Fourier processing

For wide field imaging with w term correction, the Nifty Gridder is supported and has the best performance. It is installed via pip as part of the RASCIL install process.See more information on nifty-gridder at:

https://gitlab.mpcdf.mpg.de/ift/nifty_gridder

Nifty Gridder is supported at the processing component level via:

• rascil.processing_components.imaging.ng.invert_ng()

• rascil.processing_components.imaging.ng.predict_ng()

At the workflow level, such as imaging and pipeline workflows, use context=’ng’.

RASCIL model

If only wterm needs to be corrected, the Nifty Gridder is the best option. However RASCIL has another approach. There are many algorithms for imaging, using different approaches to correct for various effects:

• Simple 2D transforms

• AW projection

• MFS variants

Since the scale of SKA is so much larger than previous telescopes, it is not clear which scaling strategies and algorithms are going to offer the best performance. For this reason, it is important the synthesis framework not be restrictive.

All the above functions are linear in the visibilities and image. The 2D transform is correct for sufficiently restricted context. Hence we layer all algorithms on top of the 2D transform. This means that a suitable framework decomposes the overall transform into suitable linear combinations of invocations of 2D transforms.

The full layering is:

• The core imaging functions are defined in processing_components.imaging.base.

• Nifty Gridder provides optimised wstack/wprojection imaging and should be the default.

• AW projection is also possible using the 2d gridder and a suitable gridding convolution function.