On Wed, Apr 05, 2006 at 04:47:07PM +0100, Dieter wrote:
> In message <[EMAIL PROTECTED]>, "Timothy Miller" writes:
> > > and list; I see a lot of legacy slots being discussed. PCI ect, is a thin=
> > g
> > > of the past. I am dead serious. IMHO you really *NEED* to be thinking
> > > PCI-Express x1 if that will handle your bandwidth, or possibly x4, or eve=
> > n
> >
> > We're not targetting anything at all. Sometimes I use "PCI" in the
> > abstract to refer to any peripheral bus.
>
> With an external box, consider Ethernet, Firewire, maybe USB.
> Avoid the PCI vs PCIe vs whatever problem. Leave the computer
> with its accustic and electrical noise in the next room.
This idea seems to me to be the key concept. The "sound card" idea
is OK for consumer-level stereo audio, gaming, and so on. For truly
high-end professional recording hardware, the sound equipment doesn't belong
in the computer, and the acoustically noisy computer doesn't belong in the
studio control room -- unless you want to design a specialized one, with big
heat sink assemblies outside the case, cooled by natural convection.
High speed Ethernet, or one of the other high-speed serial links,
breaks the dependencies between the expensive audio hardware and the
computer's interface standards. Besides getting the low-level analog
hardware out of the electrically noisy computer environment, it
future-proofs it. It cuts it loose from the constant standards churn in
PCs.
Also, it would get rid of an expensive and bulky piece of field
recording equipment: the snake. Instead of a passive connector box on the
stage, connected to the recording engineer's station by hundreds of feet of
large, heavy, and expensive custom multichannel shielded cable, there would
be a little Cat 6 Ethernet cable running from the engineer's console to an
active A/D conversion unit on stage. Even better, a media converter box to
change between UTP and free-space optical communication would get rid of the
whole hassle of laying a cable the length of the auditorium and having the
audience walk on it or trip over it. I envision a box with some 1/4" jacks
for direct mike input, and some digital input jacks for digitized stage
mikes. Input boxes should be chainable through the Ethernet interface, so
that channel expansion is just a matter of plugging more audio input boxes
into the hub.
This piece of gear could outlive many generations of computer
technology. It wouldn't need drivers as such; just a set of standards for
the protocols to be used over Ethernet. This makes it a lot easier to
support as a cross-platform peripheral. Since multichannel audio recording
is fundamentally time-division multiplex, TCP/IP doesn't really belong in
the system. ATM over Ethernet might be a better fit. Running multiple
digitizer boxes over a single subnet would be basically a problem in time
slot assignment.
Back-of-the-envelope numbers: if a single audio sample is encoded as
a 32-bit data word, and the sample rate is 192K/S, then the data rate per
channel is around 7.68Mb/s with ECC overhead. Throw in another 20% for
packet overhead, and the rate is maybe 9.6Mb/S. So a 100Mb/S Ethernet cable
would max out at 10 channels, while a 1Gb/S path could carry 104 channels.
Another necessary piece of hardware is an Ethernet-connected control
console. The mouse-keyboard-monitor concept is just too clumsy for anything
as interaction-heavy as sound mixing. A recording engineer needs a real
mixing board, with real knobs on a per-channel basis.
The next thing to figure out is partitioning the functions between
the A/D box, the engineer's console, and the computer. Many setups might
not need the computer at all. After all, big disks are cheap enough now to
pop into the engineer's console.
Now: what's going to sit between the raw serial data stream from the
A/Ds and the Ethernet interface? Maybe a power-efficient laptop CPU and a
big EEPROM to hold the firmware? Maybe a mostly-hardware implementation in
an FPGA, to do simultaneous sample-and-hold? It's gotta be done with no
moving parts and a light power drain, because it's lying on the floor right
under the mikes.
What I really don't see is where there's a place in any of this for
an open design hardware project. What isn't off-the-shelf generic parts is
really painstaking system design, involving knowing high-end audio design
practice in detail, chasing down a lot of second and third order errors and
noise sources, and getting the ergonomics and packaging right. This favors
career specialists.
On the other hand, we understand very well what a computer graphics
card is supposed to do. We don't need to do a lot of market research before
we can even begin to propose realistic concepts.
BTW, there's no such thing as a 32-bit DAC or A/D. With
thermodynamic noise in the range of a microvolt for a low-impedance mike,
that would require a full-scale signal level of +/-2148 volts. Aside from
that, physical components aren't stable enough for much over 20-bit
accuracy. What might make sense is floating poing with a 24-bit mantissa
and an 8-bit exponent; that would avoid the need to adjust levels during
mike check; it would facilitate automatic level control without sacrificing
dynamics.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
