Mobile Physical-Layer Authentication Using Channel State Information and Conditional Recurrent Neural Networks

Channel state information offers a unique characteristic that can be used for authenticating devices at the physical layer. The quickly changing response of the mobile channel presents a challenge for identifying trusted transmitters. In this paper, we propose the use of recurrent neural networks to predict these changes in order to make an authentication decision. Using previous channel response measurements, we condition the output of the network and estimate future channel responses. This work presents a novel method for physical layer authentication using two variations of a conditional generative adversarial network (CGAN) and evaluates the CGAN accuracy against networks using long short-term memory (LSTM) and gated recurrent unit (GRU) cells. Performance evaluation reveals promising results for the GRU-enabled CGAN achieving 96.2% accuracy as well as the LSTM-enabled CGAN reaching 90.9%, compared with the GRU and LSTM standalone networks, scoring 98.1% and 88.9%, respectively.