[evla-sw-discuss] Tsys & Calibration.

[Barry Clark]

I suggest that applying a Van Vleck type correction be dropped as a
software goal.

The magnitude of the correction is of order
a * c^2 * 2^{-nbits}
where a is some small constant (I rather think it might be 1/pi)
c is the zero lag correlation coefficient
n bits is the number of bits in the quantizer being corrected for

To second order, the correction shows up as a simple offset of the
whole spectrum.

I claim it is unnecessary for 8/7 bit sampling/correlation.

Rick's initial looks seem to indicate that it is unnecessary for
L band 8/4 bit sampling/correlation, although the question is a bit
up in the air for subbands containing the satellite radio signals.

3/4 bit sampling/correlation will only be used for C band and higher.
The only celestial sources that make c large enough to warrant a
correction are Cyg A and Cas A.  The only potential use of this mode
for these sources is to search for extremely wide, but faint absorption
lines.  For this use, not applying the second order term does not
compromise the reduction.  It is affected only by the even smaller third
order term.   An alternate route to reduction of this sort of 
observation is use of 8/7 bit sampling/correlation, and a really good
subband stitching.  We need subband stitching  that might well be good
enough for other purposes.  This way of doing the observation requires
four times the observing time to cover the spectrum, of course, but
It seems to me that the time the Array will spend on an
observation like this will not justify the programmer time required to
implement the Van Vleck type correction.