[evlatests] Troubles with 0.416 second integrations

[Rick Perley]

    Bob Sault and I have been investigating EVLA polarizer 
characteristics (a memo will come out Monday on the subject).   As often 
seems the case, this detailed work has revealed other issues ...

    We took the data in two sessions, the first, at C-band, was in early 
April, and used 1.67 seconds averaging.
    The second, at C and K bands, was in late April, and used 0.4166 
second averaging. 

    The data were filled as correlation coefficients.  For all 
observations, the correlator coefficient values were at the correct level. 

    After calibration, histograms of the noise were made.  In all cases, 
the widths of these, in all corrrelation products (parallel and 
cross-hand) were as expected. 
   
    BUT:

    The rms noise in all images made with the 0.417 sec averages were a 
factor of three to six too high, while the rms noise in images made with 
the 1.67 second data were close to the expected values ('close' means to 
within 20%). 

    At Ken's suggestion, I coherently averaged the 0.417 second data to 
3.33 seconds.  There was no change in the map rms noise.  Of course, the 
histogram widths narrowed by the expected factor. 

    The only explanation I can think of is that the fast-dump data has a 
correlator-based offset which is too small to be visible in the 
histograms.  But any such offset must persist much longer than 3.3 
seconds. 

    It would be prudent to not use this fast-dump option for sensitive 
imaging experiments.  (Fortunately, few users know of this option, and 
those who do are unlikely to use it.  It's mostly useful for diagnostic 
purposes).