[sdiy] Nonlinearities in IR3109 filters

[Sean Costello]

From: "karl dalen" <dalenkarl at yahoo.se>

> You assumed wrongly. Besides are you saying that digital filters
> implementations dont simulate the use of diodes in fedback paths? ..Ugh!

Well, the plots I saw of the SSM 2040-style filter was explicitly assuming
no clipping in the feedback path. The amplitude of the self-oscillation,
relative to the signal, was much higher than in the Moog filter model.

In general, digital filters have experimented with having the clipping in
all sorts of locations. Harvey Thornburg had some slides in 1998 or so that
showed having various softnesses of clipping in different parts of the
Stilson/Smith Moog filter. In these slides, the clipping tended to be
between the 1-pole filters, rather than within the filter. The feedback path
was another location shown.

Other digital filters use direct form II structures, and have the clipping
in the summer for the poles. I have seen filters with clipping feedforward
only (i.e. two cascaded 2nd order filters, with clipping between the first
and the second). Reaktor used to have a bunch of different examples of
nonlinear filters, including the feedforward concept, as well as cascaded
1-poles with soft saturation between each stage.

In fixed point, it is moot, as you will have hard clipping throughout the
structure. Be sure to turn saturating math on, otherwise the distortion will
get pretty ugly.

Sean Costello