Comments on: use fast fourier transform (fft) to perform low-pass filtering http://desk.stinkpot.org:8080/tricks/index.php/2009/12/use-fast-fourier-transform-fft-to-perform-low-pass-filtering/ tricks! Thu, 02 Feb 2012 20:46:58 +0000 hourly 1 http://wordpress.org/?v=3.1.1 By: sh http://desk.stinkpot.org:8080/tricks/index.php/2009/12/use-fast-fourier-transform-fft-to-perform-low-pass-filtering/comment-page-1/#comment-22593 sh Tue, 16 Aug 2011 10:09:29 +0000 http://desk.stinkpot.org:8080/tricks/?p=313#comment-22593 Could you please explain more about what you wrote and how does it work? I used it and the result is good but i don't know how does it work. Could you please explain more about what you wrote and how does it work?
I used it and the result is good but i don’t know how does it work.

]]> By: Spencer http://desk.stinkpot.org:8080/tricks/index.php/2009/12/use-fast-fourier-transform-fft-to-perform-low-pass-filtering/comment-page-1/#comment-11671 Spencer Mon, 14 Dec 2009 02:05:49 +0000 http://desk.stinkpot.org:8080/tricks/?p=313#comment-11671 A few suggestions - cutting off frequencies like that effectively multiplies by a rectangular window, which will smear a lot of content back in the time domain. It's for this reason that low pass filter design usually multiplies the ideal lpf in the frequency domain (i.e., rect window cutoff at the proper frequency) with a hann window or something similar. Check out this thread http://www.dsprelated.com/showmessage/42364/1.php for some more in depth thoughts on the subject. Also, the "real" would not be necessary if you ensure conjugate symmetry before the ifft(). For even N, keep the first N/2+1 elements and set the last N/2-1 to conj(fliplr(s_fft_data_v(2:N/2))). I think that would be the correct syntax in your case, but you probably want to check the details to make sure. I do hope these ideas are helpful and apologies if I'm pointing out something you're already aware of and intentionally omitting! A few suggestions – cutting off frequencies like that effectively multiplies by a rectangular window, which will smear a lot of content back in the time domain. It’s for this reason that low pass filter design usually multiplies the ideal lpf in the frequency domain (i.e., rect window cutoff at the proper frequency) with a hann window or something similar. Check out this thread http://www.dsprelated.com/showmessage/42364/1.php for some more in depth thoughts on the subject.

Also, the “real” would not be necessary if you ensure conjugate symmetry before the ifft(). For even N, keep the first N/2+1 elements and set the last N/2-1 to conj(fliplr(s_fft_data_v(2:N/2))). I think that would be the correct syntax in your case, but you probably want to check the details to make sure.

I do hope these ideas are helpful and apologies if I’m pointing out something you’re already aware of and intentionally omitting!

]]>