Electronic – Issue with multichannel LM317 circuit

The point of the current limiter is that, if the current is higher than desired, it will reduce voltage, which with most Ohmic loads will reduce the current draw.

This uses the fact that voltage drop == resistance times current, and the regulator reduces voltage if it sees a voltage on the reference pin that is greater than the reference (1.2V? something like that)

Does this mean that the voltage out from the current limiter is at least 1.2V less than the input? Yes, it does!

However, if you have a microcontroller, and don't need "immediate" feedback (your uC response time is OK) then you can simply sample the voltage drop across a current sense resistor (something small like 0.1 Ohms) to estimate current, and drive any variable resistor, like a regular power transistor, using an analog output from your uC (or a filtered PWM.)

You probably want another solution, like an existing programmable potentiometer in the feedback loop of an opamp, for example. Use a current sense resistor that's small (0.1 Ohms?) and program the amplification of the voltage drop using an opamp. When amplification is 10x, then you get to 1.2 volts at 1.2 amps; when the amplification is 100x, then you get to 1.2 volts at 120 mA.

The LM317 is a linear regulator, which is really inappropriate for significant power, like what a model train requires. You say the train draws 300 mA. That means at 3.3 V drop accross the regulator it would dissipate 1 W, which would require a heatsink.

I'm not sure why your regulator is dying. I thought (haven't checked) that the LM317 has integrated thermal shutdown, so it should shut itself off if overheated. Perhaps it doesn't. Check the datasheet.

In any case, a switching power supply is the right answer. I am actually working on a product right now for the model train market that includes a track power supply. I am starting with 48 V. That gets bucked down to whatever the user selects as the track power supply voltage, then a separate H bridge circuit implements the polarity flipping for the DCC protocol. I can dump a few Amps onto the track and the board gets warm, but not hot. Once this becomes a product, the schematic will be made public. If you really want to see what I have so far, I can make it public unofficially, but I don't like showing internal designs until everything has been worked out and properly tested.

On a separate note, why not use DCC? That completely gets around the problem of "sending one of the trains in reverse due to the common rail".