[evlatests] AIPS Weights, etc.

[Rick Perley]

    I've looked at the AIPS weights, and the reported system 
temperatures, from
my Thursday data, and did a rough correlation of these with the measured
rms amplitudes and/or phases across the passband. 

    Basic result:  There is little to no correlation between AIPS 
weights, reported system
temperatures, and actual measured dispersions of the amplitude or phase.  
Of the three possible combinations, the best is between aips weights and 
measured
noise -- but this is far short of what I'd like to see. 

    Some details.  I used IF 'B' throughout.   All data were calibrated 
on a point source. 

    1) L-band.  Only antenna 14 was available (due to the 16B problem).  
It ranked 12th
best in AIPS weights, and 16th in rms.  Right in the median, as expected. 

    2) C-band.  Antenna 16 was out for reasons unknown.  Antenna 14 
ranked 3rd in
aips weights, and 2nd in rms phase noise.  Not bad! 

    3) X-band.  Antenna 16 was 3rd in aipsweight, and 15th in spectrum 
noise. 
Antenna 14 was 21st ub aipsweight, and 13th in spectrum noise.  A poor 
correlation!

    4) K-band.  Antenna 14 was 17th in aipsweight, and 9th in noise. 
Antenna 16 was 23rd in aipsweight, and 19th in noise. 
    For this band, the spread in aipsweights is particularly large -- 
the spread from the
typical worst to typical best is a factor of about 25 (corresponding to 
a factor of 5 in
Tsys or antenna efficiency).  The spread in spectral noise measurements 
is at most
a factor of 2. 

    I am leery of putting much faith into these convenient aips weights. 

    On the other hand, the rms values I determined were from 3 MHz-wide 
spectral
channels across the full 50 MHz bandwidth.  I can imagine that gain 
variations and
perhaps uncompensated spectral features will skew the values.   Also, 
with only
14 channels to make the measurement, we're getting close to the 
'small-N' problem. 

    I suggest using some of my time tonight to repeat these, using much 
narrower
BW (and more channels) to get a more reliable estimate of the true noise.