Functional Link ANN based Models for fast QoT estimation on Multiband Optical Networks

The adoption of multiband (MB) technology is critical to assure that the existing optical fiber infrastructure capacity can be fully exploited in a cost-effective way. In addition, the emergence of digital twin (DT) with smart and automated applications will enable more efficient use and planning of the network resources (e.g., optical transceivers, spectrum). For this last proposition to be achieved, fast and accurate lightpath quality-of-transmission (QoT) estimation is critical. To avoid using complex and time-consuming analytical methods (e.g., Split Step Fourier Method), we propose two novel machine-learning (ML) models for QoT prediction on MB C+L+S optical networks. The models are based on functional link artificial neural network (FLANN) structures which are known for their simplicity and ability to model linear and non-linear patterns, such as the ones introduced by impairments from fiber transmission. These models can be trained directly with in-phase and quadrature (IQ) constellation data or with the respective gaussian-based features. Using an optical link with variable span count, both FLANN models are benchmarked by deep neural network (DNN) based OCATA model and shown to be highly accurate in QoT estimation. In addition, the proposed models are able to significantly reduce the training time in comparison with the conventional split step methods and even DNN based OCATA, making them more adequate for applications with large data pools, such as a MB optical network.