[alma-config] Difference images TNG

[Mark Holdaway]

I may be jumping the gun here, but based on a visual inspection of the
difference images, it appears that the spiral configurations win in most
cases.  We obviously need to go through the process of evaluation to
quantify this. Ideally, we would have a second round of work, where we try
to understand why things are bad and work on ways to improve them. So, it
seems that using a deconvolution mask or using a default image for Mars
would be the sorts of things you do in round two.  Someone (Bryan?)
pointed out that the masks are not known a prior, and need to be set
either automatically (possibility of disaster!), by an expert user, or a
combination of the two, and may need to be adjusted iteratively.  Of
course, this is exactly the problem the imaging pipeline people will find
themselves in.  Exept for the simplest of cases, automatic mask generation
is probably outside the scope of this work.  

So, what do we do?  Run with the current results, or go another level
in sophistication?

Another dimension which we have said we would address is error modeling.
What is the status of work in that dimension?

	-Mark


> Hello.
> 
>  In response to Bryan and Mark's queries, all the images are treated the
> same when CLEANing, i.e. no masks, no boxes, no initial models. By way of an
> example, below is an extract of the batch file for CLEANing the KBM51_4-23B7
> case. Note however that much of the sensible defaults are embodied in the
> nuvco3mm parameter set. The parameters used are hopefully as decided by
> consensus.
> 
> Now having said that all images are treated the same, there is no problem
> with using model-specific parameter sets when CLEANing. In the batch file
> extract below you will note that some adverbs are in upper case- these are
> read in from an external file of CLEAN parameters, which may be specified on
> a per-model basis in the setup file that my scripts use to create the batch
> file (the parameters shown are read in by default when the input file is
> specified as 'none').  As the external file is read immediately before
> running IMAGR any of that task's adverbs may be modified at this point.
> 
> So if it is necessary and desirable to redo all the CLEANing using initial
> models and CLEAN boxes, fine, although abviously the less the better!
> Speaking as a relative novice at this game, if the Mars results are poor
> because of lack of short spacings and show no distinction between the K and
> C configurations, is that not the bottom line... i.e. worry about Mars at
> other resolutions rather than using CLEAN to make a silk purse out of a sows
> ear. (I'm sure there is more to it than that, I'm just hoping someone will
> tell me!)
> 
> Moving along to the metrics as discussed at the last telecon, I have done a
> few FFTs (I presume we are interested in power spectra rather than
> considering real and imaginary components). Also I have found an excellent
> AIPS task (IMEAN) which allows me to export image statistics and indeed
> histogram data to a named file, thus allowing max, min, and rms data to be
> listed with plots in the future.I have also been looking at calculating
> Fidelity images and their histograms as highlighted by Stephane, and also
> Leonia's 'fidelity-range' measure, and this all seems fairly
> straightforward.
> This essentially uses IMSTAT or IMEAN to yield the rms of the difference
> image already calculated, IMCLP to apply the 0.7x rms threshold to the
> abs(difference image), COMB to calculate the ratio, and IMEAN to reveal the
> histogram and new rms etc.
> 
> The same principles can be employed to calculate John's metrics of rms
> values for a series of (difference image)/(convolved model magic blanked at
> a certain level), assuming 'magic blanked' is synonymous with 'brutally
> thresholded'!
> 
> So I am in the position to
>     a) produce FFTs of difference images
>     b) generate fidelity images and associated numbers
>     c) calculate John's metric
>     d) redo CLEANing with new parameters, for some/all images at some/all
> resolutions
> whilst also thinking about
>     e) try some MEM deconvolutions
> 
> e) looks in jepoardy just now, but some indication of priority for the
> others would be useful along with parameters and/or initial models for d).
> In the meantime I will start scripting a) b) and c), and will probably
> produce some initial results for your perusal.
> 
> Cheers,
>        Steven
> 
> 
> Extract of batch file for CLEANing
> ===========================
> version 'myaips'
> prnumber 0
> prtask 'imagr'
> go clrmsg
> prtask ' '
> get nuvco3mm
> cellsize 0.02
> inname 'KBM51_4-23b7'
> inclass 'nuvco'
> inseq 0
> indisk 0
> bchan 0
> outname ' '
> outseq 0
> outdisk 11
> in2name ' '
> in2class ' '
> in2seq 0
> in2disk 11
> selfreq 345000
> NITER 10000
> NBOXES 0
> CLBOX 0
> GAIN 0.1
> MINPATCH 256
> BMAJ 0.101
> BMIN 0.101
> BPA 0
> PHAT 0
> FACTOR 0
> CPARM 12,0
> go imagr
> inclass 'ICLN'
> inseq 0
> blc 256
> trc 768
> plev 0
> outfile 'myarea:KBM51_4-23b7.ICLN.PS'
> go cntr
> functype 'NE'
> go lwpla
> outfile 'myarea:KBM51_4-23b7.ICLN2.PS'
> go grey
> functype 'NE'
> go lwpla
> functype 'LN'
> 
> Steven Heddle, Newington, Holm Road, Kirkwall, Orkney KW15 1PY
> Tel: 01856 877119             Email: steven at heddle97.freeserve.co.uk
> Fax: 0870 4039359           URL: http://www.heddle97.freeserve.co.uk
> 
>