Mark wrote: ...> I agree a listening test would be interesting. > > But I'll also muse for a bit. > > My thought experiment is to think about the transient (impulse > response) of 5 kHz low pass filters fed with impulses. > > I classify the filters as min phase vs linear phase and also as sharp > cutoff vs gradual cutoff. > > SHARP CUTOFF FILTERS > If the low pass filter has a sharp cutoff, it WILL ring due to Gibbs.Gibbs phenomenon is a term reserved to describe the ringing of a Fourier series at discontinuities. Non-monotonic amplitude response of filters is usually not called Gibbs, but I know what you mean.> A sharp cutoff min phase will post ring. A sharp-cutoff linear phase > will pre and post ring. Both sound bad.Why do you say it sounds bad when you feed the filter with impulses? What effect is it that makes you perceive it as a bad sound?> GRADUAL CUTOFF FILTERS > What about gradual cutoff filters? If we now assume a gradualcutoff> without Gibbs, which is better, min or linear phase? Standardtheory> is that the ear is not sensitive to phase information in the contextof> the relative phase of one frequency to another. I have done this > listening test and I believe it. You can take a 1.00 kHz tone andadd> a 2.01 kHz tone and watch the phases drift through making a radical > changes to the scope waveform but you don't hear any change.I know these tests. If you Hilbert transform a square wave signal, you get a periodic signal that looks quite odd, but still sounds the same as the square wave. This test shows that we cannot detect phase difference between stationary sine waves -- that's another thing than detecting differing phase responses of equal magnitude filters.> So my hypothisis is.... > > Sharp cutoff filters will always create echoes. Min phase vs linear > phase will only make a difference in the nature of the echoes. Min > phase having all post echoes and linear phase have pre and postechoes. I'm not sure about those echoes --- I've listened to lots of lowpassed material in my life, and none seemed to have been encumbered with an audible echo. Any FIR (and for that matter IIR as well) filter (gradual or steep lowpass, bandpass, allpass, etc.) is just a weighted sum of the signal with time delayed versions of itself. All of these filters introduce "echo", but we don't perceive the filtering effect as such.> To make a "good sounding" filter without echoes, you need to use a > gradual cutoff and it doesn't matter if it is min phase or linear > phase. The more gradual the frequency response change, the less > temporal distortion.Mark, I think I can guess at what you are aiming at, but your terminology is so diffuse that I find it hard to discuss it with you. What is "good sounding", and "temporal distortion"? A filter is a linear device, as such it does not introduce distortion - in my world, distortion is introduced by dynamic (non-linear) processing.> CommentsYou might want to read this article, I think it sort of describes (in great detail) what you mean (if not, please correct me): Michael Gerzon: "Why do equalisers sound different?" Studio Sound, July 1990, available at: http://www.audiosignal.co.uk/Gerzon%20archive.html Regards, Andor
ringing: minimum vs linear phase
Started by ●March 27, 2005
Reply by ●April 1, 20052005-04-01
Reply by ●April 1, 20052005-04-01
"Jerry Avins" <jya@ieee.org> wrote in message news:s6SdnS8mz6pGq9bfRVn-hA@rcn.net...>> >> Fred > > One can always adjust the delays, even if only by displacing the speakers. > To keep focus, look at a single speaker. The minimum-phase filter may > introduce ringing after a steep edge. We're accustomed to that; drums and > gongs behave that way. A little more of the same is so tolerable it's not > usually noticed. Linear-phase filters cause ringing both before and after > the sharp transition. When the program material telegraphs its punches, it > sounds very bad. That's the wrong way to make a tone control. > > Thunder is the sound of a very big spark. The rumble afterward is often a > bunch of echoes. What causes the rumble before? Would you want a bass drum > to sound like that? Let's coin a word: preverberation. > > JerryJerry, A very good and clear answer! After some thought, I figured "duh!" we are talking about transient responses here - which of course is obvious. It's only in the *steady state* that waveform phase distortion occurs - such as with a not-linear phase filter. So, one has to ask: "am I more interested in transient response or steady state response?" when asking questions like this regarding linear phase and minimum phase filters.... Fred
Reply by ●April 1, 20052005-04-01
Andor, good article, thanks... it pretty much agrees with what I was saying, "bad sounding" filters have an impulse response with pre- and post echoes.. (he calls them pre and post responses). The article does make one more interesting point, even a gradual cutoff (he calls it low Q filter) can also have pre-echoes if it has many small ripples in the passband. Mark
Reply by ●April 2, 20052005-04-02
