[evlatests] Adventures in wide-band calibration, cont.

[Rick Perley]

    This is a top-level overview.   The interesting details will be 
shown tomorrow...

    A 1.5 hour long test was run Friday night, consisting of four 
calibrator sources (3C286, 3C287, 1549+0237, and 1651+0129), plus the 
large radio galaxy 3C348.   The corrrelator was set up to provide 16 
subbands, each 128 MHz wide, with 2 MHz channel resolution, and full 
polarization.  The lowest 8 subbands covered 4488 MHz through 5512 Mhz, 
the upper 8 subbands covered 5488 through 6512 Mhz.  (The 24 MHz overlap 
was a blunder on my part -- 8*128 = 1024, not 1000).  The goals of this 
run were to test wide-band calibration techniques, and to judge the 
suitability of 3C348 ('Hercules A') as a 'demo science' source. 

    The database is impressively large -- over 100GB.  It took 10 hours 
for this 1.5 hour run to be transferred from the archive to my FITS 
disk.   The data were loaded without flags applied. 

    1.9% of the visibilities were integer zero.  Upon inspection, 
essentially all of these are associated with antenna 9, IF 'B' (subbands 
9 through 16). 
    Antenna 19 failed to provide fringes on 6 of the 36 scans.  This 
appears to be a 'failure to tune', as noise was visible on all. 
    Antenna 27, on the B IF showed a remarkable change in gain (about 6 
dB) in the middle of the first scan, after which it was stable.
    Antenna 25 is notably more unstable, on the RCP side, than any other 
antenna-IF. 

    Other than these relatively minor issues, the data quality was 
superb.  The only other flagging needed was for antenna motion. 

    Delays are all less than ~5 nsec.  CRoss-hand delays (based on 
3C286) were 5.5 nsec.  Bandpass stability (other than the RFI, and 
antenna 25 RCP) appears to be of order 0.1% (amplitude). 

    Parallel hand calibration was done using the method I'll describe 
tomorrow.  I also did polarization calibration -- PCAL (at present) 
defaults to integrating the entire 128 MHz -- a bad idea for a couple of 
reasons, noted below.  I got around this by flagging, the edge channels, 
running PCAL, then unflagging.  The result of this was quite impressive 
-- it worked!!  But there are issues, which I'll show tomorrow. 

    RFI is present, but there is very little.  The two major areas are 
near 6000 and 6250 MHz.  These are mostly washed on on longer baselines, 
but are very strong on short ones.  I haven't yet tried to sort through 
this massive data block to get a better idea of origin. 

    The bottom 2 or 3 channels of subbands 1 and 9 (the lowest subbands 
for the AC and BD IFs) show the large, non-closing offsets noted 
earlier.  These really mess up calibration and display, and are not 
handled well even by Hanning smoothing. 

    A lovely initial map was made, using just one of the 1024 channels 
... Somebody will have to show me how to properly include the other 
1023, as there are enormous spectral gradients across the source ...