Base Paper

2494 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 25, NO.
4, FOURTH QUARTER 2023
A Survey of Blockchain and Artificial Intelligence

for 6G Wireless Communications
Yiping Zuo , Member, IEEE, Jiajia Guo , Member, IEEE, Ning Gao , Member, IEEE,
Yongxu Zhu , Senior Member, IEEE, Shi Jin , Senior Member, IEEE, and Xiao Li , Member, IEEE
Abstract—The research on the sixth-generation (6G) wire- Index Terms—Blockchain, AI, wireless communications, 6G
less communications for the development of future mobile networks, secure services, IoT smart applications, spectrum
communication networks has been officially launched around management, security and privacy, smart healthcare, UAVs.
the world. 6G networks face multifarious challenges, such as
resource-constrained mobile devices, difficult wireless resource
management, high complexity of heterogeneous network archi-
tectures, explosive computing and storage requirements, privacy I. I NTRODUCTION
and security threats. To address these challenges, deploying ROM 2020, the fifth-generation (5G) wireless networks
blockchain and artificial intelligence (AI) in 6G networks may
realize new breakthroughs in advancing network performances
F achieve large-scale commercial deployment worldwide.
Academia, industry, and governments are now engaged in
in terms of security, privacy, efficiency, cost, and more. In this
paper, we provide a detailed survey of existing works on the research and development of the sixth-generation (6G) wire-
application of blockchain and AI to 6G wireless communications. less communication technology to meet the demands of
More specifically, we start with a brief overview of blockchain future networks in 2030 and beyond [1]. Compared with 5G
and AI. Then, we mainly review the recent advances in the networks, 6G networks will have ultra-high network speed,
fusion of blockchain and AI, and highlight the inevitable trend of
deploying both blockchain and AI in wireless communications. ultra-low communication delay, and wider coverage depth.
Furthermore, we extensively explore integrating blockchain and 6G networks will fully share ultra-high frequency wireless
AI for wireless communication systems, involving secure services spectrum resources such as millimeter waves, terahertz, and
and Internet of Things (IoT) smart applications. Particularly, light waves. 6G networks will also integrate technologies such
some of the most talked-about key services based on blockchain as terrestrial mobile communications, satellite Internet, and
and AI are introduced, such as spectrum management, com-
putation allocation, content caching, and security and privacy. microwave networks to form an integrated green network with
Moreover, we also focus on some important IoT smart applica- group collaboration of all things, intelligent data perception,
tions supported by blockchain and AI, covering smart healthcare, real-time security assessment, and coordinated coverage of
smart transportation, smart grid, and unmanned aerial vehicles space and earth [2], [3], [4]. Facing the 6G era, the network
(UAVs). Moreover, we thoroughly discuss operating frequencies, will usher in new application scenarios and new performance
visions, and requirements from the 6G perspective. We also
analyze the open issues and research challenges for the joint requirements. In the 6G era, an air-space-ground integrated
deployment of blockchain and AI in 6G wireless communica- network communication system will be built to realize a
tions. Lastly, based on lots of existing meaningful works, this ubiquitous network for full coverage and all scenarios [5].
paper aims to provide a comprehensive survey of blockchain and However, diverse applications and communication scenarios,
AI in 6G networks. We hope this survey can shed new light on ultra-heterogeneous network connections, and service require-
the research of this newly emerging area and serve as a roadmap
for future studies. ments for extreme performance all place higher requirements
on the bandwidth, latency, security, connection density, and
flexibility of 6G networks [6], [7], [8].
Manuscript received 24 October 2022; revised 16 March 2023 and 3 August In the 6G era, artificial intelligence (AI) [9] is becom-
2023; accepted 2 September 2023. Date of publication 14 September 2023;
date of current version 22 November 2023. This work was supported in part ing more and more important. AI relies on mining big data
by the National Natural Science Foundation of China (NSFC) under Grant for training and learning, continuously enhancing computing
62261160576, Grant 62301279, Grant 62371131, and Grant 62001109; in power to cope with higher transmission rates, and gaining
part by the Natural Science Foundation of the Higher Education Institutions
of Jiangsu Province under Grant 23KJB510020; and in part by the Royal more flexibility through continuous learning. In the future, 6G
Societys International Exchanges Scheme under Grant IEC\NSFC\211011. networks need to deal with explosive data traffic growth and
(Corresponding author: Shi Jin.) massive device connections. Real-time management and con-
Yiping Zuo is with the College of Computer, Nanjing University of Posts
and Telecommunications, Nanjing 210023, China, and also with the National trol of these massive data will result in high complexity and
Mobile Communications Research Laboratory, Southeast University, Nanjing latency overhead. Therefore, how to effectively perceive ser-
210096, China (e-mail: zuoyiping@njupt.edu.cn). vice characteristics, accurately monitor and control network
Jiajia Guo, Ning Gao, Shi Jin, and Xiao Li are with the National
Mobile Communications Research Laboratory, Southeast University, resources, and dynamically allocate wireless resources has
Nanjing 210096, China (e-mail: jiajiaguo@seu.edu.cn; ninggao@seu.edu.cn; become an important issue for 6G networks. The use of AI at
jinshi@seu.edu.cn; li_xiao@seu.edu.cn). the application layer and network layer of 6G networks makes
Yongxu Zhu is with the School of Engineering, University of Warwick,
CV4 7AL Coventry, U.K. (e-mail: yongxu.zhu@warwick.ac.uk). the network more intelligent and automated, which will be a
Digital Object Identifier 10.1109/COMST.2023.3315374 necessary way to manage and control massive wireless big
1553-877X
c 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Visvesvaraya Technological University Belagavi. Downloaded on March 22,2024 at 04:43:30 UTC from IEEE Xplore. Restrictions apply.
ZUO et al.: SURVEY OF BLOCKCHAIN AND AI FOR 6G WIRELESS COMMUNICATIONS 2495
data [10], [11]. In addition, 6G network supports large-scale AI. The IMT-2030 (6G) Promotion Group puts forward the
users, large-scale antennas, and multi-band hybrid transmis- 6G vision of “Intelligent Internet of Everything, Digital Twin”,
sion, so traditional physical layer transmission technologies pointing out that 6G will enable the efficient and intelligent
will face multiple challenges in performance, complexity, and interconnection of all things.
efficiency. This provides the possibility for AI technology to
be applied to the wireless physical layer [12], [13]. Notably,
a 6G white paper [14] released by the University of Oulu A. Previous Survey Works and Motivations
believed that AI will play an important role in 6G networks. AI model or algorithm is based on the trained intelli-
The report of [15] also indicated that it is necessary to intro- gence data. Meanwhile, blockchain is essentially a data storage
duce AI technology into complex network architectures in the method, or “hyper ledger”, which embodies data intelli-
future. gence [52]. Consequently, these two technologies, which are
Blockchain is another highly anticipated emerging technol- both closely related to data, can be effectively combined to
ogy. In fact, blockchain is a technical system that integrates complement each other and achieve technological improve-
various technologies such as chain data structure, point-to- ment [37], [38], [39]. As a trusted platform, blockchain can
point transmission, distributed storage, consensus mechanism, improve the authenticity, relevance, and validity of the data
and encryption algorithm [16]. The performance index require- used by AI. From the perspective of data, computing power,
ments of 6G networks, such as ultra-high peak rate, ultra-low and algorithms, blockchain improves the level of AI tech-
latency, ultra-high reliability, ultra-low energy consumption, nology, innovates AI collaboration models and computing
and seamless connection, make system security, data privacy, paradigms, and constructs a new AI ecosystem. With intel-
sustainability, scalability, and other aspects subject to many ligent and automatic characteristics, AI can promote the
risks and challenges [17], [18]. Blockchain technology is an natural evolution and data sorting of blockchain through the
important technical means to cope with these challenges, espe- optimization and simulation of AI algorithms. Additionally,
cially with the advantages of distributed network architecture, AI can effectively prevent the occurrence of blockchain node
intelligent node consensus, and smart contracts. The integrated forks, can handle the operation of the blockchain more effec-
application of blockchain and 6G networks provide a safe, tively, and improve efficiency intelligently. Most importantly,
intelligent, and efficient underlying technical support for the the close combination of blockchain and AI can promote
realization of the 6G network vision [19], [20]. In particular, and optimize various services and applications and also can
6G white paper [14] pointed out that 6G network requires provide a reliable, secure, and ultra-low latency intelligent
an endogenous trust network, and blockchain technology may network environment for next-generation wireless communi-
play an important role in the 6G networks to deal with a cations. Accordingly, in the future 6G networks, the research
variety of complex new privacy challenges. Blockchain is a on the simultaneous deployment of blockchain and AI is of
potential solution for privacy protection of 6G networks [21]. positive significance.
Moreover, blockchain can provide a strong guarantee for 6G Next, we briefly describe the existing survey on the adop-
networks to build a distributed, secure, and trusted transaction tion of blockchain and AI in wireless communication systems.
environment. Researchers integrated blockchain with wireless communica-
Research institutions and operators worldwide are accelerat- tions to form secure and trusted mobile networks and services.
ing the development of the cross-integration between emerging In [22], [23], [24], [25], [26], [27], a large number of reviews
technologies such as blockchain and AI with 6G networks. on blockchain-supported wireless communications have been
The IMT-2030 (6G) Promotion Group’s white paper pro- published, broadly elaborating the basic concepts, network
poses various scenarios for the application of blockchain architecture, enabling technologies, research challenges, and
technology in 6G networks, including dynamic spectrum man- future research directions. Moreover, the mutual integration
agement, ubiquitous access management, edge computing, and of blockchain and AI has also been investigated in detail
so on. China has established several international standard by multiple studies [33], [34], [35], [36], [37], [38], [39],
projects of blockchain in ITU, such as the establishment of [40], [41]. Most importantly, the disruptive integration of
“Framework of blockchain of things as decentralized service blockchain and AI for wireless communications can greatly
platform” in ITU-T SG20 and the “Reference framework for improve the network performance for a variety of services
distributed ledger technologies” in ITU-T SG16. Sprint, an and applications. Many pieces of literature [42], [43], [44],
American operator, has partnered with NXM Labs to launch [45], [46], [47], [48], [49], [50], [51] have summarized and
a 5G connected vehicle platform powered by blockchain reviewed the topic of joint blockchain and AI for wireless
technology. China Mobile and Huobi China have created a communications. However, to the best of our knowledge, none
“blockchain + Internet of Things (IoT)” identity authenti- of the existing surveys have comprehensively investigated this
cation platform. Meanwhile, 3GPP specifically defines the popular topic, especially few research emphasized the simul-
network data analysis function, aiming to provide a standard taneous deployment of blockchain and AI for next-generation
interface for the development and application of AI mod- wireless communications. For example, the research of [46]
els in wireless networks. The European Telecommunications only briefly discussed the potential of the joint application of
Standards Institute has also established an industry standard blockchain and machine learning (ML) in wireless communi-
working group to use AI for network management, expecting cation systems. Similarly, the work in [47] briefly reviewed
to achieve a high-level autonomous network with endogenous reinforcement learning (RL)-empowered blockchains applied
2496 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 25, NO. 4, FOURTH QUARTER 2023
TABLE I
C OMPARISON OF O UR W ORK W ITH E XISTING R ELATED R ESEARCH
in Industrial IoT (IIoT) networks. The authors of [51] simply services, such as spectrum management, computation
investigated the benefits of adopting blockchain with ML allocation, content caching, and security and privacy.
under the secure in-vehicle network. TABLE I displays a 4) We review the latest achievements of joint blockchain
straightforward comparison of our work with existing related and AI in 6G IoT smart applications. We extensively dis-
surveys. cuss some important 6G IoT smart applications, including
smart healthcare, smart transportation, smart grid, and
unmanned aerial vehicle (UAV).
B. Novelty and Contributions
5) On the basis of the comprehensive survey, we thoroughly
Compared with the existing aforementioned works, our sur- discuss operating frequencies, visions, and requirements
vey provides a comprehensive analysis and outlook on the from the 6G perspective. We also propose some open
current research progress of blockchain and AI for 6G wireless issues and research challenges that need to be resolved
communications. We hope that this survey has some reference for the applications of blockchain and AI to 6G wireless
significance for carrying out more innovative research in this communications, and summarize several future research
promising field. The main contributions of this article can be directions.
summarized as follows:
1) We briefly outline the basic knowledge of blockchain and
AI. First, the concept, characteristics, and categories of
blockchain and AI are introduced. Then, we separately C. Outline of the Survey
discuss the classic applications of blockchain and AI for The outline of this article is presented in Fig. 1. The
wireless communication systems. remainder of this survey is organized as follows. Section II-A
2) We systematically summarize the integration of provides an overview of blockchain, including the concept,
blockchain and AI from two directions: blockchain- characteristics, categories, and representative applications in
assisted AI and AI-assisted blockchain. Furthermore, we wireless communications. In Section II-B, we describe an
also emphasize the advantages of integrating blockchain overview of AI, taking into account the concept, character-
and AI for wireless communication systems. istics, categories, and typical applications in wireless commu-
3) We deeply elaborate on the latest developments of com- nications. In Section II-C, we discuss the mutual fusion of
bining blockchain and AI in 6G secure services. We blockchain and AI, and emphasize the abundant benefits of
specifically focus on some of the most popular 6G secure this fusion for wireless communication systems. Section III
Fig. 1. Outline of the paper.
presents the integration of blockchain and AI for wireless com- based on smart contracts means the arrival of the blockchain
munications, covering secure services and IoT smart applica- 2.0 era [54]. Blockchain 3.0 emphasizes its application to
tions. Some open issues and research challenges are discussed all aspects of society. Blockchain is essentially a distributed
in Section IV, and the future work is also addressed. Finally, super-accounting ledger [55]. This digital ledger guarantees
we conclude the main works of the survey in Section V. data security through encryption algorithms and consensus
The major acronyms used in this paper are summarized in mechanisms. Over time, the past transaction records on the
TABLE II. blockchain ledger will not be deleted and cannot be tam-
pered with. The blockchain network consists of multiple peer
nodes, and these nodes do not need to trust each other. Each
II. BACKGROUND OF B LOCKCHAIN AND AI
node independently maintains a copy of the global ledger. The
A. An Overview of Blockchain transaction data in the ledger is encapsulated by blocks. The
1) Concept of Blockchain: The concept of blockchain new block will be added to the end of the previous block in
was first mentioned in the Bitcoin white paper written by the form of a linked list, so this accounting ledger is called
Nakamoto [53], marking the birth of blockchain 1.0. Ethereum “blockchain” [56]. Taking bitcoin as an example, the typical
TABLE II
L IST OF M AJOR ACRONYMS of blockchain system used. These algorithms have their own
advantages but also have their own shortcomings, as presented
in TABLE III.
Since blockchain systems run in the Peer-to-Peer (P2P)
network where nodes do not trust each other, the initiated
transaction needs to be completed under the witness of all
nodes in the network. The transaction execution process of the
blockchain is as follows. Specifically, the node first randomly
generates its own private key and public key and constructs a
transaction through a wallet or script tool, and then uses its
own private key to sign the transaction. The signed transac-
tion is propagated between neighbor nodes through the P2P
network. Then, the node receiving the transaction verifies the
legality of the transaction, and the miner digs out a new block
according to the consensus algorithm. Next, the miners broad-
cast the new block to other nodes through the P2P network.
Other miners verify the legitimacy of the new block to decide
to discard or add to the local chain. After confirming the new
block through the nodes of the whole network, this indicates
that the new transaction is transferred successfully.
2) Characteristics of Blockchain: The development of
blockchain technology has formed a relatively complete tech-
nology stack. Blockchain has been widely concerned and stud-
ied because of its important characteristics: decentralization,
non-tampering, traceability, and anonymity [23].
Decentralization: Blockchain technology is to complete data
interaction without relying on any third-party intermediaries
or institutions. Compared with the centralized network, the
bottom layer of blockchains adopts the P2P network archi-
tecture. In the blockchain network, there is no traditional
central server to process data recording, storage, and updat-
ing. Every node is equal, and the data maintenance of the
entire blockchain network is jointly participated by all nodes.
In addition, the withdrawal of any node will not affect the
operation of the entire system, and the blockchain network
block structure is shown in Fig. 2. Each block is divided into has strong robustness.
two parts: block header and block body. The block body stores Non-Tampering: Once the transaction data is packaged on
the verified transaction data. The block header contains the the chain by miner nodes and permanently stored in the
version, hash value of the previous block, hash value of the blockchain to form an immutable historical ledger. By stor-
current block, timestamp, difficulty value, nonce, and Merkle ing the hash value of the previous block in each block, the
root. blocks are connected back and forth to form a chain structure.
In the blockchain system, the consensus algorithm ensures This special chained data structure enables all blocks storing
that each node can maintain the same transaction con- transaction data to be added to the end of the blockchain in
tent and sequence, which is the core component of the chronological order. The malicious node wants to tamper with
blockchain network [57]. Currently, the widely used and com- the data, which inevitably causes the hash value of the cur-
mon consensus algorithms are as follows: Proof of Work rent block and all subsequent blocks to change, leading to the
(PoW) [53], Proof of Stake (PoS) [58], Delegated Proof of collapse of the chain structure. Therefore, the cost of data tam-
Stake (DPoS) [59], and Practical Byzantine Fault Tolerance pering becomes extremely high, making it almost impossible
(PBFT) [60]. PoW introduces distributed node computing to modify the blockchain.
power competition to maintain data consistency and consen- Traceability: In blockchain networks, all transactions are
sus security. The core idea of PoS is that the node with the public and any node can get a record of all transactions. Except
highest stake obtains the accounting right of the block. DPoS for the encrypted private information of both parties to the
elects representatives through shareholder voting to get the transaction, all data on the blockchain can be queried through
right to keep accounts. PBFT sorts the request through the public interfaces. Blockchain uses the chain block structure
leader node, the follower node responds to the request, and the with a timestamp to store data, resulting in adding a time
response result of most nodes is the final result. In addition, dimension to data. Each transaction on the block is connected
there is no perfect consensus protocol, because the adopted to two adjacent blocks through cryptographic methods, which
consensus protocol needs to be matched according to the type guarantees that users can trace the source of any transaction.
Fig. 2. The structure of blocks.
TABLE III
C OMPARISON OF F OUR T YPICAL C ONSENSUS A LGORITHMS
Anonymity: Since the nodes in the blockchain network do blockchain is between the public blockchain and the private
not need to trust each other, there is no need to disclose the blockchain and has the characteristics of multi-center or par-
identity between the nodes. This ensures the anonymity of tial decentralization. Only members belonging to the alliance
each participating node in the blockchain system and protects can generate transactions or view blockchain information.
the privacy of the nodes. Nodes can conduct transactions with- According to the way of trust construction in different sce-
out knowing the identity of the other party. Both nodes of the narios, the blockchain can also be divided into a permissionless
transaction only need to publish their own addresses to com- blockchain and a permissioned blockchain [24]. The permis-
municate with each other. In the blockchain network, nodes sionless blockchain is also called the public blockchain, which
use asymmetric encryption technology to build trust between is a completely open blockchain. That is, anyone can join the
nodes in an anonymous environment. network and participate in the complete consensus account-
3) Categories of Blockchain: According to differ- ing process. The permissioned blockchain is a semi-open
ent application scenarios, blockchains are classified into blockchain. Only designated members can join the network,
public blockchain, consortium blockchain, and private and each member has different rights to participate. The per-
blockchain [46]. mission chain often establishes a trust relationship by issuing
Public Blockchain: The public blockchain is a completely identity certificates. This blockchain has partial decentraliza-
decentralized blockchain [61], and any user can join the tion characteristics, which is more efficient than permissionless
blockchain network. There is no official organization, man- blockchains. Private blockchains and consortium blockchains
agement agency, and no central server. Participating nodes can belong to permission chains. The comparison of the character-
freely enter and exit the network without being controlled. istics of the above-mentioned different blockchains is shown
Private Blockchain: The private blockchain is fully central- in TABLE IV.
ized [23], and only authorized and trusted nodes can join the 4) Blockchain for Wireless Communications: Blockchain
blockchain network. All nodes in the network are controlled by technology naturally has many advantages such as decen-
an organization. The operating rules and consensus mechanism tralization, traceability, distribution, and tamper resistance.
of the system are determined by the organization itself. Therefore, researchers expect to apply the blockchain to all
Consortium Blockchain: The consortium blockchain is a levels of the wireless communication system, which will
