Post by Urs HeckmannHiya,
now... what exactly are the Z-Plane filters?
First off, let me say that Z-Plane is a trademark of E-Mu, so a more
technically correct classification is required. I'd basically call them
multi-segment interpolating filters.
There are two parts to this filter architecture - the filters themselves
and the coefficient interpolator. The interpolator can store up to six
"filter presets", and then interpolate between all six presets. Think of
each preset residing in a corner of a cube, and then have three control
parameters (CVs, CCs, whatever) be able to move in the X, Y, and Z
directions. This is why they called the first z-plane synthesizer
"morpheus" - because you could morph between filter presets.
Post by Urs HeckmannI kept thinking that they are a couple of LP/HP and Peaking filters in
series, or maybe with some BP in parallel?
Specifically, the E-Mu filter architecture consists of seven two-pole IIR
sections connected in series. The first being a lowpass filter, and the
next six sections being parametric filters. Each segment has controllable
Cutoff, Q, and the parametric sections have controllable bandwidth. This
allows you to create much more detailed filter curves than your standard
-12/-24dB per octave lowpass curves of the old analog days.
Post by Urs HeckmannI think I can remember that they have like 14 poles. But then, putting
them into a single set of coefficients (multiplying the transforms)
would make them pretty instable even at double precision, no?
Post by Urs HeckmannHmmm, or am I completely mistaken?
Yes and no - you get a "14 pole filter" by adding up each 2-pole section.
Each set of coefficients described in the ARMAdillo spec on AES and the
patent are for a single 2-pole section. Rossum developed this spec (which
is really a funky form of logarithmic compression) so he could implement
the filter morphing in hardware with minimal loss of fidelity when you
interpolated between the preset coefficients. There's some additional
stuff to the biquad IIR architecture I can't talk about, which is silly...
but most competent DSP folks can figure out how to make a biquad section
like this stable across the audio band.
The cool part is when you think about it in hardware terms, it gets really
impressive. Dave Rossum basically got 64 2-pole filters into a 20,000 gate
ASIC with decent dynamic range in 1989. In 2001, E-Mu had a chip on the
drawing board called the H2.0, which increased the number of filters to
384 on a single 250K gate ASIC - and that's the magic behind the ARMAdillo
algorithm. Encoding the coefficients to make it easy for a hardware ASIC
to interpolate, and shrinking gate budgets to make the hardware more cost
effective.
Post by Urs HeckmannPost by j***@det3.netI'm
currently working on an AU that implements something similar to E-Mu's
Z-Plane using RBJ's filter classes that have been all over here for a
while.
Post by Urs HeckmannPlease, keep us up to date!
All updates for the project will be placed here:
http://proteus.det3.net/projects/sonos/
Don't laugh too hard - I like to start in marketspeak, so I know how big
of a boondoggle I'm getting myself into before I build. :-) I'll start
seriously hacking on it in a few weeks, after fall semester is over at
University.
Best Regards,
Joe Grisso
Det3 Media, Ltd.