Subject: Re: [linux-audio-dev] LADMEA Prototype
From: Steve Harris (S.W.Harris_AT_ecs.soton.ac.uk)
Date: Fri Aug 17 2001 - 16:50:46 EEST
On Thu, Aug 16, 2001 at 08:54:01PM +0100, Richard W.E. Furse wrote:
> You will have realtime/clock sync problems for instance when your clock is
> on a different box to a real audio input or output. What happens if the
> crystal on the audio card isn't in sync with the clock. Or one audio card
> with another? This is a common problem.
Yes, sorry, I wasn't making myself clear. I was planning to use other
machines only for processing, not for audio i/o, so the sync would be
redived from the bridge.
I still think that the resyncing problem is too hard to solved to my
satisfaction on a PC. I would be unhappy if sync drifting on some hardware
caused the CPU usage to jump.
> Assuming this problem is resolved somehow, there's still no guarantee of
> relative arrival times across a network (unless you want to use slow
> centralised locking techniques, which will fail for latent networks). You're
> going to have to implement latency requirements on inputs to judge when a
> client has failed/is late. If you want to stream things other than PCM audio
For the network I was thinking of isochronous IEEE1394 streams (a personal
favourite of mine, reliable, common, cheap).
> Incidentally, as I think you suspect, it *is* possible to drop or add
> samples to deal with sync problems. This is a well-known issue that hardware
> digital mixers in the real world have to face. This is partly why they are
> expensive - but the technology has been around for quite a while and we
> really ought to be able to mirror it in our software-synthesis world. A
> mixture of techniques are used, from dropping or inserting samples (bad) to
> resynthesis of a block of audio with a length change (good but costly).
Agreed it is possible, but AFAIK not all digital mixers can do it very
well, my 01v certainly can't resync without audible glitches, and doesn't
claim to. It flashes up big warnings when you don't have 1:1 sync across
the ports.
> As I've said before, the zero-copy issue I'm flexible on, I'd just prefer to
> put it off until more fundamental issues are resolved. It isn't conceptually
> hard but it is fiddly.
As long as your sure it *can* be implemented then I'm happy on that front,
something I'm not clear on: can LADMEA have inprocess applications? The
cost of context switching is less than it was, but still an issue for
large numbers of apps, and I would want to be able to implement common,
simple things as apps (eg. noise gates).
- Steve
This archive was generated by hypermail 2b28 : Fri Aug 17 2001 - 19:01:16 EEST