Hi,
My first reason to be here was actually to ask Dr Geddes if any circular that are correctly made are condemned to exhibits diffraction holes on the on axis response like this one:
http://s139.photobucket.com/albums/q288/augerpro/QSC SP-000106-GP/
Cheers
M
My first reason to be here was actually to ask Dr Geddes if any circular that are correctly made are condemned to exhibits diffraction holes on the on axis response like this one:
http://s139.photobucket.com/albums/q288/augerpro/QSC SP-000106-GP/
Cheers
M
Toole sometimes comes to watered down conclusions along the lines of "If a speaker is flat on and off axis and well behave when hemispherically and spherically measured, it will probably sound okay."
If you dig through his papers you can come to much stronger and more useful conclusions than that. I would recommend his original study where 20 speakers were rank ordered in careful listening tests and then all of their parameters where measured and compared to the rankings. It is clear there (and Toole so concludes) that axial and near axial frequency response has a strong correlation with ranking while power response does not.
There is no particular power response shape that is favored in the rankings.
In fact some of the speakers that do well have poor power response and some systems with smooth and flat power response do very poorly in the rankings. Toole does like to use power response as a convenient way to reveal resonances. Unlike reflections that give response effects specific to particular spatial location, resonances are still present after averaging the spherical response.
A better paper on the subject (more directly on point) was the one that Lipshitz and Vanderkooy did where they placed a dipole speaker on top of a conventional speaker. The dipole was turned sideways so that the listener was in its null. Using this settup they could independently manipulate direct response and later power response and come to some conclusions: Flat power response achieved by messing up the axial response of a convention directivity will sound way too bright. Flat power response achieved wihtout messing up the axial response, rather by bolstering the reverberent field, will still sound too bright. A smoothly declining power response sounded better. And finally that holes in the power response were innocuous (as Toole's data shows).
I was involved in some of the first speakers that used CD devices back in the early 80s. It would have been easy to conclude that they were better because the power response was flatter (the marketing department, in fact, ran with that). Don Keele and I resisted saying that, because no psychoacoustic reseach supports it.
When we discuss "flat power response" we need to realize that a CD device does not give flat power, other than in its particular range. There are also a lot of devices called CD when they have smooth rather than flat power response. Also, discussing flat power response can not be divorced from the particular directivity of the unit. If flat power means omnidirectional, or very low directivity, then the L and VK study says it will sound too bright. If we have flat power response with higher directivity (say a constant cardoid or dipole directivity) then we may be okay with the sound, since we aren't accentuating HF power. We need to be specific.
Everything I read says that power response should roll down hill and may have holes but shouldn't have peaks. Direct response on the listening axis is far more important.
David S.
The axial hole requires two things, a ring of diffraction, such as a mouth edge, that is everywhere equidistant, hence not square nor elliptical, and a highly coherent wavefront (not very common). So we won't see an axial hole if there are a lot of HOMs, nor if the mouth is rectangular. A highly coherent wavefront impinging on an elliptical mouth would have a small indentation, i.e. think of it as a hole spread wide - not as deep, but wider in frequency.
The depth of the hole depends on the amount of diffraction, which depends on the radius of the mouth flare. The Abbey's axial hole is deeper than the Summa's (which is negligable) for two reasons. First the mouth radius is smaller, more diffraction, but it just so happens that at the hole fdrequency there is a resonance across the mouth aperature which amplifies this diffraction through a resonance. Future Abbey's will have a larger radi which will change both the diffraction and the resonance making the axial response much more like the Summa.
I discovered this with my new measurement software which allows for a reconstruction of the source that created the field. At the hole frequency a standing wave in the plane of the mouth is clearly evident.
There is actually a notch in the crossover at this precise frequency because otherwise the power response has a peak. Thus, correcting the power response further drops the axial hole. Clearly a case where one would not want to EQ the response along the central axis to be flat.
That's really interesting. Does it show you this graphically, or do you have to fish it out of the numbers yourself?
The relationship between the mouth diameter and lip radius should be studied. Generally, if the lip is not a constant radius, then the hole will also be reduced.
