Guidelines for post-processing EVLA data in AIPS
Since August 1, 2006 EVLA antennas have been included in observations by default. During the post processing of any data including EVLA antennas there are a number of caveats observers should be aware of. In any case we strongly recommend users to use the latest version of AIPS (31DEC07), including running the AIPS midnight job regularly

EVLA antenna positions

Since their inclusion in the array by default, we have obtained improved EVLA antenna positions over time. This means that the position of the EVLA antennas (and VLA antennas, for that matter) in your data may not be the best available one, which can lead to phase errors. Especially for EVLA antennas, this error may be substantial. The AIPS task VLANT determines the antenna position corrections applied by the VLA operations staff after your observation that apply to that observation. It copies an input CL table and applies these antenna position corrections to the antenna phases. It is important that VLANT be run on CL table version 1 immediately after running FILLM, and prior to any other operation on the CL table. Note that since VLANT is a new task you must have 31DEC07 AIPS to run it.

Closure errors - continuum data

For all continuum observations we strongly recommend frequent observations of a baseline calibrator in order to be able to correct for closure errors on EVLA - VLA baselines caused by non-matched EVLA and VLA bandpass shapes.

To determine the baseline solutions, follow the following procedure:

  1. Remove obvious bad data from the baseline calibrator via the usual programs. It is not necessary to know the flux density of the source
  2. Run CALIB on the baseline calibrator source. Check the solutions for any residual errors
  3. Run CLCAL, specifying the baseline calibrator source as the calibrator. Use the defaults (2-point interpolation)
  4. To check that all is well so far, run UVPLT on the Stokes 'RR' and 'LL' visibilities, specifying the baseline calibrator source, and applying the calibration (DOCAL=1). One should see two vertically separated superimposed plots, the top one near 1.05 Jy, the bottom one near 0.95 Jy. (The high visibilities are EVLA-EVLA, and VLA-VLA baselines. The low visibilities are the VLA-EVLA baselines)
  5. Run BLCAL with the default settings. WARNING: If you are doing polarization calibration and imaging, see the note below. Specify DOCAL=1, which applies the CL table to the data prior to determining the baseline based corrections. Use SMODEL = 1,0 if the flux density was not specified prior to the CALIB step above. If one chooses to specify the calibrator source flux density, that value must be inserted in SMODEL. One can check whether the procedure worked by re-running UVPLT, again with DOCAL=1, and this time also specifying BLVER=1. One should now see a single horizontal bar, with flux density = 1.0
  6. Delete the SN #1 and CL #2 tables made above, using EXTDEST
  7. Proceed with normal calibration, and ensure that BLVER=1 for all further steps
Note:If you are doing polarization calibration and imaging, running BLCAL may cause a mis-match in the parallel-hands and cross-hands that can break PCAL and polarization imaging. If the BLCAL changes are small and mostly in amplitude, and the following PCAL seems to work, then you are probably OK. The problem occurs if the adjustments in amplitude are large or there are substantial phase corrections, as they are not propagated to the cross-hands.

Closure errors - spectral line data

As advertised, 50 MHz continuum observations show closure errors of around 8% on VLA - EVLA baselines due to non-matched bandpass shapes between VLA and EVLA antennas. This error becomes much more severe at narrower bandwidths, and we recommend observing in spectral line mode for those bandwidths, in which case an initial bandpass calibration can correct for the non-matched bandpass shapes.

It is common practice to use 'channel 0' as created by FILLM in the initial calibration stage. On data with VLA - EVLA baselines this practice is strongly discouraged since the 'channel 0' data now contain closure errors on VLA - EVLA baselines. Our recommended solution is to create an alternative 'channel 0' from which closure errors have been removed. There are two ways to do this: an easy one which requires an AIPS midnight job (MNJ) after October 1, 2007, and a more complicated one for older versions of AIPS. The difference is that in the latest versions of AIPS the task AVSPC allows application of calibration, which reduces the number of steps required.

