Cygnus A reconstructions at S band with R2D2



Cygnus A reconstructions with the R2D2 algorithm, a novel AI algorithm for high-dynamic range radio interferometry (RI) imaging. The imaged data are observations of Cygnus A at S band with the Very Large Array. The dataset consists of monochromatic estimated model (restored) images and associated residual dirty images, obtained by the three variants of the R2D2 algorithm, and benchmark algorithms.

R2D2 variants are:
(1) R2D2: a DNN series underpinned by U-Net modules;
(2) R2D2-Net: an unrolled end-to-end DNN reminiscent of the R2D2 algorithm;
(2) R3D3: a DNN series underpinned by R2D2-Net modules.

Modern algorithms used for benchmarking are:
(3) AIRI: a Plug-and-Play algorithm for RI (Terris et al. 2023);
(4) uSARA: a sparsity-based optimisation algorithm for RI (Repetti & Wiaux 2020, Terris et al. 2023);
Both AIRI and uSARA are part of the BASPLib code library.

CLEAN variants used for benchmarking are:
(5) Hogbom (Ho) CLEAN;
(6) Cotton-Schwab (CS) CLEAN;
(7) Multi-scale (MS) CLEAN.
All CLEAN variants are part of the WSClean software (Offringa & Smirnov 2017). CLEAN model component images are also provided.

Cygnus A images obtained with a joint calibration and imaging framework (Repetti et al. 2017) for the calibration of direction dependent effects (DDEs) and using either AIRI or uSARA as imaging modules, are provided for reference:
(8) AIRI_DDE-CALIM calibration;
(9) uSARA_DDE-CALIM calibration.
Date made availableApr 2024
PublisherHeriot-Watt University

Cite this