Dr Geddes thanks a lot for the clarification.
Dave,
I do know well about the papers you are referring to but I’d like to add to my point an illustration of a high end form JBL:
And more recently in this other design:
A legend can be found here:
From:
JBL Pro - Recording & Broadcast
As you can see it correlates quite well with what I tried to explain in domestic conditions:
One may find in this documents two different design goals figure 1 and 2 which are more in line with the Keel paper:
http://www.jblpro.com/catalog/support/getfile.aspx?docid=238&doctype=3
The priority was the flat on axis responsen however these results in DI (not flattish) terms are consistent with the intended use in a passively treated room (studio mixing room) which (is supposed to) guaranties the consistency of the direct sound field and (or drastic decrease in) the early reflected and reverberant sound fields.
K+H and Genelec seem also to pay some attention to the horizontal CD characteristics of their monitor for some reasons as published in their specs and corollary the flatness of the power response as well.
My take is that the partial CD (the plateau) is a gain in robustness of the design with regards to its integration in a less than optimal listening environment. I don’t think that the flatness of the designed axis response is to be overlooked but a trade-off implied by Toole and probably biased towards a flattish on axis rersponse.
All these approaches have merits and one is entitle to choose the priorities to meet one’s requirements as long as no clear results are scientifically extracted.
End of Off topic.
M
Dave,
I do know well about the papers you are referring to but I’d like to add to my point an illustration of a high end form JBL:
And more recently in this other design:
A legend can be found here:
An externally hosted image should be here but it was not working when we last tested it.
From:
JBL Pro - Recording & Broadcast
As you can see it correlates quite well with what I tried to explain in domestic conditions:
One may find in this documents two different design goals figure 1 and 2 which are more in line with the Keel paper:
http://www.jblpro.com/catalog/support/getfile.aspx?docid=238&doctype=3
The priority was the flat on axis responsen however these results in DI (not flattish) terms are consistent with the intended use in a passively treated room (studio mixing room) which (is supposed to) guaranties the consistency of the direct sound field and (or drastic decrease in) the early reflected and reverberant sound fields.
K+H and Genelec seem also to pay some attention to the horizontal CD characteristics of their monitor for some reasons as published in their specs and corollary the flatness of the power response as well.
My take is that the partial CD (the plateau) is a gain in robustness of the design with regards to its integration in a less than optimal listening environment. I don’t think that the flatness of the designed axis response is to be overlooked but a trade-off implied by Toole and probably biased towards a flattish on axis rersponse.
All these approaches have merits and one is entitle to choose the priorities to meet one’s requirements as long as no clear results are scientifically extracted.
End of Off topic.
M
Last edited:
Lets not confuse that a speaker can be designed with flat axial curves and linear hemi and omni curves with proof that it should be designed that way.(or must)
Floyd has always talked about speaker response as falling neatly into axial (or near listening window) curves, hemispherical curves, and spherical (power response) curves, but his own tests clearly show that a wide range of power responses won't remove a speaker from the top category, nor will a good power response make it well received if the axial response is sub par.
This is where this off topic bit came from: it was stated that a room correction system fixed power response holes (necessarily putting peaks in the direct/early response) and that that was a good thing.
To say that "All these approaches have merits and one is entitle to choose the priorities to meet one’s requirements as long as no clear results are scientifically extracted." ignores that a great deal of careful subjective testing has been done and real conclusions have been drawn. The question is always about determining "necessary and sufficient criterion" in loudspeaker design. Those that get interested in improving power response frequently fall into the trap of proclaiming it a necessary criterion and, worse, begin compromising axial response to improve it.
David S.
Well its graphically presented in the software, but the software has to fish it out from the numbers. The software lets me look at any frequency just by moving a slider and it then simulates the motion in real time (slowed down to cycle once a second of course. You couldn't see it otherwise.) Right now its resolution is kind of low because of the boundary conditions that I am using. I want to use an "infinite baffle" to improve the resolution, but then "infinite baffles" are hard to come by. So I am looking at maybe using a finite baffle and removing the effects of its being finite. That may be just as problematic as the current boundary condition however.
