here's a new thread to discuss this topic and not continue the "threadjacking" in the "The best cabinet material" thread.
anyone care to weigh in on this topic?
anyone care to weigh in on this topic?
I did extensive testing of a variety of HF compression drivers in these threads:
http://www.diyaudio.com/forums/multi-way/212240-high-frequency-compression-driver-evaluation.html
http://soundforums.net/varsity/4329-high-frequency-compression-driver-evaluation.html
The posts include music and test tone recordings of the various drivers at levels ranging from typical home use to full power more typical of PA use.
Those tests all used one type of horn, it would be interesting to conduct the inverse test, one driver on different types of horns, equalized flat on axis, recorded at different levels.
Art
thanks for the submission of "prior art" Art (that makes me feel like i'm stuttering)
i guess i'd better put coffee on this is going to take a while to go through but i asked for it.
(where's Mr Geddes)
i guess i'd better put coffee on this is going to take a while to go through but i asked for it.
(where's Mr Geddes)
Last edited:
Will be interested to hear what you think of the sound of the drivers compared to the original at various levels, and at what drive levels you find the distortion becomes objectionable.
I don't have the time right now, but at some point I will review the work that I have done on this topic.
ok coffee in hand. can someone help me out with understanding how to change file extensions (some of this stuff is foreign to me i'm a analog neanderthal in a digital world! also the reason i love old analog things like magnetic tape and vinyl)
This example is Win7, right click a file and mark "Rename" then Windows will color mark normal file name it expects to be edited, to get free of this use right arrow to get to extension which is letters after dot and change this combination to target extension which as in attached example below is "mp3".
If you don't see dot and extension, then first go to control panel and under folder options remove setting saying something ala "hide file extensions".
Hope it works as above and is understandable, because i have to translate my Danish Windows dialogs to English : )
If you don't see dot and extension, then first go to control panel and under folder options remove setting saying something ala "hide file extensions".
Hope it works as above and is understandable, because i have to translate my Danish Windows dialogs to English : )
Attachments
Last edited:
I have a few minutes so I will get started. I will do this in sections, because it is safer and I don't have the time to do it all at once.
Many decades ago I believed that nonlinear distortion was the limiting factor in all loudspeaker designs. That if we could just get things more linear then everything would be fine. So pursuant to this belief I studied nonlinear dynamics in great depth, including a graduate course in nonlinear acoustics (a big deal in underwater because water is about 1000 time more nonlinear than air), and reading the nonlinear bible by Schetzen "The Volterra & Wiener Theories of Nonlinear Systems", which is still the most complete text of its kind available today.
One of the first things that struck me was how much more complex nonlinear systems are than linear ones, at least an order of magnitude more complex. The possibilities for misunderstanding these systems seemed ripe and I have found that this is the case.
One experiment that was shown to me early on that never left my mind was how an electric circuit with << 1% THD could sound so bad, while we were dealing with speakers at 5-10% that didn't sound so bad. This is a key factor to keep in mind throughout this discussion.
In about the late 80's I got interested in "perceptual coding" i.e. MP3, etc. I was amazed at how understanding how we hear could allow one to reduce the data by as much as 90% and still not be audible (for the most part). It became clear to me that one absolutely had to take into account the hearing mechanism if one wanted to understand how NLD (Nonlinear Distortion) was perceived.
While I was at Knowles I looked at the perception of amplifier distortion and found that a NLD amp could actually sound better, certainly louder, than its perfectly linear cousin. Clearly something was not as it first seemed.
After I left Knowles, Lidia, my wife, and I setup a test to correlate some metrics (THD IMD, GedLee, etc.) with the perception of numerous nonlinear systems.
Got to go!
Many decades ago I believed that nonlinear distortion was the limiting factor in all loudspeaker designs. That if we could just get things more linear then everything would be fine. So pursuant to this belief I studied nonlinear dynamics in great depth, including a graduate course in nonlinear acoustics (a big deal in underwater because water is about 1000 time more nonlinear than air), and reading the nonlinear bible by Schetzen "The Volterra & Wiener Theories of Nonlinear Systems", which is still the most complete text of its kind available today.
One of the first things that struck me was how much more complex nonlinear systems are than linear ones, at least an order of magnitude more complex. The possibilities for misunderstanding these systems seemed ripe and I have found that this is the case.
One experiment that was shown to me early on that never left my mind was how an electric circuit with << 1% THD could sound so bad, while we were dealing with speakers at 5-10% that didn't sound so bad. This is a key factor to keep in mind throughout this discussion.
In about the late 80's I got interested in "perceptual coding" i.e. MP3, etc. I was amazed at how understanding how we hear could allow one to reduce the data by as much as 90% and still not be audible (for the most part). It became clear to me that one absolutely had to take into account the hearing mechanism if one wanted to understand how NLD (Nonlinear Distortion) was perceived.
