A Decentralized Framework for Radio-Interferometric Image Reconstruction
S. Wang, S. Mignot, S. Prunet, L. Di Mascolo, M. Spinelli, A. Ferrari · 2025
Read A Decentralized Framework for Radio-Interferometric Image Reconstruction by S. Wang, S. Mignot, S. Prunet, L. Di Mascolo, M. Spinelli, A. Ferrari (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-22128.txt.
Introduction
Radio telescopes of the SKA generation will pour out more data than one machine can process, so image building must be shared across computers. Wang and co-authors propose a decentralized framework that splits the work by spatial frequency. Each node holds part of the visibility data, the raw measurements, and rebuilds a near-full-resolution image by borrowing the major-cycle images already made by the other nodes.
They plug the scheme into two established reconstruction methods, multiscale CLEAN and a sparsity-based convex approach, and test it on four datasets with two ways of splitting the data. Image quality matched the serial versions while runtimes fell, most sharply on the larger datasets, where speedups approached the ideal twofold. The authors present the framework as one piece of the scalable pipeline the new telescopes demand.
Catalog record
| CREATOR | S. Wang, S. Mignot, S. Prunet, L. Di Mascolo, M. Spinelli, A. Ferrari |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-22128
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Decentralized Framework for Radio-Interferometric Image Reconstruction S. Wang, S. Mignot, S. Prunet, L. Di Mascolo, M. Spinelli, A. Ferrari Abstract The advent of large aperture arrays, such as the ones currently under construction for the SKA project, allows for observing the Universe in the radio-spectrum at unprecedented resolution and sensitivity. To process the enormous amounts of data produced by these telescopes, scalable software pipelines are required. This paper helps address this by proposing a framework that allows for decentralized radio-interferometric image reconstruction, parallelizing by spatial frequency. This is achieved by creating pseudo-full-resolution problems for each node by using the local visibilities together with previous major cycle reconstructed images from the other nodes. We apply the proposed framework to both multiscale CLEAN and sparsity regularized convex reconstruction and compare them to their serial counterparts across four different data sets of varying properties in the context of two visibility partitions. We found that the parallelization framework allows for significantly improved reconstruction times for images of similar quality. This was especially the case for our larger datasets where we were able to achieve close to the optimal 2 × 2\times speedup. I Introduction The two telescopes currently under construction for The SKA project 3 3 3 https://www.skao.int/ , SKA-Mid ( 350 MHz to 15.35 GHz) in South Africa and SKA-Low ( 50-350 MHz) in Australia, will allow us to observe the Universe \chreplaced [id=rev]atwithin these frequencies at unprecedented resolutions and sensitivities. These more detailed observations will result in large amounts of data to be processed. For SKA-Mid, projections estimate up to 2.375\;\text{TB/s} from the dishes to the beamformer and correlator engines, and 1.125\;\text{TB/s} from these to the imaging supercomputer ( Swart et al. 2022 ) . For SKA-Low, the estimated data transfer is 0.725\;\text{TB/s} from the antennas to the correlator and 0.29\;\text{TB/s} from the correlator to the imaging supercomputer ( Labate et al. 2022 ) . With these data rates, it is imperative for current image reconstruction algorithms to scale. On a coarse \chreplaced [id=min]levelscale, this involves parallelizing the sample data (i.e. visibilities) processing across multiple nodes of a cluster. This is typically performed along the embarrassingly parallel frequency and time domains, with independent reconstruction of each partition. Approaches that separate the image into facets for direction-dependent calibration, such as Cornwell & Perley 1992 ; Van Weeren et al. 2016 ; Tasse et al. 2018 , also enable parallelization in the spatial domain. This is more complicated due to the corner turn caused by each facet depending on all visibilities and vice versa \chadded [id=rev], but there has been recent work by Wortmann et al. 2024 that looks to address this. Another possible axis to parallelize is by spatial frequency. This requires more careful treatment compared to parallelizing by time and frequency, since image reconstruction methods typically need to process all the \chadded [id=rev]spatial frequencies together in order to obtain full-resolution reconstructions, leading to only a modicum of work being done in this respect ( Onose et al. 2016 ; Pratley et al. 2019 ; Van der Tol et al. 2018 ; Wortmann et al. 2024 ) . This paper addresses this by proposing a general decentralized framework to parallelize radio-interferometric image reconstruction by spatial frequency. This is achieved by filling in the missing information in each node with reconstructed images from the others, allowing each node to simultaneously reconstruct full-resolution images. It then combines these to obtain the final reconstructed image. We present our framework within the traditional major-minor loop image reconstruction paradigm, and apply it to two different image reconstruction methods, one based on convex optimization with sparsity regularization, such as in Wiaux et al. 2009 ; Carrillo et al. 2014 ; Garsden et al. 2015 ; Dabbech et al. 2015 , and the other being multi-scale CLEAN ( Cornwell 2008 ) . We evaluated our framework by comparing these parallel implementations with their serial counterparts across four different datasets. We found our framework to produce reconstructed images of comparable quality in substantially less time than the serial approaches, particularly in the case of larger datasets. The remainder of this paper is structured as follows: We provide a brief overview of radio-interferometric imaging as well as a review of the literature in Section II ; we describe our proposed parallelization framework in Section III ; we discuss how it can be applied to two existing reconstruction methods in Section IV ; Section V details our experimental setup, and Section VI presents and discusses our results; finally, we provide conclusions and discuss future avenues of research in Section VII . II Radio-interferometric imaging Radio-interferometers measure the sky using arrays of antennas (i.e. aperture arrays). Pairs of antennas (i.e. baselines) produce visibilities, the correlated instrumental response of the electrical field for some given time duration and electro-magnetic frequency. These can be related to the sky radiance distribution I(l,m) using the measurement equation ( Smirnov 2011 ) : \displaystyle\text{V}(u,v)=C_{uv}\iint D_{uv}(l,m)\frac{I(l,m)}{n}e^{-2\pi i(ul+vm+w(n-1))}\,dl\,dm, (1) \displaystyle n=\sqrt{1-l^{2}-m^{2}} where C represents the direction-independent effects, such as antenna gain, D denotes the direction-dependent effects, such as phase gradients caused by the Earth’s ionosphere, (u,v,w) is the difference between antenna coordinates in the frame where w is aligned with the phase center, and (l,m) are the spatial angular coordinates on the celestial sphere, which is also the domain of the integral. If we ignore the D and e^{-2\pi iw(n-1)} terms, Equation 1 simplifies to a two-dimensional Fourier transform, allowing us to retrieve an image of the sky emission through its inversion. This image contains artifacts due to the partial sampling caused by the nature of the antenna arrays and, especially in the case when imaging larger fields of view, by the omission of the w and D terms. Radio-interferometric imaging aims to correct for these, reconstructing an image that is usable for science. De/gridding Estimation \tilde{\imath}^{\,n} v Image update \bar{\imath}^{\,n} \hat{\imath}^{\,n+1} Figure 1: A high-level overview of the radio-interferometric pipeline. A reconstructed image \hat{\imath}^{\,n} is compared to the measured visibilities v in the de/gridding step, which outputs the residual \tilde{\imath}^{\,n} in the spatial domain. This is passed to the estimation algorithm which generates \bar{\imath}^{\,n} , an estimate of \tilde{\imath}^{\,n} without the effects of the measurement operator F^{\dagger}GG^{\dagger}F . The current reconstructed image \hat{\imath}^{\,n} is then updated to \hat{\imath}^{\,n+1} by adding \bar{\imath}^{\,n} to it. As the estimation algorithm is usually iterative, the overall pipeline often has a major-minor loop structure. The general approach of most radio-interferometric imaging algorithms is iterative and is illustrated in Figure 1 . The residual \tilde{\imath}^{\,n} between the measured visibilities v and the current image estimate \hat{\imath}^{\,n} , also termed the dirty image when n=0 , is computed in the de/gridding step. In image space, this can be expressed as \tilde{\imath}^{\,n}=F^{\dagger}G(\text{v}-G^{\dagger}F\hat{\imath}^{\,n}) (2) where F and F^{\dagger} are the Fast Fourier Transform (FFT) and its inverse respectively, and G and G^{\dagger} are the gridding and degridding operators for resampling visibilities to regular and irregular positions, respectively. There is a myriad of methods for G and G^{\dagger} , most of which also include the correction for w , with some examples being the work of Cornwell et al. 2008 ; Offringa et al. 2014 ; Van der Tol et al. 2018 ; Ye et al. 2022 . The