[daip] VLBA bandpass calibration in AIPS

[Eric Greisen]

On 06/16/2014 02:45 PM, Yuri Y. Kovalev wrote:
> Dear Eric,
>
> In Aug 2013 I contacted you about issues related to BPASSPRM values as
> applied to VLBA data sets.
> Since 2013-Dec-15, we have reduced our MOJAVE datasets with ICHANSEL =
> 0,  BPASSPRM = 0,0,0,0,1,0,0,0,0,4,1 following on your suggestions. A
> keyfile containing the setup can be found at
> http://www.vlba.nrao.edu/astro/VOBS/astronomy/may14/bl193ai/bl193ai.key
>
> We've found that with these values, the amplitudes are about 20% lower
> than they should be.
> We've verified this with near-simultaneous observations with OVRO at the
> same observing frequency.
>
> Re-reducing the latest dataset with the BPASSPRMS we used prior to
> 2013-Dec-15: ie.. ICHANSEL = 0, BPASSPRM = 0,0,0,0,1,0,0,0,0,2,1   gives
> amplitudes consistent with OVRO.  We found that in that dataset, using
> either BPASSPRM(5) = 0 or BPASSPRM(5) = 1 made less than a 1% difference
> in the amplitudes.
>
> So using all of the channels in BPASS normalization (i.e. BPASSPRM(10) =
> 4) doesn't appear to be advisable for VLBA data, as the resulting
> normalization yields significantly low amplitudes.
>
> I wonder if you could take a look at this issue and comment on it? Do
> you find problems in the code? Do you find problems in the algorithm?
> Did we make a mistake somewhere?
> It seems like (my guess) there might be something to do with the way how
> weights of the spectral channels are being used.
>
> To help you, attached please find two BPASS solution plots.
> For BPASSPRM(10)=4 (flux seems off my 20% for the current MOJAVE
> frequency setup): BPASS4.PS.gz
> For BPASSPRM(10)=2 (flux seems OK for the current MOJAVE frequency
> setup): BPASS4.PS.gz

I'm sorry if I have steered you incorrectly - I was told that the VLBA 
measured Tsys over the full bandwidth and therefore the bandpass 
function should be based on the integral of the full bandwidth.  The 
amplitude scaling is then done with the Tsys - not calibration sources 
as in the VLA.  If you have independent evidence that this is wrong, by
all means follow that which seems better.  The ICHANSEL=0, means use the 
inner 75%.  You ask about errors in the code - I will look but I doubt 
it - and also about channel weights.  Do your data have weights which 
are channel dependent?  That would be very unusual without special 
action on your part.

Eric Greisen