report e

IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO.
15, AUGUST 1, 2021 11891
Enabling Massive IoT Toward 6G:

A Comprehensive Survey
Fengxian Guo , F. Richard Yu , Fellow, IEEE, Heli Zhang , Xi Li , Member, IEEE,
Hong Ji , Senior Member, IEEE, and Victor C. M. Leung , Life Fellow, IEEE
Abstract—Nowadays, many disruptive Internet-of-Things (IoT) interactions in the future, including holographic commu-
applications emerge, such as augmented/virtual reality online nications, five-sense communications, and wireless brain–
games, autonomous driving, and smart everything, which are computer interfaces (WBCI), which will lead to a true immer-
massive in number, data intensive, computation intensive, and
delay sensitive. Due to the mismatch between the fifth gener- sion into a distant environment. At the same time, advances
ation (5G) and the requirements of such massive IoT-enabled in personal communications will promote the evolution of
applications, there is a need for technological advancements smart verticals in the fifth-generation networks (5G) to a
and evolutions for wireless communications and networking higher level, including healthcare, remote education/training,
toward the sixth-generation (6G) networks. 6G is expected to industry Internet, fully autonomous driving, and super smart
deliver extended 5G capabilities at a very high level, such as
Tbps data rate, sub-ms latency, cm-level localization, and so homes/cities. During this paradigm shift, the Internet of Things
on, which will play a significant role in supporting massive IoT (IoT) plays a vital role in enabling these emerging applications
devices to operate seamlessly with highly diverse service require- by connecting the physical environment to the cyberspace of
ments. Motivated by the aforementioned facts, in this article, communication systems [1].
we present a comprehensive survey on 6G-enabled massive IoT. Although these IoT-enabled applications will bring conve-
First, we present the drivers and requirements by summarizing
the emerging IoT-enabled applications and the corresponding nience to human life, it is an extremely daunting task for 5G
requirements, along with the limitations of 5G. Second, visions to support these applications. First, these IoT-enabled appli-
of 6G are provided in terms of core technical requirements, use cations require superior performances in terms of data rate,
cases, and trends. Third, a new network architecture provided by latency, coverage, localization, and so on. Second, they are
6G to enable massive IoT is introduced, i.e., space–air–ground– more data intensive and computation intensive, which far
underwater/sea networks enhanced by edge computing. Fourth,
some breakthrough technologies, such as machine learning and exceeds the range of ultrareliable low-latency communications
blockchain, in 6G are introduced, where the motivations, appli- (uRLLC) and massive machine-type communication (mMTC)
cations, and open issues of these technologies for massive IoT are of 5G [2]. Third, it is hard to efficiently manage massive IoT
summarized. Finally, a use case of fully autonomous driving is devices in this case. Fourth, with massive data generated, seri-
presented to show 6G supports massive IoT. ous security issues are accompanying [3]. With IoT evolving,
Index Terms—6G, blockchain, Internet of Things (IoT), 5G will gradually reach its limitations and be unable to pro-
machine learning, space–air–ground–underwater networks. vide support to most of these advanced applications, which
can be predicted from the history of previous generations. So
there is a strong motivation for the sixth-generation networks
I. I NTRODUCTION (6G) to extend 5G capabilities to a higher level to enable
NSTEAD of only exchanges of voice, image, or video massive IoT.
I in the fifth-generation (5G) mobile networks and the ear-
lier generations, researchers are exploring new forms of
Motivated by the aforementioned facts, the visions of 6G in
terms of requirements, use cases, and trends should be clearly
investigated first, since 6G is not defined yet. Then, to have
Manuscript received October 26, 2020; revised February 6, 2021; accepted massive IoT deployed in 6G, the development with respect to
February 16, 2021. Date of publication March 4, 2021; date of current architecture, breakthrough technologies, and their challenges
version July 23, 2021. This work was supported by the National Natural should be known.
Science Foundation of China under Grant 61771070. (Corresponding author:
Heli Zhang.)
Fengxian Guo, Heli Zhang, Xi Li, and Hong Ji are with the Key A. Visions of 6G
Laboratory of Universal Wireless Communications, Ministry of Education, Inspired by the robust requirements of the future IoT-
Beijing University of Posts and Telecommunications, Beijing 100876, China
(e-mail: fengxianguo@bupt.edu.cn; zhangheli@bupt.edu.cn; enabled applications and the limitations of 5G, 6G, as an
lixi@bupt.edu.cn; jihong@bupt.edu.cn). evolutionary generation, will expand and upgrade based on
F. Richard Yu is with the Department Systems and Computer 5G from every aspect, which revolutionizes not only human
Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada (e-mail:
richard.yu@carleton.ca). life but also society. First, network performance of 6G will
Victor C. M. Leung is with the College of Computer Science and Software upgrade to a superior level, e.g., higher data rate (up to Tbps),
Engineering, Shenzhen University, Shenzhen 518060, China, and also with the lower latency (sub-ms), 3-D (3-D)-ubiquitous coverage (into
Department of Electrical and Computer Engineering, the University of British
Columbia, Vancouver, BC V6T 1Z4, Canada (e-mail: vleung@ieee.org). space, sea, and even the undersea), more accurate localization
Digital Object Identifier 10.1109/JIOT.2021.3063686 (up to cm-level), more stringent privacy and security, and so
2327-4662
c 2021 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: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on March 24,2022 at 12:00:51 UTC from IEEE Xplore. Restrictions apply.
11892 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 15, AUGUST 1, 2021
on. Second, use cases in 6G will be more multitudinous and latency. From the perspective of system, edge intelligence with
complex, resulting in different and even conflicting require- machine learning is enabled by edge computing to manage IoT
ments in different use cases. Third, several trends about 6G systems using an intelligent method.
have started to emerge, which are summarized as follows.
1) More bits and spectrum, and denser networks. C. Breakthrough Technologies
2) Convergence of various communication systems.
As an omnipotent network, 6G is enhanced by a number
3) Convergence of communication, caching, computing,
of breakthrough technologies, including machine learning and
control, sensing, and localization (4CSL).
blockchain.
4) From network softwarization to network
As one of the most powerful intelligence enabling tech-
intelligentization.
nologies, machine learning has been widely used for different
5) From centralization to distribution.
aspects of the IoT-enabled applications, ranging from the
As a result, 6G maintains the ability to connect millions of
application layer and the network layer to the perception layer.
devices and applications seamlessly with performance guar-
In the application layer, machine learning is widely used for
anteed. Thus, 6G plays a major role in supporting massive
task offloading and resource allocation. In turn, edge comput-
interconnectivity in IoT with highly diverse service require-
ing in the application layer provides storage and computation
ments. To enable massive IoT, 6G will provide a new network
capability to enable edge intelligence. In the network layer,
architecture and breakthrough technologies to meet their
network intelligentization and automation are the primary
demands.
goals of the IoT systems in 6G. Machine learning is recently
being adopted in wireless systems to address the related
B. New Network Architectures for Massive IoT challenges and to pave the way for future massive IoT com-
munications [13]–[15]. Toward the future IoT networking in
With human activities expanding to the extreme environ-
6G, machine learning algorithms are widely used for multiple
ment, e.g., higher altitudes, outer space, oceans, and deep
resource allocation, power allocation, transmit scheduling,
under the sea, an ubiquitous (covering Earth, sea, sky, and
traffic offloading, and so on. In the perception layer, machine
space), everything-connected (IoE), omniscient (with various
learning is used for autonomous control for different IoT sce-
sensors), and omnipotent (4CSL) network should be built to
narios, e.g., movement control for autonomous robots (e.g.,
truly realize the connection anytime and anywhere with diverse
tactile Internet, smart factory, and remote surgery), driving aid
requirements. To achieve this goal, a four-tier network archi-
for autonomous driving, and intelligent management for smart
tecture enhanced by edge computing is provided by 6G, i.e.,
grid [16].
space–air–ground–underwater/sea networks.
In addition to intelligence, another stirring premise promised
The space tier contains various types of satellites, which
by 6G is distribution, which exactly hits the bullseye of
aims to provide Internet connections for some extreme envi-
the future massive IoT systems. The current centralized IoT
ronment, e.g., rural areas, mountains, and so on. In this tier,
network model, in which IoT devices use a single gate-
the very low Earth orbit (VLEO) satellites are promising to
way to transfer data between them and connect through a
provide high data rates, low round trip latency, and accurate
cloud server, is no longer suitable for the future massive
localization with the lowest orbit [4], [5]. The air tier con-
IoT devices and the volumes of data they share due to its
sists of UAVs, airships, and balloons, which are aerial mobile
shortcomings, i.e., high costs of centralized cloud mainte-
systems to complement the terrestrial networks with its flex-
nance and networking equipment, low interoperability due to
ibility. For example, UAVs could move closer proximity to
restricted data exchange with other centralized infrastructures,
the ground IoT devices to collect data or acting as a com-
and severe security issues due to the untrustworthy single
puting hub, achieving higher throughput rates and conserving
gateway and centralized cloud server. Blockchain, a decen-
the energy of less-capable IoT devices [6], [7]. As the main
tralized distributed ledger, is recently regarded as the key to
way to acquire services for most IoT-enabled applications, the
solve many of the problems faced with the current model
ground tier refers to the legacy wireless networks, e.g., cellu-
and improve security [17]–[19]. First, dynamic network man-
lar networks, wireless local area networks, VLC, and so on,
agement is enabled by blockchain with decentralization and
where terahertz communications are promising to achieve the
low cost. Second, with a unified authentication system in
ambitious goals of 6G [8]. In the underwater/sea tier, opti-
blockchain, interoperability among different IoT systems is
cal communications play a vital role in providing Internet
improved. Third, data stored on various nodes eliminate the
services for distributed nodes over the broad or deep sea as
single point of failure. Confidentiality, integrity, and authentic-
the water exhibits different propagation characteristics from
ity of the data are protected by immutability, anonymity, and
the land [9], [10].
encryption of the blockchain [20], [21].
For the IoT-enabled applications that require real-time oper-
ations and decentralized services, edge computing is regarded
as a key enabler [11], [12]. From the perspective of users’ D. Review of Related Overview/Survey Articles
level, edge computing directly helps IoT devices execute their To the best of our knowledge, there is no detailed survey
tasks, e.g., rendering for VR, decision making for autonomous paper dedicated to massive IoT enabled by 6G. Chettri and
driving, and so on, which outperforms the centralized cloud Bera [22] presented a comprehensive survey on IoT toward
computing technology due to its distributed nature and low 5G, while Akpakwu et al. [23] talked about 5G networks for
GUO et al.: ENABLING MASSIVE IoT TOWARD 6G: COMPREHENSIVE SURVEY 11893
Fig. 1. Roadmap of this article.
IoT from the perspective of communication technologies and applications, visions of 6G, and the role of 6G for IoT in terms
challenges. Huang et al. [24] presented an overview of wireless of network architecture and breakthrough technologies.
evolution toward green 6G networks, where new architectural
changes related to 6G and related potential technologies are E. Contributions
discussed. You et al. [25] discussed the visions and enabling In this article, we present a comprehensive survey of
technologies of 6G thoroughly, which provided an insight- massive IoT enabled by 6G. We mainly identify four
ful picture of 6G. Focusing on breakthrough technologies aspects, on which we focus drivers and requirements, visions
for IoT, Sharma and Wang [16] discussed the current issues of 6G, network architecture, and breakthrough technolo-
in ultradense cellular IoT networks and machine learning- gies. The main contributions of this article are summarized
assisted solutions. Hussain et al. [26] provided a survey on as follows.
different aspects of resource management in cellular and IoT 1) Compared with the other survey papers related to this
networks that leverage machine learning techniques. Several topic, we provide a comprehensive survey on massive
surveys [18], [19], [27] discussed blockchain for IoT from IoT enabled by 6G, where the challenges of future
different aspects. Although these works have laid a solid foun- massive IoT are reviewed and the roles of 6G for mas-
dation on 5G for IoT, 6G, and breakthrough technologies for sive IoT are presented from the perspectives of network
IoT, the future massive IoT-enabled applications and the role architecture and breakthrough technologies.
of 6G for massive IoT have not been covered in the existing 2) The visions of 6G are presented, including core tech-
surveys. To fill this gap, we investigate the future IoT-enabled nical requirements, use cases, and trends of 6G. The
technical requirements in terms of data rate, latency, etc., approaches, which will create new verticals in turn. All these
are summarized, along with use cases of 6G. Several evolving IoT-enabled applications will propose more robust
trends associated with 6G are presented, encompassing requirements to the wireless communication networks. Next,
more bits and spectrum, and denser networks, conver- we aim to present the evolution of IoT by summarizing
gence of various communication systems, convergence emerging IoT-enabled applications and their requirements.
of communication, caching, computing, control, sensing, 1) IoT-Enabled Applications and Corresponding
and localization (4CSL), from network softwarization Requirements: Instead of only exchanges of voice, image,
to network intelligentization, and from centralization to or video in the earlier and current generations, people are
distribution. exploring new forms of IoT-enabled interactions in the future,
3) A new four-tier network architecture enhanced by edge including holographic communications, five-sense commu-
computing for massive IoT is reviewed. The promising nications, and WBCI, which can lead to a true immersion
technologies in each tier are presented, including VLEO into a distant environment. Powered by near-real-time and
satellites, UAVs, terahertz communications, VLC, and true-immersive experiences in personal communication using
optical communications. holograms and five senses and autonomously operating
4) The breakthrough technologies in 6G for massive IoT machinery in the industry as new fundamental media-objects,
are reviewed, e.g., machine learning and blockchain, new verticals emerge, including smart healthcare, smart edu-
which provide intelligence and distribution, respectively. cation/training, industry Internet, fully autonomous driving,
The applications and open issues are also summarized. and super smart city/home.
Holographic Communications: The first new form of inter-
F. Paper Organization actions refers to holographic communications, which is capa-
We structure this article in a manner shown in Fig. 1. We ble of projecting full-motion 3-D images in real time. This
begin by summarizing the motivations behind 6G, where the technology captures images of people and/or objects, presents
IoT-enabled applications and corresponding requirements, lim- in reality or at a remote location, and transmits these images
itations of 5G, and necessities to develop 6G are presented and related sounds to the receiver. In this way, it makes
(Section II). It helps define the motivation and contributions objects or people along with the real-time audio information
of this article. Since 6G is not defined yet, we provide its present at a different location and appear right in front of the
visions in Section III, highlighting its technical requirements, users, resulting in a closer-to-reality experience than VR and
use cases, and trends. To achieve 6G visions and enable mas- AR. Holographic type communications will have a big part
sive IoT, modifications to existing network architecture and to play in the industry, agriculture, education, entertainment,
cloud computing technologies as well as the involvement of and in many other fields. Due to rich details to transmit, the
breakthrough technologies are essential. We discuss the sup- amount of data necessary to stream holographic media can
porting network architecture and breakthrough technologies, as be very large even after compression, which calls for very
well as open issues in Sections IV and V, respectively, aiming high throughput in the range of hundreds of gigabits per sec-
at helping researchers understand the roles of 6G. To show how ond or even terabits per second to support such capabilities.
6G supports massive IoT, a use case of fully autonomous driv- Besides, an additional ask from networks to provide reliabil-
ing is presented in Section VI. Finally, this study is concluded ity and timeliness is required to eliminate any jitter since that
in Section VII. will immediately degrade interactive applications’ behaviors.
Concisely, the requirement of holographic communications is a
II. D RIVERS synergistic mix of URLLC and eMBB to guarantee low latency
and high data rates.
Nowadays, many IoT-enabled applications emerge, such Five-Sense Communications: Despite tactile transmission
as holographic communications, five-sense communications, and traditional human interaction in terms of the exchange of
WBCI, and new verticals. The IoT system is evolving toward voice, images, and videos, researchers also state the trend of
super massive in number and convergence of 4CSL, which new forms of remote human interactions [13], the so-called
needs higher key performance indicators (KPI) and more com- immersive five-sense or five-dimension (5-D) communica-
plex usage scenarios. As known, it is a rather daunting task for tions. The five-sense media will integrate all human sense
5G to support this evolving IoT system due to its limitations. information, including sight, hearing, touch, smell, and taste.
In this section, we try to summarize the main drivers behind This technology detects sensations from the human body and
6G by introducing the evolution of IoT, limitations of 5G, and the environment and integrates sensations by using the neu-
necessities to develop 6G. rological process. Then, the information is transferred to the
receiver at a remote location, leading to a truer immersion
A. Evolution of IoT into a distant environment. Such multisensory applications
IoT in 5G is transforming and bringing industrial revolution (e.g., a remote surgery), combined with VR/AR or holographic
4.0 in every aspect of human life, including smart healthcare, communications, will constitute truly immersive services for
smart education/training, industry Internet, autonomous driv- 6G. It requires a joint design considering not only engi-
ing, smart homes, and cities. At the same time, people are neering (i.e., wireless, computing, and storage) requirements
exploring new forms of IoT-enabled interactions in the future, but also perceptual (e.g., human senses, cognition, and phys-
including holographic, five-sense, and even brain-computer iology) requirements, which jointly determine the service
performance. In view of latency and data rate, the require- Industry Internet: 6G will facilitate a variety of vertical
ments of five-sense communications are a blend of traditional industries, e.g., electricity, manufacturing, delivery, and ports.
uRLLC and eMBB with accurate localization ability, as well Based on AI, full automation will be provided by 6G with
as incorporated perceptual factors. its ultramassive connectivity capability and ultra reliability,
Wireless Brain–Computer Interfaces: Another potential which means that automatic control of processes, devices, and
form of interactions refers to WBCI, also known as wire- systems is enabled by 6G. By transmitting data to clouds
less mind-machine interfaces (WMMI), which are interfaces or edge clouds and analyzing the data, decisions are made
that use human thoughts to interact with machines and/or the intelligently to achieve automatic manufacturing. In this pro-
environments [28]. This technology first reads the neural sig- cess, error-free data transfer is ensured by 6G. With the
nals generated in the human’s mind with a certain number help of VR/AR/holographic communications, remote main-
of electrodes and then translates these acquired signals into tenance can be enabled by 6G, in which experts in remote
commands that a machine can understand [29], thus achiev- locations and workers in present can work together timely
ing control or other functions, e.g., turn on a light. WBCI is to solve problems, resulting in higher efficiency and lower
a communication pathway between the brain and the exter- costs. Another service refers to remote control, which con-
nal peripheral devices, which is a promising approach to trols machines remotely to ensure the safety of workers and
control the appliances that are used daily in smart cities, reduce costs. It requires rigorous low-latency, broadband, and
homes, and medical systems in a more simple and intelli- reliable transmission of 6G.
gent way. Beyond healthcare and smart cities/home scenarios, Fully Autonomous Driving: Equipped with multiple high-
