[evlatests] Image Symmetrization -- More to the story

[Rick Perley]

    It seemed clear that the image symmetrization seen a couple days 
back was due to using a point-source model, rather than a more accurate 
representation of the field, in the calibration procedures.
 
    There is more to the story, however. 

    I have carefully reduced the data from three nights ago (5 hours, 
single polarization, on 3C286), using sub-band 1 (1300 -- 1428 MHz), 
which is quite free of strong RFI.  The standard bandpass and gain 
calibration methods were followed. 

    The data were then reduced to 24 channels (from 256) by averaging in 
10 channel 'chunks' -- to decrease data volume, to improve calibration 
SNR, and to increase imaging speed.  BPASS was employed to provide a 
channel/time dependent calibration (5 MHz, 15 seconds), using a good 
image, and NMAPS = 1.  The resulting images are very good.  However, it 
was noted that symmetrization is seen *in the edge channels* from both 
sides of the spectrum, and that the degree of symmetrization is a sharp 
function of frequency offset from the edges.  The edge channels 
(referring to the 5 MHz wide 'chunks') show 16% symmetrization (the 
reflected image is 1/6 of the primary), and the effect disappears by the 
7th channel (i.e. 35 MHz) from the edges.  (Ratio is less than 1 in 200). 

    To confirm this is NOT a calibration effect, I split out the lowest 
frequency 'chunk' into its own database, and ran simple CALIB on it, 
using the best map determined from the central channels.  The resulting 
map retained the the image symmetrization, at the same level as before 
(1:6). 

    I hope somebody has a rational explanation for all this ...