ucla88 said:
Much of what you say is true and very good points. THD and IMD will correlate to the subjective response if they are compared within the same system. This is because the structure of the nonlinearity is not changing. So "comparitively" speaking the numbers are relavent. This is the case with the Klippel demo's - the levels change, but not structure of the nonlinearity, that is held contstant. One cannot compare 5% THD on one system with 5% on different system and say that they will sound the same - it just ain't so. Usually distortion is used to compare one driver to another one and this IS NOT a comparative situation and should not be done.
And (I say this very often) it is ceratinly possible to design a system where nonlinearity is a factor and a big factor. That's why it is important to note the SPL levels that one is doing an evaluation at in order for a comparison to be valid. This is virtually never done in practice. And, in fact, I would claim that reviewers should find that SPL level at which the speaker is audibly distorting and note that - this too is never done. The 6.5", etc. systems may be fine at 70 - 80 dB SPL, but they can't go much above this in a decent sized room. If that level is fine with you then that design is fine.
But I think that I take the strongest acception to the statement "Nonlinearity is undesireable. " This does not test out to be the case in practice. Because, when someone is asked for their "preference" they will often, if not always, prefer the distorted system. Your statement assumes that "accuracy" is the goal, which it is for me, but for many, if not most, accuracy is not the goal. Anyone who says "its what you like that matters", does not have accuracy as the goal, it is preference, and distortion is usually prefered.
Earl, thanks for starting this thread. People are asking questions I wouldn't have thought of.
I have a question that I think you've answered in part but I'll ask it anyway.
On page 227 of "Audio Transducers", you have a chart showing the relative audibility of several nonlinear characteristics which are functions of displacement and/or temperature. Variations in BL are rated as high importance, while variations in compliance and voice coil resistance are rated as medium importance.
Would your more recent work on the audibility of distortion modify the information in that table?
Thanks,
Duke
I have a question that I think you've answered in part but I'll ask it anyway.
On page 227 of "Audio Transducers", you have a chart showing the relative audibility of several nonlinear characteristics which are functions of displacement and/or temperature. Variations in BL are rated as high importance, while variations in compliance and voice coil resistance are rated as medium importance.
Would your more recent work on the audibility of distortion modify the information in that table?
Thanks,
Duke
audiokinesis said:
Hey Duke
I'd comment by saying that it was not my attention that the table was in regards to perception - only the physical effect of the nonlinearity. Based on perception I would likely change that table a lot. When that book was written I, like many, belived that nonlinearity in the transducer was a major source of poor sound quality. I studied nonlinear systems theory, made systems to measure the component nonlinearity (almost a decade before Klippel by the way) and felt that I understood every facet of what influence nonlinearity had on the physical system.
Then I went to work for a transducer company and we began to "rank order" the subjective effects of the nonlinearities - one wants to solve the problems most worth solving. And at the same time I made the Summa loudspeakers and experimented with those. To make a long story short, I/we could not find any changes to the nonlinear characteristics that had much of an audible effect, but I did find changes that did have a profound audible effect.
It doesn't take a nuclear scientist to tell you to pursue those things that make a big difference and to leave behind those things that don't. The products that I created were IMHO, the best that I have ever done, and quite competitive with anything that I have experinced anywhere. And these were done with virtually no attention to nonlinearity what-so-ever. It just doesn't seem to make any difference when the system design is right. I don't look at THD curves, I don't pay much attention to X-max, etc. In fact I don't really pay much attention to the main loudspeaker at all below about 200 Hz, other than to make it a monopole. Below that frequency the room and HOW the subs and mains are setup dominates the problem and the individual LF sources don't make much of a difference.
So in the woofer I am more concerned with how it behaves at its "edge hole" than what its Tiele-Small paramters are - I never even looked at those numbers. Even in my subs the TS parameters are pretty much irrelavent because the bandpass designs acoustics dominate the problem. Good systems design makes the drivers pretty unimportant. Of that I am convinced.
