IEEE Systems, Man and Cybernetics Magazine - January 2023 - 29

utilized. Therefore, an intelligent network infrastructure
architecture is urgently needed to facilitate the integration
of spectrum, devices, and other resources, which is
expected to achieve the ultralow latency and ultrareliable
delivery of IoT-based critical applications anytime
and anywhere.
The Cognitive Network
Construction Architecture
The cognitive network construction architecture
explored in this article can be shown in Figure 1. First,
the cognitive network brokers will purchase the original
spectrum status data from primary users and the rental
intention information from the owners of idle smart
devices, where secondary users do not need to participate
in perception. Then, they rent the cognitive computing
infrastructure (e.g., computing resources of
various devices) to perform data processing and value
mining, where the goal is to obtain the available spectrum
as well as the idle smart devices that are willing to
provide IoT services. The relay communication style
through the devices in cellular networks can make the
sub-6-GHz frequency band easy for reuse. On the other
hand, the brokers will collect network requirements
from the demand side for network construction, and
then, they also rent the cognitive computing infrastructure
to execute a match between the obtained resources
(i.e., the available spectrum and idle smart devices) and
the collected network performance requirements, where
the goal is to obtain the desired cognitive network and
sell it to the demand side.
Mechanisms for Data-Driven Cognitive
Network Architecture Systems
This section introduces a data-driven cognitive network
construction architecture that integrates the spectrum
and devices to provide rich communication resources,
where the cognitive computing technologies are used to
effectively improve the stability of cognitive networks.
As shown in Figure 1, on the one hand, when a broker
purchases enough spectrum status data from primary
users and rental intention information from the owners
of idle smart devices, it can mine the using rule of the
cognitive frequency band and plan the using strategy of
idle smart devices.
On the other hand, when these idle smart devices are
used as communication access points or relays in future
cognitive networks, they can effectively improve the
quality of future cognitive network coverage, especially in
a temporary congestion area. Moreover, when the
intelligent scheduling policies are designed and integrated
into this architecture, even fragmentary
resources (e.g., the fragmented spectrum utilization
opportunity or the irregular idle time of smart devices)
can be used to maintain a relatively stable cognitive
network topology.
A Hierarchical Architecture for Cognitive
Network Construction
For future cognitive network construction, we propose a
hierarchical architecture system that comprehensively utilizes
the computing resources of the network edge and
cloud infrastructure, fully mines the available spectrum
and devices, and intelligently schedules the spectrum and
devices to improve the topology stability. Figure 2 illustrates
the hierarchical architecture system for future cognitive
network construction, which consists of three
layers: the data acquisition, information mining, and network
service layers.
The data acquisition layer involves the collection of
spectrum status data, the idle time and rental intention of
smart devices, and cognitive network performance
requirements. Since the dispatchers of spectrum resources
(e.g., small base stations and macro base stations), serving
as the primary users, fully know the spectrum usage status,
it will be more convenient for them to actively offer the
spectrum status data. On the one hand, remuneration
should be paid to motivate the primary users to report the
spectrum status data in a stable way. On the other hand,
the owners of idle smart devices should be motivated to
offer their rental intention information. In addition, the
demand side of future cognitive networks, which may be a
representative of smart terminals in a temporary congestion
area, will report the performance requirement for a
desired cognitive network. The representative, serving as
the secondary user, possesses only the limited computing
resources and inaccurate spectrum information, incurring
the unaffordable costs of network construction and maintenance.
Therefore, it is more sensible for a secondary
user, acting as a demander, to buy network construction
and maintenance services from the brokers.
The information mining layer includes the spectrum
and device cognitive engine as well as the networking cognitive
engine. These two engines are deployed in the network
edging facilities and the cloud infrastructure rented
by the brokers. The spectrum and device cognitive engine
processes and analyzes the spectrum status data coming
from primary users and the rental intention of the owners
of idle smart devices, while the networking cognitive
engine deals with the network construction requirements
coming from secondary users. Network edge facilities are
introduced to preprocess the collected data, such as cleaning,
filtering, and fusion. Therefore, on the one hand, it
reduces the unnecessary network traffic and lowers the
storage and processing overhead of cloud facilities. On the
other hand, it also provides high-quality data for data mining
engines running in the cloud infrastructure.
The network service layer is responsible for the corresponding
match between the obtained resources (i.e., the
available spectrum and idle smart devices) and the cognitive
network construction requirements. Through the use
of network edge facilities, it can transmit the instructions
to secondary users in a timely manner. At the same time,
January 2023 IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE 29
IEEE Systems, Man and Cybernetics Magazine - January 2023

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - January 2023
