Compensation of carrier envelope phase slip using machine learning

Abstract

We demonstrate the effective compensation of carrier envelope phase (CEP) slip in a high-power femtosecond laser using a machine-learning (ML)-based control scheme. The compensation is achieved through fine dispersion tuning guided by a recurrent neural network (RNN) that predicts the temporal evolution of CEP slip, combined with a reinforcement-learning (RL) scheme that determines the optimal corrective actions. With this RNN+RL framework, the integrated CEP noise is reduced by more than a factor of two compared with a conventional proportional-integral-derivative controller. The proposed ML-based control methodology provides a versatile tool for stabilizing and optimizing various parameters in high-power laser systems.