discrete SSM2040-style filter

From Haible_Juergen#Tel2743
Sent Tue, Jan 7th 1997, 21:20

Hello again!
A happy new year to all of you!
I had the time to build some things on veroboard during the last
weeks. But before I talk about the thing I proudly announced in the
topic line, let me report about a failure first.

I tried to build a 16-stage phaser around 4 CEM3320 filter chips,
because I thought it would be so easy. The idea was to build
something like the Moog phaser, with variable stages and so on.
I wanted to use chips with a good SNR (compared to the CA3094)
and so I would not need a compander circuit.
Unfortunately, I ran into serious stability problems. *One* chip
(i.e. 4 stages, like in Barry Klein's book) works perfectly, but
if I cascade several 3320's, the whole thing turns into a radio
receiver at certain control voltages. Even without feedback applied.
Must be some feedback over the supplies, I guess. I put a *lot*
of decoupling capacitors into the circuit, but it wouldn't help.
So I gave it up, and I will use 3080's or LDRs if I will build a phaser
again one time.

That was the bad news. Now to the VCF, which was just the
contrary. But let me tell it one by one.
For my polyphonic project, I always wanted to have a 2pole
filter and a 4pole. 2pole should be a SEM-type state variable,
no doubt. But the 4pole wasn't that clear. First I wanted to
use my beloved SSM2040 chips ("Best filter chip ever") to get this
famous Prophet5/rev2 or Kobol sound. But then I thought that
someone else might want to build this synth as well, and using
ulta rare chips is very problematic then. So I was almost sure that
I would use a Moog ladder instead - my second-favorite 4pole filter.
I even had already designed this part of the pcb. But then I tried
something better, something I always wanted to try, but never found
the time so far: Building a "clone" of the SSM2040 with standard
Well, I must be careful with the word "clone", because this would
mean a one by one reproduction, which it surely isn't. I don't even
know all details of the original circuit. But at least I know the circuit
of the signal path (Dave Rossum was so kind to answer a letter
about this question), and here are some details that I think are
important for the SSM2040 sound:
(1) A minimum (for a ota/buffer filter) of pn junctions in the signal path,
     due to a very minimalistic ota topology: the ota just consists of
     a differential pair and one single current mirror.
(2) The very special distortion scheme, due to the same minimalistic
     ota topology: The ota output can only swing into the positive
     direction. The buffer is a npn darlington, so levels are shifted
     by approx. -1.4V, so you get a very unsymmetrical clipping
     at the output. This, together with the frequency-selective distortion
     of the ota-inputs, gives a very unique overdrive charactersitic.
Now this is how I build my discrete filter:
I selected 8 transistor pairs by the Moog method to +/-1mV:
4 pairs of BC550C (npn, for the differential pairs), and 4 pairs of
BC560C (pnp, for the current mirrors). The bases of the npn's
are connected to gnd by 200 Ohm resistors. The pnp's form
a current mirror for this pair, to the positive supply. (The 2040
uses 3-transistor mirrors which compensate for base currents.
But with discrete high-beta pnp's, 2-transistor mirrors are fine.)
This is the whole ota: 4 transistors ! At the ota output there is
a 1nF capacitor to gnd, and a darlington buffer made from two
BC550C transistors. The darlington works into a 500uA current
source which is made from a 7th transistor (again BC550C).
The feedback network consists of the 200 Ohm base resistors
and two 10k resistors, from the input and from the buffer output
to the base of the transistor that forms the inverting input of
the ota.
Four of these stages are connected in series. The currents into
the emitters comes from an exponential converter, built
around a  CA3086 array and a dual opamp. One transistor
for temperature compensation, 4 transistors for the 4 otas.
Emitters all tied together. I chose a reference voltage of -1V
for the expo converter, to allow some voltage drop over the
transistor array. This results in a -1V instead of 0V summing
node for control voltages. This is no problem in my PolySynth,
where the CVs will be mixed with otas anyway. For the usual
applications, you'd need another opamp for level shift.
One thing I had worried about were the offset voltages.
But I only have approx. 100mV offset at the output, and the best
thing is that this offset only changes by a few millivolts over the
entire control voltage range. So it turns out that selected transistors
are better than the monolithic transistors of the original in some
respect. Temperature drift is worse, of course. If I touch one
transistor of a pair with my finger, the offset increases to one volt.
With the limited output swing, this might be a problem. But
heating one transistor to 37 degrees and leaving the other at room
temperature should be the exception, I assume. So I am quite
optimistic that I won't need copper clamps or the like.
I also added a voltage controlled feedback loop (which is not
part of the SSM chip). I tried a 3080 which works fine, but I finally
used another discrete differential pair. Call me purist.
The sound? Unbelievable. A good and clean 4pole sound with
regular input levels, and full power brute force with overdrive.
All I ever wanted from a 4-pole filter. Without overdrive, the VCF's
own frequency dominates when Resonance is set to self-oscillation.
With higher input levels, the input signal dominates. I think this
is also a result of the ota-output clipping, and again it's exactly
what I wanted.
I also added a trimpot and a resistor from the filter input to the
feedback differential pair. This way the usual bass loss with increasing
resonance can be compensated for. I am not yet sure which amount
of compensation is the best, but it's nice to have it programmable
with a single resistor value. As I plan to make crossfades from
2pole to 4pole filters, and as the SEM-style 2pole hasn't got the bass loss,
I think some compensation will be fine.

Hmmm - a long mail that was, and probably all you want to see is the
schematics. I'll send it to Kevin and ask him to put it on the web site.