KerrNet: Fast and Accurate Estimation of Nonlinear Interference in Elastic Meshed Optical Networks Leveraging Artificial Intelligence

Due to constant demand for higher capacity in optical networks, the need for accurate and fast Quality of Transmission (QoT) estimators in optical meshed networks is increasing. Techniques based on machine learning and artificial intelligence have been proved to be efficient for developing such QoT tools in the past years. This work aims to extend our previously developed Machine learning (ML)-based framework called KerrNet, which rapidly and reliably estimates the variance of nonlinear interference (NLI) distortions of heterogeneous multi-span coherent WDM transmission systems. We extend the capabilities of our previously introduced QoT tool, called KerrNet [25-27] by using a modified version of the theoretical time-domain regular perturbation (TDRP) model to address elastic WDM scenarios with variable symbol-rate per WDM channel; we also add support for negative dispersions in the fibers, allowing for estimation of fully or partially dispersion-managed networks. Moreover, we introduce improved pre-processing and dataset generation methods, to simplify the design and construction of our ML-based solution, and present extensive numerical validations of the TDRP model.