Wide-field spectral-line imaging with modern radio interferometers remains challenging when the emission is extended, multiscale, and distributed over many overlapping pointings. In this regime, accurate reconstruction requires joint treatment of the calibrated visibilities, control of image-domain regularity, and, when necessary, recovery of missing short spacings from single-dish data. Marchal et al. present IViS (Interferometric Visibility-domain Inversion Software), a GPU-accelerated Python framework for visibility-domain joint deconvolution, and assess its performance for wide-field HI imaging with ASKAP.

The current model, Classic3D, reconstructs a non-parametric sky cube directly from calibrated visibilities by minimizing a regularizedleast-squares criterion. The forward model relies on a non-uniform fast Fourier transform (NUFFT), supports joint deconvolution across all mosaic pointings, and allows positivity constraints through bounded optimization. Short-spacing information can be incorporated through a fusion term that models single-dish data from the same sky estimate. The team validate the method on simulated data and apply it to ASKAP full-survey observations toward the Large Magellanic Cloud (LMC).

The image above shows an ASKAP image obtained with IViS (Classic3D) for the LMC at a velocity (LSR) of 238.6 km/s.  The white dashed contours annotate the 0.05 and 0.1 effective primary beam levels. The white square outlines the region used to compute spatial power spectra.  The right panel shows the short-spacing corrected image obtained using traditional feathering with Parkes Murriyang data.

Relative to linear mosaicking of independently deconvolved pointings, joint deconvolution can recover more large-scale power. When single-dish information is included, the reconstructed power spectra closely reproduce the input sky statistics over a broad range of spatial frequencies.  IViS provides a flexible visibility-domain framework for joint deconvolution of large spectral-line mosaics, and illustrates the potential of regularized inverse-problem approaches as an alternative to traditional CLEAN-based imaging for next-generation radio surveys.