This is the wrong position to take. My topic is about electric crossover order. This can be 1st,2nd 3rd etc The problems you get with phase deviation at the extreme ends of the response are different from that of combining 2 drive units together.
Why do you criticise 1st order crossovers for inadequate protection when even 4th order has problems? Any speaker has a specific polar pattern which is fixed. How can a fixed polar response mimic what we hear in real life? It cannot!
Not Fixed!
All filter networks have "problems" particularly when they are interposed in the low but variable impedance circuit that connects the power amplifier to loudspeaker driver(s). A HP filter with 24 db/oct slope will adequately protect most MR and HF drivers. The reasons for doing this have already been outlined in several earlier posts, so there is no need to repeat them here. However, it should be noted, that it is not clear, that the phase coherence in the drive signal, gained by use of a single poll, 1st. order filter, is worth the trade of increased IM distortion and reduced driver protection, particularly when the drivers themselves do not exhibit a 1st. order response as well. In fact to get phase coherence at the acoustical output may require a high order filter network that has a phase incoherence that offsets that of the drivers.
Driver polar pattern narrows as signal frequency increases. The onset of this narrowing is determined firstly by driver diaphragm size. If the baffle in which it is mounted in is not flat (including horn), this pattern will be altered but will still exhibit narrowing with frequency increase. There will always be some degree of pattern mismatch because of differing diaphragm sizes between LF, MF and HF drivers, particularly when they are operated in the crossover region where driver outputs overlap. Here LF/MF or MF/HF driver pairs, when operated in these regions, will have opposed (narrow vs. wide) polar responses. An acoustic lenses and/or CD horns may be used to mitigate this problem. I see this mismatch as a much bigger problem than pattern anomalies that may be generated between drivers by the crossover alone.
Regards,
WHG
All filter networks have "problems" particularly when they are interposed in the low but variable impedance circuit that connects the power amplifier to loudspeaker driver(s). A HP filter with 24 db/oct slope will adequately protect most MR and HF drivers. The reasons for doing this have already been outlined in several earlier posts, so there is no need to repeat them here. However, it should be noted, that it is not clear, that the phase coherence in the drive signal, gained by use of a single poll, 1st. order filter, is worth the trade of increased IM distortion and reduced driver protection, particularly when the drivers themselves do not exhibit a 1st. order response as well. In fact to get phase coherence at the acoustical output may require a high order filter network that has a phase incoherence that offsets that of the drivers.
Driver polar pattern narrows as signal frequency increases. The onset of this narrowing is determined firstly by driver diaphragm size. If the baffle in which it is mounted in is not flat (including horn), this pattern will be altered but will still exhibit narrowing with frequency increase. There will always be some degree of pattern mismatch because of differing diaphragm sizes between LF, MF and HF drivers, particularly when they are operated in the crossover region where driver outputs overlap. Here LF/MF or MF/HF driver pairs, when operated in these regions, will have opposed (narrow vs. wide) polar responses. An acoustic lenses and/or CD horns may be used to mitigate this problem. I see this mismatch as a much bigger problem than pattern anomalies that may be generated between drivers by the crossover alone.
Regards,
WHG
Mayday
The use of the word "may" is appropriate here, because consequences are dependent on the signal content of the program being reproduced, and the setting of the level (volume) control in the system used.
My intent is to critsize no one. I will take issue with misnomers and try to provide useful information concerning audio systems design, irrespective of thread or technology ownership. When appropriate, references to the work of others, that I am aware of, will be given to guide those pursueing audio perfection.
Regards,
Bill
Hi Dave,
The use of the word "may" is appropriate here, because consequences are dependent on the signal content of the program being reproduced, and the setting of the level (volume) control in the system used.
My intent is to critsize no one. I will take issue with misnomers and try to provide useful information concerning audio systems design, irrespective of thread or technology ownership. When appropriate, references to the work of others, that I am aware of, will be given to guide those pursueing audio perfection.
Regards,
Bill
Not clear, no. That may not be the goal. Yours obviously is different than many others. The issues you raise can be mitigated with good driver selection and design. It does take more care and the drivers certainly are taxed more. But that's a design decision.
There is another benefit to Butterworth, first or third, that I mentioned to which you failed to respond. If one's goal is a passive system, there is a distinct difference that may make the Butterworth better than any other passive system. That is the power response. This can have an easily measured difference on the order of 3db in the crossover region. As such, since even a first order system that varies in the stop band is still essentially first order in and around the crossover area, the power response is essentially that of a true first order system. You may not think this to be a benefit. Others just may find it better, despite other "possibly audible" issues.