the recent advent of WBCI revolutionizes this field and intro- definition cameras and high-precision radar sensors and sup-
duces new use case scenarios, ranging from brain-controlled ported by 6G, fully autonomous driving can be achieved,
movies to fully fledged multibrain-controlled cinemas [30]. which means that the vehicle performs all driving tasks
Coupled with tactile Internet or haptic communications and and there is not even a cockpit (opening up new mobil-
related ideas, in which the functions of emotion-driven devices ity possibilities for people with disabilities, for example).
can match the users’ mood, WBCI will constitute important The core functions of autonomous driving include percep-
6G use cases [31]. WBCI requires high data rates, ultralow tion, planning, and control [32]. The information (including
latency, and high reliability, as well as powerful computation the accurate vehicle location and target recognition) generated
capability, which is similar to VR/AR but much more sen- by various sensors, e.g., image sensors or cameras, millimeter-
sitive than VR/AR to physical perceptions and necessitates wave/terahertz radar, and light detection and ranging (LIDAR),
Quality-of-Services (QoS) and Quality-of-Experience (QoE) is as the input of the perception layer, which can be regarded
guarantees. as the prerequisite to realize autonomous driving. The instruc-
Smart Healthcare: 6G can help build smart healthcare tions of the planning layer include following, overtaking, and
systems, where a reliable remote monitoring system, remote accelerating, which depend on the input information, i.e., the
diagnosis, remote guidance, and even remote surgery can be information from the perception layer and the feedback from
facilitated by 6G. 6G with high data rate, low latency, accurate the control layer. The control layer is in charge of imple-
localization, and ultra reliability will help to quickly and reli- menting the specific control over the vehicles according to
ably transport huge volumes of medical five-sense data, which the instructions issued by the planning layer, including throt-
can improve both the access to care and the quality of care. tle, brake, and gear control. The key challenge to achieve
Cooperated with artificial intelligence (AI), the data can be fully autonomous driving is how to meet the stringent safety
better analyzed by doctors to make accurate diagnoses. With demands when faced with different driving conditions. With
blockchain, personal data can be privately and safely shared the help of AI, MEC, and reliable and low-latency transmission
among the world to contribute to the development of medicine. of 6G, the complex information can be handled in time by 6G.
Smart Education/Training: Smart education/training will Besides, the information can be shared safely with blockchain
benefit from 6G wireless systems because innovations, to enhance the performance of autonomous driving.
e.g., holographic communications, five-sense communications, Super Smart City/Home: The superior features of 6G will
high-quality VR/AR, mobile-edge computing, and AI, will lead to significant improvement of life quality, intelligent mon-
help build smart education/training systems. With the support itoring, and automation to accelerate the building of super
of the above techniques, it allows students to view structures smart cities and homes. A city is considered to be smart when
and models in 3-D form and even to be taught by a famous it can run intelligently and autonomously by collecting and
teacher at a remote location, thus achieving interactive and analyzing mass quantities of data from a wide variety of indus-
immersive online education. For training, by illustrating the tries, from urban planning to garbage collection, which can
processes (live) in holography and interacting with objects or make better use of the public resources, increase the qual-
other trainers, it helps learners retain more information, reduce ity of the services offered to the citizens, as well as reduce
high costs, and avoid being in dangerous environments than the operational costs of the public administrations. The use
traditional training methods. 6G can also help build intelligent of smart mobile devices, autonomous vehicles, and so on
classes, in which data are collected by sensors, and sent to the will make the cities smarter in 6G. A smart home is not
clouds or edge clouds to be analyzed. Then, the results can only simply a residential or commercial building equipped
be used to improve the quality of education to better interact with Internet-connected smart devices to help people manage
with students. and monitor a range of appliances and systems from mobile
TABLE I
C OMPARISON OF A PPLICATIONS IN 5G AND 6G volume of IoT data, which is generated in a nearly real-time
fashion. Second, in view of the quantum of IoT data, more
computation power will be needed for data processing and
analysis. Third, the aforementioned IoT-enabled applications
in 6G usually need more stringent latency requirements com-
pared with 5G. Fourth, with more data involved, the desire for
data security/privacy protection will be more strong.
From Communication, Caching, and Computing (3C) to
4CSL: For IoT, it is important to collect, store, query, under-
stand, and utilize the raw sensor data. In 5G, a number of
works studied the convergence of communication, caching,
phones but also an intelligent entity with instantaneous and and computing in the case of edge computing [33], which
distributive decision-making capabilities. Besides, people can can be used for data transmission, storage, and processing.
control light, heat, or multimedia entertainment by voice or However, it is not sufficient for IoT in 6G. For example,
just mind (brain–computer interfaces, BCI) or leave them all fully autonomous driving, which is context aware, needs to
to AI, which can analyze your behaviors to make people’s exploit the localization and sensing information to construct
lives safer and easier. 6G will make smart home a reality. To the environment accurately. Simultaneous localization and
achieve smart city/home, it poses key challenges to the con- mapping methods are required to enable holographic commu-
nectivity and coverage capability of 6G, since there are so nications or enhance the navigation of autonomous vehicles
many sensors and intelligent terminals. and drones. Remote control with ultrareliability is the key for
2) Summary: 5G is the first generation that is specially industrial Internet and five-sense communications. Hence, the
designed for vertical IoT use cases, including VR/AR, tactile convergence of 4CSL is necessary for IoT in 6G.
Internet, wearable devices, vehicle to everything, smart verti-
cals, and so on. As shown in Table I, each application of 5G B. Limitations of 5G
will evolve to be a part of 6G, which aims to boost every aspect The limitations of 5G come from not only its architecture
of human life. For example, holographic communications and but also its technologies, both resulting in a mismatch between
five-sense communications in 6G will bring a truer immersive its capabilities and the requirements of all aforementioned IoT-
experience when compared with VR/AR and tactile Internet in enabled applications. The limitations of 5G will be described
5G. Instead of vehicle to everything, fully autonomous driv- in detail as follows.
ing will bring more convenience to human life. Furthermore, 1) The first limitation lies in the best performance that 5G
6G will explore communications in space and deep sea rather could provide in terms of data rate and latency. From
than focusing on the ground like 5G. Build upon the unique the perspective of data rate, a peak data rate of 20 Gb/s
creation of new interactions, new verticals emerge and IoT in and an experienced data rate of 100 Mb/s are touted,
the future 6G will change a lot. The main differences of IoT which indeed sounds fantastic. But this projection over-
in 5G and 6G are described as follows. looks the critical fact that the capacity must be shared
Super Massive in Number: Owing to the aforementioned among multiple users and assumes unrealistic conditions
applications, new device forms emerge. First, smartphones are that result in a condition that actual throughput capacity
likely to be replaced with lightweight extended reality devices for wireless users is often only 15% of the peak data
(google glasses, for example) to deliver an unprecedented res- connection rate. In addition, the required data rate of an
olution, frame rates, and dynamic range. Second, wearable ideal VR user could reach several Gb/s and even Tb/s,
displays, mobile robots and drones, and specialized processors and it is far away from meeting the requirement of high-
will be used for high-resolution imaging and sensing to make quality VR for 5G. From the perspective of latency, a
holographic experience or telepresence a reality. Third, with latency of 1 ms is promised by 5G. However, latency
fully autonomous driving coming true in 6G, more vehicles of less than 1 ms is needed for many IoT-enabled appli-
will be connected to the Internet for ecologically sustainable cations, e.g., fully autonomous driving, tactile Internet,
transport and logistics. All these mobile devices embedded with and remote control.
various sensing abilities (called sensors) will contribute to the 2) The second limitation comes from the supporting service
growth of IoT in volume. According to the new analysis from classes of 5G, i.e., eMBB, mMTC, and uRLLC. mMTC
IHS Markit, the number of connected IoT devices will jump and uRLLC are designed for IoT. uRLLC supports low-
12% on average annually and reach 125 billion in 2030 from 27 latency transmissions with very high reliability for small
billion in 2017. With this trend continuing, it can be predicted payloads from a limited amount of user devices, which
that the number of IoT devices will be super massive in the are active according to patterns typically triggered by
future 6G, which will far exceed what 5G promises to support. outside events (e.g., alarms). While mMTC aims to
More Intensive and Sensitive: IoT in future 6G will be more connect a massive number of IoT devices, which are
data intensive, computation intensive, delay sensitive, and pri- with small data payloads and only sporadically active.
vacy/security sensitive. First, evolving from VR/AR, tactile However, most of the future IoT-enabled applications
Internet, and wearable devices to holographic communications, are data intensive, computation intensive, and delay sen-
five-sense communications and WBCI brings a surge of big sitive, e.g., augmented/virtual reality (AR/VR) online
games, which exceed the range of any of these three driving, will be involved in 6G. Hence, how to ensure
typical scenarios. Obviously, 5G will not be able to sup- privacy and security is a key challenge in the future.
port such scenarios that combine any two classes or a
synergistic mix of all three classes. For example, the C. Necessities to Start the Study on 6G
requirements of five-sense communications are a blend Besides the above drivers, we provide the following neces-
of traditional uRLLC and eMBB. sities of 6G.
3) The third limitation of 5G refers to its connection capa- Ten-Year Rule: A new mobile generation has appeared
bility, which aims to connect everything (i.e., the Internet approximately every 10 years since the first 1G system was
of everything, IoE). With technologies developing, introduced since 1982 and it needs at least ten years from
humans are exploring mountains, seas, and even space, definition to commercial deployment, which means that when
where communications should follow. Besides commu- the previous generation enters the commercial phase, the con-
nications for human, IoT devices will vastly expand cept and technology research of the next generation begins.
the geographic space for communication access, includ- Assuming that this trend continues, 6G will be a reality around
ing deploying unmanned detectors deep into the Earth, 2030. With 5G rolling out across the world, it is the perfect
sea, and space, UAVs at median altitude, autonomous time to start to study 6G.
robots that penetrate into harsh environments, and smart Catfish Effect: Unlike previous generations of mobile com-
machines that are controlled remotely. However, 5G, munication systems, 5G is also targeted at IoT/vertical industry
including the earlier generations, focuses on communi- application scenarios. With 5G networks scaling, especially in
cations over the surface, which ignore the activities in the middle and late stages of 5G, more vertical industry mem-
these extreme environments. So 5G is far away from bers will surely participate in the 5G ecosystem. Compared
connecting everything and ubiquitous connectivity is with the current conditions that are dominated by traditional
desired with time going on. operators, the in-depth participation of emerging enterprises
4) The fourth limitation of 5G is related to its coverage in (especially Internet companies with innovative thinking) in the
rural areas. People in rural areas are by far the most future will have a huge impact on the traditional communica-
underserved population when it comes to broadband tions industry, which is even revolutionary.
access. Let alone for IoT. In 2018, the federal communi- Exploding Potentials of Emerging Applications: Just as the
cations commission (FCC) found that 98% of Americans emergence of smartphones that stimulates the development of
in urban areas had access to a broadband connection, yet 3G and triggers the demand to deploy 4G, it is believed that
only 69% of rural Americans do [34], which is nowhere IoT business will also stimulate 5G at some point in the 5G
near fast enough to be useful for most modern appli- era [15], and then further stimulate the demand for future 6G
cations. In 5G, rural areas will not be left out, but it networks. To have enough imagination, it needs us to prepare
will be different there. The multi-Gb/s speeds and mas- for possible future networks and lay the technical foundation
sive capacity of 5G is an urban phenomenon to a great in advance.
extent, achieved by the huge bandwidths of mmWave
spectrum, which does not travel very far, resulting in III. V ISIONS OF 6G: C ORE R EQUIREMENTS , U SE C ASES ,
ultradense small cells. Considering the profitability, rural AND T RENDS
areas will get a form of 5G called “low-band” or “sub-6”
5G, which will have less capacity. Thus, how to provide 6G envisions a data-driven society, which is enabled by
IoT-enabled services for the last 4 billion people with near instant and unlimited connectivity to anything, ranging
broadband services in rural areas still remains unsolved from tiny static sensors to autonomous objects, from anywhere,
in 5G, which is what rural residents probably most want including the rural areas, mountains, and seas, and anytime.
and are most frustrated. With 6G not defined, this section introduces the core technical
5) The fifth limitation is the new KPI arising from emerg- requirements along with the use cases of 6G, followed by the
ing applications, including privacy and security. With the arising trends of 6G.
capabilities promised by 5G, nearly all aspects of human
life will be connected to communication networks, but A. Core Requirements and 6G Scenarios
it also provides a fertile hunting ground for criminals With technologies developing and time going on, the current
who are able to access the communication between a support scenarios will not be able to support the applications in
device and a network in order to intercept conversations the future, as stated in the limitations of 5G. Hence, new tech-
or steal data [35]. The need for robust security mecha- nical requirements will be imposed and new service classes
nisms is underscored across all network segments of 5G. should be supported in the future. In this section, core tech-
However, the research from the University of Dundee nical requirements will be introduced and new service classes
and so on found that while data protection in 5G has will be defined.
improved on that offered in the 3G and 4G versions, crit- 1) Core Technical Requirements:
ical security gaps still exist [36]. For 6G, this problem 1) Peak Data Rate: Peak data rate is one of the key techni-
will keep exacerbating, since so many sensors, which cal indicators that has been pursued since 1G. Without
are used in IoT-enabled applications, such as health- a doubt, 6G will further enhance the achieved peak data
care, remote education, smart homes, and autonomous rate, which is expected to reach up to Tb/s [15], [31].
First, it can be predicted from the history of peak data to be dependent, the most extreme case of which can be
rate improvement from 1G to 5G. Second, the demands one billion bits transmission with only one-bit error and
of big data transmission for intelligence and emerging a delay of 0.1 ms.
applications (e.g., high-quality VR/AR and holographic 6) Full Coverage: With the advancement of science and
communications) will far exceed the current wireless technology, human activities are further expanded to the
applications. extreme environment, e.g., higher altitudes, outer space,
2) Sub-ms end-to-end Latency: Low latency is pivotal to oceans, and deep under the sea. Communication nodes,
many wireless applications. For example, it makes the especially the IoT devices, will spread over a wider
users feel dizzy with a latency of more than 20 ms for area, which is inseparable from human social activi-
VR/AR users. 20 us ∼ 10 ms is required for machine- ties. As a result, an ubiquitous (covering Earth, sea,
to-machine communications (M2M). For remote surgery sky, and space), everything-connected (IoE), omniscient
or other remote control in most of the industrial use (with various sensors), and omnipotent (based on big
cases (e.g., industrial IoT and IIoT), a higher require- data and deep learning) network should be built to truly
ment in terms of latency is proposed [37], e.g., one realize the connection anytime and anywhere. The ulti-
submillisecond. mate goal of mobile communications is the provision of
3) Higher Energy Efficiency: The energy consumption of ubiquitous super connectivity on the global scale [41],
mobile networks has become an ignorable problem i.e., anywhere on Earth, outside or inside buildings, pros-
around the world [38], which will not only generate a perous cities, or remote rural areas. Hence, full coverage
huge amount of carbon emissions that is harmful to the should be provided by 6G.
environment but also accounts for a considerable part of 7) Privacy and Security: Individuals should have the
the operating expenses. First, with ultralarge throughput, rights to choose under what conditions his/her per-
ultrabroad bandwidth, and ultramassive ubiquitous wire- sonal information can be accessible by others. However,
less access points (APs), 6G will bring unprecedented privacy violations occur sometimes, e.g., unapproved
huge challenges to energy consumption. Second, vari- auxiliary utilization and unauthorized gathering of indi-
ous sensors will be ubiquitous in human’s life, which vidual data and unauthorized access of securely stored
need bulky energy to operate and manage these sensor individual data. With new applications emerging (e.g.,
networks. Hence, green communication is particularly autonomous driving, healthcare) and technologies matur-
urgent for 6G, e.g., energy efficiency of 10–100 times ing (e.g., cloud and MEC), there will involve more
(in J/bit) [39]. personal information in the future wireless networks,
4) Accurate Localization: First, to determine the location which must be protected. Therefore, an extra level of
of the mobile terminals in a cellular system is mainly in privacy and security should be provided by 6G networks
the case of an emergency call, which is also the demand compared to the current 5G and the earlier generation
of governmental institutions. Second, the location of of networks.
the mobile device can be exploited for commercial 8) Programmable Service to Deal With Different
services to obtain revenue, e.g., the so-called location- Requirements: With the breakthrough and development
based services. Examples of such services include nav- of new technologies, new business and scenarios
igation, mapping, asset tracking, geo-marketing and are spawned and users’ demands will tend to be
advertising, social networking, location-sensitive billing, more diversified and personalized. Hence, on-demand
AR, etc. Third, the location information can be also services should be provided by future 6G networks,
used for network optimization to improve network which are designed to meet the individual needs of
efficiency and communication capacity. Examples are users and provide users with extreme experiences.
network management, resource management for device Dynamic and extremely fine-grained services should
to device (D2D) communications, intelligent transporta- be enabled, which enable users with different service
tion systems, radio reconfigurable spectrum, vehicu- types, service levels, and free combinations of different
lar ad hoc networks, etc., [40]. Fourth, new applica- services according to their demands. Considering
tions, e.g., autonomous driving, autonomous factory, and this, to provide the end-users with tools allowing the
remote surgery, which involve complicated operations, individuals to configure/program personal services is a
pose a stringent localization requirement to 6G to ensure promising solution.
the safety of users and protect the values and peo-
ple, both indoors and outdoors, which can reach up to B. 6G Use Cases
the level of several centimeters for indoor localization Beyond imposing new performance metrics (e.g., techni-
and 1 m for outdoor localization. Hence, more accurate cal requirements), the new technological trends will introduce
localization capability is required in 6G. new use cases, i.e., service classes, to redefine the applica-
5) High Reliability: Many applications with remote con- tion types by morphing classical eMBB, uRLLC, and mMTC,
trol functions are emerging, such as autonomous driv- which depend on only one single constraint. Though differ-
ing, autonomous factory, and tactile Internet, which all ent researchers have different opinions on classification, what
require high reliability to ensure safety. The reliability is commonly agreed with is that the classification should be
requirements of different 6G applications are predicted based on the requirements of different applications. Following
this common rule, several novel forward-looking scenarios,

