[evlatests] L-Band Sensitivity -- Revised
Rick Perley
rperley at aoc.nrao.edu
Wed Feb 22 14:05:49 EST 2006
I have reviewed the spectral line data from the weekend tests, with the
conclusion that many of the problems noted in yesterday's report are
erroneous, or less serious than state.
1) Closure:
The spectral line data show no significant closure. The levels seen
(1 to 3%)
are compatible with those expected from real astronomical structure in the
source (or, more likely, in the background around it). The values reported
yesterday from the continuum data were caused by the residual phase slope
differences between VLA and EVLA antennas.
This has the effect of lowering the apparent amplitude of the
signal between
VLA -- EVLA antennas, causing the AIPS program CALIB to overestimate the
necessary gain correction. Application of these overestimates causes the
EVLA-EVLA baselines to show too much amplitude -- this is seen as a closure
error.
This same effect will bias the AIPS weights. The proper approach is to
utilize the spectral line data for sensitivity estimates.
2) Sensitivity:
This was done in two ways.
2-A) Aips Weights. After bandpass correction (which removes the
problems
of the phase slope noted above), the data were calibrated in the usual
sense. The
resulting aipsweights show that for IF 'A', antennas 13 14 and 16 all
lie at the bottom
end of the sensitivity distribution. The ratios between these antennas
and the median
of the distribution are: 0.76, 0.80 and 0.88 for antennas 16, 13 and
14, respectively.
These can be interpreted as a reduction in efficiency, or in (inverse)
Tsys, by these
factors. Thus, at 1465 MHz, in RCP, we do have a significant loss in
sensitivity.
However, the story is very different in IF 'B' -- at 1385 MHz.
Here, the aips
weights for these three antennas lie exactly within the median of the
overall
distribution. (Amongst these 3 antennas, 13 remains the worst, and 14
the best --
as in IF 'A'). There appears to be no problem at this frequency in RCP.
2-B) AIPS weights are *estimates* of the antenna sensitivity --
they are not
generated from real correlator output. To check that these estimates
can be
trusted, I plotted the histograms of the real part of the calibrated
visibilities for
the observation taken on blank sky. Two histograms were made -- one for
EVLA
antennas 13, 14 and 16, and one for VLA antennas 6, 8, 9, 10, and 17. These
were chosen as they span the same spacing range as the EVLA antennas.
(Note:
antenna coupling effects, and background astronomical structure can conspire
to increase the rms at short spacings, particularly at low frequencies.
This effect
is quite spectacular at 74 MHz, bad at 327 MHz, and may be notable at
1400 Mhz).
The two histograms (EVLA and VLA) have the same width, to within the
errors accompanying so little data.
I conclude that the aipsweights are a trustworthy measure of antenna
sensitivity
for both VLA and EVLA antennas, and that, for this frequency and
polarization,
we have no significant sensitivity loss.
It would be good to have an explanation for the modest loss at 1465
MHz.
3) Other effects. The Tsys/visibility amplitude effects noted
yesterday remain.
Some of these are likely related to large changes in input power to the
T304s,
as I was changing frequencies rapidly, and antenna 13 (inparticular) showed
exponential-like changes in Tsys and visibility amplitude at the
beginning of
each L-band scan, of the type expected if the input power abruptly
changes.
The peculiar spike seen in antenna 14 is not caused by this -- it is in
the middle
of a scan, and remains mysterious -- although rare.
More information about the evlatests
mailing list