ADS-B Signal Fingerprint Recognition Based on Adaptive Short-Time Fourier Transform and Residual Networks

In order to improve the security defense capability of ADS-B system, aiming at the problem of individual identification of ADS-B signal emitters, this paper proposes a signal fingerprint recognition algorithm based on adaptive shorttime Fourier transform (STFT) and residual networks. The algorithm converts one-dimensional signals into time-frequency images with superior time-frequency concentration by adaptively selecting window length and step parameters for STFT, and improves the ability to characterize the fingerprint features of radiation sources in the time-frequency joint domain. By incorporating self-attention mechanism and transfer learning into the deep residual network, the network is enabled to automatically focus on highly discriminative fingerprint features in timefrequency representations, and improve the ability to extract radiation source fingerprint features. Comparative studies with methods employing different time-frequency image parameters and diverse deep network architectures show that our algorithm can more effectively characterize and extract signal fingerprint features from time-frequency representations. Experimental results demonstrate that the proposed algorithm achieves an identification accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 7. 2 6 \%}$</tex> for real-world collected data from 187 targets.