aver_time, doscatter
Time average a UV data set.
EXAMPLE
-------

0>uvaver 240
Averaging into 240 second bins.
Selecting channels: 1 -> 1, polarization: RR.
0>

All but the first line above are informational messages from the
'uvaver' command.

PARAMETERS
----------
aver_time - The duration of the new integrations in seconds.

doscatter - Default: FALSE
            If doscatter=FALSE then the new visibility weights will be
            the sum of the old visibility weights. This is
            statistically correct under the assumption that the input
            weights equal or are proportional to the reciprocal of the
            amplitude variance. This assumption is based on the rules
            described in AIPS documentation for UV FITS files.

            If doscatter=TRUE then the new visibilities will be
            given weights derived from the data by using the scatter
            within a solution interval to derive an estimate of the
            reciprocal amplitude variance. Effective use of this
            option requires that each output bin encompass many input
            integrations. In particular, if only one visibility exists
            in a given bin, uvaver will leave this visibility in place
            with its original weight, but make it flagged.

CONTEXT
-------

There are two main things that people commonly use uvaver for. The
obvious one is to reduce redundantly high time sampling. This speeds
up subsequent processing. The less obvious use is to re-bin the
data. This is useful if the data that were read in from a FITS file
had different time-stamps on different baselines within the same
logical integration. In this case when difmap reads such a file 
the end result is that each baseline is assigned its own integration!
Not only does this waste large amounts of memory and subsequent CPU
time, but it also means that station-based editing in vplot actually
does baseline-based editing, and selfcal, when run without a solution
time will not make any corrections due to the lack data within each
integration. The solution is to re-bin the data into integrations with
selfcal.

You should be careful to choose an averaging time, that will not
smooth over real variations in the UV data. If you choose too large a
time scale, large scale structures will disappear from your map.

CAUTION: *** If uvaver fails you may loose your data ***
--------------------------------------------------------
When difmap reads a UV FITS file, it places the raw visibilities in
a scratch file. This file is then read whenever the 'select' command
is invoked, or when edits need to be applied. The uvaver command
averages data in this file and writes the averaged integrations to a
new scratch file before deleting the original. If it fails to write
the new scratch file - perhaps because of insufficient disk space,
then difmap will discard the current data. Thus you should check that
you have sufficient disk space for the new file (look at the existing
file called uvdata.scr to get an idea of the required disk space)
before running uvaver. If you have done anything significant to the
data, then it would also be prudent to save the merge file before
running uvaver.

DETAILS
-------
The uvaver command reads raw integration records from its existing
uvdata scratch file and writes binned averaged integrations to a new
uvdata file. When all integrations have been averaged, the old uvdata
scratch file is deleted to be replaced by the new one.

For each new integration read within a given solution bin, uvaver adds
to a weighted vector running mean of the data versus time and a
weighted running mean of baseline U,V,W coordinates versus time.
As each solution bin is completed the averaged integration record is
written to the output uvdata scratch file. The new U,V,W coordinates
and time-stamps are recorded in memory as before.

Solution bins are regularly spaced on a grid of width equal to the
averaging time. Input integrations are binned onto this grid.