Deep learning prediction of Stokes pulse evolution in ultrafast Raman fiber amplifiers

The nonlinear Schrödinger equation (NLSE) is extensively used to numerically study pulse evolution dynamics in ultrafast fiber lasers.Yet, the computational speed of the NLSE is relatively slow, restricting its applications in systems that rely on real-time computation for dynamic control and operation.In this work, we propose and demonstrate a deep learning approach for the prediction of Stokes pulses' evolution in a Raman fiber amplifier based on nonlinear optical gain modulation (NOGM).A four-layer fully connected neural network is developed to predict the spectral evolution of the first-order Stokes light in fiber amplifiers using different types of Raman gain fibers.The model achieves high prediction accuracy with normalized root mean square errors below 0.1, while providing up to 86 times faster computation compared to conventional NLSE methods.The network demonstrates reliable generalization capability for parameter combinations beyond the training dataset.