[daip] Re: CALIB issues

[Eric Greisen]

Carl Gross writes:

 > 1.  During the initial calibration of each single polarization data set,
 > CALIB reported the correct number of good and failed solutions.  But when I
 > calibrated a dual polarization data set in which I had flagged data in one
 > polarization but not the other, then CALIB would sometimes incorrectly
 > report those flagged data as failed solutions.  This definitely occurred in
 > only SOME such instances, and not ALL such instances.  In both cases (single
 > and dual polarization), CALIB would report the correct number of good
 > solutions.

       I have tried a number of tests and do not confirm your
observations.  For 2 polarization, 2 IF data the number of good
solutions was reported as 1620 (no failures).  If I flagged 1
polarization of 1 IF, the number of good solutions bacme 1215 (3/4 of
the total as expected).  No failures were reported in 31DEC05 or
31DEC06 although 31DEC04 reported 405 failures.

 > 
 > 2.  After I combined all six data sets and began self cal, CALIB would
 > report odd numbers as well.  If I combined one or two data sets into a
 > single DBCON UV file, the solutions that CALIB reported were dictated by the
 > parameters I detailed in #1.  If I combined three to six data sets (I did
 > not combine any more than 6 data sets, so I can't comment on anything else)
 > into a single DBCON UV file however, the story changed.  For the single
 > polarization data (i.e. the subarrays that were from the single polarization
 > data sets), CALIB reported the correct number of good solutions.  But the
 > number of failed solutions it reported obeyed this equation:  # of failed
 > solutions reported = # of actual good solutions + 2*(# of actual failed
 > solutions).  For the dual polarization data (i.e. the subarrays that were
 > from the dual polarization data sets), CALIB again reports the correct
 > number of good solutions.  But the number of failed solutions again suffers
 > from the problem described in #1.

     In the DBCON data set the single-polarization data are actually
recorded as dual polarization with 1 polarization completely flagged.
The #failed = #actual good + 2 * actual failed makes sense in the old
version where every possible solution in one polarization "failed"
plus some solutions in the good polarization also actually failed.


 > This calculations are not based on speculation; I arrived at them by
 > thoroughly plotting SN and FG tables and making sure that failed solutions
 > were indeed occurring where data were flagged.  
 > 
 > With initial calibration, I'd like to ensure as few failed solutions as
 > possible.  The fact that CALIB reports the incorrect number of failed
 > solutions for dual polarizations requires me to track down every failed
 > solution to make sure that it is indeed a failed solution and not just
 > flagged data.  When self cal'ing, I would like the know the percentage of
 > good solutions to ensure that it is increasing during every step of self
 > cal.  The issues I descbribed in #2 require me to again track down every
 > failed solution to see whether it is indeed an actual failed solution.  I
 > was wondering if you had ever heard of this before, and if so, is there a
 > work around?  If not, I will just continue to be thorough and make sure that
 > all the CALIB failed solutions are definitely failed solutions.  I know that
 > one of my colleagues here, Tracy Clarke, has experienced similar problems
 > when calibrating data with multiple IFs.  I have included her on this
 > e-mail, as she would be interested to know any thoughts you might have on
 > the issue.  Thanks for any help you can provide.
 > 
 > Carl Gross
 > Naval Research Lab
 > 
 > PS -- These issues are observed in AIPS versions 31DEC03 and 31DEC05.
 > 
 
I conclude that you are running older versions of the AIPS code - the
fact that some dates in your 31DEC05 came after the fix worries me.  I
suspect that something went wrong with your MNJ and you do not
actually have a complete version to the dates you found.

Note - I did try another test flagging 10 antennas in one
polarization.  In that case failed solutions were reported in all
versions (= 10 * number of solution times).

I suggest that you get a current version of 31DEC06 - we have added
all sorts of goodies to aips recently.

Eric Greisen