A couple of observations/questions:

-   When I was using Csound, I would simply use a large enough table
size, so that the error created by using only linear (or none at all)
interpolation, was less then the resolution of the output file type (in
my case, 16 bits). On today's machines, using 64K for a wavetable isn't
a big deal. I suppose the tradeoff is that there the RAM cache will be
overwhelmed on certain processors, hence making interpolation
preferable. 

-  At what point, does Sinc interpolation (which, is the theoretically
correct approach) become more cost effective (in terms of fidelity per
computation time) then polynomial interpolation? Without really
analyzing the tradeoffs, my initial response to a proposal to a proposal
to use say, eight-point poly interpolation, would be that surely a Sinc
interpolation with lesser points would give better results. But perhaps
not, especially given that the implementation of a Sinc interpolation
might need to involve a second table lookup to efficiently compute the
sinc function!

Well, in summary, (a) Why not just use larger table sizes where the
interpolation method isn't an issue, and (b) what are the tradeoff
points with polynomial vs. Sinc interpolations? 

Larry
    

--  Larry Troxler --  lt@westnet.com  --  Patterson, NY USA  --