Multi-Objective Parameter Self-Refinement for Digital Twin of C+L-Band Transmission System Under Measurement Constraints

Abstract

Digital twin (DT) technology has emerged as a cornerstone for autonomous optical networks; however, its predictive fidelity is critically dependent on the accuracy of the underlying physical parameters—particularly the fiber attenuation, component insertion losses (ILs), and amplifier gain profiles. These parameters are often impractical to measure directly in field-deployed systems. To address this challenge within practical constraints, we propose a two-stage, data-driven multi-parameter self-refinement framework for C+L-band transmission systems, capable of jointly refining fiber parameters and erbium-doped fiber amplifier (EDFA) gain spectra without requiring additional monitoring hardware, intrusive link reconfigurations, or service disruption. In the first stage, a neural network is constructed to learn the mapping between variations in fiber parameters (attenuation coefficient and ILs) and the corresponding changes in stimulated Raman scattering (SRS)-induced band powers measured by built-in photodetector at each EDFA, enabling fiber parameter refinement. The second stage extracts an equivalent optical multiplex section (OMS)-level EDFA gain spectrum via a bisection-based iterative process, utilizing the received power profile (RPP) difference monitored by optical channel monitor (OCM) at the reconfigurable optical add-drop multiplexer (ROADM) node. Comprehensive simulations on an OMS C+L-band system are conducted, and the results show that the proposed method reduces the generalized signal-to-noise ratio (GSNR) estimation error to a mean absolute error (MAE) of 0.10 dB and a maximum absolute error of 0.14 dB. The calibrated DT model demonstrates strong generalization capability, maintaining accuracy across 100 heterogeneous test links with diverse parameter sets, under varying launch power profiles, and in dynamic loading scenarios including partial channel cuts and full L-band shutdowns. Furthermore, experimental validation on a three-span C+L-band transmission system confirms the method's practical effectiveness, achieving GSNR MAE of 0.18 dB and a maximum absolute error of 0.23 dB after full refinement. The proposed framework established a viable pathway toward self-calibrating DT that maintain high fidelity throughout the network lifecycle. By achieving accurate parameter estimation using only existing measurement infrastructure, it proved potential applicability in current optical networks within measurement contraints.