i.e., service classes, are defined in [31], [39], and [42].
Saad et al. [31] defined four service classes: 1) mobile
broadband reliable low latency communication (MBRLLC);
2) massive URLLC (mURLLC); 3) human-centric service
(HCS); and 4) multipurpose communications, computing,
control, localization, and sensing (3CLS) and energy ser-
vice (MPS). MBRLLC provides services that require any
performance within the rate-reliability-latency space, which
merges eMBB and uRLLC. Examples of applications are
AR/VR and WBCI. mURLLC scales classical uRLLC across Fig. 2. Evolution of wireless technologies.
the device dimension that generalizes 5G uRLLC with legacy
mMTC, which brings forth a tradeoff among reliability,
latency, and scalability. Examples of mURLLC are classical require much higher data rates than 5G, and a 1000× increase
IoT, blockchain, and autonomous robots. HCS is tightly cou- in data rate should be delivered by 6G to cater to these appli-
pled with humans, the performance of which depends largely cations, such as holographic communications and WBCI [31].
on the physiology of the human users and their actions rather To support these extremely high data rate requirements in 6G,
than raw rate-reliability-latency metrics. Typical application new spectral bands will be explored [44]. In this context, the
refers to WBCI. MPS aims to provide 3CLS services includ- terahertz spectrum (0.1–10 THz) has been envisioned as a key
ing their derivatives, and can also offer energy to tiny devices technology to fulfill this requirement [45], [46], as well as the
via wireless energy transfer, which is important to the appli- visible light spectrum (430–790 THz) and unlicensed spectrum
cations, such as telemedicine, special cases of AR/VR, and (e.g., 2.4 GHz/5–7 GHz/57–71 GHz). It is worth noted that the
connected robotics and autonomous systems. handsets will combine more modes of operation from multiple
Zhang et al. [39] introduced five service classes, including frequencies in addition to cellular, WiFi, and Bluetooth [47].
further eMBB (FeMBB) to provide higher data rate, extremely Providing a massive amount of available bandwidth up to sev-
URLLC (eURLLC) to meet the demand of lower latency, eral THz (three orders of magnitude higher than unregulated
ultra mMTC (umMTC) to further scales the device amount, regions at mmWave frequencies), terahertz spectrum has the
long-distance and high-mobility communications (LDHMC) to potential to provide several Tbps data rates. By using the ter-
support deep-sea sightseeing and space travel, and extremely ahertz spectrum, the communication range of which is limited
low-power communications (ELPC) to enable e-health and to 50 m [48], an unprecedented level of network densifica-
nanorobots. tion, and base stations (BSs) miniaturized and embedded all
In the opinions of Zong et al. [42], revolved core require- around us will be needed [49]. At the same time, hundreds
ments for 6G will lead to ubiquitous mobile ultrabroadband of simultaneous beams will be a part of 6G, which helps
(uMUB), ultrahigh data density (uHDD), and ultrahigh-speed- yield much higher data rates. In addition, the licensed spec-
with-low-latency communications (uHSLLC). uMUB aims to trum, unlicensed spectrum, and VLC will play crucial roles in
provide ultrahigh data rate and ultralow latency ubiquitously achieving the ultrahigh data rate.
by integrating space and terrestrial networks. uHDD requires Convergence of Various Communication Systems: In the ear-
huge wireless capacity and high reliability, while uHSLLC lier generations of mobile communications, researchers have
aims at ultrahigh rates and ultralow latency. tried to integrate the satellite networks with the terrestrial wire-
It is worth noting that the above definitions are quite differ- less networks [50], as well as in 5G [51], though it was never
ent and not considerate, since the affiliation of one application integrated into cellular handsets [47] due to a series of compli-
may be confusing based on the above definitions. In summary, cations in terms of technical and political challenges. In 6G,
novel service classes aim to provide compound services, which this trend will continue, which will extend to a convergence
depend on more than one single constraint and, more impor- of various communication systems, focusing not only on ter-
tantly, are with varying granularities [43], as well as to support restrial wireless networks but also on airborne communication
new functions, such as localization, mapping, computing, pri- networks [52]. Airborne communication networks are engi-
vacy, and so on. What is concrete is that 6G will provide more neered to utilize various aircraft, including satellites, airships,
complex and heterogeneous services than ever before. and balloons on high-altitude platforms, and UAVs on low-
altitude platforms [52], while a terrestrial wireless network is
C. Trends of 6G a typical heterogeneous wireless network comprised of macro-
The applications above, limitations of 5G, and the develop- cells, microcells, picocells, or femtocells, WiFi, VLC APs, and
ment of new technologies will lead to new system-wide trends, even D2D devices. In this converged system, each communi-
which will be discussed in the following. cation technology maintains its advantages and disadvantages,
More Bits and Spectrum, and Denser Networks: Fig. 2 which could compensate each other to improve the system
presents the evolution of wireless technologies. From the his- performance, e.g., coverage.
tory of mobile communications from 1G to 5G, it shows the Convergence of Communications, Caching, Computing,
trend of pursuing higher data rates. This trend will continue in Control, Sensing, and Localization (4CSL): As stated by
6G. As discussed above, most of the driving applications of 6G Saad et al. [31], the past four generations of cellular systems
maintained one exclusive function: wireless communications. management mechanisms are desired. With emerging IoT appli-
When it comes to 5G, more functions are involved, e.g., cations containing personal information and the popularity of
caching and computing. It is believed that 6G will completely cloud services, future wireless networks are generating and
disrupt this premise through a convergence of diverse func- carrying massive data. In view of privacy, the data are usu-
tions, including communications, caching, computing, sensing, ally stored on mobile devices, which needs distributed training
control, and localization. The drivers behind this convergence when applying AI [55], i.e., distributed computing and intel-
mainly refer to the inherent features, e.g., tracking, control, ligence. A new mechanism is needed to enable the sharing
localization, and computing, of applications, such as AR, of a large amount of information without jeopardizing users’
autonomous driving, and smart homes. Hence, 6G is envi- privacy, which is also required to be distributed due to the
sioned to be a multipurpose system that could deliver multiple complexity. A promising solution refers to blockchain, which
4CSL services and functions. Especially, sensing and localiza- could enable safe data sharing with its nature of distribution
tion services will enable 6G systems to provide users with a and immutability. In addition, FCC states that blockchain can
3-D mapping of the radio and physical environments. Hence, help remake spectrum access, which reboots the old binary
to tightly integrate and manage 4CSL functions is vital to 6G, system of licensed or unlicensed airwaves [49], thus improv-
and it will be enabled by the evolutions pertaining to previous ing spectral efficiency. Besides spectrum, blockchain can be
trends. also used for resource management in terms of computing,
From Network Softwarization to Network Intelligentization: caching, and so on.
It is envisioned that 6G will take network softwarization
toward network intelligentization, i.e., a new level of soft- D. Full Picture of 6G
warization [53]. In 5G, two key technologies, SDN and NFV, A schematic diagram of future 6G wireless systems is
have shifted modern communication networks to software- shown in Fig. 3, which includes applications, services, archi-
based virtual networks, which is enabled by network slicing to tecture, and management. As shown in this figure, the appli-
create multiple virtual networks atop a shared physical infras- cations and services encompass every aspect of human life
tructure to meet the diverse and heterogeneous requirements. and society, which are both human centric and machine cen-
In 6G, diverse capabilities, such as communication, caching, tric. It provides more ways, such as through gestures, voices,
computing, and even energy harvesting, should be supported and minds, for communicating and interacting with mobile
by the network entities to support the AI-based applications. users and intelligent devices [39], and the interaction forms are
More advent IoT functionalities, including sensing, control, diverse, such as hologram videos and five senses. As a result,
localization, data collection and analytics, caching, and com- immersive experiences and super smart verticals based on IoT
puting, should be supported by 6G. Furthermore, intelligent will be supported by 6G. For the network architecture, 6G
surfaces and terahertz communications will be embraced in is envisioned to incorporate multiple innovative technologies,
6G. All these contribute to a more complex and heterogeneous such as space–air–ground-underwater/sea networks and cloud
future network, which calls for a flexible, adaptive, and more technologies. First, the integration of space, air, ground, and
importantly, intelligent architecture. As a result, softwarization underwater/sea networks will constitute a large-dimensional
is not going to be sufficient for 6G, which needs to be fur- 6G network, which will provide ubiquitous and unlimited
ther improved to meet these challenges. Recent successes have connectivity. By adding terahertz spectrum, visible light spec-
motivated AI to form part of 5G. However, the combination of trum, and unlicensed one, full spectra will be featured by this
AI and 5G only scratch the surface and is only expected to be network architecture. To make full use of full spectra, mas-
operated in isolated areas, which involve powerful computing sive MIMO, full duplex, and beamforming are necessary to
facility and massive training data. It is envisioned that a part achieve high network capacity and data rates, which need to
of 5G will realize some form of AI and AI will become one be adjusted and upgraded of course. Especially, the usage of
of the core components in 6G [54], from the application layer terahertz spectrum will lead 6G to be even denser. For network
to the physical layer, i.e., ubiquitous intelligence. management, facing with more types of terminals and network
From Centralization to Distribution: A trend from central- devices, more complex and diverse business types, and more
ization to distribution has arisen in 6G, from the architecture to complex and huge networks, intelligence and distribution are
technologies. In 5G, the weakness of the centralized architec- needed by incorporating machine learning, blockchain, and
ture and management mechanisms has been initially revealed, SDN/NFV.
an example of which is the proposal of distributed SDN, 6G vision can be summarized into immersive connec-
MEC, and edge caching to meet the increasing complexity. tivity, ubiquitous connectivity, intelligent connectivity, and
Besides the complexity, the drivers behind the distribution distributed connectivity, which constitute the overall vision of
of 6G are omnifarious, including the development of IoT 6G, “Wherever you think, everything follows your heart” [15].
applications, the increasing amount of users, and the intro- In summary, 6G will be the first generation that fully achieves
duction of AI and blockchain. In 6G, various functionalities, the digital transformation of societies by providing ubiquitous,
e.g., communication, caching, computing, sensing, localiza- near-instant, ultrareliable, and secure wireless connectivity for
tion, and so on, are integrated, and the networks tend to be both humans and machines. The incorporation of innova-
more complex and heterogeneous, especially with the increase tive technologies, such as space–air–ground–underwater/sea
of mobile users. To operate and manage huge networks with networks, machine learning, blockchain, and so on, will make
stringent requirements and limited resources, fully distributed it multidimensional, human intelligent, and fully distributed.
Fig. 3. Schematic diagram of future 6G wireless systems.
IV. N ETWORK A RCHITECTURE FOR 6G-E NABLED industry in the future wireless communications [57], [58]. The
M ASSIVE I OT: S PACE –A IR –G ROUND –U NDERWATER ongoing VLEO constellation projects, e.g., SpaceX [58] and
N ETWORK W ITH E DGE C OMPUTING OneWeb [59], plan to launch over thousands of VLEO satel-
In this section, we first introduce the space–air–ground– lites over the Earth, aiming to construct a massive VLEO
underwater/sea network and edge computing. Then, the overall constellation and cooperate with the traditional terrestrial
network architecture is described by explaining how this operators to support seamless and high-capacity communi-
network architecture can support the massive IoT in 6G. cation services. In the future 6G, massive VLEO satellites
Finally, we discuss some open issues faced by the system will be deployed in the sky to provide both diverse geo-
design. graphic coverage and enough capacity to support a wide range
of broadband communication services for residential, com-
A. Space–Air–Ground–Underwater/Sea Networks mercial, institutional, governmental, and professional clients
globally [60].
A space–air–ground–underwater/sea network is a large-
dimensional network, which consists of four tiers, i.e., a 2) Air Tier-UAVs: The air tier consists of UAVs, airships,
space-network tier formed by various satellites, an air-network and balloons, which is an aerial mobile system to complement
tier constituted by various flying BSs (e.g., high-altitude plat- the terrestrial networks by providing broadband wireless com-
forms, mobile airborne cells, UAVs, and so on), a terrestrial- munications. Compared with terrestrial networks, it has the
network tier consisting of all kinds of legacy BSs, and an advantages of low cost, flexibility, easy deployment, large cov-
underwater-network tier, including underwater hubs, ships, and erage, and no infrastructure required to offer wireless access
so on. services on a regional basis. Among these carriers, UAVs
1) Space Tier-Massive VLEO Satellites: The space tier is attract the focus of academic and industry due to its salient
composed of all kinds of satellites, constellations, and their attributes of strong line-of-sight (LOS) connection links, and
corresponding terrestrial infrastructures (e.g., ground stations). additional design degrees of freedom with the controlled
These satellites are deployed at different orbits and maintain mobility [61].
different characteristics, which can be classified into four cat- UAVs originated mostly in military applications. In recent
egories based on the altitude, i.e., geostationary Earth orbit years, the use of UAVs has quickly expanded into commu-
(GEO), medium Earth orbit (MEO), low Earth orbit (LEO), nication, commercial, scientific, recreational, agricultural, and
and VLEO satellites [56]. Among these four kinds of satel- other domains, the applications of which include acting as a
lites, VLEO satellites are promising to provide high-rate data relay, broadcasting, cargo transport, weather control, forest fire
services and accurate localization due to the lowest orbit. recognition, traffic detection, emergency search, and rescue.
Considering the outstanding performance in terms of To deploy UAVs in wireless communication networks, it offers
extremely low round trip latency and low cost of VLEO satel- cost-effective services with no or limited infrastructure, which,
lites, this technology attracts the attention of academia and as a relay, can extend the coverage of the traditional cellular
networks. UAVs can be considered as a substitute to a wireless which uses the visible light (430–790 THz) as a carrier for
recovery network in cases of terrestrial disruption. Compared the data, and offers a 1000 times greater bandwidth com-
with satellite communications, UAV communications are more pared to the RF communications [65]. The communication
cost effective and provide lower latency and better signal- through visible light holds special advantages when compared
to-noise ratio [62]. In addition, the feature of UAVs, i.e., to the other existing forms of wireless communications, such
short-range LOS, provides better communication channels and as several tens of Gb/s data rate [66], cost efficiency [67],
mobility provides flexibility in deployment. Thanks to these security [65], and large geographical distribution [65]. The
advantages, many wireless applications for UAVs have been high data rates, ease of availability, and low cost of VLC
proposed, such as mobile relays, cache in the air, mobile make it a relevant wireless communication technology and
computing cloudlets, and so on. a possible complementary technology that would cater to all
3) Ground Tier-Terahertz Communications and VLC: The kinds of future applications [68]. As a consequence, it has
ground tier refers to the legacy wireless networks, e.g., cellular been widely considered as an enabling technology of 6G [13],
networks, wireless local area networks, VLC, and so on, which [24], [39], which can be used for indoor positioning, human–
is still the main way to acquire services for most IoT-enabled computer interface (HCI), vehicular communications, and even
applications. 6G will be enhanced with higher spectrum bands underwater communications.
(e.g., mmWave and terahertz bands), better modulation and 4) Underwater/Sea Tier-Optical Communications: The
channel coding, and new spectrum utilization (e.g., flexible underwater tier aims to provide Internet services for distributed
spectrum sharing technology) to improve its system capacity. nodes over the broad or deep sea, including seabed sensors,
Terahertz communications use the terahertz electromag- relay buoys, autonomous underwater vehicles, and remotely
netic spectrum that lies in the boundary region between radio operated underwater vehicles. All these nodes collaborate with
frequency (RF) and optical frequency, which is envisioned each other to fulfill sensing, processing, and communication
as a key technology to enable real-time applications for 6G tasks for environmental monitoring, disaster precaution, off-
by alleviating the spectrum scarcity and capacity limitations shore exploration, and military operations [9]. Compared with
of current wireless systems [13], [45], [46]. Compared with the acoustic and RF methods, optical communications, which
mmWave communications, terahertz communication maintains employ optical waves as the transmission carrier, could pro-
a wider bandwidth, which induces various inspiring advan- vide the highest transmission data rate, lowest link delay,
tages. First, it has the potential to support unprecedented and lowest implementation costs for bidirectional underwater
data rates from tens of Gb/s to several Tbps than mmWave communications [10] due to water exhibiting different propa-
communications, which can enable innovative applications for gation characteristics from the land. Especially, VLC has been
diverse scenarios. Second, it is easy to track beams for indoor proven the possibility for underwater communications [69] by
wireless communications, which will enormously affect the employing a single-photon avalanche diode for transmission in
mobility of the wireless communication systems. Third, the pure seawater, which maintains high data rate and low power
reflection paths can be utilized by terahertz communication consumption [70].
systems to enhance link gains, especially for indoor appli-
cations. Fourth, thanks to the short transmission range of B. Edge Computing
terahertz communications, interference in terahertz band-based As a dynamic global network infrastructure with self-
networks can be substantially cut down with highly direc- configuring capabilities, IoT makes all things communicate
tional beams, resulting in terahertz band-based networks to with each other based on standard and interoperable commu-
be noise limited rather than interference limited, which fur- nication protocols and realize information sharing and decision
ther increases the performance. Fifth, extremely directional making [71], which involves data acquisition, exchange, and
beams narrow beams can be used to partially secure the data analysis. In addition to data analysis, edge computing plays a
at the physical layer in addition to extending the communi- pivotal role in supporting massive computation-intensive IoT-
cation range [63]. As a result, more antenna elements (i.e., enabled applications. First, these applications propose more
massive multiple-input and multiple-output, MIMO) can be robust requirements in terms of computation power, such as
packed at terahertz frequencies than at microwave (including holographic communications, high-precision VR, autonomous
mmWave), the formed beams of which can be narrower. It driving, smart factory, and so on. Second, the caching power
further facilitates the development of other applications, e.g., of edge computing could help further relieve the heavy bur-
detection radars. Due to the short transmission range of ter- den caused by massive data uploaded to the cloud [72]. Third,
ahertz communications, 6G envisions an ultradense network edge intelligence can be enabled by edge computing, which is
than ever before, which could further enhance the connectivity expected to provide intelligent services at the network edge,
capacity and data rates of 6G. In summary, all these inspiring such as training machine learning models for decision making
advantages make it ideal for transmission among IoT devices. and analyzing big data generated by massive IoT devices at
Another attractive technology for indoor IoT transmissions the network edge. In summary, the introduction of edge com-
refers to VLC. VLC, a class of optical wireless communi- puting can significantly reduce the latency, save bandwidth
cations, can be utilized as one of the promising alternative resources, and protect privacy to some extent.
methods to the existing RF-based wireless communications, As noted, this section aims to introduce the network archi-
especially for indoor communications [64]. VLC can use tecture for massive IoT provided by 6G. Since edge computing
the existing illuminating devices for communication purposes, has been studied thoroughly, we only briefly introduce edge
Fig. 4. Network architecture for 6G-enabled massive IoT: space–air–ground–underwater/sea networks with edge computing.
computing, focusing on its functions and advantages. For technologies promise superior performances for IoT, there
details on edge computing, we recommend the readers to refer are some open issues needed to be addressed before fully
to [72]–[76], and [77]. achieving these visions.
1) VLEO Satellites Communications: Satellite communica-
C. Overall Network Architecture tions are usually suffered from long latency, which consists
The overall network architecture for massive IoT in the of transmission time, propagation time, processing time, and
upcoming 6G is shown in Fig. 4, which consists of space– queuing time. Though VLEO satellite communications main-
air–ground–underwater/sea networks and edge computing. tain the lowest latency due to the lowest deployed altitude
As described above, the space–air–ground–underwater/sea compared with GEO, MEO, and LEO satellites, it is far from
network extends the traditional cellular networks by integrat- enough to meet the demands of some time-sensitive applica-
ing space, air, and underwater/sea networks, which cover all tions. Hence, it is viable to integrate with other technologies,
natural space, such as space, air, land, and ocean. It will lead such as in-network caching. Not only technical challenges,
to ubiquitous connectivity, that is, providing information assur- e.g., long latency, are associated with VLEO satellite com-
ance for any user who can access any subnetworks. With munications but also the ones related to standardization and
resorting to terahertz spectrum, visible light spectrum, and political factors.
unlicensed spectrum (for both communication and sensing), 2) UAV Communications: There are different kinds of
this network can provide ultrahigh data rate and low latency UAVs for different purposes. For example, consumer low-
for some data-intensive and mission-critical IoT applications. cost UAVs are with limited capabilities in terms of payload
Edge computing endows tiny IoT devices with additional and flying time, which are suitable for photo/video shoot-
computational and caching capabilities through computation ing applications for entertainment. Commercial or military
offloading. Rather than outsourcing storage and processing UAVs are capable of traveling long distances with heavy
functionalities of IoT data to a third party with cloud comput- payloads carried, which are mainly for strategic operations,
ing, edge computing can realize edge storage and processing such as surveillance and wide-area communication coverage,
for the cumbersome traffic created by massive IoT and fill the involving high data throughput requirements, cargo delivery,
gap between centralized clouds and distributed IoT, such as and so on. How to achieve a balance among flying time,
single point of failure, reachability, lack of location aware- carried payload, and associated costs is essential where a
ness, and latencies associated with core networks. As a result, joint optimization of these three important metrics is desir-
the incorporating of edge computing can enable applications able in the future. Although some works have been done
with the needs for mobility management, location awareness, in this direction, several years are still needed to close this
geodistribution, scalability, and ultralow latency. gap. Despite the technical aspects, regulations concerning pri-
vacy, data protection, and public safety should be also paid
D. Open Issues attention to.
As discussed above, VLEO satellite, UAVs, terahertz, opti- 3) Terahertz Communications: The research on terahertz
cal communications (especially VLC), and edge computing communications needs further investigation, the achievement
play important roles in supporting massive IoT. Though these of which relies on various aspects.
Hardware Design: Transistor and hardware materials need architectures, performance analysis, and challenges in space-
to comply with excellent high-frequency characteristics, the ground- or air-ground-integrated networks [79]. Few of them
potential materials of which is graphene. The researchers may concerned the case of space–air–ground networks, let alone
also find more suitable materials. In addition, nanoantennas space–air–ground–underwater networks. As noted, the het-
are desired to enable massive MIMO. erogeneity, time variability, and self-organization in space–
Channel Modeling: It is hard to establish the channel and air–ground–underwater networks bring new challenges, which
noise modeling in terahertz bands. A mixture of models that will seriously affect the network performances, e.g., latency,
consider path-loss mechanisms may be the solution to this throughput, and reliability for data transmission. That is, more
problem. research works are expected to concern such issues involving
Medium Access Control (MAC) Layer Design: In view of all four network tiers.
the relatively wide terahertz bands, the properties of different
carrier frequency windows will be rather different. The MAC V. B REAKTHROUGH T ECHNOLOGIES
layer protocols and network deployment schemes need to be
Considering the increasing growth of IoT-enabled applica-
redesigned. To improve the coverage and support the seam-
tions in terms of volume and requirements, it is essential to
less connection, new error control mechanisms and networking
integrate breakthrough technologies to meet their demands,
strategies should be developed, due to the high path
the most two promising of which refer to machine learn-
loss.
ing and blockchain. In this section, we mainly introduce the
Network Deployment Modeling: The future network will
motivations, applications (or functions), and open issues when
be multitier and heterogeneous. In 5G small cells, it tends
applying these two technologies to massive IoT.
to be user centric, which will be more emphatic in future
6G with even smaller cells. In addition, future networks will
be deployed in 3-D. Thus, the current Poisson point process A. Machine Learning
will not be applicable. A new network deployment model is 1) Motivation: The term “edge big data” results from the
required to consider the above facts. overwhelming volume of data, which is generated and col-
4) VLC: As discussed in [70], LEDs in VLC have lected by massive IoT devices at the network edge [80], [81].
interesting properties to be used as sensors. The usage of VLC As we know, machine learning is a data-driven technique, i.e.,
for sensing has attracted more focuses in recent years. In view the machine learning models need to be trained and tested by
of the complexity, it can be combined with WiFi or other using massive data. In this case, it just so happens that we
techniques to improve accuracy. cloud apply machine learning to massive IoT to fully unleash
Hybrid Systems: Adopted as a complementary technology the potential of edge big data by data analytics, which in turn
to the current RF-based systems, new hybrid VLC systems provides the opportunity to develop intelligent applications at
are pointed out as a future trend. To enhance the quality of the network edge.
communications, VLC is expected to cooperate with wired- From the perspective of communication networks, 6G will
based-network devices, RF, and Ethernet networks. become more complex and heterogeneous, as more techniques
Standardization: The research on VLC has conducted sev- are emerging, such as caching, edge computing, and terahertz
eral years. However, actual solutions are not interoperated due communication networks, which provide more possibilities as
to the different VLC hardware and testbeds. To better organize well as pose more challenges for network organization, man-
and optimize the research field, it calls for standardization. agement, and optimization [82]. With computing and caching
5) Optical Communications: First, optical communications capabilities throughout the network, pervasive intelligence is
cannot perform well for scenarios with non-LOS (NLOS) expected to enable 6G, driving a fully autonomous future wire-
links, while for LOS scenarios, precise pointing between less system. With machine learning being such a craze, it has
the transceiver is essential. Second, the performance of been considered as a key approach in [31], [39], and [53] to
optical communications for underwater scenarios can be realize 6G from a user perspective. With intelligence intro-
severely degraded due to the absorption and scattering effects duced to 6G, machine learning enables wireless networks
of seawater, misalignment errors, channel turbulence, and learning capabilities to autonomously make optimal deci-
other impact factors, resulting in recurrent communication sions to adapt to the complex and varying environment, i.e.,
failures [10], [78]. To this point, how to ensure the reliability self-adaptive, self-upgrade, self-aware, and predictive ability.
of optical communications for underwater networks should be In summary, for both IoT applications and communication
further studied. networks, machine learning will play vital roles in building
6) Integration: Despite the challenges associated with indi- an intelligent IoT system to deliver smart services in the 6G
vidual tier, the integration of the four tiers faces unprece- massive IoT realm.
dented challenges due to its specific characteristics of high 2) Roles of Machine Learning: In wireless communication
heterogeneity, time variability, and self-organization. Since networks, machine learning can be used for prediction, big
the space–air–ground–underwater networks are significantly data analysis, and decision making [15]. Considering its attrac-
affected by the limited, heterogeneous, and unbalanced tive advantages, machine learning has been proposed for 5G,
resources in all four network segments, to obtain the best the roles of which are summarized in Table II. As we can
performance is difficult for traffic delivery. As known, existing see, machine learning is proposed to be applied to the whole
works are mainly focusing on the space–air–ground-integrated network, ranging from the physical layer to the application
TABLE II TABLE III