If you are refering to me, thats not what I stated. I said that the fact that a power response hole made a difference means that it was audible, not that its right or wrong either way. Only that if holes in power response were not audible then it should not have entered into the data at all. But it did. Which is right or wrong is irrelavent.
Cool, thanks. Have you posted these somewhere? I seem to remember something like it. I"m also reminded of Wilson E. Kock's "Seeing Sound" experiments.
Your quote:
"Sean Olive showed that power response holes do make an audible difference in his testing of room EQ systems. The eystems that "fixed" this hole were deemed to "sound better". "
Fixing a hole (via room EQ) can only mean filling it in by placing a complimentary peak in the direct sound at the same frequency.
David S
"Fixed" is in quotes, as is "sounds better" to imply that I do not give any credance to those claims, only that they were stated. As far as I am concerned, the only take-away point is that the power response hole was a factor in the data and as such had to be audible. Whether it was "fixed" or made worse has nothing to do with my point.
It should be obvious that if a power response hole is audible that not having the hole is prefered. The importance of this preference is undetermined.
Your whole argument rests on one premiss - that a crossover is inaudible. For, if it is inaudible then it makes no difference what frequency it is at. However, if it is audible, which I certainly believe is always the case, then the frequency it is at does make a difference and there are better and worse places to put it.
Your whole point about the hole being better or worse in Sean's test is irrelavent IMO. But for what irt is worth, I agree with you, fixing the whole with electronics is exactly the wrong thing to do and it will only make things worse. Moving it somewhere that it is less objectionable is exactly what I am suggesting be done.
Last edited:
I was claiming that power response holes are generally inaudible. You were using Sean's room EQ success as some proof that holes were audible and filling them (by inference) was a good thing. Let's both agree that Sean wasn't really saying that and his liking a room EQ system isn't proof one way or the other about power response, at least relative to axial response.
My claim is that crossovers can be easily designed to give seamless performance and that, on axis, the crossover point can be so good as to be undetectable. I can do that, many other people on this forum can do that, and I assume that you can do that too.
That being the case, then the only means of detecting a crossover (that I can think of) is to 1)move off to an unlikely (wrong) listening position. 2) Be able to hear the shift in group delay as you cross between units, or 3) be able to detect a dip in the power response.
If any of these things are the case you still have to be convincing in showing that the undesireable abberation (1, 2 or 3) has frequency ranges of higher acuity and frequency ranges of lesser acuity i.e. good and bad frequencies for a crossover.
With so many systems designed over the years and all having crossover points chosen for reasons of expediency (driver - best response ranges, power handling, matching polars, whatever) I haven't heard anyone come up with a theory of a no-go regions for crossover points. It just feels more like an audiophile notion than one that has real science behind it.
David S.
My claim is that crossovers can be easily designed to give seamless performance and that, on axis, the crossover point can be so good as to be undetectable. I can do that, many other people on this forum can do that, and I assume that you can do that too.
That being the case, then the only means of detecting a crossover (that I can think of) is to 1)move off to an unlikely (wrong) listening position. 2) Be able to hear the shift in group delay as you cross between units, or 3) be able to detect a dip in the power response.
If any of these things are the case you still have to be convincing in showing that the undesireable abberation (1, 2 or 3) has frequency ranges of higher acuity and frequency ranges of lesser acuity i.e. good and bad frequencies for a crossover.
With so many systems designed over the years and all having crossover points chosen for reasons of expediency (driver - best response ranges, power handling, matching polars, whatever) I haven't heard anyone come up with a theory of a no-go regions for crossover points. It just feels more like an audiophile notion than one that has real science behind it.
David S.
Last edited:
Hi,
I am NOT denying that flattish FR is one of the goals... I thought that “biased towards flattish on axis response” was clear enough to explain that it should be on top of the priorities.
It seems to me that if one says the flat on axis response is the only really important criteria for designing speaker one is actually putting aside a part of the corpus of the advances made during the past 30 years or so (Toole as quoted, Bech...).