For reference let me point to the article by John K on power response. For a passive system, there is no system with a better power response than an Butterworth if the drivers are not coaxial/coincident. Note that in every case, the higher order L-R has a worse power response than does the Butterworth one order below. It may be that the better power response of the first order Butterworth is more audible than the other issues of concern, so it's not clear that any crossover type is better or worse than others, since it depends on the goals of the designer.
From John's page:
Dave
Last edited:
What's Audible? vs. What's Inaudible?
If a claim is to be clear and certain, then it requires quantified comparative evidence to support it. Anecdotal testimonies won't dispel doubt. There is not even industry consensus that phase incoherence in a typical higher order crossover is audible.
What are the criteria being used for driver selection? There are tradeoffs being made here as well to support the shallow high pass slopes used, particularly bandwidth vs. efficiency that intern effect rise times, overshoot, ringing and transient response of the drivers selected.
As the polar response of all drivers narrows with frequency, matching polar response between drivers in the crossover region is not possible unless a acoustic horn or lens is used; and even then, the results may be of marginal benefit.
In the case of uniform power response, out of necessity, on axis intensity (SPL) will rise with frequency due to the resulting frequency dependant concentration of constant power across a diminishing wave front. I suspect the acoustics of the listening space will determin if this is benificial.
If you pose a question to me, and I read it, I will probably answer it. However, you can forget the suggestion that I am avoiding an issue you raise and that somehow that omission diminishes credibility. I answer several post here most every day and I suspect from time to time I will miss a few. In those cases, a kind reminder will suffice.
Regards,
WHG
If a claim is to be clear and certain, then it requires quantified comparative evidence to support it. Anecdotal testimonies won't dispel doubt. There is not even industry consensus that phase incoherence in a typical higher order crossover is audible.
What are the criteria being used for driver selection? There are tradeoffs being made here as well to support the shallow high pass slopes used, particularly bandwidth vs. efficiency that intern effect rise times, overshoot, ringing and transient response of the drivers selected.
As the polar response of all drivers narrows with frequency, matching polar response between drivers in the crossover region is not possible unless a acoustic horn or lens is used; and even then, the results may be of marginal benefit.
In the case of uniform power response, out of necessity, on axis intensity (SPL) will rise with frequency due to the resulting frequency dependant concentration of constant power across a diminishing wave front. I suspect the acoustics of the listening space will determin if this is benificial.
If you pose a question to me, and I read it, I will probably answer it. However, you can forget the suggestion that I am avoiding an issue you raise and that somehow that omission diminishes credibility. I answer several post here most every day and I suspect from time to time I will miss a few. In those cases, a kind reminder will suffice.
Regards,
WHG
My point is that from the beginning your responses in regard to the benefits of first order vs. others, the origin of the thread, seemed to be not much more than to dismiss first order without indicating that there may be benefits, with comments such as
Is any of this obvious? I would say no and I'm sure were Dunlavy still around he would have quite a bit to say in disagreement as well. First order has limitations as do all systems. One needs to know them and take them into account and I suspect that sort of input may be what the OP was seeking.
With regard to tradeoffs, of course they are being made. That was an early point of mine, but done in a way to indicate that they do not preclude a good system of what is considered to be a first order system. Your approach as I read it is to maximize response from the selected drivers and do so in the digital domain. Fine. Do you think that this is the input being sought by the OP? Possibly, but I suspect not.
Now I may be wrong, but it appeared to me that your input had the result of being dismissive of first order due to your preferences, whatever that may be, rather than simply stating the benefits or demerits in an informative way to allow the OP to make a decision based on the input. Your comments quoted above are an indication. That seemed to be your focus and to me is not being helpful to someone who may be interested in a first order system as even most in the industry use as a term of reference to those systems.
I certainly agree with many of your statements, but I don't agree with the dismissive approach.
Dave
Is any of this obvious? I would say no and I'm sure were Dunlavy still around he would have quite a bit to say in disagreement as well. First order has limitations as do all systems. One needs to know them and take them into account and I suspect that sort of input may be what the OP was seeking.
With regard to tradeoffs, of course they are being made. That was an early point of mine, but done in a way to indicate that they do not preclude a good system of what is considered to be a first order system. Your approach as I read it is to maximize response from the selected drivers and do so in the digital domain. Fine. Do you think that this is the input being sought by the OP? Possibly, but I suspect not.
