Dynamic Spectrum Access in Cognitive Radio Networks Using Deep Reinforcement Learning and Evolutionary Game

With the rapid development of wireless communication technology, the low utilization of spectrum resources and the high demand for spectrum have always been an urgent and paradoxical problem to be resolved. In order to alleviate this conflict, cognitive radio technology has been proposed. In this paper, we propose a new method of distributed multi-user dynamic spectrum access in cognitive radio network through combining deep reinforcement learning with evolutionary game theory. This method utilizes the Deep Q-network (DQN) as the main framework, and each user independently performs DQN algorithm to select channel. Through dynamic spectrum management, the utilization of spectrum resources can be effectively improved. In addition, we introduce the replicator dynamic using evolutionary game theory into the setting of the reward function for reinforcement learning, so as to effectively balance the collaboration among users. The simulation results show that the proposed algorithm can significantly reduce the collision rate of cognitive users and effectively increase the system capacity.