RFFDDPM: A Robust Framework for Radio Frequency Fingerprint Extraction and Open-Set Authentication in LTE-V2X Networks

In the field of vehicular networks, ensuring se-cure communication through device authentication has become increasingly critical. Traditional Radio Frequency Fingerprint (RFF) extraction methods often struggle under dynamic and unpredictable channel conditions, leading to overfitting and poor generalization, especially in open-set scenarios where unknown devices and varying environments must be handled. Current techniques are insufficient in mitigating channel effects, limiting their practical applicability in LTE-V2X systems. This paper addresses the gap by developing a novel deep learning frame-work for robust RFF extraction and open-set authentication in LTE-V2X environments, leveraging advanced generative models. Here, we propose the RFFDDPM framework, which integrates Denoising Diffusion Probabilistic Models (DDPM) with a Chan-nel Removal Guidance Module (CRGM) to eliminate channel effects and enhance RFF extraction. The CRGM utilizes both time-domain and frequency-domain features to guide DDPM during denoising, ensuring that the RFF characteristics are pre-served while the channel effects are removed. Our experimental results, conducted on both indoor and outdoor datasets, show that RFFDDPM significantly outperforms existing methods in terms of accuracy and robustness, particularly in high-speed, non-line-of-sight (NLoS), and open-set scenarios.