Pyramid and Hartmann-Shack wavefront sensor with artificial neural network for adaptive optics

A three-sided pyramid prism is used as a wavefront sensor in non-modulating mode with an artificial neural network (NN) as a control unit in an adaptive optical setup. The NN is composed of a convolution network for feature extraction from the pyramid signal and a three-layer feed-forward back-propagation net. The network is trained with over 15,000 examples of signal-wavefront pairs. Presenting wavefronts of 6 mm pupil diameter and a RMS error of under 200 nm to the adaptive optical system with the trained network, it can correct the aberrations within three loops to a wavefront with 60 nm root mean square (RMS) error.