[sdiy] 2164 4P LPF perfected

[David G. Dixon]

Hey Team,

I am a happy camper today, and since the list has been so quiet lately, I
thought I'd send a wee message and toot my own horn a little.

As you probably know, I have designed a cascaded-stage 4-pole lowpass filter
with 8 quadrature outputs based on the SSM2164 quad VCA.  There are two
innovations in this design which distinguish it from other COTA-type
filters:

a) distributed resonance gain to equalize the quadrature output amplitudes
b) reverse exponential gain-VCA response to smooth out resonance control

You probably also know that I've been developing a multimode matrix filter
add-on for this filter.  However, as part of my simulations, I realized that
the distributed gain aspect of my filter would give the wrong responses in
the multimode interface when the resonance was turned up.  Hence, I altered
my circuit by adding a switch so that one could select between distributed
resonance gain and conventional resonance feedback.  This switch does two
things in "conventional feedback" mode:

a) it grounds the VC pins of the first three resonance gain VCAs to give
unity gain through the first three stages, and

b) it adds input resistors in series with the mixer input and the fourth
stage output on the fourth stage's gain summer to increase the current
through that summer by a factor of 2.82 (twice root 2), thereby increasing
the overall gain on that summer at oscillation from 1.41 to 4.

The switch is 3PDT, with one pole grounding the first three gain VCAs and
the other two adding the summer input resistors in parallel.

The good news is that this works beautifully, but I've discovered something
very interesting that I didn't know before about this circuit.  When the
resonance is turned all the way down, the filter gives identical results
regardless of the position of the switch (of course).  However, when
conventional feedback is employed, the input signal is sacrificed as the
resonance is increased, such that the signal nearly disappears when the
filter is near the point of self-oscillation.  However, when distributed
resonance gain is employed, the signal remains at unity gain across the
entire spectrum of resonance.  That means that resonance can be changed
manually or controlled with CV and there will be no decrease in audibility
of the filtered signal.

The very first filter I ever built had this distributed gain feature (it was
the 13700 4P LPF, based on OTAs rather than VCAs), so I've never actually
used a filter that suffered from this loss-of-signal problem (I've never
bought a module or even a PCB).  Indeed, I never realized just how lame
these filters can be with increasing resonance until I installed this switch
on my filter.

The only downside of distributed gain is that self-oscillation is just a
little bit less robust at low frequencies than in the conventional feedback
case, and thus requires the resonance trimmer to be set a bit higher if one
wants to use the filter as an LFO.  Part of this has to do with the reverse
exponential response as well, which allows for less overdrive at the top end
of resonance.  It may be worthwhile to replace these trimmers with small
panel pots.  In fact, it would be nice to put the tuning trimmer on the
panel as well.

In any case, I think this circuit is now just about perfect.  My goal was to
have a COTA-type filter which would be a robust 4P filter, an excellent
quadrature oscillator, and a suitable core for a matrix filter interface,
all without compromise, and I think I've achieved all three.  Please email
privately if you would like more information.