Erling
Whether it is or isn't is not the point. Most of us in this discussion have designs which are "the best in the world" - it's strange how many of those there are!
Actually don't think so - there is imense support for others around here - it's just that usually one can speak from his own limited experience only - most are even aware of this limitation as well...
Michael
Last edited:
Erling,
why would you go (seemingly) backwards from your Mjao (with a dedicated midrange) to CS2 clone (crossing 15" Eminence to a dome tweeter is bound to have problems, power handling, polar response, radiated power, acoustic signature/distortion characteristics around crossover, you name it) ? Apart from a pure academic exercise ? After all, even Emerald Physics themselves abandoned the concept.
Curious,
Bratislav
So this quasi LR3 would have a driver overlap halfway between LR2 and LR4? With nearly symmetric lobing? When you offset the tweeter you are referring to its AC versus the woofer AC? So the tweeter would truly be behind the woofer in a real speaker? If we left the polarity normal is there some place farther ahead we could place tweeter and still keep its impulse in relatively the same time as the woofer, or does the polarity flip delay it 180 degrees with the offset causing another 90 degrees, making the whole thing work?
Asking again in case you missed my question Earl: can I take this to mean you have decided to use the LR family of filters?
Brandon
Yes, I did miss this.
As I have said over and over again I don't obsess over filter names. I do what works acoustically given the wildly variant amplitude and phase of a CD waveguide near cutoff. Electrically the filters are a mess not resembling anything that I have ever seen. Acoustically the summed impulse response is extremely compact with basically only the ringing of the woofers HP. I like to called them Semi-quasi-odd-order-Einsteinium-linear-phase-Gaussian - but that's why I stay away from the "name" thing.
Driver are either in-phase or out-of-phase, there is no middle ground - "phase quadrature" is not a possibility. On some of my systems the drivers are in-phase and on others out-of-phase. As soon as you talk about "on-axis", then you have to specify a field point and things like displacement of the drivers, etc. come in to play. My point is simply that simplifying the problem down to "named" filters with defined relationships like phase (and some peoples absolute ban on "out-of-phase" drivers) is totally missing the point. Waveguides put the source well back of the baffle, and they have significant frequency dependent delays, while the woofer is general on the baffle - is this not all part of the phase problem? My eyes just gloss over with all this talk of phase linearity, and "Quasi" this and that. It might make sense, but its all irrelavent to me.
Earl, I think Brandon (and most of the rest of us with a clue) understands that 'named' crossover types apply to the acoustical response not the electrical transfer function. Anyone with any modern software (LspCAD, SoundEasy, Passive Crossover Designer, Speaker Workshop, etc.) is using real measured magnitude and phase response of the drivers on the baffles and optimizing electrical components to give the desired acoustical magnitude and phase.
Rather than assuming we're all stupid and that we're referring to electrical transfer functions, please do us the courtesy of assuming we know what's going on because we do.
Rather than assuming we're all stupid and that we're referring to electrical transfer functions, please do us the courtesy of assuming we know what's going on because we do.
Last edited:
That's really a mouthfull. I have the same situation here. For some reason I just don't go by the standard format either. 
Earl I understand what you are saying about using non-ideal, and often asymmetric slopes, most of here do that as matter of course. But we use the slopes to accomplish some ideal-typically a large frontal lobe. So we change the actual response of the driver from ideal (whether LR4 or whatever), but in doing so keep the driver phase near some ideal relationship. For example perfectly in phase (like LR filters). At least out to some angle where you just can't overcome effect of the driver rolloff.
I get that. Whether you start with ideal acoustic slopes and then tweak from there to maintain smooth response as far off axis as possible, or you come from the opposite way and jsut keep tweaking slopes (phase relationships) until you get reasonably smooth response in some frontal arc, you MUST be converging on the same ideal realtionship of some kind. Otherwise you would end up with a prefectly flat on axis response that may suddenly peak or dip just 10 degrees off axis, then suddenly better another 10 degrees off, etc. In other words terrible lobing. A crazy wavy mess off axis.
So...What is your ideal? Perfectly in phase drivers on axis with symmetric lobing, that gradually deteriorates off axis? If so that is pretty much the LR relationship, whether or not the actual slopes of the driver responses are textbook LR. At least that is how I look at it. What confuses me is you once said you use BW3, which at the time made sense as I fiugred you used it's lobing behavior to offset the tilt caused by the waveguide in the first place. Were you talking about using a 3rd order electrical network? As you said I don't really care about that, it's the acoustical response that I'm talking about.
