Linkwitz have also talked about the issue of phase distortion. He's on the side that suggest phase distortion is inaudible.
http://www.linkwitzlab.com/phs-dist.htm
http://www.linkwitzlab.com/phs-dist.htm
I think we're talking about two different things here
Yes, that has been his position. I haven't tried it, so I can't say for myself. I do know that I find 2nd order slightly better than 4th, but both distort phase.
However, that's not related to the question in the original post. The question was related to inverting the connection of a driver in a 2nd order crossover, a requirement if a design target is to be met in that situation. The phase distortion is inherent to all but transient-perfect crossovers. Crossovers all have requirements that must be met such as the aforementioned inverted connections for 2nd order. Otherwise, a target as a goal (an acoustic one to be specific) is meaningless and we all might as well just blindly make random crossovers until we hear something pleasing.
Dave
banana said:
Yes, that has been his position. I haven't tried it, so I can't say for myself. I do know that I find 2nd order slightly better than 4th, but both distort phase.
However, that's not related to the question in the original post. The question was related to inverting the connection of a driver in a 2nd order crossover, a requirement if a design target is to be met in that situation. The phase distortion is inherent to all but transient-perfect crossovers. Crossovers all have requirements that must be met such as the aforementioned inverted connections for 2nd order. Otherwise, a target as a goal (an acoustic one to be specific) is meaningless and we all might as well just blindly make random crossovers until we hear something pleasing.
Dave
Maybe I am one of the 1 of 100 but I find accurate acoustical transient reproduction audible. Just as a rhetorical question for other readers,
"Do you know that you cannot hear accurate impulse or transient response reproduction?"
The impulse response is the key. Design your loudspeakers for accurate acoustical reproduction of impulse responses. To do this they will have to preserve polarity and be closely aligned or compensated for air path delay through crossover regions. Get as close as you can. If you are close enough, then you will have both flat frequency response and flat phase response. Assuming here that all other important design criteria have been met. Transducers still have to sum acoustically for example.
If you invert polarity, then you will not be close to accurately reproducing transients or impulse responses.
Mark
P.S. I would lump phase distortion with linear distortion. Borderline oxymoron alert. At best, sloppy and confusing use of terminology. I am surprised someone hasn't started referring to frequency distortion.
"Do you know that you cannot hear accurate impulse or transient response reproduction?"
The impulse response is the key. Design your loudspeakers for accurate acoustical reproduction of impulse responses. To do this they will have to preserve polarity and be closely aligned or compensated for air path delay through crossover regions. Get as close as you can. If you are close enough, then you will have both flat frequency response and flat phase response. Assuming here that all other important design criteria have been met. Transducers still have to sum acoustically for example.
If you invert polarity, then you will not be close to accurately reproducing transients or impulse responses.
Mark
P.S. I would lump phase distortion with linear distortion. Borderline oxymoron alert. At best, sloppy and confusing use of terminology. I am surprised someone hasn't started referring to frequency distortion.
PB2 said:
Headphones.
I have done this a bit over speakers too myself, with the same result.
I am not saying that it would be impossible to hear with any signal, actually I managed to point it out with reasonable statistical significance once using the "crossover frequency" ~120 Hz and there is a "buzzing" sound in the recording (as I recall it it was a classical piece with double-basses playing). Anyway it was terribly terribly hard to hear and something that would definitely have been left unnoticed without the possibility for direct comparison.
But anyway, as I said, anyone who wonders about this should build the thing and listen to it. It is an eye-opener.
We need to be clear here
If that is the goal. It sounds simple, but in practice it's not simple and is, in fact, never truly achieved. There is a limited number of crossovers that fall into the category of transient-perfect systems, but they all fall apart somewhere along the way.
I'm not sure what you mean by preserve polarity. The most basic transient-perfect system is a 1st order Butterworth. However, the polarity can be reversed, say on the tweeter, without affecting the summed response on the design axis. The drivers in the crossover region remain in phase quadrature. The lobe is inverted, but then the summed response is not transient perfect anywhere off-axis no matter the lobe position. Any transient perfect crossover fails off-axis as does the summed response for most, if not all crossovers in the off-axis.
Not necessarily true, as explained above.
Because that's not correct. It's a time-delay phenomenon that does not show up in the frequency domain steady-state amplitude response. For example, a 4th order Linkwitz-Riley crossover properly designed so the acoustic response of the system sums to a linear response without any peaks/dips in the steady-state response has no linear distortion due to the acoustic crossover. There is no positive nor negative gain at any frequency. The impulse and step responses to transients show the impact, but the frequency domain does not.
