Wavelength and Polarization Diversity OTDR for Accurate Fiber Faults Identification in Data-Center Networks

Abstract

As data-center networks (DCNs) continue to scale in capacity and complexity, rapid and accurate fiber faults identification is becoming increasingly important for maintaining service continuity, minimizing downtime, and enabling automated maintenance. Optical time-domain reflectometry (OTDR) is widely used for fiber monitoring and fault localization. However, conventional OTDR relies on single-wavelength and polarization-insensitive measurements, and therefore struggles to distinguish different fault types with similar trace patterns, limiting precise fault diagnosis. This limitation is particularly critical for soft failures whose signatures can be easily confused with other impairments. To address this challenge, we propose a wavelength- and polarization-diversity OTDR (WPD-OTDR) framework for intelligent fiber fault identification in DCNs. By probing the wavelength- and polarization-dependent responses of fiber faults, the proposed system generates multi-dimensional fault signatures unavailable to conventional OTDRs. These features are then processed by a machine learning classifier for automatic fault recognition. Six representative fault types in data-center optical links are experimentally emulated and evaluated, including cases that are difficult to distinguish using traditional OTDRs. Experimental results show that conventional OTDR achieves an overall identification accuracy of only 72.30%, with per-class accuracies below 60% for certain fault types with highly-similar trace patterns, such as macrobending, microbending, and bad splicing. In contrast, the proposed WPD-OTDR clearly separates these fault classes and improves the overall identification accuracy to 99.23%. These results demonstrate that wavelength and polarization diversity can overcome conventional OTDR ambiguity, offering a promising solution for automated and reliable fiber fault management in future high-capacity DCNs.