Malaysia.
15 2
Department of Systems and Computer Engineering, Carleton University, 1125 Colonel By Dr., Ottawa, ON, K1S 5B6, Canada.
16
3
School of Engineering, Warwick University, UK.
4
Cognitive Systems Research Centre, School of Engineering/Computer Science, London South Bank University, 103 Borough Road, London SE1
17 0AA, UK.
18 Corresponding authors: Mau-Luen Tham (e-mail: thamml@utar.edu.my).
19 This work was supported in part by the Universiti Tunku Abdul Rahman (UTAR), Malaysia, under UTAR Research Fund (UTARRF)
(IPSR/RMC/UTARRF/2021C1/T05) and in part by the British Council under UK-ASEAN Institutional Links Early Career Researchers Scheme with project
20 number 913030644 and partially supported by NSERC Canada).”
21
22
23
ABSTRACT 6G networks can support global, ubiquitous and seamless connectivity through the
24
convergence of terrestrial and non-terrestrial networks (NTNs). Unlike terrestrial scenarios, NTNs pose
25
unique challenges including propagation characteristics, latency and mobility, owing to the operations in
26
spaceborne and airborne platforms. To overcome all these technical hurdles, this survey paper presents the
27
use of artificial intelligence (AI) techniques in learning and adapting to the complex NTN environments. We
28
begin by providing an overview of NTNs in the context of 6G, highlighting the potential security and privacy
29
issues. Next, we review the existing AI methods adopted for 6G NTN optimization, starting from machine
30
learning (ML), through deep learning (DL) to deep reinforcement learning (DRL). All these AI techniques
31
have paved the way towards more intelligent network planning, resource allocation (RA), and interference
32
management. Furthermore, we discuss the challenges and opportunities in AI-powered NTN for 6G networks.
33
Finally, we conclude by providing insights and recommendations on the key enabling technologies for future
34
AI-powered 6G NTNs.
35
36
37 INDEX TERMS Non-Terrestrial Networks (NTNs), Artificial Intelligence (AI), 5G/6G, Unmanned
38 Aircraft System (UAS), Resource Allocation (RA), Reinforcement Learning (RL), Deep Learning (DL)
39
40
41 I. INTRODUCTION (NTNs), which include Low/Medium/Geostationary (LEO,
42 The evolution of wireless communication technology has MEO, GEO) satellites, high altitude platform stations
43 been rapid in recent years, with mobile network operators (HAPS), unmanned aircraft systems (UASs), or a
44 deploying the fifth-generation (5G) technology worldwide. combination of these technologies [3]. NTNs can provide
45 The Third Generation Partnership Project (3GPP) has uninterrupted communication and high data transfer rates
46 begun the standardization of 5G-Advanced, which is in remote, disaster-stricken, and rural areas where
47 expected to offer higher data rates, lower latency, increased terrestrial networks are not available. Recent studies have
48 capacity, and more efficient spectrum utilization than any shown that NTNs can offer high availability and low
49 of its predecessors [1]-[2]. However, due to limited latency, making them an ideal candidate for 6G
50 coverage areas and geographic constraints, it is challenging communication systems [4]-[5].
51 to guarantee ubiquitous coverage everywhere with existing Despite these advantages, NTNs present significant
52 network infrastructures. In times of natural disasters, challenges related to their deployment and management.
53 connectivity outages are widespread, which can be One of the primary challenges is how to deploy and manage
54 detrimental to critical actions necessary for saving lives and efficiently, including the physical and ground-based
55 properties. To address these challenges, research on 5G- infrastructure, such as antennas, base stations (BSs), and
56 Advanced and sixth-generation (6G) communication backhaul networks. Furthermore, the dynamic nature of
57 networks are shifting towards non-terrestrial networks NTNs, with platforms moving at high speeds and in
58
59 VOLUME XX, 2017 For Review Only 1

60
 IEEE Access Page 24 of 42

1
2
3 different directions, introduces additional challenges overview of the vision, requirements, and challenges of 6G
4 related to signal propagation, interference, and handover wireless networks are explained in [18] and [19]. The current
5 management [6]-[7]. status and future directions of 6G wireless networks are
6 Researchers are turning to artificial intelligence (AI) as a reviewed in [20] and [21]. Furthermore, [22] and [23] discuss
7 promising solution to address these challenges. AI can help the key enabling technologies for 6G networks, such as
8 to optimize the performance of NTNs by analyzing data terahertz (THz) communication, visible light communication
9 from various sources, predicting network behavior, and (VLC), and wireless power transfer. The applications,
10 adapting to changing conditions [8]. For example, AI can technologies, and challenges of 6G wireless networks are
11 enable intelligent power management and beamforming to explored in [24] and [25]. The challenges and opportunities
12 maximize resource utilization while minimizing in 6G networks, including ultra-reliable, low-latency, and
13 interference and maintaining the quality of service (QoS) massive machine-type communication, are discussed in [26],
14 [9]. Also, AI can provide advanced analytics and automated [27], and [28]. A comprehensive 6G wireless communication
15 decision-making to enhance the efficiency and reliability of survey is presented in [29] and [30]. The concept and
16 NTNs. standardization of 6G networks, including new spectrum
17 The integration of AI in cellular networks is still in its early bands and wireless technologies, are explained in [31] and
18 stages compared to other fields, primarily due to wireless [32]. In [33] and [34], the authors explore the opportunities
19 networks’ complexity and time-varying nature [10]. The and challenges of 6G networks, including the use of satellite
20 multi-dimensional topology of the next-generation wireless communication and energy-efficient design. A
21 networks adds an additional layer of complexity to the comprehensive survey of 6G networks, including new radio
22 existing communication networks, making it challenging to access technologies and security challenges, is presented in
23 solve problems that arise in real networks. Nevertheless, AI [35] and [36]. In [37], the authors present a comprehensive
24 techniques can be exploited to overcome these challenges survey of 6G wireless networks, including new antenna
25 and provide efficient solutions. NTNs, being an integral technologies and network slicing. Similarly, [38] explains
26 part of next-generation wireless networks, holds great the five facets of the new wireless generation, along with its
27 potential for the application of AI. However, implementing research challenges and different opportunities for the new
28 these algorithms in real-world environments while ensuring wireless generation.
29 reliable vertical connectivity between ground and space The aforementioned studies do not incorporate the use of AI
30 networks can present practical challenges. Proper AI in their analysis. In contrast, this survey paper analyzes AI's
31 solutions must complement theoretical advancements in utilization in NTNs to enhance 6G wireless communications
32 communication systems design to achieve optimal networks, as summarized in Table 1. Firstly, we discuss the
33 performance in future networks. potential benefits and challenges of integrating AI in NTNs,
34 including its impact on network performance, reliability, and
35 A. RELATED WORKS AND PAPER CONTRIBUTION security. Secondly, we show different AI techniques that can
36 Several studies and surveys have been conducted to explore be applied to NTNs, such as machine learning (ML), deep
37 the possibility of NTNs for 6G wireless communications. learning (DL), and deep reinforcement learning (DRL).
38 However, it is very rare to explore the concept of AI for Finally, this survey paper discusses the potential applications
39 NTNs. Instead, numerous research has focused on network of AI-powered NTNs in various industries, including
40 architectures, standards, regulations, and use cases. For healthcare, transportation, and smart cities. We analyze the
41 instance, [11] and [12] give an overview of different NTN use cases for AI-powered NTNs in these industries,
42 use, including satellite communication, aerial drones, and including how they can improve efficiency and reduce costs.
43 terrestrial devices. In [13], challenges associated with The main contributions of this survey paper can be
44 satellite communication in 6G are examined, including summarized as follows:
45 power constraints, latency issues, and frequency allocation. 1) We provide an overview of NTNs, including their
46 The works in [14] and [15] explore the potential impact of introduction in the context of 6G networks, their role in
47 6G networks on various industries, including agriculture, enhancing network performance, and their key features
48 transportation, and healthcare. Unlike [14] and [15], the and requirements. Additionally, we analyze the unique
49 works in [16] and [17] focus on the potential security threats security and privacy concerns associated with NTNs,
50 of NTNs by discussing the technical aspects of NTNs, such providing valuable insights into their intrinsic nature.
51 as the different radio access technologies available and the 2) We discuss the AI approaches by explaining the
52 requirements and challenges of integrating 6G with existing fundamental aspects of AI techniques used in the NTNs.
53 technologies. Overall, these works provide valuable insights As a result, we can select appropriate AI approaches for
54 into the potential of NTNs for 6G and the various challenges dealing with various NTN issues.
55 to consider for their successful implementation. The
56
57
58
59 VOLUME XX, 2017 For Review Only 1

60
Page 25 of 42 IEEE Access

1
2
3 Table 1. Summary of Related Works on 6G
4 Ref Year Contribution and Main Focus
5 [22] 2019
A vision for 6G networks in 2030 for superior performance and enabling emerging services and applications is discussed. The focus is to
propose a large-dimensional, autonomous network architecture integrating various networks and advanced technologies.
6
The security and privacy issues associated with 6G networks are being explored as next-generation solutions due to the limitations of 5G
7 [17] 2020
networks. The main focus is to discuss four key aspects of 6G networks and their associated security and privacy issues.
8 [19] 2020
Vision, technology trends, and challenges for 6G are discussed. The main focus is providing the key enabler of a ubiquitous intelligent
9 mobile society and suggesting a roadmap for the 6G standards.
Recent advances in 6G wireless systems are discussed. The main focus is to present a taxonomy of key technologies and open research
10 [21] 2020
challenges and propose practical guidelines such as neural networks and blockchain-based secure business models.
11 A discussion of 6G wireless communication technologies is provided, emphasizing fundamental breakthroughs at the physical layer. The
12 [31] 2020 main focus is to provide an overview of these technologies, including holographic radio, terahertz communication, large, intelligent
13 surface, and orbital angular momentum.
The limitations of 5G and the need to develop the 6G wireless system are discussed. The main focus is to provide the vision for 6G and
14 [32] 2020
outline a research agenda for enabling the new services and technologies required.
15 The potential of UAVs in beyond 5G and 6G wireless networks are discussed. The main focus is to highlight the use of cellular networks,
[13] 2021
16 advanced technologies, machine learning, and non-terrestrial networks to support UAVs in 6G.
A fiber-wireless network architecture is presented based on full spectrum, fully adaptive, and coordinated radio access networks (RANs).
17 [23] 2021 The main focus is to offer promising scenarios such as NR-free space optical backhauling and indoor systems via visible light
18 communication for high-speed data link and VLC-aided positioning systems.
19 [24] 2021
The need for 6G to overcome the limitations of current cellular networks and support high-bandwidth applications are discussed. The main
20 focus is to provide an overview of system requirements, potential technologies, and recent research progress.
The overviews of reconfigurable intelligent surfaces (RISs) for 6G wireless networks are explained. The main focus is to provide the use
21 [25] 2021
case of RISs to create a favorable propagation channel and improve performance gains.
22 A comparison of 5G and 6G technologies, including terahertz communication, RIS, and blockchain, are presented. The main focus is to
[26] 2021
23 illustrate how IRS can enhance signal quality by controlling passive reflecting elements and how blockchain can enhance system security.
24 The possibility of integrating terrestrial and NTNs is discussed as a means of improving user experience and connecting unconnected
[16] 2022 devices. The main focus is identifying the opportunities and challenges for defining and orchestrating a new 3D wireless network
25 architecture.
26 The challenges of 5G technology and the potential benefits of 6G technology for edge networks in processing real-time applications are
[28] 2022
27 examined. The focus is on integrating ultra-reliable 6G technology into edge computing networks.
The potential of IoT devices and energy-efficient 6G wireless communication in transforming smart cities into super-smart cities is
28 [33] 2022 conferred. The main focus is to review key technologies and applications, including quantum communication, blockchain, and VLC and
29 identifies promising trends for using 6G through IoT devices in smart cities.
30 [39] 2022
An overview of 6G mobile networks, including motivations, use cases, requirements, and research projects, are reviewed. The main focus
31 is on the transition from 5G to B5G and on the advanced features that will be required for 6G.
The role of wireless backhauls in 5G networks and its integration with new technologies like UAV, HAPS, mmWave, mMIMO, and
32 [40] 2022 beamforming are presented. This article focuses mainly on rural connectivity, mobile edge computing, and security issues related to wireless
33 backhaul in 5G and B5G.
34 [34] 2023
The design of an energy-efficient resource allocation system for NTNs is explained. The main focus is to maximize system energy
efficiency by collaboratively optimizing user equipment association, power control, and UAV deployment.
35
Researchers are exploring five Facets of 6G to develop next-generation solutions, i.e., next-generation architectures, networking, IoT,
36 [38] 2023 wireless positioning and sensing, and deep learning applications. The main focus is to review promising techniques and architectures,
37 address vulnerabilities, and advocate for multi-component Pareto optimization for optimal solutions.
38 [41] 2023
Distributed edge learning (EL) techniques and their integration with advanced communication optimization designs for B5G wireless
networks are explored. The main focus is to present the open problems and emerging application opportunities for the B5G network.
39 We discuss the use of AI in NTNs for 6G communication networks. The main focus of the survey is:
40 1. NTNs role in 6G networks and unique security/privacy concerns.
41 This work
2. Applicable AI approaches for NTN problems and proper technique selection.
42 3. AI-enabled NTN research avenues and superiority over traditional methods.
4. AI-based NTN resource allocation case studies and research.
43 5. Future open issues in AI for NTNs, considering constraints for its maximum potential.
44
45 3) We provide a holistic overview of AI-enabled NTN such as network optimization and resource allocation.
46 research and a motivating argument for their
47 implementation. We explore the challenges of NTNs, B. PAPER ORGANIZATION
48 outline the issue associated with traditional methods, and This paper is organized as follows. Section II provides an
49 provide ideas for the superiority of AI techniques. overview of NTN and its potential role in 6G networks.
50 4) We summarize relevant case studies and existing Different security and privacy concerns of NTNs are
51 research used to solve the resource allocation problem discussed in Section III. In Section IV, we introduce an
52 associated with AI in NTNs. overview of AI techniques that can be used to optimize the
53 5) We summarize the main challenges and opportunities of performance of NTNs. Section V discusses the recent
54 using AI for NTNs. In addition, we provide a set of research in AI-powered NTN for 6G networks. Section VI
55 recommendations for future research directions in this identifies the key research directions and challenges in this
56 area. We also identify potential AI applications in areas field, and Section VII concludes the paper. The structure of
57 the paper is shown in Fig. 1.
58 3
59 For Review Only
60
 IEEE Access Page 26 of 42

