DSP-based ODN characterization for coherent PONs with high-splitting loss

Coherent passive optical networks (PONs) not only allow for very high line rates but also offer benefits in terms of their operation. One example of this is using the deployed transceivers to accurately characterize the optical distribution network (ODN) to ensure network reliability and operational efficiency. This paper demonstrates a digital signal processing (DSP)-based framework for the ODN characterization in coherent PONs using Gaussian process regression (GPR), particularly targeting deployments with high-ratio power splitters that introduce loss up to 21 dB. This approach validates that the large loss introduced by a power splitter can be precisely localized by sweeping the assumed loss position and minimizing the error between measured post-DSP received signals and forward-propagated simulated signals using GPR, with a maximum median error of 0.17 km in the experimental demonstration. Building upon this framework, we extend the model to bidirectional ODN characterization to realize attenuation factor estimation and loss quantification by implementing loss localization for both downstream and upstream directions. Results indicate accurate bidirectional ODN characterization with the maximum loss localization and quantification errors of 0.21 km and 0.24 dB, respectively. Furthermore, the GPR model interpolation capacity and the trade-off between loss localization performance and computational complexity are discussed, validating that the GPR model maintains accurate and reliable loss localization with low complexity.