[evlatests] Results from the deep L-band observation of 3C147

[Rick Perley]

    Flushed with success following the nice results taken at C-band of 
some strong sources, I convinced Michael to spend 6 hours integrating on 
3C147, a strong,  unresolved, and completely unpolarized calibrator.  
Four subbands, with full polarization, were observed.  The four subbands 
were centered at 1308, 1436, 1820 and 1948 MHz -- chosen as the four 
subbands most free of RFI.  In fact, the first and last, although mainly 
clear of RFI, still have quite a lot, so my analysis was done on the 
central two, which are indeed almost 100% clear. 

    All 12 WIDAR-connected antennas worked throughout the 6-hour run, 
which extended from 07:15 through 13:16 IAT (i.e. 1:15 to 7:15 AM).  
There was no shadowing during this time. 

    In general, the data quality appear excellent (from the usual 
cursory inspection routines:  LISTR, UVPLT, VPLOT, etc.).  There was a 
spectacular 'event' at 09:00 IAT, in RCP only on all antennas, plots of 
which I'll show tomorrow.  This event was flagged. 

    Delays and phases are very constant (other than the three I reported 
on yesterday).   Bandpasses are for the most part very stable.  In 
short, all looked good for outstanding images.  But looks are deceiving ...

    I split out a 10 MHz-wide 'pseudo-continuum' chunk, centered on the 
central channel, for which *only* the gross delay was removed (a single 
delay for the entire run).  The data were then amplitude/phase 
calibrated, using a 1 minute average, to ensure that fast time 
irregularities would not be shifted around.  (This simplified processing 
was done to prevent any hint of complicated calibration procedures from 
contaminating the results.  However, the results below are seen equally 
in 'fancy' method calibrated data as well). 

    An image of a *single* 500 kHz channel provides a good map:  Dynamic 
range (DR) = 36,000:1.  Sounds good.  But it turns out that this same DR 
is obtained *** no matter how many channels are  utilized***.  Thus, 
summing up 4, or 16, or 64, or even all (256) channels results in the 
same image, and the same dynamic range.  Both IFs are similarly limited. 

    Looked at in detail, most baselines have a severe case of 
'wobble-itis'.  That is, there are sinusoidal oscillations in amplitude 
and/or phase which dominate over the noise when bandwidths are 1 MHz or 
more.  Some of these are simply spectacular, with pk-pk amplitude 
oscillations of 2.5 Jy (10% of the source flux density), and/or 6 
degrees in phase.  Running FRPLT shows that the oscillations are rarely 
pure (various harmonics are often seen), and that the oscillation 
frequency changes with time. 

    Some baselines are much worse than others.  Notable bad ones are:  
19 x 25 (amplitude only!), 5 x 27, 15 x 19, 27 x 28 (phase only!), 2 x 
15, 8 x 9, 2 x 19, 2 x 25, 3 x 28, 5 x 28 ... (the list goes on ...).  
There are no obvious spatial relationships here.  The *maximum* fringe 
frequency that the D-configuration can give is 0.2 Hz.  Nearly all of 
the oscillations (wobbles) reported above have a higher frequency than 
this -- I claim from this that these wobbles are not due to any source 
in the sky, unless coupled in through some sort of non-linearity.  Note 
also that these wobbles were seen through the entire run, most of which 
was taken before sunrise.  No change in characteristics were noted upon 
sunrise.  I don't see how the sun can have anything to do with this. 

    Recall some conclusions made from the fast dumps that Michael 
provided us 2 weeks ago:  The wobble frequencies are proportional to sky 
frequency, and the wobble amplitudes are much stronger at lower 
frequencies.  (This from analysis of X and C band data).  It seems 
likely that our L-band wobbles are an extension of these trends. 

    Any hopes that baseline-calibration (BLCAL) would solve our problems 
(as it did so well at C-band) were quickly dashed.  BLCAL doubled the 
dynamic range, so the noise level was reduced to 0.3 mJy -- still at 
least an order of magnitude too high. 

    I then 'clipped' all the data whose amplitudes exceeded what appears 
to be the thermal noise (about the mean visibility).  This process 
removed most (but not all) of the most strongly affected wobbling 
baselines.  The result from this was another doubling of the dynamic 
range -- we're still a factor of many tens away from where we should be. 

    I conclude we have a serious issue here.  Although one may argue 
that DR = 50,000:1 is not too shabby (and it isn't), it's nowhere near 
good enough.   I think a renewed effort to root out what is going wrong 
here is needed.