[evlatests] Some clarity on 3 bit sampler issues ...

[Rick Perley]

    Michael and Ken ran a test yesterday afternoon in which:

    1) The power levels to the samplers were set at appropriate levels, 
as judged by state counts, and
    2) The stationboard requantizer gains were adjusted to optimum levels. 

    This was done, separately, for both 3-bit and 8-bit paths. 
    The observations were taken in the 'narrowband' mode, with 1 subband 
at 7056 MHz (thankfully free of RFI).  Observations were made of 3C286 
and a nearby piece of blank sky. 

    The results have considerably clarified what is happening. 

    A)  RMS noise levels. 

    I give the two 'noise matrices', with 8-bit on the left, and 3-bit 
on the right, and make some comments below.

                           8-Bit                                        
             3-Bit
---------------------------------------------------------------------------------------------------------------------------------------
       |      12          15             22              28       
|          12                15              22           28
-----------------------------------------------------------------------------------------------------------------------
12   |       X          .15          .15             .16        
|          X                .22             .45         .25
15   |     .15           X           .13            .14         
|         .17                X              .33         .18
22   |     .15          .13            X            .15         
|         .18               .16               X         .37
28   |     .16          .14          .14              X         
|         .19               .17             .16          X
---------------------------------------------------------------------------------------------------------------------------------------
    Two observations from this:

    1) In the LCP side, we have good uniformity in the noise values for 
both 8-bit and 3-bit paths.  But, the 3-bit path is about 15 to 20% 
noisier.
    2) On the RCP side for 3-bit, we have -- as usual -- values for 22A 
which are far too high.  But I argue also that 12A is also too high, so 
there is only one 'representative' value in that matrix -- 15 x 28, 
whose noise value of 0.18 is right in line with the LCP correlators.  
This interpretation is strongly supported by other evidence, given 
below.  Bottom Line -- I believe the RCP-LCP asymmetry that we have been 
focusing on for the past week is a *** red herring***. 

    B) Gains -- as determined from cross-correlation power. 

    If all antennas are identical, and all system parameters set 
equally, we should see the same cross-power amplitudes, and hence the 
same calibration gains.  Significant deviations from uniformity then 
give us useful information. 
    Below I list the calibration gains (in power, not amplitude, 
units).  Note that a low number means the antenna's correlated power 
(that is, the power from the source which is provided to the correlator) 
is high. 

                  8-Bit                    3-Bit
------------------------------------------------
12R        36                            69
12L        32                            44
15R         24                          36
15L        27                           37
22R        28                          144
22L        27                            34
28R        30                            44
28L        32                             37
-------------------------------------------------

    Note that 3-Bit gains are generally higher -- indicating that the 
power from 3C286 to the corrrelator is lower -- and that 12R and 22R are 
much worse than any others. 

    C) Autocorrelation Spectra.

    Back in VLA days, autocorrelation spectra were never utilized, as 
they had a 'bad rep' -- they gave weird results.  But in principle, 
these should give us useful information -- when properly calibrated, 
they should tell us the antenna SEFD. 
    It appears that this is what the WIDAR correlator's autocorrelation 
spectra are giving us.  Below I give the calibrated autocorrelation 
spectral amplitudes, in Jy.

                       8-Bit                   3-Bit
-------------------------------------------------------------------------
12R               280                      500
12L               260                      330
15R              220                        270
15L              220                        260
22R              250                      1200
22L              230                        260
28R             250                        350
28L              250                       310
----------------------------------------------------------------------

     A few spot checks shows a ***perfect*** correlation between the 
geometric means of the products of  autocorrelation amplitudes (above) 
and the noise values in the matrix at the top.  In other words, the 
(easily obtained) calibrated autocorrelation amplitudes are an excellent 
predictor of system sensitivity (as they should be).   We seem to have 
gained an even better way to determine system sensitivity than the old 
(VLA) route of utilizing normalized correlation coefficients. 

     Bottom line here is:  There in indeed extra noise in the 3-bit 
path, amounting to about 15 to 20% over the 8-bit path. 
There is (I claim) no asymmetry between RCP and LCP -- we have a couple 
of bad antennas/samplers on the RCP side.   The origin of the additional 
noise is yet to be determined.