Fast and Accurate Estimation of Performance Degradation Due to PMD Based on Neural Network Method

We propose a fast and accurate method based on neural networks to estimate the performance degradation due to PMD. This method provides precise results and is 1000 times faster in computation than traditional method.

I. Introduction

With the increasing demand for global communication capacity, the establishment of ultra- high-speed and extensive optical communication networks has become a common goal for both academia and industry. Today, fiber optic communications are advancing toward larger bandwidth and higher symbol rate. Nokia Bell Labs have successfully achieved a 260-GBaud single-wavelength coherent transmission in 2022[1]. In traditional 50 GHz wavelength division multiplexing (WDM) narrowband systems, the value of the differential group delay (DGD) and the direction of polarization principal state (PSP), which used to characterize the polarization mode dispersion (PMD) are almost constant at different frequencies. It means that the DGD and PSP between different frequency components are strongly correlated. However, high symbol rate systems require larger bandwidths, necessitating consideration of all-order PMD effects rather than just first-order PMD. In such systems, the DGD and PSP between different frequency components are no longer correlatedl”, In addition, the temporal evolution of PMD also have an impact on large bandwidth systems, which is mainly due to rotation of the state of polarization (RSOP) in the channel. In extreme cases, the RSOP rate can exceed 5.1 Mrad/ s when lightning strikes near the channel [3]. Therefore, high symbol rate and fast time-varying channel have serious effects on the system, making it more difficult to compensate and evaluate performance degradation.