Andor wrote:> Bob Cain wrote: > >>Philosophy? I don't understand your relegation of the >>question to philosophy when there are clear physical >>implications about when within a sampled sequence things are >>heard after filtering. > > > It is philosphizing (musing) like Aristoteles practiced it: think hard > about a physical problem and then come up with a result based solely on > "logical" deduction.You're wrong. First I saw it on the audio waveform display of some binaural recordings I'd done and equalized with linear phase filters to attempt to preserve timing info as I thought linear phase would do. What I saw prompted me to do some listening tests of the same filters linear and minimum phase to see if the difference audible. I eventually did do randomizing of blocks within a filtered piece, some filtered one way and some the other, and although the effect is subtle I was able to reliably tell which was which. This is not an acceptable protocol for publication but it satisfied me that there was an identifiable effect. I found that I preferred the minimum phase because it sounded crisper without added high frequency content. It was empirical observation and experimentation, not at all what you are describing. I no longer think that the effect of linear phase is that of preserving timing relationships such as we have long thought it was.> So howcome you are so convinced that linear phase eq "smears" bass drum > hits?That's words based on what I saw to describe an effect I subsequently heard. They may be inadequate but they are all I have and they sit right with what I heard.> I believe that this is what your eyes tell you to hear.That possibility is exactly why I suggested DBT. Why are you being antagonistic? Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
Reply by ●April 2, 20052005-04-02
Bob, I want to (mostly) amplify what you wrote. Bob Cain wrote:> > > Andor wrote: > >> Bob Cain wrote: >> >>> Philosophy? I don't understand your relegation of the >>> question to philosophy when there are clear physical >>> implications about when within a sampled sequence things are >>> heard after filtering. >> >> >> >> It is philosphizing (musing) like Aristoteles practiced it: think hard >> about a physical problem and then come up with a result based solely on >> "logical" deduction. > > > You're wrong. First I saw it on the audio waveform display of some > binaural recordings I'd done and equalized with linear phase filters to > attempt to preserve timing info as I thought linear phase would do. > What I saw prompted me to do some listening tests of the same filters > linear and minimum phase to see if the difference audible. I eventually > did do randomizing of blocks within a filtered piece, some filtered one > way and some the other, and although the effect is subtle I was able to > reliably tell which was which. This is not an acceptable protocol for > publication but it satisfied me that there was an identifiable effect. > I found that I preferred the minimum phase because it sounded crisper > without added high frequency content. It was empirical observation and > experimentation, not at all what you are describing.I hear it too. A mathematically-minded colleague built himself a linear-phase preamp -- bass, center and treble controls -- as well as equalizer for his magnetic phono cartridge. He was disappointed in the sound and invited several of us to listen and share ideas. His thought was some flaw in the ADC-DAC chain which used the same modules (yes, modules, not chips!) that I used for instrumentation. I was able to quickly point out that the records were made with minimum-phase preemphasis and roll-off, and that restoring the amplitude with a linear-phase equalizer actually distorted the phase. Running the phono cartridge into a an analog pre-amp and that into a line input quickly showed the correctness of that idea, but extreme positions of the tone controls still sounded unnatural, Boost was worse, presumably because the offending effect was louder. We couldn't at the time assign a reason except to say vaguely. "Analog seems to sound better." Eventually, someone suggested redoing the tone-control code with IIRs, and the "analog is better" consensus went away.> I no longer think that the effect of linear phase is that of preserving > timing relationships such as we have long thought it was.Oh? The delays at all frequencies are matched. What do you mean by timing? ... Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 2, 20052005-04-02
"Jerry Avins" <jya@ieee.org> wrote in message news:2sqdnd60nNzVR9PfRVn-rg@rcn.net...