I know I'm missing something basic ;) Having begun with a very long signal ( m samples ) Done a fft on a series of windows n samples long ( m >>> n ) optionally performed some operation in the frequency domain done an inverse fft on each of the modified windows How do you "glue" things back together in the time domain? Thank you.
Inverse FFT or "Unscrambling the egg"
Started by ●November 10, 2003
Reply by ●November 11, 20032003-11-11
On Mon, 10 Nov 2003 20:20:39 -0600, Richard Owlett <rowlett@atlascomm.net> wrote:>I know I'm missing something basic ;) >Having begun with a very long signal ( m samples ) >Done > a fft on a series of windows n samples long ( m >>> n ) > optionally performed some operation in the frequency domain > done an inverse fft on each of the modified windows > >How do you "glue" things back together in the time domain? > >Thank you. >Richard, search the web for "overlap and save" and "overlap and add". Good luck, [-Rick-]
Reply by ●November 11, 20032003-11-11
Rick Lyons wrote:> On Mon, 10 Nov 2003 20:20:39 -0600, Richard Owlett > <rowlett@atlascomm.net> wrote: > > >>I know I'm missing something basic ;) >>Having begun with a very long signal ( m samples ) >>Done >> a fft on a series of windows n samples long ( m >>> n ) >> optionally performed some operation in the frequency domain >> done an inverse fft on each of the modified windows >> >>How do you "glue" things back together in the time domain? >> >>Thank you. >> > > > Richard, > > search the web for "overlap and save" > and "overlap and add". > > Good luck, > [-Rick-] >Dang it all, knew it must be obvious. When is a sequence infinite? When it's *BIGGER* than you are ;\ Now let's see if I can do it in Scilab. P.S. References state and/or imply that ideal ratio of input points: output points is 2:1 . Is that based on computational efficiency or other factors. I.E. would a transform on window 10x what used in output yield any additional "fidelity"? "fidelity" used VERY loosely as I intend to multiply frequency domain components by either 1 or 0 in probably octave chunks. Yep, phase relationships destroyed in unspecified manner. If an implied assumption of my underlying assumptions is valid, the point is moot. Thank you all for patience.