Next-Generation Radio Interferometry 

PI: Jean-Paul Kneib (EPFL)

Co-PIs: Emma Tolley, Paul Hurley, Sepand Kashani, Matthieu Simeoni, Gilles Fourestey 

July 1, 2021 – June 30, 2024

Project Summary

The Square Kilometre Array (SKA) project is an international effort to build the world’s largest radio telescope with ultimately over one million square metres of collecting area Wit.h an expected operational phase of at least 50 years, it will be one of the most important physics machines in the 21st century. SKA will tackle some of the most fundamental scientific questions of our time. Among them: How do galaxies evolve? What is Dark Energy and what role does it play in the expansion of the Universe? Can we find and understand where gravitational waves come from? What causes planets to form around stars? Is there life out there? Individually and working in tandem with other next-generation facilities, SKA will transform our understanding of the Universe. 

As a groundbreaking observational facility, SKA introduces new astrophysical, computational, and data analysis challenges. During its operation, the SKA will collect unprecedented amounts of data in the domain of 10 Exabytes, requiring the world’s fastest supercomputers to process this data in near real time. According to the SKA estimates, the Science Data Processor workflow will need to be able to deal with a data flow rate of around 1 TB/s at full capacity, and will need a supercomputer of around 100 Pflops to transform and compress the data before it is sent to the regional centers around the world for storage and final analysis by scientists. This implies that data analysis software of SKA must be be able to leverage this computing power. 

Image synthesis in radio astronomy is done with interferometry, a powerful technique allowing observation of the sky with antenna arrays with otherwise inaccessible angular resolutions and sensitivities. A major task of the SKA Science Data Processor workflow will be reconstructing the image of the sky from the radio signals recorded across all of the receivers in the array. However, image formation is a complicated problem. Current radio-interferometry imaging software (the CLEAN family of algorithms, including WSClean)  build the image of the sky by calculating an approximate “dirty image”, assuming that all radio sources in the sky are point sources, and iteratively subtracting distortions called “dirty beam” around these points. However, these algorithms will struggle to meet the requirements of SKA: they often require thousands of iterations on the same data to produce clean image. Additionally, the increased sensitivity and resolution of SKA imply that future observations may have more complex structure distributions across the field of view than simple point source fields. Thus next-generation imaging algorithms will not only have to cope with large data, but also to leverage more complex signal models.

The Bluebild algorithm offers a more robust, efficient way to calculate the sky intensity distribution. Bluebild is a flexible spherical imager for interferometric applications. It uses principle component analysis to calculate the sky intensity matrix in a series of energy levels. This provides astronomers with images of the sky decomposed into different energy levels, allowing for easy denoising through filtering low energy levels and recombination of the eigen-images in statistically optimal ways. With low computational complexity and affinity for parallel execution, Bluebild is an excellent candidate for next-generation radio-astronomy imaging. With additional optimization and parallelization, it may even achieve real-time imaging in the field. 

The goal of this project is to optimize the image synthesis algorithms of Bluebuild to reach real-time imaging requirements of SKA. Once optimized, Bluebild can be deployed in the telescope pipeline of one of the SKA precursor projects. This will help us validate performance and accuracy in a real scientific environment, and allow astronomers to analyse the data products produced by Bluebild. Then, Bluebild can be implemented in the SKA Science Data Processor, where it would run on the SKA tier-0 sites in South Africa and Australia. 

The fast Fourier series technique of Bluebild exploits FFTs while maintaining precision, and has great potential for many HPC applications beyond astrophysics. The fast Fourier series library which will be produced as part of this project will be useful for signal processing and the HPC community.