Reconsidering the advantages of the three-dimensional representation of the interferometric transform for imaging with non-coplanar baselines and wide fields of view

Radio telescopes with baselines that span thousands of kilometres and with fields of view that span tens of degrees have been recently deployed, such as the Low Frequency Array, and are currently being developed, such as the Square Kilometre Array. Additionally, there are proposals for space-based instruments with all-sky imaging capabilities, such as the Orbiting Low Frequency Array. Such telescopes produce observations with three-dimensional visibility distributions and curved image domains. In most work to date, the visibility distribution has been converted to a planar form to compute the brightness map using a two-dimensional Fourier transform. The celestial sphere is faceted in order to counter pixel distortion at wide angles, with each such facet requiring a unique planar form of the visibility distribution. Under the above conditions, the computational and storage complexities of this approach can become excessive. On the other hand, when using the direct Fourier transform approach, which maintains the three-dimensional shapes of the visibility distribution and celestial sphere, the non-coplanar visibility component requires no special attention. Furthermore, as the celestial samples are placed directly on the curved surface of the celestial sphere, pixel distortion at wide angles is avoided. In this paper, a number of examples illustrate that under these conditions (very long baselines and very wide fields of view) the costs of the direct Fourier transform may be comparable to (or even lower than) methods that utilise the two-dimensional fast Fourier transform.