The task CLCAL is used for the interpolation procedure. This task has rather a lot of inputs too, and I have provided a reduced set in the procedure ATCLCAL. Again, if you find insufficient flexibility for your purpose in ATCLCAL, go back to CLCAL. It is probably a good idea to look at both sets of inputs to start, so that you can see what ATCLCAL doesn't let you do. ATCLCAL writes the interpolated gain solutions into the calibration (CL) table(s) which usually have entries every few minutes or so (you specified this with either ATLOD or INDXR). The CL table contains one (interpolated) solution per time entry, and there can only be one source/frequency combination per entry. CLCAL also `manages' the SN tables. This means that it merges all the SN tables (so make sure you have deleted any you don't want with EXTDEST) into one and deletes the old ones before beginning the interpolation.
Note that a different run of ATCLCAL is required for each source and its calibrator (unless you have one calibrator for more than one source). I will describe here a fairly simple case of one primary calibrator, one secondary calibrator and one program source. Extension to additional complexity is fairly obvious. Although you have the option to write as many CL tables as you like, it is generally best at this stage to write all the interpolated solutions to table 2. This makes the book-keeping as simple as possible. However, if you are prone to making mistakes, then writing a new CL table for each run of ATCLCAL may prove useful, since you need only delete the CL table that you messed up, rather than deleting table 2 which might contain the results of many runs from ATCLCAL.
Beware of trying to ask ATCLCAL to do too much. Don't try to calibrate all sources with the relevant calibrators in one go and expect ATCLCAL to do the right thing. It does its interpolation in time without regard to where sources are. Thus you would be very likely to get solutions interpolated through sources which may be in very different parts of the sky (depending on exactly how your program was set up). Do it in the most fool-proof way, a bit at a time, and, in the long run, you will save yourself a lot of trouble.
It is important to realise that CL tables are designed to be used in a `cumulative' fashion. That is, one always generates a new CL table by applying the current gain solutions to a previous CL table. Thus, gain solutions can be incrementally improved. This is the procedure used in the iterative self-calibration technique (see § 17). In the current case, you are performing the first calibration, so you apply the gain solutions to CL table 1, which is all ones and zeros for amplitude and phase, respectively, and write CL table 2. Never delete the first CL table, as it is used to go back to a pristine calibration (if you do, then you must rebuild it [and the NX table] with INDXR).
You should still be working with the frequency averaged data base (channel 0) at this point, again regardless of what the final spectral goal is.