Green Physical Layer Authentication Based on Few-Shot RFF via Asymmetric Augmentation and Frequency-Assisted Learning

Radio Frequency Fingerprint (RFF) identification has emerged as a promising technique for enhancing security in green communication. However, it typically relies on large labeled datasets, which presents an impractical requirement in resource constrained scenarios. To address this challenge, we propose Asymmetric Augmentation with Frequency-Assisted Learning (AA-FAL), a novel self-supervised framework for few-shot RFF identification. Our approach integrates a family of asymmetric time-frequency augmentation operators, informed by hardware-specific impairments, to generate realistic, identity-preserving signal variations. Additionally, we employ a dual-stream architecture governed by a neighborhood-consistency objective, which aligns time and frequency domain representations to extract complementary and discriminative features. Experimental evaluations on real-world Wi-Fi, LTE and ADS-B datasets demonstrate that AA-FAL achieves state-of-the-art few-shot performance in the 10-shot setting, with accuracies of 98.13%, 92.25% and 74.73%, respectively. Our code is available at https://github.com/SONGzyyyyyy/AA-FAL.