State variable filter (SVF) with self-oscillation.

Entry: Schematic notes
Date: Mon Jan  3 17:41:23 EST 2011

See svf.sch

This is a textbook SVF extended to gracefully handle self-oscillation
in a controlled way.

Operation:

* The R5 log voltage divider provides negative feedback to lower Q.
  Value of R5 is not critical since it's not loaded.

* R6 is for lifting that negative R5 feedback into positive when
  feedback trough R5 approaches zero.  This is to ensure oscillation
  when the R5 resonance potentiometer is turned full scale (no
  feedback).  Without this slight positive feedback I only got self
  oscillation at the higher audio frequencies.  More positive feedback
  gives more distorted self-osc waveforms.  With R5=1M we have 100k /
  1M or 0.1.  The resulting limit cycle is pretty close to sinusoidal.

* Diodes D1-D4 predictably limit the amplitude in case of oscillation
  due to positive feedback.  It's best not to leave saturation to the
  opamp.

* R1/R2 is a dual antilog potentiometer that sets the pole frequencies
  with R7 and R8 used to limit the opamp load and set the highest
  frequency.  Full resistance is low frequency, short circuit is high.
  A log pot usually has 1/10 of total value at 1/2 scale, meaning it
  covers 2 decades, hence 50k / 470 = +- 100.  Max opamp current
  source/sink is +- 1mA: a diode drop (0.6V) over the limiting
  resistor (470R).

* Input gain is set by R4/R3 which is 10 in the schematic.  This seems
  to be reasonably optimal for unity buffered guitar inputs (100mV ->
  1V), which is what I use to test.


Other remarks:

* The pot is set at 50k because that's the only dual log pot I have.
  Probably best to change to 500k or 1M (and lower the caps
  accordingly) because the currents get large at the high resonance
  frequencies.

* Overall it's my idea that log pots to set frequencies isn't such a
  good idea as the taper curve of pots is pretty bad overall and they
  are quite pricey.  s proably better to use voltage or current
  controlled pole frequencies combined with a true exponential
  converter and more predictable linear pots.

* Opamp is TL074 quad, which seems to be the cheapest opamp good
  enough for audio (Futurlec: $0.27 for 25+, Tayda: $0.20).

* While it's cheap, don't use LM324 quad: crossover distortion is
  quite severe.  Either the output needs to be biased into class A
  output (resistor to V+ or V-) or it needs to be used with a lot of
  feedback to compensate the crossover nonlinearity of the class B
  output stage.

* The circuit has ground as opamp bias but I built this with a single
  supply and a 1/2 Vcc buffered bias reference.  Just replace ground
  with ref.  For proper power supply rejection do use a zener or other
  true voltage ref, or use a battery or linearly regulated power
  supply that has no ripple current.

* To limit the self-oscillation, amplitude limiting is only necessary
  in one of the integrators, or possibly after the summing inverting
  amplifier.  However, this will lead to a different overdriven filter
  sound depending on which one is taken.  A switch could be used to
  toggle between different configurations.