[daip] more NX table problems

[Eric Greisen]

Lynn D. Matthews wrote:
> I'm running into more problems that I think are linked with the new NX 
> tables created by SPLIT.
> 
> I have SPLIT out a single channel to do a self-cal sequence, 
> and then run a P-only CALIB with 10-s SOLINT using a CC model. After 
> one or many iterations, I see that the phases look good for the first half 
> of the data, then are noisy for the second half.
> 
> Initially I thought that something went wrong during the second half of 
> the observations, but then I found the exact same behavior for comparable 
> data taken on different dates. If I delete the NX table, this problem goes 
> away. In addition, without the NX table I go from getting, e.g., 1439 good 
> solutions, 9 failed, to 1978 good, 10 failed.

CALIB does do different things in the presence of an NX table.  In 
particular, if a scan is e.g. 11 min long and you do a SOLINT of 5 min, 
the with no NX table you get a 2 5-min soln and 1 1-min one - the latter 
would be rather noisy.  With an NX table, CALIB will break the scan into 
intervals that are equal - I suspect 5.5 min in the example case.  But a 
12 minute scan might go to 3 4-min solints (I have not looked at this 
code in some time).  That might account for poorer solutions but I do 
not see why - certainly fewer solutions is expected with the NX table. 
It is my opinion that having an NX table - if it is correct - is always 
better.  I suppose that one might confuse things in the automatic NX 
table (much the same algorithm as INDXR but no user parameters offered) 
and end up with a bunch of short scans.  CALIB will not go outside scan 
boundaries and so could end up confused.  Study your NX table and the 
times that you get solutions to see how this aligns with what you are 
seeing.

I will not remove the auto NX tables unless they are really clearly a 
bad thing so we need to understand this.

Eric Greisen