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Figure 1. Article Structure
22
23
To facilitate the readability of this paper, Table 2 provides III. Geostationary Orbit (GEO) Satellites: GEO satellites are
24
the list of abbreviations with the terms associated. positioned at approximately 35,786 kilometers above the
25
equator. These satellites have an orbital period that matches
26
II. NON-TERRESTRIAL NETWORKS the Earth's rotation, allowing them to remain fixed relative
27 NTNs have emerged as a promising solution for extending to a specific point on the Earth's surface. This characteristic
28 coverage and improving connectivity in 6G networks. NTNs makes them ideal for applications that require continuous
29 are a class of wireless networks that operate using non-earth- coverage of a specific geographic area, such as television
30 based platforms to provide connectivity [3],[10],[11]. Unlike broadcasting and weather monitoring.
31 terrestrial networks, NTNs can cover large areas, operate IV. High Altitude Platform Stations (HAPS): HAPS are
32 independently of terrestrial infrastructure, and be quickly unmanned platforms that operate in the stratosphere at
33 deployed in areas without adequate terrestrial infrastructure altitudes between 17 to 22 kilometers. They have the
34 [8].
potential to provide connectivity to areas that are difficult to
35
access or where terrestrial infrastructure is not feasible.
36 A. OVERVIEW OF NTNs TECHNOLOGIES
HAPS can be used for a range of applications, including
37 I. Low Earth Orbit (LEO) Satellites: LEO satellites are
communication, surveillance, and environmental
38 satellites that orbit the Earth at altitudes between 200 and
monitoring. HAPS includes balloons and airships. Some
39 2000 kilometers. They have emerged as a promising
specific examples might include Google’s Project
40 platform for providing connectivity to remote and rural
Loon, Facebook’s Aquila, and the Stratobus airship
41 areas. Recent developments in satellite technology have
[44]-[45].
42 reduced the cost of launching and operating LEO satellites,
V. Unmanned Aircraft Systems (UASs): UASs operate
43 making it a viable option for commercial use. Some
without a pilot. UASs can be quickly deployed to provide
44 specific technologies for LEO include the International
connectivity in areas affected by natural disasters or
45 Space Station (ISS), the Hubble Space Telescope, and
emergencies. They can be used for a variety of applications,
46 the Global Positioning System (GPS) [42]-[43].
including communication, surveillance, and delivery. Some
47 II. Medium Earth Orbit (MEO) Satellites: MEO satellites are
specific examples that use the UASs are DJI Mavic, the
48 positioned in orbits between approximately 2,000 and
Parrot Bebop 2, and Lockheed Martin Indago3 [46],
49 35,786 kilometers above the Earth's surface. They offer a
[47],[48],[49].
50 compromise between the coverage area and the signal
51 delay. MEO satellites are commonly used for navigation
B. ROLE OF NTNs IN 6G NETWORKS
52 and communication purposes. MEO allows longer
6G wireless communications networks will require a variety
53 communication windows than LEO satellites, resulting in
of new technologies to meet the high data rate, low latency,
54 reduced handovers and a more stable connection for users.
and mobility requirements of the future network. One
55 MEO is often used for global navigation satellite systems
possible solution is using NTNs that offer a viable option to
56 (GNSS) like GPS, GLONASS, and Galileo, as well as for
provide these services cost-effectively and efficiently. They
57 remote sensing and communication purposes.
58 4
59 For Review Only
60
Page 27 of 42 IEEE Access

1
2
3 Table 2. List of Abbreviations impractical. Moreover, NTNs can also serve as a backup or
4 Abbreviation Definition redundant network in case of network failures or disasters,
5 3GPP 3rd generation partnership project ensuring high reliability and availability of communication
6 5G Fifth generation services.
7 6G Sixth generation To date, several research efforts have been undertaken to
8 AI Artificial intelligence explore the potential of NTNs for 6G networks. For instance,
9 AP Access point the integration of LEO satellites into 6G communications
10 B5G Beyond fifth generation networks has been studied extensively, with a focus on
11 BSs Base stations optimizing the satellite constellation design, routing
12 CNN Convolutional neural networks
algorithms, and interference management techniques [50].
13 DBN Deep belief network
Similarly, the use of HAPs in 6G networks has been
14 investigated, focusing on developing efficient
DL Deep learning
15 communication protocols, beamforming techniques, and
DRL Deep reinforcement learning
16 energy-efficient power management schemes [51]. The use
EL Edge learning
17 of UASs in 6G networks has also been explored, with a focus
FL Federated learning
18 on developing aerial base stations and efficient trajectory
GA Genetic algorithms
19 planning algorithms [52].
GPS Global positioning system
20 Overall, the role of NTNs in 6G networks is crucial for
HAPs High-altitude platforms providing ubiquitous, reliable, and high-capacity wireless
21 IDS Intrusion detection systems
22 communication services. The unique characteristics of
IoT Internet of things NTNs, such as high altitude, 3D mobility support, and broad
23
IPS Intrusion prevention system coverage, make them a promising solution for meeting the
24
ISS International space station requirements of future wireless networks. Ongoing research
25
26
LEO Low earth orbit efforts are expected to further improve the performance and
27 MARL Multi-agent reinforcement learning efficiency of NTNs in 6G networks and pave the way for the
28 MIMO Multiple input multiple output realization of the full potential of these networks.
29 mMIMO Massive multiple input multiple output
30 ML Machine learning C. KEY NTNs FEATURES AND REQUIREMENTS
31 NFV Network function virtualization NTNs are critical for ensuring their success in 6G
32 NTNs Non-terrestrial networks communication systems. NTNs architecture must be
33 ORAN Open radio access network designed to address key challenges such as 3D mobility link
34 PCA Principal component analysis reliability, latency, energy efficiency, and capacity. The
35 QoS Quality of service
NTNs system must be able to support a large number of
36 RA Resource allocation
devices with high data rates and low latency while also being
37 scalable, flexible, and cost-effective.
RAN Radio access network
38 One key feature of NTNs is the use of advanced antenna
RIS Reconfigurable intelligent surfaces
39 technologies, such as beamforming, which can enhance
RL Reinforcement learning
40 signal strength and reduce interference. Other key features
SDN Software-defined networks
41 include the use of multi-frequency bands, efficient power
SL Supervised learning
42 management techniques, and advanced modulation schemes.
SVMs Support vector machines
43 NTNs must have strong authentication, encryption, and
UASs Unmanned aircraft systems access control mechanisms to ensure security and privacy.
44 UAV Unmanned aerial vehicle
45 NTN requirements for 6G communication systems include
UL Unsupervised learning their ability to provide ubiquitous connectivity, reliability,
46 VLC Visible light communication
47 and high data rates with low latency. The system must
VNFs Virtual network functions support seamless integration with terrestrial networks and
48
are expected to play a vital role in enabling the full potential enable global coverage with minimal delay. Additionally,
49
of 6G networks. NTNs must be designed to meet the specific needs of various
50
The terrestrial network infrastructure, including 5G applications, such as the Internet of Things (IoT), smart
51
networks, has limitations in terms of coverage (e.g., cities, and connected vehicles. It must be able to support
52
remote/rural areas, sea/air, etc.). One of the key roles of high-altitude platforms, LEO satellites, and geostationary
53
NTNs in 6G networks is to complement the terrestrial orbit satellites. The system should also be resilient to natural
54
network infrastructure and overcome its limitations. NTNs disasters and cyber-attacks.
55
can provide broader coverage, higher capacity, and mobility Fig. 2 summarizes the key features and requirements of
56
support in areas where terrestrial networks are unavailable or NTNs for 6G communication systems.
57
58 5
59 For Review Only
60
 IEEE Access Page 28 of 42

1
2
3 legitimate users [53].
4 Another important aspect of authentication and authorization
5 in 6G networks is the regular updating of policies. As
6 security threats evolve over time, it is crucial to keep
7 authentication and authorization policies up to date to
8 address new vulnerabilities and prevent potential security
9 breaches. Regular updates help ensure that the network
10 remains secure against emerging threats and that any new
11 authentication methods or standards are implemented
12 effectively [54].
13 Privacy and Confidentiality: Privacy and confidentiality
14 are crucial considerations for the 6G network due to the vast
15 amount of data generated and the increased use of AI-
16 powered devices. With the proliferation of IoT devices and
17 the integration of AI technologies, 6G networks are expected
18 to handle a massive volume of data, including personal,
19 sensitive, and confidential information.
20 One of the main challenges for privacy and confidentiality in
21 6G networks is ensuring that this data is protected from
unauthorized access as it traverses across various network
22
segments. Data transmitted over the network may pass
23
through multiple nodes, edge computing systems, and cloud
24 Figure 2. Basic Features of NTN
infrastructures, increasing the risk of interception or
25 unauthorized access at any point in the network. Therefore,
26 robust security measures must be implemented to safeguard
III. SECURITY AND PRIVACY IN NTN
27 data privacy and maintain confidentiality [55].
Security and privacy are critical considerations in designing
28 Network Congestion: The 6G network’s high-speed
and implementing NTNs for 6G. The increasing reliance on
29 capabilities can cause network congestion, leading to a
advanced technologies in the telecommunications industry
30 higher risk of network attacks. With more devices connected
means that network security threats are becoming more
31 to the network, the chances of network congestion increase,
sophisticated and complex. As the next generation of
32 creating a larger attack surface [56].
wireless networks, 6G networks are expected to bring a
33 Malware and Hacking: As 6G networks become more
revolutionary transformation to the world of communication.
34 complex, the risk of malware and hacking increases, putting
In the context of 6G, security and privacy concerns are
35 the security and privacy of the network at risk. It is essential
magnified due to the large-scale deployment of networked
36 to have robust, secure technologies in place, such as
devices and the increasing complexity of network
37 encryption, authentication, and firewalls, to protect the
architectures. These concerns include data leakage, identity
38 network. Additionally, users should follow best practices for
theft, and unauthorized access to sensitive information.
39 online security, such as using strong passwords and being
Developing secure and privacy-preserving technologies is
40 aware of potential phishing scams [36].
essential to ensure the secure operation of 6G networks. The
41 Resource Allocation: 6G technology is expected to have
following is a list of security and privacy concerns for the
42 enhanced security features to ensure data is protected from
NTNs in 6G networks.
43 unauthorized access. Additionally, the network must be able
Authentication and Authorization: Authentication and
44 to allocate resources to meet users' demands efficiently.
authorization refer to verifying the identity of devices or
45 Finally, the network must be able to adapt to changing user
users and determining their level of access privileges within
46 needs and respond quickly to changing conditions [57].
the network. These mechanisms are crucial for ensuring the
47 Fault Tolerance: Identifying and mitigating faults in the
security and privacy of the network, as they prevent
48 network is essential for maintaining the network’s security
unauthorized devices or malicious actors from gaining
49 and privacy. This is done by detecting errors, isolating them,
access and potentially causing harm.
50 and recovering from them without impacting the system.
Advanced technologies and protocols will be required to
51 Fault tolerance is a key element in network security and helps
achieve strong authentication in 6G networks. One potential
52 prevent unauthorized access or misuse of the network. It also
approach is biometric authentication, which relies on
53 helps to protect against data loss [58].
individuals' unique physical or behavioral traits, such as
54 Trust Management: Trust management is vital to the 6G
fingerprints or facial recognition. These biometric
55 network security, enabling secure interactions and
characteristics can serve as strong authentication factors,
56 collaborations between different devices, services, and users.
making it difficult for unauthorized entities to impersonate
57
58 6
59 For Review Only
60
Page 29 of 42 IEEE Access

