Re: [LAD] OSC for syncing

Hi Victor,

On 06/18/2010 11:35 AM, Victor Lazzarini wrote:
>
> a question for one of your OSC gurus out there: is it possible/a good
> idea to use OSC for syncing? How good is it for that kind of thing?
> So far I have only used it for parameter control, but is there a good
> means of using it with timing accuracy?

I can't pretend to be an OSC guru, but I've been thinking and working quite a
lot on OSC and timing accuracy.

"Syncing" can be many things, but I'll take an example: let's say that a step
sequencer sends OSC messages to a sampler. The sequencer might send timestamped
messages a bit before they ought to result in playback, to compensate message
transport time, as well as latency and related issues on the sampler side.

So sequencer sends a message at instant T-1 to tell the sampler to start
playback at instant T.

First a very practical consideration. If you use liblo, then you can't rely on
the high-level API to receive timestamped messages, because it has its own
message queues and will delay the message by itself. That forbids taking latency
into account. I made a bug report about this:
http://sourceforge.net/tracker/?func=detail&aid=2858774&group_id=116064&atid=673869

So you'll have to use the liblo lo-level API, which means extra work.

Now, the clock issues. OSC timetags are absolute timestamps, so you need a
reference clock. On a single system, that will be the system clock, and you will
certainly use gettimeofday() (altough some might disagree about this).

The problem is that the sound card has its own clock, so you're dealing with two
clocks, which causes a variety of problems, as described by Fons in hi essay
about Delay Locked loops: http://www.kokkinizita.net/papers/usingdll.pdf

/Simplifying it/, those clocks do not run at the same speed, they drift. Plus,
there's a jitter between soundcard hardware interrupts and the moment your (JACK
or whatever) process callback gets called. It's not getting called as regularly
that one would expect.

To circumvent these issues, I have been working on libtimefilter (formerly
libpendule), which is an implementation of Fons' proposal:
http://code.google.com/p/libtimefilter/

Now, last but not least, almost all of these problems would become a souvenir,
if JACK supported OSC ports, as it does for MIDI, and that of course both the
sender and receiver use that.

Hope that helps,

--
  Olivier
_______________________________________________
Linux-audio-dev mailing list
Linux-audio-dev@email-addr-hidden
http://lists.linuxaudio.org/listinfo/linux-audio-dev