Laboratory demonstration of single-camera PPPP wavefront sensing using neural networks

Abstract

Laser guide stars in astronomical adaptive optics systems have the focus anisoplanatism problem, especially for telescopes larger than 4 m in diameter. The Projected Pupil Plane Pattern (PPPP) offers an alternative solution by projecting a collimated laser beam across the telescope’s entire pupil. One significant challenge is dealing with gain-related issues, necessitating the use of two beam profiles obtained simultaneously from two different distances from the telescope pupil. In this work, we explore the integration of a convolutional neural network (CNN) with experimental data emulating PPPP. We investigate how CNNs can significantly simplify the PPPP design by enabling operation with a single beam profile. These results permit the development of the PPPP concept to use a single beam profile without distance-gain degeneracy. In this work, it is shown that a 10% residual error can be achieved for test data randomly chosen over the SNR range of 4 to 12.

Keywords

• Image processing
• instrumentation
• adaptive optics
• machine learning
• astronomy