Circuit operation
 
The circuit is a differential transistor amplifier in the feedback loop
of an opamp. A diff. amp. shows the same temperature dependancy
like the exponential converter (-3300ppm/0C). The opamp works as a 
divider, with a fixed divsion, therefore the temperature 
coefficient turns into its reciprocal. This effectively cancels the 
temperature drift in the overall circuit. For best results the transistors
of the exp. converter and of the diff. amp have to be on the same temperature.
This is acomplished by using a CA3046/86 array, ensuring low temperature gradients.
Care has been taken, to find a suitable compromise between noise (leads to jitter)
and nonlinearity (v/oct precision). Maybe a solution with MAT0x transistors yields an
even better SNR.
 
However in conjunction with the expo. converter of my VCO, the nonlinearity appears
only towards the low-frequency end, causing deviations of the 1v/oct standard only for 
frequencies below the audio range. I find this is a suitable compromise.
If you disagree then you are free to tweak the values of R8,R3 and R9 to your needs.
Anyway the gain of the differential amplifier *must* be less than unity, since otherwise
the circuit may oscillate.
To obtain the pairs of matched resistors simply buy a few 5% resistors and match them best
with a DMM.
 
This circuit has several advantages over the heating method employed by several people
(refer to AN299 of National, or Rick Jansens VCO), the heater approach eats a lot of 
current (about 70mA), and still you have temperature gradients, i.e. the temperature
falls with increasing distance of the heater transistor, so this doesn't ensure that 
all transistors are on the same temperature. With this method, this is eliminated largely
by the fact that the chip does not get hot! The same applies to tempco resistors, you have 
to couple the tempco to the expo. pair, although I must say that noise and linearity are 
better with tempcos.