Wavelength attack recognition based on machine learning optical spectrum analysis for the practical continuous-variable quantum key distribution system

Abstract

The imperfect devices in the practical continuous-variable quantum key distribution (CVQKD) system leave security loopholes that an eavesdropper may exploit. For example, the wavelength attacks can be implemented successfully by utilizing the wavelength-dependent properties of beam splitters (BSs). Eliminating the potential and existing security loopholes is essential for the practical CVQKD system. In this paper, an intelligent monitoring technology based on optical spectrum analysis is proposed for the practical CVQKD system. Through the machine learning-based optical spectrum analysis technique, an abnormal optical spectrum signal can be detected automatically by using the linear discriminant analysis support vector machine (LDA-SVM) algorithm, so as to realize attack detection and intelligent monitoring of the system. Simulation and experimental results show that the original spectral data and the abnormal spectral data after the attack can be accurately identified by the LDA-SVM algorithm, and the trained model can well resist the wavelength attacks.