Adaptive Uniform Weighting: Pre-conditioning to Improve Image Fidelity

TL;DR

The paper addresses how visibility weighting in radio interferometry shapes the conditioning of the deconvolution problem and image fidelity. It introduces Adaptive Uniform Weighting (AUW), a data-driven scheme that computes a local occupancy-based density $o(r)$ and uses $w_u = w_m w_o$ with $w_m = 1/M$ to produce a more accurate dirty beam, especially when sampling is incomplete. AUW demonstrates up to 2–10× improvements in PSF quality and final image fidelity across multiple facilities and observing modes, while remaining fully adaptive with no user-tunable knobs beyond field and pixel choices. The approach is compatible with 3D imaging methods and offers a practical path to higher-fidelity interferometric imaging in next-generation facilities.

Abstract

The "dirty" image made by direct Fourier inversion of visibility data is an important first step in inteferometric imaging. This is where the "deconvolution problem" is defined and the degree to which that problem is either well- or ill-conditioned has direct consequences for the ultimate image fidelity that is achieved in practise. An under-utilised degree of freedom during Fourier imaging is the relative weights that are assigned to the visibility data. We explore the circumstances under which some adjustment of the relative weights might provide improvements to the "dirty" image, and consequently the ultimate post-deconvolution image fidelity. We develop a method to calculate a distinct effective local density estimate for each data point. When used in conjunction with a "uniform" weight correction and the desired clean beam (eg. Gaussian) tapering, it provides a significant improvement in the image quality over that provided by the current pixel-based density estimate. In many cases, particularly spectral-line observations and those with only limited sidereal tracking, this adaptive approach improves the beam quality by a factor of 2 to 10, as measured by the RMS residual relative to the best-fitting clean beam, providing an improvement in final image fidelity that is similar in magnitude.

Figures (9)

• Figure 1: Comparison of restored image fidelity with standard Uniform Weights (UW), Briggs, (BRI), Natural (NA) and Adaptive Uniform Weights (AUW) for a noise-less (top-left), low-noise (top-right), nominal-noise (bottom-left) and high-noise (bottom-right) monochromatic 4-hour tracking observation with SKA-Low. Image fidelity has been averaged in logarithmically spaced annuli.
• Figure 2: Sky model used for image fidelity simulations. The full $6.4\times 6.4$ degree field tapered with the primary beam model is shown on the left and the central portion on the right. The peak brightness is 1.5 Jy/Beam, the diffuse emission plateau has brightness of 0.2 mJy/Beam (about 25 K), the faintest discrete sources are < 1 $\mu$Jy/Beam.
• Figure 3: Comparison of imaging performance with standard Uniform Weights (UW, dashed lines) and Adaptive Uniform Weights (AUW, solid lines) for SKA-Mid. The monochromatic case is on the left and multi-frequency synthesis with 40% fractional bandwidth on the right. Bottom panels show the image RMS noise relative to Natural weighting. The other panels show the RMS residual after a Gaussian fit to the PSF for a snap-shot (top) and full track (centre) observation.
• Figure 4: Examples of average occupancy versus $(u,v)$ radius (filled circles) for different facilities, observing strategies and target beam FWHM. The fit to $o(r)$ is overlaid as the solid line and parameter values are listed.
• Figure 5: As in Figure \ref{['fig:SKA-Mid']} for the International LOFAR telescope.