[evlatests] C-Band observations of 0217, Thursday

[Rick Perley]

    John Benson managed to convert the C-band tests to UVFITS on 
Friday.  The results from the test are:

    In general, the data are of very good quality, with no serious, 
obvious problems.  The problems noted are all the same as reported 
Friday for X-band. 

    1) All 12 antennas fringed strongly in both RCP and LCP. 

    2) There are a significant number of zero amplitude/phase records -- 
roughly 0.1% of the data.  These are easy to remove. 

    3) The delays are all exactly the same as they are at X-band, (to 
with 1 nsec or so), except the signs are reversed.  Note that the very 
large values and differences between the antennas in LCP are present at 
C-band also.  Although this is a very minor problem (since FRING finds 
and removes these with high accuracy), it would be good to fix up the 
on-line delays, so the phase slopes are smaller.  (Makes editing the 
data easier, if nothing else). 

    4) Bandpass solutions are good, except (as at X-band):

       Antenna 3, LCP, subbands 3 and 4 have a strong slope in 
amplitude.  (Subbands 1 and 2 look normal, but the sensitivity is much 
reduced in these two, so I removed all four).
       Antenna 8, LCP, all subbands, as a pure, 1-cycle sinusoid (in 
frequency) bandpass. 

    I removed the LCP subbands for both antennas from the database. 

    5) Antenna gains show beautifully smooth phase variations, exactly 
as expected for the baseline lengths and frequency at this time of 
year.  The amplitude solutions show the same curious instabilities that 
are visible at X-band, to wit:
          - All antennas show non-sinusoidal variations, typically +/- 
2% in amplitude (or 4% in power).  Some antennas reach 6% at times.  All 
four IFs in both polarization show *exactly* the same variations.  All 
antennas have independent patterns.  Since the data are in no way 
normalized, we are seeing here changes in attenuation or gain.  At this 
frequency, I think we can discount opacity effects, which leaves us with 
amplifier gain variations (or perhaps, but much less likely, pointing 
problems). 

    6) Following a 2-point amplitude and gain calibration, noise 
histograms were plotted.  All antennas in both polarizations gave 
beautiful gaussian histograms, with 1sigma widths of 0.42 Jy.  A BOE 
(back-of-the-envelope) estimation is that this should be 0.33 Jy -- we 
are somewhat noisier than we should be.  However, this observation was 
on a very strong point source -- true noise estimates must be made in 
blank fields.  I suggest this be done, when convenient. 

    7) Clipping discrepant visibilities removed only 0.3% of all 
visibilities (including the edge channels, were gain variations on 
filter edges will surely cause some fluctuations) -- this is very good. 

    8) Imaging showed that again that the noise does not go down with 
sqrt(BW), after about 32 central channels are included.  With both 
polarizations, all four subbands, and 150 central channels, the noise 
ends up a factor of 2 to 3 above that expected from the histogram 
width.   The images show the single background source I'm aware of, but 
also some 'lumpy' stuff nearer the point source, or similar amplitude.  
Closure corrections reduce the 'lumps', but have little effect on the 
noise.  The edges of the images are dominated by patchy sinusoids (the 
patches are symmetric about the center, but are located well past the 
first null of the beam -- they are clearly an imaging artefact).  I'm 
not at all sure about what to expect with so little data, and such large 
fields -- only 16 minutes of data are in this test.