Signature-Aware RF Exploitation (SNARE) Fingerprinting using Deep Learning to identify UAVs

Currently, there are emerging challenges with non-cooperative unmanned aerial vehicle (UAV) operating in urban airspaces. Likewise, law-enforcement groups need UAV detection capabilities to respond for safety and security (e.g., as defined for situation awareness at airports and forest fires where protocols restrict UAVs from operating within a few nautical miles from the event). This paper presents a novel physical layer authentication solution to identify UAVs that have identical visual signatures such as the same drone type and manufacturer. Within each UAV, the radio frequency (RF) signals transmitted from UAVs have a unique signature, called RF fingerprint, that can be used to distinguish among UAVs. The proposed Signal-to-Noise Ratio (SNR) of the transmitted signal in the wireless domain knowledge signifies the equipment onboard the UAV. The SNR-Aware RF Exploitation (SNARE) method solution improves the overall performance of conventional machine learning neural network models applied to imagery. This paper compares the performance metrics of different deep learning techniques including convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), and the effect of related hyper parameters such as size of sliding window, learning rate, and SNR range. Experimental RF data collected from multiple identical UAVs hovering in different ranges from the receiver node are employed in this study. Compared to the traditional models that do not consider the received RF signal related SNR information, our proposed SNARE improves UAV classification of the CNN, DNN, and RNN models from 84% to 96%, 91% to 96%, and 80% to 86%, respectively.