But the third harmonic distortion in loudspeakers, for example, can emanate from two distinct mechanisms and have a different subjective impact because of that. That is why I have been trying to emphasize that context and interpretation of measurements is important if the measurements are to correlate with impressions of "sound quality".
As you no doubt know already, the radiation impedance forms part of the "coil" impedance. If the radiation impedance changes, as it will do as and when a diaphragm resonates, or when suspension or surround non-linearity is apparent, that change will be reflected in the coil impedance, hence cause a (small) non-linearity.
In the professional scientific and engineering world, MATLAB is ubiquitous. It provides readily usable analysis tools, but even then requires data capturing equipment and at least a moderate grasp of scripting. For me, however, this has always constituted "readily available" - others with different backgrounds may offer a different interpretation!
I suspect many of the tools used by members of this forum will require some adaption/coding too. It would be frustrating if the tools that produce the colourful spectrograms in this thread, for example, did not allow for use of Wigner Distributions or bispectral analysis, for examples. (See post #24). Hopefully people who know the tools well can explore the options available?
For those not having MATLAB available, Octave might be worth trying instead, although I suspect it will take more scripting to produce the desired results.
In several... For a start, the motor linearities due to eddys become more prevalent at mid and high frequencies, so can be discriminated simply by their frequency.
But beyond that, displacement dependent non-linearities will have an intrinsic correlation to coil velocity (hence sound pressure), and therefore a correlation to the source of the sound that is being reproduced. The eddy-dependent non-linearities have a less obvious correlation with coil velocity, and so can often be perceived as being separate from the reproduced sound source - hence produce a different subjective impression.
I have tried to avoid providing subjective assessments here and I welcome people to make their own judgements. Typically third order displacement non-linearities can be inaudible, whereas the same level of third harmonic distortion from eddy-related distortion can be recognised easily. Typically listeners report a "noise" or a "glare" or "haze". The "noise", for example, is heard as being separate from the reproduced sound source.
FWIW and IMHO this particular distortion reminds me of crossover distortion in poorly engineered class B amplifers or quanitization distortion in undithered digital processing.
Last edited:
The issue I was referring to is that the requirements for stereo are frequency dependent. In the early days of stereo, EMI incorprated "Stereosonic" shuffling, but such processing has largely ceased. Capacitive coupling between left and right channels can often provide for a similar effect, but would afford a significantly worse channel separation measure.
Purifi surround was mentioned earlier and Lars Risbo from Purifi comments it's to tackle Sd(x), which is basically amplitude modulation distortion.
https://www.diyaudio.com/community/...istortion-speaker-drivers.294787/post-7112912
Here is more about their distortion stuff https://purifi-audio.com/ushindi , look more from the papers and blog on their site. Of course Klippel has made lot's of papers and knowledge available about distortion mechanisms in transducers, perhaps many many others as well, I'm not industry pro so cannot enumerate any others
Overall, since Geddes found out THD number doesn't correlate with perceived sound quality, still some people obsess about distortion, there is a lot to the subject. Best thing would be to identify some offensive distortion from measurements, and not stress about distortion that is not audible, or perhaps even preferred. In my opinion, distortion doesn't seem to be a problem when speaker system is designed with some other goals, like being big enough for party SPL levels and address integration with room, distortion doesn't seem to be any kind of issue as byproduct. If speakers must be small, and cheap, and play 20Hz and so on, then it would be critical to evaluate distortion. Perhaps it is for any system, it just isn't that high in priority after everything else is addressed, just pay attention to it, but don't over stress about it would be the general advice from me. Perhaps the main nuget is to think that low frequency sounds make greatest excursion, which then makes the whole bandwidth of the transducer to distort.
