[daip] solve for a complex band pass

[Z.-Q. Shen]

Dear Eric Greisen,

Thank you very much for the explanations.

I did what you suggested on the inputs to BPASS and tried CPASS too.
It turns out that the low SNR signals worsen the complex bandpass
solution. I agree with you that the "ripples" seen in the plot are just
noise.

I also did the comparison between bandpass functions obtained from the
same xc and ac data but correlated with 16 spectral channel of an 8 MHz IF.
The complex bandpass solutions are not that "noisy", but still not that
satisfactory compared with the one from the ac data.

Now, I decided to take two steps in deciding a complex bandpass.
First, using ac data to solve for the amplitude part  and then solve for
the phase part only by using xc data and fixing the amplitude solution.
This can be done in CPASS.

I have another question regarding the normalization (BPASSPRM(10)
in both BPASS and CPASS) and auto-normalization (CPARM(8) in CPASS).
I didn't see any difference with and without auto-normalization in
the CPASS run. Will such normalization (auto-scale) have any effect
on the amplitude?

Thanks again,
Zhiqiang Shen


>Z.-Q. Shen writes:
> > The inputs I used for the BPASS run are something like
>
>     They were about what I suspected
>
> >
> > docalib=1 (used for solving for the complex bandpass)
> > gainu=7 (including ACCOR and FRING, but not APCAL of amplitude calibration)
>
>       I am more a VLA person and so thought this not needed - but
>correcting fringe rates across the band and with time before averaging
>for channel zero and then for the scan average may be a very good idea.
>
> > minamper=10
> > minphser=10
>
>These are a bit low.
>
> > solint=0
>
>You could try -1 and do all of the data for a single solution.  This
>may be adequate for the vLBA and your S/N does not lead me to expect
>that you will achieve b=very high spectral dynamic range anyway.
>
> > smooth=1,0
>
>One could also do wider smoothing for improved S/N.  I do not know how
>much this would help.  The actual bandpass probably does not have
>narrow features so smoothing over say 5 channels may not hurt.
>
> >
> > I also tried the single calibrator scan about 3 minutes, or restrict
> > the BPASS to the inner array (without SC,HN and MK), or cut some
> > edge channels. But unfortunately, the frequency-dependent amplitude
> > ripple in BP is still there.
>
>      In the plot I saw this "ripple" appears to be just noise.
>
> >
> >   > >BPASS2 11:10:41 Antenna  1  IF  1  corr 2  had      2088 excess 
>closure
> > errors
> >
> > Do I understand correctly that the large number given above (2088)
> > tells you how bad the bandpass solution is? I suspect such a problem
> > may be common for the 43 GHz VLBA observations. This may be related
> > to the low SNR signal for each spectral channel at 43 GHz. Can you
> > comment on this?
>
>       This number is the sum of all time samples and all baselines
>for which the corrected data differred from (1,0) by more than 10%.
>If you had 100 minutes of 0.2 second data, then you would have 30000
>times 45 such samples and 2088 failures would be a failure rate of
>0.15 per cent which would be very good.  If you have 3 minutes of
>data, then the failure rate is 5.2 % which is worrisome but not a
>disaster.  If you have 3 minutes of 1 second data then the failure
>rate is 25% which is not good.
>
> >
> > I am wondering if there is any special smoothing function within the
> > freq band.
>
>       There is also the task CPASS.  I hesitate to recommend it
>becuase I am not sure that the fitting of the functions is really
>proper, but it was designed to fit a function to the complex bandpass
>using a smaller number of parameters than the number of channels.  In
>CPASS, set BPASSP(3) 1, BPASSP(5) 1.  This does the scan average first
>and then divides by a channel 0 determined from the average which is
>better than dividing each record by its channel 0.
>
>Eric Greisen