Sun Mar 24 12:37:26 EDT 2013

Small signal analysis

I've added small signal analysis to ai-freq.rkt based at 0.  However,
this could be generalized to any point, by using ai-autodiff.rkt to
construct linear approximations.

The probing part can probably be completely replaced by the
computation of a set of derivatives.  The thing there however is to
distinguish linear from nonlinear outputs.

Let's give it a try.  Roadmap:
- Implement evaluation over offset + signal correctly
- Evaluate over normal numbers to find matrix

Computing the derivative matrix was straightforward, just folling the
same pattern.

Changing the number implementation to small-signal is a bit stranger
though.  Maybe this can now be simplified a bit.