[evlatests] 3 and 8 bit comparisons
Rick Perley
rperley at nrao.edu
Thu Sep 2 11:45:05 EDT 2010
Vivek gave a brief report on yesterday's short tests.
I've also reviewed the data, and have a few additional observations.
The observations were in an OSRO-type mode, a single subband for the
3-bit path, and a pair of subbands for the 8-bit path. CBand, near 5
GHz. BW = 128 MHz, Channelwidth = 2 MHz. Alternating between 3C286 and
a nearby blank field, one minute per field. About 30 minutes duration
for each.
A) Amplitude Stability.
Excellent for 8-bit -- no gain changes of any kind.
For 3-bit, we have the usual change of gain at the beginning, due to
warming up. Because of antenna slew, we did not see the first 2 minutes
of the run -- when the antennas got on source, all samplers *except* 22R
and 27R were fully warmed up, as no coherent gain changes were seen in
the data.
Gain stability from scan to scan at 3-bit varies a lot. 15R, 15L,
and 22L are as good as the 8-bit path. 27R is almost as good. 22R,
27L, 28R, and 28L show notable changes from scan to scan, typically of
order 5% in power, sometimes a few times higher.
B) Gain Amplitudes.
For 8-bit, all gains were nearly the same. But for 3-bit, they were
notably different. A table will be useful ...
8-bit 3-bit
-----------------------------------------
15R 6.25 6.95
15L 6.15 7.2
22R 5.80 12.0 < -- NB!
22L 6.65 5.4
27R 6.30 11.0 < -- NB!
27L 7.00 6.3
28R 6.10 7.7
28L 6.20 6.60
--------------------------------------
The sense of these gains is that a high gain means a low power ---
22R and 27R are weak, by a factor of about 5 dB in power. Remember
this when we get to the sensitivity plots ...
C) Phase stability.
Good, and the same, for both paths. (Except for antenna 27 on the
BD IFs -- I'll send a special note about this).
D) Bandpass stability
Very good for both paths. No changes over the 30 minutes above 1%,
and probably much better than this, as these short scans don't give very
high SNR. However, the individual solutions for the 3-bit path are
clearly noisier than those for 8-bit. For an explanation, read on ...
E) Sensitivity.
3C286's spectrum is known, so we can determine the absolute
sensitivity by using the blank field noise statistics. The
histogram-plotting program UVHGM is excellent for this purpose. I
selected 15 minutes of the test when all gains are stable, and
determined the baseline noise statistics, after applying both the gain
and bandpass corrections. I utilized the central 45 channels to improve
the statistics. The results are below:
rms noise, in mJy, per baseline/polarization for the 3-bit path.
RCP in the upper right, LCP in the lower left.
15 22 27 28
------------------------------------------------
15 | xx 310 240 210
22 | 190 xx 370 310
27 | 210 170 xx 250
28 | 230 190 210 xx
----------------------------------------------
For the 8-bit path, no matrix is required -- all combinations gave
160 mJy.
There is an excellent correlation between the high noise entries in
the above table and the high gain corrections listed in Section B.
Indeed, the only entry in the noise matrix which is at the correct level
(22 x 27 in LCP) is from the two antennas-IFs with the lowest gain
corrections. And the inverse is most certainly true -- the two
antenna-IFs with the highest required gain corrections (22 and 27 in
RCP) have by far the highest noise in the cross-correlations. From this
I deduce that the powers into the samplers are -- in effect -- too low.
What is puzzling to me is Ken's statement that the powers were carefully
adjusted before this experiment to give the same width in the state
counts. Do I misunderstand something?
More information about the evlatests
mailing list