Naïve Bayes Supervised Learning based Physical Layer Authentication: Anti-Spoofing techniques for Industrial Radio Systems

Abstract

Physical Layer Security (PLS) based authentication schemes are an alternative to conventional security schemes such as e.g. certificates or Message Authentication Codes (MACs). They can provide a more lightweight solution compared to traditional cryptography in order to meet the requirement of secure data transmission. However, errors in Physical Layer Authentication (PLA) techniques can occur due to adverse influences resulting from PLS schemes such as receiver noise or fading channels. Skillfull methods are therefore required in order to detect anormal system behaviour. One promising solution are supervised classification schemes. The application of Naïve Bayes (NB) classifiers for PLA is therefore proposed and evaluated within this work. Prior to that, we analyse the resource efficiency within typical Ultra Reliable Low Latency Communication (URLLC) applications and conduct a security overhead analysis. We propose strategies in order to overcome the problem of missing training data from either the URLLC user or attacker node. A real world Software Defined Radio (SDR) based testbed using OFDM (Orthogonal Frequency Division Multiplexing) is implemented in order to evaluate the performance of NB based PLA. The measurements are conducted within the german campus network frequency band (LTE band 43). Further, we conduct a hyperparameter optimization (HPO) based on random search. The investigated classifiers show promising results in terms of authentication accuracy and Receiver Operating Characteristic (ROC) curve performance.