The Kitchen Is On Fire, Baby!

[kkantor]

Guys, it's staring you all right in the face. Please, just try and take one more conceptual baby-step, and I think we might actually agree. Here is something that I believe is very, very relevant to several of the questions being debated about how to measure speakers to achieve good subjective correlation. Please, take at least a few minutes to consider what I am saying, roll it around on your tongue, before jumping in with more of the same old pissing match.

Remember, all these measurements, Allison's, Toole's, Beck's, Geddes', anyone's, are M-O-D-E-L-S of system behavior. They are NOT system transfer functions!!!!!!! ANY single type of measurement signal set, including windowed functions, can at best approximate the perceptual response to a given piece of music.

Take just one example: Music has both sharp transients, and legato, steady-state elements. Even at a fixed point in space, even in an anechoic chamber, the signal envelope of music or speech has modulation time-constants that span a wide range. Some musical events are much shorter than typical early room reflections, some are much longer than typical listening room decay times.

Why should it be that any one model is "best?" Even if it is statistically most often correlated somehow with listener preference, that DOES NOT MEAN that it is necessarily the measurement model that best quantifies the reproduction of each and every piece of music, or recording method. Forget EQ and frequency balance here, they are trivial. I am talking about integration times and the modulation transfer function of a given segment of audible information. By definition, there simply cannot be one perfect measurement technique. If there was, certainly, there would be fewer design approaches on the market.

Sorry for all the caps and underlines. I feel very strongly that this issue is both crucial and very often completely misunderstood.

-k

PS- for advanced readers: the above hints at the same dimensional transformation problem I referred to last week. Ultimately, its about transform reversibility and information entropy. It can be mathematically shown that there exists an infinite set of distinct and different transducers which can satisfy any given criteria of accuracy, if the number of channels is finite. Thus, there cannot be either one perfect measurement, or one perfect loudspeaker response.

[soundminded]

Guys, it's staring you all right in the face. Please, just try and take one more conceptual baby-step, and I think we might actually agree. Here is something that I believe is very, very relevant to several of the questions being debated about how to measure speakers to achieve good subjective correlation. Please, take at least a few minutes to consider what I am saying, roll it around on your tongue, before jumping in with more of the same old pissing match.

Remember, all these measurements, Allison's, Toole's, Beck's, Geddes', anyone's, are M-O-D-E-L-S of system behavior. They are NOT system transfer functions!!!!!!! ANY single type of measurement signal set, including windowed functions, can at best approximate the perceptual response to a given piece of music.

Take just one example: Music has both sharp transients, and legato, steady-state elements. Even at a fixed point in space, even in an anechoic chamber, the signal envelope of music or speech has modulation time-constants that span a wide range. Some musical events are much shorter than typical early room reflections, some are much longer than typical listening room decay times.

Why should it be that any one model is "best?" Even if it is statistically most often correlated somehow with listener preference, that DOES NOT MEAN that it is necessarily the measurement model that best quantifies the reproduction of each and every piece of music, or recording method. Forget EQ and frequency balance here, they are trivial. I am talking about integration times and the modulation transfer function of a given segment of audible information. By definition, there simply cannot be one perfect measurement technique. If there was, certainly, there would be fewer design approaches on the market.

Sorry for all the caps and underlines. I feel very strongly that this issue is both crucial and very often completely misunderstood.

-k

PS- for advanced readers: the above hints at the same dimensional transformation problem I referred to last week. Ultimately, its about transform reversibility and information entropy. It can be mathematically shown that there exists an infinite set of distinct and different transducers which can satisfy any given criteria of accuracy, if the number of channels is finite. Thus, there cannot be either one perfect measurement, or one perfect loudspeaker response.

"Why should it be that any one model is "best?" "

Because if it accurately predicts or describes the relationships between sound fields generated at one point in space and the resultant sound fields at others, it is correct where the others aren't. I heard Peter Snell say around 1982 that what is important is the sound field that reaches your ears. I came to that conclusion 8 years earlier.

"By definition, there simply cannot be one perfect measurement technique."

Of all the postings you have ever written here, that is by far the most outrageous. Even if it can't be perfect since no measurement of anything can be perfect in the ultimate sense of the meaning of perfect, it can be a lot better than we have now. And it can be. But first you need a much better model that will tell you what to measure that you are missing. If MIT didn't sell you the tools you need to find one for all the money they charged, don't expect me to give it to you for free.

[Shacky]

Cool stuff Ken. Would it be inappropriate to ask what some of your favorite setups are?

[soundminded]

Cool stuff Ken. Would it be inappropriate to ask what some of your favorite setups are?

I'll just bet it has to do with a speaker whose positioning is so critical, the designer had to go to Stereophile Magazine's facilities to adjust its exact spot and orientation with a micrometer before he felt it could be properly tested.

[kkantor]

Cool stuff Ken. Would it be inappropriate to ask what some of your favorite setups are?

Shacky,

I try never to endorse (or criticize) any particular products here. Also, what my favorite setups are is not related to the point I am trying to make here.

I've held back on bring up what I see as the defining issue as this discussion has progressed, but I would really like to address it now. Do you really understand what I wrote? It is not trivial, and not gobbledygook either. I brought up a very specific issue, and I illustrated it with a concrete example. I'm happy to debate my position, but not to water it, or the thread, down.

-k

[kkantor]

Because if it accurately predicts or describes the relationships between sound fields generated at one point in space and the resultant sound fields at others, it is correct where the others aren't.

You clearly don't understand what I wrote, since I am stating directly that such a "prediction or description" is mathematically not possible to make without ambiguity. (Not the ambiguity of measurement noise, or quantum uncertainty, or error... rather, inherent degrees of freedom.) Let me state it again: there are an infinite number of different loudspeaker responses that will produce >identical< results on any measurement which is based on the spatial sampling of the listening room. Period. Such measurements are attempts to parametrize and populate models of the system response. They are not the system response itself. One cannot define, measure or theorize their way out of this, no matter what. It's exactly like trying to extrapolate what is going on at 10KHz on a system sampled at 1 KHz.

