RiftNet: Radio Frequency Classification for Large Populations
Josh Robinson, Scott Kuzdeba
Abstract
Radio transmitters can be identified by their unique radio frequency (RF) fingerprints. Hardware differences and manufacturing variations between different RF devices impart the unique RF fingerprints to their transmitted waveforms, which can be exploited to accurately distinguish them. Most previous work has focused on using expert-defined physics-based features to distinguish between 100 or less individual devices, with more recent trends turning to machine learning, and deep learning in particular. In this paper, we present a deep learning approach that scales to 10,000 devices, across different signal types, with upwards of 95% accuracy, which we call RiftNet (Resampling in Frequency and Time Network). RiftNet is particularly suited to pulling out generic signal features that may be relevant to other RF machine learning tasks, beyond just fingerprinting,