Subject: Re: [gsmp-devel] Re: [linux-audio-dev] open-source like hardware
From: Erik Walthinsen (omega_AT_temple-baptist.com)
Date: Wed Jan 16 2002 - 23:20:24 EET
On Wed, 16 Jan 2002, Rene Rebe wrote:
> > This is pretty much the design I'm working on, except using a simple
> > microcontroller, not a DSP. What would the DSPs by used for for this
> > device?
>
> To run custom effects! Reverb, EQ, Vocoder, Flanger ... - whatever.
I'm still trying to figure out why you would want to put the processing in
the audio interface box. Yes, it can be low latency for a certain set of
processing on only the channels that that box deals with. But that's not
the point IMO.
A big problem I have with existing hardware is that it assumes you need
symmetric I/O's. I do live mixing, where I need 32 input channels and
maybe 8 to 12 outputs. I want to build a box that has 16 inputs. Then
build two of them. And a box with 8-12 outputs. Hook those to my
computer via firewire. This alone makes it pointless to do processing in
the I/O box.
If I need more processing power than the host computer, I'll add it via
one of several methods:
1) faster processor
2) DSP farm cards in the PC
3) a firewire connected DSP farm that's a bus intermediary between the I/O
boxes and the PC, or otherwise on the same bus.
In the long run I like solution 3 far more than 1 or even 2. Less
form-factor constraints, more scalable solution, lower latency (don't have
to traverse the PCI bus back and forth in addition to firewire).
> Oh 8051. Was really slow ..., I do not know if this chip provides
> enough CPU power to run some effets on it. But since I guess it
> doesn't provide fixed or floading point arithmetic programming effects
> migh be really hard :-(
The Triscend E5 has a 'optimized' 8051, which can run at 40Mhz with a 4
clock/cycle rate. So I dunno if it has enough power to do much while also
dealing with the firewire interface. At most, I'd only be interested in
doing low-latency monitoring, which is still a bunch of MACs, but rather
light-weight. And not necessarily the highest quality requirements
either, though it wouldn't hurt to do full 24-bit.
A more promising solution that I though of is the fact that the chip also
has FPGA bits right there, and this is how the SPI interfaces would be
built anyway. With a large enough variant of the chip (they come with
differing amounts of FPGA), you can implement the monitoring unit right
there in FPGA, for *very* low latency.
Triscend also has the A7, which is an ARM7TDMI core with FPGA logic. It
comes in the same package (probably slightly different pinout of course),
and goes to 60MHz. The design I have in mind has a central controller
board with a reconfigurable (because it's connected to the FPGA logic on
the chip) GPIO port for each group of ADCs or DACs. This makes the
central controller board swappable, so you can switch from an E5 to an A7
or even a pure FPGA, at a later date.
> A DSP (Digital Signal Processor) normally provide sophisticatted
> fixed-point (preferred by me) or floating-point support and are much
> faster (the Motorola 56000, I played with, executes all instruction in
> one cycle and two of these parallel ...)
Yup, but I would put those in a separate farm, not in the I/O. The
question there becomes how efficient can you make a mass of them globbed
together ;-)
> Feel free to email s.th. over: rene.rebe at gmx.net
Will do.
Erik Walthinsen <omega_AT_temple-baptist.com> - System Administrator
__
/ \ GStreamer - The only way to stream!
| | M E G A ***** http://gstreamer.net/ *****
_\ /_
This archive was generated by hypermail 2b28 : Wed Jan 16 2002 - 23:14:21 EET