If you averaged all your data into one channel per IF frequency, skip to § 8.2.
You need to determine the gains from the calibrators as a function of frequency as well as time. We would hope that the gains are stable in time and frequency, but generally the variation from channel to channel is more stable in time than the gain variation of a single channel with time. This is because the latter depends to a large extent on the atmosphere, and the former doesn't.
CALIB can only work on one channel at a time (or channel 0), so that you can't determine the calibration as a function of frequency with CALIB alone. Thus, CALIB is used on one channel, and the task BPASS is used to work out the variation across the band (see § 11). Hence, you need to produce a channel 0 data set for CALIB to work on, regardless of whether you want to do spectral-line or multi-frequency continuum synthesis work. If you haven't already produced the channel 0 data base, do this with AVSPC as detailed in § 6.2. Include as many (good) channels as possible in this average to allow the highest signal-to-noise ratio solutions (see discussion on how to determine this in § 6.2).