There is a lot that I would change it that book, but none more than the stuff on nonlinearity. That turned out to be a massive waste of time. The waveguide stuff is where the real gold is to be found. The sound of a speaker is dominated by its 600 - 6000 Hz performance and this region is dominated by design considerations that only a waveguide can achieve. Don't read the stuff on non-linearity, but fully grasp the chapter on waveguides. Thats what I would say today.
That and room acoustics.
gedlee said:
I meant this more as my technical preference. You're quite right that many preferences may actually gravitate towards distortion. (Just don't tell the audiophile that!) This is one of the reasons why the 6.5" 2 way is so ubiquitous. The second and third order products really are not particularly offensive and don't sound like distortion. Or, at least not like the higher order products of say, a strained tweeter. All things being equal, my personal preference would be the lower nonlinear distortion unit.
Of course, things are rarely equal!
Wow, thank you. I'm not used to authors saying "based on what I now know, you can ignore this part of my book"!
I have a follow-up question. You wrote:
"So in the woofer I am more concerned with how it behaves at its "edge hole" ..."
Is the "edge hole" the dip in response that I see in some woofers' frequency response curves, before the final peak and rolloff? Or, is it something else?
I presume it's caused by a cancellation as the wave moving in the flexing cone itself reflects back from the edge, but if not could you clarify that as well?
Thanks.
I have a follow-up question. You wrote:
"So in the woofer I am more concerned with how it behaves at its "edge hole" ..."
Is the "edge hole" the dip in response that I see in some woofers' frequency response curves, before the final peak and rolloff? Or, is it something else?
I presume it's caused by a cancellation as the wave moving in the flexing cone itself reflects back from the edge, but if not could you clarify that as well?
Thanks.
gedlee said:
This is a conceptual point that I don't think many people understand, and this lack of understanding may be muddling the discussion here.
The point is that music has a high crest factor and if you look at the amplitude distribution of the signal, it spends a hell of a lot more time near 0 than out at the peaks. Consequently low amplitude distortions matter a lot more. Notice how this is in stark contrast to say a THD measurement at a fixed SPL, which is done with a sine wave that has a very low crest factor. It's not surprising that the THD of two speakers at say 90 dB does not correlate with listening to music on them at 90 dB.
What I do wonder though, is if one could make a reasonable comparison using say, HD measurements made in 5 dB steps from 50-110 dB, weighting the low amplitude and higher order distortion more.
On a separate topic, a few people have asked somewhat indirectly if low nonlinear distortion isn't responsible for dynamics, what is? My guess is that this is almost entirely thermal. That's the way I've been leaning over time and it seems to be implied by nonlinear distortion not mattering. I know for damn sure that thermal compression is a major problem in a lot of speakers, but I can't rigorously claim that it's the dominant issue in dynamics. That is a guess. Any comment Dr. Geddes?
Rybaudio said:
Loudspeakers are different from electronics. All orders of distortion tend towards zero at lower levels in loudspeakers. There is no "zero crossing" problem with loudspeaker units. If a nonlinear distortion is not audible at a higher level or measureable, it will not be evident at a lower level. There is no benefit as far as I've seen for testing nonlinear distortion at low levels.
I'm quite sure that most people can understand that we need to handle the more dominant aspects that are in the way of more accurate sound reproduction. However, it would be wrong to throw the "linearity" issue out the window. Once we take care of the energy storage problems to a certain degree, then the linearity issues will be more obvious.gedlee said:
I find this quite interesting. One, because if we beleive that less drivers is better for a particular SPL level design, then Xmax if quite important. Second, most of the audible problems I have come across are driver related, mostly below the 100Hz and above the 15KHz range.gedlee said:
While it is true that speakers are dominated by the 600Hz-6000Hz perforemance, there are other issues outside this range that might make very donminate effects that will be detected during listening because it effects this the integrity of the signal in this range most sensitive to the ear.