I get that. Whether you start with ideal acoustic slopes and then tweak from there to maintain smooth response as far off axis as possible, or you come from the opposite way and jsut keep tweaking slopes (phase relationships) until you get reasonably smooth response in some frontal arc, you MUST be converging on the same ideal realtionship of some kind. Otherwise you would end up with a prefectly flat on axis response that may suddenly peak or dip just 10 degrees off axis, then suddenly better another 10 degrees off, etc. In other words terrible lobing. A crazy wavy mess off axis.
So...What is your ideal? Perfectly in phase drivers on axis with symmetric lobing, that gradually deteriorates off axis? If so that is pretty much the LR relationship, whether or not the actual slopes of the driver responses are textbook LR. At least that is how I look at it. What confuses me is you once said you use BW3, which at the time made sense as I fiugred you used it's lobing behavior to offset the tilt caused by the waveguide in the first place. Were you talking about using a 3rd order electrical network? As you said I don't really care about that, it's the acoustical response that I'm talking about.
Last edited:
I will find it truly interesting when you folks get hold of some EnABL'd drivers to work with. Soongsc can speak quite clearly about how they may help, or harm your activities.
On thing that is quite certain, their audible characteristics, concerning nulls and resonant nodes, is far more benign than the audible characteristics of untreated drivers. A properly treated driver has a smooth and very even audible character, and it is the one you normally only find on axis minus the occasional shrill spikes, across the swept angle of the "wave guide". By this I am referring to the angle of the walls in a cone or the mounting plate for a dome or the edges of of a flat radiator or the swept angle of a horn or the chamber and surface of a CD..
I m not suggesting that you cease where you are headed, but it might be a useful tool, to show you how you can manipulate a driver's match to the surrounding air. With tools designed to work at the interface between driver and air.
Bud
On thing that is quite certain, their audible characteristics, concerning nulls and resonant nodes, is far more benign than the audible characteristics of untreated drivers. A properly treated driver has a smooth and very even audible character, and it is the one you normally only find on axis minus the occasional shrill spikes, across the swept angle of the "wave guide". By this I am referring to the angle of the walls in a cone or the mounting plate for a dome or the edges of of a flat radiator or the swept angle of a horn or the chamber and surface of a CD..
I m not suggesting that you cease where you are headed, but it might be a useful tool, to show you how you can manipulate a driver's match to the surrounding air. With tools designed to work at the interface between driver and air.
Bud
Hello John,
I can see that you didn't give a look to my spreadsheet for a 3 ways system. That's not the common little tool that you can see to calculate crosssovers for cars... It solves completely in the complex domain the calculation of the response of the 3 ways. The influence of each cell of the crossover (LP, BP, HP) on the others is perfectly treated.
As an example give a look to the attached graph which one shows the response of a perfectly aligned 3 ways system using only 6dB/octave crossover at 800Hz and 8000Hz. You'll see that a +0.7dB bump appears in the medium and alos that the group delay shows some variation (small but real. It is equivalent to 7 mm variation of the equivalent distance travelled at the speed of sound). This is due to the influence you mentionned of one cell of the crossover on the other.
Now about your remark about what happens with real loudspeakers. I prefer in that kind of discussion to separate what is related to the crossover itself and what is related to the loudspeakers themselves. IMHO the crossover is the weak element when we pay attention to phase distortion.
That doesn't mean that in my designs I don't take care of the phase non linearity of the loudspeakers themselves. (It is the opposite actually and since long time I developped under Matlab and used my own tools to analyze the response of real loudspeakers including phase and group delay... In the document I mentionned you'll find real measurements on multiways systems...)
But, as my philosophy is influenced by the "KIS" concept (keep it simple), the solutions I used for myself are generally based on loudspeakers (+ load) used in an interval of frequency inside which their own phase non linearity cannot mess the "theorical results" of the Le Cleac'h crossover.
But I know many people who are using my crossovers with less linear loudspeakers, adding phase equalizers, amplitude equalizers... often I helped them to optimize their set-up.
You said yourself that we have to "define a target" and then to bring some modification to the original design in order to reach that goal. That the same here. I define the target as the behaviour of a quasi-optimal crossover then I study the means to reach that goal.
We are here in a thread devoted to Lynn's project. Do you think Lynn will use a complex crossover with phase equalizers added, etc. In a first approach I guess that's Lynn will try the simplest KIS optimal crossover: the 6dB/octave. I hope it will succeed in that but I have some doubts. It is also doubtful that Lynn will appreciate digital crossovers. That's why the KIS philosophy of the Le Cléac'h 3rd order crossover could be interesting. With exactly the same number of passive elements as the classical 3rd order Butterworth it brings audible ameliorations:
- good linearity of the response curve
- good "in phase" operation of the 2 loudspeakers in the interval of frequency around the common cut-off (define at -5dB for the Le Cléac'h crossover)
- quasi constant group delay below 4kHz
For a passive 2 ways crossover it needs only 3 capacitors and 3 selfs.