I have now given you two clear examples:

1- The unknown and changing relationship between the target signal's modulation transfer function to that of the measurement characterization signal(s).

2- Spatial aliasing due to insufficient sampling in both the recording and the reconstruction.

There are many others I could introduce.

If you can refute these points adequately, and on their stated terms, you will surely go down in history. (Well, of course, I mean, more even than already.) Why, I will personally sponsor your landmark paper at the next International AES. However, if you prefer to use this venue to try and insult me, I suggest you start with my height or my adolescent sexual insecurities, since I am much more sensitive to these matters than I am about my education, accomplishments or intellect. As it stands, you only got one tear out of me.

-k

BTW- for the benefit of interested parties... the approach you cite is known as the "Transauralization" model. It was seriously introduced in, if I remember correctly, the 1930's. Tomes have been written about its strengths and weaknesses as both a model and a practical technique. Its latest incarnation is generally referred to as "Wave Field Synthesis." It certainly has its applications. Home hifi is probably not one of them.

http://en.wikipedia.org/wiki/Wave_field_synthesis

[Zilch]

Was that not the intent of the AR reverberant chamber (which you might describe, if you will, to dispel any erroneous perceptions that it was a "normal" room with glass-equivalent walls), to integrate the total energy output into a single representation of the response as shown in Allison and Berkovitz (1970)?

Useful, perhaps, for the purpose of manipulating that response (which you suggest is trivial), but clearly unreliable as a direct characterization of behavior in real rooms. If we consider the integrated (well, averaged, actually) power response over multiple locations in multiple "typical" rooms as at least indicative, if not representative, of that behavior, I believe I have demonstrated that conventional anechoic measures are more predictive above the transition frequency, which does tell us something about what is actually occurring within the space.

Most researchers appear to agree that no single measurement can accurately characterize loudspeaker performance, and what we hear is even more uncertain, thus we have heuristically derived "metrics," most proprietary, comprising multiple "snapshots" of measurable behavior, which are claimed to correlate with listener preferences. It's not cause and effect, and not definitive, but purportedly effective, nonetheless, in discriminating what works well from that which doesn't.

As soon as you toss in production variables as well, the degree of uncertainty goes up, and it's not clear those are any better controlled than they ever were, particularly given the expanded "artistic license" currently practiced. We need a metric for that, as well, though the fundamentals are already evident. In simple terms of defining loudspeaker accuracy as level of divergence from input = output, however, transfer function may have greater utility than what we presently do, but from what I've seen of it, a far more precise algorithm than me watching it is required to render the data meaningful.

Are we talking the same language here?

["Rendition area" -- I like it.... ]

[kkantor]

The lines are drawn in the sand, plain as day:

On one side, Howard, Pete and I are opposed to aliasing.

Everyone else, including you, is aliasing.

1- The reverb chamber I am familiar with was at AR's Houghton Regis facility. It was a large, coffin-like chamber that sat on the floor. I think it was concrete. It had maybe 10 or 12 sides, and was designed to reduce standing waves and randomize the energy inside as much as possible. I never saw it used in actual development. I think it's days were over by the late 70's, other than for specialized purposes.

[i didn't mean "trivial" in the engineering sense. I mean, trivial to understand and deal with conceptually, especially by now.]

2- Every loudspeaker designer, most definitely including myself, selects a portfolio of objective measurements that they believe will help them meet their goals. Most often, there is a subset of tests that are always used, and dozens-to-hundreds of other measurements that are improvised during the development of a product to try and understand what is being heard.

The last thing I would do is argue a relativistic position, that all approaches are valid. Some are more valid than others. But, what has happened is that best-in-class manufacturers can now achieve pretty much what they want to, if cost isn't as object. But, has the veil lifted??? No. Glimpses of perfection are not much more common than they were 40 years ago. Where do we go from here? Give up? Say, this is as good as it gets? Of course not. Many people are trying to move understanding forward, yourself included. I just don't think there is much fertile ground yet to plow, without taking the next steps in understanding. Forward motion, not arguments for or against any particular technique of the past. If any of these techniques were going to lift the veil, they would have by now!

Please, take a look at the following, if only just to see the figure 3-1. (It's big, but it is worth it.)

http://www.deutsche-telekom-laboratories.d..._WFS_Theory.pdf

How far have we gotten by debating this same stuff, decade after decade? Clearly we are missing something. I think, three things:

- anything even approximating a standard recording methodology.

- a fundamentally correct model of what constitutes objectively accurate reproduction; ie- a model that is not mathematically flawed.

- a knowledge of human hearing that is complete enough to allow a sensible approach to the data reduction inherent in any form of reproduction.

-k

Was that not the intent of the AR reverberant chamber (which you might describe, if you will, to dispel any erroneous perceptions that it was a "normal" room with glass-equivalent walls), to integrate the total energy output into a single representation of the response as shown in Allison and Berkovitz (1970)?

Useful, perhaps, for the purpose of manipulating that response (which you suggest is trivial), but clearly unreliable as a direct characterization of behavior in real rooms. If we consider the integrated (well, averaged, actually) power response over multiple locations in multiple "typical" rooms as indicative, if not representative, of that behavior, I believe I have demonstrated that conventional anechoic measures are more predictive above the transition frequency, which does tell us something about what is actually occurring within the space.

Most researchers appear to agree that no single measurement can accurately characterize loudspeaker performance, and what we hear is even more uncertain, thus we have heuristically derived "metrics," most proprietary, comprising multiple "snapshots" of measurable behavior, which are claimed to correlate with listener preferences. It's not cause and effect, and not definitive, but purportedly effective, nonetheless, in discriminating what works well from that which doesn't.