1
2
3 As 6G networks become more complex, with a diverse range activities. Organizations can significantly reduce the risk of
4 of devices, services, and stakeholders, managing trust successful attacks and mitigate potential damage by acting
5 becomes increasingly challenging. Trust-related attacks, preemptively.
6 such as malicious entities masquerading as trusted devices or Identity Management: AI can manage identities and access
7 services, pose a significant risk. These attacks can lead to control more effectively, ensuring that only authorized
8 unauthorized access, data breaches, or disruptions in network devices and users can access the network resources. AI can
9 operations. facilitate advanced authentication mechanisms such as
10 To address the trust management challenges in 6G networks, biometric recognition, behavior analysis, and contextual
11 several approaches can be considered. One such approach is information to verify the identity of users and devices
12 the use of trust models and frameworks that assess and accurately. By leveraging AI-driven identity management
13 quantify the trustworthiness of devices, services, or entities solutions, 6G networks can enhance security, prevent
14 based on their behavior, reputation, or credentials. These unauthorized access, and mitigate privacy risks.
15 models can help establish a trust hierarchy and determine the Encryption: AI can contribute to enhancing encryption
16 level of trustworthiness associated with different entities techniques by improving encryption algorithms, key
17 within the network. management, and overall cryptographic processes.
18 Additionally, secure authentication protocols, such as public Leveraging AI algorithms can strengthen encryption
19 key infrastructure or certificate-based authentication, can be mechanisms to withstand increasingly sophisticated attacks
20 employed to verify the identities of devices or entities and and cryptographic vulnerabilities. AI can assist in
21 ensure that only trusted entities are granted access to the developing more robust encryption algorithms, optimizing
22 network [59]. key generation and distribution, and detecting potential
23 AI can play a crucial role in mitigating these security and weaknesses or patterns that attackers may exploit. This helps
24 privacy concerns. AI can be leveraged to provide enhanced create a more secure data transmission and storage
25 security and privacy features that go beyond traditional environment in 6G networks [60].
26 security measures. The following are some of the ways AI Behavioral Analysis: AI algorithms can analyze user and
27 can help resolve these concerns. device behavior patterns, allowing the network to detect and
28 Threat Detection and Prevention: Threat detection and prevent abnormal activities that could signify a potential
29 prevention play a crucial role in ensuring network security. security threat. By continuously monitoring behavior in real-
30 As the complexity of the 6G network increases; therefore, time, AI-based systems can identify suspicious actions, such
31 advanced mechanisms are required to detect and mitigate as unauthorized access attempts, unusual data transfers, or
32 potential threats in real-time. anomalous user behaviors. This proactive approach enables
33 AI can be a powerful threat detection and prevention tool in the network to respond swiftly and mitigate potential threats
34 6G networks. AI algorithms can analyze large volumes of before they can cause harm [17].
35 network data and identify patterns or anomalies that indicate In summary, the 6G network’s security and privacy concerns
36 potential security threats. By leveraging machine learning are significant and must be adequately addressed to ensure
37 techniques, AI systems can continuously learn and adapt to the network’s safety and secure operation. AI can provide
38 new attack vectors and evolving threats, effectively detecting enhanced security and privacy features, enabling the network
39 known and unknown threats. to mitigate these concerns.
40 Anomaly Detection: AI can be used to identify and flag
unusual behavior on the network, which may be an indicator IV. AI TECHNIQUES FOR NTNs OPTIMIZATION
41
of a potential security breach. The optimization process plays an important role in
42
Predictive Analysis: Predictive analysis uses AI techniques improving the performance and efficiency of wireless
43
to forecast potential security threats and take proactive networks deployed in non-terrestrial environments such as
44
measures to prevent them before they materialize. AI- satellites, drones, and balloons. Optimization can be used to
45
powered predictive analysis can identify indicators that may identify the best settings for the network, such as the power
46
lead to future security incidents or attacks by analyzing level, frequency, and type of antenna, as well as the optimal
47
historical data, network patterns, and security trends. placement of devices. This can lead to improved coverage,
48
Using ML algorithms, predictive analysis models can learn increased bandwidth, and reduced latency. Additionally,
49
from historical data to recognize patterns and correlations optimization can save energy by reducing the amount of
50
associated with security threats. These models can then power required to operate the network.
51
analyze real-time data and identify early warning signs or Numerous approaches were defined for NTN optimization.
52
anomalies that could indicate an imminent security breach. These traditional methods are based on mathematical models
53
The proactive measures taken based on predictive analysis and simulations that use prior knowledge of the network
54
can include strengthening network defenses, implementing parameters and environment to make predictions and
55
additional security controls, or raising alerts to security decisions. For example, network planning involves
56
teams to investigate potential vulnerabilities or suspicious designing the network architecture, coverage area, and
57
58 7
59 For Review Only
60
 IEEE Access Page 30 of 42

1
2
3 capacity based on the expected traffic and user requirements SL involves training a model using labelled data, where the
4 [61]. Link budget analysis involves calculating the power input-output pairs are known in advance. The model can then
5 budget and signal-to-noise ratio for each link in the network be used to make predictions on new inputs. In NTN
6 to ensure reliable and high-quality communication [62]. optimization, SL can be used for various tasks, such as
7 Antenna design involves selecting the antenna’s type, size, channel estimation [70], interference mitigation [71], and
8 and orientation based on the frequency, gain, and radiation resource allocation [72]. For example, a deep neural network
9 pattern [63]-[64]. Signal processing involves filtering, (DNN) can be trained to predict the best channel and power
10 equalizing, and modulating the signal to optimize allocation for a given set of users and resources. The
11 transmission and reception [65]. advantage of supervised learning is that it can produce
12 These traditional approaches have been used for a long time, accurate results with high precision. SL requires labelled
13 but they have some limitations when it comes to NTN data, where the input-output pairs are known in advance.
14 optimization. First, they require prior knowledge of the However, in NTN optimization, it may be difficult to obtain
15 network parameters and environment, which may not be labelled data, as the network operates in dynamic and
16 accurate or up-to-date, especially in dynamic and uncertain uncertain conditions, and the ground truth may not be
17 environments. Second, they are not adaptive enough to available. In addition, the labelled data may not be
18 handle the changing conditions and demands of the network, representative of the entire network, leading to biased or
19 as they rely on fixed and predefined models and rules. Third, inaccurate models. SL is shown in Fig. 4.
20 they may not be able to handle the heterogeneity and UL involves training a model using unlabeled data, where
21 complexity of the network, as they assume a uniform and the input-output pairs are not known in advance. The model
22 idealized network topology and behavior. Therefore, can then discover patterns and structures in the data and
23 exploring new approaches based on AI techniques can group them into clusters or categories. In NTN optimization,
24 overcome these limitations and provide more accurate and unsupervised learning can be used for various tasks, such as
25 adaptive solutions for NTN optimization [66]. These AI anomaly detection [73], network clustering [74], and traffic
26 techniques can learn from the data and adapt to changing analysis [75]. For example, a self-organizing map can be
27 environments. The AI techniques can be broadly classified used to cluster the satellites or drones based on their location
28 into three categories: supervised learning (SL) [67], and connectivity [76]. The advantage of unsupervised
29 unsupervised learning (UL) [68], and RL [69]. The basic learning is that it can find hidden patterns in data that may
30 architecture of AI can be shown in Fig. 3. not be easily visible. Unsupervised learning does not require
31 labelled data, but it relies on the assumption that the data has
32 some inherent structure or pattern that can be discovered.
33 However, in NTN optimization, the network may be too
34 complex or heterogeneous, and the data may not have clear
35 patterns or clusters that can be easily identified. In addition,
36 unsupervised learning may suffer from the problem of
37 overfitting, where the model memorizes the data instead of
38 learning the underlying structure. The basic structure of UL
39 is seen in Fig. 5.
40
41
42
43
44
45
46
47
48
Figure 4. Supervised Learning
49
50
51
52
53
54
55
56
Figure 5. Unsupervised Learning
57 Figure 3. Artificial Intelligence Architecture
58 8
59 For Review Only
60
Page 31 of 42 IEEE Access

1
2
3 RL involves training an agent to interact with an functions based on the new observations, which can result in
4 environment and learn an optimal policy to maximize a more robust and efficient operations.
5 cumulative reward [77]. In NTN optimization, RL can be As summarized in Table 3. SL can address some issues of
6 used for various tasks, such as beamforming, routing, and NTN optimization, such as anomaly detection, channel
scheduling. For example, a Q-learning algorithm can be used estimation, interference mitigation, and resource allocation.
7
to find the optimal beamforming angles and power levels for UL can address issues like network clustering and traffic
8
a given set of antennas and users [78]. analysis. However, neither of them can address the dynamic
9
One of the advantages of RL is that it can handle dynamic and uncertain environment of NTN optimization, which is a
10 critical factor for NTNs.
11 environments with large state and action spaces. However, it
is important to note that RL is typically used for discrete state DRL, on the other hand, can address all of the above issues,
12 as it can learn the optimal policy for complex and dynamic
13 and action spaces and is not well-suited for problems with
continuous state and action spaces. Additionally, while environments where the state and action spaces are high-
14 dimensional and continuous. DRL can be used for
15 RL can handle dynamic environments but may struggle with
large and complex ones. Overall, while RL has shown beamforming, routing, and scheduling tasks requiring real-
16 time decision-making and optimization. Therefore, RL,
17 promising results in a range of domains, it is crucial to
consider its applicability to the specific problem. In the especially DRL, is preferred for NTN optimization in 6G
18 networks, where the goal is to provide reliable and high-
19 context of NTN optimization, RL can be a powerful tool, but
it may not always be the best choice depending on the speed connectivity to remote and underserved areas.
20
21 complexity and nature of the problem. A basic representation
of RL is shown in Fig. 6. A. POSSIBLE APPLICATIONS OF AI IN NTNs
22 AI has the potential to revolutionize NTN optimization in 6G
23 DRL is an extension of RL that has been shown to be more
effective in handling complex and high-dimensional networks by enabling faster and more efficient
24 communication. There are several possible applications of
25 problems. In RL, the policy and value functions are
represented using simple linear or nonlinear models, AI in NTNs for 6G adoption, including Terahertz
26 Communications, Optical Wireless Communication, Remote
27 whereas, in DRL, deep neural networks are used to represent
these functions. This allows the DRL models to learn more Sensing, and Industrial Internet of Things (IIoT).
28 One possible application is satellite communication. AI can
29 complex and abstract representations of the state and action
spaces, which can result in better performance and more be used to optimize communication between satellites,
30 ground stations, and users. It can also help reduce latency
31 efficient exploration of the environment.
DRL has been able to achieve state-of-the-art performance and interference, thereby improving the system's overall
32 performance. For instance, AI can be used to optimize the
33 in various domains, such as robotics, games, wireless
communication, and natural language processing. These scheduling and routing of satellite links and predict and
34 mitigate the effects of atmospheric attenuation and weather
35 successes have motivated researchers to explore the use of
DRL for NTN optimization, where the state and action conditions.
36 Another application is drone-based connectivity. AI can be
37 spaces are often high-dimensional and continuous. By using
DRL, it is possible to learn an optimal policy for resource used to optimize the coverage and capacity of drone-based
38 networks, which can be used for various applications, such
39 allocation [79], [80], [81],[82], routing [83], and scheduling
[84] that operate in dynamic and uncertain conditions in as search and rescue, environmental monitoring, and
40 precision agriculture. For example, AI can be used to
41 NTN.
42 Another advantage of DRL is that it can handle the
heterogeneity and complexity of the network, which is Table 3. Dynamics of SL, UL and RL
43 Issues SL UL RL
44 difficult to model using traditional approaches. For example,
in a satellite network, the number and position of the Dynamic and uncertain environment x x ✓
45
satellites may change over time, the users may move in Heterogeneity and complexity of the network x ✓ ✓
46
different directions, and the signal quality may vary High-dimensional and continuous state and x x ✓
47
depending on the atmospheric conditions. DRL can adapt to action spaces
48
these changes by continuously updating the policy and value Anomaly detection ✓ ✓ x
49
Channel estimation ✓ x x
50
51 Interference mitigation ✓ x x
52 Resource allocation ✓ x x
53 Network clustering x ✓ x
54 Traffic analysis x ✓ x
55 Beamforming x x ✓
56 Routing x x ✓
Figure 6. Reinforcement Learning ✓
57 Scheduling x x
58 9
59 For Review Only
60
 IEEE Access Page 32 of 42

1
2
3 optimize the trajectory and positioning of drones and manage planning and design, where AI can be used to optimize the
4 the handover between drones and ground stations. placement and configuration of network elements such as
5 AI can also be used to optimize the altitude, trajectory, and antennas, base stations, and routers. This can help to improve
6 communication of balloon-based networks. AI can help coverage, capacity, and QoS while minimizing costs and
7 optimize the positioning and deployment of balloons and energy consumption.
8 predict and mitigate the effects of atmospheric conditions Another area is resource allocation, where AI can optimize
9 and other environmental factors. network resource allocation, such as bandwidth, power, and
10 Finally, AI can be used to optimize energy consumption, data spectrum. This can help to improve network efficiency and
11 processing, and transmission of IoT and remote sensing capacity while minimizing interference and congestion.
12 devices. This can be particularly important in limited energy AI can also be used for fault detection and diagnosis,
13 applications, such as remote and off-grid locations. AI can automatically identifying and diagnosing network faults and
14 be used to optimize the routing and processing of data and anomalies and providing recommendations for remedial
15 predict and mitigate the effects of interference and other action. This can help to reduce downtime and improve
16 environmental factors. network availability and reliability.
17 Table 4 summarizes the possible applications of AI in NTNs Finally, AI can be used for network security, automatically
18 for 6G networks. detecting and preventing security threats such as intrusion,
19 malware, and denial-of-service attacks. This can help to
20 B. AI-POWERED NETWORK OPTIMIZATION AND protect network assets and data and maintain user privacy
MANAGEMENT and trust.
21
AI-powered network optimization and management involves In summary, AI-powered network optimization and
22
the application of AI techniques to improve the performance, management have the potential to transform the way we
23
efficiency, and reliability of 6G communication networks. design, operate, and maintain communication networks. By
24
These techniques include ML, DL, and RL, which can be providing automated monitoring, analysis, and decision-
25
used to learn from network data, predict network behavior, making tools, AI can help improve network performance,
26
and make automated decisions. efficiency, and reliability and enable new applications and
27
One of the key benefits of AI-powered network optimization services.
28
is its ability to handle the complexity and heterogeneity of
29
modern communication networks. With the rise of 5G and C. AI-POWERED RECONFIGURABLE INTELLIGENT
30
the advent of 6G, networks are becoming more diverse and SURFACE WITH NTNs
31 Reconfigurable intelligent surface (RIS) is a new technology
dynamic, with a wide range of devices, protocols, and
32 that can manipulate electromagnetic waves by dynamically
services. AI can help manage this complexity by providing
33 changing their surface properties [85]. They consist of a
automated monitoring, analysis, and optimization tools.
34 planar array of small, passive, and tunable elements that can
There are several areas in which AI-powered network
35 manipulate electromagnetic waves to improve the
optimization can be applied. One of these is network
36 performance of wireless networks. RIS technology can
37 Table 4. AI Possible NTN Applications enhance wireless networks' coverage, capacity, and QoS by
38 Application Description adjusting the phase, amplitude, and direction of
39 Optimize communication between satellites, ground electromagnetic waves.
40 Satellite
stations, and users; reduce latency and interference; RIS can be used to optimize the signal strength and quality
41 communication
optimize scheduling and routing of satellite links. of NTN, improving its coverage and capacity. When RIS
42 technology is combined with NTNs, the resulting system can
Optimize coverage and capacity of drone-based
43 improve wireless communication performance even more.
networks; manage handover between drones and
44 Drone-based
This performance improvement is especially beneficial in
ground stations; enable search and rescue,
45 connectivity
areas with poor network coverage. RIS is also more cost-
environmental monitoring, and precision
46 effective than other methods of improving NTN signal
agriculture.
47 strength, making it a great solution for both small and large
Optimize altitude, trajectory, and communication of
48 networks [86]. Recent research has shown that integrating AI
balloon-based networks; provide connectivity to
49 Balloon-based
with RIS technology can improve the performance and
remote and underserved areas; predict and mitigate
50 networks
reliability of NTNs, such as those used for satellite
effects of atmospheric conditions and environmental
51 communications, deep-space exploration, and interplanetary
factors.
52 networks. These networks face unique challenges, such as
Optimize energy consumption, data processing, and
53 limited bandwidth, high latency, and harsh environmental
transmission of IoT and remote sensing devices;
54 IoT and remote
conditions that can impede data transmission and
optimize data routing and processing; predict and
55 sensing
connectivity [87]-[88].
mitigate effects of interference and environmental
56
factors.
57
58 10
59 For Review Only
60
Page 33 of 42 IEEE Access

