exp(-x) function

Depends on what is the range of parameters. Your statement "fixed point 16bit numbers" doesn't help much, because it doesn't specify any particular encoding -- is it 8 bits for the integral part and 8 bits for fractions or only 16 bits of fraction, which describe a number in the range of (-1,1) or something else?

As I said, it depends.

Best regards Piotr Wyderski

I've posted on this subject before (search google using andraka and exp) here is one such posting that is probably the closest to what you want:

Most likely that's the way to do it. An alternative would be a hyperbolic CORDIC, also explained by Ray:

Kolja Sulimma

If I needed high precision or for some reason could not use reasonably sized look-up tables, I'd consider a hybrid of these two, replacing the look-up with a CORDIC exp(x) after performing the normalize to strip off the integer part. The look up method I described in my prior post will get you plenty of precision for most applications.