[evlatests] Polarization calibration for separated 3-bit observations

[rperley]

In June, I attempted polarization calibration on a database made up from 
combining many separate short observations, each made with an identical 
SB, in 3-bit mode.  The polarization calibration failed miserably.

The cause was soon found (by Vivek and Ken) to originate in the setup of 
the individual SBs:  Contention between setting the slope equalizers and 
the T304 digital gains resulted in an essentially random phase offset 
for every antenna, for every individual observation.  With no antenna 
with the same (R-L) phase, the usual procedures for polarization 
calibration fail.

This unhappy state of affairs can be handled in calibration *if* each 
individual observation as both an unpolarized calibrator (for antenna 
calibration), *and* an observation of a strongly polarized source (to 
set the R-L phase).  In the tests referred to above, I had neither.

But if each individual SB has an observation of 3C286 (or another 
strongly polarized source of known PA), we should be able to set the R-L 
phase differences in advance, without the need for having corrected for 
the antenna polarizations first.  This should work because the 
fractional polarization of 3C286 is much greater than that of any of the 
antennas.

By happenstance, I have been able to confirm this procedure works.  I'm 
working with four SBs, each an hour long, taken in May of some GRB.  
Each has an observation of 3C286 for flux calibration, but no inclusion 
of an unpolarized source.  The local calibrator has 2 -- 3% 
polarization, so is unsuitable for the usual zero-polarization 
determination of the antenna polarization.  Simply combining the 
databases, and executing the usual polarization calibration, gave 
bizarre solutions (as expected).

The procedure which worked (well) was:

1) Calibrate each SB for parallel-hand gains, as usual.
2) Using 3C286, determine the cross-hand delays as usual, then apply the 
R-L phase determined by the mean phase difference between R and L 
channels, without attempting an antenna polarization calibration.
3) Then combine the databases (DBCON).
4) Run PCAL, to determine the antenna polarizations.
5) Again determine the R-L phase (using 3C286), and apply.

This resulted in an excellent solution, as judged by the determined 
fractional polarization of 3C286 and the nearby phase calibrator.

I had to do some 'AIPSian' tricks in the above, as it is important that 
the SN and CL table align properly when combining the databases.  No 
doubt the details will change if attempted in CASA.

-----------------------

Ken informs me that the underlying procedures for setting the slope 
equalizers and T304 attenuators will be changed soon (perhaps already 
done?).  He expects R-L phases to be much more stable between individual 
SBs -- but of course we'll need to test this once the changes are in.



On 2021-08-10 15:58, Ken Sowinski wrote:
> Yes, I remember the discussion.  The problem was that the 3-bit
> setup sequence did not consistently choose equalizer settings
> so at each iinvocation of an SB there could be a different R-L
> phase diffrence for each IF pair and antenna.  When the fix is
> in place I expect that we will be in the converse situation,
> that is only rarely will the R-L phase difference change between
> SBs.  Of course this must be veified by experience.
> 
> I have just pinged Bruce to see if the s\change is in place and
> if not when we might be able to do it.
> 
> Ken
> 
> 
> On Tue, 10 Aug 2021, rperley wrote:
>> Ken:
>> 
>> You will recall this discussion -- and its effect on polarization 
>> calibration.
>> 
>> Daniel's new GRB data, taken in May, has given me the opportunity to 
>> test how to calibrate such data.  The difference between his (Ku-band) 
>> datasets and mine is that he included 3C286 in each of the separate 
>> SBs.  However, because he did not anticipate doing polarization 
>> calibration, he did not include an unpolarized source.
>> 
>> It turned out that the four SBs spanned a reasonable range in 
>> parallactic angle, so that a 'traditional' polarization calibration 
>> might be possible, where individuals one would not, due to lack of 
>> angle coverage.
>> 
>> Simply combining the four straightaway did not work -- the 
>> polarization solutions were nonsense.  The issue of course is the 
>> 'random' R-L phase changes between the four databases.
>> 
>> So I then 'pre-rotated' the R-L phases using 3C286 -- without 
>> previously applying any polarization calibration.  This is in 
>> violation of the accepted procedures -- rotation is the final step, 
>> after antenna polarization correction.  However, since 3C286's 
>> cross-pol is larger than any single antenna, I expected the process to 
>> work.
>> 
>> And it did -- beautifully.  After pre-rotation, then normal 
>> polarization calibration, everything lined up.
>> 
>> So -- the question:  What difference will the changes you propose make 
>> to solve the problem?  Will there be any assurance that we can combine 
>> databases without the preconditions of having (at least) 3C286 in 
>> every database?
>> 
>> Rick
>> 
>> PS  After I hear back from you, I plan to send this report around.
>> 
>> On 2021-07-01 13:04, Ken Sowinski wrote:
>>>  On Wed, 30 Jun 2021, Ken Sowinski wrote:
>>> 
>>>>  On Tue, 29 Jun 2021, Rick Perley wrote:
>>>> 
>>>>>  I suppose a question that should be asked is:  Why do the slope
>>>>>  equalizers
>>>>>  change so much, for identical observations taken consecutively in 
>>>>> good
>>>>>   weather conditions?  I'm sure the system is stable enough that 
>>>>> the
>>>>>   spectral shapes should be very similar.
>>>> 
>>>>  I also wondered about that in my response to your first email.
>>>>  The explanation is in the plot that Vivek sent around yesterday.
>>>>  The servo loop is simultaneously trying to optimize T304 output
>>>>  power and bandpass slope.  However the two knobs we have 
>>>> (attenuator
>>>>  and equalizer) are not independent; turning one knob affects both
>>>>  measured quantities.  So, often the system winds up alternating
>>>>  between two equally good states until the setup scan ends.  As a
>>>>  result the final state is a matter of chance rather than certainty.
>>>>  We good try to make a smarter servo loop that knows how the two
>>>>  knobs interact, but it is not clear that it is worth the effort.
>>> 
>>>  I have just had a careful look at the arithmetic of determining
>>>  the slope and how that slope is used to implement the servo loop.
>>>  My initial impression is that while the facts above are mostly
>>>  correct the conclusions are not.  I suspect that the servo loop
>>>  'dead band' is much too small and so the servo oscillates rather
>>>  than converges.
>>> 
>>>  I will go over the arithmetic once more to be sure I am right and
>>>  then test my hypothesis if we ever have  powered correlator again.
>>> 
>>>  Kwn
>> 
>>