Deep Reinforcement Learning Based Opportunistic Routing for Cognitive Relay Networks
Jintaek Oh, Hyunjoon Suh, Shinhyeok Kang, Taewon Hwang · 2023
In this paper, we consider a cognitive radio (CR) relay network where a secondary relay network and a primary network coexist and share the spectrum. We propose a deep reinforcement learning (DRL) based opportunistic routing (OR) scheme for the secondary relay network to maximize the packet reception probability at a destination via multi-hop relay under the QoS constraint of the primary network. We model the routing problem of the secondary relay network with the QoS constraint of the primary network as a constrained Markov decision process (CMDP). To solve the CMDP, we use a Lagrangian relaxation and obtain an unconstrained MDP. We use deep Q-learning (DQL) to solve the problem. Based on Lagrangian relaxation and DQL, the proposed DRL-based OR scheme finds optimal routing decisions. Simulation results show that the proposed DRL-based OR scheme can improve the packet reception probability of the secondary relay network while satisfying the QoS constraint.