Unifying spectrum measurements for digital twin optical networks via high-dimensional feature calibration and...

The evolution of optical networks toward higher data rates, disaggregation, and software-hardware decoupling has driven operators to adopt multi-vendor strategies, which introduce significant challenges in observability, data consistency, and interoperability. Digital twin optical networks (DTONs) rely heavily on accurate and consistent monitoring data to achieve high-fidelity modeling and autonomous operation. However, in multi-vendor environments, various optical spectrum analyzers (OSAs) produce inconsistent spectral data due to differences in principles, physical structures, and manufacturing processes. This inconsistency degrades the construction of spectral-domain digital twins for optical networks and compromises the performance of optical spectrum feature analysis. To address this issue, we firstly establish a mathematical model to theoretically analyze the impact of OSAs on spectral features. Then, we design a high-dimensional feature calibration (HDFC) tool based on the multi-layer perceptron (MLP), and propose an adaptive perceptual information calibration (APIC) method based on model-agnostic meta-learning (MAML) algorithm, enabling adaptive alignment of inconsistent spectral data to a unified reference. Simulations and experimental validations demonstrate that proposed HDFC and APIC effectively mitigate data discrepancies and enhance feature extraction accuracy. Furthermore, we define use cases and message exchange workflows to integrate APIC into multi-vendor DTON operations. This work provides a critical step toward consistent perceptual data acquisition and interoperable management in next-generation optical networks.