Synergetical Use of Analytical Models and Machine-Learning for Data Transport Abstraction in Open Optical Networks

Abstract

The key-operation to enabling an effective data transport abstraction in open optical line systems (OLS) is the capability to predict the quality of transmission (QoT), that is given by the generalized signal-to-noise ratio (GSNR), including both the effects of the ASE noise and the nonlinear interference (NLI) accumulation. Among the two impairing effects, the estimation of the ASE noise is the most challenging task, because of the spectrally resolved working point of the erbium-doped fiber amplifiers (EDFA) depending on the spectral load, given the overall gain. While, the computation of the NLI is well addressed by mathematical models based on the knowledge of parameters and spectral load of fiber spans. So, the NLI prediction is mainly impaired by the uncertainties on insertion losses an spectral tilting. An accurate and spectrally resolved GSNR estimation enables to optimize the power control and to reliably and automatically deploy lightpaths with minimum margin, consequently maximizing the transmission capacity. We address the potentialities of machine-learning (ML) methods combined with analytic models for the NLI computation to improve the accuracy in the QoT estimation. We also analyze an experimental data-set showing the main uncertainties and addressing the use of ML to predict their effect on the QoT estimation.