Best regards from Paris, France
Jean-Michel Le Cléac'h
I can see that you didn't give a look to my spreadsheet for a 3 ways system. That's not the common little tool that you can see to calculate crosssovers for cars... It solves completely in the complex domain the calculation of the response of the 3 ways. The influence of each cell of the crossover (LP, BP, HP) on the others is perfectly treated.
As an example give a look to the attached graph which one shows the response of a perfectly aligned 3 ways system using only 6dB/octave crossover at 800Hz and 8000Hz. You'll see that a +0.7dB bump appears in the medium and alos that the group delay shows some variation (small but real. It is equivalent to 7 mm variation of the equivalent distance travelled at the speed of sound). This is due to the influence you mentionned of one cell of the crossover on the other.
Now about your remark about what happens with real loudspeakers. I prefer in that kind of discussion to separate what is related to the crossover itself and what is related to the loudspeakers themselves. IMHO the crossover is the weak element when we pay attention to phase distortion.
That doesn't mean that in my designs I don't take care of the phase non linearity of the loudspeakers themselves. (It is the opposite actually and since long time I developped under Matlab and used my own tools to analyze the response of real loudspeakers including phase and group delay... In the document I mentionned you'll find real measurements on multiways systems...)
But, as my philosophy is influenced by the "KIS" concept (keep it simple), the solutions I used for myself are generally based on loudspeakers (+ load) used in an interval of frequency inside which their own phase non linearity cannot mess the "theorical results" of the Le Cleac'h crossover.
But I know many people who are using my crossovers with less linear loudspeakers, adding phase equalizers, amplitude equalizers... often I helped them to optimize their set-up.
You said yourself that we have to "define a target" and then to bring some modification to the original design in order to reach that goal. That the same here. I define the target as the behaviour of a quasi-optimal crossover then I study the means to reach that goal.
We are here in a thread devoted to Lynn's project. Do you think Lynn will use a complex crossover with phase equalizers added, etc. In a first approach I guess that's Lynn will try the simplest KIS optimal crossover: the 6dB/octave. I hope it will succeed in that but I have some doubts. It is also doubtful that Lynn will appreciate digital crossovers. That's why the KIS philosophy of the Le Cléac'h 3rd order crossover could be interesting. With exactly the same number of passive elements as the classical 3rd order Butterworth it brings audible ameliorations:
- good linearity of the response curve
- good "in phase" operation of the 2 loudspeakers in the interval of frequency around the common cut-off (define at -5dB for the Le Cléac'h crossover)
- quasi constant group delay below 4kHz
For a passive 2 ways crossover it needs only 3 capacitors and 3 selfs.
Best regards from Paris, France
Jean-Michel Le Cléac'h
Attachments
Hello John,
No this quasi 3rd order LR is not in the list of the 4 quasioptimal crossovers we found until now. I'll need so further work with Francis Brooke in order to see if it can be the 5th...
My crossover don't use a cascaded structure and, please note, that I developped the first quasioptimal Le Cléac'h crossovers in 1982-1983...
But you mentionnned that this Quasi 3rd order LR crossover need phase inverting and a delay equal to 0.25 wavelength at cut-off ferquency. This is quite interesting and it is probably very near of the quasi optimal 3rd order Le Cléac'h crossover which one uses phase inversion of the tweeter too and a delay equivalent to 0,22 wavelength at the common cut-off frequency (f-5dB). Their behaviour is probably very similar...
What you say about an offset performed geometrically is perfectly true and we are aware of this since long time. (It was discussed on French forums). While I consider this as a no-problem for an audiophile listening at the sweet spot, the solution to that is to perform the delay with a delay line.
Best regards from Paris France
Jean-Michel Le Cléac'h
No this quasi 3rd order LR is not in the list of the 4 quasioptimal crossovers we found until now. I'll need so further work with Francis Brooke in order to see if it can be the 5th...
My crossover don't use a cascaded structure and, please note, that I developped the first quasioptimal Le Cléac'h crossovers in 1982-1983...
But you mentionnned that this Quasi 3rd order LR crossover need phase inverting and a delay equal to 0.25 wavelength at cut-off ferquency. This is quite interesting and it is probably very near of the quasi optimal 3rd order Le Cléac'h crossover which one uses phase inversion of the tweeter too and a delay equivalent to 0,22 wavelength at the common cut-off frequency (f-5dB). Their behaviour is probably very similar...
