Dilated Causal Convolutional Model For RF Fingerprinting

Abstract

We design a network to classify individual wireless devices based on their radio frequency (RF) fingerprints imparted on transmitted signals. The network combines a stack of dilated causal convolution layers with traditional convolutional layers which we call an augmented dilated causal convolution (ADCC) network. It is designed to work on real-world Wi-Fi and ADS-B transmissions, but we expect it to generalize to any classes of signals. We explore various aspects of the ADCC for RF fingerprinting including: classification of up to 10,000 devices, sensitivity to training set size, varying signal-to-noise ratios, and channel propagation effects.