Monday, June 22, 2009

Mulit-Resolution Clean Tutorial?

Here is a very reasonable request from an anonymous reader:

Can someone write about multi-resolution clean(MRC)..basically the steps involved and the way self cal should be carried out while using MRC ? thanks in advance.

which I can not fulfill, as I have never successfully pulled off a MRC. You might try looking at the EXPLAIN file for IMAGR in the section called "Extended Source Options".

If anyone knows better resources for learning about how to implement multi-resolution clean, please let us know!


2 comments:

James said...

While I don't know any decent resources to look at for multi-res clean, I have some experience in using it.

Inputs:
- For my data I found IMAGRPRM(11)~0.75 0.02 0.2 would work well
- Typically I use 3 resolutions, i.e. ngauss 3.
- I choose resolutions spaced by a factor of about 3, i.e. pointlike, 3x pointlike, and 9x pointlike (these are the wgauss values)
- I choose flux cutoffs (wgauss values) scaled by the beam area, so 1, 10 and 100 times (should technically be 1, 9 and 81 times) an appropriate noise level (set by the approximate min algorithm flux found at the end of the deconvolution when doing a non-MR CLEAN).

The flux cutoffs take some playing with. I set them fairly conservatively high at first, and then stop and restart IMAGR as otherwise it never ends up getting to the point source resolution. This way it will reach a cutoff flux and move on to the higher resolution. If you still want to CLEAN some more flux off at the lower resolution, then stop cleaning, and restart IMAGR.

If you restart IMAGR like this, use:
- outver 1
- bcomp 1e6
- allokay 1
- geton (catalog number of ICL001 file)
- plus your adjusted fgauss cutoff values

When you selfcal, you need to be careful. IMAGR will restore all the flux in the images to the highest resolution image. However, it does not do the same to the CC tables. The flux cleaned off each resolution is stored in the CC table of that resolution, and not restored to the corresponding CC table of the pointlike resolution image. So for your selfcal model, you need to use all images at all resolutions. For instance, if you have 3 resolutions and 41 facets per resolution, you need to use nmap 123 (3x41). Despite the fact that it has restored all the flux to the highest-resolution images, if you only use these in the selfcal you will miss out on all the flux you cleaned off from the extended emission in the low-resolution images, and your selfcal will fail horribly.

Does that help? Good luck!

ruta said...

thank you..