Quantum mechanics

Beam (structure)

Boundary value problem

Optical fiber

Applied mathematics

Helmholtz equation

Telecommunications

Computer science

Physics

Nonlinear system

Paraxial approximation

Mathematics

Optics

Artificial intelligence

Time domain

Partial differential equation

Artificial neural network

Computer vision

IEEE Communications Magazine

Due to the capability of the physics-informed neural network (PINN) to solve complex partial differential equations automatically, it has revolutionized the field of scientific computing. This article studies the applicability of PINN in optical fiber communication and presents five potential solutions for time-domain, frequency-domain, and spatial-domain modeling. By solving the nonlinear Schrödinger equation (NLSE), both forward and backward optical waveform propagation are modeled by PINN; by solving the inverse problem of NLSE, the physical parameters of fiber are identified with limited data, including attenuation, dispersion, and nonlinear coefficient. By solving the stimulated Raman scattering evolution equation, the wideband spectrum evolution is simulated for C+L-band systems. By solving the paraxial Helmholtz equation, the electric field distribution in a fiber is modeled, and five lowest-order fiber modes are studied. In comparison to conventional numerical methods, PINNs achieve the approximated accuracy while having lower computation complexity and less computation time over the same fiber length for a given initial condition. Finally, the challenges faced by PINN are pointed out, and an outlook on future research of the physics-informed approach is concluded.

Applications of Physics-Informed Neural Network for Optical Fiber Communications