[evlatests] Curious Startup Troubles for 24-hour stress test

[Rick Perley]

    I'm developing a 24-hour file to permit Operations to execute the 
standard stress test during startup.  Although Ken will shortly be 
implementing this in a much more sophisticated manner, the various 
curious troubles accompanying last night's run indicate some problems. 

    The file which was run begins at 0h and runs through 23h59m.  It 
cycles from P-band through Q band, in a sequence of increasing 
frequency, with one minute dwell for each observation.  To get around 
the 'lost-command' problem observations at P and L band are 'doubled' 
(two consecutive observations at P, followed by two at L).  All others 
are single observations.  P-band observations are of one of 3C48, 3C147, 
or 3C286 -- whichever is highest at the time.  All other observations 
are of one of two far north clean point-source objects.  There is thus a 
significant slew going from Q to P, and from P to L bands. 

    So -- last night's observation began at 02:34 IAT.  This is what 
FILLM showed us:

    1)  The first recorded data are on 3C48 at P-band.  The header 
contains the correct frequency and bandwidth.  There is only single 
record at 02:34:18, and it is flagged bad (which it is). 

    2) the next recorded data is at Q-band -- again a single record, and 
flagged bad (which it is).   But this is the wrong band -- the next 
commanded band in the observe file must be either P, or  L.  So why is 
it recorded as Q?  The recorded observation is at 02:34:55 -- not 
reasonable, since the source (0217+738) is at least one minute of slew 
time away (from 0137+331 = 3C48), yet only 37 seconds have passed from 
the P-band observation.  Furthermore, the recorded frequency is wrong by 
23.4376 MHz (half the difference between the Q-band BW and the P-band BW). 

    3) The next recorded data are at L-band -- also out of sequence, as 
P follows Q, -- at 02:35:15, 20 seconds after the (false?) Q-band 
observation.  This datum is also flagged bad (which it is), and is also 
recorded at the wrong frequency. 

    4) The next recorded data are at C-band -- in sequence now -- at 
02:35:52 to 02:37:08.  The data now look perfectly normal, except for 
one thing -- the frequency is still wrong by 23.4376 MHz.  But -- the 
correlation coefficients are exactly the same as on the following C-band 
observation (16 minutes later), so I think the data really are at the 
correct frequency and BW, but the header is wrong.  (If the observation 
really was centered at 4861.6625 MHz with 50 MHz BW, it doesn't seem 
possible that the correlation coefficients would be exactly the same as 
at the correct frequency of 4885.1 MHz). 

    Following this, all data are recorded correctly, in sequence.