Few-Shot Specific Emitter Identification Using Asymmetric Masked Auto-Encoder
Zhisheng Yao, Xue Fu, Lantu Guo, Yu Wang, Yun Lin, Shengnan Shi, Guan Gui
Abstract
Specific emitter identification (SEI) based on radio frequency fingerprint (RFF) characteristics can be used to identify different transmitters, and the deep learning (DL)-based SEI methods have achieved good performance with sufficient samples. However, these methods are difficult to identify emitters when the labeled training samples are limited. Hence, we propose a few-shot SEI (FS-SEI) using asymmetric masked auto-encoder (AMAE) to solve the few-shot problem. Specifically, we use the sufficiently unlabeled training samples which refer as the source domain to drive the training process of AMAE to obtain an RFF extractor with good feature extraction performance on the source domain, and then the pre-trained RFF extractor together with a classifier is fine-tuned using limited labeled training samples which refer as the target domain. Simulation results show that the proposed AMAE-based FS-SEI method achieves state-of-the-art identification performance compared to other supervised and unsupervised methods on the LoRa dataset with 30 categories and WiFi dataset with 16 categories. The codes can be downloaded from GitHub: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/YZS666/A-Method-for-Solving-the-FS-SEI-Problem</uri> .