partially decentralized blockchain [62] that is jointly main- achieve a substantial increase in system performance and a
tained by multiple companies or organizations. This type of true sense of the connection of everything [24]. Blockchain
TABLE IV
C OMPARISON OF D IFFERENT B LOCKCHAINS
can provide traceable communication for 6G networks, which The computing and storage capabilities of edge computing
not only facilitates network administrators to query historical are valuable network resources, which can be efficiently
resource conditions at any time, but also reduces malicious managed through the blockchain. To solve the problem of
users’ behavior of fabricating resource usage. In addition, the low efficiency of computing resource transactions, in the
blockchain uses multi-party consensus mechanisms to record blockchain-based edge-assisted IoT network, the work in [81]
the interactions between users, so as to ensure the fairness and considered using the credit-based payment for fast computing
openness of all interactions. The integration of blockchain and resource transactions. The work in [82] proposed a blockchain-
6G will provide a strong security guarantee for the construc- based multi-layer computing offloading architecture, which
tion of a safe and credible communication ecosystem. So far, enhances the collaboration between users in sharing com-
the research on the integration of blockchain and 6G network puting resources. In the blockchain-empowered multi-task
mainly involves two main aspects [25]: blockchain-enabled 6G cross-server edge computing scenario, the authors of [83]
secure services [63], [64], [65], [66], [67], [68], [69], [70], proposed two double auction mechanisms to drive end users
[71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], and edge servers to securely allocate and trade resources. The
[82], [83], [84], [85], [86], [87], [88], [89], [90], [91] and works of [84] and [85] applied the consortium blockchain
blockchain-assisted 6G Internet of Things (IoT) smart applica- and smart contracts to the vehicle edge computing network
tions [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], for resource trading, data storage, and data sharing, and to
[102], [103]. defend against malicious behaviors of vehicles. Blockchain
Secure Services: The research on the blockchain-enabled 6G was used to construct an attribute-based encryption scheme
secure services mainly involves spectrum sharing [71], [72], for secure storage and sharing of electronic medical records
[73], [74], [75], [76], [77], [78], [79], [80], computing and stor- in [86]. The authors of [87] designed a blockchain-enabled
age [81], [82], [83], [84], [85], [86], [87], [88], interference arbitrable remote data auditing scheme to provide reliable
management [89], [90], [91], and so on. Take blockchain network storage services. To deal with the privacy issues in
based spectrum sharing as an example, the authors of [71] content-centric mobile networks, the study in [88] proposed
and [72] proposed a blockchain-based verification and access a secure and efficient blockchain-inspired encrypted cloud
control protocol to complete the spectrum sharing between pri- storage solution.
mary and secondary users. The works in [73], [74] proposed a The dense deployment of 6G networks will cause serious
blockchain-based spectrum sensing as a service solution. Here, interference problems, so the use of blockchain for interference
the smart contract is mainly responsible for the following func- management is also a very important topic. The work in [89]
tions: 1) Scheduling the spectrum allocation between users and described a greedy distributed algorithm, using the blockchain
the helper to maximize system revenue; 2) Identifying whether currency mechanism and coordination protocol. This algo-
the helper is a malicious node, and ensuring the security of rithm can realize the optimal information distribution achieved
spectrum sharing. Reference [75] discussed the application of by the traditional central controller before, and eliminate the
blockchain in different spectrum access scenarios and ana- interference between users. The authors of [90] analyzed
lyzed the advantages and disadvantages of different spectrum the interference problem between transaction nodes in the
sharing mechanisms. Based on the consortium blockchain, a blockchain-based IoT network, and derived the probability
secure spectrum trading and sharing scheme for drone-assisted density function of the signal to interference plus noise ratio
communication systems was contrived in [76]. To solve the between IoT nodes and full nodes. The blockchain-based full
issue of privacy risk in spectrum sharing, [77] proposed a node deployment solution can ensure a high transaction suc-
trusted framework found on blockchain entitled TrustSAS for cess rate and overall communication throughput, and protect
dynamic spectrum access. The work in [78] introduced a smart the IoT network from security threats. In the blockchain-based
network architecture, which uses smart contracts to handle femtocell network, to avoid excessive interference from fem-
unlicensed spectrum sharing between operators and users. The tocell users to the macro base station (MBS), the MBS set
authors of [79] proposed a blockchain-based radio service a price for the interference generated by the femtocell user
model, which can reduce the management cost of dynamic in [91]. Femtocell users determined their transmission power
access systems. For wireless downlink communication systems and payment fees according to the modeled Stackelberg game.
with multiple mobile operators, the work in [80] delineated a Blockchain enabled the femtocell network to reliably make
blockchain-based dynamic spectrum acquisition scheme. payments without the involvement of intermediaries.
IoT Smart Applications: Blockchain has also been still a debated issue in academia, and there is no unanimously
introduced into many IoT smart application systems, accepted statement. Prof. N. J. Nisson of Stanford University’s
such as smart healthcare [92], [93], [94], smart trans- AI Research Center believes that “AI is the science of knowl-
portation [95], [96], [97], smart grid [98], [99], [100], edge, that is, how to express knowledge, how to acquire knowl-
UAV [101], [102], [103], and so on. For example, the authors edge, and how to use knowledge” [105]. Prof. P. H. Winston
of [92] described a blockchain-energized patient-centric of the Massachusetts Institute of Technology holds that “AI
electronic medical record management architecture, and is the study of how to make computers do intelligent jobs
completed the prototype implementation of this architecture that only humans could do in the past” [106]. From the per-
on the Hyperledger platform. The work in [93] proposed spective of knowledge engineering, Prof. E. A. Feigenbaum
a mobile edge computing (MEC)- and blockchain-based of Stanford University considers that “AI is a knowledge
distributed healthcare architecture for medical data offloading information processing system” [107]. In a word, AI is a
and data sharing. In the hospital network, the traditional comprehensive discipline, which integrates many disciplines
centralized patient identity authentication method may cause such as computer science, logic, biology, psychology, philos-
problems such as long time and high cost. To resolve these ophy, etc. AI has achieved remarkable results in applications
problems, the authors of [94] designed a distributed patient such as speech recognition, image processing, natural lan-
authentication method using blockchain. A blockchain- guage processing, automatic theorem proving, and intelligent
enabled electricity trading scheme between vehicles was robots [108], [109].
proposed in [95]. To alleviate the problem of incomplete In the early days of AI, problems that were very difficult
information sharing, the Bayesian game was also used to for human intelligence but relatively simple for computers
price electricity. The study of [96] delineated a blockchain were dealt with quickly. For example, some problems can be
storage system, which supports incremental data updates of described by a series of formal mathematical rules. The real
vehicles. This system used multiple technologies such as data challenge for AI lies in solving tasks that are easy for humans
partitioning, smart contracts, and redundant backups. To deal to perform but difficult to describe formally, such as recog-
with the security threats of the Internet of Vehicles, [97] nizing words spoken by people or faces in images. For these
considered a blockchain-assisted certificateless key agreement problems, human beings can often solve them easily and intu-
protocol, which has high security and low communication itively. In recent years, most of the major breakthroughs in
and computing costs. the field of AI can be summarized as the development and
The integration of blockchain into smart grid and UAV application of ML technology. The relationship between AI
is also a hot topic. In the smart grid system based on and ML is depicted in Fig. 3. ML provides a solution for
edge computing, to realize the private and secure commu- these relatively intuitive problems. This solution has its own
nication between grid terminals and edge servers, the work ability to acquire knowledge, that is, the ability to extract pat-
of [98] introduced a blockchain-enabled anonymous authenti- terns from raw data. Further, there is a key approach in ML,
cation and key agreement protocol. In the IoT-supported smart which can improve computer systems from experience and
grid system, the data of smart meters can be safely trans- data. This approach allows computers to learn from experience
mitted to service providers through the private blockchain- and understand the world in terms of a hierarchical concept
based access control protocol proposed in [99]. The work system, with each concept defined by its relationship to some
in [100] introduced a blockchain-empowered security demand relatively simple concepts. By allowing computers to acquire
response management scheme, which processes energy trans- knowledge from experience, human beings can avoid formally
action requests generated in the smart grid system. The authors specifying all the knowledge they need. Hierarchical concepts
of [101] proposed a blockchain-energized multi-party authen- let computers construct simpler concepts to learn complex con-
tication scheme, which can provide secure point-to-point cepts. Drawing a diagram representing how these concepts
wireless communication and reliable group communication build on top of each other, we obtain a ‘deep’ (many lay-
for UAV networks. Under the heterogeneous UAV flight ad ers) diagram. For this reason, we call this approach as deep
hoc network, the study of [102] drew up a blockchain- learning (DL) [109], [110]. ML can build AI systems running
based distributed key management scheme, which includes in complex real-world environments. DL is a specific type of
four modules: cluster key distribution, key updating, clus- ML with great power and flexibility. In DL, the big world can
ter UAV migration, and malicious UAV revocation. In order be described as a nested hierarchical concept system. This
to solve the security and privacy issues faced by energy hierarchical concept system refers to the definition of com-
micro-transactions, [103] introduced a distributed and secure plex concepts by the connection between simpler concepts,
UAV-assisted radio transmission architecture, which utilizes and the generalization from general abstraction to high-level
the directed acyclic graph and consortium blockchain. abstraction.
2) Characteristics of AI: In this subsection, we will discuss
some important characteristics of AI, including data driving,
B. An Overview of AI uncertainty, environmental perception, and scalability.
1) Concept of AI: Nowadays, AI has become a field with Data Driving: AI gradually completes the technology from
numerous practical applications and active research topics, and artificial knowledge expression to big data-driven knowledge
is booming [104]. It is difficult to give AI a scientific definition learning. AI rarely needs to rely on manual engineering, so
as rigorous as a mathematical one. Until now, “what is AI ?” is it can easily take advantage of the increment in the amount
Fig. 3. Classification of AI.
of available computation and data [111]. For example, a data- Scalability: Over time, the computer hardware and software
driven ML network regards the function to be implemented as infrastructure for AI technology have improved, and the scale
an unknown black box, replaces it with an ML network, and of AI models has grown accordingly. AI has been solving
then relies on a large amount of training data to complete the increasingly complex applications with increasing precision.
training from input to output. With the development of new learning algorithms and archi-
Uncertainty: There is a lot of uncertainty since AI has tectures developed for deep neural networks (DNNs), AI is
some similarities or differences compared with any other bound to have broader application prospects.
discipline such as mathematics, physics, cognitive, and behav- 3) Categories of AI: As depicted in Fig. 3, we first intro-
ioral psychology. Most areas of AI do not develop like duce the classification of AI. Then, we focus on several
traditional methods of mathematics, nor do they align with important branches of ML in the AI field. According to the
general models of physics. There will always be connec- classification of learning methods, ML can be divided into:
tions between AI and cognitive or behavioral psychology, but supervised learning, unsupervised learning, and RL. We also
those connections ignore the mathematical and engineering discuss some typical network architecture in DL.
themes. As a prescience, the framework of AI is not yet Supervised Learning: Supervised learning refers to training
complete. on labeled data to predict the type or value of new data [111].
Environmental Perception: The AI system should be able According to the different prediction results, supervised learn-
to generate the ability to perceive the external environment ing can be divided into two categories: classification and
with the help of sensors and other devices. AI can receive regression. In training, an objective function is usually given to
various information from the environment through hearing, measure the error (or distance) between the output and ground
vision, smell, and touch like humans, and generate necessary truth, and then its internal adjustable weights are modified to
reactions to external input such as text, voice, expressions, reduce the error via gradient descent. To improve the con-
and actions. These reactions even influence environmental or vergence speed and reduce the computational complexity, the
human decision-making. Ideally, an AI system should have stochastic gradient descent (SGD) [112] method is often used
certain adaptive characteristics and learning capabilities. That in practice. SGD randomly selects a sample to compute the
is, AI has a certain ability to adaptively adjust parameters or loss and gradient each time. Compared with more complex
update optimization models with changes in the environment, optimization techniques, this simple process of SGD often
data, or tasks. finds a good set of weights quickly. The common methods of
TABLE V
C OMPARISON OF T YPICAL ML A LGORITHMS
supervised learning are K-nearest neighbor (KNN), decision output layer, as shown in Fig. 3(a). The convolutional layer
tree, and logistic/linear regression. and the pooling layer are set alternately. In the convolutional
Unsupervised Learning: Unsupervised learning [113] is to layer, each neuron of the convolutional kernel is locally con-
do data mining without labels. One of the important func- nected to its input, and weighted and summed with the local
tions of unsupervised learning reflects in clustering, which is input through the corresponding connection weight. Then, the
simply to classify data according to different features with- bias value is added to get the output value of the neuron.
out labels. Typical methods of unsupervised learning include Because this process is equivalent to the convolution process,
K-means clustering and principal component analysis, etc. An it is called CNN. CNN is easier to train and popularize than
important premise of K-means clustering is that the difference the fully connected network between adjacent layers, and is
between data can be measured by Euclidean distance. If it widely adopted in the field of computer vision.
cannot be measured, it needs to be converted into a usable The recurrent neural network (RNN) [118] can process one
Euclidean distance measure. Principal component analysis is element of the input sequence at a time. As shown in Fig. 3(b),
a statistical method. By using orthogonal transformation, the the RNN maintains a “state vector” in its hidden units, which
variables with correlation are changed into variables with- implicitly contains historical information for all past elements
out correlation. The transformed variables are called principal in the sequence. The output depends not only on the current
components. The basic idea is to replace the initially cor- input, but also on the information available in past moments
related indicators with a set of independent comprehensive or information available in future moments. With such special
indicators. structure, RNNs are capable of providing memory for neural
Reinforcement Learning: RL [109], [114] is about obtain- networks.
ing rewards by interacting with the environment. Moreover, Additionally, the generative adversarial network
RL judges the quality of actions by the level of rewards and (GAN) [119] is also a typical DL network that aims to
then learns the optimal strategy. Agent perceives the state learn a model capable of generating fake samples on real-
information in the environment, searches for strategies, and distributed datasets. The basic structure of GAN is shown in
selects the optimal action. This causes a state change and a Fig. 3(c), which includes a generator G and a discriminator D.
return value to update the evaluation function. After complet- Both the generator and the discriminator can be implemented
ing a learning process, enter the next round of learning and by DL networks. The discriminator is used to distinguish
training. The learning process is repeated cyclically and iter- the fake samples generated by the generator from the real
atively, until the conditions of stop rule are met, and then the samples of the actual dataset. The task of the generator is
learning is terminated. For large-scale station-action pair, deep to generate sample data such that the discriminator cannot
reinforcement learning (DRL) [115] is an end-to-end percep- distinguish between real samples and fake samples. When
tion and control system with strong generality. The principle the generator produces samples that the discriminator cannot
framework of DRL is represented in Fig. 3(d). The DRL learn- distinguish from the real samples, training is balanced. The
ing process can be described as: (1) At each moment, the agent applications of GAN in basic fields such as image generation,
interacts with the environment to get a high-dimensional state, image translation, and speech images are very rich.
and uses DL to perceive the state to obtain a specific state fea- 4) AI for Wireless Communications: In this section, we
ture representation; (2) The agent evaluates the value function mainly describe the latest research progress of AI applied to
of each action based on the expected return, and maps the cur- 6G wireless communications. The combination of AI and 6G
rent state to the corresponding action through a certain policy; networks mainly contains in physical layer and upper layer.
(3) The environment reacts to this action and gets the next AI for Physical Layer: The main contents involve chan-
state. Through the continuous cycle of the above process, the nel estimation [120], [121], [122], [123], [124], signal
optimal policy to achieve the goal can be finally obtained. detection [125], [126], [127], [128], [129], channel state
We summarize the characteristics and typical structures of information (CSI) feedback and reconstruction [130], [131],
several algorithms of ML discussed above as shown in the [132], [133], [134], channel decoding [135], [136], [137],
following TABLE V. [138], [139], and end-to-end wireless communications [140],
Basic Network Architecture of DL: The basic network struc- [141], [142], [143], [144]. In massive multiple-input multiple-
ture is the convolutional neural network (CNN) [116], [117], output (MIMO) beam mmWave scenarios, channel estima-
which consists of an input layer, multiple convolutional lay- tion is extremely challenging, especially in scenarios where
ers, multiple pooling layers, a fully connected layer, and an antenna arrays are dense and receivers are equipped with
limited radio frequency (RF) links. The work of [120] pio- the complete CSI becomes extremely huge. The work of [130]
neered channel estimation by using the DL-based method presented a CNN-based CSI perception and recovery mecha-
in wireless energy transfer systems. In [120], the authors nism named CsiNet. Since then, DL-based CSI compression
developed an autoencoder-based channel estimation scheme, techniques have attracted a lot of attention [131], [132],
where the encoder is used to design pilots and the decoder [133], [134]. On the basis of [130], the authors of [131] pro-
is utilized to estimate the channel. The authors of [121] vided a real-time long short-term memory (LSTM)-based CSI
proposed a DL-based super-resolution channel estimation feedback architecture entitled CsiNet-LSTM, which employs
scheme in millimeter-wave massive MIMO systems. This temporal correlation to improve the feedback accuracy of time-
scheme utilizes DNN for beam direction-of-arrival estimation. varying channels. CsiNet-LSTM can accomplish a trade-off
In contrast, there are some other DL-based channel estima- between compression ratio, CSI reconstruction quality, and
tion schemes that combine traditional algorithms with certain complexity. On the basis of the high correlation of amplitudes
performance guarantees with DL algorithms. Reference [122] between bidirectional channels in the delay domain, DualNet
designed a learned denoising-based approximate message was proposed in [132] to use uplink amplitude information
passing (LDAMP) network. The LDAMP network takes the to help reconstruct downlink channel amplitudes. The CSI
channel matrix as a two-dimensional image as input, and inte- feedback and reconstruction algorithms in [130], [131], [132]
grates denoising CNN into the iterative signal reconstruction rely on a large amount of data for offline training, and the
algorithm for channel estimation. To improve the performance network complexity is high. The work of [133] focused on the
of sparse signal recovery, the authors of [123] proposed complexity of the neural network. The experimental results
a learned approximate message passing (LAMP) network. in [134] demonstrated that the DL-based channel feedback
LAMP directly expands the iterations of the AMP algo- framework can reduce the air time overhead by an average
rithm into the corresponding hierarchical network structure, of 73% and improve the throughput by about 69% compared
whose linear transformation coefficients and nonlinear shrink- with the 802.11 feedback protocol.
age parameters are jointly optimized by DNN. Furthermore, The authors in [135] developed a DNN-based channel
starting from the basic structure of the minimum mean square decoding method. This paper draws two conclusions about the
error (MMSE) algorithm, the work of [124] developed a DL- application of DL to channel decoding: 1) Structured codes
based channel estimator, in which the estimated channel vector such as polar codes are easier to learn than random codes;
consists of conditional Gaussian random variables with ran- 2) For structured codes, DL networks can decode untrained
dom covariance matrices. To reduce the complexity of channel codewords. However, this proposed method is neither suitable
estimation, an MMSE-based CNN network is proposed to for random codes nor codewords with long code lengths, and
compensate for the error in [124]. has great limitations. On the basis of the traditional polar code
The authors in [125] applied DNN to tackle the problem of iterative decoding algorithm, the work of [136] presented a
signal detection in orthogonal frequency division multiplexing DL polar code decoding network that separates sub-blocks.
(FDM) systems. Different from traditional wireless communi- The decoding algorithm in [136] is a highly parallel decoding
cation, [125] regarded channel estimation and signal detection algorithm. Compared with the decoding algorithm in [135], the
as a whole, and directly uses DNN to realize the mapping algorithm of [136] significantly reduced the number of train-
