Distributed spectrum sensing and access through deep recurrent Q-networks
Manish Kumar Giri, Saikat Majumder · 2022 IEEE 6th Conference on Information and Communication Technology (CICT) · 2022
Dynamic spectrum access (DSA) has gained significant prominence for spectrum utilization. In DSA avoiding collision with the PUs and maintaining coordination with other SUs are the critical problems. These problems become severe in a distributed scenario, where no centralized controller is utilized. In this paper we propose a deep recurrent Q-network (DRQN) based approach for distributed sensing and access. In the proposed approach, the SUs will learn strategies in a distributed manner with having the system dynamics. Moreover, we utilized a special type of recurrent neural network (RNN) called long short-term memory (LSTM) to realize the deep reinforcement learning (DRL). The proposed approach is a combination of DRQN and LSTM, in which SUs will take decision independently. The simulation results present the suitability of proposed approach to handle the dynamic environment. The proposed approach achieves similar results in terms of reward value, and it further provides improvement in the collision number compared to the other techniques.