waveform conversion

[Arthur Harrison]

I have a circuit we devised that converts a square wave 
into a triangle.  There's one hitch:  it takes a finite number 
of cycles to settle.  The operation is this:  the square wave 
is fed to a voltage controlled limiter.  In some cases, this is 
merely a pair of transistors arranged one on top of the other, 
so that the emitter of the top transistor feeds the collector of
the bottom one (sort of like a cascode arrangement.)  The 
square wave is fed to the lower transistor's base, so that it 
reappears at its collector, except at an amplitude controlled 
by the upper transistor (and, of course, 180 degrees  
out-of-phase.)  This point is then connected to an integrator 
comprised of an op amp, resistor, and feedback cap (the 
standard configuration.)  This integrator has a local servo 
consisting of a second op amp hooked up as a slower 
integrator, whose job it is to keep the integral's DC average 
zero (or 1/2 supply, in the case of a single-supply design).  
The integral waveform, which is now a triangle with a frequency 
equal to the square wave, is fed to a full-wave detector comprised 
of two more op amps and two diodes (the standard precision 
full-wave rectifier scheme.)  The output of the detector is a triangle 
twice the frequency of the square, with a DC peak value 
corresponding to the limiter's output amplitude.  This waveform 
is then fed to yet another integrator, which provides a DC value 
directly correlated to the limited square's amplitude.  Then, this 
value is compared to a DC reference in another op amp, and 
the output of that op amp is fed back to the upper transistor's base.
  
In tests, we were able to make this scheme track over a 470Hz to 
20kHz range.  This could be scaled down, however.  I don't know 
how applicable this idea is to music synthesis, since all those 
integrators need time to settle.  The result would be a timbre 
change at each frequency transition.  Actually, that might sound cool 
in some places.

Art Harrison
Theremin Guy