Focal plane wavefront correction for dark hole creation with a neural network on PICTURE-D

In order to directly image exoplanets, a dark hole is created in the coronagraph science camera image, which is a region where light speckles are minimized to improve the contrast ratio. Creation of a dark hole normally consists of two independent steps: sensing and correction. During sensing, the electric field (EF) in the focal plane is determined, then during correction, a command is applied to the deformable mirror(s) to minimize the EF. The PICTURE-D balloon mission makes use of the linear EF Conjugation (EFC) algorithm for EF correction. However, this algorithm has two limitations: first, the corrections require a nonlinear model, whereas EFC relies on an inverted matrix, and second, EFC may not properly describe the optical system. Therefore, a neural network (NN) is developed for EF correction on PICTURE-D and will target the issue of nonlinearity. It is shown, in simulation, that a NN can obtain a higher contrast ratio after one iteration when compared with EFC.