ROLES OF M ACHINE L EARNING IN 5G S UMMARY OF E XISTING W ORKS OF M ACHINE L EARNING IN I OT
exert control actions by interacting with the environment. The

network layer corresponds to the communication networks
for IoT, which encompasses wireless access networks and the
Internet, aiming at discovering and connecting IoT devices to
edge/cloud servers for data or control information transmis-
Fig. 5. Typical architecture of IoT systems.
sion. The application layer refers to edge computing systems,
which is in charge of data processing, storage, and decision
layer, which aims to provide automation for network manage- making for control determination.
ment. Despite these roles in 5G, machine learning will play There have been many works focusing on the applications
even more important roles in the future 6G. First, different in the three layers of this IoT architecture, a brief summary of
from the global evolution in 5G, in which the system has to be which is shown in Table III. As seen in Table III, deep rein-
rebuild when new waveforms, coding, and multiaccess proto- forcement learning (DRL) has been attracted wide attentions
cols are applied, machine learning can enable local evolution from all areas.
to upgrade the local networks, which brings more flexibil- Applications of Machine Learning in the Application Layer:
ity. Second, different from 5G, which developed an intelligent With limited energy and computation capability in IoT devices,
physical layer paradigm relying on the algorithm-hardware edge computing is essential for data analysis, prediction, deci-
coupled architecture, machine learning will enable intelligent sion making, (e.g., control instruction for IIoT and autonomous
radio based on the hardware-algorithm separation architec- driving), and computation tasks execution (e.g., rendering for
ture [53], which can adapt the environment and hardware VR videos) [83]. IoT devices could offload such kinds of tasks
in an intelligent way by replacing the conventional modula- to edge servers to save energy and reduce latency. To further
tion/coding modules with deep neural networks (DNNs). As improve the performance of IoT systems, the caching capabil-
6G becoming concrete and machine learning developing, it is ity of edge computing can be utilized by some data-reusable
believed that machine learning will have more positions in 6G. applications, e.g., high-definition live streaming and VR/AR
3) Applications of Machine Learning for Massive IoT in (i.e., rendering results [110]).
6G: The future IoT-enabled applications encompass identifi- There have been numerous works dedicated to task offload-
cation/authentication, sensing, communication, caching, coming and resource allocation in edge computing for IoT systems.
putation, and localization services [83], which establish a more Different from task offloading and resource allocation in cel-
complex ecosystem than ever before. A typical architecture for lular networks with edge computing, some different aspects
these IoT-enabled applications is shown in Fig. 5, which con- of IoT should be considered, e.g., ubiquitous coverage to
sists of three typical layers, i.e., a perception layer, a network suburban and rural areas and mobility in vehicular networks.
layer, and an application layer. The perception layer refers Space–air–ground networks are promising to provide ubiq-
to the physical systems where IoT devices acquire data and uitous services for remote IoT devices, which, however, leads
to higher complexity for task offloading. Considering this fact, transmission power, which was proven to outperform the cur-
Cheng et al. [94] proposed a reinforcement learning (RL)- rent state-of-the-art solutions only with a minimal amount of
based algorithm for task offloading, which was proven to information from the IoT system. However, the cooperation
achieve a near-optimal solution with low complexity. However, between different agents was ignored in this work.
the mobility of UAVs was ignored in this work, which intro- By using the licensed spectrum, narrowband IoT (NB-IoT)
duces flexibility in terms of implementation and offloading provides higher reliability and QoS. To enhance coverage, NB-
choices as well as management complexity. Considering a IoT chooses to increase the number of retransmissions for
hybrid edge computing platform with ground stations, ground data and control signals, which leads to throughput improve-
vehicles, and UAVs, Jiang et al. [91] proposed a deep learning- ment, but in turn, lowers the spectral efficiency. Hence, it
based algorithm to solve the resource scheduling problem, is important to extend coverage and reduce the number of
where a large-scale path-loss fuzzy c-means algorithm was retransmissions. Chafii et al. [95] proposed an RL-based
proposed to predict the optimal positions of ground vehicles dynamic spectrum access approach to increase the coverage.
and UAVs and a deep neural network was introduced for real- Based on the proposed approach, the IoT devices select the
time decision making. It was proven that the proposed methods best channel considering both the criterion of availability and
have better accuracy and could make real time and optimal the best coverage condition. However, this work focused on a
design feasible in the proposed architecture. single-player scenario, which is not practical.
In vehicle edge computing networks, vehicles act as Cooperative communication has been regarded as a promis-
mobile edge servers to provide offloading services for nearby ing solution to enhance spectrum utilization, throughput,
users, which extends the range of computation services, coverage, and so on. The cooperation among the relay nodes
as well as calls for intelligent networks. Constructing a realizes the transmission sharing, thereby improves the system
three-layer offloading framework for the Internet of vehicles, performance. In this case, it is important to select the best
Ning et al. [97] put forward a DRL algorithm to solve the relay node. Jadoon and Kim [88] applied Q-learning to
flow redirection and offloading decision problem. However, solve the relay selection problem for wireless cooperative
this work only focused on services for vehicular entertainment. networks. In this work, based on the state of previous system
Focusing on time-critical services, such as safety warnings and performance and the reward function about quality, the source
navigation suggestions, Tan et al. [98] constructed a vehicular node could determine the optimal relay with its learning capa-
network with MEC in the finite block-length coding regime bilities. However, the proposed Q-learning-based algorithm
and proposed a deep deterministic policy gradient approach to cannot deal with large-scale problems. Considering this fact,
solve the continuous resource allocation problem. However, Su et al. [92] proposed a deep Q network-based relay algo-
the aforementioned works did not apply to large-scale prob- rithm to rectify this problem, which selected the optimal relay
lems. In this case, distributed machine learning provides a from a plurality of relay nodes without any prior data or the
promising solution, e.g., multiagent RL [111] and federated need for a network model. However, this work did not con-
learning [104], which needs further exploration. sider the mobility of sensor nodes and assumed simple channel
Applications of Machine Learning in the Network Layer: conditions.
Reliable and efficient wireless communication networks are To tackle the problem of the crowded spectrum to support
essential for an IoT ecosystem, which range from short-range the increasing amount of IoT applications, cognitive networks
local area networks to long-range wide area networks. To are regarded as one of the enablers in future IoT systems [112].
transmit the tremendous amount of IoT data, efficient resource In this case, the IoT devices need to coordinate with the
control mechanisms are desired to efficiently utilize the scarce actions of the primary users with effective mechanisms in
radio, caching, and computing resources. First, the character- terms of spectrum access, transmission power control, trans-
istics of IoT devices need to be considered, such as massive in mission scheduling, and so on. Zhu et al. [93] proposed to
number, limited energy, memory, and computation resources. utilize deep Q learning to solve the transmission scheduling
Second, the diverse requirements of the IoT-enabled applica- problem in cognitive IoT networks. An appropriate strategy is
tions need to be taken into account, such as high data rate, low found by the proposed mechanism to transmit packets with dif-
latency, high reliability, privacy, and so on. In addition, different buffers to maximize the system throughput. However,
ferent applications may possess conflicting requirements. To the authors only considered one relay in this work and the
counter these challenges, a promising solution is to deploy cooperation and competition between multiple relays were
intelligence in the IoT ecosystem to enable IoT devices to ignored.
operate autonomously, e.g., to use machine learning. In the resource-constrained IoT systems, an intelligent and
In a mesh topology, better network connectivity is essential efficient resource management scheme is pivotal to guarantee
for performance improvement, e.g., throughput, which could performance, since the future networks are more complex and
be achieved by deploying more relay nodes at the cost of heterogeneous. In a two-layer heterogeneous IIoT network,
larger energy consumption and higher complexity. Hence, it a distributed Q-learning-aided network association scheme
is important to achieve a tradeoff between node coverage was proposed [89]. However, the proposed algorithm cannot
and energy consumption and complexity. Kwon et al. [99] support a massive number of IoT connections. Considering
proposed a distributed intelligent IoT connectivity solution the fact that the conventional RF networks may fail to sup-
based on DRL that enables mobile IoT devices to strategically port massive IoT connections as well as high data rate, a
make decisions on whether to activate the transmission and the collaborative RF-VLC architecture was proposed for indoor
communications [113], where RF networks offered long- to downsize the transmitted data, and deep learning is proven
distance transmissions with wide-range coverage, while VLC to have better performance for data recovery. However, it does
provided short-distance transmissions with high date rates. not apply to small data samples. To this end, Liang et al. [106]
UAV plays an important role to collect information for IoT developed a convolution-based transfer compressed sensing
devices in remote or mountainous areas. Considering the high model with the help of transfer learning, which could transfer
mobility of UAVs, which results in high dynamics for chan- a well-trained model to other related but insufficient data sets,
nel gain, it is impractical to obtain global information in the resulting in better performance but less time cost.
UAV-enabled IoT system. Hence, Cao et al. [100] proposed Machine learning plays a pivotal role in enabling
a DRL-based algorithm for channel and power allocation autonomous robots with or without intelligence, the research
with only partial network information. However, the utiliza- of which mainly focus on mobile behavior control, robotic
tion of the proposed algorithm depended on the discretization manipulation, and multirobot collaboration. Referring to gen-
of action space, which could not obtain the optimal solution. eral movement control, DRL has been widely applied to
Applications of Machine Learning in the Perception Layer: existing research, e.g., trajectory planning for UAVs [101],
The applications of machine learning in the perception layer [115] and path planning for navigation robots [102], [116].
include sensing, control, and so on. Sensing refers to the pro- Intelligent robots could perform some tasks in practice, where
cedure to collect the data generated by the IoT sensors. Control an appropriate control scheme tells the robot how to achieve
mainly refers to movement control for autonomous robots a target. For example, Mnih et al. [117] utilized DRL to teach
(e.g., tactile Internet, smart factory, and remote surgery), driv- the machine itself to play Atari 2600 games from the Arcade
ing aid for autonomous driving, intelligent management for Learning Environment, where it learned control policies from
smart grid. high-dimensional sensory input. For multirobot scenarios, col-
As the foundation of an IoT system, sensing is important laboration among these robots is essential in order to achieve
for decision-making policies, where the reliability and accu- a common goal. In this case, a distributed machine learn-
racy of the sensed data need to be ensured. Since the IoT ing algorithm may be more popular and appropriate, e.g.,
sensors are susceptible to malicious attacks, it is essential to asynchronous actor–critic algorithms (A3C) [107].
detect outliers to ensure data quality. Recently, machine learn- Machine learning is considered as the key enabler of
ing has been proven to be a powerful tool to detect outliers autonomous driving by the academia and industry, which is
in sensor data [114], which can be classified into paramet- used to control the vehicles without the help of humans,
ric methods, nonparametric methods, supervised learning, and e.g., following, overtaking, accelerating, and direction chang-
unsupervised learning. Hasan et al. [84] used logistic regres- ing (i.e., remaining in the current direction, turn right or left,
sion to detect anomaly or outlier in sensed data. Though or lane changing). To study the autonomous lane changing
it is simple and learns fast from the data, it suffers from problem for a single vehicle, Mirchevska et al. [103] proposed
the drawback of assuming a priori of the data distribution. a DRL-based approach to learn high-level decision makings,
To this end, Nesa et al. [85] proposed an outlier detection where the relative distance between the considered vehicle and
approach based on the gray systems theory. Though it does the neighboring six vehicles and the relative velocity of the
not make an assumption about the data distribution, which neighboring vehicles were input as the state of the learning
makes it more pliable for any data, a lot of training data is approach. Considering the overtaking problem, Li et al. [90]
needed to create a mapping function. Zidi et al. [86] used proposed a Q-learning-based approach to improve the over-
support vector machine (SVM) to detect faults in sensed data, taking policies to accord with the driving habits of humans.
where a cluster head made decisions on detection. However, However, these works only consider a single vehicle, where the
this vector-based approach may destroy the original struc- interactions among different vehicles are ignored. Motivated
ture and correlation within large-scale sensed data, resulting by this fact, cooperative adaptive cruise control [118], naviga-
in low detection accuracy. Thus, Deng et al. [87] proposed tion coordination through intersections [119] through interac-
an intelligent outlier detection method with one-class support tions, and cooperative lane changing on a freeway [120] have
Tucker machine and genetic algorithm, which extended SVM been studied using RL in multivehicle scenarios. However,
to tensor space with retaining the structural information of these works only directly applied distributed RL, which
data, resulting in improved accuracy and efficiency of anomaly ignored the coordination among vehicles in the process of
detection. learning. To solve this problem, Yu et al. [108] proposed a
Despite the outlier detection in sensor data, another problem strategic distributed learning solution for coordinating multiple
in sensing is the increased communication overhead and data vehicles.
redundancy, which needs an efficient method to cut down Smart grid has been the trend for future wireless networks
the size of the data set and to reduce the communication and attracted the attention of the world [121], which aims
cost. Traditional sensing algorithms are usually task specific to improve the electricity generation, consumption, transmis-
and cannot be directly applied to other field sensing tasks, sion, and distribution. Due to its uncertainty, complexities,
which is energy inefficient and time consuming. Considering and high volume of data, machine learning has revealed its
this, Wu et al. [105] developed an efficient sensing algorithm advantages in smart grid management, e.g., energy trading,
with the help of RL and metalearning, which significantly energy storage management, and demand response [83]. The
reduced the communication head with integrated communica- research and practice of deep learning and RL in smart grid
tion and computation. Compressed sensing is commonly used were reviewed in [121]. Energy trading allows the users to
switch their role between a consumer and provider depending e.g., edge computing servers, which in turn may not meet the
on its relationship between its generated and demanded energy, computation demands of machine learning algorithms.
where an autonomous control scheme is essential to ensure Centralized or Distributed: Considering the high complex-
the supply/demand balance, especially for microgrid [122]. A ity, centralized machine learning methods are not practical,
promising solution is to integrate games with machine learn- especially for large-scale IoT problems. To cater to the
ing to achieve Nash equilibrium autonomously in a centralized scalability problem, distributed machine learning plays an
or distributed way [123], [124]. To further consider prices, important role. However, it usually induces high overhead for
energy storage takes advantage of price fluctuations to buy information exchange and privacy problems. Also, it requires
or sell energy at a proper price in order to generate profit. each node to be with powerful computation capability. Hence,
Unlike previous works [125], with more practical assump- a sophisticated choice between centralized and distributed
tions, e.g., unknown electricity market prices, Cao et al. [96] machine learning-based schemes is crucial, which should be
adopted a model-free RL approach to learn an optimized based on various factors about the addressed problem.
