Article Preview
Top1. Introduction
The main motivation for the development of cognitive radio is to overcome the conflicts between spectrum demand growth and spectrum underutilization arising from legacy static spectrum allocation policies (López-Benítez, 2018).
A CR must be intelligent and able to learn from its experience by interacting with its radio frequency (RF) environment. As a result, learning is an indispensable part of the CR that can be provided using artificial intelligence and machine learning techniques. Indeed, the application of artificial intelligence and especially machine learning in cognitive radio networks (CRNs) has recently aroused great interest in the literature.
Learning aims to make machines perform tasks similar to those of an expert. The intelligent machine will perceive its environment and make decisions to maximize its own utility. According to Woods (1986), artificial intelligence focuses on deduction, reasoning, problem solving, knowledge representation and learning.
According to Abbas et al. (2015), the learning process in CRNs is illustrated in Figure 1 and can be presented as follows:
- ▪
Sensing of radio frequency parameters
- ▪
Observation of the environment and analysis of its reactions
- ▪
Learning
- ▪
Retention of decisions and observations for updating the model and finally
- ▪
Decide on resource management issues and adjustment of transmission errors accordingly (Bkassiny et al., 2013; Russel et al., 2016)
Figure 1. Learning process in CRNs (Abbas et al. (2015))
In the literature, there is some work that has focused on the application of artificial intelligence techniques in CRNs. In Zhao et al. (2012), the authors introduced the use of artificial intelligence and machine learning techniques in the CR. They also presented the possible applications and fundamental ideas around the CR.
A survey on different techniques of artificial intelligence has been presented in Bkassiny et al. (2013).
The difficulties encountered in the application of these techniques in CR, their advantages and disadvantages were also discussed in the paper.
In Gavrilovska et al. (2013), the authors studied game theory, reinforcement learning, and reasoning approaches such as Bayesian networks, fuzzy logic, and case-based reasoning in CRNs.
In 2015, the survey presented in Abbas et al. (2015) was devoted to fuzzy logic, genetic algorithms, neural networks, game theory, reinforcement learning, support vector machines, case-based reasoning, Bayesian networks, Markov models, multi agent systems and artificial bee colony algorithms.
However, we noticed that the paper previously mentioned, did not discuss the application of the particle swarm optimization (PSO), a metaheuristic widely used in CRNs. Other metaheuristics, more recent and used in CRNs, are not mentioned in Abbas et al. (2015), like firefly algorithm (FA), cuckoo search (CS), gravitational search algorithm (GSA), invasive weed optimization (IWO) and flower pollination algorithm (FPA).
As, the state of the art of a part of techniques has been presented in the paper published in 2015. In this paper, we are interested in most artificial intelligence techniques that have been used only in the last three years in CRNs (between 2016 and 2019).
The main objectives of this survey are as follows:
- ▪
Presents a comprehensive study on artificial intelligence techniques, their definitions, their applications in CRNs, their evaluations, their strengths and their challenges. Some techniques were never discussed before in the previous survey.
- ▪
Presents the main tasks of CR and their corresponding challenges.
- ▪
Categorizes the techniques presented according to the type of learning (supervised or unsupervised) and presents their applications according to the tasks of the CR.