https://www.diyaudio.com/community/...istortion-speaker-drivers.294787/post-7112912
Here is more about their distortion stuff https://purifi-audio.com/ushindi , look more from the papers and blog on their site. Of course Klippel has made lot's of papers and knowledge available about distortion mechanisms in transducers, perhaps many many others as well, I'm not industry pro so cannot enumerate any others
Overall, since Geddes found out THD number doesn't correlate with perceived sound quality, still some people obsess about distortion, there is a lot to the subject. Best thing would be to identify some offensive distortion from measurements, and not stress about distortion that is not audible, or perhaps even preferred. In my opinion, distortion doesn't seem to be a problem when speaker system is designed with some other goals, like being big enough for party SPL levels and address integration with room, distortion doesn't seem to be any kind of issue as byproduct. If speakers must be small, and cheap, and play 20Hz and so on, then it would be critical to evaluate distortion. Perhaps it is for any system, it just isn't that high in priority after everything else is addressed, just pay attention to it, but don't over stress about it would be the general advice from me. Perhaps the main nuget is to think that low frequency sounds make greatest excursion, which then makes the whole bandwidth of the transducer to distort.
Last edited:
Yes, but MATLAB in itself is not an audio programme, but rather an advanced mathematical tool. In its own it will do nothing to my best of knowledge, as opposed to e.g. ARTA, CLIO, REW e.a.. So we need stuff to run MATLAB.
Just try for yourself! Pick a piano track, for instance, and listen to the leading transient edge of the notes. Once heard, you will recognize the sonic signature ever more clearly. Comparing the same piece using an ESL might prove enlightening too if you have one available.
As far as accountable design goes, you will find driver manufacturers now go to considerable efforts to reduce this particular distortion via copper pole pieces and the like (which is the easiest and cheapest way forward, if not the best). At least they do in their more expensive drivers. Do you think manufacturers would go to such efforts if there was no sellable improvement to be gained?
Hi soundbloke,
Separation was part of the huge CD vs record player debate.
Strongly agree! Here is where an accurate system with high efficiency shows it's worth. Actively crossed, it over avoids losses in the crossover. Heating voice coils, I have measured a 2 dB drop in efficiency. That was not running the drivers at maximum power. The commercial example of the first system I've heard that sounds real is the Klipsch Jubilee.
Commercial use of engineering data has always been heavily filtered in every field. The average person who is not a professional in a field tends to seize on ideas and points of data that don't reflect the big picture reality. You see this in any forum. I greatly prefer talking to my peers where information is accepted with interest and professional curiosity. There is a good reason why information is filtered for the public, but unfortunately partial information is always put out there in order to support a commercial goal.
Channel separation. Hmmm, I would say that the maximum separation should be maintained in the reproduction chain up to the speakers. The recording mix should be preserved so at least you have a known by the time you reach the room where the speakers are. 30 dB is the approximate figure for good stereo separation, 25 dB and up for good phono cartridges, 35 dB and up for FM stereo is common. Yes, there have been products that allow the end user to "mess with" the separation. Bob Carver did a fair amount of that in his consumer products for example. Listening room separation will vary, but the real determining factor ought to be the recording engineer and producer. Mixing studios have speakers to approximate near field and large speaker systems so they can target "the average" listening environment. They have no hope of replicating every situation, but they try. We can't second guess them, and we just do our best with the listening environment. Most electronics can easily support that figure to the extent it doesn't degrade it much.
Separation was part of the huge CD vs record player debate.
Yes it was - and yet another nonsense! But as noted in post #208, it is not the magnitude that matters so much as its change with frequency.
You are implying that I have an alternative to the standard point of view. I don't.
Unfortunately, straw man arguments creating "alternative" points of view is an irritating habit some on this forum have.
Don't apologize. Search for it and read it yourself.
Dave.
The question "do measurements matter" is in my experience the most important question to be asked and the most difficult to answer.
This very simple question leads to very complex answers that are the keys to good loudspeaker design.