control policy for storage charging/discharging. The demand Optimization of Machine Learning: Due to the characteris-
response aims to control the customers’ usage patterns in tics of IoT and 6G, the implementation of machine learning
response to electricity tariffs or other incentives and reduce in 6G is not straightforward and some modifications need to
the demand during system contingencies or at peak time peri- be made. First, each machine learning model has its draw-
ods [126]. In view of the facts of the dynamic and unknown backs, and it is essential to combine several machine learning
energy prices and the demand of the users’ appliances, an techniques to provide better performance. Second, the capabil-
online demand response scheme based on actor–critic learning ities of network entities may vary in hardware, connectivity,
was proposed in [109], where the long-term load scheduling and energy. This heterogeneity is believed to bring flaws for
problem was investigated with real-time prices. However, only machine learning models. Hence, machine learning should be
one householder was considered and the competition among optimized to fix this problem. In addition, model updates can
multiple multiple utility companies needs to be explored in the be disclosed, and techniques, such as deep net pruning, dif-
future. ferential privacy, and gradient compression are recommended
4) Open Issues of Machine Learning for Massive IoT in 6G: to preserve privacy for machine learning. Third, distributed
Though there has been a flurry of research activities dedicated machine learning is considered to be more suitable for 6G
to machine learning, these works only scratch the surface, and IoT, which, however, involves massive communication
and the potential of machine learning to tackle problems for between different edge devices. Hence, distributed machine
massive IoT in 6G remains largely unexplored [127]. Some learning needs to be modified to cope with the wireless
challenges are summarized as follows. communication conditions.
Incomplete Data Set in Perception Layer: Most machine
learning methods require data for training and testing, e.g.,
supervised, unsupervised learning, RL, and so on, which can B. Blockchain
be generated or collected with different approaches. For exam- This section mainly introduces motivations to apply
ple, the more the data samples, the more accurately the blockchain to IoT, applications of blockchain for massive
machine learning algorithms can achieve. The more complete IoT in 6G, and some open issues. For a more explicit
data samples are, the more accurately deep learning can learn understanding of blockchain, refer to [18], [128], and [129].
its model. However, it is not easy to acquire the perfect data 1) Motivations of Blockchain for IoT in 6G-A Perfect
samples needed. First, the unpredictable link delays and radio Match: With the evolution of IoT, several challenges are aris-
interference may result in slow response and certain parame- ing, including high complexity in network management, poor
ter miss or data errors. Second, some private data are usually interoperability, and privacy and security vulnerabilities, which
possessed by different entities, which is to sell or for secrecy. are explained as follows.
Third, it is a privacy problem. First, IoT systems are becoming more complex and het-
Long Time Cost: The long time cost may prevent the appli- erogeneous. In the perception layer, IoT devices are hetero-
cations of machine learning, which includes training time and geneous, which involves various sensors, machines, vehicles,
inference time. The former refers to the time to train the mod- robots, and so on. In addition, massive data are gener-
els to be accurate enough, while the latter refers to the time ated with different types, i.e., structured, semistructured, and
to output the results using the trained-well models. For some nonstructured. In the network layer, the multidimensional
applications, such as autonomous driving and tactile Internet, space–air–ground–underwater networks incur extra manage-
it needs a near real-time response, e.g., usually a timescale ment complexity with full spectra. In the application layer,
of milliseconds. However, it is far more behind enough to multilevel caching and computing resources are presented
train the machine learning algorithms in time, especially for from near to far from IoT devices. In summary, all these pose
large-dimensional IoT scenarios. challenges to network management. For resource management,
Computation and Storage Hungry: Most machine learning various resources are involved, i.e., communication, caching,
algorithms require powerful computation power and storage and computing, the management of which should take sensing,
capacity. To implement machine learning on the clouds in core control, and localization into considerations as well. For data
networks incurs long latency. To avoid transmission latency, management, massive data are difficult to manage with respect
machine learning can be implemented at the network edge, to elaboration, communication/transmission, and storage.
Second, interoperability for massive IoT in 6G becomes is regarded as a promising technology of 6G to provide a dis-
difficult in the presence of multiple protocol options and cross- tributed network management platform, which can be used for
platform architectures. To provide diverse services to IoT resource management, data sharing, data storage, and so on.
devices, multiple operational platforms, radio access service Hence, blockchain is a perfect match with IoT and 6G.
providers (SPs), edge cloud vendors, caching vendors, and so 2) Applications of Blockchain for Massive IoT in 6G:
on coexist. It is essential to collaborate with different SPs and With the inherent superior properties of blockchain, i.e., dis-
vendors and exchange data between different strategic centers, tributed nature, decentralized consensus, trust-less system,
industrial sectors, and IoT systems, which is impractical in cryptographic security, and nonrepudiation guarantees, it is
the current network architecture. First, a unified authentication considered as the key enabler of massive IoT, as well as
and authorization mechanism is required to integrate individual the next revolution of future mobile communication technol-
systems of different operators. Second, diverse service provi- ogy [24]. Generally, the use of blockchain can be categorized
sion and proper payment to different SPs and vendors need to into three types: 1) recording as a distributed ledger; 2) decen-
be audited and ensured. For example, the computation offload- tralized storing; and 3) realizing automation with smart con-
ing in the heterogeneous networks with MEC involves cellular tracts [131]. The exact applications of blockchain for massive
networks and MEC providers. However, there is no trusted entity IoT are described as follows.
in this system to audit the computation offloading process or Recording as a Distributed Ledger: Blockchain is attract-
ensure the surefire payments to SPs and edge cloud vendors. ing the focus of 6G on the perspective of dynamic resource
Third, it refers to privacy and security issues. Massive data management, e.g., spectrum, computation, caching, network
are generated by massive IoT devices, which may contain con- slicing, and so on, which could be used to record transactions
fidential and private information. It is important to use IoT of resource trading and audit the process of resource trad-
data with an appropriate approach so that there is no dis- ing. Currently, spectrums are auctioned off one at a time in a
closure of user private information without users’ consent. process that is slow, complex to manage, and expensive. One
However, it is challenging to protect users’ privacy in IoT option is to share the spectrum by using different bands at dif-
due to the decentralization, heterogeneity, and complexity of ferent levels of priority. Another option is to build a real-time
IoT systems. In addition, a trend to integrate cloud technolo- spectrum market that would issue permissions dynamically
gies, i.e., fog/edge/cloud computing with IoT arises, which using AI, which will bring new efficiencies and move the
aims to empower IoT applications with extra computing and system from one of scarcity to one that can manage relative
caching/storage capabilities. In this case, if clouds are attacked, abundance. To achieve this goal, blockchain is one of the most
a massive amount of data may be leaked, breaching users’ promising solutions being looked at by the researchers. There
privacy [130]. Due to the characteristics of IoT systems, i.e., have been some works dedicated to blockchain-based spectrum
decentralized, heterogeneous, and highly complex, it is also management. In [132]–[135], the spectrum sharing problem
difficult to ensure the security of IoT. The current methods was investigated, where blockchain acted as a peer-to-peer
for providing privacy, security, and data handling rely heavily decentralized ledger to record the spectrum trading efficiently.
on the control of third-party entities, few of which can be fully However, all these works ignored an important problem, i.e.,
trusted by end users. the performance of blockchain itself. For example, the trading
With improved interoperability, reliability, privacy, security, of the spectrum is obviously frequent, which is far larger than
and scalability, blockchain is perfect for massive IoT, which the throughput of the current blockchains. Also, blockchain
could exactly deal with these challenges. The combination of itself is resource hungry in terms of computation and trans-
blockchain and IoT can bring the following merits. mission. How to ensure the advantages of utilizing blockchain
1) Decentralized Network Management: It can be achieved for spectrum sharing is essential. To create a new spectrum
by recording all network events as transactions of the market is not easy, ranging from the framework design to the
blockchain. In this way, all events will be audited optimization of the system.
without a third party, which is more cost efficient. Decentralized Storage: In traditional cloud-based IoT
2) Interoperability Across IoT Devices, IoT Systems, systems, a centralized cloud server maintains and controls all
Industrial Sectors, and SPs: This can be achieved by the data, which poses some challenges: 1) high storage capac-
building a blockchain-composite layer on the top of ity required from the centralized server and 2) sensitive data
an overlay P2P network with uniform access to dif- easy to be leaked. To handle these issues, a decentralized struc-
ferent systems by providing a unified authentication, ture may be more proper. Fortunately, decentralized storage is
authorization mechanism, and billing system. one of the most popular applications of blockchain in IoT
3) Traceability and Reliability of IoT Data: A historic systems [136]. In [137], a decentralized platform was studied,
timestamp in each data block saved in a blockchain will where blockchain was used to protect personal private data.
consequently assure the data traceability. With inher- Here, blockchain acts as a decentralized personal data man-
ent asymmetric encryption algorithms, digital signatures, agement system to ensure users owning and controlling their
and hash functions in blockchains, the integrity of IoT data. In this work, the data were stored off blockchain, the
data is enforced. In this way, the security and privacy of security of which cannot be ensured. Li et al. [138] proposed
massive IoT will be ensured. to store the IoT data in distributed hash tables (DHTs) and the
In addition, as analyzed in Section III, a trend from cen- pointer to the DHT storage address in the blockchain. In this
tralization to distribution in 6G is arising, and the blockchain work, the authentication of the data requester is handled by
the distributed blockchain nodes rather than a central server, the services of IoT-enabled applications are highly hetero-
which maintains the advantages of decentralized storage, with- geneous in terms of different technical requirements. Some
out any intervention from a trusted server, traceability, and require ultralow latency, while some care only about privacy.
accountability. However, the authentication in this work was Hence, blockchain needs to be modified with more power-
done by only verifying the identity of the data requester, which ful adaptivity to support a wide variety of services, where a
needs further improvement. generic blockchain platform for IoT and 6G is desired. Second,
Realizing Automation With Smart Contracts: A smart con- blockchain is limited by its own properties, i.e., scalability
tract is a computer program, which is self-executing, self- and delay (or time to finality), which is vital to apply it to
verifying, and tamper resistant. It takes the transactions as massive IoT. Hence, a novel consensus protocol needs to be
input and then executes the code. When some conditions are designed to improve the throughput, time to finality, security,
met, some events are triggered, in which way automation is and so on, which should meet the corresponding demands of
achieved. In IoT systems, different nodes need to cooperate IoT applications in 6G.
with each other to provide services, which necessities the
need of managing the identity of nodes and realizing security
authentication between nodes [139]. Some works have adopted C. Combination of AI and Blockchain
smart contracts for access control in IoT systems [140], [141], As two of the most disruptive technologies, AI and
where the main function of smart contracts is to manage the blockchain have its own natural limitations, which restrict
data records. Different from these works, Zhang et al. [142] their applications to 6G. To fully unleash the potential of
proposed a smart contract-based access control framework, this two techniques and meet the future intelligent, distributed,
which consisted of multiple access control contracts to achieve and secure requirements, it is essential to combine these two
distributed access control for untrustworthy IoT systems. megatrends, which will make mutual enhancement.
Zhaofeng et al. [20] proposed a blockchain-based decen- Blockchain for AI: To date, the majority of AI methods rely
tralized trust management and data usage scheme, where a on a centralized model for training and inference by uploading
smart contract was in charge of all data operation, i.e., data data collected at edge devices to data centers. This centralized
gathering, invoking, processing, transfer, and usage. In IoT nature will lead to massive communication overhead and a
networks, a blockchain-based identity management framework possibility of data tampering. As a result, the decisions made
was proposed in [143], where a smart contract was utilized to by AI may be delayed and highly erroneous [147]. In this
govern the interactions between IoT devices by implement- case, distributed AI is in dire need, where privacy preser-
ing the processes and rules. Apart from identity management, vation and network traffic congestion reduction are enabled
smart contract can be also used for autonomous energy trad- by avoiding data uploading to a central server. However, dis-
ing [144], adaptive offloading [145], IoT device behavior tributed training and inference still require communications
regulation [146], and so on. between different agents and a centralized server for data and
3) Open Issues of Blockchain for Massive IoT in 6G: model sharing. First, model updates can be disclosed. Second,
Despite these potential visions of blockchain in massive IoT, it is hard to evaluate the contributions of each agent and lacks
multiple challenges need to be addressed before its widespread motivation for data and model sharing.
applications. To deal with these problems, blockchain is considered as a
Performances of Blockchain-Based Systems: In blockchain- promising solution. By replacing the centralized server in dis-
based systems, it performs as an overlaid layer to provide tributed AI with blockchain, in which models are stored in the
services to the underlaid layer, i.e., IoT systems. Hence, we form of transactions, model sharing is operated autonomously
need to guarantee the performances of both systems, which with smart contracts, and decisions are made on blockchain,
affect each other significantly. multiple benefits are brought, which are illustrated in Fig. 6.
Resource Hungry: Blockchain is inherently resource hun- First, it will provide rational motivations for data and model
gry in terms of both computation, caching, and transmission sharing. Second, security and privacy can be ensured with
for block generation, verification, ledger storage, and the inherent asymmetric encryption algorithms, digital signatures,
consensus among nodes. How to ensure the advantage of and hash functions in blockchain. Third, decisions made by
incorporating blockchain into IoT should be investigated. distributed AI that utilize blockchain can be trusted. As a
New Security Threats: Outsourcing a large scale of services result, distributed intelligence will be enabled.
at the edge and self-organization to achieve automation both AI for Blockchain: The benefits brought by AI to blockchain
triggers new security problems, which need to be studied are mainly twofold. First, AI can manage blockchains more
furthermore. efficiently. Nowadays, the operation of blockchain on com-
Resource Management: On one hand, the resources of the puters requires a large amount of computer processing power,
networks are limited, such as computation, caching, and wire- which is inefficient. Take the hashing algorithm as an exam-
less resources, especially at the network edge. On another ple. A brute force approach is used by the Bitcoin blockchain
hand, the blockchain system itself is resource hungry. Thus, to mine blocks, which tries every combination of characters
the resources need to be allocated between the blockchain and until a transaction can be verified. Moving away from this
wireless systems to make the best use of limited resources. brute force approach, a more intelligent and thoughtful manner
Optimization of Blockchain: To make full use of blockchain to manage tasks is desired, e.g., a machine learning-powered
for IoT and 6G, it should be optimized in various ways. First, mining algorithm. Despite mining, the other tasks can also
Fig. 7. Fully autonomous driving in 6G. CAV: connected autonomous

