[evlatests] Results from T304 Attenuator and Requantizer Tests
Rick Perley
rperley at nrao.edu
Sat Oct 22 10:56:53 EDT 2011
Michael and Keith ran some tests Friday afternoon, searching for an
explanation for the closure problems noted on Cygnus A at L-band, and
for the compression of the calibration (switched power) signal when
observing Cygnus A.
The working theory is that we are overdriving the digital system at
L and S bands, in part due to setting the input powers at too high a
level -- this is due to not taking into account that at L and S bands,
the input spectrum does not fully occupy the 5 GHz BW. By setting the
input power level at a value appropriate for full band occupancy, the
spectral power density within the 1 GHz of actual FE bandwidth is a
factor of 5 too high at L-band (7 dB). This is no problem to the analog
circuitry in the wideband parts of the T304, but may be a problem when
the 1 GHz of desired bandwidth at L-band is selected -- now the power
level (not just the spectral power density) will be 7dB higher than
expected. This is the reason why the output T304 attenuators are set to
quite high levels (typically 25 to 31 dB -- the maximum).
This might not be a problem since the output attenuators have the
range to reduce the power level to the desired level -- for cold sky.
But Cygnus A adds another 6 dB of power, which will exceed the output
attenuator's capabilities, and may overdrive the station board
requantizer.
The same argument applies to S-band, but the effect is now at the 3
to 4 dB level.
The tests consisted of four parts:
1) Observe as normal: Set the attenuators on cold sky, and observe
Cygnus and a calibrator alternately.
2) Set the atttenuators on Cygnus A, and observe both sources
alternately, (with no change in attenuator level).
3) Set the attenuators on cold sky, but at a target level -7dB below
the standard (to account for the actual spectral occupancy of the input
signal). (I think this should cause the input T304 attenuator to be 7
dB higher, but leave the output attenuator unchanged?)
4) Set the attenuators on cold sky, at the usual level, but adjust
the stationboard requantizers to prevent overflow when on Cygnus A.
This level would be used for both calibrator and Cygnus A.
Results:
A) The requantizers are most certainly overflowing when the
observations are made with the standard setup (experiment 1). The
requantizer state count distribution shows huge peaks at each end. (The
sampler state counts looked nicely gaussian).
B) The 'PDif Compression' is reduced in experiments 2, 3, and 4, but
is far from eliminated. At L-band, when observing Cygnus A, the typical
PDif Compression is 10 to 20 percent -- nearly all antennas show this.
(Some antenna-IFs are as high as 50%, a few show very little
compression). The compression reduction is modest -- the typical
compression is probably 5 to 15% now.
C) The non-closing effect was determined in the following manner:
The data -- after basic calibration (but not corrected by the switched
power -- this doesn't affect closure, and has other issues -- see
below), I self-calibrated the 1445 MHz data using a 'golden' high
resolution model made with VLA data in 'ancient times'. I then plotted
Stokes 'V' -- this turns out to be quite sensitive to correlator based
problems (provided they are different on the two polarizations). I found:
Expt. 1: Large Stokes V visibilities, with 6 x 17 and
14 x 17 having values nearly equal to the total flux. (This is caused
by the LCP apparently overflowing, giving visibilities near zero).
Expt. 2: All Stokes V visibilities are less than 10 Jy
(which is less than 1% of the Stokes I).
Expt. 3: Large Stokes V found on 17 x 23 and 6 x 17.
Smaller, but still significant values are found on many other correlators.
Expt 4: A single antenna -- ea27 -- had modest (10 to
30 Jy) values of 'V' on its baselines to 2, 3, 4, 12, 18, 25, and 26 --
a behavior quite different than any of the other experiments. All other
baselines showed no 'V' at all.
But, sadly, the situation is not as simple as it seems. The 'V'
test only is sensitive to closure errors which are different between the
polarizations. Another test is to subtract the 'golden' model from the
current data (following self-cal), and looking at the residuals. This
exercise paints a much darker picture:
Large, non-closing, residuals of up to 100 Jy (about 7% of the
total flux) are seen on *some* baselines -- only a few, but their
locations are very odd. The following table lists the antennas and
baselines, and approximate residual:
Antennas Baseline Residual
---------------------------------------------------
1 x 5 w7 x w8 110 Jy
1 x 15 w7 x w9 110 Jy
5 x 15 w8 x w9 50 Jy
3 x 12 e8 x e9 50 Jy
The following baselines have residuals between 30 and 50 Jy:
15 x 20 w9 x n9
1 x 26 w6 x w7
3 x 26 w9 x w6
1 x 4 w5 x w7
3 x 4 e9 x w5
3 x 25 w4 x e9
20 x 28 n8 x n9
3 x 6 e9 x n3
--------------------------------------------------------------------------------------
The distribution of the afflicted antennas is very odd, with a great
preponderance of antennas on the west arm.
This is not RFI -- the calibrator scans -- for all setups -- gave
lovely data.
This is not due to delay errors -- the analyzed data are from a
single 2 Mhz channel at the middle of the band. (SNR is not an issue
for this source!)
This is not due to an error in the 'golden' image -- a 2nd
self-calibration, using the image made from the data itself, produced no
changes to the antenna gains.
*** Summary ***
While we were overdriving the requantizers when observing Cygnus A
-- this is not the cause of the two central problems:
a) The compression in the switched power when observing a strong
source,
b) The apparent introduction of non-closing errors when
observing a strong source.
More information about the evlatests
mailing list