1
2
3 AI-powered RIS can help to overcome some of these Crowdsensing, a collective sensing paradigm, taps into the
4 challenges by optimizing the wireless channel in real-time. power of connected individuals and their mobile devices to
5 By leveraging AI algorithms, RIS can adapt to changing gather data about the environment [96]. Through mobile
6 network conditions and dynamically adjust its configuration apps or wearable devices, individuals contribute data on
7 to improve signal quality, increase network capacity, and various aspects like traffic conditions, air quality, noise
8 reduce interference [89]-[90]. For example, AI-powered RIS levels, and social behaviors. This distributed data collection
9 can intelligently direct signals to avoid obstacles and approach provides extensive coverage and delivers real-time
10 optimize coverage, which can be especially useful in deep- insights into the physical world, empowering communities
11 space communications where distance can severely attenuate with a deeper understanding of their surroundings.
12 signal strength [91]. Moreover, AI-powered RIS can enable The relationship between edge computing, crowdsensing,
13 more efficient use of the available spectrum, which is an NTN, and AI is pivotal in advancing the capabilities of
14 important resource in NTNs [92]. By intelligently managing modern systems. In [97], a Stackelberg game-based
15 signal strength and direction, RIS can increase the capacity computation offloading method can be integrated into edge
16 of the existing spectrum, reduce the risk of interference, and computing and crowdsensing systems to optimize data
17 enable more reliable data transmission. processing at the network edge. Building upon this, the [98]
18 One of the key benefits of AI-powered RIS is its ability to proposes a Stackelberg game approach with the assistance of
19 create a more dynamic and adaptive wireless environment. UAVs, enhancing the offloading process in mobile edge
20 By constantly analyzing and adjusting the wireless channel, computing networks. The proposed approaches optimize RA
21 RIS can ensure that the network remains resilient and and enhance system performance by leveraging the insights
22 reliable, even in the face of equipment failure or gained from crowdsensing data and utilizing AI techniques,
23 environmental disturbances. Additionally, RIS can be used such as ML. Furthermore, [99] focuses on cost-
24 to create more secure and resilient NTNs by controlling and minimization-oriented computation offloading and service
25 manipulating the propagation of radio waves to prevent caching, employing an advanced ML-based approach.
26 unauthorized access and mitigate the risk of cyber-attacks Integrating such methodology with edge computing,
27 [93]. AI-powered RIS also has the potential to revolutionize crowdsensing, and NTN enables efficient resource
28 the field of space exploration. By enabling more advanced management, reduced operational costs, and intelligent
29 and sophisticated communications systems, RIS can improve decision-making based on data collected from NTN-enabled
30 the accuracy and reliability of spacecraft navigation systems, devices and sensors. Collectively, [97]-[99] contribute to
31 enabling more precise targeting and maneuvering. integrating edge computing, crowdsensing, NTN, and AI,
32 Furthermore, RIS can create more efficient and robust data fostering advancements in various domains, including smart
33 transmission systems for deep-space exploration, enabling cities, industrial IoT, and remote sensing applications.
34 more rapid and reliable transmission of scientific data and When edge computing, crowdsensing, NTNs, and AI come
35 images [94]. together, the possibilities for transformative applications
36 Although the integration of AI and RIS technology has great across domains become endless. In the realm of smart cities,
37 promise for enhancing the performance and reliability of this integration enables edge devices connected through
38 NTNs for future advancements in space exploration, satellite NTNs to monitor traffic flow, analyze real-time air quality
39 communications, and interplanetary networks, however, the data, and optimize energy consumption based on demand
40 technical challenges still need to be addressed; ongoing and availability. Advanced AI algorithms identify congestion
41 research and development in this field suggests that AI- hotspots, predict air pollution levels, and recommend
42 powered RIS will play an increasingly important role in efficient routes or dynamically adjust energy usage, fostering
43 shaping the future of wireless technology in non-terrestrial sustainable urban environments.
44 environments. Furthermore, this integration profoundly impacts public
45 safety and emergency response. In the event of natural
46 D. RELATIONSHIP OF EDGE COMPUTING AND disasters or emergencies, the combination of edge computing
47
CROWDSENSING WITH NTN AND AI and crowdsensing allows for rapid data collection and
With its decentralized approach, edge computing analysis. By harnessing data from mobile devices, sensors,
48
revolutionizes data processing and analysis by bringing and surveillance systems, AI algorithms can detect critical
49
computational power closer to the edge of the network. This situations in real-time, such as identifying earthquake-prone
50
proximity to data sources enables real-time insights and areas or predicting the spread of wildfires. This timely
51
analysis, minimizing latency and bandwidth usage and information enables emergency responders to allocate
52
making it ideal for time-sensitive applications. By leveraging resources effectively, coordinate evacuation plans, and save
53
a network of devices such as edge servers, gateways, and lives. The healthcare sector benefits tremendously from this
54
edge sensors, edge computing infrastructure collaboratively synergy. Crowdsensing facilitates remote patient monitoring,
55
processes and filters data at the edge, enhancing efficiency collecting vital signs, medication adherence, and activity
56
and scalability [95]. levels through wearable devices. Edge computing and AI
57
58 11
59 For Review Only
60
 IEEE Access Page 34 of 42

1
2
3 process this data in real-time, enabling healthcare providers predictive maintenance. In the past few years, researchers
4 to monitor patients' health conditions, detect anomalies, and have explored various AI-based approaches, such as DL, RL,
5 intervene promptly. Moreover, AI-powered algorithms and DRL to improve network performance [66],[100]. These
6 analyze population-level health data to identify patterns, approaches have been used for various applications such as
7 predict disease outbreaks, and allocate resources efficiently intelligent network slicing, load balancing, and resource
8 during public health emergencies, ultimately saving lives. allocation, enabling the network to operate optimally in
9 The symbiotic relationship between edge computing, dynamic and unpredictable environments.
10 crowdsensing, NTNs, and AI also extends to agriculture, The integration of AI into NTNs has also led to the
11 where edge devices combined with NTNs monitor soil development of new and innovative network architectures
12 moisture levels, temperature, and crop health. AI algorithms that can efficiently handle the vast amounts of data generated
13 analyze this data to optimize irrigation, predict crop yields, by 6G networks. One such architecture is the AI-powered
14 and identify disease or pest outbreaks, empowering farmers Open Radio Access Network (ORAN) [101], which utilizes
15 to make informed decisions, increase productivity, and AI to optimize network coverage and capacity and minimize
16 reduce resource waste. interference and latency. The AI-powered ORAN also
17 Edge computing enables real-time equipment performance allows for dynamic network configuration, enabling the
18 and predictive maintenance monitoring in industrial settings, network to adapt to changing network conditions in real-
19 ensuring optimal productivity and minimizing downtime. time. Furthermore, researchers have also explored the use of
20 Coupled with crowdsensing and AI, this convergence AI for network security and privacy in 6G networks. AI-
21 enhances worker safety by detecting hazards in real-time and based security solutions such as Intrusion Detection Systems
22 providing immediate alerts, mitigating accidents, and (IDS) and Intrusion Prevention Systems (IPS) can identify
23 improving overall workplace security. and prevent security threats in real-time, improving the
24 However, as these technologies advance, addressing network's overall security [102]. AI-powered privacy
25 important considerations such as data privacy and security is solutions such as differential privacy can also protect user
26 crucial. Robust security measures must be implemented to data while enabling network operators to collect valuable
27 protect sensitive information as data is collected and data for network optimization and management. In addition,
28 processed at the edge. Additionally, establishing data AI is being used to address energy efficiency challenges in
29 governance frameworks and ethical guidelines is imperative 6G networks [66]. With the increasing demand for high-
30 to ensure the responsible and transparent use of data, speed connectivity, the energy consumption of 6G networks
31 building trust, and fostering the long-term sustainability of is expected to increase significantly. However, AI-based
32 these technologies. approaches such as energy-efficient resource allocation,
33 In summary, the relationship between edge computing, dynamic sleep mode management, and intelligent power
34 crowdsensing, NTNs, and AI represents a powerful force control can help to reduce energy consumption and improve
35 that has the potential to revolutionize various industries and the overall energy efficiency of the network.
36 significantly improve the quality of life. By bringing
37 computational power closer to the source of data generation, A. NETWORK PLANNING AND OPTIMIZATION
38 leveraging the crowd's collective intelligence, extending The use of AI algorithms in network planning and
39 connectivity beyond traditional networks, and harnessing the optimization offers numerous benefits, including efficient
40 analytical capabilities of AI, this integration unlocks network management, automatic optimization, and
41 unprecedented opportunities for real-time insights, intelligent decision-making. AI-driven network planning can
42 intelligent decision-making, and enhanced efficiency. As help ensure optimal resource allocation, reduce the cost of
43 these technologies continue to advance and intertwine, their network deployment, and enhance network performance.
44 intricate synergy will shape the future of connectivity, data With the ability to analyze vast amounts of data from
45 analytics and redefine the way we interact with the digital multiple sources, such as user behavior, network traffic, and
46 world, paving the way for a smarter, more sustainable future. resource utilization, AI can enable efficient network
47 This transformative potential is not only limited to large- management and improve users' QoS.
48 scale industries but also has the ability to empower ML algorithms, such as Support Vector Machines (SVMs)
49 individuals and local communities, democratizing access to and Random forests, have been utilized to predict network
50 information and fostering innovation at every level. traffic, optimize routing, and allocate resources effectively.
51 SVMs can be used to classify network traffic based on
V. RECENT ADVANCES IN AI-POWER NTN FOR 6G various criteria, such as the source of the traffic, the type of
52
NETWORK data being transmitted, and the destination of the traffic. This
53 In recent years, significant progress has been made in
54 enables service providers to allocate network resources more
developing AI-powered NTNs. The use of AI in NTNs efficiently and reduce congestion. On the other hand,
55 provides numerous benefits, such as intelligent network
56 Random Forests can be used to predict network traffic based
management, automatic network optimization, and on historical data and network behavior, enabling service
57
58 12
59 For Review Only
60
Page 35 of 42 IEEE Access

1
2
3 providers to allocate network resources more effectively. the effectiveness of AI-based approaches in 6G networks.
4 DRL and Genetic Algorithm (GA) have also been applied to One such study focused on using AI in network slicing,
5 improve the network architecture and adapt to changing which allows network operators to partition the network into
6 environmental conditions, enhancing the network's overall virtual slices with customized functionalities and resources.
7 performance. DRL is an AI technique that involves training The study demonstrated that the use of AI-based approaches
8 an agent to act in an environment to maximize a cumulative such as DL and CNNs could significantly improve network
9 reward. This approach has been used to optimize network slicing performance, enabling better resource utilization and
10 routing, enabling the network to adapt to changing traffic enhancing the QoS for end-users [103].
11 patterns and reduce latency. Conversely, GA has been used The authors in [110] propose a cooperative transmission
12 to optimize network parameters such as the number of BSs scheme between a satellite and an aerial BS based on two
13 and the frequency allocation, enabling the network to operate unsupervised ML algorithms, namely K-means and K-
14 optimally in dynamic and unpredictable environments. medoids. The simulation results show that the proposed
15 AI can analyze data from multiple sources, such as user approach demonstrates promising gains in terms of spectral
16 behavior, network traffic, and resource utilization to identify efficiency and system sum rate. In [111], the authors
17 areas where network resources are being underutilized or consider the issue of mobility management due to the
18 overutilized. This can help service providers optimize their movements of LEO satellites. They group user equipment in
19 network architecture and allocate resources more efficiently, different clusters based on K-means clustering algorithm and
20 reducing the overall cost of network deployment. decide handover process based on the distance from its cell
21 Furthermore, AI-driven network planning can ensure center.
22 efficient resource allocation and reduce the cost of network In [104], the authors focused on using RL to optimize
23 deployment, thereby enabling service providers to deliver network resource allocation in dynamic and unpredictable
24 high-quality services while reducing operational costs. environments. The study proposed a new approach to
25 The application of AI in network planning and optimization resource allocation, which combined RL with the GA to
26 is expected to become more widespread with the enable the network to adapt to changing environmental
27 advancement of technology and the increasing demand for conditions. The experimental results demonstrated that the
28 high-speed and reliable connectivity in 6G networks. proposed approach can significantly improve network
29 However, there are several challenges that must be addressed performance by optimizing the allocation of resources to
30 to ensure the effective implementation of AI in network different network functions.
31 planning and optimization. These challenges include the The work in [105] adopted similar MARL strategy for
32 need for standardized data formats, the integration of AI dynamic network slicing. However, the proposed approach
33 algorithms with existing network infrastructure, and the accelerated the policy deployment by integrating a transfer
34 development of ethical guidelines for the use of AI in learning method. The results showed that the proposed
35 network management. approach can achieve 27% better network performance and
36 In summary, AI-driven network planning and optimization utilization compared to traditional approaches.
37 have emerged as critical research areas in developing AI- In [106], an AI-based approach for energy-efficient resource
38 powered NTNs for 6G networks. AI algorithms offer allocation in 6G networks is proposed. The proposed
39 numerous benefits, such as efficient network management, approach utilized a GA to optimize the allocation of network
40 automatic optimization, and intelligent decision-making. ML resources, considering each network function's energy
41 algorithms such as SVMs and Random Forests have been consumption. The results demonstrated that the proposed
42 utilized to predict network traffic and allocate resources approach can significantly reduce the network's energy
43 effectively, while DRL and GA have been applied to consumption while maintaining the desired level of network
44 improve the network architecture and adapt to changing performance.
45 environmental conditions. AI-driven network planning can Another study by [107] focused on using AI in network
46 ensure efficient resource allocation and reduce the cost of function virtualization (NFV) to improve network efficiency
47 network deployment, thereby enabling service providers to and resource utilization. The study proposed an AI-based
48 deliver high-quality services while reducing operational approach for the placement of virtual network functions
49 costs. However, there are several challenges that must be (VNFs) in the network, which utilized a DRL algorithm. The
50 addressed to ensure the effective implementation of AI in results showed that the proposed approach can significantly
51 network planning and optimization. improve network performance and reduce the resource
52 utilization of the network.
53 B. CASE STUDIES AND EXPERIMENTAL RESULTS In a different study [108], the authors proposed an AI-based
54 With the rapid development of 6G networks, AI-powered approach for network traffic prediction in 6G networks. The
55 NTNs have emerged as a critical area of research to improve proposed approach utilized a two-dimensional CNN-based
56 network management, optimization, and performance. long short-term memory (LSTM) to predict network traffic,
57 Several recent case studies have been conducted to evaluate which can enable better resource allocation and network
58 13
59 For Review Only
60
 IEEE Access Page 36 of 42

