Sat Mar 9 17:26:12 EST 2013

Exponentials: push drift compensation upward

Some experiments show that for the values involved, an order 5-10
Taylor series is going to be quite good and approximation.  The
remaining problem then is drift.

These functions are to be used in LFOs and envelopes in a sound synth,
so there might be some trade-offs to be exploited, i.e. for an
envelope, the end state is 0 increment, which should remain stable.
Maybe it's even possible to compute where we'll end up after one
cycle, to compensate for that externally?

So, given ^exp(), approximation of exp(), a set-point curve at block
boundaries (N), and an initial state, it is possible to compute the
expected deviation after one "cycle" of the control wave.  This can
probably be used externally to compensate the control wave form on a
larger time scale.