Physical layer authentication of Internet of Things wireless devices using convolutional neural networks and recurrence plots

Abstract

This letter addresses the problem of Internet of Things (IoT) authentication when cryptographic means are not feasible or they have limited applicability due to constraints in the context or due to the limited computing capabilities of the IoT devices. This letter presents a novel approach for the authentication of IoT wireless devices using their Radio Frequency (RF) emissions where Convolutional Neural Networks (CNN) in combination with Recurrence Plots (RP) are applied. In recent years various studies have demonstrated that wireless devices can be authenticated on the basis of their RF emissions because physical differences generate different features in the RF signal during communication. Recently Deep Learning (DP) techniques based on CNN and other algorithms have been applied to this authentication problem. In this letter we present a novel application of CNN based on the use of RP where the original time series derived from the digitized RF emissions is transformed into images on which CNN are applied. The proposed approach is applied to an experimental data set of RF emissions collected from 11 IoT devices of the same model. The results show that this technique can provide superior performance if compared to conventional machine learning algorithms based on the extraction of statistical features.