[evlatests] 3-bit degradation, (or not?)

[Rick Perley]

    I've spent a couple of hours looking at histograms from the 3/8 bit 
test data, taken a couple days ago.  There are some curious conclusions 
from this exercise. 

    The test included about 8 minutes on a blank field -- 10 individual 
scans of about 50 seconds' duration each.  For this analysis, I used the 
central 45 channels from the central subband for the B0D0, A1C1, and 
A2C2 paths.  The selected subband was centered at 13.960 GHz.  The data 
were calibrated for the bandpass and gains in identical manners for each 
of the three paths.  There is no RFI visible in this subband.  CLIP was 
run to remove outlier points (present only in the A1C1 path). 

    The histograms utilized the imaginary part of the complex visibility. 

    Some Notable Results:

    1) The A2C2 path gives much higher noise than the A1C1.    Averaging 
over all 20 working antennas, the A2C2 sensitivity is degraded by 13% 
(RCP)  and 15% (LCP) relative to A1C1.   Using a single antenna as a 
reference (ea01 in this case) the same overall degradation is seen, 
although the spread amongst the antennas is very wide.  Some are 
spectacularly bad:  in RCP, I note:  ea02, 04, 10, 19 and 20 (all 
greater than 15%).  In LCP I note:  ea07, 10, 16, 20 and 22 (all greater 
than 15%).   Although these antennas account for most of the 
degradation, there remains an overall offset.  This could be due to 
antenna ea01 (the reference), or it could be a true overall difference.  
A more extensive analysis would be needed to discern the difference.

    2)  The degradation of the 3-bit path compared to the 8-bit is 
easily seen -- but the amount depends very much on which antennas you 
choose.  Since the Hittite samplers are only in the A1C1 path, I 
compared A1C1 to B0D0 in the following.  Note in the following that 
there were only 3 'Hittite antennas' (2, 16, and 25, as ea03 does not 
have a Ku-band receiver), and of these, ea16 is clearly not working well. 

    a) Over all antennas, the degradation is 13% for RCP, and 14% for LCP. 

     b) Using ea02 as a reference ( a Hittite sampler equipped antenna, 
claimed to be working well), the overall average (i.e., ea02 to all 
other antennas) shows a degradation of only 7% on the RCP side, and 15% 
on the LCP side. 

       c) Using ea02 as a reference, the degradation as a function of 
antenna shows a very wide spread:  in the RCP side, it runs from -5% 
(i.e., 3-bits is better) to +23%.  On the LCP side, the spread is from 
-2% to +22%.   And -- quite notably -- there are six antennas on the RCP 
side for which the noise is equal, or less in 3-bits than in 8-bits. 

    d)  Taking the antennas with the least degradation on the RCP side 
as a group (ea02, 4, 10, 11 ,17, 19, 22, 23, 24, 25 and 27), the 
degradation is just 7%. 

    e) Taking the best of the above antennas (showing zero or less 
degradation w.r.t. ea02), (ea2, 4, 10, 11, 19, 22 and 25), the 
degradation is now only 3.5% -- not far from what is expected 
theoretically. 

    f) Conversely, taking the worst degraded antennas on the RCP side:  
1, 6, 12, 14 and 26 as a group, the degradation is 24%. 

    f) The above exercise (c through e) was repeated for the LCP side -- 
but the results are less spectacular -- clearly because we have many 
fewer 'really good' antennas.  The good ones are ea2, 4, 10 and 20  -- 
taken as a group, the degradation is 8% over the 8-bit path. 

    g) It's clear we need to 'fix' the worst-degraded antennas.  Here is 
a list for which the degradation (3-bits c.f. 8-bits is more than 10%):
          A1 (RCP):  1, 6, 12, 13, 14, 16, and 26.
          C1 (LCP):  6, 7, 12, 13, 16, 19, 23, 24, 25 (!), and 26. 

    So, my conclusion from this is that there is a subset of samplers 
providing near-theoretical performance.   This result is not likely to 
be due to the 8-bit samplers in those antennas being bad -- there is no 
obvious correlation between degradation and 8-bit baseline sensitivity. 
    Further, at least on the RCP (A1) side, the fraction of these which 
are Hittites is about the same as the fraction which are Teledynes!   
(But the statistics on the Hittites is very poor ...)   
    The other conclusion is that there are a (rather large) number of 
3-bit samplers which are much worse performers than others.  We need to 
tend to these.