Deep Reinforcement Learning-Based Relay Selection Algorithm in Free-Space Optical Cooperative Communications

Abstract

Relay-aided free-space optical (FSO) communication systems have the ability of mitigating the adverse effects of link disruption by dividing a long link into several short links. In order to solve the relay selection (RS) problem in a decode and forward (DF) relay-aided FSO system, we model the relay selection scheme as a Markov decision process (MDP). Based on a dueling deep Q-network (DQN), the DQN-RS algorithm is proposed, which aims at maximizing the average capacity. Different from relevant works, the switching loss between relay nodes is considered. Thanks to the advantage of maximizing cumulative rewards by deep reinforcement learning (DRL), our simulation results demonstrate that the proposed DQN-RS algorithm outperforms the traditional greedy method.