LAUSR

Deep learning provides a new scheme for specific emitter identification (SEI). However, most deep learning-based methods directly apply existing neural networks in other fields to SEI without considering its particularity. The identification-relevant information only accounts for a small proportion of the information carried by the received signal, and thus is easily interfered by the rest identification-irrelevant information during its mining process. This Letter proposes adversarial shared-private networks (ASPN) to address the most challenging case where all emitters are of the same type. By separating the normalised signal bispectrum amplitude and phase in the first quadrant into two orthogonal components, one that is shared among all emitters and the other that is private to each emitter, ASPN extracts more effective features and achieves more remarkable performances than most methods in the literature.