Channel-Prediction-Based One-Class Mobile IoT Device Authentication

Physical layer authentication (PLA) is a promising complement for the cryptographic-based authentication scheme, especially for Internet of Things (IoT) scenarios with massive devices. Traditional PLA schemes exploiting channel state information (CSI) face significant challenges in mobile communication scenarios due to the unknown variation of wireless channels. To address this challenge, we propose a PLA scheme based on Gaussian process (GP) channel prediction, where the variation of channel characteristics is tracked and predicted. Specifically, historical CSI attributes together with the transmitter's geographical information are exploited to establish a mapping to predict the next legitimate CSI for authentication. Furthermore, to overcome the impracticality of applying conventional PLA framework for authentication, where an unrealistic assumption that either the prior knowledge of the adversary's statistical channel model or even the real observations of its CSI data is required, we propose the so-called one-class authentication (OCA) scheme, which does not require any attacker's channel information. We exploit the quasideterministic radio channel generator (QuaDRiGa) simulation platform as the generator of CSI for experimental verifications. Simulation tests are performed to demonstrate that our method improves authentication performance significantly in time-varying scenarios.