A Generalizable Model-and-Data Driven Approach for Open-Set RFF Authentication

Radio-frequency fingerprints (RFFs) are promising solutions for realizing low-cost physical layer authentication. Machine learning-based methods have been proposed RFF extraction and discrimination. However, most existing designed the closed-set scenario where set of devices is remains unchanged. These can not be generalized to discrimination unknown devices. To enable from both known devices, we propose a new end-to-end deep learning framework extracting RFFs raw received signals. The comprises novel preprocessing module, called neural synchronization (NS), which incorporates data-driven with signal processing priors as an inductive bias communication-model based processing. Compared traditional carrier techniques, static, this module estimates offsets by two learnable networks jointly trained extractor. Additionally, hypersphere representation further improve RFF. Theoretical analysis shows that such data-and-model better optimize mutual information between device identity RFF, naturally leads performance. Experimental results verify significantly outperforms purely DNN-design handcrafted in terms network generalizability.