High accuracy deep learning wavefront sensing under high-order turbulence

We explore an end-to-end wavefront sensing approach based on deep learning, which aims to deal with the high-order turbulence and the discontinuous aberration caused by optical system obstructions commonly encountered in real-world ground-based telescope observations. We have considered factors such as the entrance pupil wavefront containing high-order turbulence and discontinuous aberrations due to obstruction by the secondary mirror and spider, realistically simulating the observation conditions of ground-based telescopes. By comparing with the Marechal criterion (0.075λ), we validate the effectiveness of the proposed approach. Experimental results show that the deep learning wavefront sensing approach can correct the distorted wavefront affect by high-order turbulence to close to the diffraction limit. We also analyze the limitations of this approach, using the direct zonal phase output method, where the residual wavefront stems from the fitting error. Furthermore, we have explored the wavefront reconstruction accuracy of different noise intensities and the central obstruction ratios. Within a noise intensity range of 1% –1.9%, the root mean square error (RMSE) of the residual wavefront is less than Marechal criterion. In the range of central obstruction ratios from 0.0 to 0.3 commonly used in ground-based telescopes, the RMSE of the residual wavefront is greater than 0.039λ and less than 0.041λ. This research provides an efficient and valid wavefront sensing approach for high-resolution observation with ground-based telescopes.