Physical Layer Security: Detection of Active Eavesdropping Attacks by Support Vector Machines

This article presents a framework for converting wireless signals into structured datasets, which can be fed machine learning algorithms the detection of active eavesdropping attacks at physical layer. More specifically, communication system, consists an access point (AP), K legitimate users and eavesdropper, is considered. To detect eavesdropper who breaks system during authentication phase, we first build datasets based on different features then apply sophisticated support vector (SVM) classifiers to those datasets. more specific, process received by AP define pair statistical post-processing signals. By arranging simulate entire transmission constructing features, form so-called artificial training data (ATD). SVM ATD, classify associated with nonattacks, thereby detecting presence eavesdropper. Two are considered, including classic twin-class (TC-SVM) single-class (SC-SVM). While TC-SVM preferred in case having perfect channel state information (CSI) all channels, SC-SVM realistic scenario when have only CSI users. We also evaluate accuracy trained models depending choice kernel functions, eavesdropper's power. Our numerical results show that careful parameter-tuning required exceeding probability 95%.