As soon as you toss in production variables as well, the degree of uncertainty goes up, and it's not clear those are any better controlled than they ever were, particularly given the expanded "artistic license" currently practiced. We need a metric for that, as well, though the fundamentals are already evident. In simple terms of defining loudspeaker accuracy as level of divergence from input = output, however, transfer function may have greater utility than what we presently do, but from what I've seen of it, a far more precise algorithm than me watching it is required to render it meaningful.

Are we talking the same language here?

[Shacky]

Shacky,

I try never to endorse (or criticize) any particular products here. Also, what my favorite setups are is not related to the point I am trying to make here.

I've held back on bring up what I see as the defining issue as this discussion has progressed, but I would really like to address it now. Do you really understand what I wrote? It is not trivial, and not gobbledygook either. I brought up a very specific issue, and I illustrated it with a concrete example. I'm happy to debate my position, but not to water it, or the thread, down.

-k

Ken,

I meant no disrespect. I just love good (geat would be better) sounding stereo. I'm a fan of NHT and appreciated fact that you were one of the few using acoustic suspension. I'm thoroughly enjoying Peter's Eico HF-81 - it's a gem though not with my 3a's. My JBL L36 and AR 5's (now given to my son) are a better match. So I'm always looking for stereo targets.

I don't pretend to understand all that you said. I am not a student of this subject. That said the logic of what you map out is something even a novice like me can appreciate. I'm in this soley for the enjoyment of quality sounding reproduced music. Learning bits and pieces of theory from a master like you is cream on the top!

I respect your position and will refrain from "watering down" your thread any further.

[speaker dave]

Hi Ken,

Thanks for starting a new "root of the matter" thread. This is certainly the topic we have been debating for a few weeks.

I think the fundamental debate boils down to the general (ideal) form of the frequency response of a neutral loudspeaker, one which adds no color to the sound. Arguments have been made for flat power response, flat anechoic axial response, flat in-room curves, rolled off room curves (matching concert hall measurements) and flat time windowed response. Also wide dispersion vs. constant directivity.

There was a great paper by Lipshitz and Vanderkoy where they did a simple experiment to explore just these questions. They put a KEF 104.2 ( a conventional forward firing loudspeaker) underneath an ESL 63. The ESL 63 is of course a dipole with two strongly defined nulls. It was mounted 90 degrees to the normal orrientation, such that a listener would sit in its null and hence any signal fed into it would contribute to the reverberent field of the room but not the direct sound. With this setup they could independently manipulate the direct field and the reverberent field of a loudspeaker in a room. They tried a number of scenarios and came to the following conclusions (somewhat paraphrased and simplified).

1) Flat sound power (near flat reverberent field, depending on the room constants) achieved by making the direct sound (the 104.2) bright, sounded excruciatingly bright.

2) Flat sound power by adding high frequency energy into the ESL 63 (the 104.2 stays "flat") also sound bright, but not quite as bad as above.

3) Adding some, but lesser amounts, of energy to the reverberent field could add some pleasent (acceptable) brightness or spaciousness.

4) If the sound power added to the reverberent field had holes in it's spectrum, this was nearly unnoticable.

5) If the sound power added to the reverberent field had peaks, this was noticable and undesirable.

I think these tests answer most of our questions. A shortcoming, for Howard's sake, is that they didn't pursue any responses tailored to mimic a concert hall reverberent field. Their starting assumption was that some measurement should be flat rather than contoured. So the baseline was a flat (anechoic, axial) 104.2 and they added from there.

From these experiments I would jump to a few conclusions: Flat power response is not what we want, especially if we achieve it by messing up the axial response. Wide range constant directivity isn't desirable either, because it would force us to take flat power response with flat axial response. However a smooth but rising directivity is a good thing. Peaks in the reverberent field are a bad thing but holes don't seem to matter.

Since the room curve is strongly influenced by the reverberent field it becomes an unreliable indicator of response. Remember that holes in the power response, hence in the room curve, are benign. We also know that the "ideal room curve" changes totally with the size of the room. Large theaters only sound good if the "house curve" (a far field measure dominated by the reverberent field, a combination of power response and rising room constant) rolls off dramatically.

I believe these various facts can only be reconciled if we accept that the ear has the ability to hear through to the direct sound (first arrival) and ignore the spectrum of the later sound. Some serious papers have been written that take that a step further and define a variable integrating time window that captures the direct sound only at HF and opens up to effectively capture the total room response at LF. (See Salmi, Kates and Bech) I am strongly convinced that this is the answer.

It also ties in with what Toole would say: That a good loudspeaker is smooth and flat in its axial anechoic response and smooth (not necessarily flat) in its off axis response. (Very paraphrased!)

Regarding how directional, or how wide a dispersion is correct, I think it is hard to argue any absolutes here. I am perfectly willing to let Howard like a very wide dispersion diffuse sound and let Zilch like a narrower, constant directivity derived, focused sound. With only two channels a lot of researchers conclude that room reflections, even though our small rooms aren't the concert halls, add some to the naturalness of reproduction (sound, that is). (A thought: what if a speaker had a means of sending delayed side wall reflections. Wouldn't that be magic?) Griesinger of Lexicon makes a comment about hearing a 5 channel system in a very dead room with an average absorbtion of 0.8 (near anechoic): "Ordinary stereo in this room was poor, but with side and rear loudspeakers and hall synthesis the sound was wonderful..." So with 2 channels we can use some help from the room (wide dispersion), with more channels we are probably better off without it.

I don't expect this little lecture to change anybodies strongly entrenched viewpoints on the subject. I've been watching and contributing to the debate for a couple of weeks. It doesn't seem to me that anyone hears or cares about any other viewpoint (sure, me included). It is a facinating topic and (Soundminded) not an indicator of our ignorance of the subject but more of the complexity of the matters of human perception. I would encourage everybody to search out some of these various papers. The answers are out there, it is just a matter of reading the literature and coming to a conclusion that seems to fit our observations of the (sound reproduction) world.

