Classification of Optical Transmission Anomalies with Convolutional Neural Networks and 2D Histograms

In recent times, machine learning (ML) and deep learning (DL) algorithms have been integrated into existing control and management tools of optical networks but their implementation is often challenging due to high amount of data to be processed. For example, DL has been applied to anomaly detection to detect interferences in optical networks traffic. This paper tackles the more challenging problem of anomaly classification (rather than anomaly detection) applied to the constellation diagrams of the received optical signal. To address the large volume of data in constellation diagrams, this paper proposes the application of 2D histograms as a pre-processing step to generate constellation images, fed in input to a Convolutional Neural Network (CNN). The approach is evaluated on a open data set for anomaly detection in optical networks, demonstrating a high classification accuracy of different kinds of anomalies. This paper also provides an evaluation of the performance impact of the main hyper-parameters of the proposed anomaly classification algorithm.