Deep Learning-Based Optimization of Phase-Modulated Waveforms for SBS Mitigation

To enhance the spectral uniformity and energy distribution of narrow-linewidth high-power fiber lasers and effectively suppress nonlinear effects such as stimulated Brillouin scattering, this paper proposes a deep learning-based method for phase-modulated waveform optimization. A Decoupled Echo State Network is used to construct a spectral prediction model, which captures the nonlinear mapping between phase-modulated waveforms and spectral responses. Based on the predicted spectrum, a differentiable target spectrum is constructed, and a composite cost function is designed with components including mean squared error, out-of-band energy penalty, and spectral smoothness. The predicted spectrum is iteratively optimized by the Adam adaptive gradient descent algorithm to approximate the target shape. Experimental results demonstrate that the proposed method achieves accurate spectral prediction and effective waveform refinement, significantly improves spectral smoothness and in-band energy concentration, and ensures rapid convergence of the cost function. These results demonstrate the effectiveness of the approach in suppressing stimulated Brillouin scattering in high-power laser systems.