Now I may be wrong, but it appeared to me that your input had the result of being dismissive of first order due to your preferences, whatever that may be, rather than simply stating the benefits or demerits in an informative way to allow the OP to make a decision based on the input. Your comments quoted above are an indication. That seemed to be your focus and to me is not being helpful to someone who may be interested in a first order system as even most in the industry use as a term of reference to those systems.
I certainly agree with many of your statements, but I don't agree with the dismissive approach.
Dave
There are some practical issues I think you guys should consider in all this.
First order networks sum to flat only when there is 90 degree phase shift between sections. This guarantees asymmetry of polar response. Response will sum higher than flat in one direction and fall into a hole in the other direction (of the vertical polars). If the choice is between asymmetry near the listening axis vs. a power response hole, then I would choose the symmetry of Linkwitz Riley approaches every time.
Low slope means high overlap or more octaves before a driver has faded enough to not cause off axis response ripple. There was an interesting paper by (I think) Glynn Adams formerly of B&W where he tried extremely high slope digital crossovers with imperfect summing. They had deep holes at the crossover frequency but when the holes got narrow enough they became inaudible.
When considering first order filters we have to be referring to the combined acoustical response of driver and network. Since most of our drivers will be falling off at a second order or higher rate at some frequency, then we can’t maintain a first order roll off much below resonance. If response is low enough at that point then the system response might not be impacted, then again, it might. If natural rolloffs are near the crossover point you will need to boost response. Good luck doing this with a passive network.
A first order rolloff isn’t as simple as putting a capacitor in series with a tweeter. A low order crossover will have large interaction with the driver impedance. Conjugating the driver impedance at resonance would be a minimum requirement. Higher order passive networks give more degrees of freedom and a better control of response.
Regarding Dunlavy’s comment, he was obviously talking is simple terms to a journalist. He was also disregarding the distinction between electrical filtering and acoustical response filtering, although I’m sure he understood the distinction. With conventional networks, first order is the only solution, but more complex solutions are out there, subtractive networks (with perfect drivers) would one example. The one that Dennis H. offers seems to be another good example. The fact that response is only 1st order through the crossover region is immaterial. The underlying goal isn’t strictly to be first order, but to achieve the phase linearity that a first order network can offer.
David S.
First order networks sum to flat only when there is 90 degree phase shift between sections. This guarantees asymmetry of polar response. Response will sum higher than flat in one direction and fall into a hole in the other direction (of the vertical polars). If the choice is between asymmetry near the listening axis vs. a power response hole, then I would choose the symmetry of Linkwitz Riley approaches every time.
Low slope means high overlap or more octaves before a driver has faded enough to not cause off axis response ripple. There was an interesting paper by (I think) Glynn Adams formerly of B&W where he tried extremely high slope digital crossovers with imperfect summing. They had deep holes at the crossover frequency but when the holes got narrow enough they became inaudible.
When considering first order filters we have to be referring to the combined acoustical response of driver and network. Since most of our drivers will be falling off at a second order or higher rate at some frequency, then we can’t maintain a first order roll off much below resonance. If response is low enough at that point then the system response might not be impacted, then again, it might. If natural rolloffs are near the crossover point you will need to boost response. Good luck doing this with a passive network.
A first order rolloff isn’t as simple as putting a capacitor in series with a tweeter. A low order crossover will have large interaction with the driver impedance. Conjugating the driver impedance at resonance would be a minimum requirement. Higher order passive networks give more degrees of freedom and a better control of response.
Regarding Dunlavy’s comment, he was obviously talking is simple terms to a journalist. He was also disregarding the distinction between electrical filtering and acoustical response filtering, although I’m sure he understood the distinction. With conventional networks, first order is the only solution, but more complex solutions are out there, subtractive networks (with perfect drivers) would one example. The one that Dennis H. offers seems to be another good example. The fact that response is only 1st order through the crossover region is immaterial. The underlying goal isn’t strictly to be first order, but to achieve the phase linearity that a first order network can offer.
David S.
All very good points, a number of which have been covered. Regarding passive systems, this is certainly more than most DIYers would do, but consider this Dunlavy system:
Step response on tweeter axis:
Just one example. Extreme, yes, but proof that it's not just academic? Also yes.
Dave
Extreme, admittedly, but a real system backed up by measurements.
