[daip] ACCOR question

[Leonia Kogan]

Michael,

First of all I am sorry to answer you with the delay.
I was very busy with something.

Your result looks very strange for me. ACCOR does not do anything 
different but averaging the autocorrelation during the given SOLINT and 
record the result into the new created SN table.
The averaging is provided taking into account the weights.
So the average result may be changed when you exclude some visibilities.

But I do not understand what happens with autocorrelation vis if you 
remove (by UVCOP) visibilities with weights<0.3. The autocorrelation vis 
are deleted if their weights<0.3 Are they? I'd recommend you to look at 
the weights of your autocorrelation vis.

I'd use the SN table you got after UVCOP....?

Leonia

Michael Bietenholz wrote:
> Hi Leonia - I hope you are well!
> 
> I'm asking you this because I think you wrote ACCOR.
> 
> I've just noticed something a little odd with some results from ACCOR.
> I ran ACCOR as normal on a VLBI data set (continuum; L-band; VLBA +
> GBT + EB; new correlator; 31DEC11 AIPS)
> 
> I got SN table amplitude corrections near mostly between 0.8 and 1.2,
> as expected.
> 
> Then I ran UVCOP to loose any visibilities with weights <0.3 (UVCOPPRM
> 0 0 0 0 0.3 0: keep autocorrelations).  Only a small fraction of the
> total number of visibilities (0.07%) were lost.
> 
> I would have expected that ACCOR, when run on this slightly smaller
> UVCOP'ied data set, would return almost the same answer (since we've
> only very slightly changed the visibility data set its working with).
> I found, however, that there are quite systematic differences in the
> ACCOR solutions, which are now about a few % lower (the before/after
> UVCOP amplitude difference is quite consistent througout the run).
> 
> So - my question is; which value to I believe?  I know the difference
> is relatively small.  Still the largest difference in this case is for
> the most sensitive antenna, GB, and is 10% which is getting big enough
> that it might have a sensible effect on the results.
> 
> There must be something quite systematic going on here in that the
> low-weight points have very systematically high/low autocorrelation.
> 
>                   cheers,       michael