[evlatests] Correlator Scaling Errors in IFs B and D.

[Rick Perley]

    I ran an L-band test yesterday afternoon, to check on the wider 
tuning ranges on VLA antennas enabled by the removal of the FS 
filters.    The observations were taken in spectral line mode '4', with 
12.5 MHz BW (thus, 16 channels in each of the four correlations).    The 
integration time used was 1.67 seconds. 

    It was quickly apparent that an unusual form of what initially 
looked like interference afflicted the data.  However, closer inspection 
reveals that this  is definitely not RFI, but a curious scaling error, 
affecting some antennas at particular times.   These characterstics are:

    1) All the visibility amplitudes for one or two antennas, on all 
spectral channels, on one polarization (correlation) only, are 
multiplied up by a factor of about 10.  If two antennas had this happen, 
then their correlation was multiplied up by the product of the factors 
applying to each.  That is --- we have an antenna-based scaling error.   
(In principle, the data could be salvaged by a simple self-calibration, 
if the source is strong enough). 

    2) The phases are not affected -- so the real and imaginary parts 
are equally multiplied. 

    3) The effect occurs on single integrations -- adjacent integrations 
are normal. 

    4)  The effect is seen on IF#2 (Band D) only.   The scaling error is 
seen in both B and D, but there is no correlation between them. 

    5) The effect is infrequent -- about 1% of the times are affect. 

    There are visibility spikes visible in IF#1 (A and C) also -- but 
these are much rarer, and appear completely random in antenna, time, 
baseline, amplitude, and phase. 

    For anybody wanting to look at logs, the following three times 
showed the effect (all times in IAT), for the listed antennas and 
polarization:

    00:20:44   Antennas 10 and 28, in LCP (IF D).
    00:21:06   Antennas 10 and 14, in RCP (IF B).
    00:21:28   Antennas 10 and 14, in LCP.