Distributed reinforcement learning for dynamic spectrum allocation in cognitive radio-based internet of things

Cognitive Radio (CR) with other advancements such as the Internet of things and machine learning has recently emerged as the main involved technique to use spectrum in an efficient manner. It can access the spectrum in a fully dynamic way and exploit the unused spectrum resources without creating any harm to cognitive users. In this paper, the authors develop a CR access strategy founded on the implementation of an efficient Deep Multi-user Reinforcement Learning algorithm based on a combination of a Deep neural network, Q-learning, and cooperative multi-agent systems. The proposed approach consists of two stages: the user choice algorithm to set up an agent's activation order, and the frequency choice method to select the optimal channel on the appropriate bandwidth. Reasonable implementation is proposed, and the obtained results demonstrate that the authors’ approach can improve wireless communication for all CR terminals. It shows satisfactory performances in terms of user satisfaction degree and the number of used channels and can keep the channel allocation plan always in the appropriate state. © 2022 The Authors. IET Networks published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.