Abstract

Relative phase control is the key component of power scaling via coherent beam combining, but it can also be used for beam steering, compensation of atmospheric turbulence, and mode control. In these contexts, it becomes much more demanding to derive the necessary feedback from the observation and potential complicated models are necessary. We investigated the applicability of deep reinforcement learning for phase control two channel CBC system, therefore, trained a neural-network based reinforcement learning algorithm to realize phase control.

© 2018 The Japan Society of Applied Physics