[evlatests] Switch Failure Statistics

[Rick Perley]

    Although we have made the decision to replace the band switches, I 
thought it might prove illuminating to do some basic statistics on the 
failure rate from the observations taken Friday night. 

    I use as the metric of success or failure the smoothness and 
continuity of the PDif values over the 7 hours duration of the 
observations.  Recall that I cycled around the eight frequency bands, at 
a rate of roughly one change every minute -- each band was tuned 37 
times, in the order:  Cr -- C -- S -- Ku -- K -- Q -- Ka -- L -- X.    
'Cr' stands for the referenced pointing tuning, which was at a different 
pair of frequencies than the actual observation. 

    A simple table was constructed, comprising four columns for each 
band and telescope -- the four columns are for the four IFs.  An antenna 
was declared 'unstable' if there was more than one clear deviation in 
PDif from a smooth curve.  (Usually, for those antenna with a single 
deviation, it was the first observation -- something to do with setup 
perhaps).  I then counted the number of 'failures'.  The distribution 
with band is quite striking.  The table below shows the number of 
failures, and the fractional failure rate (since not all bands are 
equipped on all antennas).  For some bands, special situations were 
observed, and these are in the 'Notes'. 

Band         Failures         Fraction        Notes
----------------------------------------------------------
L                   10                 9%             Four of the 
failures are with antenna 18 (all IFs). 
S                     5                 7     
C                    5                  4                Four of the 
failures are with antenna 25.  Note this band followed referenced pntg. 
X                   16                20               (BD side was 
tuned to 11.2 GHz, so only wideband receivers are counted). 
Ku                 11                17
K                   26                25              AC side fails much 
more commonly (40%)
Ka                 22                21              AC side fails more 
commonly (31%)
Q                  52                 50%          LCP fails much more 
commonly (67%) 
---------------------------------------------------------------------------

  'Failures' come in two flavors: 

(a) A change in gain, but stable throughout a particular observation.
(b) A time-variable change, where the synchronous power (PDif) is 
changing second-by-second, often not stabilizing at all during the 
minute-long observation. 

    The good news is that in nearly all cases, the detected change in 
PDif is reflected in the visibilities, and application of the switched 
power (mostly) restores the proper amplitudes.  There are some notable 
exceptions, where application of apparently useful PDif data greatly 
worsens the visibility data.  I'll report on this next.