ANOVA-based RF DNA analysis: Identifying significant parameters for device classification

Abstract

Analysis of variance (ANOVA) is applied to RF DNA fingerprinting techniques to ascertain the most significant signal characteristics that can be used to form robust statistical fingerprint features. The goal is to find features that enable reliable identification of like-model communication devices having different serial numbers. Once achieved, these unique physical layer identities can be used to augment existing bit-level protection mechanisms and overall network security is improved. ANOVA experimentation is generated using a subset of collected signal characteristics (amplitude, phase, frequency, signal-to-noise ratio, etc.) and post-collection processing parameters (bandwidth, fingerprint regions, statistical features, etc.). The ANOVA input is percent correct device classification as obtained from MDA/ML discrimination using three like-model devices from a given manufacturer. Full factorial design experiments and ANOVA are used to determine the significance of individual parameters, and interactions thereof, in achieving higher percentages of correct classification. ANOVA is shown to be well-suited for the task and reveals parametric interactions that are otherwise unobservable using conventional graphical and tabular data representations.