SMURFIT: Spatial-frequency Multi-band Radio Fourier Imaging Technique
Abstract
The currently under construction SKA telescopes will allow for observations at unprecedented sensitivities in their respective electro-magnetic frequency ranges, enabling both faster and deeper surveys. A cost to this is the enormous amounts of data generated, necessitating scalable data reduction pipelines. This paper focuses on the imaging pipeline, which transforms measured visibilities into images of the sky. It extends a previous imaging framework that decentralizes the imaging process by spatial frequency across two nodes, which we retroactively name SURFIT, to multiple nodes. Specifically, this paper generalizes the descriptions of the deconvolution algorithms to account for images received from multiple nodes, as opposed to just one. It also describes how to prepare the datasets so that proper load balancing is achieved, and provides a method to estimate the scalability of the framework to an even larger number of nodes. We term this overall framework SMURFIT. We evaluated our framework across two simulated datasets based on SKA telescope configurations, as well as a smaller real dataset from ALMA, and found that it demonstrated largely linear speedup in visibility processing for up to 10 nodes, while reconstructing images of roughly the same quality for all but one of our test cases. We also show how the new load balancing strategy can be used to model SMURFIT's theoretical scaling, which we demonstrate for up to 50 nodes for our simulated datasets.
Related
- Synthetic-Aperture Super-Resolution Imaging via Spatial-Frequency Shift in Near-Field Diffraction
- Determining low-ℓ p-mode frequency shifts in Sun-like stars: Enhancing the cross-correlation technique with filters
- Fourier imaging of collective spontaneous emission modes in superradiant cold atomic clouds
- Faithful reconstruction of orthogonal circular polarization holography modulated by spatial frequency in TI/PMMA photopolymers
- Modulation transfer function measurement of lenses over the full spatial-frequency range using CCD moiré
BibTeX
@misc{wang2026smurfit,
title = {{SMURFIT: Spatial-frequency Multi-band Radio Fourier Imaging Technique}},
author = {Sunrise Wang and Simon Prunet and Shan Mignot and Andre Ferrari},
howpublished = {HAL},
year = {2026},
month = oct,
doi = {10.2139/ssrn.6967740},
url = {https://hal.science/hal-05842457v1},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - SMURFIT: Spatial-frequency Multi-band Radio Fourier Imaging Technique
AU - Sunrise Wang
AU - Simon Prunet
AU - Shan Mignot
AU - Andre Ferrari
PY - 2026
DA - 2026/10/28
PB - HAL
DO - 10.2139/ssrn.6967740
UR - https://hal.science/hal-05842457v1
AB - The currently under construction SKA telescopes will allow for observations at unprecedented sensitivities in their respective electro-magnetic frequency ranges, enabling both faster and deeper surveys. A cost to this is the enormous amounts of data generated, necessitating scalable data reduction pipelines. This paper focuses on the imaging pipeline, which transforms measured visibilities into images of the sky. It extends a previous imaging framework that decentralizes the imaging process by spatial frequency across two nodes, which we retroactively name SURFIT, to multiple nodes. Specifically, this paper generalizes the descriptions of the deconvolution algorithms to account for images received from multiple nodes, as opposed to just one. It also describes how to prepare the datasets so that proper load balancing is achieved, and provides a method to estimate the scalability of the framework to an even larger number of nodes. We term this overall framework SMURFIT. We evaluated our framework across two simulated datasets based on SKA telescope configurations, as well as a smaller real dataset from ALMA, and found that it demonstrated largely linear speedup in visibility processing for up to 10 nodes, while reconstructing images of roughly the same quality for all but one of our test cases. We also show how the new load balancing strategy can be used to model SMURFIT's theoretical scaling, which we demonstrate for up to 50 nodes for our simulated datasets.
ER -