Deep Deterministic Policy Gradient-Based Physical Layer Authentication Scheme Under Unknown Attacking Environment

Physical layer authentication (PLA) is considered as a promising method to resist spoofing attacks, where the stochastic features of wireless channels are used to detect attackers. Most of the existing PLA schemes assume that the prior information of attackers is known by the receiver, which might not be realized in realistic networks. To address this issue, we propose a deep deterministic policy gradient based PLA (DPLA) scheme to identify legitimate transmitters and attackers under unknown attacking environment. Specifically, the deep deterministic policy gradient approach is employed in the proposed DPLA scheme, where two types of deep neural networks are used to adaptively adjust the authentication strategy in continuous action space by learning both the policy and the state-action value. Moreover, the double-Q approach and the delayed update policy network are integrated into the proposed scheme to reduce the overestimation bias in the value estimation process and ensure the stability of the policy learning. Simulation results show that the proposed scheme can achieve a substantial performance gain over several reference schemes.