[evlatests] Super-Fabulous Result!

[Rick Perley]

    Imaging tests done yesterday afternoon with the 3C147 L-band 
database taken last week provided excellent images, at various 
bandwidths, providing BLCAL is properly run.  These tests were done on 
various 'pseudo-continuum' databases, each centered at channel 129 (at 
the center of the subbands), with various bandwidths:  0.5, 10, 50, and 
100 MHz. 

    However, the improvement in sensitivity did not rise with bandwidth 
in the expected way.  The maximum dynamic range was about 195,000:1 (for 
the 100 MHz, single subband case) -- about a factor 5 too low. 

    Investigation of the BLCAL solutions showed they changed with 
bandwidth -- as expected if the major origin of these baseline-based 
solutions lies with the polarizers.  Plots of the cross-hand amplitudes 
show significant frequency variation in both the amplitude and phase 
(with typical frequency scales of ~10 MHz or so).   Fortunately, these 
cross-hand spectra are absolutely constant in time, suggesting that they 
can be calibrated relatively easily. 

    To check on this, I broke out from my spectral line database 11 
adjacent pseudo-continuum 'chunks', each 10 MHz wide.  Each was 
calibrated with the same model (the best map from yesterday's work), 
then BLCAL was run (using the same model), then a final self-cal to 
implement the BLCAL solution.   Some minor editing was done to remove 
discrepant data (mostly associated, it seems with the GPS L3 signal 
turning on for a few minutes ...).  Then the 11 databases were 
concatenated, and an image made.

    The result is a spectacular image with rms noise of 43 microJy (for 
a single subband), or a formal 'dynamic range' of 494,000:1.  The best 
I've ever seen.  With two subbands, the noise drops to 38 microJy.  
There are sources now clearly visible that I've never seen in any 
previous deep image. 

    But we're still not at thermal noise, missing by a factor of perhaps 
3.  (I used normal weighting, rather than 'natural', so some fraction of 
this is from the weighting).  More importantly, the noise is not uniform 
in the image, with significant regions of higher noise (by a factor of 2 
or 3) surround the brightest sources.  This could be due to frequency or 
time-structure in the baseline-based solutions, or perhaps to the 
fast-wobble time variations (which I made no attempt to remove). 

    I might try splitting into 5 MHz chunks this afternoon, and 
repeating the procedure.  (But there are probably more important tests 
to be run).