A Robust Specific Emitter Identification Method Based on Frequency-Domain Amplitude Mixing for Enhanced Domain Generalization

Specific emitter identification (SEI) leverages the inherent features of radio frequency signals to identify individual emitters.Recent progress in deep learning (DL) based SEI has significantly improved wireless device identification. However, real-world electromagnetic signals fluctuate due to varying channel conditions and time, resulting in poor performance for conventional models in unseen channels. To address this domain generalization challenge, we propose a robust SEI method based on frequency-domain augmentation, termed frequency-domain amplitude mixing (FAM). FAM applies a Fourier transform to the signal, interpolates and mixes the amplitude in the frequency domain, introduces phase perturbations, and then reconstructs the signal using an inverse Fourier transform. This augmentation generates diverse training samples, enabling the model to capture a broader range of features and improve generalization. We validate the effectiveness of FAM on a multi-domain dataset covering different time periods and receivers. Experimental results demonstrate that FAM achieves an average recognition accuracy of 80.67% on unseen domains, significantly outperforming existing domain generalization methods and highlighting its robustness in SEI tasks.