One-class Physical Layer Authentication under Time-varying Fading Channels

Physical layer authentication (PLA) is an effective complement for the upper-layer scheme due to its high reliability and low complexity. Conventional channel characteristics-based PLA schemes face serious performance degradation due to the change of wireless channel. Besides, the unrealistic assumption that the illegal channel observations are known is adopted by most current schemes. In this paper, we propose a channel prediction based one-class PLA scheme which only needs the legal channel information. Specifically, we analyse the limitation of the Gaussian process (GP) -based channel prediction method and propose to use periodically updated long short-term memory network (PULSTM) to track the channel variation and perform channel prediction. Bayes estimation is used to perform the one-class authentication after channel prediction and subcarrier selection is used to improve the authentication performance. Quasi Deterministic Radio channel Generator (QuaDRiGa) platform is exploited in our simulation part. Simulation results show the better performance of our proposed scheme than some existing schemes under dynamic scenarios.