Optimized fingerprint generation using unintentional emission radio-frequency distinct native attributes (RF-DNA)

Abstract

Device discrimination has been effectively demonstrated using classification processes acting on RF-DNA features as input sequences. Device discrimination utilizing RF-DNA classifiers requires training signals representative of the expected test signals that capture device uniqueness. Current techniques divide collected signals into uniformly distributed and sized regions prior to generating the RF-DNA feature input sequences. This paper divided the collected signals using non-uniform regions tailored to the device operations. Early results indicate that using non-uniform regions for fingerprint generation do not result in increased detection performance for the specific signals considered.