  • AIPS with no MNJ after October 1, 2007

    1. Run BPASS on the LINE multi-source dataset, creating a bandpass (BP) table. Make sure BPASSPRM(5)=0 (the default)

    2. Apply this BP table using SPLAT, creating a multisource output file. In SPLAT, make sure to specify DOCAL=0 (which causes CL table version 1 to be copied to the output file) and all sources

    3. Add an index (NX) table to the SPLAT output either by copying the NX table from the SPLAT input to the SPLAT output using TACOP, or by running INDXR on the SPLAT output

      The first option is easiest, but requires an equal number of visibilities in input and output; it fails when application of the BP table in SPLAT leads to visibilities being flagged

      The second option is tricky when pointing scans are included: standard application of INDXR will merge calibrator scans at either side of a pointing scan into one, with a possible phase jump within that scan. In that case, scan boundaries need to be explicitly specified in INFILE (see EXPLAIN INDXR)

    4. Use AVSPC to average a number of spectral line channels together into an alternative 'channel 0' dataset (FILLM uses the central 75%, which is also the default in AVSPC, but other choices over a flat section of the bandpass are fine, too). Channel specification in AVSPC is flexible allowing the user to omit channels contaminated by line emission or absorption in the calibrator passband

    5. Continue the calibration as one normally would (i.e. run CALIB etc) on the newly created 'channel 0' data

  • AIPS with MNJ after October 1, 2007

    1. Run BPASS on the LINE multi-source dataset, creating a bandpass (BP) table. Make sure BPASSPRM(5)=0 (the default)

    2. Use AVSPC to average a number of spectral line channels together into an alternative 'channel 0' dataset (FILLM uses the central 75%, which is also the default in AVSPC, but other choices over a flat section of the bandpass are fine, too). Channel specification in AVSPC is flexible allowing the user to omit channels contaminated by line emission or absorption in the calibrator passband. Apply the BP table using DOBAND=1 and make sure to specify DOCAL=0 (which causes CL table version 1 to be copied to the output file) and all sources

    3. Since application of the bandpass may flag some visibilities, the total number of visibilities in the alternative channel 0 data may be less than that in the LINE multi-source dataset. We therefore recommended to create an alternative LINE dataset as well using the task SPLAT, applying the same BP (and, if applicable, FG) table as in AVSPC

    4. Continue the calibration as one normally would (i.e. run CALIB etc) on the newly created 'channel 0' data, and copying the solutions to the alternative LINE data

These prescriptions become more complicated when taking the aliasing problem into account. In that case, we recommend flagging all EVLA - EVLA baselines prior to any calibration determination, and unflagging them afterwards, prior to applying the calibration.

So in the second case above, flagging has to take place before the bandpass determination, and unflagging right afterwards, before application of the bandpass in AVSPC and SPLAT. Then, renewed flagging and unflagging is required before and after CALIB.

Doppler tracking

After the Modcomp retirement Doppler tracking works again for all baselines. However, the potential of phase jumps on VLA - EVLA baselines, caused by the small frequency changes, remains. Our recommendation not to use Doppler tracking therefore remains in effect. If you did manual Doppler tracking (i.e. calculate the appropriate sky frequency for every individual scan), no resampling in frequency is needed. In that case you should have put calibrator scans around each scan on source to deal with the phase jumps associated with frequency changes on VLA - EVLA baselines.

During the transition, online Doppler tracking is not recommended in many cases, for a variety of reasons. In that case, you probably observed all or most scans with the same frequency, and you may need to resample your data to a common frequency/velocity axis using the AIPS task CVEL, and enter the velocity manually using SETJY. Note that if the line width is narrow compared to the channel spacing, resampling may result in unwanted sidelobes in frequency (Gibbs effect), and off-line Hanning smoothing may be needed to suppress these sidelobes at the price of a loss in frequency resolution.

Removing 25 MHz spectral line artifacts

Data taken in spectral line mode at 25 MHz bandwidth between August 2006 and November 8, 2007 and suffering from the correlator offset problem can be corrected during post-processing as long as there is no significant emission in the center of the field.

To this end the AIPS task UVMTH, written in 1989 and then mostly forgotten, was resurrected. UVMTH time-averages the visibilities in a uv data set on each baseline and correlator and adds them to, subtracts them from, multiplies them by, or divides them into the visibilities in another data set. It was recently revised to apply all of the data selection, calibration, and flagging options to the first input uv data set while averaging each baseline and correlator over all TIMERANG. The result is applied to the second data set which is read, modified, and written with no application of data selection (other than sub-array and frequency ID), flagging, or calibration. For the VLA correlator offset problem (and a similar one sometimes seen in GMRT data), the user should specify the same data set for both inputs. SOURCES should select those pointings with no significant source at the center of the field and FLAGVER should select flags for at least these sources, but all calibration should be turned off. The default subtraction operation should be selected.

Ka-band Amplitude Calibration

We currently don't have models of the various amplitude calibrators at Ka-band. Instead, models at K or Q band may be used. See here for more details.




Modified on Tuesday, 29-Jan-2013 13:57:47 MST by Gustaaf van Moorsel