tjuii wrote:> So let's see if I am getting this. There is information between the bins > that I can't see in the freq. domain and that doesn't become visible until > an IFFT to go back into the time domain.More specifically, IFFT back to the time domain, zero extend the time domain result by a few factors of two and FFT it again. You will then see samples of the function that is between the original samples. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
Frequency domain filtering (rectangular window question)
Started by ●September 21, 2005
Reply by ●September 26, 20052005-09-26
Reply by ●September 26, 20052005-09-26
Bob Cain wrote: ..> "Bin" is a terrible misnomer for frequency domain samples which > continues to persist for some reason. >.. freqles? Richard Dobson
Reply by ●September 26, 20052005-09-26
Bob Cain wrote: ...> "Bin" is a terrible misnomer for frequency domain samples which > continues to persist for some reason.I agree, but I use it for lack of a better concise term. "Sample of the Underlying Continuous Spectrum" could give us SUCS, but people would construe it as plural. Besides, that's not the whole story except in filters. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●September 30, 20052005-09-30
>If you are specifying your filter in the frequency domain >what are you doing about the phase portion of that >specification? > > >Bob >--I am not doing anything about the phase. The filter function of zeros and ones just leaves the phase of the remaining frequencies unchanged when multiply by the FFT of the signal, right (i.e. both real and imaginary parts of the FFT are the same before and after the multiplication, for the remaining freqs.)? Should I be doing something different with the phase? What I am doing seems to be working well for my purposes, but I'd love to know if there is something else that I need to think about. Thanks for all your help, everyone. This message was sent using the Comp.DSP web interface on www.DSPRelated.com
Reply by ●September 30, 20052005-09-30
tjuii wrote:>>If you are specifying your filter in the frequency domain >>what are you doing about the phase portion of that >>specification? >> >> >>Bob >>-- > > > > I am not doing anything about the phase. The filter function of zeros and > ones just leaves the phase of the remaining frequencies unchanged when > multiply by the FFT of the signal, right (i.e. both real and imaginary > parts of the FFT are the same before and after the multiplication, for the > remaining freqs.)? Should I be doing something different with the phase? > What I am doing seems to be working well for my purposes, but I'd love to > know if there is something else that I need to think about.You aren't getting that "infinitely fast roll-off and a perfectly flat pass-band", but if it's working well for your purposes, keep it up. Why did you ask your original question? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●September 30, 20052005-09-30
Jerry Avins wrote:> tjuii wrote: > >>If you are specifying your filter in the frequency domain > >>what are you doing about the phase portion of that > >>specification?...> > I am not doing anything about the phase. The filter function of zeros and > > ones just leaves the phase of the remaining frequencies unchanged when > > multiply by the FFT of the signal, right (i.e. both real and imaginary > > parts of the FFT are the same before and after the multiplication, for the > > remaining freqs.)? Should I be doing something different with the phase? > > What I am doing seems to be working well for my purposes, but I'd love to > > know if there is something else that I need to think about. > > You aren't getting that "infinitely fast roll-off and a perfectly flat > pass-band", ...Is there any rule-of-thumb on what the phase response needs to look like near any artificially sharp transition in a frequency domain filter in order to reduce ripple (caused by the sharp transition) in the continuous frequency response, if that's at all possible? Thanks. -- rhn A.T nicholson D.o.T c-O-m
Reply by ●September 30, 20052005-09-30
>Why did you ask your original question? > >Jerry >-- >Engineering is the art of making what you want from things you can get. >����������������������������������������������������������������������� >Because I want to know more than I do right now. Plus, I didn't understand what my colleague was talking about when he said that the method I was using wasn't a good one. I still think he was thinking that I was using a rectangular window in the time domain. THAT I do know is not a very effective thing to do. Thanks all for the discussion/explanations. TJ Oh, almost forgot. I asked this question buried in one of my replies above, but forgot about. Is there a rule-of-thumb about how fast to sample in comparison to the frequencies of interest (aside from the Nyquist issue, i.e. sample at twice the frequency of the frequency of interest)? This message was sent using the Comp.DSP web interface on www.DSPRelated.com
Reply by ●September 30, 20052005-09-30
rhnlogic@yahoo.com wrote:> Jerry Avins wrote: > >>tjuii wrote: >> >>>>If you are specifying your filter in the frequency domain >>>>what are you doing about the phase portion of that >>>>specification? > > ... > >>>I am not doing anything about the phase. The filter function of zeros and >>>ones just leaves the phase of the remaining frequencies unchanged when >>>multiply by the FFT of the signal, right (i.e. both real and imaginary >>>parts of the FFT are the same before and after the multiplication, for the >>>remaining freqs.)? Should I be doing something different with the phase? >>>What I am doing seems to be working well for my purposes, but I'd love to >>>know if there is something else that I need to think about. >> >>You aren't getting that "infinitely fast roll-off and a perfectly flat >>pass-band", ... > > > Is there any rule-of-thumb on what the phase response needs to look > like near any artificially sharp transition in a frequency domain > filter in order to reduce ripple (caused by the sharp transition) > in the continuous frequency response, if that's at all possible?We both know that linear-phase filters can be made arbitrarily sharp, so I don't understand what you're asking for. jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●September 30, 20052005-09-30
tjuii wrote:>>Why did you ask your original question? >> >>Jerry >>-- >>Engineering is the art of making what you want from things you can get. >>����������������������������������������������������������������������� >> > > > Because I want to know more than I do right now. Plus, I didn't > understand what my colleague was talking about when he said that the > method I was using wasn't a good one. I still think he was thinking that > I was using a rectangular window in the time domain. THAT I do know is > not a very effective thing to do. > > Thanks all for the discussion/explanations. > > TJ > > Oh, almost forgot. I asked this question buried in one of my replies > above, but forgot about. Is there a rule-of-thumb about how fast to > sample in comparison to the frequencies of interest (aside from the > Nyquist issue, i.e. sample at twice the frequency of the frequency of > interest)?The sampled signal needs to n]be filtered to meet the Nyquist criterion, The sample rate needs to accommodate the nature if the imperfect filter. 25% oversampling is marginal. For servo systems that need low lag, 5x oversampling is good and more is better. Context governs. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●September 30, 20052005-09-30
Jerry Avins wrote:> rhnlogic@yahoo.com wrote: > > Is there any rule-of-thumb on what the phase response needs to look > > like near any artificially sharp transition in a frequency domain > > filter in order to reduce ripple (caused by the sharp transition) > > in the continuous frequency response, if that's at all possible? > > We both know that linear-phase filters can be made arbitrarily sharp, > so I don't understand what you're asking for.I was refering not to arbitrarily sharp filters, but to artificially sharp filters, such as the one the OP was using which quantizes the magnitude response to either 1.0 or zero (obvious not a flat filter). My question was about whether allowing degrees of freedom in the phase response would help anything (assuming more than 1 bin of 1.0 coefficients, or course). e.g. would his box filter get any "better" if he removed all linear phase restrictions? Thanks. -- rhn A.T nicholson d.O.t C-o-M