from the received signal to the original signal bits. The work ing times and the complexity of the network structure under
in [126] investigated the signal reconstruction problem of the the condition of comparable performance. Reference [137]
MIMO system, and proposed a signal detection algorithm conducted an iterative channel decoding algorithm: belief
entitled Detection Network (DetNet). DetNet is based on the propagation (BP)-CNN. The algorithm concatenates the CNN
maximum likelihood method by adding the gradient descent with the standard BP decoder to estimate information bits in
algorithm to generate a DL network. Based on the orthog- a noisy environment. For high density parity check (HDPC)
onal AMP (OAMP) iterative algorithm combined with the codes, the performance of the BP algorithm is relatively
DL network, OAMP-Net was proposed in [127]. The pur- poor. Nachmani et al. successively proposed the BP-DNN
pose of OAMP-Net is to add adjustable training parameters algorithm [138] and the BP-RNN algorithm [139], which com-
on the basis of the original algorithm to further improve the bined the DNN and RNN networks with BP algorithms to
signal detection performance of the existing algorithm. With improve the performance of BP algorithms applied to HDPC.
the advantage of fewer trainable parameters, the model-driven Reference [140] put forward an end-to-end wireless com-
detection network [128] was designed to improve detection munication system model, which explains the feasibility of
performance by expanding a specific iterative detector and replacing the processing module of the physical layer by DNN.
adding some trainable parameters. In addition, an adaptive The authors of [141] provided a differentiable channel compu-
signal detection method named JC-Net for massive MIMO tational model, which can be used for supervised autoencoder
systems was proposed in [129]. JC-Net has a foundation for training. Since then, many non-modeled methods [142], [143]
the traditional Jacobi detector, adding trainable parameters to have been developed based on synchronization-around meth-
improve the convergence speed and perform corresponding ods, none of which require any channel knowledge and can
soft projection. In FDM networks, the BS requires to attain be directly executed on real hardware. In [144], the authors
downlink CSI feedback to perform precoding and achieve treated the communication system as an end-to-end DRL
performance gains. However, there are many configured anten- autoencoder. This technique does not require any information
nas in massive MIMO systems, so the feedback overhead of about the actual channel model.
AI for Upper Layer: In recent years, AI has been intro- wireless networks need to dynamically turn BSs on and off
duced into the upper layers of wireless communications according to user traffic demands. In view of the dynamic ran-
to tackle various problems, thereby enabling near-optimal domness of traffic demands, [162] proposed to apply DL to
network performance. For example, since artificial neural analyze and predict the traffic of each BS, and then used DRL
networks have the approximation characteristics of universal to control the switch of BSs according to the predicted traffic.
functions, [145] and [146] adopted a data-driven approach to In addition to [162], the work of [163] also introduced a DRL-
allow the training model to autonomously learn user access supported intelligent sleep strategy for BSs to reduce network
and power allocation strategies. Supervised learning requires energy consumption. In sparsely populated areas, UAVs can
predicting the labels of the training data, resulting in exces- be leveraged as air BSs to serve terrestrial communication
sive data preprocessing burden. So, from the perspective of terminals. Considering the coverage limitation of the UAV
unsupervised learning, the authors of [147] leveraged a feed- and the moving variation of the air-to-ground channel, the
forward neural network to autonomously learn the optimal authors in [164] discussed to use DRL algorithms with deep
power allocation. Bypassing channel estimation, the work Q-learning for deployment planning of air BSs.
in [148] efficiently scheduled interfering links based only
on the geographic locations of transmitters and receivers via
DL algorithms. In multi-cell systems, [149] approximated C. Integration of Blockchain and AI
optimal link scheduling and power control through DNNs. In recent years, the frontier technologies of blockchain and
Specifically, a matching link schedule was estimated using AI have aroused widespread attention and in-depth research in
the deep Q-network, and then power was allocated to the academia and industry. Blockchain technology has the char-
corresponding link schedule. acteristics of decentralization, anonymity, openness and trans-
As a data-driven ML method, DRL can directly learn parency, and immutability. However, the blockchain needs to
dynamic environmental laws and obtain optimal decisions. be improved urgently in terms of scalability, energy consump-
Therefore, DRL can endow the network with the ability to tion, and security. As a powerful analysis and decision-making
self-optimize management according to the dynamic environ- tool, AI can predict and analyze data in real-time scenarios and
ment, making intelligent communication possible. Next, we make optimal decisions. Nevertheless, the centralized structure
focus on the application of DRL in the upper layer of wire- of AI and its demand for security and credibility have greatly
less communications. For example, [150] considered a power limited the wide application of AI. Therefore, there is com-
control problem in cognitive radio. Here, to improve the spec- plementary potential for the combination of blockchain and
trum usage rate, secondary users performed communication AI. As shown in Fig. 4, we respectively elaborate and ana-
by occupying the spectrum of primary users. To satisfy the lyze from the two aspects of blockchain for AI and AI for
service quality of primary and secondary users, the author blockchain.
of [150] proposed a DRL-enabled power control scheme. 1) Blockchain for AI: From the perspective of AI technol-
Reference [150] was aimed at the single-user power control ogy, blockchain as a trusted platform can create a secure,
problem, which cannot be applied to multi-user scenarios. To immutable, and distributed system for AI. In this secure
this end, the work of [151] discussed the problem of multi-user system, no third-party participation and management are
power resource allocation in cellular networks, where the goal required, users trust each other and share data. Accordingly,
was to maximize the weighted sum-rate of the entire network. based on the huge and reliable data set, the accuracy
Moreover, [152] extended [151] to multi-user device-to-device of the agent’s decision-making is improved. We introduce
(D2D) communication scenarios. The authors of [153] used blockchain-driven AI from some aspects below, including data
DRL to analyze the user’s data request, and replaced the file management [165], [166], [167], [168], decentralized intelli-
in the cache as per the user’s request rule. A DRL-empowered gence [169], [170], [171], [172], security and privacy [173],
computing resource allocation scheme was presented in [154]. [174], [175], [176], and efficiency and scalability [177], [178],
In this scheme, IoT devices adopted DRL algorithms to deter- [179], [180].
mine the power of each computing task to be executed locally, Data Management: The massive amount of AI data lacks
and a power of 0 meant that the computing task is executed in a consolidated and efficient sharing mechanism and manage-
MEC servers. In addition, [155], [156], [157], [158] also suc- ment method. The poor maintainability of open-source data
cessfully employed DRL algorithms in the joint optimization sets leads to uneven data quality, and the data is not centralized
problem of caching and computing resources, indicating that and unified. The distributed database of blockchain efficiently
DRL has a strong application prospect in managing network collects, shares, and stores the data of each node, so that every
resources. participant on the network can access the data. This can pro-
In [159], DRL was used to realize intelligent horizontal vide AI with broader data access and more efficient data mon-
handover between BSs. The work of [160] further attempted etization mechanisms. For instance, to make the updating of
to combine access control and resource allocation, and con- AI models more efficient, based on blockchain technology, the
sidered the DRL algorithm to solve the joint optimization authors of [165] proposed the novel configurable distributed
problem of user access and channel allocation in multi-layer AI framework, where participants collaborated to construct
BS cellular networks. As the size of the network esca- datasets and used smart contracts to host continuously updated
lates, the probability of network failure also increases. The AI models. In order to break the data barriers between differ-
authors of [161] attempted to apply DRL to network fault ent mobile operators, [166] designed a blockchain-empowered
self-healing. To enhance energy efficiency and reduce costs, data sharing framework and a Hyperledger-based prototype
Fig. 4. Taxonomy of the integration of blockchain and AI.
system. This system utilized smart contract-based monitoring single point of failure problem of federated learning (FL), the
and fine-grained data access control to create a safe and reli- work in [171] provided a decentralized FL scheme entitled
able environment for data sharing. Using blockchain to help ChainFL. The ChainFL utilized blockchain to delegate the
AI manage trusted data, the work in [167] introduced a secure responsibility of storing and aggregating models to nodes on
large-scale Internet architecture called SecNet. The SecNet the network without requiring any central server. In edge AI-
can realize secure data storage, computing, and sharing, and supported IoT networks, to break knowledge silos, the authors
enhance AI with a large number of data sources. The work of [172] proposed a P2P knowledge payment sharing architec-
in [168] demonstrated a blockchain-enabled joint framework ture, which made use of the knowledge consortium blockchain
for efficient data acquisition and secure sharing. This proposed to ensure that knowledge management and market transactions
framework used DRL to achieve the maximum amount of col- are safe and efficient. This knowledge consortium blockchain
lected data, and leveraged the Ethereum blockchain technology included the new encrypted currency knowledge currency,
to ensure the security and reliability of data sharing. smart contracts, and new transaction consensus mechanism
Decentralized Intelligence: Through AI algorithms, learn- proof.
ing results and models can be obtained from massive data. Security and Privacy: For AI technology, the greater amount
Due to the distribution of IoT devices or edge comput- of having data, the higher accuracy of its training model.
ing devices and the data heterogeneity, the cooperation of However, if a small part of this data has security issues,
multiple devices is required to complete complex model train- the validity of the data will affect the system’s decision-
ing tasks. That is, different devices need to share data for data making accordingly and thus the overall performance of the
analysis and prediction. Local learning models can also be system. Fortunately, blockchain has many technologies such as
shared across devices and then aggregated. Blockchain tech- anonymity, immutability, interface access control, and signa-
nology can guarantee that AI completes the interaction of ture authentication and authorization to ensure the security and
data or models between devices in the decentralized environ- privacy of transaction data, and to provide quality assurance
ment. To realize asynchronous cooperative computing among for the data required for AI model training. The work in [173]
untrusted nodes, the work in [169] developed a decentralized, proposed a blockchain-authorized edge intelligence system,
privacy-preserving, and secure computing paradigm, which which assured the security, privacy, latency, and efficiency
adopted various technical means such as blockchain, decen- of edge device data. Here, the public blockchain guaranteed
tralized learning, and homomorphic encryption. A blockchain- the security and privacy of data of edge devices, while the
assisted distributed secure multi-party learning architecture private blockchain ensured secure communication between
was proposed in [170]. Specifically, the authors formulated edge intelligent servers. The authors of [174] demonstrated
two types of Byzantine attacks, as well as elaborated “off- a distributed DL architecture named DeepChain. DeepChain
chain” and “on-chain” mining schemes. In response to the utilized the value-driven incentive mechanism of blockchain
to encourage parties to collaborate in DL model training and [186], [187], [188], security and privacy [189], [190], [191],
share the obtained local gradients. Meanwhile, DeepChain [192], and intelligent decision [22], [193], [194], [195].
guaranteed the privacy of local gradients for each participant Scalability: Currently, as the number of transactions
and provided auditability for the entire training process. To increasing significantly, scalability is the biggest barrier to
prevent malicious attacks on AI models, the work in [175] the widespread application of blockchain technology. In
introduced Biscotti, which was a fully decentralized P2P blockchain systems, the core of scalability is to tackle the
large-scale multi-party learning scheme. The Biscotti adopted problems of transaction throughput and transaction speed.
blockchain and cryptographic primitives to coordinate the Due to the characteristics of decentralization and network-
privacy-preserving ML process among peer nodes. The authors wide broadcasting, each node on the blockchain will record
of [176] presented the blockchain-assisted asynchronous FL transactions generated by the entire network, leading to low
(BAFL) architecture, where the blockchain ensured that model efficiency. AI can introduce DRL or data sharding technol-
data cannot be tampered with, and assured decentralized ogy to propose new solutions to blockchain scalability issues
and secure data storage, and asynchronous learning acceler- and improve system efficiency. In [181], through the DRL
ates global aggregation. The proposed BAFL guaranteed that algorithm, the agent dynamically selected different consensus
each device uploaded the local model whenever the global algorithms and block production nodes, and adjusted the block
aggregation can converge the global model faster. size and time interval. The agent found optimal parameters to
Efficiency and Scalability: When using AI techniques, such improve the scalability, while ensuring the decentralization,
as DL, it is difficult for people to understand what is in the latency, and security of blockchain networks. To obtain better
black box and explain the decisions made by AI systems, so throughput, the authors of [182] adopted the deep Q network
AI cannot be verified or trusted. Furthermore, without appro- algorithm to dynamically adjust the block size, time interval,
priate incentive mechanisms, various parties may be reluctant and the number of shards to seek the optimal related param-
to participate in data training. The above-mentioned prob- eters, while meeting the security of the system. The work
lems will reduce the efficiency and scalability of AI systems. in [183] studied a DRL-enabled adaptive blockchain scheme,
Blockchain can track every link in the data processing and which improved scalability and met the needs of different
decision-making chain for explainable AI. Appropriate incen- users. Specifically, according to the service quality require-
tive mechanisms can also be introduced from blockchain. The ments of users, the DRL algorithm selected the most suitable
transparent and cost-effective incentive mechanism design can consensus protocol for blockchain systems. To overcome the
be implemented, which will greatly improve the enthusiasm limitations of existing blockchain static sharding, [184] intro-
of all parties in AI systems to take part in the training. duced a DRL-based dynamic sharding blockchain framework
A blockchain-based framework for more trustworthy and called SkyChain. In the dynamic environment of blockchain
explainable AI was presented in [177]. This framework lever- systems, this presented SkyChain can dynamically adjust the
aged smart contracts to record and manage interactions, as resharding interval, number of shards, and block size in order
well as provided consensus for trusted oracles. The proposed to maintain a long-term balance between performance and
framework also addressed decentralized storage, registry, and security.
reputation supporting services. To obtain better trusted AI, the Energy Consumption: Blockchain mining requires a large
authors of [178] designed a blockchain-enabled FL system, amount of computing power and electricity resources. At
which used the blockchain to track the source information of present, Bitcoin consumes about 2.55 billion watts of elec-
the trained model. Aiming at the incentive mechanism problem tricity every year, almost the same as the annual electricity
of FL, the work in [179] presented a reputation-based miner consumption of some small countries. If the energy consump-
selection scheme, and designed an efficient incentive mech- tion problem cannot be solved well, the value of the blockchain
anism by adopting a multi-weight subjective logic model to itself will be diluted. To avoid excessive consumption of com-
evaluate the user’s credit. The proposed scheme also lever- puting resources and energy resources in this mining process,
aged the consortium blockchain to achieve secure reputation AI algorithms can understand the blockchain network process
management for miners in a decentralized manner. In [180], and architecture, and explore a more effective consensus mech-
the authors described the FLChain framework based on trust anism, which can make transactions on blockchain networks
and incentive. The FLChain saved miners’ information and execute faster.
verifiable training details for public audits. The incentive A Proof-of-Learning consensus protocol was formed by
mechanism of the FLChain encouraged honest and trustwor- combining ML algorithms in [185]. This Proof-of-Learning
thy miners. Otherwise, malicious nodes will be punished, so protocol performed model training through ML of given tasks.
as to maintain a healthy and reliable public platform. Then, the ranking was based on the minimum loss function
2) AI for Blockchain: From the perspective of blockchain value. Finally, the optimal model parameters were selected
technology, its scalability and system energy consumption can and verified by other mining nodes to achieve distributed con-
be optimized through AI algorithms. Using AI algorithms in sensus. Through combining DL algorithms, the work of [186]
blockchain networks, security vulnerabilities brought about by proposed a Proof-of-Deep-Learning consensus protocol, which
the implementation of smart contracts and consensus mech- forced the agent to conduct DL model training, and proposed
anisms can be identified and detected. We will represent the training model as a proof of effectiveness. Only when an
AI-driven blockchain from four aspects below, including scal- appropriate DL model was generated, can miners reach the
ability [181], [182], [183], [184], energy consumption [185], consensus and generate new blocks. In response to the energy
consumption problem in blockchain networks, the authors Bitcoin price. The results showed that the proposed four types
of [187] introduced a Proof-of-Useful-Work energy-saving of ML models predicted the actual Bitcoin price with a very
consensus protocol. The proposed protocol required training a low error rate. Applying ML, the work in [194] demonstrated
DL model during the mining process, and mining new blocks an Ethereum prediction model, which can predict transaction
only when the performance of the training model exceeded execution times in Ethereum systems. The transaction execu-
a given threshold. The work in [188] demonstrated an AI- tion time refers to the time frame within which a miner node
enabled node selection algorithm that exploited the nearly accepts and includes a transaction in a block. To reduce the
complementary information of each node and relied on a spe- huge risk and cost brought by the fork, the work of [195]
cially designed CNN to reach consensus. In order to ensure adopted an ML method to predict the blockchain fork, and
the decentralization and security of the network, dynamic compared the prediction accuracy of the fork by four well-
thresholds were used to obtain super nodes and random nodes. known ML methods, namely K Near Neighbor, Naive Bayes,
Security and Privacy: The decentralized power of Decision Tree, and Multilayer Perceptron.
blockchain may be at risk of abuse, especially since the smart 3) Motivations of the Integration of Blockchain and AI for
contracts and consensus mechanisms in blockchain technol- Wireless Communications: Blockchain can establish a secure
ogy are vulnerable to malicious network attacks or tampering. and decentralized resource sharing environment. AI can solve
As more and more personal data is stored in blockchain some problems with uncertain, time-varying, and complex
systems, data privacy protection becomes critical. We employ characteristics. As shown in TABLE VI, we summarize and
AI-assisted methods to identify and detect security vulnerabili- compare blockchain for 5G/6G, AI for 5G/6G, blockchain for
ties, greatly improving the security and privacy of blockchains. AI, and AI for blockchain. We conduct a critical and original
As an illustration, the work of [189] learned by extracting rel- discussion of these existing solutions, highlighting the advan-
evant features from user accounts and operation codes of a tages, disadvantages, and main findings of various solutions.
large number of smart contracts, and used the ML algorithm Although blockchain and AI are promising technologies to be
XGBoost to detect whether there is a potential Ponzi scheme applied in 6G networks, there are still many challenges and
in the smart contract. Ponzi scheme is a classic investment unresolved problems. Both blockchain and AI have attracted
fraud, and it also has a blockchain-based form. The essence significant attention recently. The combination of these two
of Ponzi is that the investment of new investors is the return of technologies may further improve the performance of 6G
old investors. On the Hyperledger Fabric blockchain platform networks. In the first place, to more systematically under-
and with a relatively low tolerance for malicious activities, stand the integration and application of blockchain and AI
the authors of [190] designed an external detection algorithm technologies for 6G networks, we summarize the benefits
based on supervised ML before the consensus protocol as of blockchain for AI and the benefits of AI for blockchain,
the consensus of the previous step. This proposed detection respectively. For details, please refer to the above subsec-
algorithm verified the new data compatibility, and discarded tions: Sections II-C1 and II-C2. Then, we briefly describe
suspicious data to improve the network’s fault tolerance for the the benefits that fusing blockchain and AI can bring to 6G
second-step consensus. In blockchain-based systems, a data networks.
fusion scheme based on collaborative clustering features was On the one hand, blockchain can improve AI in terms
represented in [191]. The data fusion scheme applied AI algo- of data management, decentralized intelligence, security and
rithms to train and analyze data clusters to detect abnormal privacy, and efficiency and scalability. Firstly, blockchain col-
intrusion behaviors. To effectively detect abnormal behavior of lects, shares, and stores data for AI, so that every participant
blockchain systems, the work in [192] proposed an encoder- on the network can access the data. Blockchain-supported
decoder-based DL model, which was an unsupervised model methods can provide AI with more efficient data manage-
trained with aggregated information extracted by monitoring ment mechanisms and wider data access. Secondly, blockchain
blockchain transactions. can ensure that AI can complete the collaborative interaction
Intelligent Decision: With the rise of blockchain technology, of data or models between devices in a decentralized envi-
more people turn to study blockchain-empowered application ronment. In addition, blockchain uses its own anonymity,
projects. Investors want to predict some important behaviors immutability, interface access control, signature authentica-
of blockchain systems, such as cryptocurrency prices, trans- tion and authorization, and other technologies to safeguard
