[sdiy] Moog filter research

[jh.]

Hi,

I spent the last couple of weekends doing some circuit analysis on Moog
filters. It's been some years since I built my last Moog-style filter
cascade.
My interest was clearly revived by the Minimoog Voyager's dual LPF
approach.

After some simulations (I posted) and after some experiments with two
different LPFs I decided to build a compact little dual Moog cascade
with shared controls and inputs plus Separation control. (I think the
Voyager's minimal user interface with just one extra knob for a dual
VCF is truely brilliant! Especially as you practically get the sound
of a single LPF when Separation is at zero, without any switching.)

So far the background. The interesting stuff begun when I wondered
which of the many Moog filters to choose. For some reason, I wanted
to try the old 904A filter. I'm quite sure that the Voyager doesn't use
this (think of a bulky switch to select 3 sets of 8 cpacitors ...) - and
that's where the Voyager influence ends. I was caught in doing an
in-depth analysis of the old modular 904A filter. And this produced
some surprising (at least to me) results.

I remember a thread about the Andromeda, Mike Peake speaking of
differences between various Moog filters. (You modelled the Andromeda's
VCF after the 904A, too, didn't you ?). I remember a question
of mine in another thread whether switching the range on a 904A
does just transpose the cutoff frequency or also alters the sound.
(Meanwhile I'm convinced that it *does* change the sound, but more
about it later.) And there's of course the general knowledge that
the differential amplifier after the filter cascade makes a difference.
This is often quoted on the list, and explains why later Moog
instruments generally sound quite different from the Minimoog
or Modular.

The pivot point is the loading of the ladder. A load resistance coupled
to the final ladder stage changes the frequency response. This is
neglectible
at high cutoff frequencies (where the ladder is low impedance), but
comes into play at low cutoff frequencies (where the ladder is high
impedance). This causes the reduction of loop gain, and thus the decrease
of resonance, at lower frequencies. But there are two different
ways the ladder can be loaded: A differential amplifier has a small
signal input resistance, and it has offset currents on both its inputs.
If we compare two filters with AC coupling in the differential amplifier,
the Minimoog VCF and the 904A, we see that the Mini has its AC coupling
right after the ladder and before the amp. The 904A has its first amp
stage (two emitter followers) DC coupled to the ladder, and then there is
an AC coupling to the second amp stage. Therefore the Mini VCF's
amp has a rather small dynamic input resistance, but no bias current
to load the ladder. The loading causes a decrease of Resonance at low
frequencies, but only down to a certain point: because the AC coupling
increases the input impedance of the differential amp at very low
frequencies.
The 904A's amp has a rather high input impedance (emitter followers
with no resistors from base to GND), but it draws a considerably
bias current. I simulated the circuit, based on typical beta values of the
transistors that were used, and I found that the bias current is in the
2uA range ! (There is no hint in the service manual that these transistors
would be selected for beta, so it would vary quite a bit from one 904A
to the next.)
What does this mean ? The uppermost stage of the ladder is fed an extra
4uA (2uA on each side) from the differential amplifier. This is added
to the normal control current thru the ladder. Whenever the control
current is high, these extra 4uA won't make a difference. But when
the control current is low, the pole of the upper stage is shifted
to a higher frequency then the remaining three !
So when is the effect of this most noticable ? Low currents means
low cutoff frequency. But that's not the whole story. You can get
the same cutoff frequency with a small capacitor and a small current,
or with a larger capacitor and larger current. (I assume that the rather
high leakage currents of early transistors was the reason for Moog
to use rather large capacitor values. Up to 1.2uF in the 904A, compared
to 10nF in the FET-buffered Moog Source.)
So my conclusion is that when you switch the range on a 904A
(which means switching between sets of capacitors in the ratio
of 1:4:16), you're switching the sound as well as the range. You
won't immediately notice, because the range switching dominates.
But I encourage you to make the following experiment:
(Actually, I'm asking for somebody to make these experiments
and share the results, as I don't have the real hardware here.)

==============================================
Experiment 1:

Switch the range of a 904A from 2 to 3, and apply a CV of -2V
to one of the inputs at the same time. Try it at various Resonance
and cutoff settings, and try to find a difference. The same
for Range 1 and 2, with a CV of +2V when Range 1 is set.
Is there a difference ? How big is the difference ? Is it more
prominent at low cutoff frequencies and high resonance settings ?
===============================================

Meanwhile, I have almost finished my version of the dual VCF.
Instead of switching capacitor banks, I used a trick: A quad voltage
controlled current source is used to create a variable bias current
for loading the ladder. This can either be switched together with
an ordinary cutoff transposing (+/-2V CV), or it can be applied
independently with a potentiometer. So far, I get the most convincing
results when I just emulate the capacitor switching, i.e. when I link
the bias current source to the transpose switch. The sound reminds
me of certain recordings which were positively made with a Modular,
and which cannot be reproduced with my Minimoog clone nor
Taurus clone. But this might as well be wishfull thinking, and so
I'd like to have this backed up by some more experiments from
experienced Moog Modular users (If somebody likes to do this):

==============================================
Experiment 2:

Set the Resonance to self oscillation at a rather high frequency
(2kHz) in Range 3. Don't just set it on the treshold of oscillation;
solid feedback is needed. Now reduce the cutoff frequency and
note when the oscillation stops. (If it doesn't stop, start with
less Resonance amount at 2kHz, but keep the same setting for
all measurements in this experiment.) Note the lowest frequency
that has self oscillation.
Do the same thing in Range 2 (do *not* change the resonance
potentiometer setting !), and note the lowest frequency pof self
oscillation for range 2.
Do the same thing for Range 1.

Repeat these 3 measurements with Resonance fully turned up
and note the 3 lowest oscillation frequencies.
===============================================

Switching the capacitors also means a limitation of the available
frequency range, because the current thru the ladder is limited.
I'm not speaking of the range of self oscillation here, just an
upper limit of cutoff requency dependig of the "Range" switch.

According to my simulations, Range 3 should go far into the
ultrasonic, Range 2 should still cover the whole audio band.
But Range 1 (wuth the 1.2uF caps) should be limited to
approximately 5 ... 8 kHz. If the results are right, this could not
be further increased with external CVs, either. If so, this
can cause a special effect when sweeped with an envelope
or a modulation oscillator, as the filter response would
follow a waveshaped (clipped) version of the modulation signal
when Range 1 is chosen. Of course this will make a huge
sonic difference with wide sweeps, whether resonance stays
prominent in the upper audio range or escapes into the ultrasonic.
Can ynybody confirm this?

===============================================
Experiment 3:

See how far the filter goes up in Range 1 setting. Use self oscillation,
and use internal and external CVs to push the theoretical cutoff
frequency up, and note at which frequency it's stuck in practice.
===============================================

So, this was a long mail, and I hope somebody found it interesting.
When somebody is interested enough to make my little experiments,
this would be even better. It's like to get confirmation, or correction,
for my findings. I don't own a Moog Modular, but I want to see how
close I can come with emulations. I would also be interested
if these thinks have been considered in other emulations (Andromeda.
Mike ? Any comments ?)

And finally:

=================================================
Experiment IV:
This is Kate's, of course. (;->)
=================================================

JH.