[mmaimcal]Re: [Almasci] Antenna Acceleration
Richard Hills
richard at mrao.cam.ac.uk
Thu Nov 20 06:22:31 EST 2003
Dear Tetsuo,
I believe your understanding of the messages is correct.
I very much agree that a simple servo loop working on position error would
certainly get into trouble trying to follow a trajectory like this. My
simple-minded understanding of the way round this (which I hinted at in my
earlier message) is that the servo should look at the actual velocity,
acceleration and "jerk" required, as well as the position, and work out what
current it should apply to the motors to achieve this. It won't get it
perfectly right but it should be close, because of imprefections in the system
and the things you point out like differences between antennas, etc. The
position servo then has to deal with these imperfections. We did this at a
simple level on JCMT (there we were just using "feed-forward" of the velocity
but not the acceleration) and it works reasonably well when properly tuned up.
We found this was better than having two modes - one for rapid motions and one
for e.g. tracking a given object.
I also agree that if there is a problem with executing fast scans then we
should certainly look at the definition of the interface again. As things
stand we only tell the telescope what we want it to do about 1/20th of a
second ahead of time. Now clearly the higher level software knows what it is
planning for much longer than this. At least a few seconds ahead and
typically several minutes. If it would help the servo to know in advance then
it ought to be possible to tell it, but obviously this will complicate the
interface.
I note also that if one is doing a repetative pattern, which we very often
will be, then in principle the servo could learn from what happened on e.g.
one turn-around how to improve its performance on the next attempt.
What I don't know is what has actually been implemented, or what performance
is either predicted or achieved. I think we don't have the right circulation
on this mail discussion and we need to bring in someone who does know about
this topic - presumably Steffano and/or Jeff Magnum should be asked who is the
right person. The trouble is that they are all rather busy at the moment.
In response to Robert Laing's question - I am sure that simulations have been
done and that enough must be known to be able to generate a simple generic
model, but again I don't know who has their finger on this.
Best Richard
More information about the mmaimcal
mailing list