ML-Based Detection and Categorization of Complex Mechanical Vibrations via State of Polarization Analysis in Optical Networks

Paper in proceeding, 2025

Modern optical networks form the critical backbone of global communications, enabling high-speed data transmission for a wide range of applications. Despite their inherent advantages in bandwidth and scalability, these networks are not immune to physical-layer vulnerabilities. Mechanical disturbances, both accidental and intentional, can compromise service quality or serve as gateways for more severe cyber-physical attacks. Thus, there is a growing need for intelligent, real-time monitoring solutions capable of detecting and interpreting subtle anomalies in optical fiber infrastructures. This paper presents a Machine Learning (ML)-based State of Polarization (SOP) monitoring approach for the identification and classification of complex mechanical vibrations in optical fiber networks. We address the real-world challenge of mixed-frequency and overlapping vibration signatures, arising from benign activities, malicious attacks, or simultaneous events, by collecting 14 distinct polarization signatures under various physical scenarios. A diverse set of supervised ML classifiers is evaluated, with Histogram Gradient Boosting (HGB) achieving the highest performance at 88.33% accuracy.