action confirmation times, and blockchain forks. We consider the security and privacy of transaction data in the AI system.
embedding AI algorithms into blockchain systems and allow- Finally, blockchain can track every link in the data process-
ing AI to optimize or make decisions for the entire system, ing and decision-making chain for explainable AI. Transparent
which is more conducive to investors making correct policies. and cost-effective incentive mechanisms can also be designed,
In [22], without knowing the details of the blockchain network which will effectively upgrade the efficiency and scalability of
model, a multi-dimensional RL algorithm was proposed to AI systems.
solve the mining problem with the Markov decision process. On the other hand, AI can enhance the performance of
The designed algorithm can obtain the near-optimal mining blockchains in terms of scalability, energy consumption, secu-
strategy solution in the time-varying blockchain network. In rity and privacy, and intelligent decision. First of all, AI
both short-term (1 day and 7 days) and medium-term (30 days can introduce DRL or data sharding technology, to propose
and 90 days) time periods, the authors of [193] employed clas- new solutions for blockchain scalability issues and ameliorate
sification and regression models of ML to predict the trend of system efficiency. Furthermore, in order to avoid the massive
TABLE VI
C OMPARISON OF E XISTING S OLUTIONS
resource consumption of blockchain, AI algorithms can dis- communication systems. Through this section, we have gained
sect the blockchain network process and architecture, as well valuable insights into the opportunities and challenges of lever-
as explore a more effective AI-based consensus mechanism aging blockchain and AI in wireless communication systems.
based on AI, so that transactions on the blockchain can be exe- We have also recognized the importance of considering differ-
cuted faster. Next, AI-assisted methods are used to identify and ent integration approaches and identifying suitable use cases
detect security vulnerabilities, which remarkably augments the to maximize the potential of these technologies. Overall, this
security and privacy of blockchains. Ultimately, we can con- section provides a solid foundation for further exploration and
sider embedding AI algorithms into blockchain systems and analysis of the integration of blockchain and AI in wireless
letting AI optimize or make decisions for the entire system, communication systems.
which is more conducive to investors making make correct
decisions.
According to the above analysis, the amalgamation of III. I NTEGRATION OF B LOCKCHAIN AND AI FOR
blockchain and AI has complementary potential. Blockchain W IRELESS C OMMUNICATIONS
can conduct a secure, immutable, and distributed system For the existing problems of blockchain and AI, the inte-
for AI technologies. In this system, users trust each other gration of these two technologies can complement each other.
and share data. Based on a huge and reliable data set, the Facing the 6G era, the network will meet new applica-
performance of AI algorithms and decision-making can be tion scenarios and new performance requirements. Diverse
effectively upgraded. Using AI algorithms, the scalability and applications, communication scenarios, ultra-heterogeneous
energy consumption issues of blockchain can be mitigated, and network connections, and service requirements for extreme
its security vulnerabilities can also be identified and detected. performance all put forward higher demands on mobile com-
The integration of AI and blockchain is not only to enhance munication networks [46]. The merger of blockchain and AI
each other, but also to push and optimize various services and can not only play to their respective advantages [196], [197],
applications for 6G scenarios in the process of mutual pro- but also better bring optimization and improvement to various
motion. In this case, a reliable, secure, and ultra-low latency services and applications in 6G networks [47], [48], [198]. In
network environment can be provided for 6G wireless com- this section, we will discuss broadly the applications of merg-
munications. Consequently, the research on the integration of ing blockchain and AI in 6G networks, including 6G secure
blockchain and AI is extremely important and worth expecting services [199], [200], [201], [202], [203], [204], [205], [206],
in 6G networks. [207], [208], [209], [210], [211], [212], [213], [214], [215],
In this section, we have provided a comprehensive overview [216], [217], [218], [219], [220], [221], [222], [223], [224],
of the fundamental concepts, characteristics, and categories [225], [226], [227], [228] and 6G IoT smart applications
of blockchain and AI. We have also discussed the classic [229], [230], [231], [232], [233], [234], [235], [236], [237],
applications of both technologies in wireless communication [238], [239], [240], [241], [242], [243], [244], [245], [246],
systems. Then, we systematically summarized the integration [247], [248], [249], [250], [251], [252], [253], [254], [255],
of blockchain and AI from two directions: blockchain-assisted [256], [257], [258], [259], [260], [261] as depicted in Fig. 5.
AI and AI-assisted blockchain. Furthermore, we analyzed The convergence of the integration of blockchain and AI for
the advantages of integrating blockchain and AI for wireless wireless communications is illustrated in Fig. 6. Furthermore,
Fig. 5. Taxonomy of the integration of blockchain and AI for wireless communications.
we thoroughly discuss operating frequencies, visions, and automatically learn user behavior patterns or further make
requirements from the 6G perspective. optimal decisions for users. Through the use of AI, behav-
iors such as users’ mobility and data/computing traffic can
be dynamically predicted, so as to optimize the allocation of
A. Secure Services wireless resources.
As mentioned in the previous section, the combination of In 5G beyond and 6G wireless communications, the
blockchain and AI can not only promote each other, but also authors of [199] designed a blockchain- and AI-supported
provide better services. In 6G networks, wireless resources dynamic resource sharing architecture. The low-cost and low-
such as spectrum, computing, and caching are some of the complexity hierarchical blockchain is an enabling platform
most concerned services. The development of 6G networks for dynamic resource sharing. AI is employed to optimize
will bring explosive growth of user data, and security and data management in the process of dynamic resource sharing.
privacy services are also the keys to improving the overall The proposed architecture in [199] can successfully implement
performance. In this subsection, we will focus on some key dynamic spectrum sharing. Simulation results show that DRL
6G secure services, where blockchain and AI are simultane- can effectively maximize the user’s profit margin compared to
ously applied, including spectrum management [199], [200], the traditional Q-Learning algorithm and stochastic decision-
[201], [202], [203], [204], computation allocation [205], [206], making. There is a competitive relationship between multiple
[207], [208], [209], [210], content caching [211], [212], [213], operators. Therefore, the operator’s spectrum utilization rate
[214], [215], [216], and security and privacy [217], [218], and infrastructure deployment efficiency are deeply low. 6G
[219], [220], [221], [222], [223], [224], [225], [226], [227], networks will enable more flexible mobile network deploy-
[228]. TABLE VII presents an analysis of the integration of ments through spectrum and infrastructure sharing among
blockchain and AI for secure services. operators. Under the 6G network of multiple mobile operators,
1) Spectrum Management: Radio spectrum resources are the work in [200] developed a blockchain- and AI-empowered
scarce resources. Spectrum is widely used by various radio multi-plane framework for open spectrum and infrastructure
technologies and services, resulting in increasing demands sharing. The developed framework of [200] consists of user
for spectrum resources in various industries and fields. plane, infrastructure plane, operator plane, blockchain plane,
Facing ever-increasing demands for radio spectrum, spec- and AI plane. As a case study of the developed multi-
trum management has never been more challenging. Given plane framework, the authors of [200] utilized deep RNN
that traditional fixed spectrum allocation strategies lead to and blockchain technology to design a workflow for dynamic
inefficient spectrum usage, dynamic spectrum management is spectrum management among multiple operators. Simulation
proposed as an encouraging approach to alleviate the spec- results indicate that the designed intelligent dynamic spectrum
trum scarcity problem [262]. Blockchain and AI are two management workflow can provide more equitable bandwidth
promising enabling technologies for solving spectrum manage- allocation for all users compared with static and semi-
ment issues. Blockchain can be applied to spectrum auctions, intelligent workflows. The authors in [201] proposed a general
which improves security and decentralization, and reduces edge intelligent privacy-preserving framework, which is inte-
spectrum management costs. On the other hand, AI technolo- grated blockchain with FL and can be specifically applied to
gies represented by DL and DRL are very powerful and can spectrum resource sharing. Spectrum sharing information is
Fig. 6. The convergence of the integration of blockchain and AI for wireless communications.
recorded in the blockchain as a transaction, and consumers was proposed in [205]. In large-scale distributed SDNIIoT
pay spectrum leasing fees to providers through the blockchain. networks, the authors of [205] presented a novel permissioned
FL can not only learn computational results from data, but blockchain-energized consensus mechanism. This mechanism
also provide optimized spectrum sharing strategies. In the synchronizes local views among different SDN controllers,
digital twin edge network, the work of [202] provided a per- and finally achieves a consensus on the global view. To fur-
missioned blockchain-based FL architecture. To improve the ther improve the throughput of the blockchain system, a joint
communication efficiency, the authors of [202] also designed optimization problem of view change, access selection, and
an efficient asynchronous aggregation model and DRL-based computational resource allocation was constructed in [205].
algorithm to optimize user scheduling and spectrum resource To this end, a dueling deep Q-learning method was proposed
allocation. For the problem of adaptive resource allocation, ref- to deal with the joint problem. In the network environment
erence [203] presented a blockchain-based MEC framework, with more and more IoT devices, reference [206] introduced
where DRL method is utilized to tackle the joint optimization a general system framework for blockchain-assisted edge com-
problem of spectrum resource allocation and block generation. puting. In this framework, the complete procedure of transac-
In 6G networks, reference [204] discussed the possibility of tions between IoT side and edge nodes is specified step by
distributed ledger technology and ML techniques to promote step. Furthermore, the authors of [206] provided a smart con-
the coexistence of licensed and unlicensed spectrum. tract for resource allocation in the private blockchain network.
2) Computation Allocation: Computing resource is one of The problem of allocating edge computing resources to data
the key resources in 6G wireless communications. Combining service users was described as a continuous-time Markov deci-
blockchain and AI can allocate and offload computing sion process. In [206], the designed smart contract adopted
resources more efficiently. For instance, to facilitate the the RL algorithm and asynchronous advantage actor-critic
scalability and flexibility of resources, SDNIIoT, which algorithm to tackle the problem of edge computing resource
integrates software-defined networking (SDN) into the IIoT, allocation. Compared with some traditional algorithms, the
TABLE VII
A NALYSIS OF I NTEGRATION OF B LOCKCHAIN AND AI FOR S ECURE S ERVICES
proposed algorithm can distinguish multiple service quality the security and offloading requirements in mobile edge-cloud
requirements of different service users, thereby ameliorating IoT networks, the work in [208] developed a secure com-
the allocation efficiency of computing resources. putation offloading scheme by combining blockchain and
To enhance security resource management for edge users DRL. In this scheme, the computing tasks of mobile IoT
in a distributed manner, the work of [207] demonstrated a devices can be offloaded to the cloud or edge servers. To
blockchain-guided offloading mode, which maximizes data upgrade the security of data offloading, a trusted smart
availability. This mode alleviates the non-probabilistic hard- contract-empowered access control mechanism was presented
ness problem of data availability due to cooperative and in [208]. This mechanism prevents cloud resources from being
probabilistic data offloading. The data offloading process accessed by illegal offloading devices. Again, for example,
occurs in the edge network assisted by blockchain. In this the authors of [208] formulated the computation offloading,
data offloading mode, Naive Bayes’ learning was employed edge resource allocation, bandwidth allocation, and smart con-
to linearly classify offloaded and non-offloaded instances to tract cost as a joint optimization problem. This joint problem
obstruct service delays and unnecessary backlogs. Aiming at can be solved using an advanced DRL algorithm with a
double-duling Q-network and a optimal offloading strategy Furthermore, considering vehicle mobility, the work of [213]
for all IoT devices can be obtained. Under the vehicular constructed a vehicle-to-vehicle content caching optimization
fog computing network, [209] introduced a blockchain- and problem, and applied the advanced DRL algorithm to obtain
ML-assisted task offloading framework. The framework uti- the optimal caching strategies.
lizes smart contracts and Merkle hash trees to facilitate fair In machine-to-machine communication networks, a
task offloading and mitigate various security attacks. Then, blockchain- and edge computing-enabled network framework
to tackle the task offloading optimization problem, an intelli- was proposed in [214]. In this architecture, edge computing
gent task offloading algorithm based on online learning was improves data caching and computing capabilities, and
delineated in [209]. Without requiring the information and blockchain ensures data security and efficiency. To reduce
CSI of the vehicle fog computing server, this algorithm can latency, the authors of [214] framed the joint optimization
learn the long-term optimal unloading strategy and effectively problem of content caching, computation offloading, and
reduce the unloading delay, queuing delay, and switching blockchain scheduling as a discrete Markov decision process.
cost. In an air-to-ground integrated power system, the work Then, a dueling optimization algorithm inspired by a dueling
of [210] formulated a joint optimization problem of device- deep Q-network was adopted to solve this joint optimization
side task offloading and server-side resource allocation. Then, problem. To polish up the file caching hit ratio, a compression
it demonstrated an electromagnetic interference-aware com- algorithm named CREAT applied to the caching was provided
putational offloading algorithm by combining blockchain and in [215]. CREAT integrates FL and blockchain, which can
semi-distributed learning. cache files by predicting the popularity of different files
3) Content Caching: Caching is introduced into the through the FL algorithm and speed up the response to
6G communication architecture. Specifically, by deploying file requests from IoT devices. In the meantime, blockchain
caching in terminals, BSs, and core network gateways, pop- technology ensures the security of data transmitted by IoT
ular content is cached to the location closer to the user. devices and gradients uploaded by edge nodes. Additionally,
Content caching can realize the local response of some user an advanced compression algorithm is adopted in [215] to
requests, reduce the transmission delay of the requested con- compress the uploaded gradients, so as to speed up the
tent, improve the user experience, and balance the network training process of FL. The most critical reason for the
load. However, the current content caching strategy is rel- hindered development of user sharing-depended caching
atively static. Therefore, the caching performance depends solutions is the lack of incentive mechanism. To upgrade user
heavily on the popularity of the content and lacks the per- sharing-depended caching performance, the work of [216]
ception of users’ personalized demands. At the same time, the provided a blockchain- and smart contract-guided distributed
caching deployment in the mobile environment brings great cache sharing incentive mechanism. In this mechanism, D2D
security risks to the user’s data privacy. Blockchain and AI and MEC caching nodes incentivize their cache sharing
are key enabling technologies to address these challenges. willingness by receiving expected rewards. Then, to depress
For example, in [211], a blockchain-guided edge intelligence consensus latency and undertake confidence, a partially PBFT
framework entitled AI-Chain for 6G wireless networks was consensus protocol was suggested. Furthermore, both the
pioneered, which integrates DL and blockchain. The frame- cache placement problem and the scene selection problem
work benefits from the transferability of DL. Specifically, each can be described as Markov decision processes [216]. Then,
lightweight edge node trains neural network components, and the DRL algorithm with deep Q-Network was presented to
then shares local learning results on the blockchain. To demon- deal with these problems.
strate the effectiveness of the proposed framework, the work 4) Security and Privacy: Security and privacy vulnera-
in [211] applied AI-Chain to handle the joint resource allo- bilities increase with the scale of wireless communication
cation problem of networking, edge computing, and content systems. Accordingly, the security and privacy of commu-
caching. Aiming at the privacy leakage problem in cognitive nication participants have become an important issue. The
vehicle networks, [212] provided a blockchain-inspired content combination of blockchain and AI can provide more effective
caching architecture. Under this framework, road side units solutions to security and privacy challenges 6G in wireless
(RSUs) and vehicles that provide content cache the content communication systems [217]. As an illustration, a distributed
in advance and broadcast it to surrounding vehicles. Vehicles heterogeneous IoT network architecture by integrating per-
with content requirements can selectively download the related missioned blockchain and AI was designed in [218], so
content. Once the content transaction is completed, the trans- as to add additional security performance. Here, blockchain
action record is written to the blockchain and broadcast to all is applied to share and store data of IoT devices. On the
RSUs and vehicles. To enhance the cache hit ratio, the cogni- hand, ML algorithms can detect Malware and cyberattacks
tive engine can sense the content demands of the underlying of distributed IoT networks and can classify these anoma-
vehicles. Then, in [212], the perception data is analyzed by ML lous behaviors in real time. On the other hand, the existing
and DL algorithms, and predictive cached results are presented identity authentication mechanisms have the problems of sin-
to RSUs and vehicles that provide content. In vehicle edge gleness and poor adaptability. Based on this consideration,
computing networks, a secure and intelligent content caching the authors of [219] introduced a secure user authentication
scheme by combining permissioned blockchain and DRL was mechanism called ATLB with the help of transfer learning and
demonstrated [213]. In this scheme, vehicles accomplish con- blockchain. ATLB described layered blockchain to implement
tent caching and BSs sustain the permissioned blockchain. the privacy protection of the authentication mechanism with
collusion attack and Sybil attack. In addition, to reduce the more and more attention from academia and industry. IoT is a
model training time, reference [219] added transfer learning huge network formed by combining various information sens-
to optimize the authentication mechanism and build a trust- ing devices with the Internet, realizing the interconnection
worthy and intelligent blockchain. In the edge service network of people, machines, and things at any time and any place.
of IIoT, a distributed ML scheme guided by blockchain can In this subsection, we will extensively discuss some impor-
guarantee the security and privacy of data processing of tant 6G IoT smart applications supported by both blockchain
multiple resource-constrained devices [220]. To reduce the and AI, including smart healthcare [229], [230], [231], [232],
response delay of edge services, the proposed scheme of [220] [233], [234], [235], [236], smart transportation [237], [238],
employed blockchain to replace cloud servers as trusted third- [239], [240], [241], [242], [243], [244], [245], [246], smart
party institutions. Moreover, a smart contract-based incentive grid [247], [248], [249], [250], [251], [252], [253], and UAV
mechanism was applied to encourage multiple devices to [254], [255], [256], [257], [258], [259], [260], [261]. The
participate in computing tasks. Furthermore, a size-weighted analysis of integration of blockchain and AI for IoT smart
aggregation strategy was discussed to validate and integrate applications is shown in TABLE VIII.
model parameters, thereby improving model accuracy. The 1) Smart Healthcare: At present, due to the impact of the
SM2 public key cryptosystem was applied in [220] to complete new crown pneumonia epidemic, the integration of emerging
the privacy protection of model parameters in edge services. technologies and medical scenarios is accelerating. New mod-
There are also security and privacy challenges for data in the els such as telemedicine and intelligent pre-diagnosis have
software-defined Internet of Vehicles. To address these chal- become rigid needs. Thus, the smart medical industry has ush-
lenges, the work in [221] developed a spatial crowdsourcing ered in a new round of outbreaks. The fusion of two emerging
framework guided by multiple blockchains and DRL together. technologies, blockchain and AI, achieves leap-forward devel-
In [222], blockchain and DL provided two levels of secu- opment of smart healthcare in areas including electronic health
rity and privacy for collaborative intelligent transportation records, health insurance, biomedical research, drug supply,
systems. For the first level, smart contracts were employed procurement process management, and medical education. For
for secure and intelligent data communication. For the second instance, a multi-party electronic health record sharing frame-
level, to prevent cyber attacks, LSTM auto-encoders in DL work entitled BinDaas was proposed in [229]. This framework
encoded data into new formats. To mitigate the precipitately integrates two technologies, blockchain and DL. Here, the
growing security challenges of in-vehicle networks, a novel blockchain stores the patient’s electronic health record data in
blockchain- and AI-empowered trust management architecture a secure manner. DL provides future disease risk predictions
was developed in [223]. for patients based on past repositories. Also, a lattice-based
More interesting, many works [224], [225], [226], [227], key and signature verification method was developed in [229]
[228] have proved that the combination of blockchain and FL to fight against quantum and collusion attacks. The authors
can greatly enhance data security and privacy. For example, to of [230] designed a distributed secure e-health architecture
defeat the privacy leakage problem in traditional device fail- named Healthchain-RL by combining blockchain and DRL.
ure detection, the blockchain-authorized FL scheme of [224] The blockchain in the designed architecture assembled hetero-
can verify the integrity of the data. Then, an innovative geneous healthcare institutions with dissimilar demands. At
centroid distance-weighted joint averaging algorithm can alle- the same time, the configuration of the blockchain network
viate the data heterogeneity problem in device fault detection. was optimized in real time through the online enlightened
In [225], an adaptive trust model was constructed by com- policy-making DRL algorithms, so as to accomplish a bal-
