Physical Layer Authentication via Conditional Variational Auto-Encoder for AIS

Abstract

Automatic identification system (AIS) with satellite components is exposed to increasing external attackers due to its high degree of openness. Radio frequency fingerprint identification (RFFI) offers a new perspective on AIS security solutions as a physical layer authentication scheme. However, existing RFFI methods are constrained by sensitivity to noise or interference, and these methods are cumbersome to tune when directly applied to satellite AIS (SAT-AIS). To this end, this paper proposes an RFFI method based on a conditional variational auto-encoder (CVAE). Specifically, the fractional lower-order cyclic spectrum as an extension of the conventional cyclic spectrum is used as a feature transformation to highlight the RFF features in the time-frequency domain. In addition, CVAE with multiple attention mechanisms is used to adaptively compress and extract discriminative RFF features to improve the RFFI accuracy. Experimental results show that the proposed method can yield 98.28% accuracy at $E_{b}/N_{0}$ of 10 dB, with high reliability and robustness under noise environment.