Step response on tweeter axis:
Just one example. Extreme, yes, but proof that it's not just academic? Also yes.
Dave
Its Magic
I will stick to the attribution of systems, and not those that design and build them. From microphone to media and then on to a loudspeaker, the system that delivers the music to us is not 1st. order, nor are the instruments that produce it, and the ears that hear it. But, how amazing it is, that the facsimile can be so real, despite our efforts to improve it.
Regards,
WHG
I will stick to the attribution of systems, and not those that design and build them. From microphone to media and then on to a loudspeaker, the system that delivers the music to us is not 1st. order, nor are the instruments that produce it, and the ears that hear it. But, how amazing it is, that the facsimile can be so real, despite our efforts to improve it.
Regards,
WHG
I can not for the life of me see any sense in what you wrote! They are also not 2nd, 3rd or 4th nor are they digital! So what has 1st order got to do with your statement. Maybe I missed something.
Terry
Whether you care to use the generally accepted industry reference really isn't important. Your response is focused on that, not on my main point, that from the beginning your approach was to dismiss the possibility, due to your preferences as evidenced by the quotes cited, that these "non-first order" (or whatever you care to call them) can without question be designed to be excellent systems that are highly praised and very, very successful. History by Dunlavy and others is proof of that, that part's not debatable. It seems to me that your input has not been in many ways very helpful to the OP's query IMO.
Dave
It is true that a number of systems have either used proper first order acoustical rolloff or other linear phase approaches. Some have been successful. Thiel does it well and is successful. B&O did phase link with limited success. Technics dabbled in linear phase for a while. Dunlavy made a good product. Conversly many have made good products that aren't linear phase.
To me the more important questions would be: "What compromises must we make to achieve linear phase?" and "Are the benefits worth the effort?"
Clearly low slope will mean driver power handling must be better since driver excursion will increase below crossover. Drivers will generally need wider bandwidth and response extended well above and below crossover. This is a cost factor. Crossovers will be as complex, if not more so since the electrical target functions will be complicated by the need to flatten response beyond the driver's natural rolloff.
System complexity generally goes up. If drivers need to behave well beyond a typical range then a 3-way or 4-way system becomes the easiest route. Not many linear phase 2-ways out there.
What do you gain? The odd technical paper tells you that you might hear a difference on specialized test signals. The same papers we tell you that you won't hear a difference on recorded music.
Finally, and this is killer for me, the (near) off axis performance is generally worse and likely asymmetrical (unless larger symmetrical arrays are used). I know I will hear vertical response variation and want to minimize it.
Yes, you can do it but is it good value/good design or just pursuit of an audiophile ideal of no particular benefit?
David S.
can you give an explanation for why you believe that? If its true why does it matter? If you're a serious listener you will probably be sitting in one position during the listening session.
Not a preferred solution. Except in a limited number of spots where it exactly cancels from the associated drivers, the brick wall cutoff Gibbs ringing will obtrusively cause coloration.
AES E-Library: Application of Digital Filters to Loudspeaker Crossover Networks
Authors: Wilson, R. J.; Adams, G. J.; Scott, J. B.
Not their conclusion.
"This paper addresses the subject of finite duration impulse response (FIR) and infinite duration impulse response (IIR) filters used to implement loudspeaker crossover filters. The filters have been tested on the TMS32020 digital signal processor. In the case of FIR filters, a very compact form of coding has been developed which requires only one convolution to give both the high and low pass filter outputs. Listening tests were performed to evaluate the improvement in the off-axis frequency response obtained by using very high slope filters. An FIR filter for equalizing a tweeter response was also designed."
Rhonda Wilson has been a main theoritician at Meridian for years. Glynn Adams, sadly, died soon after this paper.
David S
I am a serious listener but I like to be able to either stand or sit when listening.
The first order linear phase system are typically quadrature summing so from the ideal axis, crossover response will rise 3dB in one direction and fall into a nearby response hole in the other.
I just believe it is a basic tenant of good speaker design: A good system will achieve flat and smooth response over as wide a listening window as possible. I would always trade off linear phase performance against better response in a reasonable listening window.
David S.
I don't think they even address this issue in the paper. Secondly, the pre-ringing can sound particularly nasty with brick wall filters in the middle of the audio pass band when the cancellation is not essentially perfect.