1
2
3 management. The experimental results showed that the AI-powered NTNs in 6G networks.
4 proposed approach can achieve higher prediction accuracy Multi-objective optimization for AI-powered NTNs: 6G
5 compared to traditional approaches. networks are expected to support diverse applications with
6 In [109], the authors focused on the use of AI in network different quality-of-service requirements. Multi-objective
7 function placement for edge computing in 6G networks. The optimization can help balance conflicting objectives, such as
8 study proposed an AI-based approach for the placement of energy efficiency, spectral efficiency, and reliability, in AI-
9 network functions in the network, which utilized a Q- powered NTNs.
10 learning algorithm. The results showed that the proposed Intelligent network planning and deployment: AI-
11 approach could significantly improve network performance. powered NTNs can leverage intelligent algorithms to
12 The above-explained case studies are summarized in Table automate the network planning and deployment process.
13 5. This includes using ML to identify the optimal network
14 topology, antenna placement, and resource allocation to
VI. CHALLENGES AND OPPORTUNITIES IN AI- minimize interference and maximize coverage.
15
POWERED NTNs FOR 6G NETWORKS AI-driven resource allocation and management: The use
16 The development of the 6G networks is rapidly advancing,
17 of AI can help to efficiently allocate and manage network
and AI has emerged as a key technology in realizing the resources, including frequency bands, power levels, and
18 vision of intelligent and efficient 6G networks. AI-powered
19 computing resources. This can improve network
network management is expected to provide numerous performance and reduce energy consumption.
20 benefits, including improved network performance,
21 Federated learning for distributed NTNs: Federated
enhanced user experience, and reduced operational costs. learning is an ML technique allowing devices to
22 However, achieving these benefits comes with significant
23 collaboratively learn a shared model without sharing data.
challenges and requires overcoming various technical and This can be used to train AI models for NTNs in a privacy-
24 practical obstacles. Therefore, this survey paper discusses
25 preserving and energy-efficient manner.
the challenges and opportunities in AI-powered network Network slicing for AI-powered NTNs: Network slicing
26 management for 6G networks.
27 allows multiple logical networks to be created on top of a
28 shared physical infrastructure. This can help to meet the
A. KEY RESEARCH DIRECTIONS FOR AI-POWERED
29 diverse requirements of different applications in AI-powered
NTNs
30 NTNs, such as low latency, high bandwidth, and high
As AI technologies continue to evolve, they offer
31 reliability.
tremendous potential to revolutionize the way future 6G
32 Edge intelligence for AI-powered NTNs: Edge computing
networks are designed, deployed, and managed. AI-powered
33 can perform AI computations at the network edge, reducing
NTNs is an emerging research area that aims to leverage AI
34 latency and improving energy efficiency. This includes using
techniques to enhance the performance and efficiency of 6G
35 edge devices, such as BSs and user equipment, to perform AI
networks. However, realizing the full potential of AI-
36 computations and make real-time decisions.
powered NTNs requires identifying key research directions
37 Explainable AI for NTNs: The development of explainable
that can overcome the challenges posed by the highly
38 AI can enhance the transparency and interpretability of
dynamic and complex nature of 6G networks. This section
39 NTNs powered by AI. This includes developing algorithms
discusses some of the most promising research directions for
40 that can explain how AI models make decisions and identify
41 Table 5. Summary of AI Use Cases potential biases in the data.
42 AI Related Work(s) Relevance to NTNs
AI-driven security for NTNs: The use of AI can help to
43 [103] - LSTM Smart network slicing, node
improve the security of NTNs, including detecting and
44 SL [106] - CNN clustering or traffic prediction for
mitigating attacks, identifying vulnerabilities, and enhancing
45 [108] - CNN-LSTM increasing energy efficiency.
privacy and data protection.
46 AI for network optimization and self-healing: AI can be
Cooperative transmission
47 used to continuously monitor and optimize network
UL [110] – K-means between terrestrial networks and
48 performance and detect and repair faults automatically. This
NTNs.
49 includes using AI to predict and prevent network failures,
Handover of user equipment
50 optimize routing and traffic flow, and manage network
UL [111] – K-means between satellites and ground
51 congestion.
stations.
52 AI for user experience and behavior analysis: AI can be
[104] - DQN Smart resource management for
53 RL used to analyze user behavior and preferences and predict
[105] - MADRL reducing multi-cell interference.
54 and personalize services and content. This includes using AI
[107] - DQN Computation offloading via
55 to optimize user engagement, enhance content delivery, and
RL [109] - Soft Actor coverage deployment or multi-
56 improve overall user experience. The key research directions
Critic access edge computing (MEC).
57 for AI-Powered NTNs are summarized in Fig. 7.
58 14
59 For Review Only
60
Page 37 of 42 IEEE Access

1
2
3 human factors involved in network management, including
4 user behavior, network administrator skills, and
5 organizational structures. Organizations must ensure that AI-
6 powered networks are designed to meet the needs of the users
7 and the organization. This requires understanding the skills
8 and knowledge of the network administrators and users and
9 the organizational structure and culture.
10 Resource limitations: AI-powered networks may require
11 significant computational resources, which can be a
challenge in resource-constrained environments.
12
Organizations must ensure that AI-powered networks are
13
designed to operate within the resource constraints of the
14
environment. Additionally, organizations must consider AI-
15 powered networks' power consumption and environmental
16 impact.
17 Lack of standardization: There is a lack of standardization
18 in AI-powered network technologies and protocols, which
19 can hinder interoperability and scalability. Organizations
20 must work together to develop standards for AI-powered
21 networks that promote interoperability and scalability.
22 Bias and fairness: AI-powered networks must address bias
23 and fairness issues in decision-making, particularly in areas
24 Figure 7. Key Research Directions
such as resource allocation and network management.
25 Organizations must ensure that AI-powered networks are
26 designed to be fair and unbiased and that decisions made by
B. CHALLENGES AND OPEN ISSUES IN DEPLOYING
27 AI-POWERED NTNs the network are transparent and explainable.
28 While AI-powered NTNs have significant potential, there Ethical concerns: AI-powered networks must consider
29 are also several challenges and open issues that must be ethical concerns related to the use of AI, including
30 addressed. These include but are not limited: transparency, accountability, and potential unintended
31 Data privacy and security: AI techniques require large consequences. Organizations must ensure that AI-powered
32 amounts of data to train models, but ensuring data privacy networks are designed to be transparent and accountable and
33 and security is a significant concern. Organizations must that they do not have unintended negative consequences.
34 ensure that sensitive data is protected and not accessible to Regulatory and legal frameworks: AI-powered networks
35 unauthorized personnel. Additionally, organizations must be must comply with regulatory and legal frameworks related
36 transparent about the data they are collecting, how it will be to privacy, security, and other issues. Organizations must
37 used, and who has access to it. ensure that AI-powered networks are designed to meet
38 Trustworthiness: AI-powered networks must be regulatory and legal requirements and that they are
39 trustworthy and reliable, with clear accountability transparent and accountable to regulators and other
40 mechanisms in place to ensure the network functions as stakeholders.
41 intended. Organizations must ensure that AI-powered Lack of domain expertise: Developing AI-powered
42 networks are transparent and explainable so that users can networks requires domain expertise in networking and AI,
43 understand how the network is making decisions. which can be challenging to find in a single individual or
44 Additionally, organizations must be transparent about the organization. Organizations must ensure that they have
45 limitations of AI-powered networks so that users understand access to the necessary expertise to develop and deploy AI-
46 when the network may not be appropriate for specific use powered networks and that they are able to collaborate with
47 cases. other organizations to share expertise and resources. We
48 Integration and interoperability: AI-powered networks have summarized these challenges and open issues in Fig. 8
49 must be able to integrate and interoperate with existing for better and easier understanding.
50 networks and devices, including legacy systems.
51 Organizations must ensure that AI-powered networks are C. OPPORTUNITIES FOR FUTURE RESEARCH
52 designed to work with existing infrastructure and can operate Despite the challenges and open issues, AI-powered NTNs
53 within existing networks. Additionally, organizations must for 6G networks present numerous future research
54 ensure that AI-powered networks can communicate with opportunities. In the following, we list such opportunities:
55 other devices and networks using standard protocols. Development of new AI techniques: AI-powered NTNs for
56 Human factors: AI-powered networks must consider the 6G networks require the development of new AI techniques
57
58 15
59 For Review Only
60
 IEEE Access Page 38 of 42

1
2
3 For example, AI-powered network management can
4 optimize network performance and resource allocation for
5 virtual reality applications or enable intelligent traffic
6 management for connected vehicles.
7 Standardization: This is an important area for future
8 research in AI-powered NTNs for 6G networks.
9 Standardization involves the development of consistent and
10 uniform technology and protocols across the industry. This
11 is important to ensure the interoperability and scalability of
12 AI-powered network technologies. In other words,
13 standardization ensures that different AI-powered network
14 management systems can work together seamlessly and that
15 they can be easily scaled up or down depending on the
16 network size and complexity. To achieve standardization,
17 researchers can work on developing new standards and
18 protocols that can facilitate the adoption and deployment of
19 AI-powered NTNs for 6G networks. This can involve
20 creating standardized communication protocols, data
21 formats, and interfaces that can be used across different
22 network management systems. Researchers can also work
23 with industry stakeholders and regulatory bodies to create
24 Figure 8. Summary of Challenges and Open Issues industry-wide standards and guidelines that can promote the
25 development and adoption of AI-powered NTNs for 6G
that can address the unique challenges of 6G networks. These networks. By achieving standardization, AI-powered NTNs
26
challenges include the need for real-time processing, large- for 6G networks can become more efficient, effective, and
27
scale data analysis, and dynamic network management. reliable, ultimately leading to a better user experience and
28
Research in this area can lead to the development of new improved network performance.
29
algorithms and models that can provide better performance, Sustainability: AI-powered NTNs for 6G networks must
30
accuracy, and efficiency in network management. also consider the sustainability of network management,
31
Integration with emerging technologies: The integration including the use of renewable energy sources, the reduction
32
of AI-powered network management with emerging of energy consumption, and the optimization of network
33
technologies such as edge computing, blockchain, and the resources. Researchers can work on developing new
34
IoT can create intelligent and efficient networks. Research in techniques and protocols that can ensure the sustainability of
35
this area can lead to the development of new architectures AI-powered network management.
36
and protocols that can leverage the benefits of these Explainability and transparency: AI-powered network
37
emerging technologies. For example, edge computing can management should be explainable and transparent to ensure
38
reduce latency and improve network response time, while trust and accountability. Researchers can work on
39
blockchain can provide a secure and decentralized network developing new techniques and models that can provide
40
management. insights into how AI-powered network management makes
41
Adoption of new network architecture: AI-powered NTNs decisions and provide explanations for these decisions.
42
for 6G networks requires adopting new network Resilience: AI-powered NTNs for 6G networks must be
43
architectures optimized for AI-powered network resilient to cyber-attacks and other security threats.
44
management. Research in this area can lead to the Researchers can work on developing new techniques and
45
development of new architectures and protocols that can protocols that can ensure the resilience of AI-powered
46
provide better scalability, reliability, and performance in network management and protect against these threats.
47
network management. For example, software-defined Collaboration and partnerships: AI-powered NTNs for
48
networks (SDNs) and NFV can provide greater flexibility 6G networks require collaboration and partnerships between
49
and programmability in network management, which AI- academia, industry, and government to address the complex
50
powered techniques can leverage. challenges involved in network management. Research in
51
Exploration of new applications: AI-powered NTNs for this area can lead to the development of new collaborative
52
6G networks can enable a wide range of new applications, models and frameworks that can facilitate sharing of
53
such as immersive virtual reality, connected vehicles, and knowledge, resources, and expertise.
54
smart cities. Research in this area can lead to the Sustainability: AI-powered NTNs for 6G networks must
55
development of new use cases and applications that can also consider the sustainability of network management,
56
leverage the benefits of AI-powered network management. including the use of renewable energy sources, the reduction
57
58 16
59 For Review Only
60
Page 39 of 42 IEEE Access

