[evlatests] Highly Unstable Gains at High Frequencies (aka UX Converter Switch Problem)

[Rick Perley]

     Some important new information has been obtained concerning the 
long-standing switch problem in the UX converter.


A) Executive Summary (for the impatient).

     The amplitude gain instability noted long ago when switching bands 
amongst Ku, K, Ka and Q bands is now shown to:

     a) Affect phases with amplitudes of tens of degrees.

     b) Be completely independent between polarizations (so that the 
effect above can hugely affect polarimetry).

     c) Affect *only* the A1C1 and A2C2 IF channels.  The data taken in 
B1D1 and B2D2 are completely stable.


B) Details (for those whose inquiring minds want to know):

     1) Introduction:

     The 'flux density' program data, taken years ago, showed that the 
antenna gains at Ku, K, Ka and Q bands commonly fluctuate on time scales 
as short as 1 second, with amplitudes of tens of percent.  The effect 
was independent between polarizations.  Some antennas were considerably 
worse than others.  The effect is not seen at X, C, S, or L bands.  It 
was subsequently deduced that the effect is caused by faulty switches 
within the UX converter.

     The switched power system detects these changes, and (in principle, 
although not in practice) could be employed to (mostly) remove the 
variable gain.

     This problem seemingly affects only data taken with programs which 
those change bands.  ('Seemingly' means that since I never observe at a 
single band, I cannot personally conclude the effect is not present for 
those observing at a single high frequency band).

     2) New Data:

     The results here come from the 3C273 program data, taken this 
month.  In this we (Jean Eilek and I) observe 3C273, a nearby calibrator 
(J1224+0330) and the standard calibrator 3C286, at X through Ka bands.  
Because this is a precision experiment (for both total intensity and 
polarization), I have scrutinized the data with unusually high care.  
(!)  Some of the problems found have already been reported via this 
list.  (Warning:  There are more to come!)

     The data are taken with 1-second averaging, and utilize the full 8 
GHz of bandwidth afforded by the JVLA electronics.  To enable precision 
calibration, I solved for delay and phases for every single integration 
record.  Amplitude gains are subsequently derived, again for every 
1-second integration.

     Basic results are illustrated in the six attached plots, taken from 
the Ka-band data.  The K and Ku band data show exactly the same 
characteristics.

     3) Results:

     The way things *ought* to be are shown in the three plots named *B2D2*:

     a) Ka-B2D2-R-L-Phs.png shows the R-L phase (phase difference 
between the opposite polarizations) from the delay solution, for each 
1-second solution.  3C286 in red, J1224+0330 in green, 3C273 in blue.  
'Phase Rocking' is not visible since the solution bandpass is centered 
at the central channel.  Atmospheric phases (which vary rapidly) are not 
seen since this is a difference plot.  Antenna 04 was selected as the 
reference (for reasons to be noted below).

     b) Ka-B2D-RL-Amp.png shows the amplitude solutions difference, for 
the same antennas.

     c) Ka-B2D2-R-L-PDif.png shows the difference in the switched power, 
for these antennas.

     For all three, the system is quite stable.

     The picture is *very different* for the A1C1 pair, as shown in the 
remaining three plots.

     d) Ka-A1C1-R-L-Phs.png shows the R-L phase difference for antennas 
3, 10, and 28, using ea04 as the reference.  Phase differences of 15 -- 
20 degrees are noted on all three, with timescales of seconds to tens of 
seconds.  Some other antennas have larger phase difference.  Nearly all 
antennas show the problem.  (There are about six antennas which seem to 
be stable).

     e) Ka-A1C1-RL-Amp.png shows the difference in amplitude solutions.  
The short-term variability in the phases is exactly reproduced in the 
amplitudes.

     f) Ka-A1C1-R-LPDif.png shows the difference in the switched power 
monitor data.  There is an excellent correspondence in the short-term 
variability between these and the phase and amplitude differences.  
(Note that ea04's PDif differences are smooth and uniform -- a good 
means to find the stable antennas).

     Review of all the data taken in the program show that:

     * Only the A1C1 and A2C2 IFs show this problem.  B2D2 and B1D1 are 
clean for all three high frequency bands observed in this program.

     * The X-band data do not demonstrate any variability.  (I utilized 
B1D1 and B2D2 for the 3-bit samplers, and A0C0 for the 8-bit samplers).

     * There is no correlation in the variability between opposite 
polarizations on any antenna, at any band.

     4) Implications.

     It has been argued that, since the amplitude fluctuations are 
detected by the switched power system, the gain fluctuations can be 
removed.  If if this were viable, such a correction scheme will not 
detect the accompanying phase fluctuations.   Besides this, the switched 
power data are quite noisy, and it is generally recommended that the 
data be signifcantly smoothed prior to application.  This cannot be done 
if the fluctuation timescales are of order 1 second.

     The phase fluctuations are significant, although generally not 
large enough to seriously degrade the system sensitivity. They will 
significantly degrade image fidelity -- however, for sources strong 
enough for which this is important, self-calibration should accurately 
remove both the amplitude and phase effects.

     The effect on accurate polarimetry can be more severe, if the 
'wrong' antenna is chosen as the phase reference.  A 'wrong' antenna is 
defined as one which suffers differential phases offsets (here, caused 
by the bad switches).   Identifying the 'right' reference antenna is not 
as simple as may be thought -- since nearly all antennas are 
differentially changing phase on different time scales, uniquely 
identifying an antenna whose two polarizations are stable can be 
challenging.  I suspect examination of the PDif differences (as noted 
above) may be the best way to find the 'right' reference antenna.

     5) The RIGHT Solution

     Replace those affected switches.




-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ka-A1C1-RL-Amp.png
Type: image/png
Size: 71210 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0006.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ka-A1C1-R-L-PDif.png
Type: image/png
Size: 92059 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0007.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ka-A1C1-R-L-Phs.png
Type: image/png
Size: 63563 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0008.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ka-B2D2-RL-Amp.png
Type: image/png
Size: 66710 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0009.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Ka-B2D2-R-L-PDif.png
Type: image/png
Size: 93546 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0010.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: K1-B2D2-R-L-Phs.png
Type: image/png
Size: 53583 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20180426/03136485/attachment-0011.png>