Fig. 6. Benefits of integrating blockchain to AI. vehicles. TBS: terrestrial base station. UAV: unmanned aerial vehicle. ML:
machine learning.
be enhanced with AI, e.g., AI-powered smart contracts [148].

With AI embedded, blockchain will become more efficient,
resulting in better performance. Second, AI could make full learning deployed on edge servers could help solve tasks that
use of the blockchain encrypted data. Information is held on CAVs could not deal with, which could also help enhance
a blockchain database in an encrypted state, which provides the performance of blockchain embedded on edge servers.
high security for sensitive data, but in turn, proposes extra Blockchain can help store important information for CAVs
challenges for data usage. To make things safer, we should and model parameters for machine learning (in an encrypted
reduce such incidents in which unencrypted data are exposed state, of course), and audit the offloading process of CAVs,
in any part of data processing. Nowadays, a new field of AI is which ensures the security and privacy of this system.
emerging, which aims at building algorithms capable of oper- As shown in Fig. 7, fully autonomous driving can be
ating on encrypted data [149]. With AI working with encrypted achieved in the following procedure.
data, it will further enhance data security and make full use 1) The vehicles use terahertz spectrum or other unlicensed
of blockchain database. spectrum for sensing data, e.g., the location and velocity
of the surrounding vehicles, the conditions of the road
surface, and the traffic light. Then, the sensed data are
VI. C ASE S TUDY transferred to the application layer through the network
To show how 6G can support IoT-enabled applications, we layer with the help of the space–air–ground–underwater
present a case study on fully autonomous driving in 6G, which networks, which provide high data rate, low latency, full
is shown in Fig. 7. This system enables fully autonomous coverage, and high reliability.
driving through offloading from connected autonomous vehi- 2) These data can be stored on blockchain, which preserves
cles (CAVs) to edge servers via the space–air–ground– privacy for CAVs.
underwater/sea networks, which includes three layers, CAV 3) Machine learning models are trained with these sensed
layer, network layer, and application layer. In the CAV layer, data on blockchain.
an autonomous vehicle is equipped with various sensors, 4) CAVs offload computation tasks to edge servers via the
including LIDAR, RADAR, cameras, GPS, ultrasonic sensors, space–air–ground–underwater/sea networks.
and so on, which convert some events and physical envi- 5) The application layer uses trained machine learning
ronment into electrical signals for measurement, i.e., sensed models to make intelligent decisions for driving assis-
data. The network layer refers to the space–air–ground– tant, e.g., lane changing and overtaking (i.e., strategy
underwater/sea network, which aims to provide connectivity inference).
to CAVs. Especially, the ground tier is the main solution. 6) Offloading processes are stored on blockchain, which
The air tier provides flexible coverage for CAVs, which can provides audition and motivation for offloading services.
also help sense the environment. The space tier can be used 7) The inference results, e.g., the control information, are
for localization and coverage for extreme areas. The appli- delivered to the vehicles through the network layer. In
cation layer enables machine learning and blockchain by this way, fully autonomous driving is achieved. In addi-
providing computing and storage capabilities. Here, machine tion, access control, data sharing between different cars
and edge servers, and resource sharing can be con- [9] D. Pompili and I. F. Akyildiz, “Overview of networking protocols for
trolled by blockchain, which could further enhance the underwater wireless communications,” IEEE Commun. Mag., vol. 47,
no. 1, pp. 97–102, Jan. 2009.
performance of this system. [10] Z. Zeng, S. Fu, H. Zhang, Y. Dong, and J. Cheng, “A survey of under-
water optical wireless communications,” IEEE Commun. Surveys Tuts.,
vol. 19, no. 1, pp. 204–238, 1st Quart., 2017.
VII. C ONCLUSION [11] P. Porambage, J. Okwuibe, M. Liyanage, M. Ylianttila, and T. Taleb,
In this article, we did a comprehensive survey on mas- “Survey on multi-access edge computing for Internet of Things real-
ization,” IEEE Commun. Surveys Tuts., vol. 20, no. 4, pp. 2961–2991,
sive IoT enabled by 6G. We provided an overview of the 4th Quart., 2018.
drivers and requirements, where the emerging applications [12] M. Ejaz, T. Kumar, M. Ylianttila, and E. Harjula, “Performance
and limitations of 5G were summarized. These applications and efficiency optimization of multi-layer IoT edge architecture,” in
Proc. 2nd 6G Wireless Summit (6G SUMMIT), Levi, Finland, 2020,
include holographic communications, five-sense communi- pp. 1–5.
cations, WBCI, smart healthcare, smart education/training, [13] E. C. Strinati et al., “6G: The next frontier: From holographic mes-
industry Internet, fully autonomous driving, and super smart saging to artificial intelligence using subterahertz and visible light
city/home. Then, a discussion on the visions of 6G in terms communication,” IEEE Veh. Technol. Mag., vol. 14, no. 3, pp. 42–50,
Sep. 2019.
of core technical requirements, use cases, and trends is car- [14] M. Z. Chowdhury, M. Shahjalal, S. Ahmed, and Y. M. Jang, “6G wire-
ried out. To meet the demand of the IoT-enabled applications, less communication systems: Applications, requirements, technologies,
especially the full coverage requirement, 6G provides a four- challenges, and research directions,” Sep. 2019. [Online]. Available:
http://arxiv.org/abs/1909.11315.
tier network architecture enhanced by edge computing. As [15] Y. Zhao and G. Yu, and H. Xu, “6G mobile communication
an omnipotent network, 6G also provides some breakthrough network: Vision, challenges and key technologies,” SCIENTIA SINICA
technologies, including machine learning and blockchain, to Information, vol. 49, no. 8, pp. 963–987, 2019.
enable intelligence and distribution in future IoT systems, [16] S. K. Sharma and X. Wang, “Toward massive machine type com-
munications in ultra-dense cellular IoT networks: Current issues and
which play vital roles in the whole IoT architecture, from the machine learning-assisted solutions,” IEEE Commun. Surveys Tuts.,
perception layer and network layer to the application layer. vol. 22, no. 1, pp. 426–471, 1st Quart., 2020.
Finally, we present a use case of fully autonomous driving to [17] F. R. Yu, J. Liu, Y. He, P. Si, and Y. Zhang, “Virtualization
for distributed ledger technology (vDLT),” IEEE Access, vol. 6,
show how 6G supports massive IoT. pp. 25019–25028, 2018.
In summary, research on 6G and massive IoT is quite broad [18] M. S. Ali, M. Vecchio, M. Pincheira, K. Dolui, F. Antonelli, and
and a number of challenges lay ahead. Nevertheless, it is in M. H. Rehmani, “Applications of blockchains in the Internet of Things:
favor of the wireless community to swiftly address the chal- A comprehensive survey,” IEEE Commun. Surveys Tuts., vol. 21, no. 2,
pp. 1676–1717, 2nd Quart., 2019.
lenges and go forward. This article attempts to briefly explore [19] M. Wu, K. Wang, X. Cai, S. Guo, M. Guo, and C. Rong, “A com-
the technologies related to 6G and IoT, and to discuss future prehensive survey of blockchain: From theory to IoT applications
research that may benefit the pursuit of this vision. We hope and beyond,” IEEE Internet Things J., vol. 6, no. 5, pp. 8114–8154,
Oct. 2019.
that our discussion and exploration here may open a new
[20] M. Zhaofeng, W. Lingyun, W. Xiaochang, W. Zhen, and Z. Weizhe,
avenue for the development of 6G and IoT, and shed lights “Blockchain-enabled decentralized trust management and secure usage
on effective and efficient designs therein. control of IoT big data,” IEEE Internet Things J., vol. 7, no. 5,
pp. 4000–4015, May 2020.
[21] J. Xie et al., “A survey of blockchain technology applied to smart
R EFERENCES cities: Research issues and challenges,” IEEE Commun. Surveys Tuts.,
vol. 21, no. 3, pp. 2794–2830, 3rd Quart., 2019.
[1] Q. Qi, X. Chen, C. Zhong, and Z. Zhang, “Integration of energy, com- [22] L. Chettri and R. Bera, “A comprehensive survey on Internet of Things
putation and communication in 6G cellular Internet of Things,” IEEE (IoT) toward 5G wireless systems,” IEEE Internet Things J., vol. 7,
Comm. Lett., vol. 24, no. 6, pp. 1333–1337, Jun. 2020. no. 1, pp. 16–32, Jan. 2020.
[2] A. H. Sodhro, S. Pirbhulal, L. Zongwei, K. Muhammad, and N. Zahid, [23] G. A. Akpakwu, B. J. Silva, G. P. Hancke, and A. M. Abu-Mahfouz,
“Towards 6G architecture for energy efficient communication in IoT- “A survey on 5G networks for the Internet of Things: Communication
enabled smart automation systems,” IEEE Internet Things J., early technologies and challenges,” IEEE Access, vol. 6, pp. 3619–3647,
access, Sep. 18, 2020, doi: 10.1109/JIOT.2020.3024715. 2017.
[3] X. Li, T. Chen, Q. Cheng, S. Ma, and J. Ma, “Smart applications in [24] T. Huang, W. Yang, J. Wu, J. Ma, X. Zhang, and D. Zhang, “A survey
edge computing: Overview on authentication and data security,” IEEE on green 6G network: Architecture and technologies,” IEEE Access,
Internet Things J., vol. 8, no. 6, pp. 4063–4080, Mar. 2021. vol. 7, pp. 175758–175768, 2019.
[4] N. H. Crisp et al., “The benefits of very low earth orbit for earth obser-
[25] X. You et al., “Towards 6G wireless communication networks: Vision,
vation missions,” Progr. Aerosp. Sci., vol. 117, pp. 1–23, Aug. 2020.
enabling technologies, and new paradigm shifts,” Sci. China Inf. Sci.,
[Online]. Available: http://dx.doi.org/10.1016/j.paerosci.2020.100619
vol. 64, pp. 1–77, Nov. 2020.
[5] S. M. Dakka, “VLEO satellites—A new earth observation space
systems commercial and business model,” in Proc. 36th Int. Commun. [26] F. Hussain, S. A. Hassan, R. Hussain, and E. Hossain, “Machine learn-
Satellite Syst. Conf. (ICSSC), Niagara Falls, ON, Canada, 2018, ing for resource management in cellular and IoT networks: Potentials,
pp. 1–11. current solutions, and open challenges,” IEEE Commun. Surveys Tuts.,
[6] M. Samir, S. Sharafeddine, C. M. Assi, T. M. Nguyen, and A. Ghrayeb, vol. 22, no. 2, pp. 1251–1275, 2nd Quart., 2020.
“UAV trajectory planning for data collection from time-constrained IoT [27] W. Viriyasitavat, L. D. Xu, Z. Bi, and D. Hoonsopon, “Blockchain
devices,” IEEE Trans. Wireless Commun., vol. 19, no. 1, pp. 34–46, technology for applications in Internet of Things—Mapping from
Jan. 2020. system design perspective,” IEEE Internet Things J., vol. 6, no. 5,
[7] Z. Na, Y. Liu, J. Shi, C. Liu, and Z. Gao, “UAV-supported clustered pp. 8155–8168, Oct. 2019.
NOMA for 6G-enabled Internet of Things: Trajectory planning and [28] J. R. Wolpaw, N. Birbaumer, D. J. McFarland, G. Pfurtscheller, and
resource allocation,” IEEE Internet Things J., early access, Jun. 23, T. M. Vaughan, “Brain–computer interfaces for communication and
2020, doi: 10.1109/JIOT.2020.3004432. control,” Clin. Neurophysiol., vol. 113, no. 6, pp. 767–791, 2002.
[8] L. Yan, C. Han, and J. Yuan, “Hybrid precoding for 6G terahertz com- [29] S. Guruprakash, R. Balaganesh, M. Divakar, K. Aravinth, and
munications: Performance evaluation and open problems,” in Proc. 2nd S. Kavitha, “Brain controlled home automation,” Proc. Int. J. Adv. Res.
6G Wireless Summit (6G SUMMIT), Levi, Finland, 2020, pp. 1–5. Biol. Eng. Sci. Technol. (IJARBEST), vol. 2, no. 10, pp. 430–436, 2016.
[30] P. Zioga, F. Pollick, M. Ma, P. Chapman, and K. Stefanov, [52] X. Cao, P. Yang, M. Alzenad, X. Xi, D. Wu, and H. Yanikomeroglu,
“‘Enheduanna—A manifesto of falling’ live brain-computer cinema “Airborne communication networks: A survey,” IEEE J. Sel. Areas
performance: Performer and audience participation, cognition and emo- Commun., vol. 36, no. 9, pp. 1907–1926, Sep. 2018.
tional engagement using multi-brain BCI interaction,” Front. Neurosci., [53] K. B. Letaief, W. Chen, Y. Shi, J. Zhang, and Y.-J. A. Zhang, “The
vol. 12, p. 191, Apr. 2018. roadmap to 6G: AI empowered wireless networks,” IEEE Commun.
[31] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems: Mag., vol. 57, no. 8, pp. 84–90, Aug. 2019.
