[evlatests] 3-bit sensitivity

[Rick Perley]

    The Friday night '3-bit' test has revealed much information.  Here I 
report on the obtained sensitivities. 

    I've derived the sensitivities via histograms of the noise when 
observing a blank field.  The observing frequency used for this was 
14.45 GHz, for all three 'paths' -- the 8-bit, and the two 3-bit paths.  
All three employ the same receiver and downconversion paths.  The 
separation occurs within the T304 module.  Any differences observed must 
originate in or after that module. 

    I measured the noise histograms for all 30 parallel-hand correlator 
products for each of the three paths.   For the 8-bit path, the rms 
noise amongst all 30 correlators was extremely uniform:  rangiong from 
134 to 155 mJy, corresponding to an SEFD range of 250 to 290 Jy -- spot 
on the expectations from system temperature and antenna efficiency. 

    As expected, the noise histograms for the 3-bit paths are all 
higher.  But there are interesting patterns! 

    Looking at A2C2 first:  The median degradation amongst all 
correlators is by a factor of 1.13 to 1.14 (fit by eye!).  However, it 
is quite clear that two antenna-IFs are considerably worse than the 
others:  7R and 14L.  If we exclude these two, then the median 
degradation is 1.10, with a range from 1.03 to 1.14.  We expect a loss 
of about 3 to 4% due to the 3-bit quantizers (is this the right factor?). 

    A curiously different pattern emerges from A1C1, both better and 
worse.    First -- the poorer results from 7R and 14L are again seen.  
But added to this is a much degraded noise in 23R.  Upon inspection, it 
is found that this is due to rapid time variability in the noise -- 
there appear to be 'up' and 'down' states, separated by seconds of 
time.  I haven't attempted to derive the noise in the 'down' states -- 
likely it is what we expect.  Removing these three antenna-IFs from the 
noise measurements results in a median noise degradation remarkably 
small!  My 'chi-by-eye' fit shows a mediatn degradation of only 1.05 to 
1.07 -- only slightly worse than we expect. 

    I will review these interesting results later today. 

    A later email will better document some less pleasant features of 
these 3-bit data -- internal birdies, and curious differences in 
polarization between the two 3-bit paths.