[evlatests] Testing faster setup times (attenuators and requantizers)
Rick Perley
rperley at nrao.edu
Tue Nov 26 16:32:26 EST 2013
Ken et al. have improved the setup times for 3-bit observations,
which should provide considerable improvement in observing efficiency,
especially for programs which change the correlator modes. The (as yet
unofficial) new recommendation is that 60 seconds be allocated for the
first observation of any frequency tuning, and 20 seconds be allocated
for the first observation following a change of correlator mode or
frequency tuning.
I was asked to give this new regimen a stiff test. This was done
last evening. The results are very good.
For the test, I observed four bands: L, S, C, and X. The first two
were observed in 8-bit mode, the last two in 3-bit mode. The sequence
of observations were as follows:
0) Slew to source at X-band (narrow).
1) Setup for 1 minute at S band (narrow)
2) Setup for 1 minute at L-band (narrow)
3) Setup for 1 minute at C-band (wide)
4) Setup for 1 minute at X-band (wide)
5) Move to calibrator (X-band, narrow)
6) Referenced pointing on calibrator (X-band, narrow)
Following this, five loops of the following were done:
1) Calibrator at L
2) Source at L
3) Source at S
4) Calibrator at S
5) 20 seconds at C-band (wide) for requantizer
6) Calibrator at C (wide)
7) Source at C (wide)
8) 20 seconds at X-band (wide) for requantizer
9) Source at X (wide)
10) Calibrator at X (wide)
To judge the effectiveness of this, I reviewed the visibilities and
switched power for selected antennas and spectral windows.
Results:
A) First-scan gain setups.
At L and S bands, the attenuator level setting changes were all
completed in less than 15 seconds. There was a rather wide range in
final power (PSum), however.
At L-band: For four antennas x 2 IFs that I looked at
carefully, the spread in power ranged from 6 through 23 counts, with
most between 8 and 10. The optimum level is said to be 14 counts.
At S-band: The spread in powers ran from 5 to 10 counts --
all are well below what I had believed to be the optimum.
At C and X bands, the behavior is very different than at L or S:
There are many different, oscillating power levels attempted before the
system settles down to a single stable value. In all cases, this
process was completed within 50 seconds.
At C-band, all power levels settled on were 10 to 15 counts.
But at X-band, a very different behavior: About half the
antennas had power levels at 100 or more *kilo*counts! Despite this,
the output visibilities are in the correct range -- because the
requantizer gains were set at a corresponding high level (1.5, rather
than 0.011). Further on in the file, these crazy levels disappeared,
and all seems normal ...
B) Requantizer Gain setups.
For both C and X bands, these occurred very quickly: 8 to 10
seconds. And most of this time was presumably taken up in rotating the
subreflectors. (The motion was from S to C, and from C to X, so both
changes had small rotations). For both bands, the resulting output
powers were very uniform amongst the antennas -- 14 to 15 counts.
Conclusion:
From this test, the 60 seconds initial setup is just sufficient.
The 20 second requantizer change is generous, but only because of the
chosen bands in my test.
More information about the evlatests
mailing list