[evlatests] More curiosities ...

[Rick Perley]

    Being at loose ends, I looked a bit more at Michael's test data, 
taken in spectral line mode '4', at 25 MHz BW. 

    I did the usual editing, bandpass, and gain calibration.  Because 
Ant 1A has sampler issues, I looked only at IF #2.  I removed antenna 
22, which has extremely wild gain fluctuations in all IFs. 

    I split out channel 4 into a separate database for close analysis.  
I ran self-cal on this to remove any residual gain and phase fluctuations. 

    The noise histogram had an rms width of 119 mJy in Stokes 'I' 
(imaginary part), and 109 mJy for the real part, which is within 20% of 
what I expect. 

    A 1K x 1 K image with 1 arcsecond/cell, with natural weighting, 
provided a rather ratty looking image (i.e., not limited by noise) whose 
rms at the edges was 0.88 mJy, and 1.97 mJy near the point source.  
These are too high by factors of about 4 and 9 respectively compared to 
the expected = 0.22 mJy (= 110/sqrt(Npts)).    There are no other 
visible objects in the image. 

    I then ran BLCAL three times:

    a) Averaging all 90 minutes of on-source data
    b) Averaging for 30 minutes (i.e. on BLCAL per scan)
    c) Averaging for 10 minutes (3 BL solutions per scan).

    These had dramatic effects on the image:

    a) With a whole-observation BL solution, the noises dropped to 0.48 
(outer) and 0.94 (inner) mJy.  Still way above the noise.
    b) With single-scan BL solutions, the noises dropped for 0.37 and 
0.23 mJy -- close to the expected levels.
    c) With 10-minute BL solutions, the noises dropped to 0.30 and 0.11 
(!) mJy -- at or below the expected levels.  The noise images derived 
from (b) and (c) are very non-uniform -- still 'lumpy' in the outer 
regions, and far too smooth and uniform in the inner regions.  This I 
attribute to insufficient SNR for the solutions, causing some part of 
the real noise to be absorbed into the baseline-based solutions.

    Nevertheless, the dramatic improvement in DR obtained by applying 
the BLCAL program indicates a low and variable level of 'closure', which 
is affecting the VLA correlator.  Whether this is the issue with WIDAR 
remains unknown.