1
2
of energy consumption, and the optimization of network [7] S. Kota and G. Giambene, “6G Integrated Non-Terrestrial Networks:
3 Emerging Technologies and Challenges,” 2021 IEEE Int. Conf.
4 resources. Research in this area can lead to the development Commun. Work. ICC Work. 2021 - Proc., 2021, doi:
5 of new techniques and protocols that can ensure the 10.1109/ICCWorkshops50388.2021.9473743.
sustainability of AI-powered network management. [8] M. Giordani and M. Zorzi, “Non-Terrestrial Networks in the 6G Era:
6 Challenges and Opportunities,” IEEE Netw., vol. 35, no. 2, pp. 244–
7 In summary, AI-powered NTNs for 6G networks present
251, 2021, doi: 10.1109/MNET.011.2000493.
8 significant challenges and opportunities. Addressing the [9] R. Giuliano and E. Innocenti, “Machine Learning Techniques for
challenges and leveraging the opportunities will require Non-Terrestrial Networks,” Electron., vol. 12, no. 3, 2023, doi:
9 10.3390/electronics12030652.
10 significant research and development efforts, but the
[10] C. X. Wang, M. Di Renzo, S. Stańczak, S. Wang, and E. G. Larsson,
11 potential benefits of intelligent and efficient networks are “Artificial Intelligence Enabled Wireless Networking for 5G and
12 worthy of these efforts. Beyond: Recent Advances and Future Challenges,” IEEE Wirel.
Commun., vol. 27, no. 1, pp. 16–23, 2020, doi:
13 10.1109/MWC.001.1900292.
VII. CONCLUSIONS
14 [11] O. T. H. Alzubaidi et al., “Interference Challenges and Management
In this paper, we comprehensively surveyed AI-powered in B5G Network Design: A Comprehensive Review,” Electron., vol.
15
NTNs for 6G wireless communication networks. We first 11, no. 18, 2022, doi: 10.3390/electronics11182842.
16 [12] A. Vanelli-Coralli, A. Guidotti, T. Foggi, G. Colavolpe, and G.
discussed the background and motivation for using NTNs in
17 Montorsi, “5G and beyond 5G Non-Terrestrial Networks: Trends and
6G networks and their deployment and management
18 research challenges,” 2020 IEEE 3rd 5G World Forum, 5GWF 2020
challenges. We then highlighted the key features and - Conf. Proc., pp. 163–169, 2020, doi:
19 10.1109/5GWF49715.2020.9221119.
requirements of NTNs, including the role of
20 [13] M. Mozaffari, X. Lin, and S. Hayes, “Toward 6G with Connected
LEO/MEO/GEO satellites, HAPS, and UAVs in 6G
21 Sky: UAVs and Beyond,” IEEE Commun. Mag., vol. 59, no. 12, pp.
networks. Then, we discussed various AI techniques for 74–80, 2021, doi: 10.1109/MCOM.005.2100142.
22 M. Banafaa et al., “6G Mobile Communication Technology:
optimizing NTNs, including ML, DL, RL, and DRL, and [14]
23 Requirements, Targets, Applications, Challenges, Advantages, and
explored the recent advances in AI-powered NTNs for 6G
24 Opportunities,” Alexandria Eng. J., vol. 64, pp. 245–274, 2023, doi:
networks. We also identified key research directions,
25 10.1016/j.aej.2022.08.017.
challenges, and opportunities in AI-powered NTNs for 6G [15] B. Barakat et al., “6G opportunities arising from internet of things use
26 cases: A review paper,” Futur. Internet, vol. 13, no. 6, pp. 1–29, 2021,
networks.
27 doi: 10.3390/fi13060159.
Our survey sheds the light that AI-assisted techniques are the
28 [16] G. Geraci, D. Lopez-Perez, M. Benzaghta, and S. Chatzinotas,
key ingredients for realizing 6G since AI facilitates NTNs to “Integrating Terrestrial and Non-terrestrial Networks: 3D
29 Opportunities and Challenges,” IEEE Commun. Mag., pp. 1–7, 2022,
operate optimally in dynamic and unpredictable
30 doi: 10.1109/MCOM.002.2200366.
environments. However, the reliability of AI models depends
31 [17] M. Wang, T. Zhu, T. Zhang, J. Zhang, S. Yu, and W. Zhou, “Security
on the quantity and quality of training dataset. How to create and privacy in 6G networks: New areas and new challenges,” Digit.
32 Commun. Networks, vol. 6, no. 3, pp. 281–291, 2020, doi:
a synthetic training dataset which mimics real-world 6G
33 10.1016/j.dcan.2020.07.003.
observations will be crucial. In this context, utilizing digital
34 [18] H. Tataria, M. Shafi, A. F. Molisch, M. Dohler, H. Sjoland, and F.
twin to generate virtual replicas of real-world 6G Tufvesson, “6G Wireless Systems: Vision, Requirements,
35
communications data could be an interesting avenue for Challenges, Insights, and Opportunities,” Proc. IEEE, vol. 109, no. 7,
36 pp. 1166–1199, 2021, doi: 10.1109/JPROC.2021.3061701.
future research.
37 [19] S. Chen, Y. C. Liang, S. Sun, S. Kang, W. Cheng, and M. Peng,
38 “Vision, Requirements, and Technology Trend of 6G: How to Tackle
REFERENCES the Challenges of System Coverage, Capacity, User Data-Rate and
39 Movement Speed,” IEEE Wirel. Commun., vol. 27, no. 2, pp. 218–
[1] X. Lin, “An Overview of 5G Advanced Evolution in 3GPP Release
40 18,” IEEE Commun. Stand. Mag., vol. 6, no. 3, pp. 77–83, 2022, doi: 228, 2020, doi: 10.1109/MWC.001.1900333.
41 10.1109/MCOMSTD.0001.2200001. [20] N. Promwongsa et al., “A Comprehensive Survey of the Tactile
Internet: State-of-the-Art and Research Directions,” IEEE Commun.
42 [2] I. Rahman, S. M. Razavi, O. Liberg, C. Hoymann, H. Wiemann, C.
Surv. Tutorials, vol. 23, no. 1, pp. 472–523, 2021, doi:
Tidestav, P. Schliwa-Bertling, P. Persson, and D. Gerstenberger, “5G
43 evolution toward 5G advanced: An overview of 3GPP releases 17 and 10.1109/COMST.2020.3025995.
44 18,” Ericsson Technol. Rev, vol. 2021, no. 14, pp. 2–12, 2021. [21] L. U. Khan, I. Yaqoob, M. Imran, Z. Han, and C. S. Hong, “6G
Wireless Systems: A Vision, Architectural Elements, and Future
45 [3] M. Mahdi Azari, S. Solanki, S. Chatzinotas, O. Kodheli, H. Sallouha,
Directions,” IEEE Access, vol. 8, pp. 147029–147044, 2020, doi:
A. Colpaert, J. Fabian Mendoza Montoya, S. Pollin, A. Haqiqatnejad,
46 A. Mostaani, E. Lagunas, and B. Ottersten, “Evolution of Non- 10.1109/ACCESS.2020.3015289.
47 Terrestrial Networks from 5G to 6G: A Survey,” IEEE Commun. [22] Z. Zhang et al., “6G Wireless Networks: Vision, Requirements,
48 Surv. Tutorials, vol. 24, no. 4, pp. 2633–2672, 2022, doi: Architecture, and Key Technologies,” IEEE Veh. Technol. Mag., vol.
14, no. 3, pp. 28–41, 2019, doi: 10.1109/MVT.2019.2921208.
49 10.1109/COMST.2022.3199901.
[23] Y. W. Chen, R. Zhang, C. W. Hsu, and G. K. Chang, “Key Enabling
[4] M. B. Janjua, A. E. Duranay, and H. Arslan, “Role of Wireless
50 Communication in Healthcare System to Cater Disaster Situations Technologies for the Post-5G Era: Fully Adaptive, All-Spectra
51 Under 6G Vision,” Front. Commun. Networks, vol. 1, no. December, Coordinated Radio Access Network with Function Decoupling,”
IEEE Commun. Mag., vol. 58, no. 9, pp. 60–66, 2020, doi:
52 pp. 1–10, 2020, doi: 10.3389/frcmn.2020.610879.
10.1109/MCOM.001.2000186.
[5] M. Vaezi et al., “Cellular, Wide-Area, and Non-Terrestrial IoT: A
53 Survey on 5G Advances and the Road Toward 6G,” IEEE Commun. [24] H. H. H. Mahmoud, A. A. Amer, and T. Ismail, “6G: A
54 Surv. Tutorials, vol. 24, no. 2, pp. 1117–1174, 2022, doi: comprehensive survey on technologies, applications, challenges, and
research problems,” Trans. Emerg. Telecommun. Technol., vol. 32,
55 10.1109/COMST.2022.3151028.
no. 4, pp. 1–14, 2021, doi: 10.1002/ett.4233.
[6] G. Araniti, A. Iera, S. Pizzi, and F. Rinaldi, “Toward 6G Non-
56 Terrestrial Networks,” IEEE Netw., vol. 36, no. 1, pp. 113–120, 2022, [25] S. Basharat, S. A. Hassan, H. Pervaiz, A. Mahmood, Z. Ding, and M.
57 doi: 10.1109/MNET.011.2100191. Gidlund, “Reconfigurable Intelligent Surfaces: Potentials,
58 17
59 For Review Only
60
 IEEE Access Page 40 of 42