What you say about an offset performed geometrically is perfectly true and we are aware of this since long time. (It was discussed on French forums). While I consider this as a no-problem for an audiophile listening at the sweet spot, the solution to that is to perform the delay with a delay line.
Best regards from Paris France
Jean-Michel Le Cléac'h
Hi there JMLLC, the range of possible passive crossovers ranges from 2nd to 4th order (electrically), with notch filters where necessary. Not interested in 1st order, since the out-of-band excursion control is so poor.
I've been reading the theoretical discussion with some interest. In my experience, drivers are so far from flat they require quite a bit of EQ in the crossover region, so the resulting acoustic filter approaches an ideal form, but never quite gets there (Zeno's Paradox applied to filters). So a real-world system system can approximate, say, a LR4, but is never exactly the same as an ideal textbook filter.
Choosing the appropriate electroacoustic filter system is a complex tradeoff between inter-driver phase spread (and thus polar pattern in the vertical plane), control of off-axis polar deviations, phase, time, and group-delay distortion with respect to the original waveform, and excursion control for the highpass driver. I don't see any "best" topology, given these overlapping (and at times conflicting) requirements.
In principle, 1st order has the least time distortion (with respect to the original waveform), but there is a steep price to be paid in control of out-of-band excursion and complex polar patterns with real-world drivers. I've never found any drivers that were so flat and so forgiving that a 1st-order crossover was the best solution - I've always had to use moderately steeper slopes to get reasonable control of excursion and an overlap region where inter-driver phase angles were reasonably well controlled.
One thing that's been missing from all of this theory is control of out-of-band excursion. This is a serious enough problem for direct-radiator tweeters with their inherent 12 dB/octave increase in excursion per octave, but is worse for horn tweeters. Diaphragm excursion increases at an extremely rapid rate below horn cutoff, so it is a good idea to prevent much power from reaching the HF driver at or below this critical frequency.
The horn cutoff combined with the required electrical filtering results in a very steep overall (summed) transfer function for the HF driver. A modest 1st order electrical filter will allow far too much power into the below-cutoff region, thus degrading headroom and IM distortion unnecessarily. The minimum requirement appears to fall between 2nd and 4th-order electrical filtering, or some kind of notch filter tuned to the Fs of the driver and/or the horn cutoff frequency. In terms of excursion control, the optimum filter might actually be some kind of elliptical filter, with the zero tuned to the Fs of the driver.
If I'm counting up things correctly, the electroacoustic summed response results in a 6th to 8th order highpass filter two octaves or more below the desired crossover, and a more moderate slope closer to the actual crossover. That doesn't seem to fall into any of the classical filter types, but a sort of stagger-pole filter with two sections with different alignments.
Short of a digital FIR filter, I don't see any way to avoid this sort of dual-slope filter for horn drivers, since the horn always adds its own sharp rolloff below the cutoff frequency, and the driver must be electrically protected from out-of-band excursion. Granted, different horn profiles have different behaviors through and below cutoff, but they all have cutoff frequencies, and resulting loss of diaphragm control below those frequencies.
I've been reading the theoretical discussion with some interest. In my experience, drivers are so far from flat they require quite a bit of EQ in the crossover region, so the resulting acoustic filter approaches an ideal form, but never quite gets there (Zeno's Paradox applied to filters). So a real-world system system can approximate, say, a LR4, but is never exactly the same as an ideal textbook filter.
Choosing the appropriate electroacoustic filter system is a complex tradeoff between inter-driver phase spread (and thus polar pattern in the vertical plane), control of off-axis polar deviations, phase, time, and group-delay distortion with respect to the original waveform, and excursion control for the highpass driver. I don't see any "best" topology, given these overlapping (and at times conflicting) requirements.
In principle, 1st order has the least time distortion (with respect to the original waveform), but there is a steep price to be paid in control of out-of-band excursion and complex polar patterns with real-world drivers. I've never found any drivers that were so flat and so forgiving that a 1st-order crossover was the best solution - I've always had to use moderately steeper slopes to get reasonable control of excursion and an overlap region where inter-driver phase angles were reasonably well controlled.
One thing that's been missing from all of this theory is control of out-of-band excursion. This is a serious enough problem for direct-radiator tweeters with their inherent 12 dB/octave increase in excursion per octave, but is worse for horn tweeters. Diaphragm excursion increases at an extremely rapid rate below horn cutoff, so it is a good idea to prevent much power from reaching the HF driver at or below this critical frequency.
