[evlatests] Troubles with AR646 -- last night

[Rick Perley]

    This is the first of many, many night-time observations at L-band, 
mode '4', 25 MHz BW, 3.3 sec integration. 

    In IF 'D' (LCP Stokes, IF#2), some very large values are seen on 
some baselines, at some times.  Nothing like this is seen in RCP, which 
essentially eliminates RFI as a source.  The first IF is not affected. 

    The 'signal' has some curious characteristics:

    1) It abruptly comes and goes.  There is no rise or decline time.  
Either a record has it, or it doesn't.  Perhaps 5% of the timestamps are 
affected. 
    2) Only a minority of correlators are affected.  When one of these 
has it, they all have it, and vice versa.
    3) The amplitudes are steady, and about 3 orders of magnitude too 
high.  (Not subtle!)
    4) The phases are always one of:  0, 90, 180, or 270, and are 
constant in time for any given channel.
    5) The phases advance by exactly 90 degrees, as a function of 
increasing channel number, for any given affected baseline. 

    It seems to me this must be a correlator malfunction.  I now recall 
seeing a smaller version of this in some of my previous tests.