1
2
Applications, and Challenges for 6G Wireless Networks,” IEEE Earth Orbit Satellite Access,” IEEE Commun. Mag., vol. 59, no. 12,
3 Wirel. Commun., vol. 28, no. 6, pp. 184–191, 2021, doi: pp. 36–42, 2021, doi: 10.1109/MCOM.001.2100298.
4 10.1109/MWC.011.2100016. [44] D. Renga and M. Meo, “Can High Altitude Platform Stations Make
5 [26] B. Ji, Y. Han, S. Liu, F. Tao, G. Zhang, Z. Fu, and C. Li, “Several 6G Sustainable?,” IEEE Commun. Mag., vol. 60, no. 9, pp. 75–80,
Key Technologies for 6G: Challenges and Opportunities,” IEEE 2022, doi: 10.1109/MCOM.002.2101048.
6 Commun. Stand. Mag., vol. 5, no. 2, pp. 44–51, 2021, doi: [45] C. Palanci, F. Akleman, and G. K. Kurt, “High Altitude Platform
7 10.1109/MCOMSTD.001.2000038. Station (HAPS) to Satellite Channel Models for 6G Networks,” IEEE
8 [27] S. Mumtaz, V. G. Menon, and M. I. Ashraf, “Guest Editorial: Ultra- Aerosp. Conf. Proc., vol. 2022-March, 2022, doi:
Low-Latency and Reliable Communications for 6G Networks,” IEEE 10.1109/AERO53065.2022.9843327.
9 Commun. Stand. Mag., vol. 5, no. 2, pp. 10–11, 2021, doi: [46] B. Li, Z. Fei, and Y. Zhang, “UAV communications for 5G and
10 10.1109/MCOMSTD.2021.9464926. beyond: Recent advances and future trends,” IEEE Internet Things J.,
11 [28] M. Adhikari and A. Hazra, “6G-Enabled Ultra-Reliable Low-Latency vol. 6, no. 2, pp. 2241–2263, 2019, doi: 10.1109/JIOT.2018.2887086.
12 Communication in Edge Networks,” IEEE Commun. Stand. Mag., [47] A. Sharma et al., “Communication and networking technologies for
vol. 6, no. March, pp. 67–74, 2022. UAVs: A survey,” J. Netw. Comput. Appl., vol. 168, p. 102739, 2020,
13 [29] M. Alsabah, M. A. Naser, B. M. Mahmmod, S. H. Abdulhussain, and doi: 10.1016/j.jnca.2020.102739.
14 M. R. Eissa, “6G Wireless Communications Networks : A [48] R. Shrestha, R. Bajracharya, and S. Kim, “6G Enabled Unmanned
15 Comprehensive Survey,” IEEE Access, vol. 9, pp. 148191–148243, Aerial Vehicle Traffic Management: A Perspective,” IEEE Access,
2021, doi: 10.1109/ACCESS.2021.3124812. vol. 9, pp. 91119–91136, 2021, doi:
16 [30] W. B. H. M. A. H. H. D. Jiang, “The Road Towards 6G : A 10.1109/ACCESS.2021.3092039.
17 Comprehensive Survey,” IEEE Open J. Commun. Soc., vol. 2, pp. [49] X. JIANG, M. SHENG, N. ZHAO, C. XING, W. LU, and X. WANG,
18 334–366, 2021, doi: 10.1109/OJCOMS.2021.3057679. “Green UAV communications for 6G: A survey,” Chinese J.
[31] Y. Yuan, Y. Zhao, B. Zong, and S. Parolari, “Potential key Aeronaut., vol. 35, no. 9, pp. 19–34, 2022, doi:
19 technologies for 6G mobile communications,” Sci. China Inf. Sci., 10.1016/j.cja.2021.04.025.
20 vol. 63, no. 8, pp. 1–19, 2020, doi: 10.1007/s11432-019-2789-y. [50] X. Fang, W. Feng, T. Wei, Y. Chen, N. Ge, and C. X. Wang, “5G
21 [32] W. Saad, M. Bennis, and M. Chen, “A Vision of 6G Wireless Embraces Satellites for 6G Ubiquitous IoT: Basic Models for
Systems: Applications, Trends, Technologies, and Open Research Integrated Satellite Terrestrial Networks,” IEEE Internet Things J.,
22 Problems,” IEEE Netw., vol. 34, no. 3, pp. 134–142, 2020, doi: vol. 8, no. 18, pp. 14399–14417, 2021, doi:
23 10.1109/MNET.001.1900287. 10.1109/JIOT.2021.3068596.
24 [33] M. M. Kamruzzaman, “Key Technologies, Applications and Trends [51] G. K. Kurt et al., “A Vision and Framework for the High Altitude
25 of Internet of Things for Energy-Efficient 6G Wireless Platform Station (HAPS) Networks of the Future,” IEEE Commun.
Communication in Smart Cities,” Energies, vol. 15, no. 15, p. 5608, Surv. Tutorials, vol. 23, no. 2, pp. 729–779, 2021, doi:
26 2022. 10.1109/COMST.2021.3066905.
27 [34] S. S. Hassan, D. H. Kim, Y. K. Tun, N. H. Tran, W. Saad, and C. S. [52] S. A. Al-Ahmed, M. Z. Shakir, and S. A. R. Zaidi, “Optimal 3D UAV
28 Hong, “Seamless and Energy Efficient Maritime Coverage in base station placement by considering autonomous coverage hole
Coordinated 6G Space-Air-Sea Non-Terrestrial Networks,” IEEE detection, wireless backhaul and user demand,” J. Commun.
29 Internet Things J., vol. 10, no. 6, pp. 4749–4769, 2022, doi: Networks, vol. 22, no. 6, pp. 467–475, 2020, doi:
30 10.1109/JIOT.2022.3220631. 10.23919/JCN.2020.000034.
31 [35] P. Porambage, G. Gur, D. P. Moya Osorio, M. Livanage, and M. [53] I. Natgunanathan, A. Mehmood, Y. Xiang, G. Beliakov, and J.
Ylianttila, “6G security challenges and potential solutions,” 2021 Jt. Yearwood, “Protection of Privacy in Biometric Data,” IEEE Access,
32 Eur. Conf. Networks Commun. 6G Summit, EuCNC/6G Summit 2021, vol. 4, pp. 880–892, 2016, doi: 10.1109/ACCESS.2016.2535120.
33 pp. 622–627, 2021, doi: [54] N. H. Mahmood et al., “Machine type communications: key drivers
34 10.1109/EuCNC/6GSummit51104.2021.9482609. and enablers towards the 6G era,” Eurasip J. Wirel. Commun. Netw.,
35 [36] V. L. Nguyen, P. C. Lin, B. C. Cheng, R. H. Hwang, and Y. D. Lin, vol. 2021, no. 1, 2021, doi: 10.1186/s13638-021-02010-5.
“Security and Privacy for 6G: A Survey on Prospective Technologies [55] M. A. Lopez, G. N. N. Barbosa, and D. M. F. Mattos, “New Barriers
36 and Challenges,” IEEE Commun. Surv. Tutorials, vol. 23, no. 4, pp. on 6G Networking: An Exploratory Study on the Security, Privacy,
37 2384–2428, 2021, doi: 10.1109/COMST.2021.3108618. and Opportunities for Aerial Networks,” 2022 1st Int. Conf. 6G
38 [37] L. U. Khan, I. Yaqoob, N. H. Tran, Z. Han, and C. S. Hong, “Network Networking, 6GNet 2022, no. Pdpa, pp. 1–6, 2022, doi:
Slicing: Recent Advances, Taxonomy, Requirements, and Open 10.1109/6GNet54646.2022.9830402.
39 Research Challenges,” IEEE Access, vol. 8, pp. 36009–36028, 2020, [56] G. A. Akpakwu, B. J. Silva, G. P. Hancke, and A. M. Abu-Mahfouz,
40 doi: 10.1109/ACCESS.2020.2975072. “A Survey on 5G Networks for the Internet of Things:
41 [38] L. LI-HSIANG SHEN; KAI-TEN FENG; Hanzo, “Five Facets of Communication Technologies and Challenges,” IEEE Access, vol. 6,
6G : Research Challenges and Opportunities,” ACM Comput. Surv., pp. 3619–3647, 2017, doi: 10.1109/ACCESS.2017.2779844.
42 vol. 55, no. 11, pp. 1–39, 2023, doi: 10.1145/3571072. [57] N. Panwar and S. Sharma, “Security and Privacy Aspects in 5G
43 [39] S. A. Abdel Hakeem, H. H. Hussein, and H. W. Kim, “Vision and Networks,” 2020 IEEE 19th Int. Symp. Netw. Comput. Appl. NCA
44 research directions of 6G technologies and applications,” J. King 2020, 2020, doi: 10.1109/NCA51143.2020.9306740.
Saud Univ. - Comput. Inf. Sci., vol. 34, no. 6, pp. 2419–2442, 2022, [58] R. Banu and T. Vladimirova, “Fault-tolerant encryption for space
45 doi: 10.1016/j.jksuci.2022.03.019. applications,” IEEE Trans. Aerosp. Electron. Syst., vol. 45, no. 1, pp.
46 [40] B. Tezergil and E. Onur, “Wireless Backhaul in 5G and Beyond: 266–279, 2009, doi: 10.1109/TAES.2009.4805278.
47 Issues, Challenges and Opportunities,” IEEE Commun. Surv. [59] M. Ylianttila, R. Kantola, A. Gurtov, L. Mucchi, I. Oppermann, Z.
48 Tutorials, vol. 24, no. 4, pp. 2579–2632, 2022, doi: Yan, T. H. Nguyen, F. Liu, T. Hewa, and M. Liyanage, ‘‘6G white
10.1109/COMST.2022.3203578. paper: Research challenges for trust, security and privacy,’’ 2020,
49 [41] W. Xu, Z. Yang, D. W. K. Ng, M. Levorato, Y. C. Eldar, and M. arXiv:2004.11665.
50 Debbah, “Edge Learning for B5G Networks With Distributed Signal [60] I. Ahmad, J. Suomalainen, P. Porambage, A. Gurtov, J. Huusko, and
51 Processing: Semantic Communication, Edge Computing, and M. Hoyhtya, “Security of Satellite-Terrestrial Communications:
Wireless Sensing,” IEEE J. Sel. Top. Signal Process., vol. 17, no. 1, Challenges and Potential Solutions,” IEEE Access, vol. 10, no.
52 pp. 9–39, 2023, doi: 10.1109/JSTSP.2023.3239189. August, pp. 96038–96052, 2022, doi:
53 [42] E. Juan, M. Lauridsen, J. Wigard, and P. E. Mogensen, “5G New 10.1109/ACCESS.2022.3205426.
54 Radio Mobility Performance in LEO-based Non-Terrestrial [61] M. Ericson et al., “6G Architectural Trends and Enablers,” Proc. -
Networks,” 2020 IEEE Globecom Work. GC Wkshps 2020 - Proc., 2021 IEEE 4th 5G World Forum, 5GWF 2021, pp. 406–411, 2021,
55 pp. 1–6, 2020, doi: 10.1109/GCWkshps50303.2020.9367419. doi: 10.1109/5GWF52925.2021.00078.
56 [43] X. Lin, S. Cioni, G. Charbit, N. Chuberre, S. Hellsten, and J. F. [62] T. Patra, “Link Budget Analysis for 5G Communication in the
57 Boutillon, “On the Path to 6G: Embracing the Next Wave of Low Tropical Regions,” Wirel. Commun. Mob. Comput., vol. 2020, pp. 1–
58 18
59 For Review Only
60
Page 41 of 42 IEEE Access

1
2
9, 2020. UNIVERSITI TUNKU ABDUL RAHMAN, no. June. 2022.
3 [63] M. Ikram, K. Sultan, M. F. Lateef, and A. S. M. Alqadami, “A Road [81] A. Iqbal, M. L. Tham, and Y. C. Chang, “Convolutional neural
4 towards 6G Communication — A Review of 5G Antennas , Arrays , network-based deep Q-network (CNN-DQN) resource management
5 and Wearable Devices,” electronics, vol. 11, no. 169, pp. 1–29, 2022. in cloud radio access network,” China Commun., vol. 19, no. 10, pp.
[64] Z. Wang, S. Liu, and Y. Dong, “Compact Wideband Pattern 129–142, 2022, doi: 10.23919/JCC.2022.00.025.
6 Reconfigurable Antennas Inspired by End-Fire Structure for 5G [82] A. Iqbal, M. L. Tham, and Y. C. Chang, “Resource allocation for joint
7 Vehicular Communication,” IEEE Trans. Veh. Technol., vol. 71, no. energy and spectral efficiency in cloud radio access network based on
8 5, pp. 4655–4664, 2022. deep reinforcement learning,” Trans. Emerg. Telecommun. Technol.,
[65] Y. U. Jianjun and Y. Wei, “Digital Signal Processing for High-Speed vol. 33, no. 4, pp. 1–16, 2022, doi: 10.1002/ett.4417.
9 THz Communications,” Chinese J. Electron., vol. 31, no. 3, pp. 534– [83] R. A. Nazib and S. Moh, “Reinforcement Learning-Based Routing
10 546, 2022, doi: 10.1049/cje.2021.00.258. Protocols for Vehicular Ad Hoc Networks: A Comparative Survey,”
11 [66] B. Mao, F. Tang, Y. Kawamoto, and N. Kato, “AI Models for Green IEEE Access, vol. 9, pp. 27552–27587, 2021, doi:
12 Communications Towards 6G,” IEEE Commun. Surv. Tutorials, vol. 10.1109/ACCESS.2021.3058388.
24, no. 1, pp. 210–247, 2022, doi: 10.1109/COMST.2021.3130901. [84] R. F. Atallah, C. M. Assi, and J. Y. Yu, “A Reinforcement Learning
13 [67] V. Verdhan, “Introduction to Supervised Learning,” Univ. Technique for Optimizing Downlink Scheduling in an Energy-
14 Massachusetts, vol. 3, no. 1, pp. 1–46, 2014, doi: 10.1007/978-1- Limited Vehicular Network,” IEEE Trans. Veh. Technol., vol. 66, no.
15 4842-6156-9_1. 6, pp. 4592–4601, 2017, doi: 10.1109/TVT.2016.2622180.
[68] H. U. Dike, Y. Zhou, K. K. Deveerasetty, and Q. Wu, “Unsupervised [85] Y. Liu et al., “Reconfigurable Intelligent Surfaces: Principles and
16 Learning Based On Artificial Neural Network: A Review,” 2018 Opportunities,” IEEE Commun. Surv. Tutorials, vol. 23, no. 3, pp.
17 IEEE Int. Conf. Cyborg Bionic Syst. CBS 2018, pp. 322–327, 2019, 1546–1577, 2021, doi: 10.1109/COMST.2021.3077737.
18 doi: 10.1109/CBS.2018.8612259. [86] T. Kürsat, K. Kurt, R. Ali, and Y. Halim, “Reconfigurable Intelligent
[69] H. Zhang and T. Yu, “Taxonomy of Reinforcement Learning Surfaces in Action for Non-Terrestrial Networks,” IEEE Veh.
19 Algorithms,” Deep Reinforcement Learning, Springer, 2020, pp. Technol. Mag., vol. 17, no. 3, pp. 45–53, 2022.
20 125–33. [87] K. W. Park, H. M. Kim, and O. S. Shin, “A Survey on Intelligent-
21 [70] D. Dampahalage, K. B. S. Manosha, N. Rajatheva, and M. Latva- Reflecting-Surface-Assisted UAV Communications,” Energies, vol.
Aho, “Supervised Learning Based Sparse Channel Estimation for Ris 15, no. 14, pp. 1–20, 2022, doi: 10.3390/en15145143.
22 Aided Communications,” ICASSP, IEEE Int. Conf. Acoust. Speech [88] C. C. Gonzalez, S. Pizzi, M. Murroni, and G. Araniti, “Multicasting
23 Signal Process. - Proc., vol. 2022-May, pp. 8827–8831, 2022, doi: Over 6G Non-Terrestrial Networks: A Softwarization-Based
24 10.1109/ICASSP43922.2022.9746793. Approach,” IEEE Veh. Technol. Mag., vol. 18, no. 1, pp. 91–99, 2023,
25 [71] A. E. Perez, A. Estrada Moscoso, N. G. Gonzalez, and J. J. G. Torres, doi: 10.1109/MVT.2022.3232919.
“Training Approaches in Supervised Learning for ICI Mitigation in [89] S. K. Das et al., “Comprehensive review on ML-based RIS-enhanced
26 Gridless Nyquist-WDM,” 2020 IEEE Photonics Conf. IPC 2020 - IoT systems: basics, research progress and future challenges,”
27 Proc., no. 53, pp. 2–3, 2020, doi: 10.1109/IPC47351.2020.9252366. Comput. Networks, vol. 224, p. 109581, 2023, doi:
28 [72] J. Jang, J. H. Park, and H. Jong Yang, “Supervised-Learning-Based 10.1016/j.comnet.2023.109581.
Resource Allocation in Wireless Networks,” Int. Conf. ICT Converg., [90] X. Wang, X. Li, and V. C. M. Leung, “Artificial intelligence-based
29 vol. 2020-Octob, pp. 1022–1024, 2020, doi: techniques for emerging heterogeneous network: State of the arts,
30 10.1109/ICTC49870.2020.9289481. opportunities, and challenges,” IEEE Access, vol. 3, pp. 1379–1391,
31 [73] Y. Zhang, J. Wang, Y. Chen, H. Yu, and T. Qin, “Adaptive Memory 2015, doi: 10.1109/ACCESS.2015.2467174.
Networks with Self-supervised Learning for Unsupervised Anomaly [91] B. Siniarski et al., “Need of 6G for the Metaverse Realization,”
32 Detection,” IEEE Trans. Knowl. Data Eng., vol. 4347, no. 12, pp. 1– arXiv:2301.03386, vol. 7, pp. 1–31, 2022, [Online]. Available:
33 13, 2022, doi: 10.1109/TKDE.2021.3139916. http://arxiv.org/abs/2301.03386
34 [74] W. Zhang, L. Jiao, F. Liu, S. Yang, W. Song, and J. Liu, “Sparse [92] P. K. Thiruvasagam et al., “Open RAN: Evolution of Architecture,
35 Feature Clustering Network for Unsupervised SAR Image Change Deployment Aspects, and Future Directions,” arXiv:2301.06713, pp.
Detection,” IEEE Trans. Geosci. Remote Sens., vol. 60, no. 5226713, 1–28, 2023.
36 pp. 1–13, 2022, doi: 10.1109/TGRS.2022.3167745. [93] Y. O. Imam-Fulani et al., “5G Frequency Standardization,
37 [75] M. U. Khan, M. Azizi, A. Garcia-Armada, and J. J. Escudero-Garzas, Technologies, Channel Models, and Network Deployment:
38 “Unsupervised Clustering for 5G Network Planning Assisted by Real Advances, Challenges, and Future Directions,” Sustainability, vol.
Data,” IEEE Access, vol. 10, no. 2, pp. 39269–39281, 2022, doi: 15, no. 6, p. 5173, 2023, doi: 10.3390/su15065173.
39 10.1109/ACCESS.2022.3165799. [94] S. Coveney and K. Roberts, “Lightweight UAV digital elevation
40 [76] M. Nair, T. Cappello, S. Dang, V. Kalokidou, and M. A. Beach, “RF models and orthoimagery for environmental applications: data
41 Fingerprinting of LoRa Transmitters Using Machine Learning with accuracy evaluation and potential for river flood risk modelling,” Int.
Self-Organizing Maps for Cyber Intrusion Detection,” IEEE MTT-S J. Remote Sens., vol. 38, no. 8–10, pp. 3159–3180, 2017, doi:
42 Int. Microw. Symp. Dig., vol. 2022-June, no. 1, pp. 491–494, 2022, 10.1080/01431161.2017.1292074.
43 doi: 10.1109/IMS37962.2022.9865441. [95] A. Traspadini, M. Giordani, and M. Zorzi, “UAV/HAP-Assisted
44 [77] M. L. Tham, A. Iqbal, and Y. C. Chang, “Deep reinforcement Vehicular Edge Computing in 6G: Where and What to Offload?,”
learning for resource allocation in 5G communications,” 2019 Asia- 2022 Jt. Eur. Conf. Networks Commun. 6G Summit, EuCNC/6G
45 Pacific Signal Inf. Process. Assoc. Annu. Summit Conf. APSIPA ASC Summit 2022, pp. 178–183, 2022, doi:
46 2019, no. November, pp. 1852–1855, 2019, doi: 10.1109/EuCNC/6GSummit54941.2022.9815734.
47 10.1109/APSIPAASC47483.2019.9023112. [96] Z. Ma, Y. Wang, J. Li, and Y. Liu, “A Blockchain-Based Privacy-
48 [78] S. Tripathi, S. Member, O. J. Pandey, S. Member, and R. M. Hegde, Preserving Incentive Mechanism for Internet of Vehicles in Satellite-
“An Optimal Reflective Elements Grouping Model for RIS-Assisted Terrestrial Crowdsensing,” 2021 7th Int. Conf. Comput. Commun.
49 IoT Networks Using Q-Learning,” IEEE Trans. Circuits Syst. II ICCC 2021, pp. 2062–2067, 2021, doi:
50 Express Briefs, pp. 1–5, 2023, doi: 10.1109/TCSII.2023.3251373. 10.1109/ICCC54389.2021.9674460.
51 [79] A. Iqbal, M. L. Tham, and Y. C. Chang, “Double Deep Q-Network- [97] H. Zhou, Z. Wang, N. Cheng, D. Zeng, and P. Fan, “Stackelberg-
Based Energy-Efficient Resource Allocation in Cloud Radio Access Game-Based Computation Offloading Method in Cloud-Edge
52 Network,” IEEE Access, vol. 9, pp. 20440–20449, 2021, doi: Computing Networks,” IEEE Internet Things J., vol. 9, no. 17, pp.
53 10.1109/ACCESS.2021.3054909. 16510–16520, 2022, doi: 10.1109/JIOT.2022.3153089.
54 [80] A. Iqbal, "DEVELOPMENT OF DEEP REINFORCEMENT [98] H. Zhou, Z. Wang, G. Min, and H. Zhang, “UAV-Aided Computation
LEARNING BASED RESOURCE ALLOCATION TECHNIQUES Offloading in Mobile-Edge Computing Networks: A Stackelberg
55 IN CLOUD RADIO ACCESS NETWORK" AMJAD IQBAL Game Approach,” IEEE Internet Things J., vol. 10, no. 8, pp. 6622–
56 DOCTOR OF PHILOSOPHY ( ENGINEERING ) LEE KONG 6633, 2023, doi: 10.1109/JIOT.2022.3197155.
57 CHIAN FACULTY OF ENGINEERING AND SCIENCE [99] H. Zhou, Z. Wang, H. Zheng, S. He, and M. Dong, “Cost
58 19
59 For Review Only
60
 IEEE Access Page 42 of 42

