On Oct 21, 9:39 am, "baeksan" <breakcha...@yahoo.com> wrote:
> ...
> I will try out 4 stages and see if it improves.
> ...

I'd suggest 8:7; 4:3; 5:7 to keep the maximum intermediate sampling
frequency down.

> ...
> Theory vs actual implementation seems to be
> a whole different ball game.
> ...

You mention cpu time and c code. If you are measuring times on a
desktop computer, you may need to try to pick data structure sizes to
allow the entire chain to fit in cache. Big data structures that force
write back to main memory after each stage and reading to cache going
into each stage can slow the processing considerably.

Dale B. Dalrymple
http://dbdimages.com