Yea, let's not worry about if its audible or not and just discuss the totally academic exercize of "doing it just because we can - maybe". That audiblity stuff is too difficult. Of course, in the end its all that really matters, but heck, lifes too short to worry about important things.
Again, too many variables. The better question would be to test a series of loudspeakers as you suggest and then pre-process the input so that all the speakers will reproduce a square wave (measured at the listening
position) and then see if the conclusions were the same. If they are, then see if there is a preference for the speaker deemed the best with and without phase correction. To avoid potential bias due to digital processing the signal would be digitally processed in both series of tests but in one case no phase corrections would be applied and in the other it would.
I think that I'll run this test at DIY NE this fall. 😀
Now I've really got to get to playing around with SE v17. If only I didn't have to replace 6 windows, finish the basement entrance, repair some deck steps, install a ceiling fan, install a whole-house fan... 🙁
Dave
The problem with getting a speaker to reproduce ideal squarewaves, or even evaluating them for this, is that the whole concept assumes you are working with a system that has one input and one output (like an amplifier). A speaker has one input and an infinite number of outputs.
With room reflections, diffractions, speaker directivity, and other spatial effects, ideal square wave response can be at one spot in space and will rapidly fall apart even a little bit away from that spot. Like moving your head a little even.
With room reflections, diffractions, speaker directivity, and other spatial effects, ideal square wave response can be at one spot in space and will rapidly fall apart even a little bit away from that spot. Like moving your head a little even.
Yes, the issue probably reduces to the impact of the response on-axis unless there is also some corresponding and audible change in the system power response. I don't know what impact might occur with, for example, a 3-way system that uses LR2 crossovers when the phase is linearized and all else it the same. That's the beauty of this difference, it will be one of the few instances in which one can make a comparison with "all else being equal". I would think that any audible differences would be apparent with speakers well out into the room that minimize the influence of reflections in the midrange and up insofar as localization and image cues are concerned. Maybe the change will be more apparent from the change in the bass highpass delay, I have no idea at this point.
In any case, it will add a topic not yet undertaken at our DIY event. I've tended to avoid "competitions and tests" to keep them more genial, relaxing and enjoyable. This sort of test will be more out of curiosity than any effort to prove anything to anyone. But if the difference is not too subtle if audible, well, that will just fuel the fire. 😱
Dave
Hi Bill
Your starting to echo my beliefs to the letter. Thinking of a loudspeaker as a SISO device is never going to lead to anything useful. One has to come to terms with the fact that audio electronics is always SISO but only headphones even come close to this for the loudspeakers. What is possible and reasonable for one has no equivalence to the other. Take EQ for example - works great for SISO, doesn't work in the arbitrary SIMO case.
I don't disagree with the SISO/SIMO concept, but what the speaker does and what diffraction and reflections, etc.. do are spearate issues. Certainly we try to limit the effect of diffraction, and Earl, you of all people try to limit the effects of early reflections in you designs (i.e. your designs attempt to be SISO with regard to radiated sound over a fairly wide window). If we just say there is not point in trying to achieve wave form fidelity over some listening window, then why not just say nothing matters? After all, if you move you head a little the amplitude response will change so why bother to design for a target amplitude anywhere? Obviously I am asking a retorical question. The point is that achieveing square wave response is just a means of saying that wave form fidelity is achieved relative to that design point. Degradation off axis occurs in any speaker, to a greater or lesser extent, depending on the design. In a conventional speaker or one corrected to linear phase the degradition is the same. Why not start with a speaker that at least radiated with wave form fidelity?
As for reflections, they are delayed images of the radiated sound. If the direct sound has wave form fidelity, so will the reflected sound. How they sum at the listeners ear is another matter.
This kind of black and white extremist arguement is not very productive. One can always argue that if everything is "perfect everywhere" then there isn't a problem, but that gets us nowhere, because nothing is perfect and the "black and white" arguements stop right there. Audio, like the real world, lies somewhere between the two extremes and its not productive to argue only at the extremes.
