Comparison of Statistical and Machine Learning Techniques for Physical Layer Authentication

In this paper we consider authentication at the physical layer, in which authenticator aims distinguishing a legitimate supplicant from an attacker on basis of characteristics set parallel wireless channels, are affected by time-varying fading. Moreover, attacker's channel has spatial correlation with supplicant's one. setting, assess and compare performance achieved different approaches under conditions. We first use two statistical decision methods, prove that using large number references (in form estimates) levels fading is not beneficial security point view. then classification methods based machine learning. order to face worst case scenario provided no forged messages during training, one-class classifiers. When instead training includes some messages, resort more conventional binary classifiers, considering cases such either labelled or not. For latter case, exploit clustering algorithms label set. The both nearest neighbor (NN) support vector (SVM) techniques evaluated. Through numerical examples, show same probability false alarm, (OCC) achieve lowest missed detection when small exists between main adversary one, while advantageous channels large.