Wed May 15 22:05:49 EDT 2013

Delay lines

Should be simple 1-1 ops.  The lenght needs to be available to the
z-transform evaluator!  If they share an input, they can be optimized
into tapped delay lines.

The thing is: single-sample delays all come from the state feedback
operator.  Essentially, other than unit delay should be implemented
the same.

Something isn't right: two concepts are mixed.

A simple z(n) operator won't work, since this can't do output

Maybe the current approach is really the best way to go: a delay line
is a matrix signal.