[evlatests] Scan boundary checks

[Michael Rupen]

We did three tests today to check whether data labeling is consistent
at scan boundaries.

In each case we observed 3C84 continuously.  The first two scans were
1min, and all subsequent scans were 30sec long.  We alternated between
standard attenuator settings (-30 dBm) and -35 dBm.

In an ideal world we naively expect to see abrupt amp changes at scan
boundaries, with no associated phase changes.

1- Straight OPT script (no mods except to give the attenuator settings
   described above), 1sec dumps

   In this case the last record (1sec integration) is missing for every scan,
   and the first record of each scan has amp partway in between the low and
   the high amp states.  Apart from the first and last records, the
   data behave as naively expected.

   These oddities are likely due to the mismatch between LST and UT.
   Correlator integrations are in UT, counted from the previous UT midnight;
   LST start times will be slightly offset from those UT integrations.
   Martin thinks the CBE will simply toss a record which isn't as long
   as it should be, leading to the lost last record.  The first
   record is actually a mix of the previous and the current scan, leading
   to an intermediate amplitude.

   This hypothesis was confirmed by test #2...

2- As #1, but with the OPT script modified so that all scan begin
   on UT 30second boundaries.

   In this case the last record of each scan is fine, but the first
   record of each scan is mis-labeled: that record ought to have been
   associated with the previous scan.

   [We also see slight (few degree) phase jumps on some baselines when
   the attenuator changes, followed by inverse phase jumps (i.e., return to
   the previous phase state) when the attenuator changes back again at the
   next scan.]

3- As #2, but modified to use 100msec integrations, and using only four
   antennas.

   Again the last record of each scan is fine, but the first
   record of each scan is mis-labeled: that record ought to have been
   associated with the previous scan.

   The fact that this is integration-based suggests that this
   is a CBE problem, but we'll see what Martin says on the morrow.

             -- Michael