bining blockchain and FL, which treated personal trust as a ance between security, delay, and cost. The work in [231]
probability. Moreover, under the constraints of certain security investigated a blockchain- and DL-enabled secure and intel-
standards, the trust value of the terminal devices in different ligent healthcare diagnosis scheme. This healthcare diagnosis
networks was evaluated. The study in [226] took advantage of scheme mainly involves three main steps: 1) sharing medi-
blockchain and FL to design a distributed multi-party collabo- cal images based on orthogonal particle swarm optimization
rative data sharing scheme. This scheme built and shared data (OPSO) algorithm; 2) running hash value encryption through
models through FL without directly displaying the original neighborhood indexing sequence algorithm; 3) performing
data, thus realizing data privacy protection. Furthermore, per- medical diagnosis by using OPSO-DNN algorithms.
missioned blockchains supported secure data retrieval, thereby The study of [232] leveraged the advantages of blockchain
further controlling access to shared data and depressing the and FL to constitute a detection model for computed tomog-
hazard of data disclosure. Similarly, integrating blockchain and raphy (CT) scans of COVID-19 patients. The constituted
FL can also be applied to enhance the security and privacy model identified COVID-19 patients from lung CT images
of data during data transmission [227] and vehicle intrusion by applying the capsule network-supported segmentation and
detection [228]. classification method. Therein, a global DL model was trained
from data collected from different hospitals and facilities
using FL algorithms and blockchain was used to authenti-
B. IoT Smart Applications cate data from different sources. Likewise, for screening and
IoT has become a fundamental component of future wireless monitoring the COVID-19 epidemic, the authors in [233]
communication networks. Various smart IoT applications have developed a distributed collaborative healthcare architecture
great potential to provide exciting services, which is receiving guided by blockchain and FL. Different from the detection
TABLE VIII
A NALYSIS OF I NTEGRATION OF B LOCKCHAIN AND AI FOR I OT S MART A PPLICATIONS
of CT images in [232], [234] integrated blockchain and architecture called BATS, which introduced smart contracts
DNN to extract feature data from existing datasets, thereby to maintain the integrity and reliability of data stored on the
helping to diagnose severe diseases such as COVID-19 and blockchain. During emergencies in traffic jams, UAVs can
blood cancer. Detecting infectious diseases is difficult in transport some light healthcare items, such as medicines and
remote and resource-poor rural areas. Meanwhile, smart- surgical tools.
phones are predicted to be one of the main tools driving 2) Smart Transportation: With the rapid development of
improvements in healthcare delivery. Therefore, an end-to- information technology, smart transportation has also ushered
end DeoxyriboNucleic Acid (DNA) diagnostic platform based in more development opportunities. Smart transportation refers
on smartphones was proposed in [235]. In this platform, to the full use of big data, IoT, cloud computing, blockchain,
DL provided automatic detection of infectious disease DNA AI, and other technologies in the field of transportation.
molecular test results and their analysis. Blockchain was used Smart transportation can fully guarantee traffic safety, give
for secure data connection and management, thereby increas- full play to the efficiency of transportation infrastructure, and
ing the credibility of the entire diagnostic platform. Most improve the operational efficiency and management level of
critically, the authors of [235] also verified the feasibility of the the transportation system. In this part, we focus on the role
proposed platform through field tests in rural areas. AI algo- of blockchain and AI in promoting smart transportation. To
rithms can predict the type of disease and surgery based on improve the quality of service for vehicles in 6G networks,
the patient’s basic symptoms and historical health records. For the study in [237] discussed a collaborative service compo-
example, the extreme gradient boosting algorithm was applied sition approach by combining blockchain and RL. Here, the
in [236] to classify diseases. In this reference, it described blockchain was applied to announce combined tasks under
a blockchain- and AI-enabled drone-aided smart telesurgery certain constraints of service requests, ensuring that adjacent
nodes interact securely and record transactions. To quicken decentralized communication system for autonomous vehicles.
the procedure of service composition path choice, the authors In the proposed system, the local on-vehicle ML model
of [237] employed an RL algorithm to pick the optimal solu- updates are interchanged and validated with other vehicles in
tion closest to the node request. To ward off traffic congestion, the distributed manner. At the same time, the proposed system
reference [238] described a blockchain-guided secure crowd- made full use of the consensus mechanism of the blockchain
sourcing scheme. This scheme encouraged users to voluntarily and can complete the update of local vehicle ML models with-
take part in traffic forecasting by sharing traffic information out any third-party server. To enhance the reliability of edge
to earn tokens. Meanwhile, users can also spend these tokens data sharing between vehicles, a hybrid blockchain- and FL-
to acquire the required traffic information from the network. assisted secure data sharing framework was designed in [245].
Then, with the help of an LSTM neural network, the study In this framework, the hybrid blockchain was composed of the
of [238] fused the results of a feed-forward artificial neural permissioned blockchain and the local directed acyclic graph
network trained on historical data to prognosticate traffic con- of vehicle operation, which can further upgrade the security of
gestion probabilities on real-time data. Similarly, the work in-vehicle data. Furthermore, to further polish up the training
in [239] considered integrating blockchain, RL, and edge efficiency, the authors of [245] introduced an asynchronous
computing to alleviate the traffic congestion problem. FL scheme for model learning from edge data, and preferred
Currently, vehicle positioning has challenges of low accu- the better participating node through the DRL algorithm. The
racy and network congestion in data sharing. To address these work of [246] demonstrated a layered blockchain- and FL-
challenges, a blockchain- and DNN-assisted smart vehicle co- embedded vehicle knowledge sharing system. This system
localization scheme was provided in [240]. This scheme was adopted a hierarchical blockchain to record the FL model.
benchmarked with multiple traffic signs, and the DNN algo- Then, integrating a proof-of-learning-based consensus proto-
rithm was applied to correct position of vehicles. For the col with high-precision hierarchical FL effectively prevent the
localization errors of multiple vehicles, the work of [240] waste of a large amount of computing resources. Also, the
demonstrated a DL-inspired method for distance computation knowledge sharing process among vehicles was constructed
and prognostication to turn down localization errors for com- as a multi-leader and multi-follower non-cooperative game
mon vehicles on the unchanged road segment. Additionally, problem in [246].
to actualize the safety information sharing between vehicles, 3) Smart Grid: The construction of a smart grid provides
the corresponding mechanisms of message asking, message a strong guarantee for improving the related functions of
selecting, message sharing, and punishment mechanism were smart cities, thus further accelerating the pace of urban intel-
presented based on smart contracts. In the in-vehicle self- ligence. As a network with extensive coverage, the smart grid
organizing environment, an advanced blockchain-authorized should realize the interaction with users, the intelligentiza-
distributed software-defined security architecture was delin- tion of power grid equipment, the full automation of power
eated [241]. Then, the dueling deep Q-learning algorithm with production, and the greening of energy, so as to comprehen-
prior experience playback was adopted in [241] to procure sively improve the level of informatization and intelligence of
the optimal strategy while satisfying the requirement of maxi- the power grid. Modern communication technology is fully
mizing the system throughput. In autonomous driving vehicle utilized to build a safe, reliable, green, and efficient smart
networks, the sharing and storage of massive video data is grid. The application of blockchain and AI to the smart grid
terribly difficult. To cope with these difficulties, [242] com- can improve the quality and efficiency of grid engineering,
bined blockchain and multi-access edge computing to build thereby enhancing the stability of the smart grid system.
a video analytics architecture in autonomous driving systems. For example, the work of [247] systematically discussed
This architecture completed the secure storage and sharing the enabling role of blockchain, AI and IoT in improv-
of video data with the help of smart contract. Then, the ing smart grid performance. In the environment of smart
study of [242] formulated the joint optimization problem of power, [248] demonstrated an advanced privacy-preserving
video offloading and resource allocation as a Markov decision scheme by integrating two emerging technologies, blockchain
process. Moreover, a high-level DRL algorithm with asyn- and DL, which consisted of a two-level privacy mechanism and
chronous advantage actor-critic was proposed to tackle this an anomaly detection mechanism. Specifically, the first-level
joint problem. Improper lanes for self-driving cars can cause privacy mechanism applied a blockchain found on an enhanced
tragic accidents, and thus, the authors of [243] demonstrated PoW consensus protocol to validate data integrity and dimin-
an autonomous lane changing system assisted by blockchain ish data poisoning attacks. To avoid inference attacks, the
and collective learning. Blockchain ensured data security for second-level privacy mechanism converted the raw data into
autonomous vehicles while encouraging vehicle resources to an encoded form with the assistance of a variational autoen-
join in collective learning. Then, an advanced algorithm based coder. Then, in [248], an LSTM-inspired anomaly detection
on the deep deterministic policy gradient was used in [243] to mechanism employed two public datasets to train and validate
address the lane changing problem, so as to achieve optimal the output of the two-level privacy mechanism. On the plat-
autonomous driving policies. form of smart grid, a power management system for electric
Interestingly, the fusion of blockchain and FL power- vehicles merging blockchain and AI was creatively demon-
fully polishes up the performance of intelligent transportation strated in [249]. The system applied artificial neural networks
systems [244], [245], [246]. For instance, on the basis of and FL to prophesy the electricity consumption of electric
blockchain-enhanced FL, the study in [244] proposed a fully vehicles. At the same time, the blockchain can incorporate all
distributed electric vehicles to form a smart energy storage this approach, FL only shared the data model uploaded by the
framework. The blockchain traded memory and time for the authenticated UAV instead of directly sharing the original data.
security and transparency performance of the proposed power The authors of [255] made full use of multi-signature smart
management system. To protect the smart grid from malicious contracts to practice distributed cross-domain UAVs’ identity
attacks, the study of [250] designed a blockchain- and DL- authentication. And, to surmount the single point failure, these
guided reliable energy exchange architecture called DeepCoin. multi-signature smart contracts were also employed to perform
In DeepCoin, the blockchain adopted the PBFT algorithm to aggregations of global model updates. In the IoT environment,
accomplish the consensus in the P2P energy system, thereby the work of [256] proposed a distributed dynamic resource
promoting users to voluntarily trade redundant energy to other management and pricing system that integrated blockchain-
adjacent users. In [250], the blockchain also applied bilinear as-a-service (BaaS) and UAV-authorized MEC. Specifically,
pairing, short signature, and hash function to complete the pri- MEC servers are installed on both the ground BSs and the
vacy protection of smart grid users. Then, DeepCoin tracked UAVs acting as the air BSs to process some blockchain
down cyber-attacks and deceitful transactions in smart grids tasks. BaaS combined blockchain and cloud computing so
through the RNN-based DL algorithm. that resource management and pricing can be handled on BSs
Recently, microgrids are small-scale power systems that uti- with MEC servers in [256]. Then, in the case of incomplete
lize renewable energy sources to distribute electricity near information, the interaction process of resource management
users. To effectuate the real-time scheduling of energy in and pricing between BSs and peers of the proposed system
microgrids, the authors of [251] provided a blockchain- and was expressed as a stochastic Stackelberg game with multiple
ML-assisted scheme for forecasting P2P energy transactions. leaders.
This energy transaction scheme was modeled and completed Additionally, UAV swarms or low earth orbit (LEO)
on the permissioned blockchain network entitled Hyperledger satellites are extremely vulnerable to security threats.
Fabric. At the same time, smart contracts performed real- Therefore, reference [257] elucidated a blockchain- and FL-
time scheduling of distributed energy and controllable loads. assisted knowledge sharing and collaborative learning scheme.
In addition, RNN, LSTM, and bidirectional LSTM-powered Specifically, this scheme considered the influence of the num-
ML algorithms are employed to forecast energy requirements ber of miners, block transfer, and the mobility of UAVs/LEOs,
in microgrids while downgrading electricity transportation the authors of [257] derived the probability of regular forks and
expenses. In [252], a distributed energy trading and man- optimized the energy consumption of PoW computation for
agement system for smart microgrids named FogChain was blocks. More importantly, in [257], an advanced FL-enabled
presented. FogChain adopted blockchain to formulate a decen- algorithm was used to complete the resource allocation of
tralized energy trading platform and applied edge-based AI mobile mining. And, the coordination gains of blockchain
methods to draw distributed controllers for microgrids. In and FL for UAV swarms was illustrated. The work of [258]
the same direction, the work in [253] combined blockchain integrated blockchain, AI, and UAV swarms to design an
and ML to address data sharing, processing, and forecasting autonomous detection framework for infectious diseases. Here,
problems in microgrid systems. UAV swarms can expand coverage and lessen human par-
4) UAV: In recent years, with the development of ticipation. This detection framework applied a lightweight
information technology, UAV has formed an intelligent aircraft blockchain and two-stage security authentication mechanism
that combines multiple technologies such as flight control, to remote areas where the network is scarce to degrade the
network communication, and electric power. UAVs can be burden of UAVs. Then, a DL-inspired algorithm was pro-
regarded as flying IoT devices, and have been widely applied vided to autonomously detect disease prevalence in [258].
in military, agriculture, forestry, transportation, meteorology, The security of RF signal transmission between UAVs and
and other fields. UAV has the advantages of low cost, high the accuracy of identification and detection are challenged.
dynamics, and deployment flexibility. However, UAV com- Consequently, a UAV recognition and detection architecture
munication faces many threats such as being susceptible by combining blockchain, deep DNN, and edge computing
to interference, inability to cover large areas, and unstable was elucidated in [259]. This architecture applied blockchain
communication. The combination of blockchain and AI pro- to protect the security of data transmission. Moreover, the deep
vides new research ideas for alleviating these threats. For DNN was adopted for training by using the collected RF signal
example, under the UAV Internet system, reference [254] intro- data from UAVs in different flight modes. Then, the trained
duced a security scheme for information transmission between model was downloaded to edge devices to identify UAVs and
UAVs, which integrated blockchain and DL. This scheme detect their flight modes. There are trade-off problems in terms
applied a zero-knowledge-proof-based blockchain to maintain of quantity, energy consumption, coverage area, and height
the security and privacy of data dissemination between UAVs. when installing BSs on the UAV side (UAV-BS). To address
Moreover, the DL-inspired miner selection algorithm in [254] this deployment problem of UAV-BS, the study of [260]
can obtain the optimal miner node strategy, thereby shorten- demonstrated a blockchain- and ML-guided smart placement
ing the block generation time and transaction submission time. scheme for UAV-BS. More interestingly, in 6G networks, the
To overcome the security challenges of cross-domain UAVs’ work in [261] combined blockchain and FL can provide a
authentication, a novel secure identity authentication approach new idea for UAV-assisted construction of disaster response
by leveraging blockchain-backed FL was provided in [255]. In systems.
C. Analysis of Operating Frequencies, Visions, and operating frequencies and does not establish specific frequency
Requirements From the 6G Perspective band ranges. As 6G technology continues to be researched
1) Operating Frequencies From the 6G Perspective: and developed, future studies and standardization efforts will
Considering the operating frequencies of AI and blockchain provide more specific and detailed guidance.
for 6G is crucial from 6G perspective, as 6G networks have 2) Visions From the 6G Perspective: The vision of 6G is
higher requirements for high-speed data transmission and to build a highly intelligent, highly connected, and highly
processing. However, there are currently no established stan- adaptive network to meet the needs of future society and
dards or fixed ranges for the operating frequencies of AI industries. From the 6G perspective, the visions include sev-
and blockchain in 6G. Since 6G technology is still in the eral key aspects. Ultra-high rates and ultra-low latency:
research and standardization phase, there is limited discussion The vision of 6G is to achieve higher data transmission rates
and research on this specific topic in the existing literature. and lower communication latency to support a wide range
Current works [2], [5], [6], [7], [8], [263], [264] primarily of applications, including enhanced mobile broadband, vir-
focus on the communication characteristics, spectrum range, tual and augmented reality, high-definition video, etc. Through
high-speed and low-latency network transmission, the data
and key technologies of 6G. The operating frequencies in 6G
processing and interaction capabilities of AI and blockchain
are expected to encompass a wide range of spectrum, includ-
will be enhanced, supporting more complex and real-time
ing low-frequency, mid-frequency, high-frequency, as well as
application scenarios. This will provide a stronger foundation
millimeter-wave and terahertz bands. However, the current
for integrated AI and blockchain applications. Super con-
work does not specifically address the operating frequencies
nectivity: The vision of 6G is to establish a super-connected
of AI and blockchain in 6G. Until further research and stan-
network that enables highly interconnected devices. The super-
dardization developments, specific discussions and research on
connected network includes D2D, device-to-infrastructure, and
the operating frequencies of AI and blockchain for 6G may
device-to-cloud connections. This will provide more connec-
remain limited.
tivity options and broader coverage for AI and blockchain
Although specific references may be limited, we can provide
services and applications. Powerful intelligence and adapt-
a general discussion on the operating frequencies of AI and
ability: The vision of 6G is to build an intelligent network
blockchain in 6G networks. Millimeter-wave and Terahertz
with edge computing and distributed intelligence capabilities.
Bands: 6G networks will utilize high-frequency millimeter-
This will enable real-time processing and decision-making of
wave and terahertz bands to achieve higher data transmission AI and blockchain technologies at the network edge, reducing
rates and lower latency. The use of these frequency bands latency and improving performance. Moreover, the intelligent
will impact the operating frequencies of AI and blockchain, network will be able to adaptively optimize based on applica-
requiring consideration of their performance and adaptabil- tion requirements, providing more efficient support for AI and
ity in these bands. Power Consumption and Resource blockchain services and applications. Security and privacy
Management: 6G networks demand high performance while protection: The vision of 6G is to ensure network secu-
minimizing power consumption. When determining the oper- rity and privacy protection to address the growing security
ating frequencies of AI and blockchain, a comprehensive threats. In the services and applications of AI and blockchain,
consideration of power consumption and resource management security and privacy protection are crucial. 6G will provide
is necessary to achieve efficient computation and communica- stronger security mechanisms, including encryption, iden-
tion. Adaptive Adjustment Strategies: Given the dynamic tity authentication, and access control, to ensure the secure
nature and diverse application requirements of 6G networks, and reliable operation of AI and blockchain services and
adaptive adjustment strategies are crucial for the operating applications.
frequencies of AI and blockchain. By continuously monitoring In summary, the vision of 6G is an evolving and devel-
and analyzing network conditions, application demands, and oping concept, and there is no unified definition or standard
resource availability, it is possible to dynamically adjust the yet. Therefore, our discussion is primarily based on current
operating frequencies of AI and blockchain to meet real-time research and academic discussions [2], [5], [7], [263] pro-
requirements and optimize network performance. vide a general 6G vision for blockchain and AI services and
The operating frequencies of AI and blockchain for 6G applications.
will be determined based on specific application scenarios and 3) Requirements From the 6G Perspective: To achieve the
requirements. For example, in the fields of IoT, UAV or smart 6G vision discussed above, it naturally leads to the demand
cities, different frequency bands may be used to support the for higher bandwidth, lower latency, super connectivity, high
6G secure services and 6G IoT smart applications by integrat- security, and high reliability in 6G. Specifically, in the topic
ing of AI and blockchain. The specific operating frequencies of blockchain and AI for 6G, the requirements of 6G include
will depend on communication requirements, device char- the following key aspects. High bandwidth demand: With the
acteristics, as well as the needed bandwidth and capacity. increasing adoption of AI and blockchain services and applica-
Therefore, from the 6G perspective, determining the operating tions, there is a growing need for higher bandwidth to support
frequencies of AI and blockchain requires considering multiple large-scale data transmission, real-time decision-making, and
factors and referencing the development of future 6G stan- complex computations. Low latency requirement: AI and
dards and the demand of practical application. Notably, the blockchain services and applications require real-time respon-
discussion provided above only offers general perspectives on siveness. To support these services and applications, 6G needs
to provide lower communication latency to ensure fast data purpose is to provide beneficial inspirations and references for
transmission and processing capabilities. Large-scale connec- future innovation research. We also explore potential research
tivity capability: With the proliferation of IoT devices and AI directions in the future.
applications, 6G needs to have the ability to connect a mas-
sive number of devices to facilitate interconnection and data A. Towards Blockchain
exchange. High security and privacy protection: Security