A last note on Allison and Vilchur. I think they made a great contribution to our understanding of how the low frequency system interfaces with the room. I think Allison came up with a practical approach (woofer near boundary, midrange far) that gets around the real issues they documented. We owe them for that. The thought that a speaker should mimic the balance of a concert hall isn't (shouldn't be) a major part of their legacy and is more likely a notion forced by practical issues at the time, such as low tweeter sensitivity and inadequate program material.

Finally Ken, I am aware of MTF as roughly the optical equivalent to frequency response. What do you mean by it in acoustics? Ditto alliasing? (Just curious)

Regards,

David

[soundminded]

You clearly don't understand what I wrote, since I am stating directly that such a "prediction or description" is mathematically not possible to make without ambiguity. (Not the ambiguity of measurement noise, or quantum uncertainty, or error... rather, inherent degrees of freedom.) Let me state it again: there are an infinite number of different loudspeaker responses that will produce >identical< results on any measurement which is based on the spatial sampling of the listening room. Period. Such measurements are attempts to parametrize and populate models of the system response. They are not the system response itself. One cannot define, measure or theorize their way out of this, no matter what. It's exactly like trying to extrapolate what is going on at 10KHz on a system sampled at 1 KHz.

I have now given you two clear examples:

1- The unknown and changing relationship between the target signal's modulation transfer function to that of the measurement characterization signal(s).

2- Spatial aliasing due to insufficient sampling in both the recording and the reconstruction.

There are many others I could introduce.

If you can refute these points adequately, and on their stated terms, you will surely go down in history. (Well, of course, I mean, more even than already.) Why, I will personally sponsor your landmark paper at the next International AES. However, if you prefer to use this venue to try and insult me, I suggest you start with my height or my adolescent sexual insecurities, since I am much more sensitive to these matters than I am about my education, accomplishments or intellect. As it stands, you only got one tear out of me.

-k

BTW- for the benefit of interested parties... the approach you cite is known as the "Transauralization" model. It was seriously introduced in, if I remember correctly, the 1930's. Tomes have been written about its strengths and weaknesses as both a model and a practical technique. Its latest incarnation is generally referred to as "Wave Field Synthesis." It certainly has its applications. Home hifi is probably not one of them.

http://en.wikipedia.org/wiki/Wave_field_synthesis

"I have now given you two clear examples:

1- The unknown and changing relationship between the target signal's modulation transfer function to that of the measurement characterization signal(s). "

If your statement equates to the Acoustic Energy Field Transfer Function between two points under identical circumstances is not a fixed relationship, it is absurd. If that were true, the same acoustic field propagated at the same source point in space on one occurrance would yield a different result at the same test point on another occurrance. Since I count twenty-six variables in my equations that affect this function, where a change in any one of them (someone in the audience moving one seat over alters the boundary values for example) it is difficult to duplicate these events unless they are done in immediate succession. Fortunately it is not true because if it was, we might as well all pack up and go home, there would be no point in even trying to understand what is happening since the universe would not be rational, its laws changing constantly and capriciously over time.

"2- Spatial aliasing due to insufficient sampling in both the recording and the reconstruction."

I am not exactly sure what you are talking about here. I am only familiar with aliasing of electrical signals such as the aliasing of two rf signals on adjacent carrier frequences in receivers with insufficient selectivity in their IF amplifier stages where they cross modulate each other. If you are referring to the sampling rate of digital signals in RBCD, that is a different debate altogether. In that case it is easy to prove that RBCD is more than adequate for recording any type of music, its time base resolution at least 10% greater than the best hearing capability and its amplitude resolution twenty times greater than even John Atkinson's claim, 100 times greater than my claimed hearing accuity, and 200 times greater than audiologists claim is typical for listeners with unimpaired hearing. Its dynamic range and frequency range also adequate for all musical signals of all known music. If you are talking about acoustical interference patterns of resultant sound fields from different sources modulating each other, that is a characteristic of live music as well and it is the failure to reproduce it which is the distortion. As I have posted elsewhere, I have rejected all current methods of measurement of acoustic fields in this regard because the are based on presumed models that are inadequate to describe the fields in their entirety, the components of fields, and the relationships between them. Perhaps you could explain exactly what you mean as this is my hobby, not my profession and I am not familiar with the definition of this term in this context.

I am not interested in insulting you personally but there are times when you seem to be deliberately obtuse, your tenacity to your flawed arguments uncharacteristic of a well trained electrical engineer. Perhaps one problem is that acoustics is not a specialty of electrical engineering but of mechanical engineering. It took me a long time to realize that. Perhaps that in part also explains why progress has been so slow compared to progress in most areas of electrical engineering.

I will admit that I resented your reference to your academic and professional credentials in support of one of your arguments as they had no bearing whatsoever on the validity of your argument. I felt it was unworthy of someone who asserts himself to be a trained professional and who should be able to defend his ideas on their merits alone. Nor was I intimidated by you which seemed to have been your intent because both that argument holds no currency with me and because my own educational training and professional experience are every bit the equal of yours and maybe then some.

[Pete B]

Hi Ken,

Thanks for starting a new "root of the matter" thread. This is certainly the topic we have been debating for a few weeks.

I think the fundamental debate boils down to the general (ideal) form of the frequency response of a neutral loudspeaker, one which adds no color to the sound. Arguments have been made for flat power response, flat anechoic axial response, flat in-room curves, rolled off room curves (matching concert hall measurements) and flat time windowed response. Also wide dispersion vs. constant directivity.

There was a great paper by Lipshitz and Vanderkoy where they did a simple experiment to explore just these questions. They put a KEF 104.2 ( a conventional forward firing loudspeaker) underneath an ESL 63. The ESL 63 is of course a dipole with two strongly defined nulls. It was mounted 90 degrees to the normal orrientation, such that a listener would sit in its null and hence any signal fed into it would contribute to the reverberent field of the room but not the direct sound. With this setup they could independently manipulate the direct field and the reverberent field of a loudspeaker in a room. They tried a number of scenarios and came to the following conclusions (somewhat paraphrased and simplified).

