Computational Methods and Graphical Processing Units for Real-time Control of Tomographic Adaptive Optics on Extremely Large Telescopes.

Abstract

Ground based optical telescopes suffer from limited imaging resolution as a result of the effects of atmospheric turbulence on the incoming light. Adaptive optics technology has so far been very successful in correcting these effects, providing nearly diffraction limited images. Extremely Large Telescopes will require more complex Adaptive Optics configurations that introduce the need for new mathematical models and optimal solvers. In addition, the amount of data to be processed in real time is also greatly increased, making the use of conventional computational methods and hardware inefficient, which motivates the study of advanced computational algorithms, and implementations on parallel processors. Graphical Processing Units (GPUs) are massively parallel processors that have so far demonstrated a very high increase in speed compared to CPUs and other devices, and they have a high potential to meet the real-time restrictions of adaptive optics systems. This thesis focuses on the study and evaluation of existing proposed computational algorithms with respect to computational performance, and their implementation on GPUs. Two basic methods, one direct and one iterative are implemented and tested and the results presented provide an evaluation of the basic concept upon which other algorithms are based, and demonstrate the benefits of using GPUs for adaptive optics.