Machine learning for self-tuning mode-locked lasers with multiple transmission filters

Abstract

We develop an adaptive control and self-tuning procedure for mode-locked fiber laser systems using multiple transmission filters. Each transmission filter set consists of two quarter-wave plates, a passive polarizer, and a half-wave plate to generate nonlinear polarization rotation (NPR). The energy performance of a fiber laser can be significantly increased by incorporating multiple NPR filters. Critical for self-tuning is the ability to properly characterize the average cavity birefringence, and, although the existed self-tuning algorithms can successfully classify the birefringence of single filter configuration, they cannot achieve real-time recognition of the cavity birefringence for multifilter laser systems. To remedy this issue, we propose three birefringence classification algorithms based upon learned libraries of observed dynamic patterns, including a uniform, a hierarchical, and a dynamic selection procedure from such patterns. A maximum seeking algorithm is then constructed to determine the optimal (maximal) wave plate(s) and polarizer(s) settings. Thus, the adaptive control and self-tuning scheme is designed as a combination of maximum seeking and dynamic library selection algorithms. Numerical implementation shows that the proposed self-tuning scheme achieves stable, high-energy mode-locking while circumventing the multipulsing instability.