Hey y'all --
So this is one of those times that my lack of serious math chops is coming round to bite me, and none of my usual books is helping me out. I'm hoping someone has some thoughts.
I'm trying to approximate either exp(-1/(n+1)) or 4^(-1/n+1). I can convince myself I don't care which. n is an integer from 1-65535, and the result should be fixed-point fractional, probably U0.18. The function output is always between 0-1, and goes up like a rocket for small n before leveling off to a steady cruise of >0.9 for the rest of the function domain.
I'm working in an FPGA, so I've got adds, multiplies, and table lookups from tables of reasonable size (10s of kb) cheap, but other things (divides especially) are expensive. I can throw several clock cycles at the problem if need be.
Taylor series attacks seem to fail horribly. I feel like there may be some answer where answers for n in [1,127] gets a direct table lookup, and n in [128,65535] gets some other algorithm, possibly with a table boost. Or somehow taking advantage of the fact that log(1-f(n)) is related to log(n)?
Anyone have any thoughts?