Currently, linear state space modeling is used for focal plane wavefront estimation and control of high-contrast imaging system. Although this framework has made great strides in the past decades, it fails to track the nonlinearities from the deformable mirrors and the light propagation, which to some extent influences the accuracy of the electric field estimation and the speed and robustness of the controller. In this paper, we propose the application of neural networks to identify and optimally control a high-contrast imaging system. Based on the E-M algorithm and reinforcement learning techniques, we develop a new nonlinear system identificaton method and a corresponding nonlinear neural network controller. Simulation and experimental results from Princetons High Contrast Imaging Lab (HCIL) are reported to demonstrate the utility of this algorithm.
|