Predicting nonlinear multi-pulse propagation in optical fibers via a lightweight convolutional neural network

Area of Science:

• Nonlinear optics
• Computational physics

Background:

• Ultrashort pulse propagation in optical fibers is crucial for many applications.
• Traditional numerical methods are computationally intensive.
• Need for rapid modeling methods for complex pulse evolution.

Purpose of the Study:

• To develop a lightweight convolutional neural network (CNN) for characterizing nonlinear multi-pulse propagation.
• To enable both forward and inverse mapping of multi-pulse evolution.
• To reduce the computational burden of simulating ultrashort pulses in highly nonlinear fiber.

Main Methods:

• Designed a lightweight CNN architecture.
• Utilized initial multi-pulse temporal profiles for forward mapping.
• Employed propagated multi-pulse profiles for inverse mapping.
• Simulated complex random multi-pulse evolution using Gaussian pulses in 4-level pulse amplitude modulation.

Main Results:

• Achieved accurate forward and inverse mapping of multi-pulse propagation.
• Demonstrated excellent generalization and prediction performance on unlearned test sets.
• Reported maximum absolute errors of 0.026 for forward mapping and 0.01 for inverse mapping.
• Validated the CNN's ability to handle group velocity dispersion and self-phase modulation.

Conclusions:

• The lightweight CNN effectively models nonlinear multi-pulse propagation in optical fibers.
• The proposed method offers a computationally efficient alternative to traditional numerical solutions.
• This approach has significant potential for predicting the evolution of arbitrary complex multi-pulses.