To think that system design in more important than the driver is like say the whole human body is more important than the heart, or the whole car design is more important than the engine.
audiokinesis said:
The classic problem in a loudspeaker cone happens like this:
The outer edge, which is basically unsupported, because at these higher frequencies the surround is not even there, begins to move more than the body of the cone - its starting to go into resonance. This causes a rise in output. Then very quickly the edge starts to go out of phase with the main cone body and a cancellation occurs when the outside is moving with an equal and opposite volume displacement with the inside. This can sometimes result in a very deep hole.
Various changes in the surround mass and cone stiffness, etc. can modify this result to some extent, and it can even be made to almost equal out to a fairly flat response, but if you look carefully you can always see this point in either the axial response or the polar response.
Above this first breakup the cone is basically resonating in a multitude of modes and could almost be called chaotic. Things in this frequency region can be better or worse, but for all practical purposes above the classic edge hole, the driver has ceased to rigid and IMO useful. I would never use a driver above this point.
There is also the classic "spider hole" where a spider resonance sucks out the energy - by creating a large impedance - which results a dip in the entire response of the driver. How well the driver controls these problems is a big factor in my choice. And I have found that you have to measure the driver yourself, because seldom will the manufaturers data show this. There are so many ways to hide something like this that if they can they will, and they do. Find just the right measuring distance and these can disappear.
Rybaudio said:
You can take and clip a signal with a pure third order nonlinearity and it has virtually no effect on the perceived dynamics. Perhaps I'll post some examples some day.
The dyanamics issue is a touchy one, but what I have found is that high efficiency systems seem to have a better dynamic range. Perhaps this is because they take less watts to achieve a given SPL so there is less heat that has to be dumped along the way. I started measuring thermal responses in loudspeakers and I found that this did seem to correlate better to perceived sound quality than frequency response or THD.
My plan is to simulate the thermal compression and do a blind subjective test on this to determine the extent to which we can correalte this effect with perecption. No one has ever actually done this. I suspect that thermal modulation is a bigger factor than we suspect.
Let me give you another critical piece of information. Loudspeaker do exhibit sub-harmonics. But if you study nonlinear systems theory sub-harmonics cannot exist is a time-invariant system. SO what causes this? An expert in nonlinear theory that I asked about this believed that the likely candidate is thermal modulation, which creates a time-variant system at the frequency of the sub-harmonic. Sub-harmonics - by my distortion metric - would be almost completely unmasked and highly audible. Food for thought don't you think?
ucla88 said:
Just one word of warning. It is possible to have a zero crossing effect in a loudspeaker where the spider snaps through the zero point, kind of like an oil can. I've seen this in some car speakers. It sounds so bad that you cannot even believe it. Some very strange things CAN happen, but for the most part we have learned how NOT to do these things and drivers are pretty much free from the really abominable design flaws.
In electronics, I only want to see the distortion as the level falls, not as it goes up. This is usually a problem because of the noise floor. Your can't look at THD + noise because the noise will mask the highly audible problems. You have to look at the actual spectrum.
Dr. Geddes is bringing up a critical point about voice-heating heating. It is important to remember that VC heating has an associated time-constant - it heats up in milliseconds, but can take several seconds to cool down again. The time-constant for the cool-down is controlled by the percentage of the VC that is not in the gap (which acts as a loosely-coupled heat-sink), whatever air-flow that might be present due to deep bass excursions, and many other factors.
If the cone or diaphragm isn't moving much (MF and higher excitation frequencies), there will be little or no air-flow past the VC, and cool-down is more a function of adjacent structures gradually absorbing the infrared emission from the VC. If Ferrofluid is used in the gap, then convected-fluid cooling appears, although there is always a risk of localized boil-off if temperatures are high enough. (I've seen baked-on Ferrofluid in the gap, so it's no myth.)
This time-domain "slurring" is more characteristic of Class AB output-transistor thermal drift (and associated misbiasing under dynamic conditions) than typical loudspeaker distortion mechanisms, most of which which disappear as soon as the signal excitation goes away.
