Building a digital twin for large-scale and dynamic C+L-band optical networks

Abstract

Bridging the gap between the real and virtual worlds, a digital twin (DT) leverages data, models, and algorithms for comprehensive connectivity. The research on DTs in optical networks has increased in recent years; however, optical networks are evolving toward wideband capabilities, highly dynamic states, and ever-increasing scales, posing huge challenges, including high complexity, extensive computational duration, and limited accuracy for DT modeling. In this study, the DT models are developed based on the Gaussian noise (GN) model and a deep neural network (DNN) to perform efficient and accurate quality of transmission estimations in large-scale C+L-band optical networks, facilitating effective management and control in the digital platform. The DNN-based model obtained the estimated generalized signal-to-noise absolute errors within 0.2 dB in large-scale network simulation, specifically a 77-node network topology. Additionally, compared to the GN-based model, the testing time by using the DNN-based model has been significantly reduced from tens of minutes to 110 ms. Moreover, based on the DT models, multiple potential application scenarios are studied to ensure high-reliability operation and high-efficiency management, including optimization and control of physical layer devices, real-time responses to deterioration alarms and link faults, and network rerouting and resource reallocation. The constructed DT framework integrates practical analysis and deduction functions, with fast operation and accurate calculation to gradually promote the efficient design of optical networks.