RF Fingerprint Recognition Based on Adaptive Wavelet Threshold Denoising

Radio frequency fingerprint (RFF) has attracted the attention of researchers because of its robustness towards hostile attacks. Model-based and data-based methods have been proposed for RFF extraction and authentication. However, the performance of most existing methods degrades when the signal-to-noise ratio (SNR) is low. To solve this problem, a denoising algorithm is deployed before RFF extraction. Wavelet Threshold Denoising (WTD) is one of the most widely used algorithms, but the selection of the appropriate wavelet basis is a time-consuming task, and existing wavelet bases may not be suitable for the received signal. Hence, we propose a denoising module, called Adaptive Wavelet Threshold Denoising (AWTD) and combine it with a recognition network, ResNet18, to form an RFF recognition framework. Experimental results verify that the AWTD algorithm outperforms typical WTD. In addition, we connect the AWTD module with other recognition networks. The simulation results indicate that the AWTD module is able to be a plug-and-play preprocessing module for other recognition networks and can improve the recognition accuracy.