In recent years, the research and application of blockchain
and privacy are crucial in AI and blockchain services and
has begun to grow explosively. Blockchain is considered to
applications. 6G needs to provide robust security mechanisms,
be the key technology leading the current information Internet
including identity authentication, encryption, and secure trans-
to the value Internet. Although blockchain has great potential
mission, to ensure the confidentiality and integrity of data.
for 6G networks, there are still some challenges that limit its
High reliability and robustness: 6G should exhibit high reli-
widespread application in 6G networks.
ability and robustness to handle various network environments
As the number of transactions increases significantly, scal-
and cope with interferences and failures. This will ensure the
ability is the biggest hurdle limiting the widespread adoption
stability and reliability of AI and blockchain services and
of blockchain technology in 6G networks. For example, the
applications.
Bitcoin and Ethereum systems process an average of 7 to 20
In summary, the aforementioned requirements [5], [8],
transactions per second, far behind the Visa system that han-
[263], [264] are based on the 6G perspective of integrating
dles tens of thousands of transactions per second. Moreover,
AI and blockchain services and applications. As 6G technol-
the multi-copy feature of the blockchain requires a large
ogy continues to evolve and standardization efforts progress,
amount of additional storage space, which increases storage
these requirements may be further refined and supplemented.
costs. This will result in limited space utilization, making it
In this section, we have broadly discussed the services
difficult to support large-scale applications. There have been
and applications of merging blockchain and AI for 6G
many researchers improving the scalability of blockchains
networks. This includes the integration of blockchain and
by using techniques such as sidechains, lightning networks,
AI for 6G secure services, such as spectrum management,
sharding, pruning, and directed acyclic graphs. However, these
content caching, computation allocation, and security and
methods still have their own problems, such as how to properly
privacy. We have also examined the 6G IoT smart applica-
divide the tiles, and which transactions to prune. In addition,
tions of merging blockchain and AI, covering areas such as
the blockchain is a high-energy-consumption industry. As an
smart healthcare, smart transportation, smart grid, and UAV.
illustration, the blockchain system based on the PoW con-
Furthermore, we have thoroughly discussed the operating
sensus mechanism relies on the computing power contributed
frequencies, visions, and requirements from the 6G perspec-
by the blockchain nodes. However, only part of the computing
tive. In summary, this section highlights the complementary
power has been rewarded, and other computing power is doing
nature of blockchain and AI in addressing existing challenges
useless work, which wastes a lot of resources. This problem
and meeting the evolving requirements of the 6G era. The
of high energy consumption problem affects the popularization
integration of blockchain and AI in 6G networks demon-
and application of blockchain in 6G.
strates the potential to enhance secure services and enable
Security attack is the most important problem faced by
advanced IoT applications. Additionally, the analysis of oper-
the blockchain so far, such as Bitcoin’s 51% attack and bot-
ating frequencies, visions, and requirements provides valuable
net attack. The asymmetric encryption mechanism of the
insights for shaping the future of 6G networks. By studying
blockchain will become more and more fragile with the
the integration of blockchain and AI for 6G networks, we have
development of mathematics, cryptography, and computing
gained a deeper understanding of their synergistic capabilities
technology. Security issues are also a great threat to the further
and the opportunities they present for transforming wireless
application of blockchain to 6G networks. The data transac-
communications.
tions recorded on the blockchain are open and transparent,
which is beneficial for data sharing and verification, but not
IV. O PEN I SSUES , R ESEARCH C HALLENGES , conducive to the privacy protection of user information. As
AND F UTURE W ORK more and more personal data is stored in blockchain-powered
6G networks, privacy leakage becomes another key issue.
Integrating blockchain and AI in 6G wireless communica-
tions is currently a hot research topic. In the previous sections,
we focused on investigating the possibilities of combining B. Towards AI
blockchain and AI. Moreover, we extensively discussed the In recent years, AI, especially DL, has achieved great suc-
integration of blockchain and AI for wireless communica- cess in computer vision, natural language processing, speech
tions, involving secure services and IoT smart applications. recognition, and other fields. The researchers expect to apply
Research on integrating blockchain and AI for wireless com- AI to all levels of the 6G system, thereby generating an intelli-
munications is still emerging, but future works need to address gent communication system, realizing the true interconnection
some questions and challenges. In this section, based on exten- of everything, and meeting people’s ever-changing demands
sive research works in current literatures, we summarize the for data transmission rates. However, there are still many chal-
possible open issues and research challenges for integrat- lenges and unsolved issues in implementing and managing
ing blockchain and AI in 6G wireless communications. The complex intelligent communication systems.
Inefficient data management schemes and high overhead of blockchain and AI poses certain challenges for AI embed-
information exchange among communication participants are ded in blockchain to optimize the execution decisions of 6G
key bottlenecks in the development of AI technology. The AI- networks. Therefore, in future research, a new solution is
based solutions, such as ML methods, usually require large required to deal with the contradiction between the certainty
amounts of training data, which need to be collected and of smart contracts and the randomness of AI algorithms. The
implemented on a centralized server with sufficient storage new solution can handle approximate calculations for smart
and computing resources. Nevertheless, current wireless com- contracts and design consensus protocols for each participat-
munication systems do not have access to massive amounts of ing node of blockchain. The purpose of the new solution
data to train models. In heterogeneous networks, aggregating is to output the decision results under the 6G network with
data from different sources to train models is also an open specific certainty, high accuracy, and high precision. With
problem. In 6G networks, users may have different service the explosive growth of data in 6G networks, the process-
quality requirements in different scenarios. For example, in ing of large amounts of data in blockchain- and AI-assisted
video streaming applications, users demand high throughput wireless communication systems is a terribly large challenge.
and low latency at the cost of security. However, in payment Typically, blockchain is applied to securely collect and store
softwares, users require high security, even at the expense of large amounts of data, and AI uses these data for model
throughput. In this direction, designing a cross-layer, action- training processing. The massive unlabeled and unclassified
based AI protocol for different applications is a key issue to datasets are intractable for AI training. At the same time,
satisfy various service demands while balancing the network the blockchain also presents a potential bottlenecks in stor-
resources of 6G networks. ing these large-scale distributed data. For example, the data
Notably, in the current era of rapid development of AI, data recorded on the blockchain is open and transparent. The
security and privacy issues are receiving increasing attention. blockchain-based storage method is beneficial for data sharing
6G network realizes interconnected intelligence by supporting and verification, but not conducive to data privacy protec-
AI functions, and adopts a centralized network architecture, tion. Furthermore, at present, most works in related kinds
which is vulnerable to hacker attacks. Moreover, 6G network of literature often only optimize a single performance index
needs to collect a large amount of user data for training in a single wireless communication system. The collabora-
through billions of devices. The training data involves a large tive optimization of multiple performance indexes of multiple
amount of personal information, so AI can easily lead to pri- wireless communication systems is ignored. However, with the
vacy leakage of user data. Using distributed technology to maturity and in-depth research of blockchain and AI technolo-
design AI-enabled 6G network architectures can actualize a gies in the future, the solutions proposed for the performance
decentralized security and trustworthy mechanism. Without of different wireless communication systems can be combined
sending all data to the cloud computing center, the distributed with each other. Thereby, the goal of co-optimizing multiple
technology processes data where it is generated, which can performances of multiple systems can be achieved.
alleviate the problem of privacy leakage to a certain extent. Observing the existing literature, blockchain- and AI-
Nevertheless, communication needs to exchange the knowl- assisted wireless communications are still in the infant stage.
edge information perceived by both parties and update the Many works apply blockchain to create a trusted environ-
knowledge bases of both parties collaboratively, which also ment for wireless communications, and provide prediction,
leads to the risk of privacy leakage of local data. How to optimization, identification, detection, and decision-making for
develop an efficient coordination mechanism among commu- wireless communication systems through AI algorithms. Few
nication participants without causing any private data leakage works have really deeply integrated blockchain and AI into
remains an urgent problem to be solved. wireless communications. Whether we look at the current tech-
nical indicators of blockchains or the actual implementation of
AI and 6G networks, there are still many uncertainties to truly
C. Towards Blockchain- and AI-Assisted Wireless realize the integration and implementation of blockchain and
Communications AI technologies for 6G networks. The potential outcomes of
Integrating blockchain and AI brings new opportunities to the fusion of blockchain and AI for 6G networks are also
6G networks, as well as some open issues and research chal- difficult to assess. Therefore, while actively investigating the
lenges. For example, the natural conflict between blockchain integration of blockchain and AI for 6G networks, we must
and AI, the processing of a large amounts of data, and the also look at it rationally and focus on practical implementation.
collaborative optimization of multiple systems and multiple In the future, we will continue to take an organic combina-
indicators. Furthermore, the effectiveness and feasibility of tion and flexible and innovative approach to truly realize the
blockchain- and AI-assisted wireless communications still practice and exploration of the integration of blockchain and
need to be verified by large-scale practice of wireless com- AI in 6G wireless communications.
munication networks.
There are some conflicts with the combination of blockchain
and AI. For example, the execution results of smart contracts D. Future Work
in blockchains are often deterministic. While, the execu- In future work, we still need to tackle the technical barriers
tion results of AI algorithms are usually uncertain, random, of blockchain and AI. We can deeply investigate the match-
and unpredictable in most cases. The contradiction between ing and joint optimization of blockchain- and AI-powered 6G
networks in terms of performance indicators, security, stability, AI was conducted, spotlighting spectrum management, com-
etc. We can also further research and ensure the healthy and putation allocation, content caching, and security and privacy
sustainable development of blockchain and AI technologies services. In addition, we also covered the latest achievements
in 6G networks. In addition, the Federal Communications of blockchain and AI empowerment in various IoT smart
Commission, at the 2018 U.S. Mobile World Congress, applications. We made an exhaustive analysis from four sce-
emphasized that 6G can introduce blockchain technology narios: smart healthcare, smart transportation, smart grid, and
into spectrum sharing [265]. Research institutions such as UAV. Furthermore, we have thoroughly discussed operating
the Institute of Electrical and Electronics Engineers and the frequencies, visions, and requirements from the 6G perspec-
France’s Spectrum Regulator have also begun to explore the tive. Finally, we have pointed out several open issues, research
application of blockchain to manage spectrum [75]. Thus, we challenges, and potential research directions toward blockchain
will focus on the fusion of blockchain and AI to obtain smarter and AI for 6G networks.
and more distributed dynamic wireless resource allocation. 6G In summary, this survey attempts to comprehensively
can combine emerging advanced technologies such as cloud explore the technologies related to blockchain and AI for
computing, edge computing, and big data to promote the devel- wireless communications. Although the research on integrat-
opment of blockchain and AI technologies. Accordingly, the ing blockchain and AI for 6G networks is still in its infancy,
mutual promotion and integration of blockchain, AI, and 6G it is obvious that blockchain and AI will significantly uplift
is also one of the key research directions in the future. With the performance of various services and applications in 6G
the further improvement of blockchain and AI, academia and networks. We believe our study will shed valuable insights into
industry will continue to transform theory into technology the research challenges associated with blockchain- and AI-
and put it into practice. We believe that future integration of assisted 6G networks as well as motivate interested researchers
blockchain and AI will be more in-depth, and the scenarios and practitioners to devote more research efforts to this
applied to 6G networks will be more abundant. Deploying promising area.
Blockchain and AI in 6G networks will bring more surprises
and possibilities to our lives.
In this section, we have outlined the open issues and R EFERENCES
research challenges associated with integrating blockchain and [1] S. Dang, O. Amin, B. Shihada, and M.-S. Alouini, “What should 6G
AI in 6G wireless communications. Future work should focus be?” Nat. Electron., vol. 3, no. 1, pp. 20–29, Jan. 2020.
[2] P. Yang, Y. Xiao, M. Xiao, and S. Li, “6G wireless communications:
on innovative solutions for scalability and energy efficiency in Vision and potential techniques,” IEEE Netw., vol. 33, no. 4, pp. 70–75,
blockchain, as well as exploring techniques for security and Jul./Aug. 2019.
privacy protection. Additionally, efficient data management [3] F. Tariq, M. R. Khandaker, K.-K. Wong, M. A. Imran, M. Bennis, and
M. Debbah, “A speculative study on 6G,” IEEE Wireless Commun.,
schemes and cross-layer AI protocols need to be developed to vol. 27, no. 4, pp. 118–125, Aug. 2020.
address AI-related challenges. The integration of blockchain [4] W. Jiang, B. Han, M. A. Habibi, and H. D. Schotten, “The road towards
and AI for 6G networks is still in its early stages, and 6G: A comprehensive survey,” IEEE Open J. Commun. Soc., vol. 2,
pp. 334–366, 2021.
future research should aim to overcome technical barriers and [5] Z. Zhang et al., “6G wireless networks: Vision, requirements, architec-
uncertainties, promote mutual promotion and integration, and ture, and key technologies,” IEEE Veh. Technol. Mag., vol. 14, no. 3,
explore practical applications. Overall, addressing these chal- pp. 28–41, Sep. 2019.
lenges and furthering the integration of blockchain and AI in [6] D. C. Nguyen et al., “6G Internet of Things: A comprehensive survey,”
IEEE Internet Things J., vol. 9, no. 1, pp. 359–383, Jan. 2022.
6G networks will bring significant advancements to wireless [7] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems:
communications. Applications, trends, technologies, and open research problems,” IEEE
Netw., vol. 34, no. 3, pp. 134–142, May/Jun. 2019.
[8] M. Z. Chowdhury, M. Shahjalal, S. Ahmed, and Y. M. Jang, “6G wire-
less communication systems: Applications, requirements, technologies,
V. C ONCLUSION challenges, and research directions,” IEEE Open J. Commun. Soc.,
This survey reviewed the latest progress of blockchain and vol. 1, pp. 957–975, 2020.
[9] N. Kato, B. Mao, F. Tang, Y. Kawamoto, and J. Liu, “Ten challenges in
AI for 6G wireless communications. We began our com- advancing machine learning technologies toward 6G,” IEEE Wireless
prehensive survey with a basic overview of blockchain and Commun., vol. 27, no. 3, pp. 96–103, Jun. 2020.
AI. Specifically, we briefly described the concepts, charac- [10] H. Yang, A. Alphones, Z. Xiong, D. Niyato, J. Zhao, and K. Wu,
“Artificial-intelligence-enabled intelligent 6G networks,” IEEE Netw.,
teristics, and categories of blockchain and AI. The recent vol. 34, no. 6, pp. 272–280, Nov. 2020.
developments in applying blockchain and AI to wireless com- [11] S. Khan, A. Hussain, S. Nazir, F. Khan, A. Oad, and M. D. Alshehri,
munications, respectively, were also showcased. To thoroughly “Efficient and reliable hybrid deep learning-enabled model for con-
gestion control in 5G/6G networks,” Comput. Commun., vol. 182,
explore the possibility of combining blockchain and AI, we pp. 31–40, Jan. 2022.
started with two aspects of blockchain-assisted AI and AI- [12] J. Guo, Y. Zuo, C.-K. Wen, and S. Jin, “User-centric online gossip train-
aided blockchain. We also highlighted the motivations for ing for autoencoder-based CSI feedback,” IEEE J. Sel. Topics Signal
integrating blockchain with AI for 6G wireless communica- Process., vol. 16, no. 3, pp. 559–572, Apr. 2022.
[13] J. Guo, C.-K. Wen, and S. Jin, “CAnet: Uplink-aided downlink channel
tions. Next, we then extensively discussed the simultaneous acquisition in FDD massive MIMO using deep learning,” IEEE Trans.
deployment of blockchain and AI in 6G wireless commu- Commun., vol. 70, no. 1, pp. 199–214, Jan. 2022.
nication systems, involving secure services and IoT smart [14] M. Latva-Aho, K. Leppänen, F. Clazzer, and A. Munari (Univ.
Oulu, Oulu, Finland). Key Drivers and Research Challenges for
applications. In particular, a comprehensive exploration of the 6G Ubiquitous Wireless Intelligence (White Paper). (2019). [Online].
widely popular secure services supported by blockchain and Available: http://urn.fi/urn:isbn:9789526223544.
[15] (ITU, Geneva, Switzerland). Framework for Evaluating Intelligence [39] S. K. Singh, S. Rathore, and J. H. Park, “BlockIoTIntelligence:
Levels of Future Networks Including IMT-2020. (Feb. 2020). [Online]. A blockchain-enabled intelligent IoT architecture with artificial intelli-
Available: https://www.itu.int/rec/T-REC-Y.3173/en gence,” Future Gener. Comput. Syst., vol. 110, pp. 721–743, Sep. 2020.
[16] J. Leng, M. Zhou, J. L. Zhao, Y. Huang, and Y. Bian, “Blockchain [40] A. A. Hussain and F. Al-Turjman, “Artificial intelligence and
security: A survey of techniques and research directions,” IEEE Trans. blockchain: A review,” Trans. Emerg. Telecommun. Technol., vol. 32,
Services Comput., vol. 15, no. 4, pp. 2490–2510, Jul./Aug. 2022. no. 9, Apr. 2021, Art. no. e4268.
[17] M. Wang, T. Zhu, T. Zhang, J. Zhang, S. Yu, and W. Zhou, “Security [41] Q. Yang, Y. Zhao, H. Huang, Z. Xiong, J. Kang, and Z. Zheng, “Fusing
and privacy in 6G networks: New areas and new challenges,” Digit. blockchain and AI with metaverse: A survey,” IEEE Open J. Comput.
Commun. Netw., vol. 6, no. 3, pp. 281–291, Aug. 2020. Soc., vol. 3, pp. 122–136, 2022.
[18] M. Conti, E. S. Kumar, C. Lal, and S. Ruj, “A survey on security and [42] B. K. Mohanta, D. Jena, U. Satapathy, and S. Patnaik, “Survey on
privacy issues of bitcoin,” IEEE Commun. Surveys Tuts., vol. 20, no. 4, IoT security: Challenges and solution using machine learning, artifi-
pp. 3416–3452, 4th Quart., 2018. cial intelligence and blockchain technology,” Internet Things, vol. 11,
[19] T. Hewa, G. Gür, A. Kalla, M. Ylianttila, A. Bracken, and M. Liyanage, Sep. 2020, Art. no. 100227.
“The role of blockchain in 6G: Challenges, opportunities and research [43] P. Tagde et al., “Blockchain and artificial intelligence technology
directions,” in Proc. 2nd 6G Wireless Summit (6G SUMMIT), Levi, in e-health,” Environ. Sci. Pollution Res., vol. 28, pp. 52810–52831,
Finland, Mar. 2020, pp. 1–5. Sep. 2021.
[20] Y. Zhu, G. Zheng, and K.-K. Wong, “Blockchain-empowered decen- [44] S. S. Gill et al., “Transformative effects of IoT, blockchain and artificial
tralized storage in air-to-ground industrial networks,” IEEE Trans. Ind. intelligence on cloud computing: Evolution, vision, trends and open
Informat., vol. 15, no. 6, pp. 3593–3601, Jun. 2019. challenges,” Internet Things, vol. 8, Dec. 2019, Art. no. 100118.
[21] M. Ylianttila et al., “6G white paper: Research challenges for trust, [45] A. D. Dwivedi, R. Singh, K. Kaushik, R. R. Mukkamala, and
security and privacy,” 2020, arXiv:2004.11665. W. S. Alnumay, “Blockchain and artificial intelligence for 5G-enabled
[22] J. Wang, X. Ling, Y. Le, Y. Huang, and X. You, “Blockchain-enabled Internet of Things: Challenges, opportunities, and solutions,” Trans.
wireless communications: A new paradigm towards 6G,” Nat. Sci. Rev., Emerg. Telecommun. Technol., Jul. 2021, Art. no. e4329. [Online].
vol. 8, no. 9, Sep. 2021, Art. no. nwab069. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/ett.4329
[23] M. Wu, K. Wang, X. Cai, S. Guo, M. Guo, and C. Rong, “A com- [46] Y. Liu, F. R. Yu, X. Li, H. Ji, and V. C. M. Leung, “Blockchain and
prehensive survey of blockchain: From theory to IoT applications machine learning for communications and networking systems,” IEEE
and beyond,” IEEE Internet Things J., vol. 6, no. 5, pp. 8114–8154, Commun. Surveys Tuts., vol. 22, no. 2, pp. 1392–1431, 2nd Quart.,
Oct. 2019. 2020.
[24] D. C. Nguyen, P. N. Pathirana, M. Ding, and A. Seneviratne, [47] F. Jameel, U. Javaid, W. U. Khan, M. N. Aman, H. Pervaiz,
“Blockchain for 5G and beyond networks: A state of the art survey,” and R. Jäntti, “Reinforcement learning in blockchain-enabled IIoT
J. Netw. Comput. Appl., vol. 166, Sep. 2020, Art. no. 102693. networks: A survey of recent advances and open challenges,”
[25] K. Yue et al., “A survey of decentralizing applications via blockchain: Sustainability, vol. 12, no. 12, p. 5161, Jun. 2020.
The 5G and beyond perspective,” IEEE Commun. Surveys Tuts., vol. 23, [48] Y. Wu, Z. Wang, Y. Ma, and V. C. Leung, “Deep reinforcement learning
no. 4, pp. 2191–2217, 4th Quart., 2021. for blockchain in Industrial IoT: A survey,” Comput. Netw., vol. 191,
[26] M. Tahir, M. H. Habaebi, M. Dabbagh, A. Mughees, A. Ahad, and May 2021, Art. no. 108004.
K. I. Ahmed, “A review on application of blockchain in 5G and beyond [49] A. Miglani and N. Kumar, “Blockchain management and machine
networks: Taxonomy, field-trials, challenges and opportunities,” IEEE learning adaptation for IoT environment in 5G and beyond networks:
Access, vol. 8, pp. 115876–115904, 2020. A systematic review,” Comput. Commun., vol. 178, pp. 37–63,
[27] S. A. Bhat, I. B. Sofi, and C.-Y. Chi, “Edge computing and its con- Oct. 2021.
vergence with blockchain in 5G and beyond: Security, challenges, and [50] A. El Azzaoui, S. K. Singh, Y. Pan, and J. H. Park, “Block5GIntell:
opportunities,” IEEE Access, vol. 8, pp. 205340–205373, 2020. Blockchain for AI-enabled 5G networks,” IEEE Access, vol. 8,
[28] P. Sharma, S. Jain, S. Gupta, and V. Chamola, “Role of machine pp. 145918–145935, 2020.
learning and deep learning in securing 5G-driven Industrial IoT [51] M. Dibaei et al., “Investigating the prospect of leveraging blockchain
applications,” Ad Hoc Netw., vol. 123, Dec. 2021, Art. no. 102685. and machine learning to secure vehicular networks: A survey,” IEEE
[29] Y. Sun, J. Liu, J. Wang, Y. Cao, and N. Kato, “When machine learning Trans. Intell. Transp. Syst., vol. 23, no. 2, pp. 683–700, Feb. 2022.
meets privacy in 6G: A survey,” IEEE Commun. Surveys Tuts., vol. 22, [52] Y. Yuan, T. Zhou, A. Zhou, Y. Duan, and F. Wang, “Blockchain
no. 4, pp. 2694–2724, 4th Quart., 2020. technology: From data intelligence to knowledge automation,” Acta
[30] V. P. Rekkas, S. Sotiroudis, P. Sarigiannidis, S. Wan, Automatica Sinica, vol. 43, no. 9, pp. 1485–1490, Sept. 2017.
