DDPG-PER for an IRS-Aided Secure Wireless Communication

Improving the security of communications data in a wireless network is one of the most important uses of reconfigurable intelligent surfaces (IRS), especially in next generation networks like 6G. In this context, we propose an approach based on deep a reinforcement learning (DRL) algorithm to determine the optimal phase shift of IRS elements. We used the Deep Deterministic Policy Gradient (DDPG) variant, which adapts to continuous data and improves the secrecy rate by considering, in the algorithm, the best sample obtained via a Prioritized Experiment Replay (PER).