1) Flat sound power (near flat reverberent field, depending on the room constants) achieved by making the direct sound (the 104.2) bright, sounded excruciatingly bright.

2) Flat sound power by adding high frequency energy into the ESL 63 (the 104.2 stays "flat") also sound bright, but not quite as bad as above.

3) Adding some, but lesser amounts, of energy to the reverberent field could add some pleasent (acceptable) brightness or spaciousness.

4) If the sound power added to the reverberent field had holes in it's spectrum, this was nearly unnoticable.

5) If the sound power added to the reverberent field had peaks, this was noticable and undesirable.

I think these tests answer most of our questions. A shortcoming, for Howard's sake, is that they didn't pursue any responses tailored to mimic a concert hall reverberent field. Their starting assumption was that some measurement should be flat rather than contoured. So the baseline was a flat (anechoic, axial) 104.2 and they added from there.

From these experiments I would jump to a few conclusions: Flat power response is not what we want, especially if we achieve it by messing up the axial response. Wide range constant directivity isn't desirable either, because it would force us to take flat power response with flat axial response. However a smooth but rising directivity is a good thing. Peaks in the reverberent field are a bad thing but holes don't seem to matter.

Since the room curve is strongly influenced by the reverberent field it becomes an unreliable indicator of response. Remember that holes in the power response, hence in the room curve, are benign. We also know that the "ideal room curve" changes totally with the size of the room. Large theaters only sound good if the "house curve" (a far field measure dominated by the reverberent field, a combination of power response and rising room constant) rolls off dramatically.

I believe these various facts can only be reconciled if we accept that the ear has the ability to hear through to the direct sound (first arrival) and ignore the spectrum of the later sound. Some serious papers have been written that take that a step further and define a variable integrating time window that captures the direct sound only at HF and opens up to effectively capture the total room response at LF. (See Salmi, Kates and Bech) I am strongly convinced that this is the answer.

It also ties in with what Toole would say: That a good loudspeaker is smooth and flat in its axial anechoic response and smooth (not necessarily flat) in its off axis response. (Very paraphrased!)

Regarding how directional, or how wide a dispersion is correct, I think it is hard to argue any absolutes here. I am perfectly willing to let Howard like a very wide dispersion diffuse sound and let Zilch like a narrower, constant directivity derived, focused sound. With only two channels a lot of researchers conclude that room reflections, even though our small rooms aren't the concert halls, add some to the naturalness of reproduction (sound, that is). (A thought: what if a speaker had a means of sending delayed side wall reflections. Wouldn't that be magic?) Griesinger of Lexicon makes a comment about hearing a 5 channel system in a very dead room with an average absorbtion of 0.8 (near anechoic): "Ordinary stereo in this room was poor, but with side and rear loudspeakers and hall synthesis the sound was wonderful..." So with 2 channels we can use some help from the room (wide dispersion), with more channels we are probably better off without it.

I don't expect this little lecture to change anybodies strongly entrenched viewpoints on the subject. I've been watching and contributing to the debate for a couple of weeks. It doesn't seem to me that anyone hears or cares about any other viewpoint (sure, me included). It is a facinating topic and (Soundminded) not an indicator of our ignorance of the subject but more of the complexity of the matters of human perception. I would encourage everybody to search out some of these various papers. The answers are out there, it is just a matter of reading the literature and coming to a conclusion that seems to fit our observations of the (sound reproduction) world.

A last note on Allison and Vilchur. I think they made a great contribution to our understanding of how the low frequency system interfaces with the room. I think Allison came up with a practical approach (woofer near boundary, midrange far) that gets around the real issues they documented. We owe them for that. The thought that a speaker should mimic the balance of a concert hall isn't (shouldn't be) a major part of their legacy and is more likely a notion forced by practical issues at the time, such as low tweeter sensitivity and inadequate program material.

Finally Ken, I am aware of MTF as roughly the optical equivalent to frequency response. What do you mean by it in acoustics? Ditto alliasing? (Just curious)

Regards,

David

Excellent post David.

There should be a #6 in my opinion:

6. Non-flat on axis SPL, peaks and dips, are audible when of sufficient amplitude.

Let me offer my understanding of Ken's use of aliasing. Aliasing in a sampled system

irreversibly damages the signal due to spectral overlap. Typical speakers, in typical home rooms irreversibly

damage the signal due to early reflections that cause overlap of acoustical information in the time domain.

All real rooms are reflective to some degree, and all practial speakers provide off axis energy and therefore

the signal will always be damaged in home listening situations. This should not be interpreted to suggest that

anechoic is correct or a solution either.

[speaker dave]

Excellent post David.

There should be a #6 in my opinion:

6. Non-flat on axis SPL, peaks and dips, are audible when of sufficient amplitude.

Let me offer my understanding of Ken's use of aliasing. Aliasing in a sampled system

irreversibly damages the signal due to spectral overlap. Typical speakers, in typical home rooms irreversibly

damage the signal due to early reflections that cause overlap of acoustical information in the time domain.

All real rooms are reflective to some degree, and all practial speakers provide off axis energy and therefore

the signal will always be damaged in home listening situations. This should not be interpreted to suggest that

anechoic is correct or a solution either.

Thanks Pete B, we should add your #6 for sure (yet not take it for granted that others would assume it to be true).

I get your explanation of aliasing, more figurative than literal.

Thanks,

David

[soundminded]

Excellent post David.

There should be a #6 in my opinion:

6. Non-flat on axis SPL, peaks and dips, are audible when of sufficient amplitude.

Let me offer my understanding of Ken's use of aliasing. Aliasing in a sampled system

irreversibly damages the signal due to spectral overlap. Typical speakers, in typical home rooms irreversibly

damage the signal due to early reflections that cause overlap of acoustical information in the time domain.

