Quality of Transmission prediction using Machine Learning models in future optical networks

Abstract

With the increase in global traffic, the pressure on the operators to provide services with optimal capacity is also increasing. The information about the quality of transmission is required to make routing decisions such as deciding whether to establish the lightpath or not. The traditional approaches used to estimate the QoT of new lightpaths are coupled with high margins for accurate estimation but this results in under-utilization of network elements. Machine learning models can be used to predict the QoT accurately with low margins resulting in better utilization of network elements thus increasing the efficiency of network. Models are trained on the data obtained from previous knowledge of already deployed network and is then used to predict the QoT of lightpaths. We apply Machine Learning models on synthetic data of real network topologies. We expect to predict the QoT (Generalized Signal to Noise Ratio (GSNR)) of lightpaths with minimum error