[evlatests] Gain stability issues
Rick Perley
rperley at nrao.edu
Wed Feb 3 14:13:43 EST 2010
The decision to not normalize correlator cross products in WIDAR
raises the issue of receiver gain stability, and the need to monitor
switched power for the purpose of correcting product fluctuations due to
gain changes. I've talked with Bob Hayward on these issues, and have
briefly reviewed our antenna tests to estimate gain variability.
The receiver group has not done testing for long-term gain
stability. Bob's impressions from SOIDA tests, when repeated on
hour-long intervals, is that output power is stable to ~1% or better --
and this instability might be due to system temperature changes rather
than receiver gain changes.
The on-sky tests at various bands that Bob and I have done over the
past few years gives a better measure -- but in general these apply only
to the front-end, and do not include any gain variabilities in the IF.
I found three tests where we have observed with the cold load on over
fairly widely separated times -- this is a good test for receiver gain
changes since the load (source) is absolutely stable (liquid nitrogen)
-- although again a change in output power could conceivably come from
system temperature change, rather than gain.
For the three tests (there may be more, but I have hundreds of these
plots, and sorting through them all will take some time), the news is
generally good: At 43 GHz, over 2 hours, the cold-load power changed by
less than 0.5%. At 33 GHz, over 30 minutes, a change of 0.4% was
observed. However, on a third test, at 23 GHz, I note a change of 2%
over 1.5 hours.
Overall, I would judge that the front-ends are gain stable to ~1% or
better on hourly time scales. In would judge this as perfectly
acceptable for most early science -- without the ability to monitor gain
changes -- provided the T304 attenuators are indeed kept fixed for any
given band throughout an observing run.
Unknown is the variability in the IF chain. Perhaps a knowledgeable
person can offer an opinion here.
In the new scheme, system temperatures are not needed for gain
corrections, but will be needed to establish proper weighting for
imaging. How essential is this? At the lower frequencies,
gain-elevation effects are quite small, and the antennas and receivers
are all very similar. I offer the opinion that detailed Tsys values are
not essential for early science at these frequencies. At high
frequencies, the variations become much larger, particularly at the
upper end of Q-band where we have significant atmospheric opacity. But,
given the general uniformity (and knowledge) of zenith system
temperatures, and our general understanding of opacity, a model should
suffice to allow reasonable estimation of system temperature, sufficient
for most applications. (Is such a program available in AIPS?)
A related issue is the stability of the switched power source. In
the end, we will rely on this for accurate gain correction, and accurate
estimation of system temperatures. Bob H. tells me that the noise
diodes are not in a deliberately temperature stabilized environment.
However, their power coefficients with temperature are not expected to
be large (he will find some numbers for us later today), and in general,
we don't expect large changes in ambient temperature in the vertex
room. The situation for the EVLA here is not different than the VLA,
except that the vertex room is now larger and more exposed to outside
temperature changes. We do have some history to rely on here: Each of
the flux density runs that I calibrate have shown amazing repeatability
in the visibility ratios as determined throughout the typical 30-hour
observations -- at the lower frequencies where the SNR is very high,
variations in the flux density ratio between any two sources are much
less than 1% over 30 hours or more -- indicating that the noise diodes
must be stable to this level over that time duration. The two most
recent flux density runs made heavy use of EVLA antennas. For these, it
is certainly notable that many (but not all) EVLA antennas were not as
stable as they were when in VLA times -- this might be due to variations
in switched power, but may also be due to back-conversion from the
digital to analog data needed to feed them to the VLA correlator. We
won't know about this until we begin testing with the new system.
There is much to learn here. But my sense is that the systems will
be sufficiently stable (provided attenuators don't change) that
unmonitored cross-products should suffice for early science. (But I'd
feel better if the full mechanisms were available soon...)
More information about the evlatests
mailing list