IF on-axis flat FR is sufficient, EQing to ruler-flat would be THE panacea and we both and other members of the AES around would be out of jobs since all AV receivers could do an automated EQ with this goal on any 30USD FOB HK 5.1 pack...
BTW the paper mentioned earlier from Olive would already be in conflict with the Flat goal regardless of the power response issue.
Also passive treatment is far from being common place in living rooms so in my synthesis of the state-of-the art, power response or part/model of it, among other things, is not to be forgotten.
It seems possible to get a less flat on axis FR and flattish power response and still have at least as good ratings than a similar speaker with very flat FR and no so flat power response:
Recently the Behringer B2031 is getting traction as the default low end monitor a la NS10 not quite popular yet though.
You can find data here Zaph|Audio
If accuracy is a safe and the most sensible bet when it come to designing a speaker with the scientific method it is very unfortunately far from being a popular call.
Here is an example outside the loudspeaker realm if accuracy were the goal no one would have bought the Monster Beat by DrDre "noise cancelling" Studio Headphone, I am not talking simply marketing, you can read the user reviews...
the JBL graphs shown previously might also give a hint at the targeted audience with the DI flat up to very high frequency
M
I am NOT denying that flattish FR is one of the goals... I thought that “biased towards flattish on axis response” was clear enough to explain that it should be on top of the priorities.
It seems to me that if one says the flat on axis response is the only really important criteria for designing speaker one is actually putting aside a part of the corpus of the advances made during the past 30 years or so (Toole as quoted, Bech...).
IF on-axis flat FR is sufficient, EQing to ruler-flat would be THE panacea and we both and other members of the AES around would be out of jobs since all AV receivers could do an automated EQ with this goal on any 30USD FOB HK 5.1 pack...
BTW the paper mentioned earlier from Olive would already be in conflict with the Flat goal regardless of the power response issue.
Also passive treatment is far from being common place in living rooms so in my synthesis of the state-of-the art, power response or part/model of it, among other things, is not to be forgotten.
It seems possible to get a less flat on axis FR and flattish power response and still have at least as good ratings than a similar speaker with very flat FR and no so flat power response:
An externally hosted image should be here but it was not working when we last tested it.
Recently the Behringer B2031 is getting traction as the default low end monitor a la NS10 not quite popular yet though.
You can find data here Zaph|Audio
If accuracy is a safe and the most sensible bet when it come to designing a speaker with the scientific method it is very unfortunately far from being a popular call.
Here is an example outside the loudspeaker realm if accuracy were the goal no one would have bought the Monster Beat by DrDre "noise cancelling" Studio Headphone, I am not talking simply marketing, you can read the user reviews...
the JBL graphs shown previously might also give a hint at the targeted audience with the DI flat up to very high frequency
M
Last edited:
Frequency response, cumulative spectral decay, directivity, diffraction, are all very important factors, sometime certain characteristics are more dominant than others, so for an audio system, including but not limited to audio speakers, generally there is a development cycle that occurs.
Hopefully in the process, certain technical criteria will gradually be established, and the criticality of levels are sorted out. Something that has not happened to date.
An externally hosted image should be here but it was not working when we last tested it.
Hopefully in the process, certain technical criteria will gradually be established, and the criticality of levels are sorted out. Something that has not happened to date.
Last edited:
"Biased towards" is suggesting that you would measure the on axis response and the power response then average the two with weighting towards the on-axis response, and this would give a response that matches our perception. There is no evidence that this is valid, particularly if done without regard for frequency.
I don't think anyone has said on axis frequency response is the only thing that matters. Only that it most highly corresponds with rankings of loudspeakers from best to worst in listening tests.
The problem with this assertion is that no room auto EQ systems even measure the actual (anechoic) on-axis response of a speaker. Not even close.
There are several reasons why room auto EQ systems don't work, some technical, some basic fundamental errors in defining what it is a room EQ system should and should not be doing.
