AI-Driven Fault Detection in Coherent Optical Systems Using Deep Convolutional Networks
Abstract
The next generation networks require more bandwidth than ever before, which can only be attained through coherent optical communication systems. Nevertheless, these systems are also very prone to various impairments which include laser phase noise, polarization mode dispersion and nonlinear impairments, and proper fault detection is therefore an important activity in ensuring reliable transmission. Conventional methods of monitoring are based on optical performance monitoring(OPM) parameters or on a modelbased approach that may fail in non-linear or dynamical operating features. The proposed paper introduces using deep convolutional neural networks (CNNs) as an artificial intelligence (AI)-based system fault detector of coherent optical systems. The presented approach makes use of eye diagrams, constellation diagrams, and time-series as the inputs to CNNs which allows automatic extraction of features and classification of the fault states. In simulation experiments with different signal-to-noise ratios (SNR), modulation formats and fiber impairments, the proposed CNNbased fault detection is shown to have over 96 percent classification accuracy, which is better than support vector machines (SVMs) and shallow neural networks. The findings indicate that deep learning can be used to facilitate intelligent fault-tolerant coherent systems, which establishes the path to autonomous optical networks (AONs).