Hello,
I don't want to perturbate the interesting discussion but want just to indicate that one of my excel spreadsheet to simulate a 3 ways loudspeakers system (allowing the choice of the crossovers type, order, gain, polarity of the loudspeaker, delay between loudspeakers... ) displays the expected 100Hz square wave resulting from the choice of the parameters.
This version is downloadable for free on my friend Nicolas Davidenko's website at:
http://nicolas.davidenko.perso.sfr.fr/outils/filtre_carre.zip
It enables to compare your simulation to few quasioptimal crossovers of which I am the author.
Best regards from Paris
Jean-Michel Le Cléac'h
I don't want to perturbate the interesting discussion but want just to indicate that one of my excel spreadsheet to simulate a 3 ways loudspeakers system (allowing the choice of the crossovers type, order, gain, polarity of the loudspeaker, delay between loudspeakers... ) displays the expected 100Hz square wave resulting from the choice of the parameters.
This version is downloadable for free on my friend Nicolas Davidenko's website at:
http://nicolas.davidenko.perso.sfr.fr/outils/filtre_carre.zip
It enables to compare your simulation to few quasioptimal crossovers of which I am the author.
Best regards from Paris
Jean-Michel Le Cléac'h
Last edited:
But the summation of lots of these reflections on top of the direct signal, with mixes of the reflection images insanely varying with location, has little resemblance to a square wave. Unless we are talking "square wave" only to the third or fifth harmonic of low frequency fundamentals and just close up enough to swamp the reflections.
Maybe if we are talking that an impulse shape needs to be preserved, for some reason, so they can arrive sequentially, that might make sense to me. If the meaning is "for square waves at all audio frequencies to be ideally reproduced", then I guess it's the same as the IR being ideal, though. If you can somehow see the square waves without the reflections contributing.
We have discussed this before. You have a LP filter (which has a region of constant GD) and an inverted polarity HP. At low frequency the response looks like a decent square wave but the crossover is not linear phase. It is not sufficient to look at a single frequency, well below the x-o point and make such a conclusion. Rather examine a series of frequency which start below the x-o point and spane the crossover frequency as ipresented eariler for a 1k x-o:
An externally hosted image should be here but it was not working when we last tested it.
And there is also the non-constant GD introduced by the low frequency cut off of the woofer which is not accounted for in you simulations. Even with a true TP crossover, if the GD associated with the woofer cut off is not corrected low frequency square waves will not approach flat top response.
Hello john k.
I was not waiting for answer to my message from you. As you said we yet discussed the topics.
For sure my method is not equivalent to a linear phase filter but it is the best simple aproximation and it can be passively implemented which is not the case with linear phase crossovers. (Furthermore: compared to a classical 3rd order crossover, there is no added complexity with a le CLéac'h crossover).
You refused to consider psychoacoustics but the inverted polarity of the tweeter (assuming a 2 ways system ) only has effect at very high frequency (above 4kHz) inside a frequency range where group delay distortion (which is small with my filter) is not audible. Below 4kHz the group delay of the Le Cléac'h crossover is very flat and around the common cut off of the HP and the LP the loudspeakers operates quasi in phase.
Your remark about the group delay rise (generally under 100Hz) of the bass is general and there is no reason why to consider it in the specific case of a Le Cléac'h crosover and not in the case of the other crossovers...
Then about the question: "are linear phase crossovers the best solution?" , the answer is not so clear as many said in that threads.
A part of the population seems to dislike some artifacts due to that kind of crossover and specially pre ringing... (which is more important when the cut off of the low pass filter is low in frequency).
Best regards from Paris, France
Jean-Michel Le Cléac'h
I was not waiting for answer to my message from you. As you said we yet discussed the topics.
For sure my method is not equivalent to a linear phase filter but it is the best simple aproximation and it can be passively implemented which is not the case with linear phase crossovers. (Furthermore: compared to a classical 3rd order crossover, there is no added complexity with a le CLéac'h crossover).
