when should one perform fft based filtering?
in [DSP]
|
From: Randy Yates on 24 Jul 2010 21:19 "bharat pathak" <bharat(a)n_o_s_p_a_m.arithos.com> writes: > what are the problems associated with FFT based filtering? > > and when should one decide to switchover from convolution > based time domain filtering to FFT based filtering? It depends on things like how many hardware multipliers you have available on your target platform, but the usual rule-of-thumb is when you get to an impulse response on the order of 100 samples or more. -- Randy Yates % "And all you had to say Digital Signal Labs % was that you were mailto://yates(a)ieee.org % gonna stay." http://www.digitalsignallabs.com % Getting To The Point', *Balance of Power*, ELO
From: Vladimir Vassilevsky on 24 Jul 2010 23:15 Hello Dr. Rune! It is good to see you again. Where have you been? VLV
From: Rune Allnor on 25 Jul 2010 04:31 On 25 Jul, 05:15, Vladimir Vassilevsky <nos...(a)nowhere.com> wrote: > Hello Dr. Rune! > > It is good to see you again. Where have you been? > > VLV Moving house and vacation. Starting in a new job august 1st. It will be a lot of data analysis, but not much DSP. Rune
From: Rune Allnor on 25 Jul 2010 08:39 On 25 Jul, 03:16, Randy Yates <ya...(a)ieee.org> wrote: > Rune Allnor <all...(a)tele.ntnu.no> writes: > > that combine the two approaches, by doing FFT-based filtering on > > sub-frames of a very long sequence. I suspect these methods are > > remnants from the times when one could actually get data sets that > > were larger than the available RAM, > > You mean like 2010? I didn't find anywhere in the OP's original query > that limited the platform to something like a PC. And even then, as you > yourself noted, you don't really want to do it all in one FFT. You edited away the essential statement of mine: > > and so needed > > both time- and memory-efficient methods for offline processing. I don't really see why one would want to do offline work on embedded devices. And I can't see how overlap-add/save would be more efficient on embedded devices, as their capacity for efficient computations would be offset by infklexibilities / awkwardness with respect to flexible progra control structures. Sure, embedded devices might be able to do stuff like FFTs very efficiently, but they would struggle with the control loops necessary to fuse subsequent frames. Again, the only relevance I can see for overlap-add/save these days, is to reduce numerical noise in large *offline* work. Rune
From: robert bristow-johnson on 25 Jul 2010 12:43
On Jul 25, 8:39 am, Rune Allnor <all...(a)tele.ntnu.no> wrote: .... > Sure, embedded devices might be able to do stuff like FFTs > very efficiently, but they would struggle with the control > loops necessary to fuse subsequent frames. bookkeeping. > Again, the only relevance I can see for overlap-add/save these > days, is to reduce numerical noise in large *offline* work. how about a real-time convolution reverberator based on an measured impulse response of an actual large room (like a venerated auditorium or cathedral)? FIR with a couple hundred thousand taps? r b-j |