[mmaimcal]Re: Instrumental Delay offsets

[John Conway]

On Tue, 1 Jun 2004, Mark Holdaway wrote:

>
> Just a few isolated comments which address several emails
> over the last few days (but before looking at John's most
> recent e-mail):
>
> * Concerning Dick's calculations about the effects of baseline
> * errors after fast switching, which will reduce the effect of
> * the baseline errors on the phase, but that phase error will
> * be slowly varying:
>
> The residual phase errors after fast switching should be mostly
> (ie, 90%) random, and hence will average down in a reasonable
> way.  This was considered in setting the 300s drift specification.
> The "slowly varying" residuals due to baseline errors won't average
> down, so we need to "count" them differently -- ie, we'd like them
> to be MUCH smaller than the atmospheric errors.
>
> * Concerning Larry's point about additional measurements to
> * help average down the errors:
>
> I think that we should make a paradigm shift based on this comment: we
> need to do the 1-hour Conway run to get the baselines determined well
> enough to begin observations, but we then use all subsequent high SNR fast
> switching calibrator observations to check/update/average-down the
> position solutions.  (The SSR people will not smile when we tell this to
> them.)
>

Of course we can (and should) use whatever data we have to continually
update the antenna positions. Self cal solutions using single cal obs
close to  transit won't be optimum however. Remember we need to estimate
at least three parmeters from the data so long horizon-to-horizon
tracks are really needed, and will not be common for ALMA as for the
VLA.  Specifc 'geodetic' sequences are much better than long tracks.

As I currently envision it, every 2.5 days 4 antennas are moved
during a morning, then...

1) We do first pointing and position determination in early afternoon,
qualifying antennas to particpate in observations.

2) We do a second geodetic run in the middle of the night,
when phase errors are minimum, using one or
two additional antennas from the array and taking 30 min to 1 hour.
Perhaps pointing is improved then also.

3) About every 7.5 days (or maybe longer)  we do a full position
calibration simultaneosly of all antennas again for 30min-1hour
- this  stops position rrors accumulating.

Although this all seems a lot of cal to do, it only involves
a handful of antennas, in our analysis of the continuous reconfiguation
scheme (ALMA memo Conway 2000) we noted that the number of baseline-hours
lost to antenna calibration was a fraction of a percent of the total
available and roughly comparable between continuous and conventional
burst reconfiguration.

I'm confident that in the first cal we can get good enough positions
to qualify antennas for observations -i.e get no decorrelation
over the bandpass (Marks 'Thing 1' below) - however it sets a spec
on the minimum number of spectral channels stored for continuum
expts.

 When the pipeline  is run depends on your imaging/astrometry
requirements. My guess is that for 90% of expts the required DR will be
good  enough using the antenna cal in 1) so the pipeline can be
done immediately, maybe for 10% it will be good enough after 2) so
the pipeline can be done after 12hrs and 1% maybe will require
3). It is important to remember that there will only ever be
a maximum of 4 out of 64 antennas with inaccuate positions at any
given time, and our spec of 65microns is a very severe one for
non-THz observations. As I said the impact of position errors
should be expressed as a DR/astrometry  limit and I am doing some
simulations of that.

  John



> There are TWO Things that go into a specification for antenna
> position determination:
>    (1) If the baselines are really far off, you get decorrelation over
> 	the bandpass, you lose sensitivity.  This is not nearly
> 	as stringent as the specification that John is working
> 	towards.
>    (2) If the baselines are in error, you make a slowly changing
> 	phase error after fast switching, as noted by Dick.
>
> Thing (1) is permanent, we have to meet that spec before any astronomical
> observations take place -- I don't have any equations or numbers at this
> moment (at home), but I bet that John's scheme will get the positions much
> better than the requirements of Thing (1).
>
> Thing (2) can be corrected after the fact (and is routinely corrected
> after the fact at the VLA).  SO: after reconfiguration, there may be a
> delay in the pipeline imaging as we use hours of fast switching
> data to refine the position determinations.  ONE PROBLEM: most
> observations will be close to transit, making it difficult to
> determine some components of the baselines.  BUT: as most observations
> are close to transit, we sort of don't have to get those components
> that are hard to get.
>
>    -Mark
>
>
>
>
>
>
>