A Specific Emitter Identification Method Based on Self-Supervised Representation Learning

Specific emitter identification techniques excel in discerning between various devices through their unique radio frequency fingerprints (RFF), thereby enhancing the efficiency of communication among devices. However, in non-cooperative communication environments, the labeled emitter signal is often scarce or even unavailable. We design an effective self-supervised learning (Self-SL) approach based on contrastive learning for SEI to address the extreme scenario with no labeled samples. Specifically, we employ data augmentation in conjunction with deep neural networks featuring contrast loss to extract generic RF fingerprint features from unlabeled data, enabling the discrimination of various devices. Experimental results demonstrate that the acquired generic features can attain 91% recognition accuracy using just a simple linear classifier.