Online machine learning-based physical layer authentication for MmWave MIMO systems

Transmitter authentication is significantly critical to secure millimeter wave (mmWave) multiple-input multiple-output (MIMO) systems. By utilizing the mmWave MIMO channel and carrier frequency offset (CFO), this paper proposes a novel online machine learning-based physical layer authentication scheme to validate transmitter identities. In particular, with the help of the Gaussian kernel method and representer theorem, we formulate the physical layer authentication model as a binary classification model. Based on the developed classification model and a constructed convex objective function, we then design an online algorithm to achieve updating authentication parameters, and thus establish a novel online machine learning-based physical layer authentication scheme. Based on the developed authentication model and theories of statistical hypothesis testing, we statistically derive the analytical expressions for false alarm and detection rates for the proposed authentication scheme. Finally, we carry out extensive experiments to verify the authentication performance in terms of the false alarm and detection rates. In addition, we investigate the related authentication efficiency issues (e.g., convergence process) and the capability of resisting against spoofing attacks in mmWave MIMO systems.