A Transformer Based Approach for Open Set Specific Emitter Identification

Abstract

Specific emitter identification (SEI) can verify the identity of emitters and plays an important role in a wide range of military and civilian fields. Most recently, there has been great interests in applying deep learning techniques for SEI. However, the vast majority of existing work limits SEI to closed-set classification, where deep networks only perform classification in a limited set of known emitters. These algorithms are not capable of handling transmitters that appear outside the closed set, compromising the security of the communication system once such emitters within the communication range. In this paper, we propose a novel deep learning method for SEI based on the open-set recognition. By using a combination of an improved Transformer and the modified intra-class splitting (ICS) method, our proposed method can identify unknown class of signals while maintaining a high accuracy of known classes. Numerical experiments demonstrate its ability to reject signals from unknown emitters with high accuracy on USRP-generated datasets, and its classification performance exceeds that of various existing open-set classification algorithms.