While I was at Knowles I looked at the perception of amplifier distortion and found that a NLD amp could actually sound better, certainly louder, than its perfectly linear cousin. Clearly something was not as it first seemed.
After I left Knowles, Lidia, my wife, and I setup a test to correlate some metrics (THD IMD, GedLee, etc.) with the perception of numerous nonlinear systems.
Got to go!
So somewhere about 2000, Lidia and I did a test comparing three different metrics of about 40 different nonlinear "types" using about 40 subjects (both expert and novice - curiously enough the experts in this test were not any better than the novices.). The evaluations were all double blind and computer driven. The three metrics were THD, IMD and a metric that was derived based on psychoacoustic principles of masking.
The original papers are on my website, I believe, if not I can put them there. The bottom line is that neither THD or IMD had any correlation with the subjective preference of the material and, in fact, THD showed a weak negative correlation to preference, i.e. the more distortion the better people liked it, but this was not significant. The psychoacoustic based metric had a high degree of correlation and especially so if the extreme values of very severe distortion were eliminated.
What these papers showed was that the usually measures of NLD are useless, but that it is possible to quantify NLD in a way that has meaning - but nobody does that and the calculations to get this valid metric are not trivial.
What struck me most of all was the fact that the types of NLD that were the most audible - at very low levels of THD - were likely to only occur in electronics and, in fact, are quite common and quite insidious in that domain. But in the mechanical domain the kinds of NLD that loudspeakers have, are extremely benign. Basically it appeared that NLD perception in loudspeakers was not likely to be a significant effect.
I need to point out here that it is not as if loudspeakers cannot have perceptible NLD, they can, its that the most common forms of NLD are not that problematic. And in fact, it is actually not too hard to eliminate the more problematic forms, such as flux modulation and Inductance variations. BL and excursion based NLD does not seem to ever be a big issue, and can always be avoided by just not using a driver with too little excursion capability.
One can always make a loudspeakers NLD audible, but one can just as easily make one without it.
Next will be the B&C study of NLD in compression drivers.
The original papers are on my website, I believe, if not I can put them there. The bottom line is that neither THD or IMD had any correlation with the subjective preference of the material and, in fact, THD showed a weak negative correlation to preference, i.e. the more distortion the better people liked it, but this was not significant. The psychoacoustic based metric had a high degree of correlation and especially so if the extreme values of very severe distortion were eliminated.
What these papers showed was that the usually measures of NLD are useless, but that it is possible to quantify NLD in a way that has meaning - but nobody does that and the calculations to get this valid metric are not trivial.
What struck me most of all was the fact that the types of NLD that were the most audible - at very low levels of THD - were likely to only occur in electronics and, in fact, are quite common and quite insidious in that domain. But in the mechanical domain the kinds of NLD that loudspeakers have, are extremely benign. Basically it appeared that NLD perception in loudspeakers was not likely to be a significant effect.
I need to point out here that it is not as if loudspeakers cannot have perceptible NLD, they can, its that the most common forms of NLD are not that problematic. And in fact, it is actually not too hard to eliminate the more problematic forms, such as flux modulation and Inductance variations. BL and excursion based NLD does not seem to ever be a big issue, and can always be avoided by just not using a driver with too little excursion capability.
One can always make a loudspeakers NLD audible, but one can just as easily make one without it.
Next will be the B&C study of NLD in compression drivers.
Last edited:
Thanks for that Byrtt - I see the example is one of my recordings from the 10F FAST 🙂
How do you do this on a Mac? Or an iPhone? I don't think it possible to change the extension on a file downloaded on an iPhone. Lame isn't it?
Back to the topic ...
When I started working with B&C we wanted to know how to improve the compression drivers. So we setup a test comparing different levels of distortion and different drivers in a nine position array of sources and levels. The results of this work were published in the JAES, sometime in the mid 2000's. The test was double blind and had bout 30 students with "normal" hearing rate each of the nine examples. In these examples the levels were not excessive since we were measuring in a plane wave tube and the SPL's were going above what the smaller mics could handle. 28 volts was the maximum (about 50 watts, 16 ohms) which is not out of reach for a 2" throat driver. Hence this was not a test at the drivers limits, but more like what one might find in a home (even though this was not B&Cs market - well tests never come out like you expect them to!)
A multi-dimension ANOVA ( I think that's right) test was performed on the results. The results said that there was no effect on perception with level, i.e. nonlinearity, but there was across the drivers. This meant that the test was not null in that differences were statistically significant among the tests, just not across level. There was also no inter- dependency of the variables.
Say what you will about this test, within the range of its applicability the results are significant. It isn't going to hold for very high levels of sound, but they will hold for normal home levels. Since there was correlation between drivers the test signal was sufficient to show significant differences.
