IEEE Computational Intelligence Magazine - February 2020 - 45

Abstract-The explosive growth of wireless
In general, dense active users or machine devices
devices motivates the development of the
internet-of-things (IoT), which is capable of
will bring two major challenges: increased
interconnecting massive and diverse "things"
probability of access collisions and complicated
via wireless communications. This is also called
resource management.
massive machine type communications (mMTC)
as a part of the undergoing fifth generation
services to a large amount of dense wireless users or machine
(5G) mobile networks. It is envisioned that more sophisticated
devices. In general, dense active users or machine devices will
devices would be connected to form a hyperconnected world
bring two major challenges: increased probability of access
with the aids of the sixth generation (6G) mobile networks. To
collisions and complicated resource management. When
enable wireless accesses of such IoT networks, artificial intelliaccess collisions occur, users/devices will necessarily relaunch
gence (AI) can play an important role. In this article, the frameradio access procedures leading to increased transmission
works of centralized and distributed AI-enabled IoT networks
latency and degraded quality of service (QoS). For delay-senare introduced. Key technical challenges, including random
sitive data, latency is a very important factor causing the failaccess and spectrum sharing (spectrum access and spectrum sensure of the underlying transmissions. Dense active users or
ing), are analyzed for different network architectures. Deep reinmachine devices will bring additional difficulties in wireless
forcement learning (DRL)-based strategies are introduced and
resource management, including scheduling and power conneural networks-based approaches are utilized to efficiently
trol due to the additional complications caused by the
realize the DRL strategies for system procedures such as specunderlying wireless channel and inter-user interference. Furtrum access and spectrum sensing. Different types of neural netthermore, in order to effectively conduct resource manageworks that could be used in IoT networks to conduct DRL are
ment, channel measurements need to be conducted to obtain
also discussed.
the underlying channel state information, the overhead of
which could be substantial with a large number of active
I. Introduction
users and devices. Second, IoT networks should be able to
he emerging waves of Internet-of-things (IoT) parahandle heterogeneous wireless users or machine devices.
digm are attracting increasing attention from both
Diverse wireless devices may co-exist in IoT networks with
industry and academia via its promising applications
different QoS requirements and priorities. For example, by
of intelligently connecting things or objects around
proper resource management, IoT networks should give prius without human interventions [1]. With the automatic
ority to guarantee reliable transmissions of emergency inforinteractions among numerous devices, such as radio frequency
mation, such as health care and fire alarms.
identification (RFID), sensor networks, and mobile phones,
The existing methods to tackle the aforementioned issues
IoT can assist our daily life in various ways including ubiquimainly rely on complicated algorithms, mechanisms and
tous healthcare, control of home equipment, environmental
schemes [13], [14] where specific models are used. Due to the
monitoring, smart grid, and so on [2]-[4]. Nowadays, the evolution of the fifth generation (5G) wireless networks becomes
heterogeneous nature of IoT networks, such model-depenthe main driver for world-wide IoT development [5], [6].
dent solutions can not effectively adapt to real environment.
Even though the surge of 5G networks has not fully arrived
Furthermore, algorithm-based approaches may be prohibiyet, it is time to envision what will appear in the sixth genertively complicated preventing them from being adopted in
ation (6G) wireless networks [7]. Towards this end, our paper
practice. On the other hand, artificial intelligence (AI)-based
will not only provide solutions to technical challenges in 5G
approaches can play an important role to handle resource
mobile networks but also provide visions on how these solumanagement among a massive number of devices. Accordingtions may resolve issues in future 6G networks.
ly, deep neural networks and big data analytics are naturally
According to Gartner, there will be more than 30 billion
suitable for information extraction from high-dimensional
IoT devices by the year 2025 [8]. Accordingly, the 5G network
data and complicated decision makings [15]. We do believe
and future 6G networks are expected to support a massive
that the AI-supported hyper-connected world would be an
number of smartly connected IoT devices [9], [10]. Our existintegral core part of future 6G wireless networks. In this artiing wireless technologies may break down when such a trecle, more efficient and overhead-friendly approaches using AI
mendous number of devices (things) attempt to access the
will be explored and introduced.
network. To achieve the massive access of IoT devices, it is
The remainder of this article is organized as follows. In Secimportant to redesign and improve our existing networks [11].
tion II, we introduce the frameworks of both centralized and
Compared with conventional wireless communication netdistributed IoT networks. Then, in Section III, potential techworks targeting mainly at human communications, IoT netnical challenges of centralized IoT networks are analyzed, and
works, which also allow machine-to-machine communications,
the corresponding solutions based on AI technologies are
have unique characteristics and capability requirements [12].
introduced to address these challenges. In Section IV, we introFirst, IoT networks should be able to provide wireless access
duce multiple AI-based schemes to tackle technical challenges

FEBRUARY 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

IEEE Computational Intelligence Magazine - February 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - February 2020