Physics-informed neural networks for deterministic modeling of polarization division multiplexed fiber transmission systems

The coupled nonlinear Schrödinger equation (CNLSE) governs signal propagation in polarization division multiplexed (PDM) optical fiber systems, yet poses significant numerical challenges. This paper introduces physics-informed neural networks (PINNs) as a novel framework for deterministic modeling of PDM transmission. Through validation across single-pulse evolution, communication sequences, and full PDM systems, PINNs demonstrate deterministic accuracy (<i>R</i><i>M</i><i>S</i><i>E</i>=0.0044∼0.0129 and <i>s</i><i>p</i><i>e</i><i>c</i><i>t</i><i>r</i><i>a</i><i>l</i><i>e</i><i>r</i><i>r</i><i>o</i><i>r</i><i>s</i><4<i>%</i>) while overcoming traditional limitation. They eliminate the split-step Fourier method (SSFM)'s step-size dependencies and data-driven methods' statistical uncertainties. By preserving physical determinism through embedded PDE constraints, PINNs establish a new paradigm, to our knowledge, for reliable fiber-optic system modeling.