Thu Feb 18 16:43:52 CET 2010

Control Structures or Network Combinators?

Before this can host useful code, some control structures need to be
introduced.  However, it seems better to start on the right foot and
not allow arbitrary loops or recursion and only provide network

Essentially, the basic abstraction is MIMO functions.  The combinators
then should ``duplicate structure'' meaning they should build
directed networks, not state machines.

Is it possible to push feedback (i.e. for IIR filters) outside of a
composite network?  Networks should be pure, so state can be managed
separately, and network composition is simplified.

I.e. the working model is a state-space model, where computation is
pure and state,input,output are explicit.

This seems OK _if_ subsampling is handled appropriately.

-> Hint: take a look at CUDA.