[evlatests] Effects of S-band DSR on visibility data.

[Rick Perley]

    Two trial stop-band filters were installed on ea21 and ea24 to 
assess their effect on the EVLA data.  The two filters are from 
different manufacturers, and both cover the range from 2310 through 2360 
MHz. 
    The test was to point at a northern calibrator, and nearby blank sky 
for about 10 minutes each, alternating back and forth between the two.  
Because the time of the observation wasn't known in advance, I used the 
most northerly source of reasonable flux I could find:  J1153+8058.  
Unfortunately, the observation was made at an HA of 10.5, meaning the 
elevation was only 25 degrees, with the result that both antennas with 
the filters were partially shadowed.   (These antennas are on E1 and W1, 
respectively).  The shadowing means that amplitude calibration is a bit 
dubious, and that there are cross-talk issues evident on all short 
spacings, including the one between the two filtered antennas. 
    Despite this, most of the results are quite clear.  Note that in 
pointing to the far north, I arranged the DSR signals to be about as 
weak as they can be -- the two XM satellites are geostationary (so in 
the south, near dec = -6), while the three Sirius satellites are in a 
highly elliptical orbit, seen roughly from NE through SE. 

    I looked for three different effects:

    1) The effect of the filters on the data within the subband 
containing the DSR emissions.

    2) The effect of the filters on data in adjoining subbands, but 
within the same IF. 

    3) Whether there are any differences in subbands within IFs which 
contain DSR and subbands within IFs which do not contain the DSR signals. 

    Results:

    1) The filters certainly do the job they are intended to do -- the 
cross-power spectra in subband 3 (2.245 -- 2.373 GHz for the setup I 
used) for baselines containing either ea21 and ea24 have a 'black hole' 
where the emissions are.  Unfortunately, there are no useful 
autocorrelation spectra for unfiltered antennas, since the DSR signals 
cause the accumulators to overflow.  (Ken says he will fix things so I 
can get useful autocorrelation spectra in the future).   Bandpass 
solutions were made, and show that the channels more than a few MHz away 
from the DSR signals give good solutions -- for antennas with, and 
without, filters.  That we get good solutions even within the affected 
subband suggests that the DSR signals are not strong enough to degrade 
observing -- and this was borne out by generating noise histograms for 
baselines containing the filtered antennas, and comparing these to 
baselines with unfiltered antennas.  In general, it appears that the 
data within the subband, but not directly wiped out by the DSR 
emissions, are perfectly good.  I generated images of the blank field 
for those unaffected channels, and although they are not quite as good 
as images made in unaffected subbands, they certainly show the 
background sources. 
    The baseline 21 x 24 has much higher noise than any baselines 
combining an unfiltered antenna to a filtered one -- but it turns out 
this is a proximity effect (as 21 and 24 are adjacent) -- other 
baselines of similar length between unfiltered antennas show exactly the 
same rise in noise. 
    The switched power (both PDif and PSum) for unfiltered antennas is 
certainly messed up -- and this problem is nicely corrected by the 
filters.  However, this is not a significant advantage, as one can use 
adjacent 'clean' subbands to monitor system gain variations. 

    2) Given the above, it is no surprise that other than the wings of 
the filters modifying the phase, there is no discernible effect of the 
filters on adjacent subbands.  Noise on baselines with the filters looks 
just like that on baselines without.  The filters are expected to show 
phase effects in the wings outside the stop-band -- this is easily seen, 
with a phase slope of about a half-turn over ~30 MHz seen on both sides, 
and on both filters. 

    From these results, it seems the filters provide no useful 
advantage, so far as the quality of the visibility data are concerned.  
They would still be useful (and needed) if the switched power in the 
'clean' subbands have been affected by the DSR signals.  I looked 
carefully at PSum, PDif, and Tsys for 'clean' subbands in the antennas 
with, and without, the filters.  No differences can be seen. 

    Finally, I arranged another experiment where subbands within the 
affected IF were observed with a new tuning which excluded the DSR 
signals, to see if their presence caused imaging problems even in 
subbands far from the DSR signals.  Specifically, there were two tunings:
    1) The first used two IF pairs (AC and BD), tuned to 2.5 and 3.5 
GHz, providing 2 MHz channel resolution.  This setup utilized 128 
MHz-wide subbands, with 2 MHz resolution, and provided the data reported 
above.
    2) I made a second tuning, now using 64 MHz-wide subbands, with the 
two IFpairs centered at 2.744 and 3.192 GHz, thus providing 1 MHz 
resolution.  This observation followed the first by a few minutes. 

    I then compared the noises for frequencies common between the two 
setups.  The results show that the noises for the second setup (with 1 
MHz-wide channels) is a few percent lower than that seen in the first 
setup (properly including the sqrt(2) difference due to different 
channelwidths).  But this improvement is the same for frequencies below 
3 GHz as above 3 GHz.  (In the first experiment, the DSR signals are in 
the A/C IFpair, but not in the B/D pair.  In the second, they are absent 
in both IFpairs).  I conclude that the noted improvement is unrelated to 
the presence of the DSR signals. 

    Bottom Line:  At least for observations in the far north, I see no 
useful improvement in data quality for antennas including the DSR 
stop-band filters. 

    But before abandoning this initiative, I suggest another test, to be 
done after the reconfiguration is complete:  To observe a source, and 
nearby cold sky, at a position closer to dec=0.  The filtered antennas 
will be further apart, so that shadowing and proximity effects will be 
essentially removed.