Neural network adaptive optics for the Multiple Mirror Telescope

The Multiple Mirror Telescope (MMT) consists of six co-mounted 1.8 m telescopes from which the light is brought to a combined coherent focus. Atmospheric turbulence spoils the MMT diffraction-limited beam profile, which would otherwise have a central peak of 0.06 arcsecond full width at half maximum, at 2 μm wavelength. At this wavelength adaptive correction of the tilt and path difference of each telescope beam is sufficient to recover diffraction-limited angular resolution. Computer simulations have shown that these tilts and pistons can be derived by an artificial neural network, given only a simultaneous pair of in-focus and out-of-focus images of a reference star formed at the combined focus of all the array elements.¹ In this paper we describe an adaptive optics system, based on this approach, which we have developed for the MMT. We also report on some successful tests of neural network wavefront sensing on images acquired in the laboratory and at the telescope, and initial real-time tests of the adaptive system at the telescope, in particular a demonstration of the adaptive stabilization of the mean phase errors between two mirrors which resulted in stable fringes with 0.1 arcsecond resolution.