> Bob, > > I want to (mostly) amplify what you wrote. > > Bob Cain wrote: >> >> >> Andor wrote: >> >>> Bob Cain wrote: >>> >>>> Philosophy? I don't understand your relegation of the >>>> question to philosophy when there are clear physical >>>> implications about when within a sampled sequence things are >>>> heard after filtering. >>> >>> >>> >>> It is philosphizing (musing) like Aristoteles practiced it: think hard >>> about a physical problem and then come up with a result based solely on >>> "logical" deduction. >> >> >> You're wrong. First I saw it on the audio waveform display of some >> binaural recordings I'd done and equalized with linear phase filters to >> attempt to preserve timing info as I thought linear phase would do. What >> I saw prompted me to do some listening tests of the same filters linear >> and minimum phase to see if the difference audible. I eventually did do >> randomizing of blocks within a filtered piece, some filtered one way and >> some the other, and although the effect is subtle I was able to reliably >> tell which was which. This is not an acceptable protocol for publication >> but it satisfied me that there was an identifiable effect. I found that >> I preferred the minimum phase because it sounded crisper without added >> high frequency content. It was empirical observation and >> experimentation, not at all what you are describing. > > I hear it too. A mathematically-minded colleague built himself a > linear-phase preamp -- bass, center and treble controls -- as well as > equalizer for his magnetic phono cartridge. He was disappointed in the > sound and invited several of us to listen and share ideas. His thought was > some flaw in the ADC-DAC chain which used the same modules (yes, modules, > not chips!) that I used for instrumentation. > > I was able to quickly point out that the records were made with > minimum-phase preemphasis and roll-off, and that restoring the amplitude > with a linear-phase equalizer actually distorted the phase. Running the > phono cartridge into a an analog pre-amp and that into a line input > quickly showed the correctness of that idea, but extreme positions of the > tone controls still sounded unnatural, Boost was worse, presumably because > the offending effect was louder. We couldn't at the time assign a reason > except to say vaguely. "Analog seems to sound better." Eventually, someone > suggested redoing the tone-control code with IIRs, and the "analog is > better" consensus went away. > >> I no longer think that the effect of linear phase is that of preserving >> timing relationships such as we have long thought it was. > > Oh? The delays at all frequencies are matched. What do you mean by timing? >Jerry, I posted a response but in another slot in this thread. I think the real issue here is that we're using mixed metaphors. We talk about phase and we talk about delay in the same breath as we talk about pretransients, etc. So, here is a try at unifying the discussion: The usual definition of phase only applies *after* the transient effects are gone. That doesn't mean that we can't measure phase after a relatively short time, but it does mean that we can't measure it in too short a time. Linear phase *does* preseve timing relationships - but only after the transient. I don't know, might it be said that minimum phase preserves *transient timing relationships* better? It certainly preserves *something* better but what should that be called? Maybe one good model would be a simple delay. Here everything is preserved, transient, amplitude and phase. And, a simple delay is a linear phase device of the simplest type. A very high fidelity recording fits into this category - with the delay being arbitrary and very long indeed! And, except that the delay is arbitrary, we can't measure phase relationships until some time after the playback is initiated - particularly at the lowest frequencies. We can contrast this with a linear phase lowpass filter where there is a nonzero pretransient. I learned something in this discussion because I had not understood how minimum phase filters affect the psychoacoustics - as compared to linear phase filters. Much the same as the example given. So, to apply what I've learned here, this is what might be said(?): Minimum phase filtering is preferable in audio systems or systems where latency and pretransients are important because minimum phase filters minimize pretransients (at the cost of changing the phase relationships of individual spectral components). The latter is apparently (much?) less noticeable than the former. Somehow I recall someone saying that the ear is not phase sensitive. So, while a heard waveform might have amplitude or envelope differences because of phase distortion, does the ear perceive it? I'm sure there are pathological cases where it does. However, for general sounds, is this a major factor? The suggestion was that it isn't. So, maybe that's all there is to it. It's probably important to say something about "minimum phase". Minimum phase means minimum delay at each frequency doesn't it? Whereas, linear phase means flat delay and we know that the delay is longer than the minimum at each (or most) frequency(ies). So, when there is a wideband transient we want all of the frequency components to be delayed the least in order minimize the delay of the transient. To one degree or another, the transient will be distorted in amplitude but what does that mean? If the transient includes a long precursor or pretransient then that could be viewed as an amplitude distortion also couldn't it? It's just of a different type. Fred
Reply by ●April 2, 20052005-04-02
