Fast and Accurate Waveform Modeling of Long-Haul Multi-Channel Optical Fiber Transmission Using a Hybrid Model-Data Driven Scheme

• DOI: 10.1109/jlt.2022.3168698
• Publication Date: 2022-04-19T19:34:13Z
• Authors: Hang Yang; Zekun Niu; Haochen Zhao; Shilin Xiao; Weisheng Hu; Lilin Yi
• Keywords: Signal Processing (eess.SP); FOS: Computer and information sciences; Computer Science - Machine Learning; 0103 physical sciences; FOS: Electrical engineering, electronic engineering, information engineering; Electrical Engineering and Systems Science - Signal Processing; 01 natural sciences; Machine Learning (cs.LG)

The modeling of optical wave propagation in fiber is a task fast and accurate solving the nonlinear Schr\"odinger equation (NLSE), can enable system design, digital signal processing verification waveform calculation. Traditional full-time full-frequency information split-step Fourier method (SSFM), which has long been regarded as challenging long-haul wavelength division multiplexing (WDM) communication systems because it extremely time-consuming. Here we propose linear-nonlinear feature decoupling distributed (FDD) scheme to model WDM channel, where channel linear effects are modelled by NLSE-derived model-driven methods data-driven deep learning methods. Meanwhile, proposed only focuses on one-span distance fitting, then recursively transmits achieve required transmission distance. demonstrated have high accuracy, computing speeds, robust generalization abilities for different launch powers, modulation formats, numbers distances. total running time FDD 41-channel 1040-km 3 minutes versus more than 2 hours using SSFM each input condition, achieves 98% reduction time. Considering multi-round optimization adjusting parameters, complexity significant. results represent remarkable improvement open up novel perspectives solution NLSE-like partial differential equations physics problems.