[Difx-users] Handling lin-pol data with difx2fits

[Rick Perley]

     It is of course not necessary to assume either linear or circular 
systems!  The proper analysis is quite generic as to the state of the 
antenna polarization.  This manifests itself in the magnitude of the 
off-diagonal terms in the antenna Jones matrix. If the antenna receives 
linear, then it is useful to adopt a linear basis for the analysis, as 
the off-diagonal terms are then very small.  The same holds true for 
circular.  Small off-diagonal terms lets us make reasonable 
approximations in the subsequent 'Mueller' matrix (a bad term, but it 
has come into fashion to use it), simplifying the subsequent analysis 
and calibration.

     For a mixed circular/linear array, most, or all off-diagonal terms 
will be comparable to the diagonal -- I haven't thought much about this 
for a long time, but might do so now ...

     Converting the correlations from one basis to the other is 
certainly the most attractive-appearing way to handle this.  As noted 
earlier, I'm not sure of the ramifications in calibration (i.e., the 
practicalities of the matter).  I had done some thinking on this a year 
or so ago, but was distracted by other projects.  Perhaps I'll start 
working on this again.  The subject is a timely one for 'ngVLA' studies.

     Rick



On 04/12/2019 07:59 AM, Mark Kettenis wrote:
>> From: Rick Perley <rperley at nrao.edu>
>> Date: Fri, 12 Apr 2019 07:42:15 -0600
>>
>> Mark et al.:
>>
>> I don’t know anything about how CASA deals with this.
> It should handle linears since that's what ALMA has ;).  But I'm
> fairly confident most of the calibration code would fall flat on its
> face in the case of mixed XY/RL.
>
>> I’ve also thought about the post-correlation basis conversion
>> method, (and have analyzed this in one of my many unseen memos) —
>> but I would like to think a bit more about the consequences in
>> calibration.  (It’s obviously completely valid for pure linearly
>> polarized feeds, but not so clear for ‘real,dirty’dipoles).
> As far as I can tell Ivan does consider leakage in his paper:
>
>    https://arxiv.org/abs/1601.04266
>
> But I don't think this method has been tested on dipoles...
>
> Cheers,
>
> Mark
>
>> I’m cc’ing Eric on this.
>>
>> Rik
>>
>> Sent from my iPad
>>
>> On Apr 12, 2019, at 6:42 AM, Mark Kettenis <kettenis at jive.eu> wrote:
>>
>>>> From: Rick Perley <rperley at nrao.edu>
>>>> Date: Thu, 11 Apr 2019 13:58:31 -0600
>>>>
>>>>      All:
>>>>
>>>>      AIPS now does to linear polarization quite nicely.  See the EVLA
>>>> Memo (#207) that Eric Greisen and I just submitted.
>>>>
>>>> Many changes were needed to ensure the correct combinations, and order,
>>>> were made.  Eric can provide details, if needed.
>>>>
>>>>      Rick
>>> Hi Rick & others,
>>>
>>> Of course the assumption made in 4.1 that all antennas are identically
>>> oriented on the sky doesn't hold for a typical VLBI array.  If that
>>> assumption is reflected in the AIPS code then calibration and
>>> interpretation of VLBI data with linear polarisations in AIPS won't be
>>> straightforward.  And I believe the data structures still won't be
>>> able to represent mixed linear-circular observations.
>>>
>>> While the MeasurementSet in principle allows proper representation of
>>> such observations I don't think CASA can actually handle such data
>>> properly either.
>>>
>>> This is why I think that conversion from linear to circular after
>>> correlation (as implemented by Ivan Marti-Vidal in his PolConvert
>>> tool) is still the most viable strategy to handle mixed-polarisation
>>> observations, possibly including all-linear VLBI observations.
>>>
>>> Cheers,
>>>
>>> Mark
>>>
>>>>> On 04/11/2019 01:44 PM, Leonid Petrov wrote:
>>>>> Dear Walter,
>>>>>
>>>>>> It is true that the poltype is hard coded.  With some experimentation we
>>>>>> might be able to change that to X/Y as aspropriate.  At one point I
>>>>>> believe AIPS had trouble with other than R or L.  We can look into this
>>>>>> for DiFX 2.7.
>>>>>    If AIPS at the moment has troubles with other than R and L, we can
>>>>> ask Eric just to patch AIPS to treat X and Y or H and V as R and L on the flight
>>>>> temporarily until better time come, if it is necessary. It should not take more
>>>>> than 5 minutes. From the other hand, I got an impression that some folks
>>>>> already used AIPS for processing lin-pol data. Am I wrong? If yes, how did they
>>>>> do it?
>>>>>
>>>>>    I have 6 lin-pol experiments on my table and a hundred or so to come within
>>>>> weeks. We have to fix it now, i.e. today.
>>>>>
>>>>>> As noted above, the FITS format is unable to describe such data, so this
>>>>>> is not possible.
>>>>>    Although I heard a combinations of words "possible" and "not", I still cannot
>>>>> grasp what do they mean together. We need amend FITS-IDI, I understand it.
>>>>> So what? Even the US Constitution was amended, and more than once. Let us
>>>>> discuss how to do it. And we need fix difx2fits: today it tells that it does not
>>>>> like a combination of lin- and cir- data and stops. Yes, we already have the
>>>>> lin-cir and cir-lin data that need to be processed, and we plan to get more data
>>>>> like that. I already rolled my sleeves up and ready to do it. If you or anybody
>>>>> else can provide ideas and/or guidance, warn about pitfalls, contribute code, etc,
>>>>> I will appreciate it. But I cannot just do nothing and say: "these are the data that
>>>>> I cannot process". Failure is not an option.
>>>>>
>>>>> Sincerely,
>>>>> Leonid
>>>>> 2019.04.11_15:38:06
>>>> _______________________________________________
>>>> Difx-users mailing list
>>>> Difx-users at listmgr.nrao.edu
>>>> https://listmgr.nrao.edu/mailman/listinfo/difx-users
>>
>> _______________________________________________
>> Difx-users mailing list
>> Difx-users at listmgr.nrao.edu
>> https://listmgr.nrao.edu/mailman/listinfo/difx-users