[evlatests] More 3-bit test results

[Rick Perley]

    A major goal of yesterday's test -- no reported by me -- was to 
examine the X-hand autocorrelations and compare their amplitudes to the 
parallel-hand autocorrelations for the same antennas.  It is stated 
that, given the signal replications of LCP into RCP, these should all be 
the same.  I examined yesterday's data for this, but these X-hand 
amplitudes were only a tiny fraction of the parallel hand.  Ken and I 
determined the cause was an enormous X-hand delay, which shifted the lag 
function's peak outside the delay span.

    This was corrected this morning, and the test has been re-run.  Results:

    A) The 'rms noise matrices' tell exactly the same story as yesterday 
-- turning of the lobe rotators causes the noise to drop by 20%, to the 
level expected by theory.  However, the same caveat applies -- we don't 
*know* the real noise dropped, since we can't measure the gains when the 
rotators are stopped.  (But somebody will have to explain why stopping 
the lobe rotators changes the system gain by 'just the right amount'). 
    An extra note here:  the noises for RR and LL are identical *except* 
for antenna 12, where the RR noises are 15% higher than LL.  This can 
only be due to the sampler on that side (12R), since the same signal is 
fed to both R and L. 

    B) Turning off the lobe rotators certainly stabilizes the 
'cross-auto' spectra.  (With fringe rotators on, the 'cross-auto' 
spectra have two different levels, which they alternate between on 
timescales of ten seconds or so). 
    But -- the cross-spectra levels are nowhere near the auto-spectra.  
Here is a table, with the typical value in Jy  (a value in the upper 
200s would be considered right).

     Ant.        RR             LL            RL            LR
----------------------------------------------------------------
      12         450            300           200            200
      15         280            275           180            180
      22         275            275           170            170
      28        300             300           190            190

    Note the ratio of (cross/self) is about 0.64, far from the value of 
1.0 we hope for. 
    Is it possible the X-hand delay is still large enough to have 
shifted part of the lag function outside the window?  The cross-hand 
spectra still show quite a vicious slope in the phase...

    Note also the high value for RR:  450 Jy -- more evidence that that 
sampler has troubles.