Computation

            

1. The procedure is the same whether you have converted the correlations to Stokes parameters (IQUV; not recommended) or left them as linear polarizations (but called them circulars, RR, LL, RL, LR), as the relevant tasks account for this. In the Stokes case, solutions are only found for I (as we don't have polarization models). In the latter case, two sets of solutions are found, one for XX (called RR) and one for YY (called LL). In addition, CALIB will determine the gain solutions for all the IFs contained in your data. There is no IF selection.
2. If you look carefully at the CALIB inputs, you will see that averaging over channels can be done within CALIB (with the bchan and echan) adverbs. However, I encourage you to use a separate channel 0 data base, rather than use this option. The reason is that the averaging switches are fairly well buried in CALIB's inputs, and they are poorly documented and easy to get confused over. The safe route is via AVSPC.
3. In this sub-section you should be working with the channel 0 data, regardless of your final spectral goal. The task CALIB determines the antenna gains as a function of time from the calibrator sources, and writes the solutions into the solution (SN) table ready for interpolation to the program source(s). If you examine the inputs to CALIB, you will find that its adverbs are as generously allocated as in ATLOD. CALIB is a very general task, and many of the inputs are irrelevant to our purpose. However, rather than continually scroll through all these unwanted adverbs, I have provided some AIPS procedures which hide many of these adverbs.

   The CALIB procedure is called ATCALIB. See § 2.2 for some useful information on the procedures. Most of the inputs have the same name as in CALIB; a couple are changed because they are normally poked into those aparm, bparm, cparm or dparm arrays. Obviously, ATCALIB has lost some flexibility compared with CALIB, but it should do what you want for this first calibration (if not, use CALIB). For example, it assumes that the model for each calibrator source is a point source at the phase centre; this is the usual premise for initial calibration (see § 3.1). If there is no flux density in the SU table for a particular calibrator, ATCALIB will assume that 1 Jy is the correct flux density; that is, application of the gains to that source would yield a mean visibility amplitude of 1 Jy. The only source for which you should have a correct flux density in the SU table is the primary calibrator, 1934-638 (see § 5). Later on, you will account for the 1 Jy assumption by comparison with the primary calibrator. This is called boot strapping the flux density scale (see § 9).

4. Because of ATCALIB's importance, I will discuss all of its adverbs.
   • Fill in inname, inclass, inseq and indisk with the GETNAME verb. Make sure you get the channel 0 data base.
   • Select the desired calibrator sources with calsour and calcode. If you have set calcode values as discussed in § 5, then it is often most convenient to leave calsour blank, and put calcode='*' to select all the primary and secondary calibrators. Otherwise, fill in the calibrator source names in calsour.
   • Set the freqid (see LISTR summary sheet for these). Remember each freqid must be processed separately so that one run of ATCALIB for each frequency is required.
   • Initially, you will probably want to solve for all the data so leave timerang as all zeros. Otherwise, timerang has space for a start time (day, hour, minute and second) and, similarly, an end time.
   • Select the antennas to solve for with antennas. Generally, you will want them all, so leave this at zero too.
   • Calibrators are not always perfect point sources, depending exactly on the size of the baselines that you observed them with. As you go to longer baselines, fine structure may become apparent. As you go to shorter baselines, extended structure may become apparent. uvrange allows you to select those baselines for which the source is strictly a point source. To assess this, use some UVPLT amplitude versus uv-distance plots. Be careful in setting the uvrange though, because you must have enough baselines to get good solutions (the more baselines the more the problem is overdetermined). However, because the compact array is just that, compact, the point-source model should be fairly good on its longest baselines and you may need to worry about excluding only the shortest baselines if the visibility amplitudes show signs of short baseline structure.
   • Set docalib=-1 as you don't have any calibration to apply yet and ignore gainuse which specifies which calibration table to apply if docalib is true. Generally, you would only set docalib=1 while doing self-calibration iterations; it is used to apply a previously determined calibration. For basic calibration as you are doing here, leave it negative.
   • Set flagver to zero and the highest numbered FG table will be applied.
   • refant should be set to the number of the reference antenna, for which the phase will be set to zero (see § 3.1). It should be present throughout the observation, have a good signal-to-noise ratio, and be free of gain jumps. You might as well use the antenna chosen as the reference during the observations, but it is not vital. If you leave refant at zero, the antenna chosen will be arbitrary.
   • Usually, we solve the calibration equations with a least-squares algorithm; this is obtained with soltype=' '. However, sometimes AT data processed like this has been prone to giving large numbers of failed solutions. The L1 solutions, invoked with soltype='L1', in which just the modulus of the difference of the model and data is minimized instead of the square, appears more robust. This is probably because sometimes the real and imaginary parts of the visibilities are not drawn from good Gaussian distributions. The L1 solution is less affected by outliers and fails less often. You should try the least squares solution first, and if that does not prove efficacious, try the L1 solutions.
   • solint controls the solution integration time (in minutes) over which the data are averaged. Leave this at zero, and all the data in a scan will be averaged. This is the usual approach, as calibrator scans are generally just a few minutes long (howver, see discussion in § 8.3).
   • doscalar controls whether the data are scalar or vector averaged over the solution interval. In a vector average, the real and imaginary parts of the visibilities for each baseline are averaged, and then the amplitude and phase can be formed from the averaged visibilities. In a scalar average, the amplitudes for each visibility are formed first and then averaged. However, the averaged phase is still made with a vector average, as a scalar average for a circular quantity like phase is not very meaningful. If doscalar > 0 then a scalar average will be done, otherwise a vector average of the data is taken.

