[evlatests] Some results from 3 and 8 bit tests

[Rick Perley]

    Friday was a '3-bit' day.  Michael and Ken set things up so that we 
had parallel data streams, with 3-bits through either the AC or BD IFs, 
and 8-bits through either BD or AC.  The correlator was arranged to give 
exactly the same frequency and subband coverage, so direct comparisons 
could be made:  1024 MHz width, with 8 subbands of 128 MHz each.  
Spectral resolution was 2 MHz.  Frequency span was 5488 through 6512 
MHz.  This means the two strong microwave links were in-band -- not a 
good thing for these tests (but useful in confirming we were tuned 
correctly, I suppose).  I suggest the next C-band test run from 4.5 
through 5.5 GHz...

    Four antennas are outfitted with 3-bit samplers (in both 
polarizations):  15, 22, 27, and 28.  After setup, two tests were run -- 
one with a long stare at 3C84.  Vivek is reviewing this one.  The other 
test was 30 minutes in duration, alternating between 3C286 and a blank 
field about 35 arcminutes to the north.   The following notes come from 
this short observation. 

    Basic results are:

    1) Basic data quality looks good, in that all four antenna fringed 
nicely and steadily when on the calibrator.  There are essentially no 
'blank' (integer zero) records.  'Dead' time at the beginning of scans 
was exactly as expected for the short slew (less than 10 seconds). 

    2) Delays for 8-bit data were minuscule.  Delays for 3-bits were 
large (8 nsec) in RCP, and huge on the LCP side:  typically 200 nsec. 

    3) Bandpasses are quite good -- the low frequency rolloff seen in 
the 8-bit path is not present in the 3-bit path -- as expected.  There 
is amplitude continuity between subbands in both polarization (in 
3-bits).  Phase continuity is present in RCP, but not in LCP -- there 
are 20 degree dips at the low frequency side in each subband, and 20 
degree rises in the high frequency side.  This doesn't seem to be a 
result of the very large delay for this polarization. 

    4) Calibration solutions looked pretty good -- but a closer look is 
needed to determine stability.  Phase stability appears normal. 

    5) What is *NOT* good is the noise in the off-source blank field.  
For the 8-bit path, the single-channel noise (as nicely determined by 
the histogram program UVHGM) is 0.17 Jy -- this quantity is closely 
reproduced in both polarizations on all six baselines.   
    In the 3-bit path,  the rms noise is much higher on nearly all 
baselines.    The problem is especially notable in RCP, where the six 
baselines show rms noises of 0.28 to 0.43 Jy -- typically two times too 
high.  Antenna 22 is especially bad.  In LCP, the situation is much 
better -- noise runs from 0.17 to 0.25 Jy. 

     One other easy conclusion comes from these data -- we have been 
concerned about the possibility that the roll-off in the 8-bit path seen 
at the low frequency edge might result in higher noise in imaging -- 
that is, the rolloff is not just a reduction in gain (with the noise and 
signal depressed together) due to some attenuation in the back-end.  In 
checking the noise histograms for the 3 and 8 bit data, I checked the 
noise as a function of channel for the 8-bit path -- I regret to report 
that there is a large decrease in sensitivity within subband 1, getting 
sharply worse with decreasing frequency.  It appears the 'rolloff' does 
indeed result in decreased sensitivity in the bottom ~100 MHz.  A closer 
look will be done later.