[evlatests] EVLA Flagging Issues
Rick Perley
rperley at nrao.edu
Thu Jan 31 14:11:07 EST 2008
The (long) 'stress test' run yesterday evening has revealed clearly
some significant issues with EVLA flagging. I attempt a summary below.
One 'global' issue have already been aired thoroughly. I simply
summarize it here:
1) The 'lost command' problem, which occurs when going to P-band
from another band, from P-band to another band, and on any scan
following U-band. This results in bad data which are not flagged.
About 10% of antenna-IFs are affected at any one instance. There is no
predictability of which ones are affected. The fundamental cause is
well understood, but its resolution await Barry and Hichem's return from
travel. For the time being, observers need to 'double up' their
observations when the change takes place to minimize the effect (but not
guarantee its absence).
2) Last evening's data showed the following at all bands:
a) 16 B and D were oscillating in amplitude and phase -- and not
flagged. The cause is under investigation.
b) 27A was not fringing at any band -- and all data were correctly
flagged.
c) Antenna 11 went out near the end -- apparently a known
subreflector issue. The data were flagged.
3) We then find a number of band-dependent effects:
a) X-Band: All flags set by the system were correct, and there
were no data (other than 16B and D) which were bad and not flagged.
b) C-Band:
* Antenna 26 A and C were flagged out on one scan, but the data
appear to be fine. However, on this same scan, 26B and D did not
fringe, and the data were not flagged! Do we have some monitor point
reversed?
* 14B and D were flagged out on two scans, but the data appear
fine. OTOH, ant 23 and 25 on all IFs were flagged on one scan --
correctly.
* 1D was flagged throughout the whole observation. The data
were stable, but at 1/4 the proper amplitude. So there is a real
problem, caught by the system. (No other band showed any problem on
this antenna-IF).
c) K-Band:
* At this band, all flags set were correct, and except for
the 'first observation past U-band problem', there were no data that
should have been flagged.
d) Q-Band:
* As noted elsewhere, VLA antenna 8 gave no fringes, but the
data were flagged as 'good'.
* Antnenas 16, 17, 24 and 25, on IFs A and C only, gave no
fringes on a number of scans (not all the same for these antennas), but
there data were not flagged. On the B and D IFs, all data were good. I
would presume this is a LO tuning issue -- but if so, should not the
system detect the incorrect setup, and apply the flags?
e) L-Band: We have a most curious (and perhaps alarming) situation
at this band. Approximately 5 to 10% of all antenna-IFs have been
flagged bad by the system for an entire scan. (In other words, if we
attempted 10 observations with the 12 EVLA antennas, about 10% of the 10
obs *12 ant *4 IFs antenna setups were flagged as bad). The really
interesting part about this is that in nearly all cases, these flags are
set for a single IF only -- not all four at one time, or any pair (such
as R,L, or R1 and R2). The flags affect only one scan at a time. Only
the EVLA antennas are having this happen. In nearly all cases, careful
review of the data show that there is indeed a drop in correlated
amplitude -- in most cases, the drop is quite small -- a few percent.
There is a possible higher flagging probability of the B and D IFs --
but the difference from A and C is small.
So -- what is going on here? Ken believes it likely we are
seeing RFI-caused power fluctuations. If the AGC voltage varies beyond
a certain range, a flag is applied. This is done independently for each
IF. But if this is the case, why are only the EVLA antennas being
affected? If due to the increased EVLA bandwidth, then the RFI is
likely below 1.2, or above 1.9 GHz.
f) P-Band: I didn't have the fortitude or patience to attempt a
detailed comparison of flagged vs. unflagged data. But it's clear we
have numerous instances of antennas flagged bad which are good (antennas
19, 23, 24, 25 and 26 stood out here), and some of antennas which are
bad but not flagged (24D is a good example of this).
More information about the evlatests
mailing list