Transistor as rheostat?

So the recent thread here http://forum.arduino.cc/index.php?topic=191309.0 got me to thinking about using transistors in the linear region as rheostats. Obviously, this is not a straightforward proposition or else transistor-based digipots would be cheaper to manufacture than the resistor-network digipots commonly made today. But Runaway Pancake in http://forum.arduino.cc/index.php?topic=118107.25 seems to hint that it could be possible, at least for some values of "possible".

I'm guessing that the transistor-as-rheostat has intractable problems, perhaps such as varying in resistance as the collector-emitter voltage and/or current changes?

Well, rheostat is just a two terminal potentiometer, the wiper and 1 leg. http://en.wikipedia.org/wiki/Potentiometer

So you're intentionally varying the resistance. If have a fixed voltage and say Gnd connected to it, the current will vary as the wiper position changes. Similarly, if you have a fixed current going thru it, the voltage across it will vary as the wiper position changes. What did you have in mind to do with this rheostat?

CrossRoads:
What did you have in mind to do with this rheostat?

Nothing specific – I’m mainly just curious why digipots are made from resistor networks (and a bunch of switching transistors) instead of a couple of linear transistors.

EDIT: Found more discussion of transistor-as-rheostat here: http://forum.arduino.cc/index.php?topic=190688.0 , still nothing conclusive except “there are better ways”.

Transistors are not linear devices… BJTs are exponential in base-emitter voltage and
FETs have current depending on the square of the gate-source voltage (to a first
approximation).

The nearest you’ll find is balanced networks used to make analog multiplier chips,
used mainly for modulation and demodulation - or remote controlled volume controls.
Also transconductance opamps work along the same lines…

May i ask what it is you’re going to be powering?..

cjdelphi: May i ask what it is you're going to be powering?..

Nothing in particular, mainly just curiosity. But, I can think of one potential application -- changing the power of an IR emitter. PWM cannot be used directly, because you're already using a 50% duty cycle 38kHz PWM signal to get past the receiver's DC filter.

You could ground the LED through one PWM pin with an RC low-pass filter to set the output current, and power the LED through your 38kHz PWM pin, but you rapidly hit the 20-40mA pin limit (many IR LEDs can be pulsed at 1A). So you're going to need a transistor anyway -- why not put the RC filter on the transistor base, which would require much smaller values for the cap and resistor?

Yes, the transistor will heat, but, so would your rheostat!

MarkT: Transistors are not linear devices.... BJTs are exponential

Hm, that's a bit inconvenient. While software can compensate to some degree, you are still going to lose a lot of resolution at one end of the range.