Machine-Learning-Assisted Optical Fiber Communication System

Abstract

With the development of software-defined networking and coherent transmission, to name only a couple of emerging technical and technological areas, optical networks have rapidly expanded during the past few years. To handle with the enormous increment, several sections of optical transmission networks have been addressed via machine learning. Techniques such as support vector machine and KNN algorithms are widely used in fiber-induced nonlinear mitigation, which can cause enormous financial loses if the problem is not solved properly. Optical performance monitoring is another essential area in optical networks, which often adopts multitasking, while failure management, where anomaly detection takes place, rely on both supervised and unsupervised learning methods. The overview includes a brief synopsis of four types of learning methods, including supervised learning, unsupervised learning, semi-supervised learning and reinforcement, as well as the most recent advancements in methodologies used in optical fibre communication. At the end of the analysis, it is demonstrated that ML algorithms are selected based on the challenges present and we have to consider multiple factors when choosing a suitable algorithm. The combination of the two fields brings prosperity to each other.