Siamese Network-based Open Set Identification of Communications Emitters with Comprehensive Features

Abstract

Radio frequency (RF) fingerprint-based identification schemes is of great significance in the domain of wireless devices, and we use them to complete device identification and authentication tasks. They extract the features caused by the inherent hardware circuits, which are very hard to forge. Recently, a large number of RF fingerprint identification methods based on deep learning have emerged. However, they make the close world assumption that the communications emitters appeared in the test data must have appeared in training. This assumption does not hold in open world classification. In this paper, we propose an open set identification method based on Siamese Network, dealing with the open set identification of communications emitters. Our method can extract more comprehensive signals features and has a better open set recognition effect. In addition, our method is more robust. In the case of increasing openness, it can still ensure sufficient discrimination between close set and open set classes.