The first error is the faulty assumption that the room itself needs EQ at high frequencies (above ~500Hz) and that it is valid to attempt to EQ the "room" response at these high frequencies. The second error is that nearly all auto eq systems measure only the steady state response of the room, not the actual on-axis response of the speaker, despite there being no evidence that the steady state response correlates to what we hear at high frequencies. (In fact quite the opposite) These two errors go hand in hand.
With both the end goal and the measurement technique being wrong its no wonder that an auto-eq system is no panacea for a poor sounding system. For example, such a naive system will attempt to "fill in" power response holes at high frequencies in the process introducing on axis response peaks.
Only below 300-500Hz should any auto eq system attempt to make corrections for the room based on steady state measurements. Above 500Hz the only valid corrections are correcting for errors in the response of the speaker not the room, and these cannot be based on steady state non-anechoic measurements.
If the speaker already has an exemplary on axis response and smooth polar response above this frequency there is nothing for the auto eq system to do, and nothing that it should be doing regardless of the room, however most if not all current auto eq systems will still attempt to make corrections based on the measured "room" response, thus making an excellent speaker worse. (The very flat on axis response will be made worse in a way that depends highly on the room and specific measuring setup, microphone position, speakers power response etc)
Could a room auto-eq system correct a speaker accurately above 500Hz in-situ ? I don't think it could. For it to do so it would need to measure a pseudo-anechoic response of the speaker at high frequencies using gated measurement techniques, however with a speaker in its normal resting place near boundaries, and the microphone near the listening end of the room there simply isn't enough reflection free time to measure below about 3Khz. Even at 1 metre there isn't much reflection free time if the speaker is sitting on the floor near boundaries.
A sliding window where window time is short at high frequencies and gets longer with lower frequencies gives a better approximation of how we perceive tonal balance, but doesn't solve the problem of ripples in the gated response due to reflections, which means a lot of smoothing would need to be applied, thus dramatically decreasing accuracy.
Finally any single point measurement, even assuming it had sufficient reflection free time to measure sufficiently low in frequency with sufficient resolution, would not be able to distinguish between response aberrations that are due to polar response problems (such as poor driver phasing, cabinet diffraction, driver on/off axis response problems) and would make ill-informed "corrections". For example a speaker that had an on-axis diffraction dip at some frequency may end up getting it boosted, resulting in an off axis peak at that frequency at every other angle.
Only if the speaker had a very smooth and well behaved polar pattern with no on-axis diffraction dips could such a single point measurement unambiguously "improve" the response, but such a well behaved speaker is likely to already have a very flat response and not require EQ in the first place, while a speaker with a wildly non-flat on axis response that might benefit from EQ is likely to have polar response aberrations that would make such a single point measurement inaccurate.
Finally, part of what is important when achieving flat frequency response on axis is narrow band flatness. In other words its not simply enough to have a flat 1/3 octave averaged response. It may sound tonally neutral overall, but if a multitude of high Q resonances or narrow-band diffraction interference effects are disguised by the averaging, the quality will still be poor.
A steady state measurement as taken by room EQ systems of necessity must be 1/3 octave averaged because the narrow band response in a room varies wildly +/- 10/20dB over very small fractions of an octave. As soon as you apply such smoothing resolution of narrow-band response aberrations is lost, and you cannot correct for them. (Only if they were resonances could you correct for them anyway mind you, if they're due to diffraction you can't)
All these reasons illustrate why a room auto eq system can never make a silk purse out of sows ear, but they don't make flat, smooth on-axis response any less important, however that response must be measured pseudo-anechoically, along with full polar responses to make sense of the measurements.
Other than correcting bass frequencies, or making "some" improvement to speakers with extremely poor on-axis response, automated room eq is nothing more than a gimmick to aid sales. An already well designed speaker will not benefit from automatic room eq at high frequencies, and has just as much chance of being made worse.
Last edited:
I disagree. Just gate the impulse response and we're looking at quasi anechoic data, at least for high frequencies. And, if a boundary is too close, then it IS part of the direct sound.
I believe that newer room EQ systems "do it right". Just look at the magnitude response of Audyssey XT32 filters. They do exactly what you're describing. Correct the room at lower frequencies and correct the speakers at higher frequencies.
Last edited: