Efficient feature extraction of radio-frequency fingerprint using continuous wavelet transform

In securing wireless communication, radio-frequency (RF) fingerprints, rooted in physical-layer security, are seriously affected by various types of noise. As a result, effective RF fingerprint extraction and identification for device authentication present a significant challenge. To address this, we propose a comprehensive and robust approach using continuous wavelet transform (CWT) for RF feature extraction, along with U-Net for RFF identification. Initially, the received signal undergoes CWT into a stable time-frequency representation, while the U-Net algorithm is employed to denoise in RFF feature extraction and identification. The experiment results show, remarkable accuracies of 95.4% and 89.5% are achieved (SNR@ 10dB and 5dB), respectively, for 11 Wi-Fi devices with the same model. This underscores the potential of the proposed algorithms to enhance wireless communication security, providing a valuable contribution to RFF identification.