Application of Machine Learning to Real-Time Removal of Atmospheric Blurring of Astronomical Images

Turbulence in the atmosphere severely degrades the imaging properties of large astronomical telescopes. At visible wavelengths, the resolving power of even the largest telescope is no greater than that of an instrument of 30-cm aperture. In principle, diffraction limited imaging can be recovered through the techniques of adaptive optics in which the shape of a deformable optical element in the light path is changed in real time to cancel the distortion. We have begun to explore the use of machine learning for predicting and correcting the optical distortion in real time. At the Multiple Mirror Telescope, we have demonstrated that an artificial neural network can derive the shape of the distorted optical wavefront on the basis of the far-field image irradiance. Moreover, simulations indicate that if a network is given additional information on the shape of wavefront distortions in the immediate past, it will also be capable of predicting the future behavior of the optical distortion, thus removing the degrading effect of the time lag inherent in the real-time servo loop.