[evlatests] Speeding up Correlator Configurations

[Ken Sowinski]

On Fri, 5 Jul 2013, Rick Perley wrote:

>    The correlator gurus have found a way to considerably speed up
> correlator reconfigurations.  (Ken can provide the details).

It is more accurate to say that the true speed of correlator 
reconfigurations has now been revealed.  Since the early days of
Widar we suppressed data from the baseline boards for 10 seconds
after a reconfiguration to be sure that no inconsistent data was
allowed in the archived data.  Lately we have experimented with
reducing that interval and demonstrated that even with no suppression
all the data is valid and is properly assigned to scans.  Rick's
test is a confirmaton that this is so.

As Rick remarks, this gain is often made moot by flagging due to
subreflector rotation, so the greatest benefit is when reconfiguring the
correlator without a change of band.  It is worth noting that the time
to rotate the subreflector from X band to any of the high frequency bands
exceeds ten seconds; while the worst case for Ku band is five seconds
to go to Ka band.  Two lessons derived from this is that Ku band
is a better choice for referenced pointing, and we should investigate
whether the ACU upgrade allows faster rotation of the subreflector.

We will probably change the default blanking time in Widar to a much
smaller interval next week.

>    Ken has executed a pair of short observations designed to measure
> the improvement.  This consisted of a ~1 hour observation utilizing my
> 'flux density' script.  The method here is to change bands rapidly
> (typically 30 seconds on each band) for a number of sources.    This
> short section of the big script was run twice, once with the new system,
> and again with the old.  This permits direct comparison of the durations
> over which data were available from the correlator.
>
>    The results indicate that the configuration change has been sped up
> by about 10 seconds.  So, for the old method, the configuration change
> from S to Ku band took 14 seconds.  Now it takes 3.  (Nice!)
>
>    Below I give the detailed times for each of the band changes in my
> script.
>
>    Transition        Difference
> ---------------------------------------
>    Ka to L                 11 sec
>    X to C                    8 sec
>    C to S                     0 sec (no configuration change)
>    S to Ku                  11 sec
>    Ku to K                  9 sec
>    K to  Q                   0 sec  (no configuration change)
>    Q to Ka                  0 sec  (no configuration change)
> ----------------------------------------------------
>
>    Note that most of these savings won't be seen if one is changing
> bands -- the subreflector motion is typically ~5 to 10 seconds between
> bands.  The big advantage is for changing modes without changing band.
>
>
>
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
>