Scaling convex optimisation for radio astronomy

Adrian Jackson, Pierre-Antoine Thouvenin, Ming Jiang, Abdullah Abdulaziz, Arwa Dabbech, Yves Wiaux

Research output: Chapter in Book/Report/Conference proceedingConference contribution

22 Downloads (Pure)

Abstract

Aperture synthesis by interferometry in radio astronomy allows observation of the sky by antenna arrays with otherwise inaccessible angular resolutions and sensitivities, providing a whole wealth of information for astrophysics and cosmology. At the target resolution and dynamic range of interest to upcoming telescopes, image cubes will reach close to Petabyte sizes, with data volumes orders of magnitude larger, possibly verging on the Exabyte scale. In this context, convex optimisation theory offers modern algorithmic structures that can handle extreme data volumes by distributing the computation across large computer systems, while explicitly incorporating complex image prior models to regularise the problem. Motivated by these advantages, we introduce a C++ library and set of applications, dubbed Puri-Psi, as a production implementation of the recently proposed RI imaging approach HyperSARA, previously available as a proof-of-concept MATLAB implementation. Puri-Psi enables scaling up to very large data volumes by parallelising across a large number processes and compute nodes. We validate Puri-Psi against the MATLAB implementation, and evaluate its performance in a representative high performance computing setting.
Original languageEnglish
Title of host publicationProceedings of ISC HPC 2021
Number of pages10
Publication statusSubmitted - Dec 2020

Fingerprint

Dive into the research topics of 'Scaling convex optimisation for radio astronomy'. Together they form a unique fingerprint.

Cite this