[evlatests] Accuracy of Switched Power Application
Rick Perley
rperley at nrao.edu
Wed Apr 4 15:23:01 EDT 2012
I have now reviewed the accuracy of the switched power application
at all bands. The results are rather interesting, and certainly
indicate we have some work to do to get this system into a reasonably
uniform state.
Method:
I used my flux calibrator observations, taken in mid January. The
model opacities, system delays, bandpass function, and switched power
were all applied. To avoid the complication of the elevation dependency
(see below for discussion), I used an observation of 3C286 at 60 degrees
elevation to calculate the system gains.
The weather during this run was clear, and winds were light to
moderate. The time of the observation of 3C286 was 3AM in the morning
-- about ideal for a reliable estimate. Referenced pointing was
determined on 3C286 itself. I utilized the publicly available models
for the gain estimate -- these are not critical as the array was in DnC
configuration at the time.
For each band, I selected two frequencies to generate gains from --
one from the 'AC' side, the other from the 'BD' side. For all unflagged
antenna gains, I drew histograms of the gain distribution. Analysis of
the dependencies (see Memo 145 for details) shows that if:
a) The antenna ('AIPS') gain is too high, then the applied value of
Tcal is too low, and/or the applied value of the antenna efficiency is
too high.
b) The antenna gain is too low, then the Tcal value is too high,
and/or the antenna efficiency is too low.
A perfect application means the AIPS gain is 1.0. For most bands,
the median correction is within 5% of this value (as Vivek and I have
always claimed). But for most bands, there is a disturbingly wide range
of values (as everybody else who has looked at this has always
claimed). A gain correction less than 1 means the visibility values
are higher than they should be.
For the table below, I give the median correction from all the
unflagged antenna gains, and a list of especially discrepant antennas.
Band/IF Frequency Median Gain Especially bad antennas
-----------------------------------------------------------------------------------------------
L-A 1465 0.95 Antennas 19, 25, and 28
L-C 1465 0.96 Same as above
L-B 1865 0.96 5, 14, 19, 25, 28
L-D 1865 0.95 8, 14, 19, 25, 28
S-A 2565 1.00 7, 9, 10, 12, 16, 24 28
S-C 2565 0.98 7, 9, 10, 12, 16, 28
S-B 3565 0.94 7, 9, 10, 12, 15, 16
S-D 3565 0.92 7, 9, 10, 12, 15, 16
C-A 4885 0.96 24
C-C 4885 0.98 none
C-B 6885 0.94 17
C-D 6885 0.94 none
X-A 8435 0.94 2, 3, 4, 9, 17, 20, 21, 22, 27
X-C 8435 0.94 2, 3, 4, 9, 12, 20, 21,
22, 23, 27
X-B 11064 0.96 14
X-D 11064 0.92 17
U-A 14965 0.96 7, 17, 19, 22, 23, 24
U-C 14965 0.99 13, 19, 22, 23, 25
U-B 17422 1.02 17, 19, 23
U-D 17422 1.01 17, 23
K-B 22485 1.35 all of them!
K-D 22485 1.35 all of them
K-A 25836 0.85 17, 20
K-C 25836 0.85 17, 20
A-B 28386 1.03 2, 3, 5, 7, 13, 14, 15, 17, 20
A-D 28386 1.03 2, 3, 9, 14, 15, 17, 28
A-A 36435 1.00 2, 3, 8, 9, 14, 15, 17, 24, 26
A-C 36435 1.00 2, 3, 8, 12, 13 14, 28
Q-B 43340 1.02 3, 20, 22
Q-D 43340 1.02 3, 16, 20, 22
Q-A 48425 1.15 3, 16, 20, 22
Q-C 48425 1.15 16, 20, 22
------------------------------------------------------------------------------------------------------
Comments:
1) The median value discrepancy (from 1.0) could be due to an
incorrect efficiency used in the correction procedure, or a repeatable
offset in the values for Tcal. The lower frequencies all have values
less than 1.0, indicating the tabled efficiencies are too high, or the
Tcals too low. The opposite effect is seen at high frequencies (except
K-band, which is special).
2) K-band is different than all others. On the B-D side (22.5 GHz),
the spread in gains is huge -- from 0.9 through 2.0, with a uniform
distribution. Such a spread can only be due to incorrect Tcal values --
I cannot imagine the antenna efficiencies covering such a range
(especially since it goes as the square of the gain range!). On the A-C
side, the spread is not as wide (0.6 through 1.2 -- although this is
fractionally the same), and the median is offset in the opposite way.
The K-band receivers are all 'old' -- would this have resulted in all
incorrect Tcal values?
3) The high-frequency Q-band median is high. This is likely due to
either an erroneously high efficiency (most likely, in my opinion), or
to an incorrect opacity estimate (in this case the calculated opacity is
too low).
4) The list of discrepant antennas was derived by looking at the
histograms. For some bands, it's hard to discriminate between 'good'
and 'bad' -- the distribution is broad and continuous. Notably
non-uniform bands are S, X (likely an old vs. new receiver issue), K,
and Ka.
Recommended Actions:
1) The discrepant antennas need to be addressed first. In some
cases, the discrepancies are extremely large -- even exceeding a factor
of two (both high and low) in amplitude (a factor of four in equivalent
power).
2) We should confirm the median offsets are due to to incorrect
efficiencies, and adjust this parameter. (Alternatively, confirm some
offset in Tcal measures, and adjust this).
3) The problem with K-band needs an explanation.
4) An automated procedure to generate these gains needs to be set
up. Emmanuel is doing this as part of the array checkout procedure that
we are developing.
More information about the evlatests
mailing list