A Cyberinfrastructure platform to meet the needs of data intensive radio astronomy on route to the SKA

GMRT Deep Polarization Field

GMRT Deep Polarization Field

GMRT Deep Polarization Field

  • Public

12 January 2011

Last updated 3167 days ago by Russ Taylor

We were given a few hours of time to test the 32 MHz full polarization mode, since to this point it has never actually been used for science observations.  I set up an observation of 3C286 and our polarization calibrator 1549+506.  If all went well this will be a good obervation to derive the polarized flux density of 3C286 at 610 MHz and to test the polarization calibration process in casa.

Importing the data

Read the uv fits file created with gvfits (see notes on main page) into casa using importuvfits.  This does not apply the on-line flags, but when I tried importgmrt and specified the flag file, 100% of the data was flagged.   importuvfits seems to work fine, although casa thinks that there are 32 antennas. 

The observations were taken at 581 MHz (channel 1) so the centre frequency is 597 MHz.  There seems to have been a mix up in the LO setting.  The centre frequency was supposed to be 616 MHz.   Apparently the first LO was set to 530 MHz when it should have been 545 MHz.


Used viewer to look at the measurement set.  The data in general look pretty good.  There are two antennas missing. And there is some rfi over some ranges of channels and some times.   Channel 0 has very high values (200 Jy), so I will remove this from the subsequent analysis.

Warning.  When manual flagging data using flagdata note that the antenna number to be entered should be the antenna number displayed in viewer +1.  So if the viewer says that antenna 13 is bad.  Then enter antenna = '14' in flagdata.

Removed antennas 11 and 14 (numbers used in flagdata) as these were showing very amplitudes in viewer (displayed as antennas 10 and 13).  

Then ran autoflagging using flagdata2.  This has to be run four times for each sources with rfi_clipexpr set to each of 'ABS RR', 'ABS LL', 'ABS RL' and 'ABS LR'.  I think there is a way to do this in one go using python lists but I haven't taken the time to figure that out yet.  I ran it initially using the default rfi_time_amp_cutoff = 4 and rfi_freq_amp_cutoff = 3.  But this  flagged lots of data that looked ok to me.  Reset to rfi_time_amp_cutoff = 8 and rfi_freq_amp_cutoff = 5.  This looks like it did a good job.  The obvious rfi visible in viewer was pretty much all detected and flagged and not much else was touched.

Polarization Leakage

Ran the data through gmrtcalib to derive polarization leakage using 1549+506.  Here are images of the leakage amplitudes versus channel and frequency. Click on an image to view the fits files directly.

image image

    The grayscale parts of the images are below 10%.  The amplitudes are generally below 10-20% with a few high points that get to 30% (red starts at about 30%).  Exceptions are antennas 29 and 30.  Antenna 30 is very high (of order 100%) over the whole band,  Antenna 29 gets very bad (70%-80%) at the upper end of the band.
    These amplitudes are sufficiently high that the linear approximation for the D-terms is no longer a good approximation.

    After removing antennas 29 and 30 I calibrated 3C286 using 1549+506 as the polarization calibrator. The image below shows the spectral profile of 3C286 in each Stokes parameter.  No correction for the abolute position angle (R-L phase) has been made since 1549_506 is an unpolarized source so only the D-terms themselves are calibrated.


    Stokes I upper left, Stokes Q upper right, Stokes U lower right and Stokes V lower left.  Stokes Q and U show the oscillation due to the instrumental rotation of the position angle with frequency (the instrumental rotation measure?).  Is some of this ionospheric?  Surprisingly there is very strong signal in Stokes V changing linearly across the band.  The error is zero at band centre and grows to about 2% of I at the band edges. Since the V signal is derived from RR - LL, an error in the R and L gains will produce a signal in V.  Note also the slope in Stokes I which is RR +LL.  It looks like the R and L gains are correct only for the centre of the band and have an error that grows linearly to +/- 2%.  Where does this come from?

    Gain calibration is a single correction over the band.  It is assumed that the bandpass correction removes frequency-dependent co-polar gain effects.  Is this an issue with the bandpass calibration? 

    Ran an X-pol solution using 3C286 as the position angle calibrator.  Resulted in the following spectral behaviour.  Stokes Q and U now are relatively constant across the band, although there are sizable variations of order 20% with frequency.