If you saw two measurements of a summed response in the frequency domain, both with flat summed amplitude response, you could not determine whether one, both or neither of them were from transient-perfect systems. In the frequency domain amplitude response, the two are the same. The phase response will show the phase group delay, but that's all it is, a group delay. This is what causes the impulse and step responses to be imperfect, but it's not a linear distortion causing it, it's a time delay distorting the response.
For example, a first order transient-perfect system has time delays in each leg. The phase is not linear phase in either leg. However, the summed response is a minimum-phase response. It's also not a linear phase response because all speaker systems ultimately are at best 2nd order bandpass in nature. Now if you want to say that the only system capable of properly reproducing a square wave is a DSP-corrected linear phase system, you'd still be wrong due to the artifacts introduced in the DSP processes and the ultimate bandpass nature of real drivers even when corrected. It depends on how far you want to go in defining transient-perfect if that's the goal.
Dave
If that is the goal. It sounds simple, but in practice it's not simple and is, in fact, never truly achieved. There is a limited number of crossovers that fall into the category of transient-perfect systems, but they all fall apart somewhere along the way.
I'm not sure what you mean by preserve polarity. The most basic transient-perfect system is a 1st order Butterworth. However, the polarity can be reversed, say on the tweeter, without affecting the summed response on the design axis. The drivers in the crossover region remain in phase quadrature. The lobe is inverted, but then the summed response is not transient perfect anywhere off-axis no matter the lobe position. Any transient perfect crossover fails off-axis as does the summed response for most, if not all crossovers in the off-axis.
Not necessarily true, as explained above.
Because that's not correct. It's a time-delay phenomenon that does not show up in the frequency domain steady-state amplitude response. For example, a 4th order Linkwitz-Riley crossover properly designed so the acoustic response of the system sums to a linear response without any peaks/dips in the steady-state response has no linear distortion due to the acoustic crossover. There is no positive nor negative gain at any frequency. The impulse and step responses to transients show the impact, but the frequency domain does not.
If you saw two measurements of a summed response in the frequency domain, both with flat summed amplitude response, you could not determine whether one, both or neither of them were from transient-perfect systems. In the frequency domain amplitude response, the two are the same. The phase response will show the phase group delay, but that's all it is, a group delay. This is what causes the impulse and step responses to be imperfect, but it's not a linear distortion causing it, it's a time delay distorting the response.
For example, a first order transient-perfect system has time delays in each leg. The phase is not linear phase in either leg. However, the summed response is a minimum-phase response. It's also not a linear phase response because all speaker systems ultimately are at best 2nd order bandpass in nature. Now if you want to say that the only system capable of properly reproducing a square wave is a DSP-corrected linear phase system, you'd still be wrong due to the artifacts introduced in the DSP processes and the ultimate bandpass nature of real drivers even when corrected. It depends on how far you want to go in defining transient-perfect if that's the goal.
Dave
...but, I have to add:
That does not mean that I think there is no audible difference between 1st and 2nd order crossovers. I think that the explanation for these differences are to find in the amplitude response and the better suppression of the stop band, rather than in the phase response.
That does not mean that I think there is no audible difference between 1st and 2nd order crossovers. I think that the explanation for these differences are to find in the amplitude response and the better suppression of the stop band, rather than in the phase response.
MarkMcK said:
Some level-matched controlled tests with all-pass filters in the midrange demonstrated that neither I nor any of my guinea pigs could hear the difference in or out of circuit. I can't make sweeping statements (only tested at one frequency with 6 subjects and on one system with full-range ESLs, and just myself with headphones), but it seems to be a whole lot less important than frequency response, somewhat less important than distortion, and enormously less important than polar pattern.
Hartono said:
Ah, yes there can be several purposes with allpass filters. Mine was to examine the audible effects on a 2nd order LR-crossover filter. The sum of the branches in such a filter has an identical response as the allpass filter I used.
So, rather than comparing a system with 2nd order filters with a system having a 1st order filter, I compared the allpass filter with a bypass connection. By doing so I eliminate the effect of the drivers and their behaviour in the stopband. In this way I can listen to the effects of phase only.
This is completely different from using allpass filters as a compensation in a crossover filter.
Hi Svante
Listening to headphones is NOT the same as listening to speakers with the same time issues.
Loudspeakers have sources which in addition to each having there own magnitude and phase, are also separated in space in X, Y and Z planes.