You refused to consider psychoacoustics but the inverted polarity of the tweeter (assuming a 2 ways system ) only has effect at very high frequency (above 4kHz) inside a frequency range where group delay distortion (which is small with my filter) is not audible. Below 4kHz the group delay of the Le Cléac'h crossover is very flat and around the common cut off of the HP and the LP the loudspeakers operates quasi in phase.
Your remark about the group delay rise (generally under 100Hz) of the bass is general and there is no reason why to consider it in the specific case of a Le Cléac'h crosover and not in the case of the other crossovers...
Then about the question: "are linear phase crossovers the best solution?" , the answer is not so clear as many said in that threads.
A part of the population seems to dislike some artifacts due to that kind of crossover and specially pre ringing... (which is more important when the cut off of the low pass filter is low in frequency).
Best regards from Paris, France
Jean-Michel Le Cléac'h
We have discussed this before. You have a LP filter (which has a region of constant GD) and an inverted polarity HP. At low frequency the response looks like a decent square wave but the crossover is not linear phase. It is not sufficient to look at a single frequency, well below the x-o point and make such a conclusion. Rather examine a series of frequency which start below the x-o point and spane the crossover frequency as ipresented eariler for a 1k x-o:
And there is also the non-constant GD introduced by the low frequency cut off of the woofer which is not accounted for in you simulations. Even with a true TP crossover, if the GD associated with the woofer cut off is not corrected low frequency square waves will not approach flat top response.
IMO it isn't necessary to be able to accurately reproduce a squarewave. But it would be fine if the speaker as a whole behaves like a minimum-phase device.
This speaker would still reproduce squarewaves wit reasonable accuracy (with some tilt and it wouldn't show infinitely steep rising and falling edges either). But it wouldn't introduce any additional group delay distortion.
I did this with my speakers - simply because I know how to do it.
Regrading group delay: I think it is well accepted that not the group delay itself is a problem (at least not in the orders of magnitude that we are dealing with). But the group-delay DISTORTION matters.
A clear figure for how much at what frequency would be below perception doesn't exist yet - at least not regrading the group delay distortion like it is introduced by speakers.
The tests by Bluert and Laws were done using a peaking sort of group delay distortion. While the group-delay of certain crossovers like LR does indeed have a small peak as well - its shape differs quite remarkably from the one used in aforementioned tests.
Maybe a figure showing the allowable rate-of change (dGD/df) could be found.
If research would show that some "flattening" of the group-delay distortion is sufficient for it to stay below perception then such a blameless crossover could be made with minimal effort.
Regards
Charles
This speaker would still reproduce squarewaves wit reasonable accuracy (with some tilt and it wouldn't show infinitely steep rising and falling edges either). But it wouldn't introduce any additional group delay distortion.
I did this with my speakers - simply because I know how to do it.
Regrading group delay: I think it is well accepted that not the group delay itself is a problem (at least not in the orders of magnitude that we are dealing with). But the group-delay DISTORTION matters.
A clear figure for how much at what frequency would be below perception doesn't exist yet - at least not regrading the group delay distortion like it is introduced by speakers.
The tests by Bluert and Laws were done using a peaking sort of group delay distortion. While the group-delay of certain crossovers like LR does indeed have a small peak as well - its shape differs quite remarkably from the one used in aforementioned tests.
Maybe a figure showing the allowable rate-of change (dGD/df) could be found.
If research would show that some "flattening" of the group-delay distortion is sufficient for it to stay below perception then such a blameless crossover could be made with minimal effort.
Regards
Charles
Jean-Michel:
There are two issues, the (nonlinear) phase rotation due to the crossover, and that due to the system band pass response (woofer HP and tweeter LP characteristics). Your approach addresses, in part, only the phase rotation of the crossover. What I have presented addresses the entire system. The speaker is linear phase from below audibility to above 20 K Hz.
