[daip] Re: FW: Questions about SAD (AIPS) inputs

[Eric Greisen]

Becky Brown writes:

 > Could you direct me to the correct resource for answers to these
 > questions?
 > 
The correct resource for all aips questions is daip at nrao.edu.  I
handle most of what comes in but this address goes to the full group
and others may help answer esp if I am on travel.

 > 
 > -----Original Message-----
 > From: Rich White [mailto:white at radar.nrl.navy.mil] 
 > Sent: Wednesday, July 13, 2005 9:07 AM
 > To: becky.brown at nrl.navy.mil
 > Subject: Questions about SAD (AIPS) inputs
 > 
 > Hi Becky.
 > You mentioned yesterday that one of your acquaintances might be familiar
 > 
 > with the SAD program in AIPS.  Here are some of the questions  I still
 > have 
 > about the inputs to this program:
 > 
 > 1. NGAUSS - How should this value be set for a scene with an unknown
 > amount 
 > of sources?  What if # of sources in the image are expected to vary 
 > somewhere between 0 - 50?

    This is just a limit - don't worry about it.  If you expect no
more than 50 set it to 75 or so.

 > 2. CPARM - How does one determine how high to set the highest value, how
 > 
 > low to set the lowest, and why bother with intermediate steps?

    This does depend on your image.  Usually the first level is say
50% of the peak or so and the last level is on order 3 sigma.  At
least one level in between may help to avoid failure due to trying to
fit Gaussians to low-level woofly noise areas.

 > 3. DOWIDTH - Under what circumstances is it good to set to 1?

     All.  Unless your Gaussians are of a known width you should solve
for the width.

 > 4. GAIN -  How does one know what to set this value to?

     Depends on how your noise and unceratinties behave.  In Clean
radio images the noise is higher on source than off and so GAIN is an
allowance for that.  Not too important I suspect most people use 0.1
because that is the initial value of this adverb.

 > 5. DPARM - How does one generally determine these settings besides using
 > a 
 > trial and error approach?


      I suspect that you can set initial guesses reasonably but trial
and error is ultimately the approach.

Eric Greisen