[daip] solve for a complex band pass

[Eric Greisen]

Z.-Q. Shen writes:

 > I am trying to run BPASS on an xc data to solve for a complex band pass.
 > Somehow, I can not obtain satisfactory results. The amplitude part of the
 > solution shows quick variation cross the total band of 8 MHz with 256 spectral
 > channels. The observations were made at 43 GHz with the VLBA. However,
 > I can obtain amplitude band pass response function using ac data only, which
 > looks good. Below is part of message from running BPASS using the xc data.
 > The calibrator I used is NRA0530 and PKS 1921-293. No amplitude calibration
 > (tsys and gain) was applied. I don't understand some lines like
 > 
 > >BPASS2 11:10:41 Antenna  1  IF  1  corr 2  had      2088 excess closure errors
 > 
 > Does this mean there are errors in the data which will worse the solution?
 > I have used the Hanning smoothing. Except at the edge channels, the profile
 > of the amplitude part from the complex BP is very similar to the one from the
 > ac data (see appended plots). Actually, the variation of the amp vs freq can be
 > seen from the output of the correlator (xcbandfile.ps). Is this because the
 > calibrators are still not strong enough in each spectral channel? If so, is 
 > there
 > anyway we can try to solve out the phase part of the band pass function?

 > >BPASS2 11:10:24 Excess closure errors defined at 10.00 percent and  10.0 
 > >degrees

 > >BPASS2 11:10:29 Dividing the spectral data by channel 0

So far as I can tell, BPASS has worked moderately correctly.  Your
calibrator source appears to be weak and to be resolved so that the
fringes are even weaker on the longer baselines (more distant
antennas).  You did not send your inputs to BPASS so it is hard to be
certain of what all you did.  What is clear is that you set MINAMPERR
and MINPHSER to fairly small numbers (10) - appropriate to VLA
bandpasses at longer wavelengths with strong calibrators.  The program
counts up each baseline-time sample that, when the determined bandpass
is applied, is not reduced to (1,0) within 10% and/or 10 degrees
phase.  If all 45 baselines had this then you must have had at least 42
time samples averaged in the solution interval.  With 0.2 sec
integrations, that would be only 8 seconds of data - I assume your
calibrator scans are rather longer but your integration time in the
data set may also be > than the minimum 0.2.

With a weak calibrator, you should use a solution interval SOLINT=0 to
do the full scan.  Each time sample is divided by the average of the
channels so that continuum time changes due to things like atmospheric
phase and source visbility are fully compensated.  Unfortunately, this
division also adds noise and a Ricean bias (complex division is phase
subtraction and actual division by an amplitude which is therefore
biased away from 0).  Setting BPASSPRM(5) = 1 turns off this division
and setting BPASSPRM(10) = 1, causes it to normalize the amplitude
portion of the bandpass.  Doing this requires either a good source
model or a source sufficiently  unresolved that its visibility does
not change much over a scan and also good atmospheric phases.  The
31DEC01 version of AIPS offers more options for dealing with this
problem but would still require good time stability.

The autocorrelation bandpass is based on an enormous signal and so is
noise free.  But it does not represent the xc data well on some
antennas (see PT).  Nonetheless, it may be wise to use it for the
present and to determine other calibrations.  When the atmospheric
phases have been calibrated and FRING and ACCOR, then you may be able
to use a good model for the calibrator rather than channel 0 and then
do long integrations for a complex bandpass.

Eric Greisen