A primary propose in a sound reproduction chain in the context of high end audio is to induce a pleasant experience in the listener. Measurements are useful to the degree that they help you move closer to that goal of a pleasant listener experience. Measurements are harmful to the degree they distract you from reaching a improved listener experience. If this sounds trite long experience has taught me the question asked is in fact most fundamental in importance to creating superior loudspeakers. The sound reproduction chain includes the listener and this bring psychoacoustics into the what must be considered in producing a pleasant listener experience. No two humans experience sound in the exact same way and so it is a perfectly valid and expected that there will be differences in opinion of what a pleasant listener experience is between any two "samples" of humans. The correct "target" sound for any sound reproduction system must then be recognized as the likely experience of a average of listeners not any one absolute "correct" sound or listener. Our difficulty in the acceptance of listener variation causes much friction and distraction in the audio industry resulting in "warring camps" with the only "correct" answers.
This is not to say measurement are unimportant, far from it they are our only path to improvement however how to effectively use measurements is devilishly complex.
I find it helpful to reflect on audio reproduction by considering the human sense of taste.
Most people will agree that there is no "correct" way everyone should experience any one dish of food.
Our sense of hearing like taste is deeply involved in the human experience and so is deeply individual.
Consider a good chef. They must know the properties of their ingredients in great detail.
They understand how ingredients interact with each other (often down to a chemical level) and the resulting affects on the dinners experience. And they will personally taste the dish as it progresses and use their human experience of tasting to guide the cooking process. Lastly it is accepted that not everyone will enjoy any one dish.
Ok I have taken a long time getting here. The question asked "do measurements matter" in my experience needs to be re-framed to "what measurements matter, in what order of priority do they matter and how do the impairments measured interact"
Frequency response is at the top of the list and a monotonic curve is a very important indicator of loud speaker performance. So loudspeaker A with a more limited frequency range and a monotonic curve would be in preference to loudspeaker B with a wider but bumpy frequency response. This is because non-monotonic curves usually indicate underlying resonance's in the loudspeaker or enclosure design.
Published curves of driver frequency response from manufactures are heavily influenced by test conditions and so it is unfortunately hard to compare two drivers from published curves. In the end however wide monotonic curves are what is desired.
This leads to what in my experience is the second most important aspect, that of transient non-linear distortions as discussed by mikets42 and others.
Useful data from manufactures is even less available for this type of impairment. Again test conditions strongly influence results making between manufacture comparisons less useful. In speakers transient non-linear distortions are primarily the result of driver and enclosure issues however a test setup to produce useful results can be challenging.
I have compared measurements of transient non-linear distortions in both a anechoic chambers and real rooms and in my experience measurement in normal rooms are less useful as room reflections and noise degrade the data. I know the makers of these systems state the software will eliminate room reflection but my experience is to take such claims with a grain of salt. Near-field measurements are normally used in real rooms to improve the signal to noise however real driver radiation patterns are not very often unchanged between near filed and a few meters back from the driver or loudspeaker. So your data may be accurate if you listen to the speaker a few inches from the driver but how often is that a realistic listening position?
Third in importance is harmonic distortion. Surprisingly it can be harder to produce distortion measurements that relate to the listener experience than one would wish. Total distortion is of very limited value as it is the harmonic makeup that is the key. Tiny amounts of energy in the harmonic distortion are surprising audible created by for example air currents in the magnet gap as the voice coil moves, rustling noises caused by the spider flexing, little sounds made by the tinsel leads bending as the cone moves and air movements caused as the surround changes shape and thus the volume of air it holds changes. There are many more with few of these being able to captured on measurement systems to the level sufficiently low enough to be below what a experienced designer can hear with a unaided ear. Sure the current measurements systems can capture many of theses effects at high or gross levels but we drop into the "dirt" of the data well before a experienced designer can no long hear the problem.
So measurements are very important as long as we keep in mind how incredibly limited they can be in going the last few miles toward that pleasant listener experience.
This very simple question leads to very complex answers that are the keys to good loudspeaker design.
