Jan,

You were right on target! I did a very simple application that at every 20ms
cycle, it increases the PWM by one degree (I know, the units are kind of
messy, but I guess you'll understand). When it reaches 60 degs, it starts
decreasing until -60 and then repeats forever. I also have a constant to
control angular speed, and I've tested with slow and fast speeds (up to the
servo limit speed) and they all work flawlessly.

I feel that I wasn't too polite not answeing your questions, so here they
are:

> What resolution (in degrees or ms or whatever) do you need ?
> +/- 1 deg ?
> +/- 0.1 deg ?

1 degree is fine

> How many descreet steps ? 100 ?

from -90 to 90, therefore 180 discrete steps, although I think -60 to 60 is
what I will end up using.

> I would probably not try to send a measurment it "real" units,
> but something more efficent.

Right now I'm sending a word with the pulse width in microseconds, what
would you suggest?

> > If the current angle is let's say -20 and I send a message
> > "go to +60", the servo responds immediately and its motion is
> > very smooth (I guess it goes at its regular operating speed -
> > 0.19sec/60deg), but if I grab my pointer on the user
> > interface and drag it, the motion of the servo is very bumpy
> > and jittery. It will end up where I release my GUI pointer,
> > but the trajectory will be rough. The same roughness can be
> > perceived if you are measuring the signal on the oscope.
>
> What "signal" ? The PWM signal probably, right ?

right

> Note that sending characters out from a serial port on a
> (Windows ?) PC isn't realy a "real time" process. My guess
> is that you actualy have a "bumpy" stream of characters
> comming out of the PC. Can you add a "logging" feature to
> your PC program that logs exactly whats sent out ? Best would
> be a serial line logger with timestamps on each byte.

It already has a simple one, it logs a list of messages sent and all of them
are in increasing order (if I'm increasing the angle, of course). I also did
another test by sending a stream of messages created on linear time, instead
of dragging the pointer of my GUI application, it is still bumpy, but you're
right, the problem is somewhere there I guess.

> I'm not sure what speed your PIC is using, but getting an INT
> fromt eh USART, saving the byte and RETFIE'ing whould not take
> many us's, so that should not mess up your hardcoded timing
> code that much anyway.

I guess I could do another test using the "wiper" code I just did (-60 to
+60 forever) by including again the USART code in there but not using it to
change the servo angle, but I agree with you that the problem is not likely
to be there.

> No, I'd take a closer look at whats actualy is sent out from
> the PC.

Do you have any suggestion for which tool to use in this case?

Thanks!

Padu

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist