It’s a Bird, It’s a Plane, It’s "That" UAV: RF Fingerprinting During Flight
Jerry Gu, Nasim Soltani, M. Yousof Naderi, Kaushik Roy Chowdhury · 2021 55th Asilomar Conference on Signals, Systems, and Computers · 2021
UAVs are being rapidly deployed in many surveillance-related and monitoring applications worldwide. Thus, identifying a known UAV in a larger pool of devices is important. This task is often complicated in dense urban environments or low-level flight deployment cases where traditional radar-based detection becomes difficult. In this work, we describe deep convolutional neural network architectures and pre-processing steps suitable for capturing RF signals from in-flight UAVs for detection of the type of the UAV. Our objective is to leverage subtle discriminative features that may be embedded in the signal transmissions through RF fingerprinting for identifying (i) if the test signal comes from an unseen UAV (with respect to the training set), and (ii) if it is from a seen UAV, then we identify the label used during training time. Unlike static data collections, the mobile scenario is more challenging due to the rapid fluctuations in the wireless channel and Doppler effects which impair successful classification. We study the efficacy of our approach under different distances, flight/mobility patterns, interference conditions to emulate real-world situations with high fidelity. Our experimental dataset contains signals from seven different make/models, collected within an RF anechoic chamber.