Compare screen-based DD imaging with facet-based DD imaging for SKA1-LOW

To achieve high-quality imaging results with the SKA, we need to understand the trade-offs between, e.g., computational demands and image quality across the field-of-view for various available imaging approaches. Here we look into the impact of screen-based vs. facet-based DD correction during imaging.

• SKA-LOW simulated data with known A-terms are imaged using IDG
• SKA-LOW simulated data with known A-terms are imaged using WSClean as-is
• WSClean is tuned to do facet-based imaging (stretch)
• SKA-LOW simulated data with known a-terms are imaged using facet-based approach in WSClean (stretch)
• Report on comparison of the results

• W-tiling was implemented in IDG. This technique reduces the memory usage significantly and reduces execution time by a factor ~2.5. This allows to make images of 30k-by-30k and larger on "normal" compute nodes.
• A bug was solved that caused incorrect imaging for non-centred facets.
• To isolate imaging from calibration, we needed to reconstruct the beam and ionospheric screen used in the simulation. For this, we used the OSKAR infrastructure to do this. Unfortunately, a coordinate issue in OSKAR was solved between the time the SKA-LOW simulated data sets were created and the time at which we tried to reconstruct the beams and screens that went into that simulation causing imperfect reconstruction.
• For this reason, we decided to follow another path as outlined in AST-181, which will be demonstrated at System Demo 9.1 on December 16, 2020.