Machine Learning based Decision Stratigies for Physical Layer Authentication in Wireless Systems

Abstract

In this paper, machine learning (ML) based decision strategies for physical layer authentication are presented. The intelligent authenticators learn the channel features and then classify the received message based on the channel attributes into two categories, legitimate or illegitimate. The training set construction using different features of the estimated channel fading coefficients explored. In addition, ML based physical layer authentication is compared with the statistical discriminative function formulated in binary hypothesis test with a pre-defined threshold. Simulation results demonstrated that the performance of intelligent authenticators outperform the statistical decision scheme as significant improvement can be achieved in the detection rate with minimum false alarm rate. The overall authentication accuracy measured in terms of the area under the receiver operating characteristic curve (AVC) confirmed the superior performance of the the support vector machine (SVM) based physical layer authentication compared with other ML approaches. In addition, it is concluded that using two distinct features improves the authentication performance compared with feature space constructed only from test statistic metrics.