Preringing is an issue, and an interesting one. For the artifacts of preringing to be present in the acoustic signal there must be misalignment of the sources. This means the artifacts will be come apparent in the off axis response, and primarily, for a well designed speaker with drivers positioned vertically, in the vertical off axis response. As the crossover frequency becomes lower the degree of misalignment for a giver off axis position becomes less relative to the wave length at the crossover frequency. Thus, as the crossover frequency becomes lower misalignment and preringing become a non-issue.
At higher frequency what is observed is that wave form fidelity may be lost. When compared to a conventional crossover it may appear that the change in the wave form, compared to the on axis response, is more severe than the change with conventional crossover (maybe not). However, the fidelity of the wave form to the original from the linear phase system, IMO, still better off axis than the conventional system will be on axis. That there may be some audibility effect of this actually bolsters the case for transient accurate crossovers in general since if this off axis degradation is audible, being less severe that the loss of fidelity of conventional crossover, then surely the same individuals who can hear such artifacts would also be expected to hear the differences, on axis, between TP crossover and conventional crossovers. You can think of it in another way as well. If you look at the GD of a conventional crossover then there is typically a region of constant GD below the mid/tweeter x-o point. Above the x-o point the GD tends to zero. In a sense, conventional crossover always prering as the high frequency components of the response are always presented to the listener before the low frequency components.
Earl,
I agree that the reflections will only be the same if the speaker is truly CD. This is a factor in all cases, linear phase or not, where the reflected sound is not the same at the direct.
Bill,
The square wave is of no interest to me as far as music reproduction is concerned. It is just a tool used to show the character or the ability of the speaker system to accurately reproduce the input signal, at lease over some reasonable window. A square wave makes it much easier to assess what is happening at low frequency. If we accept the idea that directional clues and transient information is contained in the direct sound then such wave form fidelity should be of advantage.
FWIW here is the impulse response of a simple 2-way system, 5" woofer, 1" tweeter, 2k LR4 crossover. The amplitude has been (digitally) equalized flat on the design axis to that of a 50 Hz high pass response in both cases. In the linear phase case the phase has also been corrected to linear from below 20 Hz to above 20k Hz.
As you can see the linear phase system reproduced the reference pulse very well with only moderate off axis horizontal degradation. Vertically both systems show degradation. If the differences between A and C due to off axis "preringing" are audible, I find it difficult to believe that differences between A and D are not audible. (Vertical off axis is 15 degrees above.) Also, both system have the same frequency response on and off axis.
There are two issues, the (nonlinear) phase rotation due to the crossover, and that due to the system band pass response (woofer HP and tweeter LP characteristics). Your approach addresses, in part, only the phase rotation of the crossover. What I have presented addresses the entire system. The speaker is linear phase from below audibility to above 20 K Hz.
Preringing is an issue, and an interesting one. For the artifacts of preringing to be present in the acoustic signal there must be misalignment of the sources. This means the artifacts will be come apparent in the off axis response, and primarily, for a well designed speaker with drivers positioned vertically, in the vertical off axis response. As the crossover frequency becomes lower the degree of misalignment for a giver off axis position becomes less relative to the wave length at the crossover frequency. Thus, as the crossover frequency becomes lower misalignment and preringing become a non-issue.
At higher frequency what is observed is that wave form fidelity may be lost. When compared to a conventional crossover it may appear that the change in the wave form, compared to the on axis response, is more severe than the change with conventional crossover (maybe not). However, the fidelity of the wave form to the original from the linear phase system, IMO, still better off axis than the conventional system will be on axis. That there may be some audibility effect of this actually bolsters the case for transient accurate crossovers in general since if this off axis degradation is audible, being less severe that the loss of fidelity of conventional crossover, then surely the same individuals who can hear such artifacts would also be expected to hear the differences, on axis, between TP crossover and conventional crossovers. You can think of it in another way as well. If you look at the GD of a conventional crossover then there is typically a region of constant GD below the mid/tweeter x-o point. Above the x-o point the GD tends to zero. In a sense, conventional crossover always prering as the high frequency components of the response are always presented to the listener before the low frequency components.
