[evlatests] L-Band WIDAR Woes

[Rick Perley]

    I spent most of yesterday carefully working on our 1-hour L-band 
dataset taken on the northern point-source, 0217+738, in the hope of 
finding some sort of clues to help us figure out what is wrong. 

    I failed. 

    But perhaps the methods I used, and the results obtained, may be of 
some interest or use, so I review these in some detail.   I apologize 
for the length and depth of detail, but we probably need to record all 
this to prevent future repetitions ...

    1) The duration of the test was about 1 hour, at HA = 9.  The source 
was in the NW, at a low elevation (about 20 -- 30 degrees).  The data 
were taken June 29 in a single block, with scans of length 1 minute.  
The integration time was 1 second.  The source strength is 2.4 Jy -- a 
strong calibrator, known to have no structure, in an empty part of the 
sky (other than about 100 mJy of background sources within the 30 
arcminute primary beam). 

    2) Four sub-bands, each of 1024 channels were set up. Two of these, 
centered near 1050 and 1565, were so full of RFI that I made no attempt 
to calibrate them.  The other two, centered near 1180 and 1950, are 
relatively clean.  I used the program PHNEG to negate the phases for all 
visibilities. 

    3) The basic spectra are exactly as expected, with RFI in all the 
usual places, and 'open regions' contained stable data.  The delay 
finding program FRING was run on all four sub-bands, with stable 
results.  With antenna 1 as reference, antennas 2 and 9 both had large 
delays (about 90 nsec!), all the others were less than 10 nsec.  The 
stability of these delays over the length of the run, and between the 
subbands, is very good -- about 1 nsec.  I used CLCOR to correct for 
these delays, and checked that the resulting phase gradients were 
negligibly small using POSSM.  I checked every single baseline to make 
sure there were no correlator-based unique delay offsets.  All looks 
good at this point. 

    4) CALIB was run, using a small (100 channel) subset of channels 
taken from mid-band.  I limited the solution to long spacings,  as it 
appears these are less susceptible to the issues discussed below.  Other 
than some curious phase jumps within a minute of the start of the run, 
all antenna phases are continuous and smooth.  There are no jumps or 
ramps.  The amplitudes show numerous examples of 1db T304 attenuator 
changes (these are *very* annoying!!!).   CALIB solutions were made 
every second, the results were smooth with a 10-second boxcar, and 
interpolated to the CL table in discrete time segments chosen to avoid 
smoothing over the gain jumps.   The reasons for the smooth were to 
avoid chasing noise, and to reduce the tendency of the antenna-based 
calibration procedure to move correlator-based effects to antennas (a 
problem that small-number arrays are especially vulnerable to). 

    5) BPASS was run to provide a normalized bandpass function for each 
antenna.  A single solution, for each antenna and each sub-band was 
generated.  These look entirely normal for the two RFI-free subbands. 

    6) I then generated three 'pseudo-continuum' databases from each of 
the two RFI-free subbands -- of 1, 12, and 50 MHz width each, for 
further analysis, using the program AVSPC.   UVPLTs of the three 
databases from the high frequency sub-band showed clean visibilities 
corresponding to a point source.  All visibility amplitudes were at the 
expected level,and all phases centered near zero.  I had hopes of 
getting clean images.  These were dashed upon viewing the results from 
IMAGR.  The 1 MHz BW image (large enough to cover the entire primary 
beam) gave a rather poor dynamic range (DR) of 2000:1.  This should be 
enough to show the background sources (which are a factor of about 250 
below the peak of the calibrator) -- but none were seen.  Only 'lumps 
and bumps'.  The 12 MHz BW image had noise which was lower by the 
expected factor (the DR is 7000:1), but *** still no background sources 
*** -- these should be easily seen with at least 10:1 contrast against 
the noise.  The 50 MHz BW image was not significantly different than the 
12 MHz BW image.  Although the map residuals gave no indication that 
'closure' problems are responsible, I ran the closure program BLCAL 
anyway.  This made no perceptible different in the resulting image. 

    7)  The results from the low-frequency triplet of pseudo-continuum 
databases gets more depressing, or interesting (depending on your point 
of view).  The visibility plots immediately showed the present of *** 
enormous *** deviant amplitudes, particularly on the short spacings.   
These deviants are not 'spikes', but sinusoidally varying with 
amplitudes up to 10 times the amplitude of the calibrator -- that makes 
them 25 Jy.  These large deviant amplitudes are matched by similar 
changes in phase, which are enough to rotate the phases continuously 
around 360 degrees.  These deviations are strongest by far in the 1 MHz 
BW dataset, but are still visible in the others.  The amplitude of these 
oscillations are not constant through the hour -- they vary over a 
factor of about ten in a rather continuous manner.  They are not due to 
any 'stationary' external interference -- an image made at the north 
pole shows no excess flux.  These excess visibilities cannot be due to 
Cas or Cyg -- they are sufficiently far away to be attenuated by a 
factor of at least 1000, more likely 10000, reducing them to sub-Jy 
level.   It is perhaps possible that they are due to solar interference 
-- the observations were taken in late afternoon -- except that some 
baselines (2 x 18 -- E2 x N16 is an excellent example) show a clear 
double beat, with a long period variation of period 90 seconds 
superposed on a fast oscillation of 4 seconds.  I don't know how any 
single external source can do such a thing.   In any event, I'm not 
aware that the sun is up to anything recently -- if any of you have 
information on this, let me know.   We'd have to be seeing nearly MegaJy 
structure on the Sun in order to get ~25 Jy residuals on our northern 
source, which is about 50 degrees away from the sun. 

    So , in sum, I'm completely baffled by what we're seeing here.  Note 
that an observation of this source taken with the VLA correlator on June 
30 (the very next day!) provided a beautiful image, with no problems 
whatever! 

    Some thoughts on what to try next:

    1) An observation of a good calibrator with accompanying 
observations on nearby blank sky.  This will establish if we're seeing 
an additive or multiplicative effect.  I note that the Cygnus A 
observations seemed devoid of the fast oscillations noted above -- this 
might argue for an additive effect, or that the problem was absent when 
Cygnus A was observed.  (Note -- this was at night ...).    Although 
0217+738 is a good source, I think 3C147 (0542+498) may be better, as it 
is 10 times stronger, and hence easier to calibrate in the face of RFI 
and effects of undetermined origin. 

    2) I suppose it is somehow possible that some bizarre aliasing 
effect is at work here -- it would have to alias the continuum via some 
super-strong semi-continuous RFI spike (not seen in any spectra!).  We 
could eliminate all this by band-limiting the RF signal prior to its 
sampling.  Bob Hayward tells me he is pretty sure he can round up as 
many as 10 RF filters with BW less than 100 MHz.   If we do this (and 
doing so would clearly be an act of desperation ...), it will need to be 
done on a maintenance day, and take some considerable organization.  
Fortunately, we're in C-config now, so traveling amongst the antennas is 
quick.