[evlatests] Phase flips; frequency averaged phases.

[Vivek Dhawan]

A consultation with Eric has cleared this up - a red herring,
unrelated to the WIDAR troubles. I thought it worth documenting,
as the testing is getting to where the optical depth of software
is significant.

The effect described  - the model not following the data from
which it was made - is not a frequency scaling effect but a UVW
computation side-effect.

When one plots data with clean-component model, there is a switch
nmaps +1, "plot model only where data exists"    or
nmaps -1, "plot model everywhere"

nmaps +1 uses the UVW from the data.
nmaps -1 computes UVW approximately, not using the full routine in
          UVFIX. Hence the apparent discrepancy in phase and closure
          phase between the model & data.

No fix is planned, just user caution.

Vivek Dhawan wrote:
> I was wondering, based on evidence below, whether the scaling
> of phase with frequency offset within a subband was correct -
> perhaps I'm seeing evidence of the same thing Rick suspects.
> 
> Test data: Cyg A at C band, July 5, calibrated & imaged by Rick.
> 
> The intent was to check whether the data and model had the same
> sign of phase offset and phase slope from the offset double source.
> In short, they do on every baseline, so the phase flip is truly
> global and not sneaking in on some baselines that then mess up
> the image.
> 
> In the same data, it is clear that:
> 
> 1. the model follows the data pretty well, but often misses by
> tens of degrees near the many nulls from the double source.
> 
> 2. The 4 IFs are spread over ~500 MHz, each IF (subband) selected
> for imaging was 1.1MHz (9 channels wide) from the 128MHz.
> The data show the change of structure with frequency, the model
> also changes but still misses the date as above. The signal is
> very strong so this is a many-sigma discrepancy.
> 
> 3. The closure phases of model/data are quite discrepant, in some
> cases by upto 180 deg. I speculated that for some reason, say mislead
> by the header information, the scaling of model with frequency is not
> done right. I was still working on this, but Rick & Barry's message may
> be pointing in the same direction so I thought it worth putting out
> before I have a conclusion.
> 
> Vivek.
>