A Digital Twin Assisted Link Status Visualization Framework for Failure Monitoring in Optical Networks: The ROADM Case

Reconfigurable optical add-drop multiplexers (ROADMs) based on wavelength selective switches (WSSs) serve as the cornerstone for dynamic signal routing in all-optical networks. With growing demands for higher capacity and flexibility, the deployment of large-scale, high-degree ROADM networks has significantly increased. However, this expansion increases the impact of failures in ROADMs, which may cause widespread service disruption across all fibers connected to a ROADM node. Therefore, it is crucial to monitor and visualize the status of all ROADMs in optical networks. Current approaches are limited by their reliance on scarce real-system failure data or the need for additional monitors. To address these challenges, in this paper, we propose a physics-based digital twin assisted link status visualization (DT-LSV) framework for monitoring ROADM status, specifically focusing on filter shift (FS) and filter tightening (FT) failures. Our approach is only based on the receiver digital signal processing (DSP) and configuration parameters, eliminating the need for additional monitors. In the absence of labeled failure data, DT-LSV utilizes normal data, and as labeled failure data accumulates, it gradually learns to further enhance performance. By integrating a parameter learning module into the theoretical model, we construct a physics-based digital twin (DT) model. Using this model and real data from receiver DSP, we formulate and solve an optimization problem for accurate ROADM status visualization. Our framework is validated through extensive experiments, achieving a 100% accuracy in localizing > 5 GHz FS (with a 0.66 dB signal-to-noise ratio (SNR) penalty) and > 8.5 GHz FT anomalies (with a 1.31 dB SNR penalty). The root mean square error (RMSE) of failure magnitude estimation is 0.60 GHz for FS and 0.41 GHz for FT. Additionally, the high efficiency of the framework is demonstrated, the importance of the parameter learning module is underscored, and the effectiveness of the continuous learning capability is highlighted. These results show that the proposed DT-LSV facilitates dynamic and continuous monitoring of ROADM status through the network lifecycle.