All real rooms are reflective to some degree, and all practial speakers provide off axis energy and therefore

the signal will always be damaged in home listening situations. This should not be interpreted to suggest that

anechoic is correct or a solution either.

"Let me offer my understanding of Ken's use of aliasing. Aliasing in a sampled system

irreversibly damages the signal due to spectral overlap. Typical speakers, in typical home rooms irreversibly

damage the signal due to early reflections that cause overlap of acoustical information in the time domain."

My measurement method solves that problem. I think it's one of the cleverest ideas I've ever had. Had it not been for that insight, the rest of what I invented would have been useless because there would have been no way to determine parameters for a real case. Between the time I developed my mathematical model and the time this method occurred to me, I had little hope that it would ever amount to anything. It was literally a case of do or die.

[Zilch]

With only two channels a lot of researchers conclude that room reflections, even though our small rooms aren't the concert halls, add some to the naturalness of reproduction (sound, that is). (A thought: what if a speaker had a means of sending delayed side wall reflections. Wouldn't that be magic?)

There's the rub, of course, as also suggested by Pete B: while early proximal sidewall reflections contribute to ASW spaciousness, with their short delays, they also interfere with the direct axial response and smear the image, which is why I argue against uncontrolled wide dispersion. While it seems that we are able to "hear through" anomalies such as combing (Toole offers an excellent thesis as to how), there's no ignoring or rationalizing away the impact of early lateral reflections upon both the measured and perceived response, and I believe most everyone agrees they are adverse. So, how can we enhance spaciousness without these consequences? Toole and Geddes point to a solution: attenuate the early near wall reflection with more moderate constant directivity and emphasize the low IACC delayed first lateral reflection from the OPPOSITE sidewall. Read up here:

http://www.gedlee.com/downloads/Cum%20laude.pdf

Toole provides an excellent graphic presentation to assist in understanding these concepts in his book, and I have suggested EconoWave as an economical platform for empirically exploring them, in everyone's own laboratories and listening spaces.

http://www.audiokarma.org/forums/showthread.php?t=150939

Geddes measured and published the waveguide polars, so the directivity is well defined with 7.5° resolution out to 90°, 180° beamwidth, horizontal, vertical, and oblique. The balance between direct vs. reflected sources and their effect upon ASW are easily adjusted by varying the toe-in angles and distances to sidewalls, with constant directivity maintaining the spectral balance essentially constant:

http://www.classicspeakerpages.net/IP.Boar...ost&id=4574

[kkantor]

Ken,

I meant no disrespect. I just love good (geat would be better) sounding stereo. I'm a fan of NHT and appreciated fact that you were one of the few using acoustic suspension. I'm thoroughly enjoying Peter's Eico HF-81 - it's a gem though not with my 3a's. My JBL L36 and AR 5's (now given to my son) are a better match. So I'm always looking for stereo targets.

I don't pretend to understand all that you said. I am not a student of this subject. That said the logic of what you map out is something even a novice like me can appreciate. I'm in this soley for the enjoyment of quality sounding reproduced music. Learning bits and pieces of theory from a master like you is cream on the top!

I respect your position and will refrain from "watering down" your thread any further.

No disrespect taken at all! Just explaining why I didn't answer. I often have to hold my tongue on the internets, when it comes to product opinions and choices. (I'm asked almost daily.) For various reasons, I try to concentrate on general issues that do not promote or devalue any manufacturer. Over the years, I have owned and appreciated a very wide variety of speakers, from various coasts and continents. At the moment my professional focus is on guitar amps, so I don't have a permanent reference system set up. I find that I can't do much of anything else when I am listening to music I love, which generally means late nights, laptop > headphones.

-k

[kkantor]

Makes sense to me. This is one reason that while I do (well, did, since I am retired from the business) place a lot of stock in the moving-microphone technique I use to initially check out the room-curve smoothness of the speakers I was reviewing for The Sensible Sound, I did not place those curves on a platform and declare them the reference standard for performance. Indeed, when I first started reviewing I did not print the curves I ran at all, and kind of did a decription thing like what Julian Hirsch did years ago. While some people felt that Hirsch was doing a copout (and I was one of them), when I started to do the work myself I understood his reasoning.

Later on, I decided to print the curves, but only after I published a special commentary article (actually, two of them, because the article was so long that it had to be split in two) that included over a dozen curves of speakers I had auditioned that measured just OK, good, and great. I then made it a point to note that such curves were only a starting point, and not the end all of speaker evaluating. When doing subsiquent reviews of speakers I would always advise readers to go back to that article I did that explained the limitations of any measurement curves, including mine.

To get down to the basics when wrapping up my speaker reviewing work, I would always do careful, level-matched A/B comparisons of the speakers under review, using as references either my main-system, wide-dispersion Allison IC-20 models, my smaller-system, much narrower dispersing Dunlavy Cantata models, or, if I was reviewing some more budget oriented speakers, the moderately priced and more conventional NHT ST4 speakers that normally were located in my living room. I felt that all three sets were good references (sometimes I would use more than one of the pairs to refine my comparison work), although I also acknowledged their limitations when doing those comparisons. I also made it a point to use the same super-grade recordings in each comparison series, along with some new releases that I felt were also demo grade. (I often ended up doing record reviews in the same magazine with those newer recordings.)

I can see the point of both approaches to speaker dispersion (wide vs narrow, with the IC-20s and Cantatas being good examples of each approach), and said so in my assorted reviews. Both can work well. That is why I had those Cantatas in my second system for several years, even though my main system continued to use the Allison IC-20 models.

