Hi all,
an article i read the other day reminded me of something that I
wish I had more time to look into - "running FFTs"
The way mp3 encoding is at the moment you have a buffer, you do an
fft on it, you throw out X number of old samples and put X new samples
into the buffer, and then do another FFT.
However, FFT theory states that if you aren't FFT'ing a whole new
buffer, (ie some of the buffer is the same before, just time shifted) then
you don't have to do the whole FFT again, you only have to do a bit of it.
The windowing and filtering of the buffer might prevent this from
being viable in mp3 though.
Check out "running fourier transforms" in November98's Electronics
World for a bit of info. I'm sure a book on DSP will give a much better
explanation of what is involved though.
later
mike
