Deep Reinforcement Learning Based Resource Allocation for Intelligent Reflecting Surface Assisted Dynamic Spectrum Sharing

Abstract

For cognitive radio (CR) systems, it is challenging to achieve high data rate for secondary user (SU) while avoiding the cross-link interference to primary user (PU). In this paper, we propose an intelligent reflecting surface (IRS) aided dynamic spectrum sharing scheme, where an IRS assisted SU link dynamically shares the spectrum opportunities with a PU link based on the eavesdropped automatic repeat request (ARQ) feedback from the PU. We model the dynamic spectrum sharing problem as a Markov decision process (MDP), where we jointly optimize the SU transmit power and IRS reflect beamforming to maximize the cumulative expected achievable rate of the SU link subject to the outage constraint of the PU link. We propose a deep reinforcement learning (DRL) algorithm to achieve the optimal spectrum sharing policy. Experimental results show that the proposed proximal policy optimization (PPO) algorithm outperforms the traditional policy gradient (PG) and deep deterministic policy gradient (DDPG) algorithms by 100% and 46% in terms of spectrum utilization efficiency, respectively.