Experimental study of machine-learning-based detection and location of eavesdropping in end-to-end optical fiber communications

Abstract As the significant infrastructure of the communication network, the security of the optical fiber is critical. Optical fiber eavesdropping is a common attack in optical communications, which can easily lead to leakage of information. To deal with this problem, this paper proposes a Machine Learning (ML)-based scheme to detect and locate physical-layer eavesdropping. To improve the efficiency and accuracy of eavesdropping detection, both Optical Performance Monitoring (OPM) data and eye-diagrams are adopted as input data. Three different ML classifiers are designed and tested to realize eavesdropping detection, location and split ratio recognition, respectively. To demonstrate the feasibility of the proposed scheme, an experiment is conducted in end-to-end fiber transmission system with Coherent Optical Orthogonal Frequency Division Multiplexing (CO-OFDM). The eavesdropping is simulated and performed by placing optical coupler in different positions. In addition, different splitting ratios are considered, including 95/5, 90/10. Results indicate that the ML scheme achieves 100% and 92.76% accuracy in eavesdropping detection and location, respectively.