[evlatests] 3 vs. 8 bit tests, continued ...

[Rick Perley]

    Ken established earlier today that the problems with the 3-bit test 
reported on last evening were reproduced today, and are a result of the 
configuration mapper.  (Please correct me if I'm wrong here...)

    He then retreated to a much simpler 'OSRO' style setup, essentially 
reproducing the initial 3-bit tests, done last month.  He made two short 
observations, one in 3-bit mode, and the other with 8-bits, each with 
128 MHz BW and 2 MHz channels, with a single spectral window for 3-bits, 
and two spectral windows (IFs) for the 8-bit (but at the same 
frequency.   Observations alternate between 3C286 and a nearby blank 
field, to enable good estimates of the noise.   Each file was observed 
for about 30 minutes. 

    The frequency of the 3-bit test was 5448 MHz, for 8-bits it was 4936 
MHz. 

    Major Results:

    A)  The 8-bit data are of extremely good quality, with stable gains 
and phases as expected from a cloudy summer afternoon.  The noise 
histograms for all 6 baselines, in both polarizations, have a 1-sigma 
width of 23 mJy -- this is about right for the the combination of 2 MHz 
channel width and 1 second integration. 

    B) The 3-bit data are much less stable, and are somewhat noisier:

    - All antennas/polarizations show the 'warm-up' phenomenon, whereby 
gains change rapidly over a ~5 minute duration before stabilizing.  The 
different samplers show quite different ranges in their gain changes:  
15R hardly changes at all (less than 5% in power).  22R and 27R change 
by over a factor of 2 in power (3.3 dB).    In all cases, the fringe 
power is declining during this interval, so the 'sampler gain' drops.  
There is no apparent change in the rms noise however -- the process 
doesn't seem to affect sensitivity. 
    -  We have step functions in gain for most antennas -- but these are 
not by 1dB, so they are not due to attenuator setting changing.  Typical 
gain changes are by 5% in amplitude -- (0.4 dB), or less.  One jump took 
place in the middle of a scan and affected 15L, 22L, and 27L.  However, 
at this moment, antenna 28 went offline (was flagged due to low 
amplitudes), so these events are probably related.  The change in RCP 
gain at this time was minimal.  Another gain change occurred 20 minutes 
later -- affecting the three remaining antennas (15, 22, and 27) all 
differently. 
    - Phase stability is the same as for the 8-bit data. 
    - The noise histograms, done in the same way as for the 8-bit data, 
show the rms width to be typically 26 mJy -- 12% noisier than the 8-bit 
data.  Some baselines pairs are much worse however -- 15 x 22 in RCP was 
38 mJy.  15 x 28 in RCP was 32 mJy, and 31 mJy in LCP. 
    - Autocorrelation spectra were available only for antennas 15 and 
27.  Of these four (2 antennas x 2 polarizations), the internal RFI 
birdie was seen only on 15RCP.  All others looked fine.  (But, we have 
only a single spectral window, so the frequency span is small).  The 
birdie changed in frequency at the beginning in the manner now known as 
part of the thermal stabilization. 
    - The cross-hand autocorrelations look normal, albeit with a large 
delay (which is not surprising, given that the R-L phases have not been 
adjusted).