An Artificial Neural Network for Phase Recovery from HST Stellar Images

Abstract

During the last two years, we have developed and refined a novel approach to estimate phase distortion across an optical beam directly from focused images of starlight. The method, applicable to real-time atmospheric compensation of large telescopes using guide stars, relies on a nonlinear neural network processor to determine the phase from two distorted point spread functions, one at the exact focus of the telescope and one intentionally out of focus. Real-time phase retrieval is possible because the network is trained using simulated data to recognize and predict the near-field phase from the characteristic shapes and features of far-field images.