[evlatests] Cross-hand phase convention...

[George Moellenbrock]

All-

On a whim, I temporarily reversed the sign of the parallactic angle phase 
correction in CASA and re-applied the calibration to the C1310 dataset. 
This had two gratifying effects:

1. The rotation of R-L phases (i.e., 2X the linear polarization position
    angles) in 3C286 and 1310+323 during the 75-min observation falls to
    zero.  I.e., no residual rate.  This is not especially surprising since
    applying the parallactic angle correction with the opposite sign (cf
    data) doubled the rate.  Reassuring, nonetheless.

2. More interestingly (and subtle), the error in the _difference_ in the
    these two sources' R-L phases (i.e., 2X the difference in their lin pol
    pos angles) relative to POLCAL results from one day later falls to ~1
    deg, well within the reported POLCAL errors (~5 deg).  The larger error
    in this difference noted yesterday (~10 deg) for inverted parang phase
    sign (relative to data) is thus indeed a result of the effective
    doubling of the par ang ang difference between the two sources, which
    are relatively close on the sky (4.7 degrees apart).  The observed poln
    pos ang difference between the sources still has the incorrect sign
    because I haven't bothered to globally reverse the visibility phase
    sign in my dataset*.

*I did not recalculate all of my other calibration, all of which is solved 
in the phase frame of the (mirror-universe) telescope.  I.e., all are 
independent of parallactic angle, regardless of sign, so I did not have to 
recalculate it.  Had I PHNEG'd, I would have had to conjugate all of my 
other calibration too, or re-solved.  Had I used a polarized source for 
the D-term solution (and required a solution for its polarization), I 
would have had to resolve that term.

It bears noting that while 3C286 provides enough SNR to clearly see the 
few degrees of parallactic angle rotation in an average over baselines and 
per-subband bandwidth (probably sufficient to tolerably separate source 
and instrumental poln at ~100 MHz resolution), I don't think there is 
enough SNR _per_channel_ (at the native 0.5 MHz resolution) for a reliable 
freq-dep D-term solution, and part of the aim of this reduction was to see 
the D-term freq-dep, which is considerable.  Some intermediate frequency 
averaging might be a nice balance between sensitivity and resolution for 
the purposes of D-term calibration using short observations of 3C286.

So, I reiterate that if the net effective _relative_ wiring/organization 
of parallel- and cross-hands is fundamentally the same in current WIDAR 
and WIDAR0 (I appreciate that there has been lots of "rewiring"), and if 
the recent phase sign flip in the CBE has been implemented uniformly for 
parallel- and cross-hands, the EVLA should be able to support accurate and 
effective freq-dep D-term calibration transfer for linear polarization, 
and position-angle calibration transfer with a precision of ~1 degree or 
better.

I'm looking forward to testing this claim with actual WIDAR data!  (The 
requested dataset will have good par ang coverage and thus permit 
quantifying the accuracy of the D-term calibration by providing solutions 
from multiple sources.)

-George



On Thu, 25 Feb 2010, George Moellenbrock wrote:

>
> Folks,
>
> In Sept, we observed 3C286 and some nearby calibrators at C-band in full 
> polarization (C1310plus.55098.blahblah, Sep 24) with WIDAR0.  I reported on 
> the polarization calibration (in CASA) of this data in a test meeting in 
> Octoberish, illustrating the strong frequency-dependence of the instrumental 
> polarization, differences between old and new C-band polarizers/receivers, 
> and the effectiveness of deriving the D-terms (channelized!) and correcting 
> data with them, including transfer among sources.
>
> Recall that the _net_ phase of the WIDAR0 parallel hands had the incorrect 
> sign at this time, and so Stokes I images (with sufficient structural detail 
> to tell) came out reflected through the origin. This _polarization_ dataset 
> is a good and more subtle test of the phase sign convention, at least for the 
> state of the correlator, etc. at that time.  This is because the cross-hand 
> phase for a polarized source should show the 'correct' evolution with time 
> (where 'correct' here means being consistent with the sign convention for 
> parallactic angle phase correction in CASA or AIPS). Also, the RL phase 
> differences between sources of known polarization position angle should be 
> consistent with prior knowledge.  If the cross-hand phase has the wrong sign, 
> then the source poln p.a. will rotate in the wrong direction, and 
> polarization imaging that integrates over significant parallactic angle 
> coverage will de-cohere. (An unpolarized source, or a snapshot of a polarized 
> source [no parang coverage], will image coherently, but suffer the same image 
> reflection that the Stokes I image does, assuming all correlations are 
> 'organized' the same way through the correlator and labelled correctly in the 
> output.  We've confirmed most of the latter via good bandpass corrections of 
> the cross-hands.)
>
> For my reduction, I did NOT negate the visibility phase a la PHNEG. The 
> D-term calibration relied upon an unpolarized source because, despite about 
> 75min of data, it was observed at an hour angle far from transit that yielded 
> very little parallactic angle coverage on 3C286.  However, I've had a closer 
> look and there is, in fact, sufficient parallactic angle coverage (and SNR) 
> on 3C286 (11% polarized) to detect the parallactic angle rotation in the 
> cross-hand phase.  The cross-hand phase of 3C286 calibrated for everything 
> but the parallactic angle phase (bandpass, delays including R-L delay [these 
> were quite large then], complex gain and D-terms) shows about 4 or 5 degrees 
> of monotonic variation.  When the parallactic angle correction is also 
> applied, the phase rate doubles, rather than going to zero.  Similarly, the 
> position angle difference between 3C286 and 1310+323 has the incorrect 
> sign---note that the small parallactic angle variation enabled a reasonably 
> coherent average of these data over time, even with the spuriously doubled 
> parallactic angle phase rate.  (I recall reporting in Oct that the difference 
> had the ~correct magnitude, but perhaps being a bit vague about the sign.

> Today, I am less impressed with the magnitude of the difference.  The 
> difference is off by
> ~ 10 degrees cf POLCAL results of the ~same epoch---this is probably due
> to a modest difference in the parallactic angles of these sources that gets 
> doubled by the spurious parang correction.  But I think the difference is 
> close enough to be reasonably confident about the _sign_ of the difference. 
> In any case, the p.a. phase-rate double is the better evidence.)  These facts 
> do not change the conclusions about D-term
> calibration effectiveness reported previously.  It worked.
>
> So, this shows that the WIDAR0 cross-hands required the same phase flip as 
> the parallel hands.  This should be confirmed for the current WIDAR.  So far, 
> only the (recently revised) parallel-hand phase sign has been confirmed for 
> WIDAR.  Michael expects soon to take some full polarization data on 3C286 and 
> its neighbors to re-confirm the parallel-hand phase sign, confirm the 
> cross-hand phase sign, confirm the D-term solution veracity, and confirm the 
> transferability of R-L phase among sources (i.e., the source poln. p.a. 
> differences).
>
> -George
>
>

--