부산대학교 도서관

Wireless communication system are facing more and more security threats, and protecting user privacy becomes important. The goal of covert communication is to hide the existence of legitimate transmission as a practical approach. Inspired by the great success of deep reinforcement learning (DRL) on handling challenging optimization problems, DRL is used to optimize covert communication performance. To achieve this, a model-free and off-policy deep deterministic policy gradient (DDPG) algorithm is proposed to maximize the covert rate under covert constraint. The transmit beamformer vector of legitimate transmitter and the phase shifts matrix of intelligent reflecting surfaces (IRS) are the outputs of DRL neural networks. The DRL can learn from the environment and adjust transmit beamformer vector and phase shifts matrix to maximize covert communication performance. Simulation results demonstrate that the proposed DDPG algorithm can achieve comparable performance with two benchmarks algorithm.