PSFAN: Prototype-Based Source-Free Alignment Network for Cross-Receiver Specific Emitter Identification

Traditional specific emitter identification (SEI) often suffers from performance degradation in cross-receiver scenarios due to domain shifts caused by receiver variations. To this end, we propose a source-free domain adaptation framework, termed prototype-based source-free alignment network (PSFAN), for cross-receiver SEI. Specifically, our method leverages prototypes learned from a pre-trained source model as category feature representations to guide the alignment of the target domain by minimizing feature discrepancies, quantified using multi-kernel maximum mean discrepancy (MK-MMD). Furthermore, we enforce category consistency by constraining the target classifier with prototype-distance vectors to enhance the discriminative ability of the target model. The target model is adapted through this alignment process and subsequently deployed to recognize signals from a new receiver. Experimental results demonstrate that PSFAN significantly improves SEI performance in cross-receiver scenarios.