Howard Ferstler

I did some live recordings a few weeks ago for a work project. Both, acoustic instruments and electric. I was aiming for as much realism as practical for a one-day location session, and plenty of time was devoted to mic setup. Being an old fart, I still record to tape, (DA-38 in this case), and brought 8 channels with me. I can now listen to various configurations side by side. The ribbon X-Y stereo pair, a Royer SF-12, captured a lush, wonderful sense of the space and soundstage, that is in many ways extraordinarily realistic. But something is missing. The AKG C414B-XLS's captured a degree of presence and detail that is quite lifelike. But something is missing. The Neumann's really nailed the balance and dimensionality of the instruments. But something is missing. The 1/2" omni capsules, calibrated virtually flat to >20KHz captured every nuance. But something is missing. I guess I'm just a lousy recording engineer....

[kkantor]

"I will admit that I resented your reference to your academic and professional credentials in support of one of your arguments as they had no bearing whatsoever on the validity of your argument. I felt it was unworthy of someone who asserts himself to be a trained professional and who should be able to defend his ideas on their merits alone. Nor was I intimidated by you which seemed to have been your intent because both that argument holds no currency with me and because my own educational training and professional experience are every bit the equal of yours and maybe then some.

Well, I certainly apologize if I did either try to intimidate you or try to substantiate my arguments with a resume. I don't remember doing either, but sorry if I did.

In fact, the people I consider my most esteemed personal mentors come from a rather diverse range of academic backgrounds, from none at all to very acclaimed.

-k

[kkantor]

Hi Ken,

Thanks for starting a new "root of the matter" thread. This is certainly the topic we have been debating for a few weeks.

I think the fundamental debate boils down to the general (ideal) form of the frequency response of a neutral loudspeaker, one which adds no color to the sound. Arguments have been made for flat power response, flat anechoic axial response, flat in-room curves, rolled off room curves (matching concert hall measurements) and flat time windowed response. Also wide dispersion vs. constant directivity.

There was a great paper by Lipshitz and Vanderkoy where they did a simple experiment to explore just these questions. They put a KEF 104.2 ( a conventional forward firing loudspeaker) underneath an ESL 63. The ESL 63 is of course a dipole with two strongly defined nulls. It was mounted 90 degrees to the normal orrientation, such that a listener would sit in its null and hence any signal fed into it would contribute to the reverberent field of the room but not the direct sound. With this setup they could independently manipulate the direct field and the reverberent field of a loudspeaker in a room. They tried a number of scenarios and came to the following conclusions (somewhat paraphrased and simplified).

1) Flat sound power (near flat reverberent field, depending on the room constants) achieved by making the direct sound (the 104.2) bright, sounded excruciatingly bright.

2) Flat sound power by adding high frequency energy into the ESL 63 (the 104.2 stays "flat") also sound bright, but not quite as bad as above.

3) Adding some, but lesser amounts, of energy to the reverberent field could add some pleasent (acceptable) brightness or spaciousness.

4) If the sound power added to the reverberent field had holes in it's spectrum, this was nearly unnoticable.

5) If the sound power added to the reverberent field had peaks, this was noticable and undesirable.

I think these tests answer most of our questions. A shortcoming, for Howard's sake, is that they didn't pursue any responses tailored to mimic a concert hall reverberent field. Their starting assumption was that some measurement should be flat rather than contoured. So the baseline was a flat (anechoic, axial) 104.2 and they added from there.

From these experiments I would jump to a few conclusions: Flat power response is not what we want, especially if we achieve it by messing up the axial response. Wide range constant directivity isn't desirable either, because it would force us to take flat power response with flat axial response. However a smooth but rising directivity is a good thing. Peaks in the reverberent field are a bad thing but holes don't seem to matter.

Since the room curve is strongly influenced by the reverberent field it becomes an unreliable indicator of response. Remember that holes in the power response, hence in the room curve, are benign. We also know that the "ideal room curve" changes totally with the size of the room. Large theaters only sound good if the "house curve" (a far field measure dominated by the reverberent field, a combination of power response and rising room constant) rolls off dramatically.

I believe these various facts can only be reconciled if we accept that the ear has the ability to hear through to the direct sound (first arrival) and ignore the spectrum of the later sound. Some serious papers have been written that take that a step further and define a variable integrating time window that captures the direct sound only at HF and opens up to effectively capture the total room response at LF. (See Salmi, Kates and Bech) I am strongly convinced that this is the answer.

It also ties in with what Toole would say: That a good loudspeaker is smooth and flat in its axial anechoic response and smooth (not necessarily flat) in its off axis response. (Very paraphrased!)

Regarding how directional, or how wide a dispersion is correct, I think it is hard to argue any absolutes here. I am perfectly willing to let Howard like a very wide dispersion diffuse sound and let Zilch like a narrower, constant directivity derived, focused sound. With only two channels a lot of researchers conclude that room reflections, even though our small rooms aren't the concert halls, add some to the naturalness of reproduction (sound, that is). (A thought: what if a speaker had a means of sending delayed side wall reflections. Wouldn't that be magic?) Griesinger of Lexicon makes a comment about hearing a 5 channel system in a very dead room with an average absorbtion of 0.8 (near anechoic): "Ordinary stereo in this room was poor, but with side and rear loudspeakers and hall synthesis the sound was wonderful..." So with 2 channels we can use some help from the room (wide dispersion), with more channels we are probably better off without it.

I don't expect this little lecture to change anybodies strongly entrenched viewpoints on the subject. I've been watching and contributing to the debate for a couple of weeks. It doesn't seem to me that anyone hears or cares about any other viewpoint (sure, me included). It is a facinating topic and (Soundminded) not an indicator of our ignorance of the subject but more of the complexity of the matters of human perception. I would encourage everybody to search out some of these various papers. The answers are out there, it is just a matter of reading the literature and coming to a conclusion that seems to fit our observations of the (sound reproduction) world.

A last note on Allison and Vilchur. I think they made a great contribution to our understanding of how the low frequency system interfaces with the room. I think Allison came up with a practical approach (woofer near boundary, midrange far) that gets around the real issues they documented. We owe them for that. The thought that a speaker should mimic the balance of a concert hall isn't (shouldn't be) a major part of their legacy and is more likely a notion forced by practical issues at the time, such as low tweeter sensitivity and inadequate program material.

