It looks like a faded old photograph: a reddish-black background peppered with pale orange specks. The dots, which are different shapes, sizes and shades of orange, seem arbitrary. But this image is at the forefront of South African radio astronomy.
Each speck – there are more than 1,300 – is a galaxy. They lie in a distant part of the Universe, representing less than 0.01% of the total sky. For comparison, this is roughly the same percentage as Costa Rica occupies on the world map. The “faded photo” was the first light image produced by the 16 active antennae of South Africa’s 64-dish MeerKAT, a radio telescope which will eventually form part of the Square Kilometre Array (SKA). When complete, the SKA will be the largest scientific instrument on Earth.
Each antenna, which is in the shape of a large satellite dish, collects radio waves from distant galaxies. These signals are collected together, and are automatically amplified and compared with those of every other antenna in the array. MeerKAT, for example, will focus all of its antennae on a particular patch of sky, and track that patch for hours at a time as the patch travels across the sky. It will then create an image of that piece of sky.
When MeerKAT is included in the SKA, scientists will be able to create even more detailed images of the Universe.
The SKA promises to investigate some of humanity’s most enigmatic scientific questions: how do galaxies evolve, how can we better detect gravitational waves, and what is dark matter? All of these would be great advances in our understanding of the universe and how it works, but none of the work can be done unless we can accurately interpret the signals received by the antennae.
But accurate interpretation is not that simple: When the Earth rotates, the sky above also rotates as it passes overhead. This rotation cannot be tracked accurately, which means that the telescope’s image is slightly distorted. The ionosphere, a charged layer of the upper atmosphere, adds to the complexity because it interferes with waves of certain frequencies and cuts off lower frequencies, much like a filter for radio waves. Compounding these problems, the physical antennae themselves add to the image errors: the way that the dishes move on their mounts – north/south or east/west – means that the “eyes” of the dishes zig-zag as they track a patch of sky.
These imperfections look like blurry halos around the specks on our image of the sky. Distant galaxies become smudges of light, rather than clearly defined points. Around particularly bright spots, the blur can even hide fainter galaxies.
These imperfections can be corrected for, but the process is complex. Scientists do have a way to correct the errors, but it only corrects the image perfectly at a specific point. It does this by looking at the halo at the point and removing it from the image. This process assumes that the blur affects the entire image, not just the part around that particular point, which means we could end up with an image that is even more inaccurate than the one we started with.
We could correct each pixel of the picture individually, but this will take far too long – fixing one galaxy currently takes hours. When it comes to the SKA, this would be impossible. The SKA will generate some of the largest data streams in the world – more than the current global daily internet traffic. Also, most of the sky does not contain galaxies, so we would be wasting a lot of time.
Another solution would be to divide the image into a series of smaller images – each with a smaller collection of pixels – and correct each smaller image separately.
This, too, has problems: breaking the image into a grid means that, like the point-at-a-time approach, a lot of time and computing power would be wasted because most of our smaller images do not have galaxies.
At the Rhodes University Centre for Radio Astronomy Techniques & Technologies (RATT), we aim to develop the best way to break up the image that reduces both the error and the number of sky pieces. Since the error depends on how bright a galaxy is and how close it is to other galaxies – because other galaxies are more likely to be hiding in a bright galaxy’s halo – we break it into pieces first. Each piece contains one of the brightest galaxies in the image. The more pieces, the less distortion we are likely to have, and the more accurate the image will be.
This problem is not unique to radio astronomy, although the data processing required for the likes of the SKA is not a recent invention. In 1854, the physician John Snow used this method to determine the source of the London cholera outbreak. In 1908, it was generalised by Russian mathematician Georgy Voronoi.
A Voronoi Diagram allows us to identify the brightest galaxies in an image, and then break the image up around them. It is a bit like like breaking a large chocolate-chip biscuit up so that each broken segment has a big chip of chocolate in it. The bigger the chocolate chip, the bigger the fragment of biscuit.
In terms of cosmology, this means that the brighter the galaxy, the larger its piece of the image can potentially be.
But what happens if two bright galaxies are close enough that their halos overlap? Then it would be better, both for the computer and the final image, if they were in the same piece. So the next step is then to find all cases like this and join these types of pieces back together.
Despite finding an easier way to navigate the image and data, time is still an issue: it would take many hours to do these calculations, especially when the image contains thousands of galaxies. That’s where gaming comes in.
Graphics processing units, or GPUs, were initially designed for gaming, but quickly became widely used for research and parallel programming because they are great at running a lot of small, similar tasks at the same time. So while GPUs have allowed people all over the world to kill zombies, explore worlds, and build civilisations in stunning detail, they will also help us to further people’s understanding of the Universe. For me, it allows us to create a Voronoi Diagram and find the best pieces of an image to merge together much faster.
The use of GPUs and more clever ways of problem solving are particularly necessary in the field of radio astronomy, especially with a fully operational MeerKAT and SKA on the horizon. In a system that will soon generate data on a scale researchers have never seen before, we need ways to cope with this data as quickly and efficiently as possible.