Fred Marshall wrote:> "Jerry Avins" <jya@ieee.org> wrote in message > news:2sqdnd60nNzVR9PfRVn-rg@rcn.net... > >>Bob, >> >>I want to (mostly) amplify what you wrote. >> >>Bob Cain wrote: >> >>> >>>Andor wrote: >>> >>> >>>>Bob Cain wrote: >>>> >>>> >>>>>Philosophy? I don't understand your relegation of the >>>>>question to philosophy when there are clear physical >>>>>implications about when within a sampled sequence things are >>>>>heard after filtering. >>>> >>>> >>>> >>>>It is philosphizing (musing) like Aristoteles practiced it: think hard >>>>about a physical problem and then come up with a result based solely on >>>>"logical" deduction. >>> >>> >>>You're wrong. First I saw it on the audio waveform display of some >>>binaural recordings I'd done and equalized with linear phase filters to >>>attempt to preserve timing info as I thought linear phase would do. What >>>I saw prompted me to do some listening tests of the same filters linear >>>and minimum phase to see if the difference audible. I eventually did do >>>randomizing of blocks within a filtered piece, some filtered one way and >>>some the other, and although the effect is subtle I was able to reliably >>>tell which was which. This is not an acceptable protocol for publication >>>but it satisfied me that there was an identifiable effect. I found that >>>I preferred the minimum phase because it sounded crisper without added >>>high frequency content. It was empirical observation and >>>experimentation, not at all what you are describing. >> >>I hear it too. A mathematically-minded colleague built himself a >>linear-phase preamp -- bass, center and treble controls -- as well as >>equalizer for his magnetic phono cartridge. He was disappointed in the >>sound and invited several of us to listen and share ideas. His thought was >>some flaw in the ADC-DAC chain which used the same modules (yes, modules, >>not chips!) that I used for instrumentation. >> >>I was able to quickly point out that the records were made with >>minimum-phase preemphasis and roll-off, and that restoring the amplitude >>with a linear-phase equalizer actually distorted the phase. Running the >>phono cartridge into a an analog pre-amp and that into a line input >>quickly showed the correctness of that idea, but extreme positions of the >>tone controls still sounded unnatural, Boost was worse, presumably because >>the offending effect was louder. We couldn't at the time assign a reason >>except to say vaguely. "Analog seems to sound better." Eventually, someone >>suggested redoing the tone-control code with IIRs, and the "analog is >>better" consensus went away. >> >> >>>I no longer think that the effect of linear phase is that of preserving >>>timing relationships such as we have long thought it was. >> >>Oh? The delays at all frequencies are matched. What do you mean by timing? >> > > > > Jerry, > > I posted a response but in another slot in this thread. I think the real > issue here is that we're using mixed metaphors. We talk about phase and we > talk about delay in the same breath as we talk about pretransients, etc. > So, here is a try at unifying the discussion: > > The usual definition of phase only applies *after* the transient effects are > gone. That doesn't mean that we can't measure phase after a relatively > short time, but it does mean that we can't measure it in too short a time. > > Linear phase *does* preseve timing relationships - but only after the > transient. I don't know, might it be said that minimum phase preserves > *transient timing relationships* better? It certainly preserves *something* > better but what should that be called? > > Maybe one good model would be a simple delay. Here everything is preserved, > transient, amplitude and phase. And, a simple delay is a linear phase > device of the simplest type. A very high fidelity recording fits into this > category - with the delay being arbitrary and very long indeed! And, except > that the delay is arbitrary, we can't measure phase relationships until some > time after the playback is initiated - particularly at the lowest > frequencies. > We can contrast this with a linear phase lowpass filter where there is a > nonzero pretransient. > > I learned something in this discussion because I had not understood how > minimum phase filters affect the psychoacoustics - as compared to linear > phase filters. Much the same as the example given. So, to apply what I've > learned here, this is what might be said(?): > > Minimum phase filtering is preferable in audio systems or systems where > latency and pretransients are important because minimum phase filters > minimize pretransients (at the cost of changing the phase relationships of > individual spectral components). The latter is apparently (much?) less > noticeable than the former. Somehow I recall someone saying that the ear is > not phase sensitive. So, while a heard waveform might have amplitude or > envelope differences because of phase distortion, does the ear perceive it? > I'm sure there are pathological cases where it does. However, for general > sounds, is this a major factor? The suggestion was that it isn't. So, > maybe that's all there is to it. > > It's probably important to say something about "minimum phase". Minimum > phase means minimum delay at each frequency doesn't it? Whereas, linear > phase means flat delay and we know that the delay is longer than the minimum > at each (or most) frequency(ies). So, when