Deep Reinforcement Learning for Secrecy Energy Efficiency Maximization in RIS-Assisted Networks

This paper investigates the deep reinforcement learning (DRL) for maximization of the secrecy energy efficiency (SEE) in reconfigurable intelligent surface (RIS)-assisted networks. An SEE maximization problem is formulated under constraints of the rate requirement of each (legitimate) user, the power budget of the transmitter and the discrete phase shift coefficient of each reflecting element at the RIS by jointly optimizing the beamforming vectors for users and the artificial noise vectors for eavesdroppers as well as the phase shift matrix. The considered problem is first reformulated into a Markov decision process with the designed state space, action space and reward function, and then solved under a proximal policy optimization (PPO) framework. Numerical results are provided to evaluate the optimality, the generalization performance and the running time of proposed PPO-based algorithm.