TESPOSDA-SEI: tensor embedding substructure preserving open set domain adaptation for specific emitter identification

Specific Emitter Identification (SEI) is an important method for secure authentication of devices in wireless networks. However, the lack of empirical identification models for many unknown devices further affects the speed and accuracy of their authentication. This work aims to propose an unsupervised open-set domain adaptation (UOSDA) based method to solve the open-set SEI problem with unknown devices appearing in the test set and few shots in the training set. The basic principle is to learn tensor embedding shared feature space and preserving inter-class substructure, which perform feature space mapping under the joint source and target domain led by mapping error minimize in the source domain. Then, in the shared space, the known and unknown targets are divided by the double clusters method of structure prediction and nearest class prototype. Specifically, this Tensor Embedding Substructure Preserving Open Set Domain Adaptation (TESPOSDA) consists of three parts, tensor substructure based invariant feature learning, unsupervised clustering based on known target intra-class structure prediction and neighbor prediction, UOSDA to refine the predicted labels. Finally, experiments are conducted on the real ADS-B dataset to demonstrate the effectiveness of TESPDA.