Receiver-Agnostic Radio Frequency Fingerprint Identification for Zero-Trust Wireless Networks

Zero-trust has emerged as a promising security paradigm for next-generation networks (NGN). However, conventional cryptographic schemes struggle with continuous and dynamic authentication due to their coarse granularity and cumbersome processes. Radio frequency fingerprint identification (RFFI), as a prospective solution, enables physical-layer user-transparent identity authentication. Whereas, facing the dynamic topology and device mobility of NGN, such as Internet of Vehicles (IoV), Drone networks, etc., there exists a current deficiency in addressing the significant performance degradation across different receivers. In this paper, we propose a novel RFFI scheme for zero-trust continuous authentication in dynamic NGN environments, enabling unified high-performance cross-receiver identification. A two-stage unsupervised domain adaptation model is designed to extract receiver-independent transmitter-specific features. The receiver-side impact on RFFI, modeled as domain shift, is addressed through adversarial training for global alignment and local maximum mean discrepancy (LMMD)-based subdomain adaptation for eliminating subdomain confusion. Moreover, we further optimize RFFI through data augmentation to enhance robustness, multi-sample fusion inference to handle dynamic uncertainties, and an adaptive few-sample selection strategy for efficient fine-tuning. Extensive experiments on public datasets demonstrate the excellent performance of our proposed scheme in cross-receiver zero-trust wireless networks.