there is a wideband transient we > want all of the frequency components to be delayed the least in order > minimize the delay of the transient. To one degree or another, the > transient will be distorted in amplitude but what does that mean? > If the transient includes a long precursor or pretransient then that could > be viewed as an amplitude distortion also couldn't it? It's just of a > different type.Fred, Thanks? for starting me thinking. (The question mark indicates that maybe we're thinking too much.) When listening is the criterion, we usually want the best fidelity we can get -- low distortion and flat response both in frequency and phase. There are exceptions. Some may want to boost the bass to emphasize the oom-pah-pah quality that they like. Others, guitarists, say, may want to generate distortion. Let's ignore the use of filters and nonlinearities to generate special effects. The criterion of "naturalness" doesn't apply to those cases anyhow. So why use filters at all? Basically, as equalizers. When the original frequency distortion is linear phase (absorption of highs in a dead room?), use a minimum-phase equalizer. When the original frequency distortion is minimum phase (most physical networks; phono preemphasis), use a minimum-phase equalizer. I doubt that one can get better sound than by those rules in those simple cases. Crossover filters aren't as simple as that. Their results depend strongly on speaker placement, so they can't be treated in isolation. Crossover filters are rarely sharp enough to generate much preverb (OK, so it's not my coinage) and are rarely set so low that preverb persists for long. Nevertheless, I don't think one does much better in general than Linkwitz-Reilly, and that happens to be minimum phase. Whatever, I think that is needs to be decided by listening case by case, that it often doesn't matter, and that advertised use of linear phase is very often hype for the ignorant. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 2, 20052005-04-02
Fred Marshall wrote: (big snip)> Minimum phase filtering is preferable in audio systems or systems where > latency and pretransients are important because minimum phase filters > minimize pretransients (at the cost of changing the phase relationships of > individual spectral components). The latter is apparently (much?) less > noticeable than the former. Somehow I recall someone saying that the ear is > not phase sensitive. So, while a heard waveform might have amplitude or > envelope differences because of phase distortion, does the ear perceive it? > I'm sure there are pathological cases where it does. However, for general > sounds, is this a major factor? The suggestion was that it isn't. So, > maybe that's all there is to it.I would say not very phase sensitive. My understanding is that the nerve impulses for low frequencies are generated on a certain part of the cycle, and for high frequencies only on some cycles. I believe that the timing information (to do stereo imaging) is done separately from sound (frequency, etc.) information. I don't know how much effect phase has on each, but it might be very different.> It's probably important to say something about "minimum phase". Minimum > phase means minimum delay at each frequency doesn't it? Whereas, linear > phase means flat delay and we know that the delay is longer than the minimum > at each (or most) frequency(ies). So, when there is a wideband transient we > want all of the frequency components to be delayed the least in order > minimize the delay of the transient. To one degree or another, the > transient will be distorted in amplitude but what does that mean? > If the transient includes a long precursor or pretransient then that could > be viewed as an amplitude distortion also couldn't it? It's just of a > different type.I am not so sure about this, but it sounds fine to me. -- glen
Reply by ●April 2, 20052005-04-02
Jerry Avins wrote:> Oh? The delays at all frequencies are matched. What do you mean by timing?Energy begins to emerge from a drum before it is hit. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
Reply by ●April 2, 20052005-04-02
Fred Marshall wrote:> It certainly preserves *something* > better but what should that be called?Exactly what I've come to wonder.> Minimum phase filtering is preferable in audio systems or systems where > latency and pretransients are important because minimum phase filters > minimize pretransients (at the cost of changing the phase relationships of > individual spectral components). The latter is apparently (much?) less > noticeable than the former. Somehow I recall someone saying that the ear is > not phase sensitive.But it is certainly TOA sensitive which belies that to some extent.> So, while a heard waveform might have amplitude or > envelope differences because of phase distortion, does the ear perceive it? > I'm sure there are pathological cases where it does. However, for general > sounds, is this a major factor? The suggestion was that it isn't. So, > maybe that's all there is to it.As with all things, it depends. Upon what is a question not well enough explored. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
