[daip] SDCLN and additional questions

[Eric Greisen]

Rene Giessuebel wrote:
> Hello,
> 
> sorry in advance if some questions might be a bit stupid, this is my
> first try reducing interferometry data.
> I have VLA data (C-Band, D configuration) of an extended source. I want
> to clean the image using the task SDCLN
> I get the following error:
> 
> pc2001> SDCLN1: BMSHP ERROR: CENTER OF BEAM =  0.49187E-04
> 
> Could you please explain to me what it means (what is the problem), and
> how I can fix it?
> The map-file I used was created using IMAGR with NITER 0

SDCLN has, I suspect, not been used in quite a while.  Image plane 
Cleaning just does not work as well as uv-plane Cleaning but it should 
be faster.  The error message says that the Beam image has value 
0.4918.. at pixel NX/2, NY/2 + 1.  Since that is forced to be 1.0 this 
seems to be an error - did you sub-image the output of IMAGR?  SDCLN 
really requires images that are a power of 2 and that are (linearly) 2 
times bigger than the area you wish to Clean.
> 
> I tried to use UVMAP as described in the help/explain of SDCLN, but it
> complained:
> pc2001> UVMAP1: ERROR: I cannot process packed UV data, use SPLIT
> even after using SPLIT ... I tried to do self-calibration and did some
> additional flagging on the "initial" split-file, so I want to use that
> one ... or is that the cause?
> 
UVMAP is an antique and does not do weighting well - always use IMAGR 
for serious imaging.
> Or would you advise to use IMAGR for SDI Clean anyway?
> 
Yes - or try multi-scale which I think works better unless your source 
really looks like a flat plateau with some bumps.

> Also I did not fully understand how to use cleaning boxes. Some people
> say to use boxes as small as possible around the point-sources, others
> advise to use big boxes and claim that small boxes somehow bias the
> cleaning towards small structures(?)
> Anyway, I played around a bit and found that I get the best results by
> arbitrarily setting rather big boxes around the region I want to clean
> next, starting around a couple of brighter sources and always selecting
> around the strongest sources I see in the residual ... so it still is a
> semi-structured way of doing it, but seems "wrong" to me.
> Any advise on where I can read up on why and how to use boxes? I am
> really confused with all the different types of boxes and fieldsizes one
> can define for cleaning actually ... and what does NBOXES do when I set
> more then one box in the TV?

Setting the boxes as you go is probably better although signal strength 
selects what is Cleaned if you start with a lot of boxes.  The general 
belief is that relatively tight boxes are a good idea.  If you simply 
let IMAGR Clean the entire area, it will produce an image with lower rms 
by sucking flux from real sources and creating erroneous sources with 
that flux.  Boxes can be too tight - I was once forced to Clean CasA 
with an image plane Clean in an image that was just slightly too small.
The tight box that I was forced to use, turned CasA into a square 
structure!  As soon as you set boxes, FLDSI no longer matters unless you 
are using multiple facets in which case FLDSI can change the size of the 
images.  The adverb NBOXES is not set by IMAGR when you increase the 
number of boxes.  I use BOXFILE and OBOXFILE (often the same file) so 
that when I make new boxes they are remembered for next time.
> 
> Another question I have concerns subtraction of sources.
> I have a source in the north of the image which produces beautiful but
> unwanted patterns over bif parts of the image even after cleaning.
> I tried to remove the source from the UV-file, but the method I found
> requires a good model from the cleaning-components. But if I had good
> CC for that source, I obviously wouldn't want to remove it ... ;-)
> Is there another way?
> Would it be worthwhile doing self-calibration on this source only just
> to remove it, and then use these CC to remove it from an UV-file with
> "normal" calibration? I don't have really strong sources (strong as in
> "good for selfcal") though, so selfcal probably doesn't do much anyway.
> 

You cannot subtract the source from the UV data unless you have some 
sort of model for it.  You may need to make the image with 2 facets, one 
centered on the source to the North.  Put a box to Clean that source and 
any others near it in that facet and all other boxes in the other facet.
Then use PEELR to self cal the Northern source only.  What has probably 
happened is that squint and pointing errors have made the source near 
the edge of the beam have different gains than the sources at the middle.

If there is not enough flux for self-cal, then I do not have any tricks 
up my sleeve.  I am a fan of the multi-scale Clean so give it a try. 
There are references and clues to its use in the help/explain file.

Eric Greisen