Physically large voice coils and efficient loudspeakers seem like the only way to minimize the problem, unless you consider offbeat solutions like redesigning loudspeakers for low-noise forced-air cooling flowing through the gap (like modern computers and their forced-air-cooled and heatsinked microprocessors). It would only take a small flow of air to dramatically cool the voice coil - by 20 to 30C. Re-engineering the driver would be the hard part.
P.S. Especially like the 600 Hz to 6 kHz comment made earlier. That's where 90% of the effort should go. Good writing, good points that need making.
If the cone or diaphragm isn't moving much (MF and higher excitation frequencies), there will be little or no air-flow past the VC, and cool-down is more a function of adjacent structures gradually absorbing the infrared emission from the VC. If Ferrofluid is used in the gap, then convected-fluid cooling appears, although there is always a risk of localized boil-off if temperatures are high enough. (I've seen baked-on Ferrofluid in the gap, so it's no myth.)
This time-domain "slurring" is more characteristic of Class AB output-transistor thermal drift (and associated misbiasing under dynamic conditions) than typical loudspeaker distortion mechanisms, most of which which disappear as soon as the signal excitation goes away.
Physically large voice coils and efficient loudspeakers seem like the only way to minimize the problem, unless you consider offbeat solutions like redesigning loudspeakers for low-noise forced-air cooling flowing through the gap (like modern computers and their forced-air-cooled and heatsinked microprocessors). It would only take a small flow of air to dramatically cool the voice coil - by 20 to 30C. Re-engineering the driver would be the hard part.
P.S. Especially like the 600 Hz to 6 kHz comment made earlier. That's where 90% of the effort should go. Good writing, good points that need making.
gedlee said:
Did anyone ever claim that? It should go without saying IMO... if we talk about people with basic training and experience in audioengineering.
Most systems.
Of course it is.
Unfortunately (as I'm sure you know) many reviwers are part of the companies marketing department (indirect) so don't hold your breath for scientificall testing by the majority. Other than that it's a little bit more complicated than you seem to imply. In order to know if the DUT is audibly colored from distortion you'd need a reference wihtout any distortion. That is easily done with bypass tests of transmission links such as electronics but not with speakers which are electroacoustic decoder systems.
Below 200Hz or so I'd agree. Above 200Hz or so a 6.5" with a good dome tweeter is ok. HD about 0.1-0.5% from 90dB up to 105dB. 90dB continuous without significant thermal compression and peaks/transients above 110dB without compression.
Again I start to wonder what kind of drivers you have used/measured. I asked about that tweeter that you used as a representant for all dome tweeters in that compression mesurement but got no answer. You seem to disqualify high quality drivers if they are expensive.. which makes the discussion kind of strange if it's not stated that the discussion is within a specific budget.
Until they experience a low distortion system with good material..
A strange statement.. I think it would be more correct to say that a speaker is dominated by its specific shortcomings (or relative lack of such) which may vary a lot between different brands and models.
Do you describe a particular driver in the post that I quoted the above from? You seem to be slightly dogmatic and generalize a lot about audio. There are drivers that have a smooth flexing behaviour that allows for extended bandwith with low distortion and good dispersion from a relatively large cone.
If clipping of a signal results in audible distortion and reduced dynamics or not obviously depends on how severe the clipping is.
Is that from listening or do you have data to support it? I know many claim what you do but I have not seen any evidence. My own measurments indicates the opposite is possible.. not always of course but sometimes.
But the low moving mass also means low thermal mass.
Sounds interesting.. I'm looking forward to your results.
What you really meant to write was that as far as YOU know, no one has, right...?
With some systems and in some occasions.. I agree.
Would you mind clarify what you mean with subharmonics, and in what situations/with what kind of signals this appears?
What puzzles me (if I understand this correct) is that you believe that these non masked signals would be highly audible.. but sum and difference tones from intermodulation is not??
Which is totally cool... as long as everyone is clear about objectivity and subjectivity.
/Peter