[daip] CLCAL problem.
Craig Walker
cwalker at nrao.edu
Mon Oct 9 17:41:40 EDT 2006
As Bill says, this is indeed tricky.
Separating the smoothing of amplitudes is probably fine and I don't
think we need to think about that.
We could make the single band delays and the phases independent if the
phases referred to the bandpass weighted center of the band so they are
an average phase. In this case, when the phase slope across the band
changes, the average phase does not. In the more usual case where the
reference frequency is near the edge of the band, any fringe fit
solution phase will contain an offset from the average phase that is due
to the phase slope (delay). That offset phase should be adjusted when
the delay is changed, or you are left with a phase fluctuation that is a
leftover from the delay fluctuations and has nothing to do with the
average phase. Second order issues related to the location of the
weighted center of the band rear their heads here. That depends on both
the bandpass shape and on the channel range selected (channels are not
even a SNSMO input because normally the SN table doesn't care about
channels).
Phase smoothing should probably be done after the delay smoothing and
after any phase corrections from the delay smoothing have been applied.
Often in VLBI, we zero the phase and rate form the FRING solution,
either with the zeroing options in FRING or in SNCOR. That, I presume,
removes the phase fluctuations from the delay that are related to the
offset of the reference frequency from the average frequency as
described above. If zeroing is done, you probably don't want to alter
the phases when smoothing the delays. This would not be a problem if
the reference frequency were at the average frequency, but it generally
is not so it gets a bit tricky. I suspect that SNSMO should determine if
zeroing has been used and, if so, not make phase adjustments when the
delay is smoothed.
My understanding of multiband delay is that it is never actually used to
apply calibration to the data. It just is a place to hold the multiband
delay fit results from FRING or MBDLY based on data that are otherwise
contained in IF to IF phase slopes and in the single band delays. The
results are then used by such programs as CL2HF that pass them to the
geodesy world. Am I right? I'm worried about possible double dipping
if you adjust phases when the multiband delay is smoothed. The single
band delay and phase smoothing may have already done the job. This
suggests 2 possible actions. The multiband delay could be smoothed
without doing phase adjustments. That is the desired action for
obtaining more accurate multiband delays by integrating. Or the
multiband delay could be zeroed to force a redetermination by
(re)running MBDLY.
Assuming that rates are referenced to the average time of the data, and
if that is also the SN table data time, I think rates can be smoothed
independently of the other parameters including the phase.
Of course, as Bill points out, the Right-Left phase difference needs to
be preserved if polarization work is contemplated. That is easy if the
reference antenna does not change and all phase like parameters for that
antenna are kept at zero. I presume SNSMO and CLCAL take the necessary
steps to preserve the RL difference if there is a change of reference -
such as keeping the new reference antenna phase like values fixed at
their values the last time there was overlap with the primary reference.
As for the code snippets in the messages from you and Leonia,
unfortunately I'm not exactly sure what the different variables such as
FZERO and FOFF are exactly. It is true that if you have some known
reason to change the delay by some amount, such as CLCOR would typically
have, you want to change the phase by the delay times frequency. But
smoothing is somewhat different. I think it is fair to assume that data
that are being smoothed are the result of a fit and so you probably want
smoothed delays not to shift the average phases, certainly not by the
very large amounts that frequency times delay would imply. And as noted
above, I suspect that nothing to do with the explicit multiband delays
should be allowed to affect the phases since the multiband delays are
not used in calibration but the phases are (of course, when CLCOR
adjusts delays, it should adjust single and multiband delays and phases
by the appropriate amounts).
Something tells me this won't be the end of this story!
Cheers,
Craig
Eric Greisen wrote:
> I have found the source of the trouble and I do not know what is the
> right thing to do.
>
> SNSMO smooths delays separately from rates separately from amp/phase.
> So if phases are 0 coming in they are 0 going out except for
> multi-band delays (see below).
>
> CLCAL smooths each separately but in the same subroutine. Then it
> says if the single band delay has changed, the phases should change
> multiplied by FOFF(iif). That factor seems wrong to me - should it
> not be Fzero + FOFF(iif)? That aside, should one change phases for a
> smooth of delay? Note that multi-band delay changes phases in SNSMO:
>
> IF ((IPHASE.NE.FBLANK) .AND. (AMP.NE.FBLANK)) THEN
> C Multiband delay correction
>
> IF ((IFEND.GT.IFBEG) .AND. SMOPHS .AND.
> * (RECR(MB1KOL).NE.FBLANK)) THEN
> PHASE = PHASE + TWOPI * (FREQS(I)-FREQS(1)) *
> * RECR(MB1KOL)
>
> but not in CLCAL. Now I am really confused.
>
> What do people thing is correct?
>
> Note Craig that the CUBE mode has real difficulty extrapolating so
> sensible phases in the smoothed SN table become crazy in the CL
> table.
>
> Eric Greisen
--
---------------------------------------------------------------------
R. Craig Walker Array Operations Center
cwalker at nrao.edu National Radio Astronomy Observatory
Phone 505 835 7247 P. O. Box O
Fax 505 835 7027 Socorro NM 87801 USA
---------------------------------------------------------------------
More information about the Daip
mailing list