Open-Set Specific Emitter Identification Under Cross-Receiver Conditions

In specific emitter identification (SEI) tasks, hardware variations across receivers cause significant distortion in radio frequency fingerprint (RFF) features, substantially degrading the performance of conventional closed-set classification methods. Furthermore, the presence of unknown emitters in open environments introduces additional challenges to the recognition system. To address these issues, this paper proposes an open-set SEI method for cross-receiver conditions that synergistically combines domain-adaptive feature learning with open-set classification to achieve robust recognition of known classes while effectively rejecting unknown emitters. The proposed method employs a ResNet18-1D-based feature extractor with domain adversarial training to mitigate receiver-induced domain shifts. Additionally, a cascading framework is constructed, integrating extreme value theory(EVT) and a softmax classifier. In this framework, the EVT is employed for unknown class detection through tail extremal analysis, while the softmax classifier performs fine-grained discrimination of known classes. Experimental results on real-world signal data collected from 12 USRP devices demonstrate that the proposed method achieves $88 \%$ accuracy for known class identification and $99.5 \%$ accuracy for unknown class detection. Currently, research on SEI in this scenario is nearly absent. Thus, the proposed method presents a new solution for SEI in complex electromagnetic environments.