Optical constellations generated by Deep Learning models trained with datasets generated through simulation are compared to experimentally collected ones. The obtained high accuracy enables its application for optical time domain analysis in complex network scenarios.

We extend the OCATA time domain digital twin to the frequency domain. It is shown that deep learning-based models enable to predict the linear and nonlinear impairments affecting the optical constellations, whereas analytical models are suitable to estimate the filter penalties affecting the optical spectrums. Illustrative results show the benefits of the proposed approach for failure detection and identification.

Multi-band transmission is attracting the interest of research community because of its prospects to increase network infrastructures life. Indeed, current deployed systems mainly exploit the C band, thus deployed fibers can still offer a wide range of available spectrum. In this paper, quality of transmission (QoT) estimation is considered as fundamental control step for light path setup in order to check the proper feasibility of transmission along a computed light path. We demonstrate multi-band networking with channels on C and S bands in a testbed including a re-circulating loop. As a QoT estimator, the open-source GNPY module is extended in order to consider the typical impairments of a long-haul multi-band transmission and validated in this multi-band scenario for assisting connection setup with quality of transmission estimation. The measurements on the testbed demonstrated the correct operations of the multiband data plane testbed, particularly with reference to the GNPy estimations and thus validate the QoT estimator for C+S band transmission.

We exploit the intrinsic advantages of a time and frequency domain digital twin to detect degradations and to estimate their severity. Noticeable performance shown for filter failures confirms the usefulness of this approach.