[evlatests] 3bit vs 8bit sensitivies -- the 'final word'

[Rick Perley]

    The data from the longest (and hopefully last!) 3-bit/8-bit test are 
now calibrated and the results ready for distribution. 

    In this note, I summarize the sensitivity losses due to the 3-bit 
path.  There are no surprises here. 

    I've compared the noise histograms, using the imaginary part of the 
correlation, for all parallel-hand products while on the empty noise 
field.  I used the central part of the central subband -- 14910 MHz.  
There is no visible RFI in the utilized channels.  Standard bandpass and 
gain calibration were done, and I've ensured that the 'PDif' compression 
problem has been avoided, so there should be no offsets in the 
calibration between 8-bit and 3-bit data. 

    There are two major points (which have been made before by Vivek) to 
emphasize:
   
1) The loss of sensitivity in the A1C1 path is 10% (not including 
'special' antenna-IFs which I'll discuss in point 2).  The loss of 
sensitivity in the A2C2 path is 15% (again, not including 'special' 
antenna-IFs). 

2) There is a very clear antenna-polarization effect, whereby some 
antenna-IFs are much more degraded than others.  The outstanding 
perpetrators are:
            ea26-A1:  mean degradation is 25%!
            ea14-A1:  mean degradation is 18%
            ea06-C1:  mean degradation is 15%
            ea23-C1:  mean degradation is 15%
            ea18-A2:  mean degradation is 48% !!!
            ea26-C2:  mean degradation is 25%!
    Is there anything 'special' about these samplers?  Unusual bias?  
Unusual rms? 
    Of course, there are also antenna-IFs for which the degradation is 
less than the mean.  ea07-C2 is only 8%, ea18-C2 is 7%, for example.