Finally Ken, I am aware of MTF as roughly the optical equivalent to frequency response. What do you mean by it in acoustics? Ditto alliasing? (Just curious)

Regards,

David

1- I appreciate the thoughtful post. I'm less convinced than you, et al, "that this is the answer," only because I no longer believe there is such an answer, in lieu of a better definition of the question. There are recordings that will shine when the playback is anechoic and the impulse response is clean and coherent. There are other recordings that will do their best when the room adds lateral reflections and reverberation. Either "side" in the debate can drag out plenty of examples to make their case.

2- Lipshitz buttonholed me in a stairwell of the NY Hilton after my very first AES paper. I was rather intimidated! He told me he thought I was on the wrong track, and that low IACC didn't sound good to people. After stuttering a while, I replied that this was the job of the recording engineer, who could make a recording with any amount of IACC they wanted. If the speakers degrade IACC, all hope and control of situation was lost. Information was gone. After a while, I believe he came to agree. I assert the same point about reverberation. When one starts relying on the speaker to add this stuff, all recordings start to sound like the speaker/room combination, to some extent. Do we really want that?

3- "Spatial Aliasing" has little to do with spectral aliasing and A/D converters, other than sharing a mathematical basis and thus a framework for analogy. Spatial aliasing happens when one attempts to reconstruct a space using too few data points. (IE- less than two sample points per unit of desired spatial resolution.) It comes into play in CAD, cartography, optics and in hifi. To make a long story short, spatial aliasing is the underlying reason that phantom images are unavoidable in stereo playback; they can be considered as artifacts of under-sampling in the spatial domain. (No, it is not true that they can be entirely eliminated anechoically, though they can be mitigated as a practical matter.)

4- Modulation Transfer Function is a measure that is often used in certain fields of audio, and which yields some valuable perceptual correlation. For some reason, speaker people seem to avoid many of the advances seen in other fields of audio perception, and still debate about things like pink noise spectra, radiation patterns and "time alignment." Sigh. I understand that its a daunting curtain to lift, and that commercial interests have little to gain, and much to lose, by undermining confidence in various dogma. It's the Occam's Razor of human nature. People default to good vs. evil. Third parties almost never succeed.

Page 4, below:

http://www.acoustics.hut.fi/teaching/S-89.3320/KA10.pdf

-k

[Zilch]

From these experiments I would jump to a few conclusions: Flat power response is not what we want, especially if we achieve it by messing up the axial response. Wide range constant directivity isn't desirable either, because it would force us to take flat power response with flat axial response. However a smooth but rising directivity is a good thing. Peaks in the reverberent field are a bad thing but holes don't seem to matter.

See Toole Fig. 18.22 contrasting Yamaha NS-10M and JBL 4301, two small two-way "monitors" with nearly identical directivities. NS-10M was optimized for flat power response, resulting in an overly bright axial response, whereas 4301 was optimized for flat on-axis response.

In a discussion with one of the design engineers, the author was told that the bookshelf loudspeaker, the NS-10M, was intended to be listened to at a distance in a normally reflective room. The bass contour allowed for some bass boost from a nearby wall, and the overall frequency response was tailored for a listener in what would then be called the reverberant sound field, which, it was thought in those days, was best characterized by sound power.

Since directivity is the difference between axial response and power response, it is possible for all of them to be flat. Flat directivity is rare, however, though by one view, it was the Villchur/Allison target. I believe it fails for other reasons....

[Zilch]

Yes, that would be a magic sounding speaker, and of course Ken designed a speaker like that when he was at AR. I never heard a pair, but the concept is terrific.

http://www.kenkantor.com/publications/magi...gic_speaker.pdf

[kkantor]

Here is the updated link to the paper I referenced in the above message. But, worth it.

http://www.hauptmikrofon.de/HW/Wittek_thesis_201207.pdf

(Seems they changed the server directory.)

Also, I will try to attach a JPG clip taken from the thesis, that I just love!

-k

[Zilch]

Interesting stuff, Ken.

If two laterally displaced guitars play the same note, do we perceive a phantom third one between them?

[kkantor]

Interesting stuff, Ken.

If two laterally displaced guitars play the same note, do we perceive a phantom third one between them?

Yes, it is interesting. Of course, we are a long, long way from knowing if it will lead anywhere, but I appreciate that good effort is still being expended toward improving sound reproduction. And, I hope it inspires us speaker guys to try and intellectually incorporate whatever is learned along the way.

Re: your tongue-in-cheek question, a true story: earlier this week, I stopped to notice a strange sound in my office, and was amused when I figured out what it was. My tech was sitting at one end of the conference table working away on his Mac laptop. Mine was at the other end of the table, maybe 8 feet away, on the other side. When I walked by, I got a strong and stable central image from the stereo fan noise, floating in mid-air above the table. A phantom laptop, as it were.

-k

[Zilch]

I note that your "Magic" speaker for AR sought to eliminate the adverse impacts of early lateral reflections upon spectral balance via moderate-width constant directivity and actively generate later ones in the range of 20 ms for ASW enhancement. Research indicates that even later ones, 80 ms and longer, are necessary for development of artificial LEV.

Recognizing that you did this work nearly 25 years ago, is this still a valid approach in your view? Looking up in Toole, I see that the Geddes alignment yields a low IACC contralateral first reflection in the range of 10 - 12 ms within the optimum rendition zone with little, if any, attenuation relative to the direct source, all without the complications inherent in independently generating spaciousness cues. I've seen earlier references to this approach as well, as you have observed, but it's only recently that advances in waveguide theory and design have provided the requisite devices and understanding to economically implement it in a well-controlled manner. It seems to work for me; what is your analysis?

I know you want us to get beyond this and on to more fundamental precepts, but my problem is apparent here and elsewhere: it's all I can do to advocate even the "trivial" without being branded a basher for the purpose of summary dismissal....