Applications, trends, technologies, and open research problems,” IEEE [54] F. Tariq, M. Khandaker, K.-K. Wong, M. Imran, M. Bennis, and
Netw., vol. 34, no. 3, pp. 134–142, May/Jun. 2020. M. Debbah, “A speculative study on 6G,” Feb. 2019. [Online].
[32] S. D. Pendleton et al., “Perception, planning, control, and coordination Available: http://arxiv.org/abs/1902.06700.
for autonomous vehicles,” Machines, vol. 5, no. 1, p. 6, 2017. [55] X. Wang, R. Li, C. Wang, X. Li, T. Taleb, and V. C. M. Leung,
[33] R. Yang, F. R. Yu, P. Si, Z. Yang, and Y. Zhang, “Integrated blockchain “Attention-weighted federated deep reinforcement learning for device-
and edge computing systems: A survey, some research issues and chal- to-device assisted heterogeneous collaborative edge caching,” IEEE J.
lenges,” IEEE Commun. Surveys Tuts., vol. 21, no. 2, pp. 1508–1532, Sel. Areas Commun., vol. 39, no. 1, pp. 154–169, Jan. 2021.
2nd Quart., 2019. [56] P. K. Chowdhury, M. Atiquzzaman, and W. Ivancic, “Handover
[34] 2018 Broadband Deployment Report, Federal Commun. Commission, schemes in satellite networks: State-of-the-art and future research direc-
Washington, DC, USA, Feb. 2018. tions,” IEEE Commun. Surveys Tuts., vol. 8, no. 4, pp. 2–14, 4th Quart.,
[35] I. Ahmad, S. Shahabuddin, T. Kumar, J. Okwuibe, A. Gurtov, and 2006.
M. Ylianttila, “Security for 5G and beyond,” IEEE Commun. Surveys
[57] M. Lin, Z. Lin, W.-P. Zhu, and J.-B. Wang, “Joint beamforming for
Tuts., vol. 21, no. 4, pp. 3682–3722, 4th Quart., 2019.
secure communication in cognitive satellite terrestrial networks,” IEEE
[36] D. Younger. (2018). 5G Study Raises Security Concerns for
J. Sel. Areas Commun., vol. 36, no. 5, pp. 1017–1029, May 2018.
Next Generation of Mobile Communication. [Online]. Available:
https://www.dundee.ac.uk/news/2018/5g-study-raises-security- [58] SpaceX V-Band Non-Geostationary Satellite System, SpaceX,
concerns-for-next-generation-of-mobile-communication.php Hawthorne, CA, USA, Nov. 2018.
[37] M. Aazam, K. A. Harras, and S. Zeadally, “Fog computing for 5G [59] A. Pasztor, OneWeb Satellite Startup to Set Up Manufacturing in
tactile Industrial Internet of Things: QoE-aware resource allocation Florida, Wall Street J., New York, NY, USA, Feb. 2016.
model,” IEEE Trans. Ind. Informat., vol. 15, no. 5, pp. 3085–3092, [60] FCC Boosts Satellite Broadband Connectivity and Competition in the
May 2019. United States, FCC News, Washington, DC, USA, Nov. 2018.
[38] Z. Ma, Z. Zhang, Z. Ding, P. Fan, and H. Li, “Key techniques for [61] D. Zhai, H. Li, X. Tang, R. Zhang, Z. Ding, and F. R. Yu, “Height
5G wireless communications: Network architecture, physical layer, and optimization and resource allocation for noma enhanced UAV-aided
MAC layer perspectives,” Sci. China Inf. Sci., vol. 58, no. 4, pp. 1–20, relay networks,” IEEE Trans. Commun., vol. 69, no. 2, pp. 962–975,
Apr. 2015. Feb. 2021.
[39] Z. Zhang et al., “6G wireless networks: Vision, requirements, architec- [62] A. A. Khuwaja, Y. Chen, N. Zhao, M.-S. Alouini, and P. Dobbins, “A
ture, and key technologies,” IEEE Veh. Technol. Mag., vol. 14, no. 3, survey of channel modeling for UAV communications,” IEEE Commun.
pp. 28–41, Sep. 2019. Surveys Tuts., vol. 20, no. 4, pp. 2804–2821, 4th Quart., 2018.
[40] J. A. del Peral-Rosado, R. Raulefs, J. A. López-Salcedo, and [63] V. Petrov, D. Moltchanov, J. M. Jornet, and Y. Koucheryavy,
G. Seco-Granados, “Survey of cellular mobile radio localization meth- “Exploiting multipath terahertz communications for physical layer
ods: From 1G to 5G,” IEEE Commun. Surveys Tuts., vol. 20, no. 2, security in beyond 5G networks,” in Proc. IEEE Conf. Comput.
pp. 1124–1148, 2nd Quart., 2018. Commun. Workshops (INFOCOM WKSHPS), Paris, France, Apr. 2019,
[41] A. Jamalipour and T. Tung, “The role of satellites in global IT: pp. 865–872.
Trends and implications,” IEEE Pers. Commun., vol. 8, no. 3, pp. 5–11, [64] O. Ergul, E. Dinc, and O. B. Akan, “Communicate to illuminate: State-
Jun. 2001. of-the-art and research challenges for visible light communications,”
[42] B. Zong, C. Fan, X. Wang, X. Duan, B. Wang, and J. Wang, “6G Phys. Commun., vol. 17, pp. 72–85, Dec. 2015.
technologies: Key drivers, core requirements, system architectures, [65] A.-M. Cǎilean and M. Dimian, “Current challenges for visible light
and enabling technologies,” IEEE Veh. Technol. Mag., vol. 14, no. 3, communications usage in vehicle applications: A survey,” IEEE
pp. 18–27, Sep. 2019. Commun. Surveys Tuts., vol. 19, no. 4, pp. 2681–2703, 4th Quart.,
[43] FG-NET-2030, “Network 2030—A blueprint of technol- 2017.
ogy, applications and market drivers towards the year [66] B. Schrenk, M. Hofer, F. Laudenbach, H. Hübel, and T. Zemen,
2030 and beyond,” ITU, Geneva, Switzerland, White Paper, “Visible-light multi-Gb/s transmission based on resonant cavity LED
May 2019. [Online]. Available: https://www.itu.int/en/ITU- with optical energy feed,” IEEE J. Sel. Areas Commun., vol. 36, no. 1,
T/focusgroups/net2030/Documents/White_Paper.pdf pp. 175–184, Jan. 2018.
[44] I. F. Akyildiz, J. M. Jornet, and C. Han, “Terahertz band: Next fron- [67] P. H. Pathak, X. Feng, P. Hu, and P. Mohapatra, “Visible light
tier for wireless communications,” Phys. Commun., vol. 12, pp. 16–32, communication, networking, and sensing: A survey, potential and chal-
Sep. 2014. lenges,” IEEE Commun. Surveys Tuts., vol. 17, no. 4, pp. 2047–2077,
[45] Z. Chen et al., “A survey on terahertz communications,” China 4th Quart., 2015.
Commun., vol. 16, no. 2, pp. 1–35, Feb. 2019.
[68] S. U. Rehman, S. Ullah, P. H. J. Chong, S. Yongchareon, and
[46] H. Elayan, O. Amin, R. M. Shubair, and M.-S. Alouini, “Terahertz
D. Komosny, “Visible light communication: A system perspective—
communication: The opportunities of wireless technology beyond
Overview and challenges,” Sensors, vol. 19, no. 5, p. 1153, Mar. 2019.
5G,” in Proc. Int. Conf. Adv. Commun. Technol. Netw. (CommNet),
Marrakech, Morocco, Apr. 2018, pp. 1–5. [69] C. Wang, H.-Y. Yu, and Y.-J. Zhu, “A long distance underwater visible
[47] Q. Bi, “Ten trends in the cellular industry and an outlook on 6G,” IEEE light communication system with single photon avalanche diode,” IEEE
Commun. Mag., vol. 57, no. 12, pp. 31–36, Dec. 2019. Photon. J., vol. 8, no. 5, pp. 1–11, Oct. 2016.
[48] V. Petrov, A. Pyattaev, D. Moltchanov, and Y. Koucheryavy, “Terahertz [70] L. E. M. Matheus, A. B. Vieira, L. F. M. Vieira, M. A. M. Vieira,
band communications: Applications, research challenges, and standard- and O. Gnawali, “Visible light communication: Concepts, applica-
ization activities,” in Proc. 8th Int. Congr. Ultra Modern Telecommun. tions and challenges,” IEEE Commun. Surveys Tuts., vol. 21, no. 4,
Control Syst. Workshops (ICUMT), Lisbon, Portugal, Oct. 2016, pp. 3204–3237, 4th Quart., 2019.
pp. 183–190. [71] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and
[49] J. Eggerton. FCC’s Rosenworcel Talks Up 6G. Accessed: Sep. 14, M. Ayyash, “Internet of Things: A survey on enabling technologies,
2018. [Online]. Available: https://www.multichannel.com/news/fccs- protocols, and applications,” IEEE Commun. Surveys Tuts., vol. 17,
rosenworcel-talks-up-6g no. 4, pp. 2347–2376, 4th Quart., 2015.
[50] M. L. Roberts, M. A. Temple, R. F. Mills, and R. A. Raines, “Evolution [72] T. Qiu, J. Chi, X. Zhou, Z. Ning, M. Atiquzzaman, and D. O. Wu,
of the air interface of cellular communications systems toward 4G “Edge computing in Industrial Internet of Things: Architecture,
realization,” IEEE Commun. Surveys Tuts., vol. 8, no. 1, pp. 2–23, advances and challenges,” IEEE Commun. Surveys Tuts., vol. 22, no. 4,
1st Quart., 2006. pp. 2462–2488, 4th Quart., 2020.
[51] R. Xie, Q. Tang, Q. Wang, X. Liu, F. R. Yu, and T. Huang, [73] T. Taleb, K. Samdanis, B. Mada, H. Flinck, S. Dutta, and D. Sabella,
“Satellite-terrestrial integrated edge computing networks: Architecture, “On multi-access edge computing: A survey of the emerging 5G
challenges, and open issues,” IEEE Netw., vol. 34, no. 3, pp. 224–231, network edge cloud architecture and orchestration,” IEEE Commun.
May/Jun. 2020. Surveys Tuts., vol. 19, no. 3, pp. 1657–1681, 3rd Quart., 2017.
[74] X. Wang, Y. Han, V. C. M. Leung, D. Niyato, X. Yan, and X. Chen, [95] M. Chafii, F. Bader, and J. Palicot, “Enhancing coverage in narrow
“Convergence of edge computing and deep learning: A comprehensive band-IoT using machine learning,” in Proc. IEEE Wireless Commun.
survey,” IEEE Commun. Surveys Tuts., vol. 22, no. 2, pp. 869–904, 2nd Netw. Conf. (WCNC), Barcelona, Spain, 2018, pp. 1–6.
Quart., 2020. [96] J. Cao, D. Harrold, Z. Fan, T. Morstyn, D. Healey, and K. Li, “Deep
[75] A. Alwarafy, K. A. Al-Thelaya, M. Abdallah, J. Schneider, and reinforcement learning-based energy storage arbitrage with accurate
M. Hamdi, “A survey on security and privacy issues in edge-computing- lithium-ion battery degradation model,” IEEE Trans. Smart Grid,
assisted Internet of Things,” IEEE Internet Things J., vol. 8, no. 6, vol. 11, no. 5, pp. 4513–4521, Sep. 2020.
pp. 4004–4022, Mar. 2021. [97] Z. Ning et al., “Deep reinforcement learning for intelligent Internet of
[76] L. U. Khan, I. Yaqoob, N. H. Tran, S. M. A. Kazmi, T. N. Dang, and Vehicles: An energy-efficient computational offloading scheme,” IEEE
C. S. Hong, “Edge-computing-enabled smart cities: A comprehensive Trans. Cogn. Commun. Netw., vol. 5, no. 4, pp. 1060–1072, Dec. 2019.
survey,” IEEE Internet Things J., vol. 7, no. 10, pp. 10200–10232, [98] G. Tan, H. Zhang, and S. Zhou, “Resource allocation in MEC-enabled
Oct. 2020. vehicular networks: A deep reinforcement learning approach,” in Proc.
[77] P. Mach and Z. Becvar, “Mobile edge computing: A survey on archi- IEEE INFOCOM Conf. Comput. Commun. Workshops (INFOCOM
tecture and computation offloading,” IEEE Commun. Surveys Tuts., WKSHPS), Toronto, ON, Canada, 2020, pp. 406–411.
vol. 19, no. 3, pp. 1628–1656, 3rd Quart., 2017. [99] M. Kwon, J. Lee, and H. Park, “Intelligent IoT connectivity: Deep
[78] M. Z. Chowdhury, M. K. Hasan, M. Shahjalal, M. T. Hossan, and reinforcement learning approach,” IEEE Sensors J., vol. 20, no. 5,
Y. M. Jang, “Optical wireless hybrid networks: Trends, opportunities, pp. 2782–2791, Mar. 2020.
challenges, and research directions,” IEEE Commun. Surveys Tuts., [100] Y. Cao, L. Zhang, and Y.-C. Liang, “Deep reinforcement learning for
vol. 22, no. 2, pp. 930–966, 2nd Quart., 2020. channel and power allocation in UAV-enabled IoT systems,” in Proc.
[79] J. Liu, Y. Shi, Z. M. Fadlullah, and N. Kato, “Space-air-ground inte- IEEE Global Commun. Conf. (GLOBECOM), Waikoloa, HI, USA,
grated network: A survey,” IEEE Commun. Surveys Tuts., vol. 20, no. 4, 2019, pp. 1–6.
pp. 2714–2741, 4th Quart., 2018. [101] J. Hu, H. Zhang, and L. Song, “Reinforcement learning for decentral-
[80] Z. Chang, L. Lei, Z. Zhou, S. Mao, and T. Ristaniemi, “Learn to cache: ized trajectory design in cellular UAV networks with sense-and-send
Machine learning for network edge caching in the big data era,” IEEE protocol,” IEEE Internet Things J., vol. 6, no. 4, pp. 6177–6189,
Wireless Commun., vol. 25, no. 3, pp. 28–35, Jun. 2018. Aug. 2019.
[81] X. Cao, L. Liu, Y. Cheng, and X. S. Shen, “Towards energy-efficient [102] J. Xin, H. Zhao, D. Liu, and M. Li, “Application of deep reinforcement
wireless networking in the big data era: A survey,” IEEE Commun. learning in mobile robot path planning,” in Proc. Chin. Autom. Congr.
Surveys Tuts., vol. 20, no. 1, pp. 303–332, 1st Quart., 2018. (CAC), Jinan, China, 2017, pp. 7112–7116.
[82] X. Wang, X. Ren, C. Qiu, Y. Cao, T. Taleb, and V. C. M. Leung, “Net- [103] B. Mirchevska, C. Pek, M. Werling, M. Althoff, and J. Boedecker,
in-AI: A computing-power networking framework with adaptability, “High-level decision making for safe and reasonable autonomous lane
flexibility, and profitability for ubiquitous AI,” IEEE Netw., vol. 35, changing using reinforcement learning,” in Proc. 21st Int. Conf. Intell.
no. 1, pp. 280–288, Jan./Feb. 2021. Transp. Syst. (ITSC), Maui, HI, USA, 2018, pp. 2156–2162.
[83] L. Lei, Y. Tan, K. Zheng, S. Liu, K. Zhang, and X. Shen, “Deep [104] S. Yu, X. Chen, Z. Zhou, X. Gong, and D. Wu, “When deep rein-
reinforcement learning for autonomous Internet of Things: Model, forcement learning meets federated learning: Intelligent multitimescale
applications and challenges,” IEEE Commun. Surveys Tuts., vol. 22, resource management for multiaccess edge computing in 5G ultra-
