[evlatests] Sensitivity and Dropouts with Phase Switching off

[Ken Sowinski]

> From: Barry Clark <bclark at aoc.nrao.edu>

> >     For reasons unknown, the data from both 14 and 16, with the
> > phase switching on, was of very poor quality.  Amplitude dropouts --
> > total dropouts -- occurred regularly in both antennas.  These are
> > different than the 10-second interval, partial 'attenuation' that we've
> > reported before.  Thus, unhappily, we have no 'before' state.
> 
> The dropouts I observed earlier were not "total" but reductions by a
> factor of a few.  There is also the interesting fact that the AOC-VLA
> link was saturated at the time, making communications slow and hazzardous.
> I suggest that the cause of the dropouts is that the computers on the
> evla subnet got confused about what the time was, and thus offset the
> phase switching by one or more 52 ms ticks.
> 
> If this is indeed the case, the best solution would be to buy a computer
> with a GPS card to serve as an NTP server.  A much less desirable solution
> would be to lock out NTP time resets on L302s except on reboot or on
> command.

Certainly the network molasses confuses the issue.  I think that
NTP keeps a log of adjustments.  We could look at that for whatever
machine at the VLA is the NTP mediator and see if it was unstable
at that time.  My memory is that NTP is pretty robust about flywheeling
in cases like that.

Rick tried to observe tow half hour segments: one with phase switching
enabled and one not.  We were only able to get a few minutes of data
at the end of the first ending at 2053 IAT.  A combination of the slow
netwrok and mistyping on my part reduced the second half hour to twenty
minutes from 2103 to 2123.  I suspect that the executor only started
when the networked cleared, which according to K Scott, occurred a minute
or two after 2100 allowing the second segment to run with a clear network.

Indeed the earlier dropout seemed to be a different kind of problem 
than what was seen in the first segment yesterday.  It is also true
that those earlier dropouts were seen when the network was not clogged.

I suppose there is nothing for it except to try this again and hope
that the network gods cooperate.

A simple stress test is to use an alternating 1-0 pattern at the 
L302 and turn off phase switching at the correlator.  If everything
works we will see no correlated signal.  If ever the L302 is in the wrong
state for one or more WG cycles we will see a correlated signal of some
fraction of the nominal response.