[evlatests] Reviewing June 3

[Rick Perley]

    I reviewed the June 3 observations -- one hour on our favorite naked 
northern source -- and learned a few more things about the 'lumpy-bumpy' 
structures which affect the maps. 

    I used only IF#3.  Bandpasses were calibrated for each scan (40 
seconds long).  Data were SPLITted off with calibration applied and 
various amounts of channel averaging.  Basic clipping was done (mostly 
to remove residual 180 degree phases jumps).  Very little data were 
removed. 

    Seven SPLITted databases were averaged over:  1, 4, 16, 64, 256, 
512, and 951 channels, to see how the  histogram and image noises change 
with bandwidth.   The images are 1024 x 1024, with 1 second cellsize.  
The noise was measured in the 'upper right corner'. 

    Results:

    1) The expected rms noise relationships in the histograms were 
followed precisely up to and including 256 channel averages (32 MHz).  
Beyond that, tails in the gaussian distributions are seen in both real 
and imaginary parts. 

    2) The map noises follow the expected relationships only to 64 
channel averages, after which they quickly flatten off.  The images, 
formally, are pretty good -- maximum DR of 40,000:1.  But the underlying 
'lumpy-bumpy' structure is clearly evident in all of the wider bandwidth 
images.  The lumpy structure is uniform throughout the map, suggesting 
it is concentrated to a particular time -- I have not yet tried to 
confirm this.

    3) Running BLCAL (averaged over the entire time frame) makes no 
difference to the images.  The closure values are extremely small -- 
typically 0.05% or less. 

    4) There are 5 discrepant baselines whose phases and amplitudes are 
clearly worse than all the others -- this is seen very easily in the 
wide bandwidth databases:  19 x 1, 2, 24 and 28, and (amazingly enough), 
1 x 23.  (This latter pair forms a long E-W baseline.  No other pair 
with either of these antennas is visibly bad). 

    5) Removing these 5 baselines makes little difference to the image!  
The rms noises are the same.  The histograms are definitely improved 
(more gaussian), but the 'tails' remain -- these must represent variable 
non-closing offsets which are approximately equally seen on all the 
remaining baselines.