Open Set RF Fingerprint Identification for Wireless Communication Devices

Abstract

Radio frequency fingerprint identification (RFFI) is a task to determine the source of a signal by extracting the radio frequency fingerprint of the emitter. It provides physical-layer non-key authentication technology for wireless communication devices, ensuring the security of wireless communications. However, traditional methods of RFFI based on deep learning cannot reject illegal and unknown emitters. In this letter, a new open set identification method called open set support vector data description (OpenSVDD) is proposed for RFFI. Adversarial reciprocal point loss is firstly used to optimize the feature distribution of the known samples to minimize the possibility of feature overlap of different known classes. Furthermore, radial basis function (RBF) kernel is used to fit the most compact classification boundaries, which achieve a reliable open set RFFI. Experimental results demonstrate that our method exhibits strong open set identification performance. Even when the openness exceeds 31%, the proposed method maintains an accuracy of 90%.