Laser Tomography in High-order Adaptive Optics: Predictive Control and Reconstructor Optimisation

Abstract

The presence of atmospheric turbulence degrades the image quality of ground- based telescopes. Adaptive Optics (AO) is a powerful method to partially correct the wavefront distortion caused by atmospheric turbulence in real-time. Owing to the technical simplicity and light computational load, existing AO systems typically use non-predictive linear controllers, and as such, temporal error remains one of the most critical error sources. The MCAO Assisted Visible Imager and Spectrograph (MAVIS) is a general-purpose MCAO instrument that is currently under development for the 8-metre Very Large Telescope (VLT) in Chile. MAVIS aims at delivering a sharp, uniform image quality over a wide field of view, demanding a tight error budget. MAVIS will be equipped with a powerful real-time computer hosting multiple GPUs. This facilitates the research of an advanced MCAO supervisory scheme to capitalise on the improved computational resources. This thesis presents the development of the optimised Learn and Apply (L&A) method - a model-based, predictive MCAO supervisory solution. The optimised L&A is based on the statistical properties of atmospheric turbulence, and operates directly in the wavefront sensor measurement space. We propose a generalised model for integrated predictive tomography, whereby the high-order and low-order measurements are combined together using the system geometry and measured statistics. The calculations involved are highly parallelisable, e.g., on multi-GPU platforms, significantly reducing the time required for the algorithm. Predictive tomography requires real-time, or at least pseudo-real-time information on the wind profile, which can be achieved with the Learn algorithm. It takes as input the wavefront measurement buffer and estimates the relevant atmospheric parameters in an iterative least- squares fitting process. We propose and implement a second-order, stochastic optimiser that uses only a subset of the input data at each iteration, and which converges much faster than existing algorithms. Scalability test results on multi-GPU platforms validate the performance of our proposed algorithm and demonstrate its feasibility for systems such as MAVIS. The classical approach for MCAO system control is to utilise split tomography, where the high- order and low-order loops are controlled independently. The generalised model proposed in this thesis allows simplifying MCAO control to a single loop, the so-called integrated controller. Our model implicitly handles the issue of tilt anisoplanatism. Additionally, we propose the slope merging method - a lightweight, heuristic alternative for correcting tilt anisoplanatism. We present numerical simulation results to compare the performance of these methods. Based on the MAVIS MCAO configuration, we implement the optimised L&A scheme in a closed supervisory loop. Realistic error sources and real-time evolution of the atmospheric turbulence statistics are taken into consideration. As shown by our results, the application of predictive control substantially improves the overall MCAO performance without any additional hardware, and only introduces marginal increase in real-time computational load. Show more