Earl,
I agree that the reflections will only be the same if the speaker is truly CD. This is a factor in all cases, linear phase or not, where the reflected sound is not the same at the direct.
Bill,
The square wave is of no interest to me as far as music reproduction is concerned. It is just a tool used to show the character or the ability of the speaker system to accurately reproduce the input signal, at lease over some reasonable window. A square wave makes it much easier to assess what is happening at low frequency. If we accept the idea that directional clues and transient information is contained in the direct sound then such wave form fidelity should be of advantage.
FWIW here is the impulse response of a simple 2-way system, 5" woofer, 1" tweeter, 2k LR4 crossover. The amplitude has been (digitally) equalized flat on the design axis to that of a 50 Hz high pass response in both cases. In the linear phase case the phase has also been corrected to linear from below 20 Hz to above 20k Hz.
As you can see the linear phase system reproduced the reference pulse very well with only moderate off axis horizontal degradation. Vertically both systems show degradation. If the differences between A and C due to off axis "preringing" are audible, I find it difficult to believe that differences between A and D are not audible. (Vertical off axis is 15 degrees above.) Also, both system have the same frequency response on and off axis.
Last edited:
Hi John - This is basically the impulse response and I don't have a problem with saying that a compact impulse response at all angles is a good goal - thats CD, just said a different way. And the fact is that this will likely yield a decent square wave - but that's just a peripheral result not the goal.
Hello John,
My post was not intended to promote my crossover but the spreadsheet I give for free. This spreadsheet , which one, many find very pedagogical, shows with great ease how every parameter plays a role in the way the crossover behaves (choice of the crossover type, of the order, of the gain and polarity applied to the loudspeaker and the delay between loudspeakers...)
Other sources of pre-ringing exists I guess, e.g.: in the case the HF loudspeaker has to compensate the pre-ringing of the LF loudspeaker and is unable to do so...
Best regards from Paris, France
Jean-Michel Le Cléac'h
My post was not intended to promote my crossover but the spreadsheet I give for free. This spreadsheet , which one, many find very pedagogical, shows with great ease how every parameter plays a role in the way the crossover behaves (choice of the crossover type, of the order, of the gain and polarity applied to the loudspeaker and the delay between loudspeakers...)
Other sources of pre-ringing exists I guess, e.g.: in the case the HF loudspeaker has to compensate the pre-ringing of the LF loudspeaker and is unable to do so...
Best regards from Paris, France
Jean-Michel Le Cléac'h
It's more than saying the impulse is compact. It is that I = O, at least over some reasonable window. Take you system for example. You strive for CD over some window. But even though you have relatively flat amplitude if you look at the radiated wave form "I" dose not equal "O" because there is time (phase) distortion associated with the crossover and the system band width limits. I'm not going to get into an argument as to who can or can not hear what. I just always believe accuracy and fidelity should be referenced to "I".
I guess I just look at it differently. I see a system where there is a spatial region of temporal perfection in the direct sound with degradation (temporal error) outside that region as one which is superior to another which is, at best, uniformly inaccurate in the time domain everywhere.
That doesn't mean I won't design/build another speaker with a conventional passive crossover and the associated phase distortion. But at least I have the option to use my PC to correct the phase is so desired.
Compact impulse, yes, but CD doesn't connote minimum-phase nor linear-phase, it's related strictly to amplitude response unless I'm mistaken. I find the discussion related to the analysis of those two to be interesting as well as CD, but the one doesn't discount the other, seems to me that they're separate issues.
I would expect that an LR4 analog crossover-based CD system vs. a linear phase phase crossover (say even that same crossover implemented with the SE UE), both using the same waveguide, would show distinctly different impulse responses, compact notwithstanding.
Dave
Edit: I just noticed that John addressed this as I was typing.
Last edited:
- Status
- Not open for further replies.