A primary propose in a sound reproduction chain in the context of high end audio is to induce a pleasant experience in the listener. Measurements are useful to the degree that they help you move closer to that goal of a pleasant listener experience. Measurements are harmful to the degree they distract you from reaching a improved listener experience. If this sounds trite long experience has taught me the question asked is in fact most fundamental in importance to creating superior loudspeakers. The sound reproduction chain includes the listener and this bring psychoacoustics into the what must be considered in producing a pleasant listener experience. No two humans experience sound in the exact same way and so it is a perfectly valid and expected that there will be differences in opinion of what a pleasant listener experience is between any two "samples" of humans. The correct "target" sound for any sound reproduction system must then be recognized as the likely experience of a average of listeners not any one absolute "correct" sound or listener. Our difficulty in the acceptance of listener variation causes much friction and distraction in the audio industry resulting in "warring camps" with the only "correct" answers.
This is not to say measurement are unimportant, far from it they are our only path to improvement however how to effectively use measurements is devilishly complex.
I find it helpful to reflect on audio reproduction by considering the human sense of taste.
Most people will agree that there is no "correct" way everyone should experience any one dish of food.
Our sense of hearing like taste is deeply involved in the human experience and so is deeply individual.
Consider a good chef. They must know the properties of their ingredients in great detail.
They understand how ingredients interact with each other (often down to a chemical level) and the resulting affects on the dinners experience. And they will personally taste the dish as it progresses and use their human experience of tasting to guide the cooking process. Lastly it is accepted that not everyone will enjoy any one dish.
Ok I have taken a long time getting here. The question asked "do measurements matter" in my experience needs to be re-framed to "what measurements matter, in what order of priority do they matter and how do the impairments measured interact"
Frequency response is at the top of the list and a monotonic curve is a very important indicator of loud speaker performance. So loudspeaker A with a more limited frequency range and a monotonic curve would be in preference to loudspeaker B with a wider but bumpy frequency response. This is because non-monotonic curves usually indicate underlying resonance's in the loudspeaker or enclosure design.
Published curves of driver frequency response from manufactures are heavily influenced by test conditions and so it is unfortunately hard to compare two drivers from published curves. In the end however wide monotonic curves are what is desired.
This leads to what in my experience is the second most important aspect, that of transient non-linear distortions as discussed by mikets42 and others.
Useful data from manufactures is even less available for this type of impairment. Again test conditions strongly influence results making between manufacture comparisons less useful. In speakers transient non-linear distortions are primarily the result of driver and enclosure issues however a test setup to produce useful results can be challenging.
I have compared measurements of transient non-linear distortions in both a anechoic chambers and real rooms and in my experience measurement in normal rooms are less useful as room reflections and noise degrade the data. I know the makers of these systems state the software will eliminate room reflection but my experience is to take such claims with a grain of salt. Near-field measurements are normally used in real rooms to improve the signal to noise however real driver radiation patterns are not very often unchanged between near filed and a few meters back from the driver or loudspeaker. So your data may be accurate if you listen to the speaker a few inches from the driver but how often is that a realistic listening position?
Third in importance is harmonic distortion. Surprisingly it can be harder to produce distortion measurements that relate to the listener experience than one would wish. Total distortion is of very limited value as it is the harmonic makeup that is the key. Tiny amounts of energy in the harmonic distortion are surprising audible created by for example air currents in the magnet gap as the voice coil moves, rustling noises caused by the spider flexing, little sounds made by the tinsel leads bending as the cone moves and air movements caused as the surround changes shape and thus the volume of air it holds changes. There are many more with few of these being able to captured on measurement systems to the level sufficiently low enough to be below what a experienced designer can hear with a unaided ear. Sure the current measurements systems can capture many of theses effects at high or gross levels but we drop into the "dirt" of the data well before a experienced designer can no long hear the problem.
So measurements are very important as long as we keep in mind how incredibly limited they can be in going the last few miles toward that pleasant listener experience.
Alas mike, as a real beach bum, I can't afford $100. I don't even have copies of my own AES papers due to a number of HD crashes. I only link to our AES papers cos they explain in much more detail where I'm coming from.
But I AM interested in your AEC measurement. Could you at least give me a hint of what it is? I'm spent most of my previous life dreaming up 'advanced' measurements and trying to correlate them to good sounding speakers.
There is a way to use Angelo's method while playing music on a system but reading between the lines, I don't think this is what you are doing.