Fault monitoring in passive optical network through the integration of machine learning and fiber sensors

Abstract

As the deployment of fiber-based broadband networks continues to accelerate, the number of network facilities too is increasing exponentially. The network of optical fiber cables keeps growing as the number of passive optical network (PON) customers increases, eventually leading to unforeseen faults. Several solutions are offered for monitoring the optical link in PON with the optical time-domain reflectometer (OTDR) as the most common for point-to-point optical link characterization. However, the OTDR approach has been found to be inadequate for point to multipoint network fault characterizations due to numerous back-reflected signals converging at the power splitter that cannot be identified simultaneously by the OTDR detector. Several machine learning (ML) methods have recently been introduced for successful monitoring of optical communication links, but much of the ML technique depends on data from network transceivers to train ML algorithms to identify and detect faults. However, the use of data information for monitoring purposes can have an impact on the consistency of the services offered. In this article, we consider the deployment of the fiber Bragg grating sensor to acquire the monitoring data samples used to train the ML technique for effective link characterization. The proposed solution has the advantage of having a separate monitoring source that is independent of the data transmission signal and guarantees transparent transmission of information. The proposed ML-based technique shows up to 99% precision for the identification of fiber defect in the PON optical distribution network.