G. K. Karagiannidis, and S. K. Goudos, “Machine learning in beyond [53] S. Nakamoto. “Bitcoin: A peer-to-peer electronic cash
5G/6G networks—State-of-the-art and future trends,” Electronics, system.” Accessed: May 2008. [Online]. Available:
vol. 10, no. 22, p. 2786, Nov. 2021. http://nakamotoinstitute.org/bitcoin/
[31] Y. Liu, X. Yuan, Z. Xiong, J. Kang, X. Wang, and D. Niyato, [54] S. Aggarwal and N. Kumar, “Blockchain 2.0: Smart contracts,” in
“Federated learning for 6G communications: Challenges, methods, Advances in Computers, vol. 121. Amsterdam, The Netherlands:
and future directions,” China Commun., vol. 17, no. 9, pp. 105–118, Elsevier, 2021, pp. 301–322.
Sep. 2020. [55] D. D. F. Maesa and P. Mori, “Blockchain 3.0 applications survey,”
[32] M. Lin and Y. Zhao, “Artificial intelligence-empowered resource J. Parallel Distrib. Comput., vol. 138, pp. 99–114, Apr. 2020.
management for future wireless communications: A survey,” China [56] F. Tschorsch and B. Scheuermann, “Bitcoin and beyond: A technical
Commun., vol. 17, no. 3, pp. 58–77, Mar. 2020. survey on decentralized digital currencies,” IEEE Commun. Surveys
[33] K. Salah, M. H. U. Rehman, N. Nizamuddin, and A. Al-Fuqaha, Tuts., vol. 18, no. 3, pp. 2084–2123, 3rd Quart., 2016.
“Blockchain for AI: Review and open research challenges,” IEEE [57] W. Wang et al., “A survey on consensus mechanisms and mining
Access, vol. 7, pp. 10127–10149, 2019. strategy management in blockchain networks,” IEEE Access, vol. 7,
[34] M. Shafay, R. W. Ahmad, K. Salah, I. Yaqoob, R. Jayaraman, and pp. 22328–22370, 2019.
M. Omar, “Blockchain for deep learning: Review and open challenges,” [58] A. Kiayias, A. Russell, B. David, and R. Oliynykov, “Ouroboros:
Clust. Comput., vol. 26, pp. 197–221, Mar. 2022. A provably secure proof-of-stake blockchain protocol,” in Proc. 37th
[35] R. Wang, M. Luo, Y. Wen, L. Wang, K.-K. R. Choo, and D. He, “The Annu. Int. Cryptol. Conf. (CRYPTO), Aug. 2017, pp. 357–388.
applications of blockchain in artificial intelligence,” Security Commun. [59] D. Larimer, “Delegated proof of stake (DPoS),” Bitshare, Blacksburg,
Netw., vol. 2021, Sep. 2021, Art. no. 6126247. [Online]. Available: VA, USA, Rep. , Apr. 2014. [Online]. Available: https://how.bitshares.
https://www.hindawi.com/journals/scn/2021/6126247/ works/en/master/technology/dpos.html#
[36] B. Xing and T. Marwala, “The synergy of blockchain and artificial [60] M. Castro and B. Liskov, “Practical Byzantine fault tolerance and
intelligence,” in Pro. SSRN, Aug. 2018, Art. no. 3225357. proactive recovery,” ACM Trans. Comput. Syst., vol. 20, no. 4,
[37] K. D. Pandl, S. Thiebes, M. Schmidt-Kraepelin, and A. Sunyaev, “On pp. 398–461, Nov. 2002.
the convergence of artificial intelligence and distributed ledger tech- [61] G. Wood et al., Ethereum: A Secure Decentralised Generalised
nology: A scoping review and future research agenda,” IEEE Access, Transaction Ledger, Ethereum, Zug, Switzerland, 2014.
vol. 8, pp. 57075–57095, 2020. [62] O. Dib, K.-L. Brousmiche, A. Durand, E. Thea, and E. B. Hamida,
[38] T. N. Dinh and M. T. Thai, “AI and blockchain: A disruptive “Consortium blockchains: Overview, applications and challenges,” Int.
integration,” Computer, vol. 51, no. 9, pp. 48–53, Sep. 2018. J. Adv. Telecommun., vol. 11, nos. 1–2, pp. 51–64, 2018.
[63] Y. Zuo, S. Jin, S. Zhang, and Y. Zhang, “Blockchain storage and [86] J. Sun, X. Yao, S. Wang, and Y. Wu, “Blockchain-based secure stor-
computation offloading for cooperative mobile-edge computing,” IEEE age and access scheme for electronic medical records in IPFS,” IEEE
Internet Things J., vol. 8, no. 11, pp. 9084–9098, Jun. 2021. Access, vol. 8, pp. 59389–59401, 2020.
[64] X. Ling, Y. Le, J. Wang, Z. Ding, and X. Gao, “Practical modeling and [87] Y. Xu, J. Ren, Y. Zhang, C. Zhang, B. Shen, and Y. Zhang, “Blockchain
analysis of blockchain radio access network,” IEEE Trans. Commun., empowered arbitrable data auditing scheme for network storage as a
vol. 69, no. 2, pp. 1021–1037, Feb. 2021. service,” IEEE Trans. Services Comput., vol. 13, no. 2, pp. 289–300,
[65] Y. Zuo, S. Jin, and S. Zhang, “Computation offloading in untrusted Mar./Apr. 2020.
MEC-aided mobile blockchain IoT systems,” IEEE Trans. Wireless [88] K. Fan, Y. Ren, Y. Wang, H. Li, and Y. Yang, “Blockchain-based effi-
Commun., vol. 20, no. 12, pp. 8333–8347, Dec. 2021. cient privacy preserving and data sharing scheme of content-centric
[66] Y. Zuo, S. Jin, S. Zhang, Y. Han, and K.-K. Wong, “Delay-limited network in 5G,” IET Commun., vol. 12, no. 5, pp. 527–532, Feb. 2018.
computation offloading for MEC-assisted mobile blockchain networks,” [89] A. E. Gamal and H. E. Gamal, “A single coin monetary mechanism
IEEE Trans. Commun., vol. 69, no. 12, pp. 8569–8584, Dec. 2021. for distributed cooperative interference management,” IEEE Wireless
[67] X. Ling, P. Chen, J. Wang, and Z. Ding, “Data broker: Dynamic Commun. Lett., vol. 8, no. 3, pp. 757–760, Jun. 2019.
multi-hop routing protocol in blockchain radio access network,” IEEE [90] Y. Sun, L. Zhang, G. Feng, B. Yang, B. Cao, and M. A. Imran,
Commun. Lett., vol. 25, no. 12, pp. 4000–4004, Dec. 2021. “Blockchain-enabled wireless Internet of Things: Performance analysis
and optimal communication node deployment,” IEEE Internet Things
[68] Y. Zuo, S. Jin, and S. Zhang, “Blockchain storage, computation offload-
J., vol. 6, no. 3, pp. 5791–5802, Jun. 2019.
ing, and user association for heterogeneous cellular networks,” IEEE
[91] Z. Liu, L. Gao, Y. Liu, X. Guan, K. Ma, and Y. Wang, “Efficient QoS
Internet Things J., vol. 9, no. 11, pp. 8191–8204, Jun. 2022.
support for robust resource allocation in blockchain-based femtocell
[69] Y. Zuo, S. Zhang, Y. Han, and S. Jin, “Computation resource alloca-
networks,” IEEE Trans. Ind. Informat., vol. 16, no. 11, pp. 7070–7080,
tion in mobile blockchain-enabled edge computing networks,” in Proc.
Nov. 2020.
IEEE/CIC Int. Conf. Commun. China (ICCC), Aug. 2020, pp. 617–622.
[92] A. P. Singh et al., “A novel patient-centric architectural framework
[70] Y. Zuo, S. Jin, and S. Zhang, “Computation offloading and user asso- for blockchain-enabled healthcare applications,” IEEE Trans. Ind.
ciation for blockchain-enabled heterogeneous cellular networks,” in Informat., vol. 17, no. 8, pp. 5779–5789, Aug. 2021.
Proc. IEEE 94th Veh. Technol. Conf. (VTC-Fall), Norman, OK, USA, [93] D. C. Nguyen, P. N. Pathirana, M. Ding, and A. Seneviratne,
Sep. 2021, pp. 1–6. “BEdgeHealth: A decentralized architecture for edge-based IoMT
[71] K. Kotobi and S. G. Bilén, “Blockchain-enabled spectrum access in networks using blockchain,” IEEE Internet Things J., vol. 8, no. 14,
cognitive radio networks,” in Proc. Wireless Telecommun. Symp. (WTS), pp. 11743–11757, Jul. 2021.
Apr. 2017, pp. 1–6. [94] A. Yazdinejad, G. Srivastava, R. M. Parizi, A. Dehghantanha,
[72] K. Kotobi and S. G. Bilen, “Secure blockchains for dynamic spectrum K.-K. R. Choo, and M. Aledhari, “Decentralized authentication of
access: A decentralized database in moving cognitive radio networks distributed patients in hospital networks using blockchain,” IEEE J.
enhances security and user access,” IEEE Veh. Technol. Mag., vol. 13, Biomed. Health Inform., vol. 24, no. 8, pp. 2146–2156, Aug. 2020.
no. 1, pp. 32–39, Mar. 2018. [95] S. Xia, F. Lin, Z. Chen, C. Tang, Y. Ma, and X. Yu, “A Bayesian game
[73] S. Bayhan, A. Zubow, and A. Wolisz, “SPASS: Spectrum sensing as based vehicle-to-vehicle electricity trading scheme for blockchain-
a service via smart contracts,” in Proc. IEEE Int. Symp. Dyn. Spectr. enabled Internet of Vehicles,” IEEE Trans. Veh. Technol., vol. 69, no. 7,
Access Netw. (DySPAN), Oct. 2018, pp. 1–10. pp. 6856–6868, Jul. 2020.
[74] S. Bayhan, A. Zubow, P. Gawlowicz, and A. Wolisz, “Smart contracts [96] Y. Yin, Y. Li, B. Ye, T. Liang, and Y. Li, “A blockchain-based incre-
for spectrum sensing as a service,” IEEE Trans. Cogn. Commun. Netw., mental update supported data storage system for intelligent vehicles,”
vol. 5, no. 3, pp. 648–660, Sep. 2019. IEEE Trans. Veh. Technol., vol. 70, no. 5, pp. 4880–4893, May 2021.
[75] M. B. H. Weiss, K. Werbach, D. C. Sicker, and C. E. C. Bastidas, “On [97] D. Chattaraj, B. Bera, A. K. Das, S. Saha, P. Lorenz, and Y. Park,
the application of blockchains to spectrum management,” IEEE Trans. “Block-CLAP: Blockchain-assisted certificateless key agreement pro-
Cogn. Commun. Netw., vol. 5, no. 2, pp. 193–205, Jun. 2019. tocol for Internet of Vehicles in smart transportation,” IEEE Trans. Veh.
[76] J. Qiu, D. Grace, G. Ding, J. Yao, and Q. Wu, “Blockchain-based Technol., vol. 70, no. 8, pp. 8092–8107, Aug. 2021.
secure spectrum trading for unmanned-aerial-vehicle-assisted cellular [98] J. Wang, L. Wu, K.-K. R. Choo, and D. He, “Blockchain-based
networks: An operator’s perspective,” IEEE Internet Things J., vol. 7, anonymous authentication with key management for smart grid edge
no. 1, pp. 451–466, Jan. 2020. computing infrastructure,” IEEE Trans. Ind. Informat., vol. 16, no. 3,
[77] M. Grissa, A. A. Yavuz, and B. Hamdaoui, “TrustSAS: A trustworthy pp. 1984–1992, Mar. 2020.
spectrum access system for the 3.5 GHz CBRS band,” in Proc. IEEE [99] B. Bera, S. Saha, A. K. Das, and A. V. Vasilakos, “Designing
INFOCOM Conf. Comput. Commun., Apr. 2019, pp. 1495–1503. blockchain-based access control protocol in IoT-enabled smart-grid
[78] T. Maksymyuk, J. Gazda, L. Han, and M. Jo, “Blockchain-based intel- system,” IEEE Internet Things J., vol. 8, no. 7, pp. 5744–5761,
ligent network management for 5G and beyond,” in Proc. 3rd Int. Conf. Apr. 2021.
Adv. Inf. Commun. Technol. (AICT), Jul. 2019, pp. 36–39. [100] A. Jindal, G. S. Aujla, N. Kumar, and M. Villari, “GUARDIAN:
Blockchain-based secure demand response management in smart grid
[79] H. Zhang, S. Leng, and H. Chai, “A blockchain enhanced dynamic
system,” IEEE Trans. Services Comput., vol. 13, no. 4, pp. 613–624,
spectrum sharing model based on proof-of-strategy,” in Proc. IEEE
Jul./Aug. 2020.
Int. Conf. Commun. (ICC), Jun. 2020, pp. 1–6.
[101] K. Gai, Y. Wu, L. Zhu, K.-K. R. Choo, and B. Xiao, “Blockchain-
[80] M. Jiang, Y. Li, Q. Zhang, G. Zhang, and J. Qin, “Decentralized
enabled trustworthy group communications in UAV networks,” IEEE
blockchain-based dynamic spectrum acquisition for wireless downlink
Trans. Intell. Transp. Syst., vol. 22, no. 7, pp. 4118–4130, Jul. 2021.
communications,” IEEE Trans. Signal Process., vol. 69, pp. 986–997,
[102] Y. Tan, J. Liu, and N. Kato, “Blockchain-based key management for
2021.
heterogeneous flying ad hoc network,” IEEE Trans. Ind. Informat.,
[81] Z. Li, Z. Yang, S. Xie, W. Chen, and K. Liu, “Credit-based payments vol. 17, no. 11, pp. 7629–7638, Nov. 2021.
for fast computing resource trading in edge-assisted Internet of Things,” [103] L. Jiang, B. Chen, S. Xie, S. Maharjan, and Y. Zhang, “Incentivizing
IEEE Internet Things J., vol. 6, no. 4, pp. 6606–6617, Aug. 2019. resource cooperation for blockchain empowered wireless power trans-
[82] D. Chatzopoulos, M. Ahmadi, S. Kosta, and P. Hui, “FlopCoin: A cryp- fer in UAV networks,” IEEE Trans. Veh. Technol., vol. 69, no. 12,
tocurrency for computation offloading,” IEEE Trans. Mobile Comput., pp. 15828–15841, Dec. 2020.
vol. 17, no. 5, pp. 1062–1075, May 2018. [104] J. Guo, C.-K. Wen, S. Jin, and G. Y. Li, “Overview of deep
[83] W. Sun, J. Liu, Y. Yue, and P. Wang, “Joint resource allocation learning-based CSI feedback in massive MIMO systems,” IEEE Trans.
and incentive design for blockchain-based mobile edge computing,” Commun., vol. 70, no. 12, pp. 8017–8045, Dec. 2022.
IEEE Trans. Wireless Commun., vol. 19, no. 9, pp. 6050–6064, [105] N. J. Nilsson, The Quest for Artificial Intelligence. Cambridge, U.K.:
Sep. 2020. Cambridge Univ. Press, 2009.
[84] S. Wang, D. Ye, X. Huang, R. Yu, Y. Wang, and Y. Zhang, “Consortium [106] P. H. Winston, Artificial Intelligence. London, U.K.: Addison-Wesley,
blockchain for secure resource sharing in vehicular edge computing: A 1992.
contract-based approach,” IEEE Trans. Netw. Sci. Eng., vol. 8, no. 2, [107] A. Barr, E. A. Feigenbaum, and P. R. Cohen, The Handbook of Artificial
pp. 1189–1201, Apr./Jun. 2021. Intelligence, vol. 1. New York, NY, USA: William Kaufmann, 1981.
[85] J. Kang et al., “Blockchain for secure and efficient data sharing in [108] Q. Mao, F. Hu, and Q. Hao, “Deep learning for intelligent wireless
vehicular edge computing and networks,” IEEE Internet Things J., networks: A comprehensive survey,” IEEE Commun. Surveys Tuts.,
vol. 6, no. 3, pp. 4660–4670, Jun. 2019. vol. 20, no. 4, pp. 2595–2621, 4th Quart., 2018.
[109] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, [135] T. Gruber, S. Cammerer, J. Hoydis, and S. T. Brink, “On deep learning-
no. 7553, pp. 436–444, May 2015. based channel decoding,” in Proc. IEEE 51st Annu. Conf. Inf. Sci. Syst.
[110] J. Zhou et al., “Multiscale network traffic prediction method based on (CISS), Baltimore, MD, USA, Mar. 2017, pp. 1–6.
deep echo-state network for Internet of Things,” IEEE Internet Things [136] S. Cammerer, T. Gruber, J. Hoydis, and S. T. Brink, “Scaling deep
J., vol. 9, no. 21, pp. 21862–21874, Nov. 2022. learning-based decoding of polar codes via partitioning,” in Proc. IEEE
[111] I. J. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Global Commun. Conf. (GLOBECOM), Singapore, Dec. 2017, pp. 1–6.
Cambridge, MA, USA: MIT Press, 2016. [137] F. Liang, C. Shen, and F. Wu, “An iterative BP-CNN architecture for
[112] L. Bottou and O. Bousquet, “The tradeoffs of large scale learning,” in channel decoding,” IEEE J. Sel. Topics Signal Process., vol. 12, no. 1,
Proc. Adv. Neural Inf. Process. Syst., vol. 20, 2007, pp. 161–168. pp. 144–159, Feb. 2018.
[113] G. E. Hinton, S. Osindero, and Y.-W. Teh, “A fast learning algorithm [138] E. Nachmani, Y. Be’ery, and D. Burshtein, “Learning to decode linear
for deep belief nets,” Neural Comput., vol. 18, no. 7, pp. 1527–1554, codes using deep learning,” in Proc. IEEE 54th Annu. Allerton Conf.
Jul. 2006. Commun. Control Comput. (Allerton), Monticello, IL, USA, Sep. 2016,
[114] Q. Xue et al., “Beam management in ultra-dense mmWave network via pp. 341–346.
federated reinforcement learning: An intelligent and secure approach,” [139] E. Nachmani, E. Marciano, L. Lugosch, W. J. Gross, D. Burshtein,
IEEE Trans. Cogn. Commun. Netw., vol. 9, no. 1, pp. 185–197, and Y. Be’ery, “Deep learning methods for improved decoding of linear
Feb. 2023. codes,” IEEE J. Sel. Topics Signal Process., vol. 12, no. 1, pp. 119–131,
[115] K. Arulkumaran, M. P. Deisenroth, M. Brundage, and A. A. Bharath, Feb. 2018.
“Deep reinforcement learning: A brief survey,” IEEE Signal Process.
[140] S. Dörner, S. Cammerer, J. Hoydis, and S. T. Brink, “Deep learning
Mag., vol. 34, no. 6, pp. 26–38, Nov. 2017.
based communication over the air,” IEEE J. Sel. Topics Signal Process.,
[116] Y. LeCun et al., “Handwritten digit recognition with a back-propagation
vol. 12, no. 1, pp. 132–143, Feb. 2017.
network,” in Proc. Adv. Neural Inf. Process. Syst., vol. 2, 1989,
pp. 396–404. [141] H. Ye, G. Y. Li, B.-H. F. Juang, and K. Sivanesan, “Channel agnos-
[117] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learn- tic end-to-end learning based communication systems with conditional
ing applied to document recognition,” Proc. IEEE, vol. 86, no. 11, GAN,” in Proc. IEEE Globecom Workshops (GC Wkshps), Abu Dhabi,
pp. 2278–2324, Nov. 1998. UAE, Dec. 2018, pp. 1–5.
[118] Z. C. Lipton, J. Berkowitz, and C. Elkan, “A critical review of recurrent [142] M. Damrath and P. A. Hoeher, “Low-complexity adaptive threshold
neural networks for sequence learning,” 2015, arXiv:1506.00019. detection for molecular communication,” IEEE Trans. Nanobiosci.,
[119] I. J. Goodfellow et al., “Generative adversarial nets,” in Proc. Adv. vol. 15, no. 3, pp. 200–208, Apr. 2016.
Neural Inf. Process. Syst., vol. 27, 2014, pp. 1–9. [143] N. Farsad and A. Goldsmith, “Sliding bidirectional recurrent neural
[120] J.-M. Kang, C.-J. Chun, and I.-M. Kim, “Deep-learning-based channel networks for sequence detection in communication systems,” in Proc.
estimation for wireless energy transfer,” IEEE Commun. Lett., vol. 22, IEEE Int. Conf. Acoust. Speech Signal Process. (ICASSP), Calgary,
no. 11, pp. 2310–2313, Nov. 2018. AB, Canada, Apr. 2018, pp. 2331–2335.
[121] H. Huang, J. Yang, H. Huang, Y. Song, and G. Gui, “Deep learn- [144] S. Mohamed et al., “Model-based: End-to-end molecular communica-
ing for super-resolution channel estimation and DOA estimation based tion system through deep reinforcement learning auto encoder,” IEEE
massive MIMO system,” IEEE Trans. Veh. Technol., vol. 67, no. 9, Access, vol. 7, pp. 70279–70286, 2019.
pp. 8549–8560, Sep. 2018. [145] Y. Zhao, I. G. Niemegeers, and S. H. De Groot, “Power allocation
[122] H. He, C.-K. Wen, S. Jin, and G. Y. Li, “Deep learning-based channel in cell-free massive MIMO: A deep learning method,” IEEE Access,
estimation for beamspace mmWave massive MIMO systems,” IEEE vol. 8, pp. 87185–87200, 2020.
Wireless Commun. Lett., vol. 7, no. 5, pp. 852–855, Oct. 2018. [146] C. D’Andrea, A. Zappone, S. Buzzi, and M. Debbah, “Uplink power
[123] M. Borgerding, P. Schniter, and S. Rangan, “AMP-inspired deep control in cell-free massive MIMO via deep learning,” in Proc.
networks for sparse linear inverse problems,” IEEE Trans. Signal IEEE 8th Int. Workshop Comput. Adv. Multisensor Adapt. Process.
Process., vol. 65, no. 16, pp. 4293–4308, Aug. 2017. (CAMSAP), Dec. 2019, pp. 554–558.
[124] D. Neumann, T. Wiese, and W. Utschick, “Learning the MMSE [147] N. Rajapaksha, K. S. Manosha, N. Rajatheva, and M. Latva-Aho, “Deep
channel estimator,” IEEE Trans. Signal Process., vol. 66, no. 11, learning-based power control for cell-free massive MIMO networks,”
pp. 2905–2917, Jun. 2018. in Proc. IEEE Int. Conf. Commun. (ICC), Montreal, QC, Canada,
[125] H. Ye, G. Y. Li, and B.-H. Juang, “Power of deep learning for chan- Jun. 2021, pp. 1–7.
nel estimation and signal detection in OFDM systems,” IEEE Wireless [148] W. Cui, K. Shen, and W. Yu, “Spatial deep learning for wire-
Commun. Lett., vol. 7, no. 1, pp. 114–117, Feb. 2018. less scheduling,” IEEE J. Sel. Areas Commun., vol. 37, no. 6,
[126] N. Samuel, T. Diskin, and A. Wiesel, “Deep MIMO detection,” in pp. 1248–1261, Jun. 2019.
Proc. IEEE 18th Int. Workshop Signal Process. Adv. Wireless Commun. [149] S. Xu, P. Liu, R. Wang, and S. S. Panwar, “Realtime scheduling and
(SPAWC), Jul. 2017, pp. 1–5. power allocation using deep neural networks,” in Proc. IEEE Wireless
[127] H. He, C.-K. Wen, S. Jin, and G. Y. Li, “A model-driven deep learning Commun. Netw. Conf. (WCNC), Marrakesh, Morocco, Apr. 2019,
network for MIMO detection,” in Proc. IEEE Global Conf. Signal Inf. pp. 1–5.
Process. (GlobalSIP), Anaheim, CA, USA, Nov. 2018, pp. 584–588.
[150] X. Li, J. Fang, W. Cheng, H. Duan, Z. Chen, and H. Li, “Intelligent
[128] J. Liao, J. Zhao, F. Gao, and G. Y. Li, “A model-driven deep learning
power control for spectrum sharing in cognitive radios: A deep rein-
method for massive MIMO detection,” IEEE Commun. Lett., vol. 24,
forcement learning approach,” IEEE Access, vol. 6, pp. 25463–25473,
no. 8, pp. 1724–1728, Aug. 2020.
2018.
[129] Q. Cao, F. Li, T. Li, W. Ji, and Y. Liang, “Adaptive signal detec-
tion method based on model-driven for massive MIMO systems,” in [151] Y. S. Nasir and D. Guo, “Multi-agent deep reinforcement learning for
Proc. IEEE 13th Int. Conf. Wireless Commun. Signal Process. (WCSP), dynamic power allocation in wireless networks,” IEEE J. Sel. Areas
Changsha, China, Oct. 2021, pp. 1–5. Commun., vol. 37, no. 10, pp. 2239–2250, Oct. 2019.
[130] C.-K. Wen, W.-T. Shih, and S. Jin, “Deep learning for massive [152] J. Tan, Y.-C. Liang, L. Zhang, and G. Feng, “Deep reinforcement learn-
MIMO CSI feedback,” IEEE Wireless Commun. Lett., vol. 7, no. 5, ing for joint channel selection and power control in D2D networks,”
pp. 748–751, Oct. 2018. IEEE Trans. Wireless Commun., vol. 20, no. 2, pp. 1363–1378,
[131] T. Wang, C.-K. Wen, S. Jin, and G. Y. Li, “Deep learning-based CSI Feb. 2021.
feedback approach for time-varying massive MIMO channels,” IEEE [153] A. Sadeghi, G. Wang, and G. B. Giannakis, “Deep reinforcement learn-
Wireless Commun. Lett., vol. 8, no. 2, pp. 416–419, Apr. 2019. ing for adaptive caching in hierarchical content delivery networks,”
[132] Z. Liu, L. Zhang, and Z. Ding, “Exploiting bi-directional channel reci- IEEE Trans. Cogn. Commun. Netw., vol. 5, no. 4, pp. 1024–1033,
procity in deep learning for low rate massive MIMO CSI feedback,” Dec. 2019.
IEEE Wireless Commun. Lett., vol. 8, no. 3, pp. 889–892, Jun. 2019. [154] J. Ren, H. Wang, T. Hou, S. Zheng, and C. Tang, “Federated
[133] J. Guo, J. Wang, C.-K. Wen, S. Jin, and G. Y. Li, “Compression and learning-based computation offloading optimization in edge computing-
acceleration of neural networks for communications,” IEEE Wireless supported Internet of Things,” IEEE Access, vol. 7, pp. 69194–69201,
Commun., vol. 27, no. 4, pp. 110–117, Aug. 2020. 2019.
[134] P. K. Sangdeh, H. Pirayesh, A. Mobiny, and H. Zeng, “LB-SciFi: [155] X. Wang, Y. Han, C. Wang, Q. Zhao, X. Chen, and M. Chen, “In-edge
Online learning-based channel feedback for MU-MIMO in wireless AI: Intelligentizing mobile edge computing, caching and communica-
LANs,” in Proc. IEEE 28th Int. Conf. Netw. Protocols (ICNP), Madrid, tion by federated learning,” IEEE Netw., vol. 33, no. 5, pp. 156–165,
Spain, Oct. 2020, pp. 1–11. Sep./Oct. 2019.
[156] Y. He, C. Liang, F. R. Yu, and Z. Han, “Trust-based social networks [178] K. Sarpatwar, R. Vaculin, H. Min, G. Su, T. Heath, G. Ganapavarapu,
with computing, caching and communications: A deep reinforce- and D. Dillenberger, “Towards enabling trusted artificial intelli-
ment learning approach,” IEEE Trans. Netw. Sci. Eng., vol. 7, no. 1, gence via blockchain,” in Policy-Based Autonomic Data Governance,
pp. 66–79, Jan.–Mar. 2020. vol. 11550. Cham, Switzerland: Springer, Apr. 2019, pp. 137–153.
[157] Y. He, N. Zhao, and H. Yin, “Integrated networking, caching, and [Online]. Available: https://doi.org/10.1007/978-3-030-17277-0_8
computing for connected vehicles: A deep reinforcement learning [179] J. Kang, Z. Xiong, D. Niyato, S. Xie, and J. Zhang, “Incentive mech-
approach,” IEEE Trans. Veh. Technol., vol. 67, no. 1, pp. 44–55, anism for reliable federated learning: A joint optimization approach
Jan. 2018. to combining reputation and contract theory,” IEEE Internet Things J.,
[158] A. Ndikumana et al., “Joint communication, computation, caching, and vol. 6, no. 6, pp. 10700–10714, Dec. 2019.
control in big data multi-access edge computing,” IEEE Trans. Mobile [180] X. Bao, C. Su, Y. Xiong, W. Huang, and Y. Hu, “FLChain: A
Comput., vol. 19, no. 6, pp. 1359–1374, Jun. 2020. blockchain for auditable federated learning with trust and incentive,” in
[159] Y. Xu, W. Xu, Z. Wang, J. Lin, and S. Cui, “Load balancing for ultra- Proc. 5th Int. Conf. Big Data Comput. Commun. (BIGCOM), Qingdao,
dense networks: A deep reinforcement learning-based approach,” IEEE China, Aug. 2019, pp. 151–159.
Internet Things J., vol. 6, no. 6, pp. 9399–9412, Dec. 2019. [181] M. Liu, F. R. Yu, Y. Teng, V. C. M. Leung, and M. Song, “Performance
[160] N. Zhao, Y.-C. Liang, D. Niyato, Y. Pei, M. Wu, and Y. Jiang, “Deep optimization for blockchain-enabled Industrial Internet of Things (IIoT)
reinforcement learning for user association and resource allocation systems: A deep reinforcement learning approach,” IEEE Trans. Ind.
in heterogeneous cellular networks,” IEEE Trans. Wireless Commun., Informat., vol. 15, no. 6, pp. 3559–3570, Jun. 2019.
vol. 18, no. 11, pp. 5141–5152, Nov. 2019. [182] J. Yun, Y. Goh, and J.-M. Chung, “DQN-based optimization framework
[161] F. B. Mismar and B. L. Evans, “Deep Q-learning for self-organizing for secure sharded blockchain systems,” IEEE Internet Things J., vol. 8,
networks fault management and radio performance improvement,” in no. 2, pp. 708–722, Jan. 2021.
Proc. IEEE 52nd Asilomar Conf. Signals Syst. Comput., Pacific Grove, [183] C. Qiu, X. Ren, Y. Cao, and T. Mai, “Deep reinforcement learning
CA, USA, Oct. 2018, pp. 1457–1461. empowered adaptivity for future blockchain networks,” IEEE Open J.
[162] J. Ye and Y.-J. A. Zhang, “DRAG: Deep reinforcement learning based Commun. Soc., vol. 2, pp. 99–105, 2021.
base station activation in heterogeneous networks,” IEEE Trans. Mobile [184] J. Zhang, Z. Hong, X. Qiu, Y. Zhan, S. Guo, and W. Chen, “SkyChain:
Comput., vol. 19, no. 9, pp. 2076–2087, Sep. 2020. A deep reinforcement learning-empowered dynamic blockchain shard-
[163] J. Liu, B. Krishnamachari, S. Zhou, and Z. Niu, “DeepNap: Data- ing system,” in Proc. 49th Int. Conf. Parallel Process. (ICPP),
driven base station sleeping operations through deep reinforcement Aug. 2020, pp. 1–11.
learning,” IEEE Internet Things J., vol. 5, no. 6, pp. 4273–4282, [185] F. Bravo-Marquez, S. Reeves, and M. Ugarte, “Proof-of-learning: A
Dec. 2018. blockchain consensus mechanism based on machine learning com-
[164] J. Wu, P. Yu, L. Feng, F. Zhou, W. Li, and X. Qiu, “3D aerial petitions,” in Proc. IEEE Int. Conf. Decentralized Appl. Infrastruct.
base station position planning based on deep Q-network for capacity (DAPPCON), Newark, CA, USA, Apr. 2019, pp. 119–124.