no. 3, pp. 1722–1760, 3rd Quart., 2020. dense network,” IEEE Internet Things J., vol. 8, no. 4, pp. 2238–2251,
[84] M. Hasan, M. M. Islam, M. I. I. Zarif, and M. M. A. Hashem, Feb. 2021.
“Attack and anomaly detection in IoT sensors in IoT sites using [105] H. Wu, Z. Zhang, C. Jiao, C. Li, and T. Q. S. Quek, “Learn to sense:
machine learning approaches,” Internet Things, vol. 7, Sep. 2019, A meta-learning-based sensing and fusion framework for wireless sen-
Art. no. 100059. sor networks,” IEEE Internet Things J., vol. 6, no. 5, pp. 8215–8227,
[85] N. Nesa, T. Ghosh, and I. Banerjee, “Non-parametric sequence-based Oct. 2019.
learning approach for outlier detection in IoT,” Future Gener. Comput. [106] J. Liang, L. Li, and C. Zhao, “A transfer learning approach for com-
Syst., vol. 82, pp. 412–421, May 2018. pressed sensing in 6G-IoT,” IEEE Internet Things J., early access,
[86] S. Zidi, T. Moulahi, and B. Alaya, “Fault detection in wireless sensor Jan. 20, 2021, doi: 10.1109/JIOT.2021.3053088.
networks through SVM classifier,” IEEE Sensors J., vol. 18, no. 1, [107] G. Sartoretti, Y. Wu, W. Paivine, T. K. Kumar, S. Koenig, and
pp. 340–347, Jan. 2018. H. Choset, “Distributed reinforcement learning for multi-robot decen-
[87] X. Deng, P. Jiang, X. Peng, and C. Mi, “An intelligent outlier detection tralized collective construction,” in Proc. Int. Symp. Distrib. Auton.
method with one class support tucker machine and genetic algo- Robot. Syst. (DARS), 2018, pp. 35–49.
rithm toward big sensor data in Internet of Things,” IEEE Trans. Ind. [108] C. Yu et al., “Distributed multiagent coordinated learning for
Electron., vol. 66, no. 6, pp. 4672–4683, Jun. 2019. autonomous driving in highways based on dynamic coordination
[88] M. A. Jadoon and S. Kim, “Relay selection algorithm for wireless coop- graphs,” IEEE Trans. Intell. Transp. Syst., vol. 21, no. 2, pp. 735–748,
erative networks: A learning-based approach,” IET Commun., vol. 11, Feb. 2020.
no. 7, pp. 1061–1066, May 2017. [109] S. Bahrami, V. W. S. Wong, and J. Huang, “An online learning algo-
[89] J. Wang, C. Jiang, K. Zhang, X. Hou, Y. Ren, and Y. Qian, rithm for demand response in smart grid,” IEEE Trans. Smart Grid,
“Distributed Q-learning aided heterogeneous network association for vol. 9, no. 5, pp. 4712–4725, Sep. 2018.
energy-efficient IIoT,” IEEE Trans. Ind. Informat., vol. 16, no. 4, [110] F. Guo, F. R. Yu, H. Zhang, H. Ji, V. C. M. Leung, and X. Li, “An
pp. 2756–2764, Apr. 2020. adaptive wireless virtual reality framework in future wireless networks:
[90] X. Li, X. Xu, and L. Zuo, “Reinforcement learning based overtak- A distributed learning approach,” IEEE Trans. Veh. Technol., vol. 69,
ing decision-making for highway autonomous driving,” in Proc. 6th no. 8, pp. 8514–8528, Aug. 2020.
Int. Conf. Intell. Control Inf. Process. (ICICIP), Wuhan, China, 2015, [111] H. Peng and X. Shen, “Multi-agent reinforcement learning based
pp. 336–342. resource management in MEC- and UAV-assisted vehicular networks,”
[91] F. Jiang, K. Wang, L. Dong, C. Pan, W. Xu, and K. Yang, “Deep- IEEE J. Sel. Areas Commun., vol. 39, no. 1, pp. 131–141, Jan. 2021.
learning-based joint resource scheduling algorithms for hybrid MEC [112] A. Somov, C. Dupont, and R. Giaffreda, “Supporting smart-city mobil-
networks,” IEEE Internet Things J., vol. 7, no. 7, pp. 6252–6265, ity with cognitive Internet of Things,” in Proc. Future Netw. Mobile
Jul. 2020. Summit, Lisbon, Portugal, 2013, pp. 1–10.
[92] Y. Su, X. Lu, Y. Zhao, L. Huang, and X. Du, “Cooperative commu- [113] H. Zhang, N. Liu, K. Long, J. Cheng, V. C. M. Leung, and L. Hanzo,
nications with relay selection based on deep reinforcement learning “Energy efficient subchannel and power allocation for software-defined
in wireless sensor networks,” IEEE Sensors J., vol. 19, no. 20, heterogeneous VLC and RF networks,” IEEE J. Sel. Areas Commun.,
pp. 9561–9569, Oct. 2019. vol. 36, no. 3, pp. 658–670, Mar. 2018.
[93] J. Zhu, Y. Song, D. Jiang, and H. Song, “A new deep-Q-learning- [114] N. Ghosh, K. Maity, R. Paul, and S. Maity, “Outlier detection in sensor
based transmission scheduling mechanism for the cognitive Internet data using machine learning techniques for IoT framework and wireless
of Things,” IEEE Internet Things J., vol. 5, no. 4, pp. 2375–2385, sensor networks: A brief study,” in Proc. Int. Conf. Appl. Mach. Learn.
Aug. 2018. (ICAML), 2019, pp. 187–190.
[94] N. Cheng et al., “Space/aerial-assisted computing offloading for IoT [115] S. Yin, S. Zhao, Y. Zhao, and F. R. Yu, “Intelligent trajectory design in
applications: A learning-based approach,” IEEE J. Sel. Areas Commun., UAV-aided communications with reinforcement learning,” IEEE Trans.
vol. 37, no. 5, pp. 1117–1129, May 2019. Veh. Technol., vol. 68, no. 8, pp. 8227–8231, Aug. 2019.
[116] T. Yan, Y. Zhang, and B. Wang, “Path planning for mobile robot’s [134] S. Han and X. Zhu, “Blockchain based spectrum sharing algorithm,” in
continuous action space based on deep reinforcement learning,” in Proc. IEEE 19th Int. Conf. Commun. Technol. (ICCT), Xi’an, China,
Proc. Int. Conf. Big Data Artif. Intell. (BDAI), Beijing, China, 2018, 2019, pp. 936–940.
pp. 42–46. [135] Z. Zhou, X. Chen, Y. Zhang, and S. Mumtaz, “Blockchain-empowered
[117] V. Mnih et al., “Playing atari with deep reinforcement learning,” 2013. secure spectrum sharing for 5G heterogeneous networks,” IEEE Netw.,
[Online]. Available: arXiv:1312.5602. vol. 34, no. 1, pp. 24–31, Jan./Feb. 2020.
[118] C. Desjardins and B. Chaib-Draa, “Cooperative adaptive cruise control: [136] Y. Zhu, G. Zheng, and K.-K. Wong, “Blockchain-empowered decen-
A reinforcement learning approach,” IEEE Trans. Intell. Transp. Syst., tralized storage in air-to-ground industrial networks,” IEEE Trans. Ind.
vol. 12, no. 4, pp. 1248–1260, Dec. 2011. Informat., vol. 15, no. 6, pp. 3593–3601, Jun. 2019.
[119] R. Kalantari, M. Motro, J. Ghosh, and C. Bhat, “A distributed, collec- [137] G. Zyskind, O. Nathan, and A. S. Pentland, “Decentralizing
tive intelligence framework for collision-free navigation through busy privacy: Using blockchain to protect personal data,” in Proc.
intersections,” in Proc. IEEE 19th Int. Conf. Intell. Transp. Syst. (ITSC), IEEE Security Privacy Workshops, San Jose, CA, USA, 2015,
Rio de Janeiro, Brazil, 2016, pp. 1378–1383. pp. 180–184.
[120] D. Moriarty and P. Langley, “Learning cooperative lane selection strate- [138] R. Li, T. Song, B. Mei, H. Li, X. Cheng, and L. Sun, “Blockchain
gies for highways,” in Proc. 15th Nat. Conf. Artif. Intell. 10th Innovat. for large-scale Internet of Things data storage and protection,”
Appl. Artif. Intell. Conf., 1998, pp. 684–691. IEEE Trans. Services Comput., vol. 12, no. 5, pp. 762–771,
[121] D. Zhang, X. Han, and C. Deng, “Review on the research and practice Sep./Oct. 2019.
of deep learning and reinforcement learning in smart grids,” CSEE J. [139] M. A. Khan and K. Salah, “IoT security: Review, blockchain solu-
Power Energy Syst., vol. 4, no. 3, pp. 362–370, Sep. 2018. tions, and open challenges,” Future Gener. Comput. Syst., vol. 82,
[122] X. Lu, X. Xiao, L. Xiao, C. Dai, M. Peng, and H. V. Poor, pp. 395–411, May 2018.
“Reinforcement learning-based microgrid energy trading with a [140] A. Azaria, A. Ekblaw, T. Vieira, and A. Lippman, “MedRec: Using
reduced power plant schedule,” IEEE Internet Things J., vol. 6, no. 6, blockchain for medical data access and permission management,” in
pp. 10728–10737, Dec. 2019. Proc. 2nd Int. Conf. Open Big Data (OBD), Vienna, Austria, 2016,
[123] E. Foruzan, L.-K. Soh, and S. Asgarpoor, “Reinforcement learning pp. 25–30.
approach for optimal distributed energy management in a microgrid,” [141] A. Ramachandran and D. M. Kantarcioglu, “Using blockchain and
IEEE Trans. Power Syst., vol. 33, no. 5, pp. 5749–5758, Sep. 2018. smart contracts for secure data provenance management,” 2017.
[124] H. Wang, T. Huang, X. Liao, H. Abu-Rub, and G. Chen, [Online]. Available: http://arxiv.org/abs/1709.10000.
“Reinforcement learning in energy trading game among smart [142] Y. Zhang, S. Kasahara, Y. Shen, X. Jiang, and J. Wan,
microgrids,” IEEE Trans. Ind. Electron., vol. 63, no. 8, pp. 5109–5119, “Smart contract-based access control for the Internet of
Aug. 2016. Things,” IEEE Internet Things J., vol. 6, no. 2, pp. 1594–1605,
[125] H. Li, Z. Wan, and H. He, “Constrained EV charging scheduling based Apr. 2019.
on safe deep reinforcement learning,” IEEE Trans. Smart Grid, vol. 11, [143] A. S. Omar and O. Basir, “Identity management in IoT networks using
no. 3, pp. 2427–2439, May 2020. blockchain and smart contracts,” in Proc. IEEE Int. Conf. Internet
[126] R. Deng, Z. Yang, M.-Y. Chow, and J. Chen, “A survey on demand Things (iThings) IEEE Green Comput. Commun. (GreenCom) IEEE
response in smart grids: Mathematical models and approaches,” IEEE Cyber Phys. Soc. Comput. (CPSCom) IEEE Smart Data (SmartData),
Trans. Ind. Informat., vol. 11, no. 3, pp. 570–582, Jun. 2015. Halifax, NS, Canada, 2018, pp. 994–1000.
[127] C. Zhang, P. Patras, and H. Haddadi, “Deep learning in mobile and [144] E. S. Kang, S. J. Pee, J. G. Song, and J. W. Jang, “A blockchain-
wireless networking: A survey,” IEEE Commun. Surveys Tuts., vol. 21, based energy trading platform for smart homes in a microgrid,” in
no. 3, pp. 2224–2287, 3rd Quart., 2019. Proc. 3rd Int. Conf. Comput. Commun. Syst. (ICCCS), Nagoya, Japan,
[128] F. Tschorsch and B. Scheuermann, “Bitcoin and beyond: A technical 2018, pp. 472–476.
survey on decentralized digital currencies,” IEEE Commun. Surveys [145] F. Guo, F. R. Yu, H. Zhang, H. Ji, M. Liu, and V. C. M. Leung,
Tuts., vol. 18, no. 3, pp. 2084–2123, 3rd Quart., 2016. “Adaptive resource allocation in future wireless networks with
[129] Y. Liu, F. R. Yu, X. Li, H. Ji, and V. C. M. Leung, “Blockchain and blockchain and mobile edge computing,” IEEE Trans. Wireless
machine learning for communications and networking systems,” IEEE Commun., vol. 19, no. 3, pp. 1689–1703, Mar. 2020.
Commun. Surveys Tuts., vol. 22, no. 2, pp. 1392–1431, 2nd Quart., [146] J. Pan, J. Wang, A. Hester, I. Alqerm, Y. Liu, and Y. Zhao, “Edgechain:
2020. An edge-IoT framework and prototype based on blockchain and smart
[130] K. Yang, X. Jia, and K. Ren, “Secure and verifiable policy update contracts,” IEEE Internet Things J., vol. 6, no. 3, pp. 4719–4732,
outsourcing for big data access control in the cloud,” IEEE Trans. Jun. 2019.
Parallel Distrib. Syst., vol. 26, no. 12, pp. 3461–3470, Dec. 2015. [147] K. Salah, M. H. U. Rehman, N. Nizamuddin, and A. Al-
[131] S. Guo, X. Hu, S. Guo, X. Qiu, and F. Qi, “Blockchain meets edge com- Fuqaha, “Blockchain for AI: Review and open research
puting: A distributed and trusted authentication system,” IEEE Trans. challenges,” IEEE Access, vol. 7, pp. 10127–10149,
Ind. Informat., vol. 16, no. 3, pp. 1972–1983, Mar. 2020. 2019.
[132] J. Qiu, D. Grace, G. Ding, J. Yao, and Q. Wu, “Blockchain-based [148] C. Qiu, H. Yao, X. Wang, N. Zhang, F. R. Yu, and D. Niyato,
secure spectrum trading for unmanned-aerial-vehicle-assisted cellular “AI-chain: Blockchain energized edge intelligence for beyond
networks: An operator’s perspective,” IEEE Internet Things J., vol. 7, 5G networks,” IEEE Netw., vol. 34, no. 6, pp. 62–69,
no. 1, pp. 451–466, Jan. 2020. Nov./Dec. 2020.
[133] Y. Choi and I.-G. Lee, “Game theoretical approach of blockchain-based [149] A. Jäschke and F. Armknecht, “Unsupervised machine learning on
spectrum sharing for 5G-enabled IoTs in dense networks,” in Proc. encrypted data,” in Selected Areas in Cryptography (SAC), C. Cid
IEEE 90th Veh. Technol. Conf. (VTC-Fall), Honolulu, HI, USA, 2019, and M. J. Jacobson Jr., Eds. Cham, Switzerland: Springer, 2019,
pp. 1–6. pp. 453–478.

report e

Uploaded by

Copyright:

Available Formats

report e

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

report e

Uploaded by

Copyright:

Available Formats

IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO.

15, AUGUST 1, 2021 11891

Enabling Massive IoT Toward 6G:

Fig. 1. Roadmap of this article.

this common rule, several novel forward-looking scenarios,

Fig. 3. Schematic diagram of future 6G wireless systems.

TABLE II TABLE III

exert control actions by interacting with the environment. The

Fig. 7. Fully autonomous driving in 6G. CAV: connected autonomous

be enhanced with AI, e.g., AI-powered smart contracts [148].

You might also like