[evlatests] Adventures in wide-band calibration and imaging

[Rick Perley]

    I've spent most of this week fiddling around with the wide-band 
Cygnus A X-band database, experimenting with the new capabilities that 
Eric has put into BPASS, for the purposes of better understanding how 
best to calibrate data in the 'WIDAR' era. 

    The database utilized was the 6-hour X-band observation of Cyg A, 
which included both its local calibrator (J2007+4029), and the primary 
calibrator 3C286. 

    The first part of this report is a recap:

    - The observations were made with 8 x 128 MHz subbands, full 
polarization, 64 channels per product, 1 second averaging.  The total 
data volume was about 144 GB. 
    - Upon loading into AIPS, FITLD reported about 50.1% of the data 
were integer zero.  Worrisome!  It turned out that the last four 
subbands were all integer zero -- some sort of correlator setup error. 
    - The remaining (good) four subbands were out of order, and were 
mislabeled w.r.t. frequency.  This was repaired via UVCOP, VBGLU, 
PUTHEAD, and UVFIX (this last step to recompute the u,v,w coords). 
    - The four remaining subbands covered the range 8.400 through 8912 
MHz.  The highest of these four lies well above the nominal bandpass of 
the receiver -- easily seen in the bandpass plots. 
    - Other than the four blank subbands, data quality was superb, with 
the  only editing required being due to antenna slew. 

    Now for the more fun bits. 

    - I decided to track and remove the 'delay clunks' (not that I think 
they're important, but mostly because I can).  This was easy for the two 
calibrators.  For Cyg A, a good model was used.  I quickly discovered 
that antenna 8 is 'the center of the EVLA' -- this antenna is located at 
N1, and it's clear from the derived delay solutions that this location 
is used by software as the array center.  Solutions were made for every 
single record (this is necessary to track the delays, as the longest 
baseline undergoes a delay 'clunk' about every 4 seconds). 
    - A nuance in AIPS was then discovered:  In order to apply solutions 
made with every integration, one needs a calibration table which is much 
more finely gridded than the data integrations!  This is because INDXR, 
which generates the CL table, pays no attention to the time stamps at 
which the observations were made.  If, for example, you have 1 second 
integrations, and you generate a 1-second CL table, there is no phasing 
attempted between these two grids.  So to ensure appropriate delay 
tracking (and until Eric can think up a way to force INDXR to align the 
integration times and CL table), one must grossly oversample the CL 
table.  I chose 0.1 seconds -- this worked well. 
    - After this basic calibration and editing, I averaged the data down 
to 10 second intervals -- this *hugely* helps the subsequent processing. 

    Eric has modified BPASS so it will employ the polynomical 
expressions for the known standard calibrators.  I proceeded this way 
(with no guarantee that this is the best approach): 

    a) SETJY is used to determine the fluxes for each subband. 
    b) BPASS was run *without normalization* (BPARM(5)=1, and 
BPARM(10)=0) to generate absolute solutions for each calibrator 
observation of 3C286.   The program now cleverly knows I want the true 
solutions, so uses the polynomial expressions to determine the solutions 
for each channel, for each subband.  (This is, in essence, CALIB for 
every channel, using the correct flux). 
    c) CALIB and GETJY were used, utilizing the central few channels 
only, to determine the spectrum of the phase calibrator J2007+4029.  I 
got 3.888, 3.908, 3.920, and 3.934 for my four subbands.  These four are 
a close enough fit to a simple power-law spectrum of spectral index 
0.266 that I used that value for the next step.  Most antennas show 
virtually  no time variability in their gains (much better than 1%), but 
some antennas are not so well behaved -- conceivably an elevation 
effect.  These deviations are highly correlated between subbands, but 
*not* between polarizations -- which argues for electronics, rather than 
pointing, as a cause. 
    d) BPASS was then used on J2007, utilizing the new adverb SPECINDX 
set to 0.266.  (I also used SETJY to force compliance between the SU 
table and the SPECINDX).   This operation provided spectacularly stable 
solutions, both over time, and between the two calibrator sources.  
(Changes in the spectrum are typically at a level less than 0.1%, 
peak-peak).  The combination of the polynomial expression for 3C286,and 
the derived spectral index for J2007 were completely sufficient for 
determining the spectral bandpass corrections. 
    e) CALIB/CLCAL was used to make final gain adjustments -- but in 
fact this has little effect, since the way I have employed BPASS 
effectively provides a 2-point calibration. 
    f) Eric's new program RLDLY immediately determined the R-L delay 
(and automatically updates the CL table). 
    g) PCAL was then used to determine the mean (relative!) antenna 
polarizations.  (AIPS can not determine D terms as a function of 
frequency -- at least not yet ... :-) ) 

    The results of all this were displayed by utilizing POSSM, the 
spectrum plotting software.  The plots are quite lovely to view, 
particularly those of Cygnus A.  The four subbands can be plotted 
adjacent (like one continuous spectrum covering 512 MHz), for any Stokes 
combination.  The most interesting results are when one views I, Q, U, 
and V, with all corrections applied.  If there is interest, I'll show 
some of these at an upcoming meeting -- certainly at the Thursday 
meeting, some of which will be dedicated to polarimetry. 
    For the calibrators, we see smooth continuous spectra, as is 
required by the methodology. 
    For Cygnus A, we see the visibility function changing (dramatically! 
-- even over only 512 MHz) for any given baseline, at any given time, as 
a function of frequency.  A good lesson in interferometry -- and an even 
better lesson on why we have to have software which 'knows' about these 
changes. 
    A even better lesson was clear in the plots of Q and U -- this will 
go into the next circular. 

    I generated an image of the J2007 calibrator.  Because I haven't 
learned yet how to implement IMAGR to account for the spectral slope, 
the resulting image, if utilizing the correctly calibrated amplitudes, 
results in an image with remarkable sidelobe structures.  To get a noise 
limited image, I had to re-determine the bandpass, forcing the spectral 
flux density to a single value (3.9 Jy) over the full 512 MHz bandpass.  
Utilizing IMAGR in a straighforward way then provides a noise-limited 
image, with 500,000:1 dynamic range.  The calibrator has a nearby 
secondary, and about 3 or 4 background images are seen.  The rms noise 
is about 7 microJy.  (I used only the three subbands safely within the 
bandpass for this image).  The integration time is slightly over 1 hour. 
    The noise in the V image is the same as in I.  But the noise in Q 
and U is a few times higher -- clearly due to small-scale variability in 
the cross-polarization, both in frequency and time, no doubt ...

    The image of Cygnus A was, as usual far from noise limited -- 
probably 100 times the noise limit.