AW: Expo conv. heater (tomg et al.)

[Haible Juergen]

	>The stability of the ARPs is due to their "linear voltage
	>   to exponential current converter" circuit, which consists of a
closely
	>   matched pair of transistors (one NPN and the other PNP) stuck
together.
	>   The heat drift of the two transistors cancels out and the output
stays
>  >  very stable. 
> 
The matched & coupled transistors, *and* a tempco resistor are
responsible for the good stability. I'm not splitting hairs here, but
it's important for my point to come ...

	>I've been messing with a circuit using a heated exponential
converter
	>lately, as well, as part of my Roland GR-300 guitar synth hacking
project.
	>I need a log(1/x) converter to turn the GR-300's pitch control
voltages
	>(which are proportional to the cycle period of the guitar string)
to a 1
	>volt/octave CV to use with everything else in the world. My analog
divider is
	>based on the one in Electronotes AN-114 (which is in turn a lot
like the
	>one in National AN-30) and uses the heater from Rick
	>Jansen's Formant VCF circuit. This heater is just like the one in
another 
	>National Application Note whose number I can't remember, but with a
temp 
	>adjustment pot. I tried to improve the analog divider by following
advice
	>from the Op-Amp Cookbook, etc., i.e. use really low drift op-amps
	>(currently it's a PMI OP-400) and polystyrene capacitors in the
feedback
	>loops.

... my point is that you probably don't need a heater or extra tempco
resistor at all. Build the log circuit of your f/V converter with a
transistor pair (but without tempco resistor or heating), and build
your expo converter the same way. If you're using an existing
synth, you might make the tempco resistor switchable to a
normal resistor. Or even better, find a way to sum a tempco-
corrected path (for keyboard playing) and a not corrected path
(for tracking to an external frequency) together.

The two errors should exactly cancel.

I would have recommended to abandon the whole V/Oct intermediate
step, but V/Oct has its advantages for combining several modulation
sources. You can add instead of multiply.
But then again, a log circuit followed by an expo circuit isn't really
much different from a multiplier ...

So I would use a 3046, two transistors for log, two for expo,
the expo pair would have one transisto's base connected to a
tempco resistor (for footage selection, sequencer input etc.),
and the other one going uncompensated to the log circuit's output.
(This does not apply to the ARP circuit, of course.)

JH.