Caustic Song file (optional):
It's a simple concept. you give an audio signal a big, hefty DC offset, on purpose. One polarity ends up with more amplitude than the other. When you distort the signal, that polarity clips first. Job done.
The bias can be positive, or negative. It doesn't really matter. I've rigged up both here. Inputs 2 and 3 (marked with song automation) of the mixer in the second row. It's easier to balance modulation, adjusting equal amounts of both.
Input 2, of the second mixer, in the third row, is a depth control.
There's an LFO in the second machine. Input 2, first mixer, third row, controls negative level.
Input 1, second mixer, third row, controls overall depth.
You can send any audio you like through it, but simpler waveforms tend to be a bit more fruitful (the FM Pair is good).
And everything else is based on this machine.