That pesky A-term: Efficiently correcting for direction-, time-, and baseline-dependent effects in radio interferometric imaging

TL;DR

This paper introduces Pigi, a GPU-based implementation of the IDG algorithm, enabling fast and accurate correction of direction-, time-, and baseline-dependent effects in radio interferometric imaging, suitable for large-scale telescopes like the SKA.

Contribution

The paper presents a new GPU-accelerated implementation of the IDG algorithm, significantly improving imaging speed and accuracy for A-term corrections in radio interferometry.

Findings

Pigi can process nearly 0.5 billion visibilities per second.

It effectively corrects for extreme ionospheric effects in simulated data.

The method is compatible with NVIDIA and AMD GPUs.

Abstract

Radio interferometers must grapple with apparent fields of view that distort the true radio sky. These so-called 'A-term' distortions may be direction-, time- and baseline-dependent, and include effects like the primary beam and the ionosphere. Traditionally, properly handling these effects has been computationally expensive and, instead, less accurate, ad-hoc methods have been employed. Image domain gridding (IDG; van der Tol et al., 2018) is a recently developed algorithm that promises to account for these A-terms both accurately and efficiently. Here we describe a new implementation of IDG known as the Parallel Interferometric GPU Imager (Pigi). Pigi is capable of imaging at rates of almost half a billion visibilities per second on modest hardware, making it well suited for the projected data rates of the Square Kilometre Array, and is compatible with both NVIDIA and AMD GPU hardware. Its accuracy is principally limited only by the degree to which A-terms are spatially sampled. Using simulated data from the Murchison Widefield Array, we demonstrate the extraordinary effectiveness of Pigi in correcting for extreme ionospheric effects, and point to future work that would enable these results on real-world data.