I am working on a precision tanh function generator to compress a wide range signal before digitization to increase resolution for signals near center and sacrifice resolution for large swings. It is based on the LM13700 dual transconductance amplifier. I am trying to reduce the non ideal aspects as much as possible and have come up with the following scheme:
I control the current into the Iabc pins using an LM385 1.25 volt as reference for the current regulator.
I control the chip temperature by controlling the voltage at the Iabc input pins (which runs around 1.3 volts at room temperature and is effectively two diode drops) using the 1.25 volt as setpoint and using the two darlingtons on the chip as heaters.
I drive the two sections inversely with the input voltage and subtract the output currents from each other (keeping the output swings very near center of the supply, to reduce Early effects on the current mirrors) so that any difference between the positive and negative current mirrors partially cancels for any signal polarity.
I haven't yet collected test data to see how closely I have achieved the tanh function, but my first impression of the operation makes me think of tube amplifiers. There is no sharp clipping for significant overdrives, so I suspect it would sound very much like an open loop tube amplifier. I can see lots of ways to combine two transconductance amplifiers (with a gain and full scale knob on each) to achieve a wide range of distortion shapes versus overload, so many different tube overload effects should be producible.