A star-test wavefront sensor using neural network analysis

Abstract

We describe a new, simple wavefront sensing method that uses a single measurement of a defocused star and a neural network to determine low-order wavefront components. The neural net is trained on computed diffracted star image data at 640 nm to output annular Zernike terms for an obscured circular aperture over a discrete range of all values. In the context of an actual star, the neural-net also provides the Fried's parameter as an estimation of atmospheric turbulence. It is shown that the neural-net can produce a robust, high accuracy solution of the wavefront based on a single measurement. The method can also be used to simultaneously determine both on-axis and field-dependent wavefront performance from a single measurement of stars throughout the field. The prototype system can run at a rate of about 1 Hz with Python interpreted code, but higher speeds, up to video rates, are possible with compilation, proper hardware and optimization. This technique is particularly useful for low-order active-optics control and for optical alignment. A key advantage of this new method is that it only requires a single camera making it a simple cost-effective solution that can take advantage of an existing camera that may already be in an optical system. Results for this method are compared to high-precision interferometric data taken with a 4D Technology, PhaseCam interferometer and with an Innovations Foresight StarWave Shack Hartmann sensor from ALCOR SYSTEM under well-controlled conditions to validate performance. We also look at how the system has been implemented to use starlight for aligning multiple mirror telescopes in the presence of atmospheric seeing.