enhancement,” in Proc. IFIP/IEEE Symp. Integr. Netw. Service Manag. [186] C. Chenli, B. Li, Y. Shi, and T. Jung, “Energy-recycling
(IM), Arlington, VA, USA, Apr. 2019, pp. 482–487. blockchain with proof-of-deep-learning,” in Proc. IEEE Int. Conf.
[165] J. D. Harris and B. Waggoner, “Decentralized and collaborative AI on Blockchain Cryptocurrency (ICBC), Seoul, South Korea, May 2019,
blockchain,” in Proc. IEEE Int. Conf. Blockchain (Blockchain), Atlanta, pp. 19–23.
GA, USA, Jul. 2019, pp. 368–375. [187] A. Baldominos and Y. Saez, “Coin.AI: A proof-of-useful-work scheme
for blockchain-based distributed deep learning,” Entropy, vol. 21, no. 8,
[166] G. Zhang, T. Li, Y. Li, P. Hui, and D. Jin, “Blockchain-based data
p. 723, Jul. 2019.
sharing system for AI-powered network operations,” J. Commun. Inf.
[188] J. Chen, K. Duan, R. Zhang, L. Zeng, and W. Wang, “An AI based
Netw., vol. 3, no. 3, pp. 1–8, Sep. 2018.
super nodes selection algorithm in blockchain networks,” Aug. 2018,
[167] K. Wang, J. Dong, Y. Wang, and H. Yin, “Securing data with
arXiv:1808.00216.
blockchain and AI,” IEEE Access, vol. 7, pp. 77981–77989, 2019.
[189] W. Chen, Z. Zheng, J. Cui, E. Ngai, P. Zheng, and Y. Zhou, “Detecting
[168] C. H. Liu, Q. Lin, and S. Wen, “Blockchain-enabled data collection PONZI schemes on Ethereum: Towards healthier blockchain technol-
and sharing for Industrial IoT with deep reinforcement learning,” IEEE ogy,” in Proc. World Wide Web Conf., Apr. 2018, pp. 1409–1418.
Trans. Ind. Informat., vol. 15, no. 6, pp. 3516–3526, Jun. 2019. [190] M. Salimitari, M. Joneidi, and M. Chatterjee, “AI-enabled blockchain:
[169] G. J. Mendis, Y. Wu, J. Wei, M. Sabounchi, and R. Roche, “A An outlier-aware consensus protocol for blockchain-based IoT
blockchain-powered decentralized and secure computing paradigm,” networks,” in Proc. IEEE Global Commun. Conf. (GLOBECOM),
IEEE Trans. Emerg. Topics Comput., vol. 9, no. 4, pp. 2201–2222, Waikoloa, HI, USA, Dec. 2019, pp. 1–6.
Oct.–Dec. 2021. [191] W. Liang, L. Xiao, K. Zhang, M. Tang, D. He, and K.-C. Li,
[170] Q. Wang, Y. Guo, X. Wang, T. Ji, L. Yu, and P. Li, “AI at the “Data fusion approach for collaborative anomaly intrusion detection
edge: Blockchain-empowered secure multiparty learning with heteroge- in blockchain-based systems,” IEEE Internet Things J., vol. 9, no. 16,
neous models,” IEEE Internet Things J., vol. 7, no. 10, pp. 9600–9610, pp. 14741–14751, Aug. 2022.
Oct. 2020. [192] F. Scicchitano, A. Liguori, M. Guarascio, E. Ritacco, and G. Manco, “A
[171] C. Korkmaz, H. E. Kocas, A. Uysal, A. Masry, O. Ozkasap, and deep learning approach for detecting security attacks on blockchain,”
B. Akgun, “Chain FL: Decentralized federated machine learning via in Proc. ITASEC, 2020, pp. 212–222.
blockchain,” in Proc. 2nd Int. Conf. Blockchain Comput. Appl. (BCCA), [193] M. Mudassir, S. Bennbaia, D. Unal, and M. Hammoudeh, “Time-
Antalya, Turkey, Nov. 2020, pp. 140–146. series forecasting of bitcoin prices using high-dimensional features: A
[172] X. Lin, J. Li, J. Wu, H. Liang, and W. Yang, “Making knowledge trad- machine learning approach,” Neural Comput. Appl., to be published.
able in edge-AI enabled IoT: A consortium blockchain-based efficient [194] H. J. Singh and A. S. Hafid, “Prediction of transaction confirmation
and incentive approach,” IEEE Trans. Ind. Informat., vol. 15, no. 12, time in Ethereum blockchain using machine learning,” in Proc. Int.
pp. 6367–6378, Dec. 2019. Congr. Blockchain Appl., Jun. 2019, pp. 126–133.
[173] R. Gupta, D. Reebadiya, S. Tanwar, N. Kumar, and M. Guizani, “When [195] S. Mohammadi and E. Rabieinejad, “Prediction forks in the blockchain
blockchain meets edge intelligence: Trusted and security solutions for using machine learning,” in Proc. 3rd Int. Conf. Mech. Elect. Comput.
consumers,” IEEE Netw., vol. 35, no. 5, pp. 272–278, Sep./Oct. 2021. Eng., Dec. 2020, pp. 1–8.
[174] J. Weng, J. Weng, J. Zhang, M. Li, Y. Zhang, and W. Luo, “DeepChain: [196] Q. Pan, J. Wu, J. Li, W. Yang, and Z. Guan, “Blockchain and AI
Auditable and privacy-preserving deep learning with blockchain-based empowered trust-information-centric network for beyond 5G,” IEEE
incentive,” IEEE Trans. Depend. Security Comput., vol. 18, no. 5, Netw., vol. 34, no. 6, pp. 38–45, Nov. 2020.
pp. 2438–2455, Sep./Oct. 2021. [197] K. Zhang, Y. Zhu, S. Maharjan, and Y. Zhang, “Edge intelli-
[175] M. Shayan, C. Fung, C. J. M. Yoon, and I. Beschastnikh, “Biscotti: gence and blockchain empowered 5G beyond for the Industrial
A blockchain system for private and secure federated learning,” IEEE Internet of Things,” IEEE Netw., vol. 33, no. 5, pp. 12–19,
Trans. Parallel Distrib. Syst., vol. 32, no. 7, pp. 1513–1525, Jul. 2021. Sep. 2019.
[176] L. Feng, Y. Zhao, S. Guo, X. Qiu, W. Li, and P. Yu, “Blockchain-based [198] S. Rathore, J. H. Park, and H. Chang, “Deep learning and blockchain-
asynchronous federated learning for Internet of Things,” IEEE Trans. empowered security framework for intelligent 5G-enabled IoT,” IEEE
Comput., vol. 71, no. 5, pp. 1092–1103, May 2022. Access, vol. 9, pp. 90075–90083, 2021.
[177] M. Nassar, K. Salah, M. H. U. Rehman, and D. Svetinovic, “Blockchain [199] S. Hu, Y.-C. Liang, Z. Xiong, and D. Niyato, “Blockchain and artificial
for explainable and trustworthy artificial intelligence,” Interdiscipl. Rev. intelligence for dynamic resource sharing in 6G and beyond,” IEEE
Data Min. Knowl. Disc., vol. 10, no. 1, Oct. 2020, Art. no. e1340. Wireless Commun., vol. 28, no. 4, pp. 145–151, Aug. 2021.
[200] T. Maksymyuk et al., “AI-enabled blockchain framework for dynamic [220] Y. Tian, T. Li, J. Xiong, M. Z. A. Bhuiyan, J. Ma, and C. Peng,
spectrum management in multi-operator 6G networks,” in Future “A blockchain-based machine learning framework for edge services
Intent-Based Networking, vol. 831. Cham, Switzerland: Springer, in IIoT,” IEEE Trans. Ind. Informat., vol. 18, no. 3, pp. 1918–1929,
Dec. 2022, pp. 322–338. [Online]. Available: https://doi.org/10.1007/ Mar. 2022.
978-3-030-92435-5_19 [221] H. Lin, S. Garg, J. Hu, G. Kaddoum, M. Peng, and M. S. Hossain,
[201] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, “Blockchain “Blockchain and deep reinforcement learning empowered spatial
and federated learning for 5G beyond,” IEEE Netw., vol. 35, no. 1, crowdsourcing in software-defined Internet of Vehicles,” IEEE Trans.
pp. 219–225, Jan./Feb. 2020. Intell. Transp. Syst., vol. 22, no. 6, pp. 3755–3764, Jun. 2021.
[202] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, [222] R. Kumar, P. Kumar, R. Tripathi, G. P. Gupta, N. Kumar, and
“Communication-efficient federated learning and permissioned M. M. Hassan, “A privacy-preserving-based secure framework using
blockchain for digital twin edge networks,” IEEE Internet Things J., blockchain-enabled deep-learning in cooperative intelligent trans-
vol. 8, no. 4, pp. 2276–2288, Feb. 2021. port system,” IEEE Trans. Intell. Transp. Syst., vol. 23, no. 9,
[203] F. Guo, F. R. Yu, H. Zhang, H. Ji, M. Liu, and V. C. Leung, “Adaptive pp. 16492–16503, Sep. 2022.
resource allocation in future wireless networks with blockchain and [223] C. Zhang, W. Li, Y. Luo, and Y. Hu, “AIT: An AI-enabled trust man-
mobile edge computing,” IEEE Trans. Wireless Commun., vol. 19, agement system for vehicular networks using blockchain technology,”
no. 3, pp. 1689–1703, Mar. 2020. IEEE Internet Things J., vol. 8, no. 5, pp. 3157–3169, Mar. 2021.
[204] G. Gür, “Expansive networks: Exploiting spectrum sharing for capacity [224] W. Zhang et al., “Blockchain-based federated learning for device failure
boost and 6G vision,” J. Commun. Netw., vol. 22, no. 6, pp. 444–454, detection in Industrial IoT,” IEEE Internet Things J., vol. 8, no. 7,
Dec. 2020. pp. 5926–5937, Apr. 2021.
[205] C. Qiu, F. R. Yu, H. Yao, C. Jiang, F. Xu, and C. Zhao, “Blockchain- [225] S. Otoum, I. Al Ridhawi, and H. Mouftah, “Securing critical IoT infras-
based software-defined Industrial Internet of Things: A dueling tructures with blockchain-supported federated learning,” IEEE Internet
deep Q-learning approach,” IEEE Internet Things J., vol. 6, no. 3, Things J., vol. 9, no. 4, pp. 2592–2601, Feb. 2022.
pp. 4627–4639, Jun. 2019. [226] Y. Lu, X. Huang, Y. Dai, S. Maharjan, and Y. Zhang, “Blockchain and
[206] Y. He, Y. Wang, C. Qiu, Q. Lin, J. Li, and Z. Ming, “Blockchain-based federated learning for privacy-preserved data sharing in Industrial IoT,”
edge computing resource allocation in IoT: A deep reinforcement learn- IEEE Trans. Ind. Informat., vol. 16, no. 6, pp. 4177–4186, Jun. 2020.
ing approach,” IEEE Internet Things J., vol. 8, no. 4, pp. 2226–2237, [227] P. Zhang, Y. Hong, N. Kumar, M. Alazab, M. D. Alshehri, and C. Jiang,
Feb. 2021. “BC-EdgeFL: Defensive transmission model based on blockchain
[207] G. Manogaran, S. Mumtaz, C. X. Mavromoustakis, E. Pallis, assisted reinforced federated learning in IIoT environment,” IEEE
and G. Mastorakis, “Artificial intelligence and blockchain-assisted Trans. Ind. Informat., vol. 18, no. 5, pp. 3551–3561, May 2022.
offloading approach for data availability maximization in edge [228] H. Liu et al., “Blockchain and federated learning for collaborative intru-
nodes,” IEEE Trans. Veh. Technol., vol. 70, no. 3, pp. 2404–2412, sion detection in vehicular edge computing,” IEEE Trans. Veh. Technol.,
Mar. 2021. vol. 70, no. 6, pp. 6073–6084, Jun. 2021.
[208] D. C. Nguyen, P. N. Pathirana, M. Ding, and A. Seneviratne, “Secure [229] P. Bhattacharya, S. Tanwar, U. Bodkhe, S. Tyagi, and N. Kumar,
computation offloading in blockchain based IoT networks with deep “BinDaaS: Blockchain-based deep-learning as-a-service in health-
reinforcement learning,” IEEE Trans. Netw. Sci. Eng., vol. 8, no. 4, care 4.0 applications,” IEEE Trans. Netw. Sci. Eng., vol. 8, no. 2,
pp. 3192–3208, Oct. 2021. pp. 1242–1255, Apr.–Jun. 2021.
[209] H. Liao, Y. Mu, Z. Zhou, M. Sun, Z. Wang, and C. Pan, “Blockchain [230] A. Z. Al-Marridi, A. Mohamed, and A. Erbad, “Reinforcement learning
and learning-based secure and intelligent task offloading for vehicu- approaches for efficient and secure blockchain-powered smart health
lar fog computing,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 7, systems,” Comput. Netw., vol. 197, Oct. 2021, Art. no. 108279.
pp. 4051–4063, Jul. 2021. [231] T. Veeramakali, R. Siva, B. Sivakumar, P. S. Mahesh, and N. Krishnaraj,
[210] H. Liao et al., “Blockchain and semi-distributed learning-based secure “An intelligent internet of things-based secure healthcare framework
and low-latency computation offloading in space–air–ground-integrated using blockchain technology with an optimal deep learning model,” J.
power IoT,” IEEE J. Sel. Topics Signal Process., vol. 16, no. 3, Supercomput., vol. 77, no. 9, pp. 9576–9596, Feb. 2021.
pp. 381–394, Apr. 2022. [232] R. Kumar et al., “Blockchain-federated-learning and deep learning
[211] C. Qiu, H. Yao, X. Wang, N. Zhang, F. R. Yu, and D. Niyato, “AI-chain: models for COVID-19 detection using CT imaging,” IEEE Sensors
Blockchain energized edge intelligence for beyond 5G networks,” IEEE J., vol. 21, no. 14, pp. 16301–16314, Jul. 2021.
Netw., vol. 34, no. 6, pp. 62–69, Nov. 2020. [233] S. Otoum, I. A. Ridhawi, and H. T. Mouftah, “Preventing and con-
[212] Y. Qian, Y. Jiang, L. Hu, M. S. Hossain, M. Alrashoud, and trolling epidemics through blockchain-assisted AI-enabled networks,”
M. Al-Hammadi, “Blockchain-based privacy-aware content caching in IEEE Netw., vol. 35, no. 3, pp. 34–41, May/Jun. 2021.
cognitive Internet of Vehicles,” IEEE Netw., vol. 34, no. 2, pp. 46–51, [234] B. Mallikarjuna, G. Shrivastava, and M. Sharma, “Blockchain tech-
Mar./Apr. 2020. nology: A DNN token-based approach in healthcare and COVID-19
[213] Y. Dai, D. Xu, K. Zhang, S. Maharjan, and Y. Zhang, “Deep rein- to generate extracted data,” Exp. Syst., vol. 39, no. 3, Jul. 2021,
forcement learning and permissioned blockchain for content caching in Art. no. e12778.
vehicular edge computing and networks,” IEEE Trans. Veh. Technol., [235] X. Guo et al., “Smartphone-based DNA diagnostics for malaria detec-
vol. 69, no. 4, pp. 4312–4324, Apr. 2020. tion using deep learning for local decision support and blockchain
[214] M. Li, F. R. Yu, P. Si, W. Wu, and Y. Zhang, “Resource optimization for technology for security,” Nat. Electron., vol. 4, no. 8, pp. 615–624,
delay-tolerant data in blockchain-enabled IoT with edge computing: A Aug. 2021.
deep reinforcement learning approach,” IEEE Internet Things J., vol. 7, [236] R. Gupta, A. Shukla, and S. Tanwar, “BATS: A blockchain and
no. 10, pp. 9399–9412, Oct. 2020. AI-empowered drone-assisted telesurgery system towards 6G,” IEEE
[215] L. Cui et al., “CREAT: Blockchain-assisted compression algorithm Trans. Netw. Sci. Eng., vol. 8, no. 4, pp. 2958–2967, Oct.–Dec. 2021.
of federated learning for content caching in edge computing,” IEEE [237] I. Al Ridhawi, M. Aloqaily, A. Boukerche, and Y. Jararweh, “Enabling
Internet Things J., vol. 9, no. 16, pp. 14151–14161, Aug. 2022. intelligent IoCV services at the edge for 5G networks and beyond,”
[216] R. Zhang, F. R. Yu, J. Liu, T. Huang, and Y. Liu, “Deep reinforcement IEEE Trans. Intell. Transp. Syst., vol. 22, no. 8, pp. 5190–5200,
learning (DRL)-based device-to-device (D2D) caching with blockchain Aug. 2021.
and mobile edge computing,” IEEE Trans. Wireless Commun., vol. 19, [238] V. Hassija, V. Gupta, S. Garg, and V. Chamola, “Traffic jam probability
no. 10, pp. 6469–6485, Oct. 2020. estimation based on blockchain and deep neural networks,” IEEE Trans.
[217] N. Waheed, X. He, M. Ikram, M. Usman, S. S. Hashmi, and M. Usman, Intell. Transp. Syst., vol. 22, no. 7, pp. 3919–3928, Jul. 2021.
“Security and privacy in IoT using machine learning and blockchain: [239] K. Tiba, R. M. Parizi, Q. Zhang, A. Dehghantanha, H. Karimipour,
Threats and countermeasures,” ACM Comput. Surveys, vol. 53, no. 6, and K.-K. R. Choo, “Secure blockchain-based traffic load balanc-
pp. 1–37, Nov. 2021. ing using edge computing and reinforcement learning,” in Blockchain
[218] N. Dhieb, H. Ghazzai, H. Besbes, and Y. Massoud, “Scalable and Cybersecurity, Trust and Privacy, vol. 79. Cham, Switzerland: Springer,
secure architecture for distributed IoT systems,” in Proc. IEEE Technol. Mar. 2020, pp. 99–128. [Online]. Available: https://doi.org/10.1007/
Eng. Manag. Conf. (TEMSCON), Novi, MI, USA, Jun. 2020, pp. 1–6. 978-3-030-38181-3_6
[219] X. Wang, S. Garg, H. Lin, M. J. Piran, J. Hu, and M. S. Hossain, [240] Y. Song, Y. Fu, F. R. Yu, and L. Zhou, “Blockchain-enabled Internet
“Enabling secure authentication in Industrial IoT with transfer learning of Vehicles with cooperative positioning: A deep neural network
empowered blockchain,” IEEE Trans. Ind. Informat., vol. 17, no. 11, approach,” IEEE Internet Things J., vol. 7, no. 4, pp. 3485–3498,
pp. 7725–7733, Nov. 2021. Apr. 2020.
[241] D. Zhang, F. R. Yu, and R. Yang, “Blockchain-based distributed [262] Y.-C. Liang, Dynamic Spectrum Management: From Cognitive Radio to
software-defined vehicular networks: A dueling deep Q-learning Blockchain and Artificial Intelligence. Singapore: Springer, Sep. 2020,
approach,” IEEE Trans. Cogn. Commun. Netw., vol. 5, no. 4, p. 166. [Online]. Available: https://doi.org/10.1007/978-981-15-0776-2
pp. 1086–1100, Dec. 2019. [263] C.-X. Wang et al., “On the road to 6G: Visions, requirements, key
[242] X. Jiang, F. R. Yu, T. Song, and V. C. Leung, “Intelligent resource technologies and testbeds,” IEEE Commun. Surveys Tuts., vol. 25, no. 2,
allocation for video analytics in blockchain-enabled Internet of pp. 905–974, 2nd Quart., 2023.
Autonomous Vehicles with edge computing,” IEEE Internet Things J., [264] C. De Alwis et al., “Survey on 6G frontiers: Trends, applica-
vol. 9, no. 16, pp. 14260–14272, Aug. 2022. tions, requirements, technologies and future research,” IEEE Open J.
[243] Y. Fu, C. Li, F. R. Yu, T. H. Luan, and Y. Zhang, “An autonomous lane- Commun. Soc., vol. 2, pp. 836–886, 2021.
changing system with knowledge accumulation and transfer assisted [265] J. Rosenworcel. “The FCC should use blockchain to manage wireless
by vehicular blockchain,” IEEE Internet Things J., vol. 7, no. 11, spectrum (2018-03-20).” Accessed: May 2020. [Online]. Available:
pp. 11123–11136, Nov. 2020. https://www.wired.com/story/the-fcc-should-use-blockchain-to-
[244] S. R. Pokhrel and J. Choi, “Federated learning with blockchain for manage-wireless-spectrum
autonomous vehicles: Analysis and design challenges,” IEEE Trans.
Commun., vol. 68, no. 8, pp. 4734–4746, Aug. 2020.
[245] Y. Lu, X. Huang, K. Zhang, S. Maharjan, and Y. Zhang, “Blockchain
empowered asynchronous federated learning for secure data sharing Yiping Zuo (Member, IEEE) received the B.S.
in Internet of Vehicles,” IEEE Trans. Veh. Technol., vol. 69, no. 4, degree in electronic information science and
pp. 4298–4311, Apr. 2020. technology from Huaiyin Normal University,
[246] H. Chai, S. Leng, Y. Chen, and K. Zhang, “A hierarchical blockchain- Huaian, China, in 2015, the M.S. degree in commu-
enabled federated learning algorithm for knowledge sharing in inter- nication and information systems from the Nanjing
net of vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 7, University of Posts and Telecommunications,
pp. 3975–3986, Jul. 2021. Nanjing, China, in 2018, and the Ph.D. degree
[247] N. M. Kumar et al., “Distributed energy resources and the application in information and communications engineering
of AI, IoT, and blockchain in smart grids,” Energies, vol. 13, no. 21, from Southeast University, Nanjing, in 2022.
p. 5739, Nov. 2020. She is currently with the faculty of the College
[248] M. Keshk, B. Turnbull, N. Moustafa, D. Vatsalan, and K.-K. R. Choo, of Computer, Nanjing University of Posts and
“A privacy-preserving-framework-based blockchain and deep learning Telecommunications. Her current research interests include blockchain,
for protecting smart power networks,” IEEE Trans. Ind. Informat., mobile edge computing, and ISAC communications.
vol. 16, no. 8, pp. 5110–5118, Aug. 2020.
[249] Z. Wang, M. Ogbodo, H. Huang, C. Qiu, M. Hisada, and
A. B. Abdallah, “AEBIS: AI-enabled blockchain-based electric vehi-
cle integration system for power management in smart grid platform,”
IEEE Access, vol. 8, pp. 226409–226421, 2020. Jiajia Guo (Member, IEEE) received the B.S.
[250] M. A. Ferrag and L. Maglaras, “DeepCoin: A novel deep learning and degree from the Nanjing University of Science
blockchain-based energy exchange framework for smart grids,” IEEE and Technology, Nanjing, China, in 2016, the
Trans. Eng. Manag., vol. 67, no. 4, pp. 1285–1297, Nov. 2020. M.S. degree from the University of Science and
[251] F. Jamil et al., “Peer-to-peer energy trading mechanism based on Technology of China, Hefei, China, in 2019, and
blockchain and machine learning for sustainable electrical power the Ph.D. degree in information and communications
supply in smart grid,” IEEE Access, vol. 9, pp. 39193–39217, 2021. engineering from Southeast University, Nanjing, in
[252] G. Gao et al., “FogChain: A blockchain-based peer-to-peer solar power 2023. He is currently a Postdoctoral Research Fellow
trading system powered by fog AI,” IEEE Internet Things J., vol. 9, with the State Key Laboratory of Internet of Things
no. 7, pp. 5200–5215, Apr. 2022. for Smart City, University of Macau. His current
[253] Z. Li, J. Xu, Y. Xie, J. Jiang, Y. Zhu, and X. Yang, “Integration of research interests include AI-native air interface,
blockchain and machine learning for microgrids,” in Proc. IEEE 6th Int. reconfigurable intelligent surfaces, and massive MIMO.
Conf. Comput. Commun. Syst. (ICCCS), Chengdu, China, Apr. 2021,
pp. 211–216.
[254] M. Singh, G. S. Aujla, and R. S. Bali, “A deep learning-based
blockchain mechanism for secure Internet of Drones environment,” Ning Gao (Member, IEEE) received the Ph.D.
IEEE Trans. Intell. Transp. Syst., vol. 22, no. 7, pp. 4404–4413, degree in information and communications engi-
Jul. 2021. neering from the Beijing University of Posts
[255] C. Feng, B. Liu, K. Yu, S. K. Goudos, and S. Wan, “Blockchain- and Telecommunications, Beijing, China, in 2019.
empowered decentralized horizontal federated learning for 5G-enabled From 2017 to 2018, he was a visiting Ph.D.
UAVs,” IEEE Trans. Ind. Informat., vol. 18, no. 5, pp. 3582–3592, student with the School of Computing and
May 2022. Communications, Lancaster University, Lancaster,
[256] A. Asheralieva and D. Niyato, “Distributed dynamic resource manage- U.K. From 2019 to 2022, he was a Research Fellow
ment and pricing in the IoT systems with blockchain-as-a-service and with the National Mobile Communications Research
UAV-enabled mobile edge computing,” IEEE Internet Things J., vol. 7, Laboratory, Southeast University, where he is cur-
no. 3, pp. 1974–1993, Mar. 2020. rently an Associate Professor with the School of
[257] S. R. Pokhrel, “Blockchain brings trust to collaborative drones and Cyber Science and Engineering. His research interests include AI/RIS enabled
LEO satellites: An intelligent decentralized learning in the space,” IEEE wireless communications and security, ISAC, and UAV communications.
Sensors J., vol. 21, no. 22, pp. 25331–25339, Nov. 2021.
[258] A. Islam, T. Rahim, M. Masuduzzaman, and S. Y. Shin, “A blockchain-
based artificial intelligence-empowered contagious pandemic situation
supervision scheme using Internet of Drone Things,” IEEE Wireless Yongxu Zhu (Senior Member, IEEE) received the
Commun., vol. 28, no. 4, pp. 166–173, Aug. 2021. Ph.D. degree from University College London, U.K.,
[259] A. Gumaei et al., “Deep learning and blockchain with edge computing in 2017. Then, she was a Postdoctoral Fellow with
for 5G-enabled drone identification and flight mode detection,” IEEE Loughborough University, U.K. She was with the
Netw., vol. 35, no. 1, pp. 94–100, Jan./Feb. 2021. Department of Engineering, London South Bank
[260] A. Aftab, N. Ashraf, H. K. Qureshi, S. A. Hassan, and S. Jangsher, University, U.K., as a Senior Lecturer from January
“BLOCK-ML: Blockchain and machine learning for UAV-BSs deploy- 2022 to October 2022 and a Lecturer from 2019
ment,” in Proc. IEEE 92nd Veh. Technol. Conf. (VTC-Fall), Victoria, to 2021. She is currently an Assistant Professor
BC, Canada, Nov. 2020, pp. 1–5. with the University of Warwick. Her research
[261] S. R. Pokhrel, “Federated learning meets blockchain at 6G edge: A interests include B5G/6G, heterogeneous networks,
drone-assisted networking for disaster response,” in Proc. 2nd ACM non-terrestrial network, and physical-layer security.
MobiCom Workshop Drone Assist. Wireless Commun. 5G Beyond, She also serves as an Editor for IEEE W IRELESS C OMMUNICATIONS
Sep. 2020, pp. 49–54. L ETTERS and IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS.
Shi Jin (Senior Member, IEEE) received the B.S. Xiao Li (Member, IEEE) received the Ph.D. degree
degree in communications engineering from the in communication and information systems from
Guilin University of Electronic Technology, Guilin, Southeast University, Nanjing, China, in 2010. She
China, in 1996, the M.S. degree from the Nanjing then joined the School of Information Science
University of Posts and Telecommunications, and Engineering, Southeast University, where she
Nanjing, China, in 2003, and the Ph.D. degree has been a Professor of Information Systems and
in information and communications engineering Communications since July 2020. From January
from Southeast University, Nanjing, in 2007. From 2013 to January 2014, she was a Postdoctoral Fellow
June 2007 to October 2009, he was a Research with The University of Texas at Austin, Austin, TX,
Fellow with the Adastral Park Research Campus, USA. Her current research interests include recon-
University College London, London, U.K. He is figurable intelligent surface assisted communications
currently affiliated with the faculty of the National Mobile Communications and intelligent communications. She serves as an Associate Editor for the
Research Laboratory, Southeast University. His research interests include IEEE W IRELESS C OMMUNICATIONS L ETTERS and Electronics Letters, and
wireless communications, random matrix theory, and information theory. He a Guest Editor for the Digital Communications and Networks.
and his coauthors were awarded the 2011 IEEE Communications Society
Stephen O. Rice Prize Paper Award in the field of communication theory,
the 2022 Best Paper Award, and the 2010 Young Author Best Paper Award
by the IEEE Signal Processing Society. He serves as an Area Editor for the
IEEE T RANSACTIONS ON C OMMUNICATIONS and IET Electronics Letters.
He was previously an Associate Editor of the IEEE T RANSACTIONS ON
W IRELESS C OMMUNICATIONS, IEEE C OMMUNICATIONS L ETTERS, and
IET Communications.

Base Paper

Uploaded by

Copyright:

Available Formats

Base Paper

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Base Paper

Uploaded by

Copyright:

Available Formats

2494 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 25, NO.

4, FOURTH QUARTER 2023

A Survey of Blockchain and Artificial Intelligence

Fig. 1. Outline of the paper.

Fig. 2. The structure of blocks.

Fig. 3. Classification of AI.

Fig. 4. Taxonomy of the integration of blockchain and AI.

Fig. 5. Taxonomy of the integration of blockchain and AI for wireless communications.

You might also like