[evlatests] More on low-level polarization issues.

[Rick Perley]

    Ken's report on the floating point arithmetic problems in the MIB 
encouraged him to wonder if the slow wind of RL and LR phases in my 
3C147 L-band polarization data might also be due to this.  Indeed, the 
observed rotation -- 5 turns in one day, corresponds to 60 microHz. 

    If so, the effect would be restricted to EVLA antennas.   The data I 
showed Monday morning showed the effect in both VLA and EVLA antennas -- 
however, a 'global, all antennas' PCAL had been run, perhaps causing the 
EVLA effect to 'leak' into the VLA. 

    To check this, I re-calibrated the VLA-only data, specifically, 
running PCAL on them alone, with a VLA antenna as reference.  Sadly, 
imaging with VLA only, and VLA-only calibrated data shows the effect is 
still there.   It's harder to see on the VLA-VLA baselines, as the 
'leakage' terms are smaller, which seems to reduce the size of the 
variable effect. 

    By looking at the baselines on which the phase rotation is seen with 
the best SNR, I think I can convince myself that the signature is 
visible in the originating data (meaning, the cross-hand data following 
parallel hand calibration, but prior to PCAL) -- as a slight sinusoidal 
oscillation -- as expected if a slowly winding vector is added to the 
large and constant antenna polarization term.

    As the differential (post-PCAL) phases are winding very uniformly 
over the 5 hours, it's doubtful that we're seeing a problem based on a 
sensitivity to elevation. 

    I want to emphasize to all that this effect is at a *very* low level 
-- the residual baseline amplitude at any one observation is perhaps 20 
mJy -- 0.1% of the total intensity.  Although all baseline phases rotate 
at the same rate, the phase offsets all appear to be different, so the 
effect on the synthesis image is reduced by roughly a factor of N -- the 
net image rms is less than 0.03% of the peak in I.  That's not bad ...