ska-sciencedataprocessor / crocodile Goto Github PK

Crocodile's 'inv' function is described as "Invert a hermitian-symetric two-dimensional grid. The hermitian symetric dimension is the second (last) index, like in numpy.fft":

AFAIUI, this is related to how the sample uv data are generated, see

The question is: what if we use uv data generated by Oskar? Do they have the same structure (hermitean-symmetric)? And if they have more general structure then how should 'inv' look like in this general case?

The way gridding works, we have a few connected characteristics of the handled data:

• In doimg, we "mirror" coordinates and conjugate visibilities. In my understanding, this is correct because it reflects "inverted" baselines.
• This causes in turn causes the result of gridding to be hermitian
• This property is what makes sure that the result image of the FFT is, in fact, real-valued

It would make the program easier to follow if this was spelled out explicitly.

The first parameter of 'overlapIndices' is 'dirty here: https://github.com/SKA-ScienceDataProcessor/crocodile/blob/master/clean.py#L73. Is this correct? Shouldn't it be 'res' instead perhaps?

As Anna commented in the Radio Cortex specification, Hogbom produces a clean image, a model image and residual. We believe this maps to comps+res, comps and res respectively, but this is never documented explicitly.

Calls to wkernelaf in synthesis.py contain hard-coded numbes 257 and 15. Those would be better understood if they had text names. Also, should those be adjustable in any way?