me wrote: >Steve Harris wrote: > >>On Sun, Nov 02, 2003 at 10:49:17PM +0100, Tim Goetze wrote: >>> the frequency response is mostly due to the oversampling filters; >>> unfortunately i am at a loss for good realtime-compatible design >>> methods for those. it would well be possible to recalculate them >>> (chebyshev prototypes) but that would take quite a few cycles, and >>> knowledge about how to move the ripple/s in a chebyshev design >>> (anyone?). >> >>Have you tried polyphase filters? I've never used them but I've heard they >>are good for oversampling. > >i don't have any precise cycle-consumption figures (a good reason to >try the polyphase filters one day) but i seem to remember having >figured that they come at equal or bigger cost for comparable stopband >attenuation.
i feel the need to clarify/correct myself: the filters employed are not chebyshev designs, they were designed the same way as 'unmatched': frequency response given, iteratively approximate the filter. they do a lot better than chebys of the same order, but it's quite impossible to recalculate their coefficients at runtime. if that is demanded, the polyphase design probably wins easily. tim
