Timothy Miller wrote:
Since we're having a bit of a lull in the discussions on OGML, I
thought I might start off a discussion about some ideas I've had that
would benefit the Linux community as well as give OGP and Traversal a
boost financially.
The Open Graphics Project has gotten everyone involved a lot of
attention. While it's still possible to get graphics cards supported
by open source drivers, that supply is dwinding. At the rate things
are going, we'll soon have no choices left. Politically and socially,
the OGP is a great idea. Economically, however, it's entirely a
different story. Because of the development and costs involved,
low-end graphics is actually not such a great place to start. The
attention we get because we want to do graphics is a major driving
force, but there are much better ways to get our stream of funds
started.
One such idea that's been brought up before is ultra high end audio.
The low-end is solved; it's called AC97 and is found in every PC
chipset you can buy today. But imagine taking what would normally
cost tens to hundreds of thousands of dollars in audio recording and
production equipment and applying the open source model to it.
Designing and producing a low-end commodity product is hard. But
designing a niche audio product that sold competitively for thousands
of dollars is relatively easy to pull off. To begin with, we now
become much less cost-sensitive for parts, so the end product can be
FPGA-based. That makes OGD1 an ideal development platform for a new
audio device; in fact, it's major overkill. Once the new design is
finished, we'd mass-produce a new board using a smaller FPGA and
include all of the audio I/O hardware directly on-board.
I know basically nothing about audio technology. But given what
little I do know, here are some things that I think would be
relatively easy to do with OGD1 and, what shall we call it, OAC1:
- 60+ audio channels (pick your combination of in and out)
- (A specialized card could handle lots more channels)
- 24-bit precision per channel
- Sample rates in the hundreds of kilohertz
- Thousands of audio samples and MIDI instruments
- Sample-based and algorithmically-generated sound-effects
- Fourier analysis, band filters, mixing, and other sorts of math
stuff that if I could name it, you'd be impressed
- Noise-free signals (because we have experience with graphics)
- Accelerated "3D sound"
- Accelerated compression/decompression
- Lots of other things
Given hardware acceleration for most parts of audio processing, plus
some excellent piece of open source sound studio software, I don't see
why we couldn't produce a combination that is as good as or better
than what you find in music recording studios, television stations,
and every other place where you might find a need for this. Imagine
how much money this could save musicians. The only thing we can't
provide is the recording room with the proper accoustics.
If a reduced version of the card sold for $1000, we'd have more than a
few gamers and multimedia enthusiasts buying it for their 50-speaker
surround-sound reality-immersion systems.
Were such a project started, I would have to carefully tune my
involvement. Since Howard, Andy, and I do not have a background in
audio technology, it would probably be best for someone with
appropriate experience to lead. But if the community can spec this
product and help us design it (we can put into hardware any algorithm
you specify), Traversal can produce it. The whole project, from start
to finish, would be developed under GPL. This would be quite a major
effort, due to the requirement for more than just drivers. Unlike the
OGP, which started out of a corporation and has retained some of that
flavor, the OAP would have to organize itself and push itself along.
In my mind, the first major problem is getting the right people in the
community together to develop the specs.
I think that the USB "sound cards" that are really nothing more than
DACs make the point that a regular processor and DACs are sufficient for
audio since 44.1 KSPS *is* sufficient for audio. A board would probably
be cheaper if it uses several CPUs since there would be parallel
processing. Although 16 bit is sufficient for a 90+ dB range. However
32 bit processing would a plus since it would eliminate loss of data
when mixing. Note that with 32 bits that floating point is not needed
since you can multiply two 32 bit integers and throw away the bottom 32
bits of the 64 bit result. So, processors with hardware integer
multiply would be a plus -- the multiply accumulate instruction which is
used for FIR filters is also ideal for a mixer.
High quality output is something that is probably important. For 5.1
audio, we should have a dedicated DSP to do the Hilbert transform and
the matrix decoding.
The parts for high quality audio CD decks could probably be used for the
analog output.
The standard high end digital to analog conversion setup is to
over-sample the 44.1 KSPS data stream (for each output channel) 8 or 16
times then run it through a DSP based FIR Bessel filter (probably 5th or
7th order with the theoretical f0 of 22.05 KHz). Then to a DAC which is
still slow by video standards and then through an active filter with f0
of 200 to 400 KHz and a slight shelf filter to flatten the frequency
response curve to 20 KHz. You need a very low noise amp for the filter
since you need to preserve the 90+ dB dynamic range.
Isolation from the digital noise in the computer is important since the
theoretical dynamic range is about 96 dB. The analog audio should be
powered by an isolating DC to DC converter and the bit streams should go
through optical isolators (along with a parallel to serial and a serial
to parallel converters) just before the over-sampling so that the audio
output ground is NOT connected to the computer's ground. It is left to
float on the sound board's end so that the ground reference is the amp's
ground -- this eliminates any possibility of ground noise or ground loop
noise except for the cables and the user will need to deal with that issue.
Another possibility is to have a sound card (which would not need to be
ultra high end) which had a processor on it which was powerful enough to
run TiMIDIty, such a card could also do OGG, MP3, etc.
--
JRT
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
