[evlatests] A Very Interesting Result!!!

[Rick Perley]

    The signals from LCP on the four 3-bit samplers were split, so RCP 
and LCP were the same. 

    Ken ran the basic experiment, but halfway through, arranged to turn 
of thee fringe rotators. 

    The results are quite striking -- when the fringe rotators were 
turned off, the rms noise levels promptly *dropped by 20%* , to just 
about the correct expected values. 

    Here are the two noise matrices:

    A) Fringe rotators on, LCP fed to both correlator inputs

           12         15        22        28
---------------------------------------------
12       X         .19        .19       .21
15       .18         X        .16       .18
22      .17         .16         X       .17
28      .18         .17       .17        X

    B)  Fringe rotators off -- LCP still fed to both polarizations

            12         15         22        28
---------------------------------------------
12        X            .16      .15       .18
15       .14          X         .13       .15
22       .14          .13        X        .14
28       .15          .14       .14         X

    The 'expected rms noise' is about 0.14 Jy -- so, except for 12RCP 
(known to be not quite right), all correlators are now at the right 
noise levels. 

    There is one important caveat:  When the fringe rotators are off, we 
cannot get any correlated signal from 3C286, so we can't calibrate the 
data.  I had to utilize the gain solutions determined 5 minutes earlier, 
when the fringe rotators were on.  We can not discriminate between a 
gain change caused by the turning off of the fringe rotators, and a real 
reduction of the noise.  But -- if a gain change did occur, it is of 
exactly the right magnitude.