Deep wideband single pointings and mosaics in radio interferometry - How accurately do we reconstruct intensities and spectral indices of faint sources?

TL;DR

This study evaluates the accuracy of different wideband radio interferometric imaging methods in reconstructing faint source intensities and spectral indices, highlighting the advantages of joint reconstruction techniques like MT-MFS.

Contribution

It provides a comprehensive comparison of wideband imaging methods using realistic simulations, demonstrating the effectiveness of joint reconstruction and correction techniques in radio interferometry.

Findings

Errors increase for weaker sources even without noise.

Joint reconstruction methods reduce errors and eliminate Clean-bias.

Simulations must be detailed to accurately reflect real observations.

Abstract

Many deep wide-band wide-field radio interferometric surveys are being designed to accurately measure intensities, spectral indices and polarization properties of faint source populations. In this paper we compare various wideband imaging methods to evaluate the accuracy to which intensities and spectral indices of sources close to the confusion limit can be reconstructed. We simulated a wideband single-pointing (C-array, L-Band (1-2GHz)) and 46-pointing mosaic(D-array, C-Band (4-8GHz)) JVLA observation using realistic brightness distribution ranging from $1\mu$Jy to $100m$Jy and time-,frequency-, polarization- and direction-dependent instrumental effects. The main results from these comparisons are (a) errors in the reconstructed intensities and spectral indices are larger for weaker sources even in the absence of simulated noise, (b) errors are systematically lower for joint reconstruction methods (such as MT-MFS) along with A-Projection for accurate primary beam correction, and (c) use of MT-MFS for image reconstruction eliminates Clean-bias (which is present otherwise). Auxiliary tests include solutions for deficiencies of data partitioning methods (e.g. the use of masks to remove clean bias and hybrid methods to remove sidelobes from sources left undeconvolved), the effect of sources not at pixel centers and the consequences of various other numerical approximations within software implementations. This paper also demonstrates the level of detail at which such simulations must be done in order to reflect reality, enable one to systematically identify specific reasons for every trend that is observed and to estimate scientifically defensible imaging performance metrics and the associated computational complexity of the algorithms/analysis procedures.