[evlatests] Set-and-Remember Failures
Rick Perley
rperley at nrao.edu
Sat Jan 21 12:42:38 EST 2012
Ken reminded me that it takes something like 40 seconds for
'set-and-remember' to actually remember the appropriate T304 attenuator
settings for any one band. I thus adjusted the durations of the initial
scans for the 30-hour 'flux densities' run to give this much time for
the first source. This appears to have worked well -- for the first 6
hours of this effort, the PDif values (in general) only slowly vary,
reflecting the real gain changes induced by temperature.
Unfortunately, at that point, the script had to be restarted. Since
all scans after the first (except at L and X bands) have durations
between 20 and 30 seconds, and there is no memory of what the settings
were prior to the restart, the new initial observations are all between
20 and 30 seconds, apparently not enough time for the levels to be
remembered. So ... what the system does when there isn't enough time to
'remember'.
The answer isn't that it picks some random value. The PDif data for
the two segments following the two restarts (the major correlator
problem was the second restart) show that the system bounces around
multiple levels, typically three. Each individual scan is apparently
set up with one of three different attenuator settings. Some antennas
may have more (the infamous 'switch issue' confuses the situation for
some antennas) possible levels.
L and X bands are less affected, since the former includes the slew
time from one source to another, and the latter includes the referenced
pointing. The data at all other bands are significantly compromised.
In principle, providing both analog and digital linearity is retained,
the PDif monitoring should remove the gain variations. I'll soon learn
how well this works in practice.
This problem affected 20 of the 30 hours data. Unaffected are the
first 6 hours, and the 4-hour 'makeup' run for the correlator problem,
both of which had the required 40-second minimum for each band on the
initial source.
This raises a question: Since it is clear from the data that the
correct power level is in general determined with ~5 seconds, why does
it take so long for the system to 'remember' this?
Finally: I'm not likely to be the last person to want short
observations with a long script. Assuming that we'll never be able to
guarantee long scripts running successfully with requiring a restart, is
there a way to remember the appropriate level settings from the earlier
portion, or to ensure that the initial scans on any script be extended
long enough for these settings to be remembered?
More information about the evlatests
mailing list