SRS-Net: a universal framework for solving stimulated Raman scattering in nonlinear fiber-optic systems by physics-informed deep learning

Stimulated Raman scattering (SRS) is a crucial nonlinear phenomenon in fiber-optic systems, impacting optical performance and enabling devices like Raman amplifiers. Existing numerical methods for solving SRS are often time-consuming and scenario-specific. This paper proposes SRS-Net, a physics-informed neural network (PINN) framework that combines the efficiency of automatic differentiation and the representation power of neural networks with the regularization of SRS physical laws. SRS-Net provides universal solutions for forward, inverse, and combined SRS problems. Simulations and experiments, including high-fidelity modeling of a 10 THz C+L-band wavelength division multiplexing system, demonstrate SRS-Net's accuracy, speed (two orders of magnitude faster than classical methods), and versatility, suggesting its potential for broader application in nonlinear PDE-governed systems.

Yuchen Song, Min Zhang, Xiaotian Jiang, Fan Zhang, Cheng Ju, Shanguo Huang, Alan Pak Tao Lau, Danshi Wang