If the sources are more than about ¼ wavelength center to center, then they do not add coherently and produce a direction dependant radiation.
Here is (the spatial interaction on account of the acoustic distances) is a major element in your ears ability to “locate” the speaker in space as opposed to hearing the image created by both speakers.
This is a tough nut, it is very hard not to have driver to driver interference related to that spacing issue but at least one can see it with a spherical radiation plot.
If no compensation is taken and except for a first order, as others pointed out, then through crossover there is about 90 degrees phase rotation per order of slope used.
It is the individual drivers phase response (which is not at zero degrees usually) added to the phase shift present in normal passive crossovers, that with very few exceptions prevents waveshape preservation over any bandwidth in passive multiway loudspeakers.
At least one horn configuration system can be made to preserve waveshape passively and not have driver to driver interference however.
Best,
Tom Danley
Listening to headphones is NOT the same as listening to speakers with the same time issues.
Loudspeakers have sources which in addition to each having there own magnitude and phase, are also separated in space in X, Y and Z planes.
If the sources are more than about ¼ wavelength center to center, then they do not add coherently and produce a direction dependant radiation.
Here is (the spatial interaction on account of the acoustic distances) is a major element in your ears ability to “locate” the speaker in space as opposed to hearing the image created by both speakers.
This is a tough nut, it is very hard not to have driver to driver interference related to that spacing issue but at least one can see it with a spherical radiation plot.
If no compensation is taken and except for a first order, as others pointed out, then through crossover there is about 90 degrees phase rotation per order of slope used.
It is the individual drivers phase response (which is not at zero degrees usually) added to the phase shift present in normal passive crossovers, that with very few exceptions prevents waveshape preservation over any bandwidth in passive multiway loudspeakers.
At least one horn configuration system can be made to preserve waveshape passively and not have driver to driver interference however.
Best,
Tom Danley
Allpass network
Originally posted by Svante: I am not saying that it would be impossible to hear with any signal, actually I managed to point it out with reasonable statistical significance once using the "crossover frequency" ~120 Hz and there is a "buzzing" sound in the recording (as I recall it it was a classical piece with double-basses playing). Anyway it was terribly terribly hard to hear and something that would definitely have been left unnoticed without the possibility for direct comparison.
But anyway, as I said, anyone who wonders about this should build the thing and listen to it. It is an eye-opener.
I have performed a test with an 80Hz allpass filter (90 degree shift at 80Hz) with an input of a half-wave (rectified) 80Hz sine wave. The effect is clearly audible with the allpass filter giving what sounds like an apparent rolloff of the high frequency content.
Originally posted by Svante: I am not saying that it would be impossible to hear with any signal, actually I managed to point it out with reasonable statistical significance once using the "crossover frequency" ~120 Hz and there is a "buzzing" sound in the recording (as I recall it it was a classical piece with double-basses playing). Anyway it was terribly terribly hard to hear and something that would definitely have been left unnoticed without the possibility for direct comparison.
But anyway, as I said, anyone who wonders about this should build the thing and listen to it. It is an eye-opener.
I have performed a test with an 80Hz allpass filter (90 degree shift at 80Hz) with an input of a half-wave (rectified) 80Hz sine wave. The effect is clearly audible with the allpass filter giving what sounds like an apparent rolloff of the high frequency content.
I thought this thread was played out, not so.
Dave, it will take some time to read your response.
For now, four issues:
(1) In years past an allpass was considered to be a broadband filter with a 90-deg phase shift over a frequency spectrum. It was a cascade of phase shifters. Phase shifters are what are being called allpass filters now. They are allpass, but very simple. They have been used in the distortion effects of guitar amps for decades.
(2) This concern with the allpass doesn't make complete sense. An allpass in this situation is a simple RC network that is linearized by an opamp. Any circuit with a time constant is going to change the shape of the wave and round it somewhat. Crossovers also work by reactance (unless they are digital filtering) and reactance by nature changes the shape of the waveform.
(3) Concerning phase response. In the 1960s there was an AES article about how phase response doesn't matter. It doesn't matter with today's equipment, or with the equipment of the 60s. I strongly suspect that the absense of phase accuracy is what causes sound reproduction to stay in its' traditional confined atmosphere. Once phase accuracy is broken it cannot be restored. I was told that high-end tape machines are adjusted for phase response. Even so, the rest of the equipment contains reactance that destroys the phase information.