     Generally, you should set doscalar=-1, but see § 8.3 for a discussion of when this may not be true (bad phase stability).

   • minant controls the minimum number of antennas that must be present in order to get a solution. Since the ATCA has only 6 antennas in the compact array (5 in the 3 km array and one 3 km further away), the loss of one or two antennas because of failure becomes important in determining the solutions. Should this occur for some period of time, you can exclude these scans by demanding that minant be, say, at least 5. This means you would then have to interpolate the solutions over a longer time baseline. It's a trade off. Experience suggests minant=3 is a good choice. But really, the bigger the better (but not more than 6 !). The default of zero means three antennas.
   • If minamper and minpherr are non zero, then residual errors are listed to the screen. For individual baselines, these are amplitude percentages from the model amplitude (the flux density for a point source) and degrees from the model phase (zero for a point source at the phase centre), respectively. These are about the only diagnostics there are from CALIB to let you know when bad solutions are occurring (except outright failure). Set these to something like 10% and 10 degrees to start with. Any baseline that returns larger residuals than these will be reported to you. The actual errors you should expect because of noise only, clearly depend on the scan time and calibrator strength (see Cornwell 1981), but the occurrence of no residual errors above a few degrees and a few per cent is typically what we aim for. These residual errors are often called closure errors.
   • docrt, if negative, allows you to list the messages from ATCALIB\ on the printer automatically . These messages (e.g., closure errors) are normally written to the terminal as well as to the message file, and turning this switch on just invokes the AIPS verb PRTMSG to print the message file. Note, however, that it clears the message file of all old ATCALIB messages before printing the new ones. If docrt is non-negative, the messages just go to the terminal and the message file as usual.

5. Now it is time to set ATCALIB in motion with the GO ATCALIB\ command. Remember that you must run ATCALIB once for each source requiring a different combination of freqid, uvrange, and antennas and write a new SN table for each run. As it computes, ATCALIB reports the closure errors to you, and at the end, it will tell you how many good and how many failed solutions it found. A failed solution is when the algorithm is unable to find a global minimum for the function that it is minimizing within its internally allocated number of iterations.

inname,inclass,inseq,indisk Fill in for averaged data base

calsour=' ' Select sources explicitly or leave blank

calcode='*' and use calcodes ('*', 'p' or 's' say)

freqid=1 One run of CALIB per freqid is necessary

timerang=0 Solve for all time ranges

antennas=0 Solve for all antennas

uvrange=0 Solve with no uv restrictions or

docalib=-1 No calibration to apply

flagver=0 Apply highest version FG table

refant=3 Specify the reference antenna

soltyp=' ' Least squares algorithm. If too many failed solutions, try 'L1'

solint=0 Averaging time in minutes. 0 means scan aver

doscalar=-1 Vector average

minant=3 Min. no. of antennas to work out a solution

minamper=5 Report amplitude errors bigger than 5%

minpherr=5 Report phase errors bigger than 5 degrees

docrt=1 List results on terminal

docrt=-1 List results on printer