Fiber fault monitoring for passive optical network using a kernel-based support vector machine

Abstract

In order to avoid service level agreement violation in a passive optical network, effective failure detection and localization is of paramount importance for early fault identification. In this paper, we focus on a technique that can be used to detect and identify a failure in the physical layer using fiber Bragg grating (FBG) sensor integrated with machine learning (ML) technology. A kernel-based support vector machine trained model is purposely developed to detect and identify faulty optical link in PON. The technique relies on the retrieved optical reflected signal from the FBG sensors which are further elaborated by the ML algorithm using MATLAB. The dataset for the training and testing of the proposed model is generated in a simulated 1 times 4 Gigabit passive optical network, with a monitoring power of -4.16 dBm at a distance of 20 km. Optimal parameters of the support vector are selected with the help of cross-validation, thus leading to the optimized non-linear decision boundary. The proposed approach is tested on a number of datasets generated from the FBG sensors and demonstrates that the model achieves a 97 to 99% accuracy compared with the observed optical reflected spectra from the FBG sensors using an optical spectrum analyzer (OSA). The model which is not too complex, could avoid the use of costly OSA when embedded into the network controller in a software define network and minimizes the monitoring cost.