"Listening tests were performed to evaluate the improvement in the off-axis frequency response obtained by using very high slope filters. "
If this is the paper I'm recalling (I'll have to dig through my files to see if I have a copy) that is just what they looked into and also their conclusion: that a two way system could be made with two abutting brick wall filters and that proper summing wasn't required. As long as the "divot" in the frequency response was sufficiently narrow it became inaudible. If true, this is good because the listening axis becomes immaterial (per the quote). Kind of a Joseph Audio Infinite Slope system on steroids.
I'm sure we would be aghast at the impulse response, but they claimed no audible effects.
David
They can claim anything they want - it's their paper, but if up to 20 cycles of pre-ringing is generally acknowledged to be so audibly problematic at 20khz, then it can't be innocuous at 2khz where the ear's sensitivity far greater in virtually all respects.
Here's a link to an AES paper from 2008 titled: "Perceptual Study and Auditory Analysis on Digital Crossover Filters" at:
http://www.acoustics.hut.fi/~mak/PUB/AES124-000056.pdf
Among the researchers' conclusions:
"FIR crossovers seem to be *highly susceptible* to off-axis errors with higher filter orders. The flight time difference of only 0.02-0.03 ms be- tween low- and highpass bands at 3 kHz was found to produce audible ringing with high FIR orders of the scale 1000-2000. Rough safety limits would be to keep the order of a linear- phase FIR filter at/under 600 at 3 kHz accord- ing to both the headphone simulation and the real loudspeaker experiment."
This would seem to be a matter of much concern to somebody who wants consistent SQ either standing or sitting while listening. As for myself, I want to do a great deal better than merely observe what is termed here 'safety limits' in any audio equipment I care to listen to. IAC, this paper's conclusions are a long way from the blanket claim of off-axis 'inaudibility' for linear phase FIR filters that you say that Wilson, Adams and Scott make, and, I believe, much more strongly grounded in fact.
Here's a link to an AES paper from 2008 titled: "Perceptual Study and Auditory Analysis on Digital Crossover Filters" at:
http://www.acoustics.hut.fi/~mak/PUB/AES124-000056.pdf
Among the researchers' conclusions:
"FIR crossovers seem to be *highly susceptible* to off-axis errors with higher filter orders. The flight time difference of only 0.02-0.03 ms be- tween low- and highpass bands at 3 kHz was found to produce audible ringing with high FIR orders of the scale 1000-2000. Rough safety limits would be to keep the order of a linear- phase FIR filter at/under 600 at 3 kHz accord- ing to both the headphone simulation and the real loudspeaker experiment."
This would seem to be a matter of much concern to somebody who wants consistent SQ either standing or sitting while listening. As for myself, I want to do a great deal better than merely observe what is termed here 'safety limits' in any audio equipment I care to listen to. IAC, this paper's conclusions are a long way from the blanket claim of off-axis 'inaudibility' for linear phase FIR filters that you say that Wilson, Adams and Scott make, and, I believe, much more strongly grounded in fact.
Last edited:
That looks like a very good paper. I've only skimed through it but it seems to cover the subject well.
Still, I'm not sure it is at odds with the Adams paper. Adams and Co. were probably using more modest filter than the 600th to 2000th order filters of Korhola and Karjalainen. Their approach was to vary the slope order with no attempt to achieve flat summing. The resultant crossover hole was immediately obvious for lower order slopes. As the crossover order was increased the response hole got narrower until it was deemed inaudible. I doubt that they got to 2000th order. If 600th order has much less impact than 2000th order (per your paper) then it is totally possible that a moderately high slope could both give an inaudible hole without the noticeable ringing of the very high order crossovers. Since the ringing doesn't have to be carefull cancelled on some exact listening axis, then the response is equally good on a range of axies.
Thats my story and I'm sticking with it.
David
Still, I'm not sure it is at odds with the Adams paper. Adams and Co. were probably using more modest filter than the 600th to 2000th order filters of Korhola and Karjalainen. Their approach was to vary the slope order with no attempt to achieve flat summing. The resultant crossover hole was immediately obvious for lower order slopes. As the crossover order was increased the response hole got narrower until it was deemed inaudible. I doubt that they got to 2000th order. If 600th order has much less impact than 2000th order (per your paper) then it is totally possible that a moderately high slope could both give an inaudible hole without the noticeable ringing of the very high order crossovers. Since the ringing doesn't have to be carefull cancelled on some exact listening axis, then the response is equally good on a range of axies.
Thats my story and I'm sticking with it.
David
- Status
- Not open for further replies.