Proactive Fiber Break Detection based on Machine Learning applied on State of Polarization represented as Quaternion

We propose to enhance a real-time highspeed optical communication system prototype based on coherent detection technologies and coupling it with machine learning to monitor mechanical events on an optical fiber, hence to proactively detect fiber breaks. The method relies on State of Polarization (SOP) monitoring via digital signal processing in a coherent receiver paired with machine learning for the event classification and enables a proactive detection of a fiber cut. We demonstrate the relevancy of representing the SOP time evolution as quaternion series to provide low complexity interpretable decision mechanism embeddable in network element while ensuring an event classification accuracy of more than 98%.