Investigation of dissipative solitons in an Er-doped fiber laser through machine-learning online optimization based on the Gaussian process

Abstract

We have applied a machine-learning online optimization method based on the Gaussian process to the numerical modeling of dissipative solitons in an Er-doped and nonlinear polarization rotation mode-locked fiber laser. Three types of dissipative solitons are achieved both experimentally and numerically in the negative, near-zero, and positive net cavity dispersion regions. Corresponding optimum cavity parameters in simulation can be determined quickly and precisely via optimization. The optimization goal is the high similarity between the experimental results and the simulation results, which is calculated by the sum of Fréchet distance of the normalized spectral waveforms and autocorrelation traces. In numerical analyses, the characteristics of the output pulse in different dispersion conditions, the pulse dynamics inside the laser cavity, and the initial process of mode-locking are also investigated. Our results demonstrate the effectiveness and universality of machine-learning online optimization based on the Gaussian process method in the optimization process of fiber laser modeling, which can further provide more insights and extend large-scale potential applications of machine-learning algorithms in fiber lasers.