CLEAN on the GPU

Observational data in radio astronomy requires significant pre-processing before it can be interpreted scientifically. The most commonly-used procedures in this field are gridding, transforming the signal recorded in frequency space by the radio antenna array to an image in the plane of the sky, and CLEANing, deconvolving the image with the point spread function of the instrument. We have implemented standard algorithms for both these tasks on the GPU using pyCUDA, achieving speedups of ~5 for gridding and ~50 for CLEANing.

Video description

Video also available on YouTube

Imaging the debris disk Fomalhaut

Below is an example image using one of the earliest published science data from the ALMA telescope array. We started with the data obtained by Boley et al. (2012) (visualized on the left) to produce an image (shown on the right) of this dusty ring. For a great description of this data and the Fomalhaut system, see Astrobites. We thank Aaron Boley and Matt Payne for use of the data.