Radio-Frequency-Based Unmanned-Aerial-Vehicle Identification Using Cyclic-Paw-Print
Abstract
A novel robust unmanned aerial vehicle (UAV) identification approach using our recently proposed cyclic-paw-print (CPP) features extracted from the radio-frequency (RF) signals is presented in this paper for air-traffic surveillance and control. In this work, a radio-frequency (RF) signal such as a UAV telemetry, track, and command (TT&C) signal and a digital data-transmission (DDT) signal emitted by a UAV is first received and segmented according to its short-time energy-to-spectral-entropy ratio (ST-ESER). Next, cyclic-spectrum analysis is applied to each selected RF signal segment to form the corresponding polyspectrum, which is further utilized to establish the cyclic-paw-print (CPP) matrices. Finally, the lightweight convolutional neural network, namely ResNet-18, which takes the produced CPPs as the input features, is trained to recognize the aforementioned RF signals so the corresponding UAVs can be identified. Monte Carlo simulation results demonstrate the superior effectiveness of our proposed new UAV identification approach in comparison with other existing deep-learning methods.
