Polarisation of 3C138 // J0521+1638

[AstroRipples]

Hi fellow MeerKAT polarisation enthusiasts,

As foreshadowed in my comment on #1604, I'm opening a new issue to document my experiences and discuss the polarisation of 3C138 // J0521+1638 with MeerKAT. This is primarily in the hope of helping others, but also to help myself, as I'm seeing some strange behaviour in some MeerKAT CBIDs, but not others.

I have both L- and S-band data from different days across several years that I'm working with, but in the plots below I'll be showing data from the following CBIDs:

L-band datasets:

• 1639674087 (2021-12-16; 4k wide)
• 1639848679 (2021-12-18; 4k wide)
• 1663451956 (2022-09-17; 32k wide, averaged to 2k before calibration)
• 1672429879 (2022-12-30; 32k wide, averaged to 2k before calibration)

S-band (S1) datasets:

• 1688875155 (2023-07-09; 4k wide)
• 1688961378 (2023-07-10; 4k wide)

To repeat before we go any further, the polarisation calibrator was always 3C138 // J0521+1638 for these datasets.

All L-band data were processed with CARACal following the standard pipeline, whereas the S-band data were processed manually in CASA using the workflow I used during my PhD for polarisation calibration of KAT-7 data. These are the same principles followed by CARACal, and tracks with other MeerKAT polarisation calibration documentation.

Despite the ~18 month gap between the L- and S-band 4k datasets, the polarisation of 3C138 lines up really nicely, as you can see below :

Figure showing the data for CBIDs 1639674087 and 1688961378:

Figure showing the data for CBIDs 1639848679 and 1688875155:

Again, the left panel shows the Stokes I flux density (left-hand axis) and linearly-polarised flux density (right-hand axis), centre panel shows the polarisation fraction, and right panel shows the polarisation angle, measured in 64 channels across the MeerKAT L- and S-bands. The orange datapoints show the L-band measurements reported in Taylor & Legodi (2024), whereas the shaded region and horizontal line show the "expected" values reported in Table 1 of the MeerKAT polarisation calibration wiki page.

So far, so good ...

Where I'm now having trouble is with the 2022 observations, taken as 32k wide data and averaged to 2k channels (2048) during download with mvftoms.py. In each of these CBs the following calibrators were observed:

• J0408-6545 (bpcal, fcal; three 10-min scans)
• J0531+1638 (xcal; one 10-min scan)
• J0453-2807 (gcal; 2-min scans every half-hour)

Calibrating with CARACal following the same configuration file, I get wildly different observed polarisation properties for 3C183:

Figure showing the data for CBID 1663451956:

Figure showing the data for CBID 1672429879:

As you can see, neither of these CBs show the correct polarisation properties (%pol, pol.ang.) for 3C138. In each case, the HV power was above the threshold of ~0.14 Jy: for 1663451956, ~0.55 Jy and for 1672429879, ~0.2 Jy based on the plots in the OPT. These observations also took place completely during the night, from ~22:00 to ~04:00 UTC and ~19:50 to ~02:00 UTC, with the scans on J0531+1638 occurring around ~01:00 and ~22:50 UTC respectively. As such, it surely can't be an ionospheric related issue.

I have exhausted just about every possible avenue of investigation for determining the cause of this discrepancy without resolution. Additional flagging doesn't change the plots seen above, changing the time solution interval doesn't change the results. At this point, the only difference I can see is that these two CBs only have J0408-6545 for bandpass/flux and leakage, whereas the others have both J0408-6545 and J1939-6342, and I'm using J1939-6342 for the leakage calibration. J0408 is unpolarised, however, so I don't see why it could be causing issues with an incorrect leakage solution.

If anyone out there can provide suggestions on what to try next here, that would be great. I haven't yet figured this out, but I feel like it should be more straightforward than this.

[mattieudv]

I tried to do polarisation calibration for your two problematic 32K datasets using a module I developed.

In the figures below for each dataset (1663451956 and 1672429879), 3 results are shown - using different scans of the unpolarised target J0408-6545 that was observed at the beginning (scan 0), middle (scan 26) and end (scan 50) of each dataset. No instance of J1939-6342 is found in either dataset.

Scan 24 (in both datasets) is of 3c138, the polarised calibrator being investigated.

My result for CBID 1663451956 is below:

And for CBID 1672429879:

I do not see an obvious cause of problems for 1663451956, but note that the gain calibrator J0453-2807 (not used by me) is 2-4% polarised. Using only a simple skymodel for J0408-6545 might deteriorate the accuracy. The applicability of transferring the flux and leakage solutions to quite different parts of the sky may play a role too (the unpolarised and polarised targets are far apart).

In the case of 1672429879, there is a more obvious problem that happens because the feed axis aligns with the measured EVPA angle at around 950MHz, limiting the telescope's ability to measure the polarisation properly. The ability to measure linear polarisation improves as the measured EVPA angle is further away from the feed's dipole orientations (parallactic angle). If a EVPA aligns with a feed axis then only the one of the two polarisations is measurable at that frequency.

[mattieudv]

Some more feedback about the initial 'good datasets' in 4K mode:

My corresponding results for CBIDs 1639674087 (L) and 1688961378 (S) is in the figure below. Note that the L-band observation is marked as a fail by quality assessment, presumably because it was interrupted by a wind stow. The wind was averaging 9m/s at the time of observing 3c138 which is not regarded as optimal night time conditions, and this increased to 12m/s gusting at 14m/s at which point the antennas were stowed.

My results for 1639848679 (L) and 1639848679 (S) shows two pairs where in the one case PKS1934-63 is used as bandpass calibrator and in the other case PKS0408-65. Differences in the flux models of these accounts for the discrepancies, and also I think PKS1934-63 is slightly polarized, and also I only use a simple one element skymodel here.

Some more info on the 32K observation 1663451956 (1672429879 mainly affected by 3c138's position angle aligning with feed axis):

Regarding 1663451956, the delay calibration is less relevant as it was done more than an hour earlier using a different calibrator (PKS1934-63) than what is available in the dataset used as bandpass calibrator. Relatively poor phase stability is noted. For example baseline m020h-m060h shows slight modulation over time of 28 degrees over 10 minutes in the PKS0408-65 track, in addition to a residual delay calibration error of 35 degrees over the bandwidth. This compares to 10 deg phase instability over 10 min and 20 deg over bandwidth, unnoticeable time-modulation, for 1672429879.