A phase-error prediction method for coherent beam combining via convolutional neural network

Abstract Coherent beam combination (CBC) with Actively Phase-Controlling is a key method to many applications. The core challenge of achieving a Diffraction-Limited output laser of CBC is to correct and lock the phase errors quickly between the sub-beams. A phase prediction method based on the convolutional neural network with a MSE-NPCD loss function is proposed to solve this problem. By using the proposed method, the phase errors can be predicted by a pair of far-field images in one-step. We also demonstrate the robustness and scalability by adding additional disturbance and expanding the scale of the sub-beams. The result shows that our method can achieve high accuracy and high speed combining effect in a tiled aperture CBC system.