The horn cutoff combined with the required electrical filtering results in a very steep overall (summed) transfer function for the HF driver. A modest 1st order electrical filter will allow far too much power into the below-cutoff region, thus degrading headroom and IM distortion unnecessarily. The minimum requirement appears to fall between 2nd and 4th-order electrical filtering, or some kind of notch filter tuned to the Fs of the driver and/or the horn cutoff frequency. In terms of excursion control, the optimum filter might actually be some kind of elliptical filter, with the zero tuned to the Fs of the driver.
If I'm counting up things correctly, the electroacoustic summed response results in a 6th to 8th order highpass filter two octaves or more below the desired crossover, and a more moderate slope closer to the actual crossover. That doesn't seem to fall into any of the classical filter types, but a sort of stagger-pole filter with two sections with different alignments.
Short of a digital FIR filter, I don't see any way to avoid this sort of dual-slope filter for horn drivers, since the horn always adds its own sharp rolloff below the cutoff frequency, and the driver must be electrically protected from out-of-band excursion. Granted, different horn profiles have different behaviors through and below cutoff, but they all have cutoff frequencies, and resulting loss of diaphragm control below those frequencies.
Last edited:
Hello Lynn,
You'll see on that link pictures of a 4 ways system, 3 ways of which are using Le Cléac'h horn (the 4th being a self powered subwoofer operating below 80Hz):
http://www.musique-concrete.com/MC/ENGalery.html
This is the "Grande Castine" designed by my friend Marco Henry.
The 3 horns are fed through 1st order passive crossovers... the system possess a 112dB/1W/2,83V efficiency and it can be used with a small power amplifier (triode SE preferred).
(If you come to ETF 2009 in France then I can arrange a listening session...)
Best regards from Paris, France
Jean-Michel Le Cléac'h
You'll see on that link pictures of a 4 ways system, 3 ways of which are using Le Cléac'h horn (the 4th being a self powered subwoofer operating below 80Hz):
http://www.musique-concrete.com/MC/ENGalery.html
This is the "Grande Castine" designed by my friend Marco Henry.
The 3 horns are fed through 1st order passive crossovers... the system possess a 112dB/1W/2,83V efficiency and it can be used with a small power amplifier (triode SE preferred).
(If you come to ETF 2009 in France then I can arrange a listening session...)
Best regards from Paris, France
Jean-Michel Le Cléac'h
Last edited:
Well, I have to admit the dynamic range of the Grand Castine looks nothing short of stupendous - the filter slope would hardly matter in a domestic context. It certainly makes the ETF tempting!
I guess I don't know what you mean by driver over lap. If you mean the slope of the roll offs, yes. AC offset, yes, if the acoustic transfer functions perfectly match the targets. In reality the offset would probably need to be adjusted to bring the phase into alignment at the crossover point. Tweeter behind woofer, yes. As I mentioned in one of my posts, this then will cause problems with the alignment when moving off axis to the left or right which is another reason I don't like the idea of this type of "time alignment". Normal polarity, move the tweeter 0.25 wave lengths forward instead of back. You will have the pretty much the same amplitude but this defeats the purpose if some ad hoc type of time alignment is the goal. You comment about the polarity flip is correct.
If the goal is odd order with symmetric polar response the problem is trivial. Just cascade any even order Butterworth response with a 1st order. The response will be -6db at the crossover point and phase will be either 90 or 270 depending on order. Set the correct HP polarity and the offset for flatest response is 0.25 wavelengths. But, if you want to quasi time align it won't work. The problem is that the offset of the tweeter must correspond to, or close to, the DC group delay of the LP filter. In the 3rd order case based on the above construction at 1k Hz the DC GD is about 0.38 msec and the offset is 0.25. In the Brooke case the DC GD is 0.3777 msec and his specified offset is 0.21 msec. But as the order of the LP increases the DC GD also increases. So the offset has to increase. It just doesn't work out. I looked at it some time ago.
I'm not saying the filter is cascaded. I just cascade the odd order Butterworth transfer function with a 1st order transfer function to obtain the desired transfer function which represents the acoustic target. The passive electrical topology to obtain the ideal filter response would be the typical odd order passive topology. What would be needed when combined with a driver's native response is another issue all together, and is why I can not separate out driver and filter response. The passive elements I gave previously were only to allow other to set up a filter and see what it does using a typical 3rd order topology.
Yep.
Dennis, I am not assuming that everybody is stupid or that they don't understand anything, I just don't fit my crossovers, electrical or acoustical into any named configuration that I know of, so there is no answer to the question.