Assessment and inspection of the gain solutions

                

1. If you incur substantial closure errors, then you should look carefully at the data for those times and baselines reported to try and find out what is wrong with it and flag it out; or just flag it out regardless. Note that the times reported by ATCALIB are the average times from the solution interval. ATCALIB has just written its solutions into the highest numbered solution (SN) table attached to the input file. The easiest way to examine these solutions is to plot the gains as a function of time with the task SNPLT. You should plot both amplitude and phase. You can plot the solutions directly on the TV, or plot the (PL) extension file created by SNPLT with MIXPL, TKPL, or TVPL.

   In amplitude, you should expect to see slow drifts in time of the order of up to %. In addition, at high frequencies and low elevations, there may be some atmospheric attenuation of the amplitudes, so that the inferred amplitude gain rises as the elevation decreases. The phase may wind more rapidly, being more subject to the effects of atmosphere and ionosphere. However, as a fiducial result, perhaps phase winds of tens of degrees over hours at 6 cm for the same source is not uncommon (you may do much better or worse).

   These plots should be examined for solutions that are obviously discrepant from the general trend. They may or may not have already made themselves known from the closure listings (but remember the times reported by ATCALIB or CALIB are the average times of the scan). Use UVPLT or VBPLT to plot all the data from the bad scan and hopefully the trouble maker(s) will be self-evident.

   It may happen that you find gain jumps. That is, an antenna suddenly jumps in gain (amplitude or phase) and then remains at the new level. Although these are not a good sign, they can be dealt with by careful calibration. Flagging them out is no good because the program source will have the same jumps, and you don't want to throw away too much program data. Note also that if a gain jump occurs in your reference antenna (all the antennas will then jump at one time instead of just one), then you should redo ATCALIB with a different reference antenna. Calibration with gain jumps will be dealt with in § 10.

   inext='sn' Plot gains in solution table

   invers=0 Plot highest table version

   sources=' ' Plot all sources

   stokes='I' Plot Stokes I gains or

   stokes='RR' RR or LL if polarization

   stokes='LL' trickery invoked

   freqid=2 Specify one freqid per run of SNPLT

   pixrange=0 Self scale plot

   ncount=6 6 plots per page (if 6 antennas say)

   xinc=1 Plot all solutions

   optype='amp' or 'phas' Plot gain amplitude or phase

   dotv=1 Plot directly on TV or

   dotv=-1 make plot extension file

2. ATCALIB will tell you if it was unable to find a solution for some time ranges. It will list the number of failed solutions. The ones that failed will not appear on your SNPLT plots. Rather than just writing them off as useless, it may be advantageous to find out why they failed. To do this, you to poke a bit harder into the system.

   You can print the SN table on the printer with PRTAB (put inext='sn' and see Appendix C. Under the columns listing the real and imaginary parts of the gain solutions, look for solutions called `INDE'. This means indefinite and is the code for failed solutions. You can then inspect the data at the relevant times (also listed in the SN table) to see if you can work out what failed. Possibilities are too much flagging and insufficient signal-to-noise ratios for CALIB to get its teeth into; if the S/N is too poor, CALIB will set the results to `INDE' (this cutoff level is set at 5 by ATCALIB but is an adverb in CALIB). Also, if minant is set to the number of antennas in the array, and one has dropped out for some period of time, failed solutions will arise.

3. Some people prefer to list their gains as numbers and look at those rather than plots. If you prefer this, use LISTR with optype='gain', dparm(1)=3 (for amplitude and rms) or dparm(1)=4 for phase and rms. Don't get confused by the scale factors that LISTR prints its results with (they are printed at the top of each page). I think that the plots provide a better qualitative feel as well as providing adequate quantitative information.
4. If after all this inspection you decide that some more flagging is necessary, go and do it; UVFLG is best for this fine detail work. Then redo the gain determination. Keep on iterating until you are happy. Usually, one or two passes are sufficient.
5. If you find some bad solutions, but cannot find any reason for them being wrong (i.e., the visibilities look fine) then you have two options for those miscreant scans. You could flag the calibrator scans, or you could flag the gain solutions in the SN table. The former is the safe way to do it, but then if you want to have another go at them later, you would need to unflag them. The task SNCOR can be used to flag actual SN table entries by setting clip levels. This may prove the quickest way to ensure bad solutions do not get applied. However, it is a less robust approach, since should you want to redo the calibration, you have to keep remembering to flag these bad solutions before applying interpolating them to the CL table. SNCOR can also be used to zero the phase part of selected solutions.

   After finishing with ATCALIB (or CALIB if you prefer), you should have produced one or more SN tables.