Explicit Feature Disentanglement for Cross-Receiver Radio Frequency Fingerprint Identification

To address the issue of radio frequency fingerprint feature drift caused by hardware variations of receivers in non-cooperative scenarios for IoT devices, this paper proposes an Explicit Feature Disentanglement Network (EFDA-Net). Unlike traditional implicit domain adaptation methods, this framework, based on mutual information, employs a dual-stream orthogonal encoder and an adversarial suppression mechanism to explicitly decompose the feature space of received signals into transmitter-inherent fingerprints and receiver-induced nonlinear distortions. Experiments demonstrate that the proposed method can effectively strip away receiver-specific effects and significantly improve device authentication performance in cross-receiver scenarios.