1
2
Minimization-Oriented Computation Offloading and Service AMJAD IQBAL received the B.Sc. degree in
3 Caching in Mobile Cloud-Edge Computing: An A3C-Based Electrical Engineering from CECOS
4 Approach,” IEEE Trans. Netw. Sci. Eng., vol. 10, no. 3, pp. 1326– University, Peshawar, Pakistan, in 2014, and
5 1338, 2023, doi: 10.1109/TNSE.2023.3255544. the M.Sc. degree in Communication and
[100] A. Heidari, N. J. Navimipour, and M. Unal, “Applications of ML/DL Electronic engineering from the University of
6 in the management of smart cities and societies based on new trends Engineering and Technology (UET), Peshawar,
7 in information technologies: A systematic literature review,” Sustain. in 2017, and the Ph.D. degree in Wireless
8 Cities Soc., vol. 85, no. February, p. 104089, 2022, doi: Communication from University Tunku Abdul
10.1016/j.scs.2022.104089. Rahman (UTAR), Malaysia, in 2022. He is
9 [101] P. H. Masur, J. H. Reed, N. K. Tripathi, and V. Tech, “Regular currently a Postdoc research fellow in the
10 Article : Artificial Intelligence in Open-Radio Access Network,” Department of Systems and Computer
11 IEEE Aerosp. Electron. Syst. Mag., vol. 37, no. 9, pp. 6–15, 2022. Engineering at Carleton University, Canada.
12 [102] P. L. S. Jayalaxmi, R. Saha, G. Kumar, M. Conti, and T. H. Kim, His research interests include resource allocation optimization, machine
“Machine and Deep Learning Solutions for Intrusion Detection and learning, deep reinforcement learning, 5G/B5G cellular networks, and
13 Prevention in IoTs: A Survey,” IEEE Access, vol. 10, no. October, Reflecting Intelligent Surface (RIS).
14 pp. 121173–121192, 2022, doi: 10.1109/ACCESS.2022.3220622.
15 [103] X. Shen et al., “AI-Assisted Network-Slicing Based Next-Generation
Wireless Networks,” IEEE Open J. Veh. Technol., vol. 1, pp. 45–66,
16 2020.
17 [104] M. K. Tefera, “Deep Reinforcement Learning-Assisted Optimization
18 for Resource Allocation in Downlink OFDMA Cooperative
Systems,” Entropy, vol. 25, no. 3, p. 413, 2023.
19 [105] T. Hu, Q. Liao, Q. Liu, and G. Carle, “Network Slicing via Transfer
20 Learning aided Distributed Deep Reinforcement Learning,” in IEEE
21 Global Communications Conference: Next-Generation Networking THAM MAU LUEN received the Bachelor of
and Internet, 2023, pp. 2909–2914. doi:
22 10.1109/globecom48099.2022.10000763.
Engineering and Doctor of Philosophy
degrees in telecommunication engineering
23 [106] A. Mukherjee, P. Goswami, M. A. Khan, L. Manman, L. Yang, and from University of Malaya. He is currently an
24 P. Pillai, “Energy-efficient resource allocation strategy in massive iot Assistant Professor with Universiti Tunku
25 for industrial 6G applications,” IEEE Internet Things J., vol. 8, no. 7, Abdul Rahman. He has been a principal
pp. 5194–5201, 2021, doi: 10.1109/JIOT.2020.3035608.
26 [107] C. Zhang, M. Dong, and K. Ota, “Fine-Grained Management in 5G:
investigator (PI) and a co-investigator of
more than 20 research and development
27 DQL Based Intelligent Resource Allocation for Network Function projects. This includes five international
28 Virtualization in C-RAN,” IEEE Trans. Cogn. Commun. Netw., vol. grants, two of which were simultaneously led
6, no. 2, pp. 428–435, 2020, doi: 10.1109/TCCN.2020.2982886.
29 [108] Y. Fu and X. Wang, “Traffic Prediction-Enabled Energy-Efficient
by him as the PI/Co-PI under the support of
ASEAN IVO and the British Council. He has
30 Dynamic Computing Resource Allocation in CRAN Based on Deep published two IEEE Transactions papers as a principal author. His research
31 Learning,” IEEE Open J. Commun. Soc., vol. 3, pp. 159–175, 2022, interests include the IoT, machine learning/deep learning/deep
doi: 10.1109/OJCOMS.2022.3146886.
32 [109] C. Sun, X. Wu, X. Li, Q. Fan, J. Wen, and V. C. M. Leung,
reinforcement learning, and beyond-5G communications.
33 “Cooperative Computation Offloading for Multi-Access Edge
34 Computing in 6G Mobile Networks via Soft Actor Critic,” IEEE
35 Trans. Netw. Sci. Eng., pp. 1–14, 2021, doi:
10.1109/TNSE.2021.3076795.
36 [110] M. Karavolos, L. Tsipi, P. S. Bithas, D. Vouyioukas and P. T.
37 Mathiopoulos, "Satellite Aerial Terrestrial Hybrid NOMA Scheme in
38 6G Networks: An Unsupervised Learning Approach," 2022 1st
International Conference on 6G Networking (6GNet), Paris, France,
39 2022, pp. 1-5, doi: 10.1109/6GNet54646.2022.9830368.
40 [111] M. K. Dahouda, S. Jin, and I. Joe, "Machine Learning-Based YI JIE WONG received the B.Eng. degree
41 Solutions for Handover Decisions in Non-Terrestrial Networks, " (Hons.) in biomedical engineering from the
Electronics, vol. 12, no. 8, 2023, doi:10.3390/electronics12081759
42 Universiti Tunku Abdul Rahman, Sungai
Long, Malaysia in 2022, where he is currently
43 pursuing the Ph.D. degree in digital
44 technology with specialization of
45 reinforcement learning-based federated
learning. His research interests include the
46 Internet of Things (IoT), machine learning,
47 federated learning, and deep reinforcement learning.
48
49
50
51
52
53
54
55
56
57
58 20
59 For Review Only
60
Page 43 of 42 IEEE Access

1
2
ALA’A AL-HABASHNA received his TASOS DAGIUKLAS received the
3 Master of Engineering degree from Memorial engineering degree from the University of
4 University of Newfoundland in 2010 and his Patras, Greece, in 1989, the M.Sc. degree from
5 Ph.D. in Electrical and Computer Engineering the University of Manchester, U.K., in 1991, and
from Carleton University in 2018. Currently, the Ph.D. degree from the University of Essex,
6 Dr. Al-Habashna is an Adjunct Research U.K., in 1995, all in electrical engineering. He is
7 Professor at Carleton University, Ottawa, a Leading Researcher and Expert in the field of
8 Canada. Dr. Al-Habashna is also a Senior smart internet technologies. He is the Leader of
Researcher at Statistics Canada, Ottawa, the Smart Internet Technologies (SuITE)
9 Canada. He has won multiple awards, including excellence and best-paper Research Group, London South Bank University, where he also acts as the
10 awards. He worked as a reviewer for many conferences and journals and as Head of the Cognitive Systems Research Centre. He has been a principal
11 a Technical Program Committee Member of multiple conferences. He has investigator, a co-investigator, the project and technical manager, a
12 also been the co-chair of the Communications and Networking Simulation coordinator, and a focal person of more than 25 internationally research and
(CNS) track of the Annual Modeling and Simulation Conference. His development and capacity training projects in the areas of fixed-mobile
13 current research interests include 5G wireless networks, machine learning, convergence, 4G/5G networking technologies, VoIP, and multimedia
14 computer vision, IoT applications, localization, multimedia communication networking. His research interests lie in the field of systems beyond 5G/6G
15 over wireless networks, signal detection and classification, cognitive radio networking technologies, programmable networks, UAVs, V2X
systems, and discrete-event modeling and simulation. communications, and cyber security for smart internet systems.
16
17
18
19
20
21
22
23 GABRIEL WAINER received the Ph.D. degree
24 (Highest Hons.) from UBA/Université d'Aix-
25 Marseille III, Marseille, France, in 1998. He held
visiting positions with the University of Arizona;
26 LSIS (CNRS), Université Paul Cézanne,
27 University of Nice, INRIA Sophia-Antipolis,
28 Université de Bordeaux (France); UCM, UPC
(Spain), University of Buenos Aires, National
29 University of Rosario (Argentina) and others. He
30 is currently a full professor with Carleton
31 University, Ottawa, ON, Canada, where he is also the head with the
Advanced Real-Time Simulation Laboratory, Centre for advanced
32 Simulation and Visualization (V-Sim). He is Editor in Chief of
33 SIMULATION (Sage) and a member of the editorial board of the IEEE
34 Computing in Science & Engineering, Wireless Networks (Elsevier), and
35 The Journal of Defense Modeling and Simulation (SCS). He is a fellow of
the Society for Modeling and Simulation International (SCS).
36
37
38
39
40
41
42
43
YONGXU ZHU received the Ph.D. degree in
44
electrical engineering from University College
45 London in 2017. From 2017 to 2019, she was a
46 Research Associate with Loughborough
47 University. She is currently a Senior Lecturer with
the Division of Computer Science and Informatics,
48 London South Bank University. Her research
49 interests include B5G/6G, heterogeneous
50 networks, UAV communications, and physical-layer security. She also
serves as an Editor for IEEE Wireless Communications Letters and IEEE
51 Transactions on Wireless Communications.
52
53
54
55
56
57
58 21
59 For Review Only
60