Reshaping Distorted Light Rays with GPUs

by Tony Kontzer

Ever notice that when you take a picture of the sunset, the sun appears misshapen, as if it’s bending and contorting just to ruin your photo?

That phenomenon has nothing to do with the camera. Rather, it’s the result of atmospheric turbulence that distorts the trajectory of light rays. And it’s been a source of constant frustration for astronomers.

Fortunately, a team from the Université Paris Diderot and the Observatoire de Paris have figured out how to use GPUs to counter this effect and coax much higher resolution images out of huge telescopes like the European Extremely Large Telescope (E-ELT), which is currently under construction and due to become operational in 2024.

telescope The $4.3 million effort, dubbed the Green Flash project and backed by several European academic and corporate partners, is focused on prototyping real-time controllers that would provide millions of commands each second to the giant telescope’s deformable mirror, which is the size of an Olympic swimming pool. The work is expected to result in significant stabilization of the telescope’s image quality.

Damien Gratadour, an associate professor at Université Paris Diderot, was at the GPU Technology Conference last month to update attendees on the project’s progress in assessing the impact of GPUs on the performance of the controllers specifically, and on the telescope in general.

Gratadour’s team typically focuses on building optical systems for large telescopes.

“We literally reshape light beams so we can get images that are as sharp as possible out of these large telescopes,” said Gratadour.

Gratadour said that GPUs are revolutionizing adaptive optics, which are used to compensate for the fast-evolving aberrations that occur in optical systems, by enabling them to process large-scale images in real time from wavefront sensors that measure those aberrations.

But since adaptive optics aren’t yet available for a telescope the size of the E-ELT, Gratadour and his team are attempting to mimic the effect by placing GPUs on the controller, which is helping to reduce jitter and is allowing for a persistent kernel so data can be analyzed constantly.

And more use for GPUs is on the horizon: Gratadour said his team plans to build a scaled-down telescope prototype with as many as six high-definition cameras, and then analyze the performance gains GPUs provide in comparison to CPUs and field-programmable gate arrays.

That may not make your photo of the sunset look any better, but it will certainly give astronomers a better look.