IEEE Computational Intelligence Magazine - February 2020 - 46

cloud units. In such networks, all the functions originally implemented in base stations
(BSs) and core networks, including storage,
signal processing, scheduling, power control,
computations, access control, and mobility
management, and so on, are migrated to the
cloud [19], [20]. With powerful cloud technologies, these functions could be executed
efficiently and robustly. Furthermore, since
only radio frequency (RF) units are remained in front ends,
the complexity and operational cost of RRH networks will
be greatly reduced. Fronthaul networks are used to establish
connections and manage information exchanges between
RRH networks and the cloud, which normally adopt wired
physical mediums, such as cable and fiber, to enable high data
rate and low latency data delivery. Based on the control information and signaling from the cloud, IoT users will conduct
data transmissions and send feedback information to the
cloud, such as channel state information (CSI), measurements
report, and ACK/NACK signaling. Based on the feedback
information, the cloud will reconfigure transmission parameters for all IoT users to maintain communication quality. All
the transmissions between IoT users and the cloud will be
forwarded by RRH networks and fronthaul networks.
Obviously, centralized control could achieve good performance for IoT networks, since complicated algorithms and
mechanisms could be employed to globally optimize network
performance [21], [22]. However, the overhead of computations and signal processing would exponentially increase with
the growing number of IoT users. Therefore, centralized IoT
networks will introduce substantial control overhead due to a
large number of IoT users. Furthermore, centralized control
will restrict the flexibility and scalability of the underlying IoT
network, since IoT users may be required to follow the same
protocol. For example, to conduct channel measurements, the
control center in Fig. 1 may need to conduct measurement
configurations for all IoT users within the network, based on
which IoT users insert user-specific pilots, also referred as to
reference signals (RSs), in their physical (PHY) radio frames
[23]. Additionally, effective measurements of user-specific pilots
require the underlying pilots from different users to be synchronized and orthogonalized.

In the future, IoT networks may be configured with
a control center to provide centralized control to be
compatible with existing cellular mobile networks, or
may be designed in a distributed fashion for better
flexibility and scalability.
that may be encountered in distributed IoT networks. Finally,
we conclude the article in Section V.
II. Network Architecture

To be compatible with existing cellular mobile networks such
as the third Generation Partnership Project (3GPP) Long-Term
Evolution (LTE)/LTE-Advanced, IoT networks may be configured with a control center to provide centralized control.
On the other hand, for better flexibility and scalability, IoT networks may be also designed in a distributed fashion. Therefore,
in this section, the frameworks of both centralized and distributed IoT networks will be discussed.
A. Centralized IoT Networks

Currently, centralized IoT networks based on cloud computing technologies or fog computing technologies are widely
studied due to their remarkable advantages [16]-[18]. As
shown in Fig. 1, cloud computing technologies based IoT
networks consist of three main components: remote radio
head (RRH) networks, fronthaul networks, and baseband

Operator 2
Operator 1

Server 1
Server 2

Scheduling
Power Control
Computing

Storage
Access Control
Signal Processing

Mobility Management

FIGURE 1 The architecture of centralized IoT networks.

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | FEBRUARY 2020

B. Distributed IoT Networks

Based on the above discussions, distributed IoT networks may
be more appealing compared with the centralized ones due to
low complexity, high flexibility, and high scalability [24]. The
architecture of distributed IoT networks can be seen most clearly in Fig. 2. Fig. 2 shows a low-complexity architecture that a
distributed IoT network can be configured to. In distributed
IoT networks, only low-complexity access points (APs) are
needed to be deployed with simplified protocol stacks. For
example, for PHY and medium access control (MAC) layers, an
AP only needs to fulfill some simple functions, such as modulation/demodulation, coding/decoding, and automatic repeat

IEEE Computational Intelligence Magazine - February 2020

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