(4) On an oscilloscope, when a signal is added to a 90-deg lag of itself it results in a 45-deg lag with increased amplitude.
Dave, it will take some time to read your response.
For now, four issues:
(1) In years past an allpass was considered to be a broadband filter with a 90-deg phase shift over a frequency spectrum. It was a cascade of phase shifters. Phase shifters are what are being called allpass filters now. They are allpass, but very simple. They have been used in the distortion effects of guitar amps for decades.
(2) This concern with the allpass doesn't make complete sense. An allpass in this situation is a simple RC network that is linearized by an opamp. Any circuit with a time constant is going to change the shape of the wave and round it somewhat. Crossovers also work by reactance (unless they are digital filtering) and reactance by nature changes the shape of the waveform.
(3) Concerning phase response. In the 1960s there was an AES article about how phase response doesn't matter. It doesn't matter with today's equipment, or with the equipment of the 60s. I strongly suspect that the absense of phase accuracy is what causes sound reproduction to stay in its' traditional confined atmosphere. Once phase accuracy is broken it cannot be restored. I was told that high-end tape machines are adjusted for phase response. Even so, the rest of the equipment contains reactance that destroys the phase information.
(4) On an oscilloscope, when a signal is added to a 90-deg lag of itself it results in a 45-deg lag with increased amplitude.
I'm quite sure it depends on how wide a band it occurs in, where it occurs, and how listening tests were conducted.banana said:Linkwitz have also talked about the issue of phase distortion. He's on the side that suggest phase distortion is inaudible.
http://www.linkwitzlab.com/phs-dist.htm
More important is what are the performance of the drivers used during these tests.
Tom Danley said:
No, I agree with that.
Actually, I think headphones would be better for detecting phase errors since they do not have an inherent crossover frequency at all, typically.
Anyway, my intention with the allpass filter is NOT to compare actual crossover filters. It is to examine the statement that a perfect impulse response is nessecary for perceived correct reproduction. I want to know if the phase error alone is detectable. I don't care about directivity or bad stopband behaviour of the drivers or other similar issues in this experiment.
For example, the impulse responses from Linkwitz' site in a previous post are the result of allpass filters, without directivity or driver properties included. It is the audibility of these filters I examine.
My opinion is that looking at impulse responses and drawing sensible conclusions from them regarding the audible result is very very hard. In fact, it is so hard that I find these types of graphs are more misleading than helping in most cases.
And again, build an allpass filter, listen to it and measure its impulse response. It is a rather convincing demo of the above.
I agree with this.Svante said:
In speaker system design, there are so many issues involved, if one masks the other, then some things are not audible because we have not really improved the source of the problem. It's looking at as much different aspects and trying to interpretate the data that is the most difficult part. But I think if one aspect can be developed so that theoretically and measurements can show that it's good, then that is the way it should be designed until there is a good reason to use a modified design to compensate for other deficiencies.
After some tries in two way systems, I will never select drivers that may require phase reversal XO designs.
soongsc said:
Good point. Through experience I am led to conclude that the
most critical area for phase response in crossovers is in the
mid-bass, the region from about 80 to 160 Hz. This is the area
where you encounter many cases of flat response but terrible
sound, and when you improve the phase response, you get
better transient attack - "punch"
I have found that reversed polarity bass always give bad punch. Sometimes it's the recording that is polarity inverted. Having measured some inductors, the low frequency impedance phase differences look interesting, I should be doing some listeing tests soon.Nelson Pass said:
Good point. Through experience I am led to conclude that the
most critical area for phase response in crossovers is in the
mid-bass, the region from about 80 to 160 Hz. This is the area
where you encounter many cases of flat response but terrible
sound, and when you improve the phase response, you get
better transient attack - "punch"
Dave wrote:
So all of these graphs are not from measurements with test equipment, but they are drawn from filter equations?
My statements were in relation to the early post by banana. The ringing at the top of the square wave looks like an electrical measurement.
When you say "acoustic responses" how can you measure them without test equipment? This is after all an electro-mechanical situation. How is acoustic resonse measured in these situations?
I'm trying to figure out how this conversation can deal only with acoustics.
So all of these graphs are not from measurements with test equipment, but they are drawn from filter equations?
My statements were in relation to the early post by banana. The ringing at the top of the square wave looks like an electrical measurement.
When you say "acoustic responses" how can you measure them without test equipment? This is after all an electro-mechanical situation. How is acoustic resonse measured in these situations?
I'm trying to figure out how this conversation can deal only with acoustics.