Enhanced Few-Shot Specific Emitter Identification via Phase Shift Prediction and Decoupling

Abstract

Specific Emitter Identification (SEI) is gaining prominence as a passive authentication technology for the physical layer of secure six-generation (6G) wireless communications. Leveraging Deep Learning (DL) for its robust data analysis and feature extraction capabilities, SEI effectively extracts Radio Frequency Fingerprints (RFFs) from received signals for identification. However, DL-based SEI faces a significant challenge due to the limited availability of labeled samples. To overcome this, we introduce an advanced Few-Shot SEI (FS-SEI) approach using Phase Shift Prediction and Decoupling (PSPD). We design a pretext task that allows an encoder to learn feature representations that include both phase shift relevant and irrelevant components from an unlabeled auxiliary dataset, processed by Short-Time Fourier Transform (STFT). In the subsequent task, we fine-tune the pretrained encoder with a classifier using a target dataset of few-shot samples. Our simulation results demonstrate that when the number of samples per category is 10 or more, the accuracy of our proposed SEI method exceeds 90%. For those interested in reproducing the results or exploring the methodologies further, the reproducible code and corresponding dataset can be downloaded from the following GitHub repository: https://github.com/IcedWatermelonJuice/FS-SEI/tree/main/PSPD.