Sat Mar 9 19:22:54 EST 2013

Exponentials: pole interpolation: linear vs. exp(linear) ?

Since all values are going to be quite small, it might be good enough
to just use linear interpolation for the pole positions instead of
multiplicative updates.

The thing is: pole Q is not going to change much.  All control effort
will be for pole frequency which is most musically interesting.

When using linear interpolation for cos/sin also, the curvature will
start to play a role when frequency difference is large.  This
introduces modulation in the Q, and reduces the average Q.

A trade-off is probably possible:
- linear interpolation for pole Qs
- circular interpolation for frequencies + normalization at border

r += dr

p_x += (c-1) p_x + s p_y
p_y += (c-1) p_y - s p_x