ccooper,

I did not say that we need to calculate the coefficients on the fly- I merely mentioned that we would have to store them somewhere and dynamically select the correct set on a sample-by-sample bosis. I didn't calculate the lowest common factors of 44100 and 96000, so my 10ms snippet was three times as long as necessary, reducing the number of necessary coefficients by a factor of 3 to "only" 6400.

My point about each filter sounding different is still valid. The whole filter bank will roll through with a 3.33ms period (300Hz).

The filter I describe is a polyphase filter.

The filter you describe is a polyphase filter- the giveaway phrase is "In practice, the DSP never even computes the samples we are throwing away."

Thanks for pointing out my factor of three, though.

Live long and prosper,

Simon.


[This message has been edited by Simon the Whale (edited 07-24-2001).]