Subject: Re: [linux-audio-dev] Realtime
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Thu Jun 29 2000 - 00:22:55 EEST
some of the requirements would be:
- dedicated CPU for audio processing, that means you would need at least a dual
CPU box.
- audio IO interfaces which provide single sample buffers ( or better double
buffering at sample level = 2 buffers of 1 sample each).
(not sure how easy this would be to do in practice (even CD players use a few
samples long FIFOs to conpensate jittering)
- the "plugin" model would not work anymore because of the high
procedure/calling overhead, thus requiring sort of
"inlining all DSP modules into a main loop without any procedure call"
In theory if would be possible to develop some DSP language which
would allow to write DSP algorithms on a per-sample basis which could
then be put in a chain, by taking the source code of all modules and
compiling it into a monolithic binary code block which would then be
executed on the host CPU.
But this scheme would make it hard if not impossible to ship
"binary plugins", since using a per-sample model, would require
a recompile when chaining different plugins in order to get decent speed.
In that case our PC becomes nothing more than a host of a dedicated
CPU which is "misused" to act like a real DSP (process sample per sample).
But the advantages would be obvious ( running within a fully fledged OS,
having access to large amounts of RAM , disk storage space etc, plus
allowing this "recompile" trick in order to give almost same flexibility as
the plugin model, but without sacrificing much speed.
or am I missing something ? :-)
Benno.
On Wed, 28 Jun 2000, Tom Pincince wrote:
> While the current architecture uses blocks, and so limits rt
> requirements and possibilities, I see a potential use for an audio
> system that provides sub-sample accuracy. As surround sound makes it's
> way into society, a significant opportunity will present itself
> regarding the psychoacoustic perception of location and movement of
> individual sounds within the sound field. The ability to stream audio
> as single samples and control the rate of playback with sub-sample
> precision would open up some impressive opportunities. This is very
> demanding. Jitter removal may be impossible since variations in the
> sample playback rate would be an integral part of the signal. Such a
> system would bring new meaning to the term "real time audio" and is so
> far beyond the current state of the art that for now it is purely
> academic. I think it is worth pursuing.
>
> Tom
This archive was generated by hypermail 2b28 : Wed Jun 28 2000 - 21:02:14 EEST