A novel solution for civilian UAV identification based on RFF and machine learning

The proliferation of civilian unmanned aerial vehicles (UAVs) in recent years has raised significant concerns regarding security threats. Radio frequency fingerprint (RFF) classification has been widely used for detecting and identifying unknown UAVs. This paper proposes an efficient UAV identification scheme that utilizes the discrete wavelet transform coefficients of differential signals (D-DWT) and machine learning. In the detection phase, the scheme extracts the preamble information of each signal frame based on energy variations and correlation detection. In the feature extraction phase, the impact of carrier frequency offset (CFO) on identification accuracy is analyzed, and the differential signals are transformed into the wavelet domain to analyze the anti-interference performance of the DWT coefficients. Experimental results demonstrate that the proposed method effectively improves RFF recognition accuracies in low SNR scenarios and exhibits excellent resistance to doppler frequency offset. Among various machine learning classifiers, the scheme achieves device classification accuracies of 97.91% and 98.38% in 0dB and 5dB scenarios, respectively.