[evlatests] End-of-Scan problem

[Rick Perley]

    I am now deriving the flux densities of the 'Planck' sources 
observed in the big December run.  One of these sources is 3C454.3, 
which is currently very strong -- about 50 Jy.  This allows a level of 
careful review at high frequencies which is otherwise difficult or 
impossible. 

    The 'Planck' sources were observed at X, K, Ka and Q bands.  After 
standard phase and amplitude self-calibration was completed, plots of 
visibility phase showed about 25% of the phases were scattered about +/- 
5 degrees, while all the rest were with 0.1 degrees about zero.  Very odd. 
    Further investigation showed that *all* of the discrepant phases 
were taken in the last two integration of each scan -- each integration 
is 3 seconds long, so up to 6 seconds of data are affected! 
    I claim the 'fuzzy noise' has nothing to do with integration length 
-- the scatter is 10 to 20 times too high. 
    I further claim that the problem indicates a real 'closure' problem 
-- even if the data belonged to another band, the object is a perfect 
point source, and the phases would calibrate out. 
    The issue is not seen in amplitudes -- these remain stable to better 
than one part in 200. 
    Finally -- the problem is seen in K, Ka, and Q bands, *** but not at 
X-band***.  What was different about X-band?  At this band (only) I did 
not change band following the observation (I did the referenced pointing 
scan).