Subject: Re: [linux-audio-dev] Software filter engines for high end audio
From: Anders Torger (torger_AT_ludd.luth.se)
Date: ke helmi 16 2000 - 07:22:12 EST
On Wed, 16 Feb 2000 sbenno_AT_em.gardena.net wrote:
> >
> > I plan to use the filter in software driven room correction systems, and
> > in fact I have already a working prototype based on the WFIR filter, found
> > at http://www.ludd.luth.se/~torger/filter.html. Complete with an MLS
> > measuring system. All GPL'd of course. I'd love to get comments and
> > questions about the system :-)
> >
> > The plan is to make a room correction system based on the FIR filter
> > later, which will be able to correct phase problems, as well as magnitude.
> > The current system does not correct phase.
> >
> > I find this area very interesting, since it is possible to make your Linux
> > machine (which must be rather fast, I use a dual P2 266, one processor per
> > channel) into a room correction system that is on par or surpasses
> > commercial systems that cost around 4000 US dollars. The problem so far
> > has been the lack of quality digital I/O cards for Linux, but soon we will
> > have a driver for RME Digi96, the best alternative (price/performance) I
> > have found for two channel systems.
> >
> > /Anders Torger
> >
>
> very interesting :-)
>
> It would be nice to know how much better the price/performance ratio
> of your solution is compared to the $4000 hardware.
>
> would this hardware be enough:
> - Dual Celeron with Abit Motherboad
> - 64MB of RAM
> - cheap gfx card (or leave it out completely)
> - cheap hd
> - Dig96
>
> I don't know how much the Digi96 costs, but do you think
> you can achieve a $1000-$2000 price ?
>
> Just like I predicted ... some "standalone audio equipement" will sooner or
> later run Linux. (because of the excellent price/performance ratio)
> :-)
>
> Benno.
The problem one have to deal with when doing software only implementation
of a DSP algorithm, is that an ordinary CPU like the Pentium is not a
DSP. An example, the WFIR filter loop in my filter is implemented with 30
instructions (executed in about 20 clock cycles), which can be implemented
with four instructions on a Motorola DSP. So one really wants to use DSPs,
but there are no cost effective options to build a single DSP-based
computer (afaik, I would love to hear about any cheap solutions out
there). What an ordinary CPU has, that an ordinary DSP don't have, is a
high clock frequency, and it is on that programs like mine survive.
The Digi96 is a 24 bit digital I/O which costs about 300 US dollars. The
hardware you mention would certainly be enough. Only 8 or 16 MB RAM would
be enough if it should be a dedicated filter engine. I have not
investigated how small the programs can get, but you after some tweaking
probably get rid of the hard drive and use a diskette instead. If you want
to use it as a filter engine, you probably want it to be silent (or have
it in another room, that is my solution), without the hard drive it should
be possible to make it entirely silent, by removing fans from processors
and transformer. If we are talking really seriously, we should also
implement the filter in real-time Linux, so we get lowest possible delay,
and highest possible realiability. The 1000 - 2000 dollar price I think is
doable for a dedicated box. However, you will need to buy and external
DAC also, if you do not own one already.
For now, I target at those who own a computer already, and want to use it
as a filter engine now and then. When you already own the computer, this
is surely a cost effective solution :-).
When comparing room equalisation system performance to commercial ones,
only one part lie in the number of filter taps. The other is in the filter
design, how to handle the measured in-room impulse response. Simply making
an inverse filter does not yeild good results. I cannot compete with Tact
Audios RCS when it comes to the number of taps, it can also handle 24 bit
input signals. Since I rely on MMX when it comes to the FIR filter (which
is not complete yet) I can only have 16 bit input, the output is of course
24 bit though. Compared to Roisters system, my filter is more powerful. I
have unfortunately not been able to listen to any of the systems, but I
believe Roister has the better filter design technology, it can correct
even excess phase, which I plan to do also. Tact Audio's system is only
minimum phase (I've heard that they are working on a linear phase system
though).
To summarise, yes it is possible to meet the price, but it is quite
complex to compare the performance of a complete room equalisation, since
it is not only dependent on pure filter engine power. You could to other
stuff than room correction with a powerful FIR filter though of course.
/Anders Torger
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:23:27 EST