The x**2 function was hard for these neurons to represent on a
short time scale, and wasn't necessary for testing dt.
So I've just replaced it with a simple negative function.
The filters on the reference outputs needed to be chosen better,
to ensure the same amount of delay in the signal.
dt. So I've just replaced it with a simple negative function.