[evlatests] A remarkable result at L-band

[Rick Perley]

    Michael observed the standard calibrator 3C286 at L-band, using 11 
EVLA antennas with WIDAR.  The four subbands, each of 128 MHz BW were 
adjacent, the lower edge of subband 1 was set to 1300 MHz.  3C286 is an 
excellent test source -- 14 Jy unresolved, with 500 mJy in background 
objects easily visible within the primary beam. 

    Subbands 1 (1300 -- 1428) and 4 (1684 -- 1812 MHz) are nearly free 
of interference, so I concentrated on these.    The calibration method 
was as follows:

    1) Basic editing to remove data taken while slewing, and antennas 
which mispointed due to power glitches, etc. 
    2) FRING was run on a short piece of data to determine the basic 
delay errors (typically 20 nsec). 
    3) CLCOR is used to apply these. 
    4) A basic calibration was done, using a subset of channels which 
are free of RFI in all four subbands.   This was done to set the 
intensity level for the clipping stage.  
    5) SPFLG was used to view the spectra (an amazing sight, to wondrous 
to describe briefly here, due to the beating between the RFI and the 
source).   Roaming about the image permitted a clip level to be set. 
    6) CLIP is run to remove strong RFI -- the upper level was set to 25 
Jy, the lower level 0.001 (this needed because about 0.1% of the 
visibilities are perfectly zero).  
    7) BPASS was used to determine the mean bandpass.
    8) Sub-bands 1 and 4 were then SPLITted out, averaging in 10 MHz 
'chunks' (20 channels), with 12 'chunks' per sub-band.  This is needed 
to remove chromatic aberration in the image. 
    9) A point-source self-cal was run on these databases -- averaging 
the 12 'chunks' in each to obtain the solution. 

    The resulting image shows the primary background sources quite well 
-- very encouraging.  But there's lots of rumble lying about, looking 
like the result of variable bandpasses. 
    I then tried using BPASS to remove variable bandpasses -- this 
provided a solution for each of the 12 'chuncks', using a solution time 
of 1 minute.   The program BPLOT shows small (few percent, few degrees) 
variations in these gains.

    The resulting 'final' image is very lovely, with many background 
sources and nice clean noise. 

    HOWEVER: 

    It is easily seen that every strong background source *** has a 
mirror image of 1/10 the strength *** present, exactly on the opposite 
side.  How can this be?  I can't see these mirrored images on the map 
made prior to the BPASS gain solution, but perhaps the sensitivity is 
poorer due to the errors.  Certainly these mirrored objects were not 
present in an image made with a pseudo-continuum database with 100 MHz 
effective bandwidth -- but on these, the background sources are quite 
attenuated due to chromatic aberration.