Machine Learning-based PHY-authentication for Mobile OFDM Transceivers

Abstract

In this paper, a machine learning (ML) framework is devised for physical layer (PHY) authentication in mobile orthogonal frequency-division multiplexing (OFDM) transceivers. The ML framework utilizes various classification models that exploit features extracted to capture the unique hardware behavior of different transmitters, namely: coarse carrier frequency offset (CFO), fine CFO, and residual CFO. These features are leveraged to train various classification models to distinguish between legitimate and non-legitimate transmitters. Since these features are mainly dependent of the hardware behavior more than the channel behavior, they are more resilient in mobility scenarios where channels are more dynamic. For validation, we adopt a software-defined radio (SDR) testbed to record the pertinent measurements in indoor and outdoor mobility environments. It is shown that the proposed approach can distinguish between legitimate and malicious transmitters effectively. Among various classifiers adopted, support vector machine (SVM) gives a true positive rate (TPR) classification of 0.97 with a false positive rate (FPR) of 0.02 in indoor environment; and a TPR of 0.93 with an FPR of 0.05 in outdoor environment.