[evla-sw-discuss] Resampler scaling

[Barry Clark]

It turns out that Brent and I have been thinking along slightly
different lines, in terms of calibration, which are, in the case
of stable subband power, equivalent.  In case of varying subband
power (either from interference within the subband or from solar
bursts) they may differ, and it is not clear to me which is better.
I therefore suggest that we implement the hooks for both.

Brent's line is that an ALC should operate within the station
board CMIB, and maintain a constant level going into the subband
resampler.  The power before the resampler is reported, and used
for calibration, probably being passed through the VCI to the 
Executor to rejoin the correlator data later.

My idea was that at the beginning of an observation, the 
resampler scaler should be set to an optimal level, and then 
left alone through the observation, with the power in the 
subband being determined by the scaler value and the autocorrelation
zero lag (from the baseband board).  (I once listened to a
Larry D'Addario diatribe about the evils of ALCs, and was semi-
convinced.)

Which of these is better may well depend on the nature of the 
subband power variations - Brent's would take care of slow 
variations without the observer having to worry about them at all,
and mine might be easier to correct for really violent variations
because what is happening now does not depend on what happened a
little while ago.


So the CMIB (and the VCI) should support three modes:
1.  No ALC - an absolute rescaling factor specified
2.  Full ALC - a desired level and a time constant for the ALC
    specified.
3.  Set-and-forget - a desired level specified, possibly with an 
    integration time before the "forget" phase also specified.

The third mode could be implemented using the first and feedback 
from the correlator via TELCAL, but doing it all in the CMIB
is certainly *much* simpler.

In all three modes, we should get monitor information about subband
power and rescaling factor back across the VCI.

By the way, does the resampler wrap, clip, or invalidate the data
in case of an overflow?