This test is consistent with the idea that nonlinearities in mechanical devices is not going to be a significant differentiator except possibly at fairly high level.
It is also interesting to note that Toole has never considered nonlinear distortion to be a "significant factor" - he said to me once, and I quote (to the best of my ability) "Nonlinear distortion is sound systems got the coverage that it deserves in my book - about one paragraph." (out of several hundred pages. Sean olive is of the same opinion as in Alex Voishvillo at JBL. The more people study this "problem," the less important it seems to become. I pay no attention to it anymore, design it out with a few tricks and forget about it.
When I started working with B&C we wanted to know how to improve the compression drivers. So we setup a test comparing different levels of distortion and different drivers in a nine position array of sources and levels. The results of this work were published in the JAES, sometime in the mid 2000's. The test was double blind and had bout 30 students with "normal" hearing rate each of the nine examples. In these examples the levels were not excessive since we were measuring in a plane wave tube and the SPL's were going above what the smaller mics could handle. 28 volts was the maximum (about 50 watts, 16 ohms) which is not out of reach for a 2" throat driver. Hence this was not a test at the drivers limits, but more like what one might find in a home (even though this was not B&Cs market - well tests never come out like you expect them to!)
A multi-dimension ANOVA ( I think that's right) test was performed on the results. The results said that there was no effect on perception with level, i.e. nonlinearity, but there was across the drivers. This meant that the test was not null in that differences were statistically significant among the tests, just not across level. There was also no inter- dependency of the variables.
Say what you will about this test, within the range of its applicability the results are significant. It isn't going to hold for very high levels of sound, but they will hold for normal home levels. Since there was correlation between drivers the test signal was sufficient to show significant differences.
This test is consistent with the idea that nonlinearities in mechanical devices is not going to be a significant differentiator except possibly at fairly high level.
It is also interesting to note that Toole has never considered nonlinear distortion to be a "significant factor" - he said to me once, and I quote (to the best of my ability) "Nonlinear distortion is sound systems got the coverage that it deserves in my book - about one paragraph." (out of several hundred pages. Sean olive is of the same opinion as in Alex Voishvillo at JBL. The more people study this "problem," the less important it seems to become. I pay no attention to it anymore, design it out with a few tricks and forget about it.
Last edited:
You spotted it 🙂 as time go really have a lot X-made sound clips.
My daughters have Mac stuff but not me, find it weird if those devices can't change extension or it maybe needs cost for a app or cloud service to be possible.
I think one important thing to note about the Klippel test for distortion audibility is that the population sample is huge and the test is double blind with comparison to reference with no distortion in binary choice. I think the results are quite statistically significant and an important finding in audio. There is definitely a majority of people who can't hear the difference with 10% or less HD. But there are quite a few people who can detect much much lower levels. For that population, speakers and systems with low HD are more important. I was quite surprised I could detect it down to -45dB. But this confirms to me it's importance - and words like distortion is "meaningless" is truly not applicable.
Nonlinear distortion is a major issue in loudspeakers, both domestic and stage/pro/club.
If this wasn't the case we would see nothing bigger than say a two way speaker with a 3.5" woofer in all homes, cars, venues.
The fact that it's hard or impossible to fully correlate THD or a few HD numbers with subjective impressions is another discussion. However, a HD curve still gives a good indication of what can be expected.
If this wasn't the case we would see nothing bigger than say a two way speaker with a 3.5" woofer in all homes, cars, venues.
The fact that it's hard or impossible to fully correlate THD or a few HD numbers with subjective impressions is another discussion. However, a HD curve still gives a good indication of what can be expected.
For the most part such a MP3 have very audible artefacts, on most types of music, and for most listeners with normal hearing and the slightest interest in good sounding music. If this is your experience I understand if you question nonlinear distortion. Maybe you have suffered hearing damage?
I don't think many experienced audio professionals argue with the fact that distortion can make some stuff sound better or more pleasing. But is that what we are dealing with? I thought high performance audio reproduction aka hifi was about removing colorations to let the production (music material) shine itself?
Last edited:
I did the Klippel test several years ago and if memory serves me I ended up at -45dB as well.
I have found that if measurements show more than 0.6 % third harmonic distortion from compression driver in a horn at 2.83 V, than it is audible and it is bad at "normal" PA level. Second harmonic distortion can be much higher, but it never sounds bad, just adds more "color".
Last edited:
The modulation distortion from a strong asymmetri (aka "2nd") can be nasty though. The old "even order is pleasing" idea is applicable on simple material in best case.. ie solo instrument and similar.
I agree with Earl, NLD is for the most part a non issue, unless one pushes the driver beyond its intended use. For PA use it can become an issue but at home, nope. For a CD used as intended I just cannot see how NLD could become a problem unless it has serious design flaws or is broken. It is of course possible to read Earls post like the Devil reads the Bible but the point being made regarding the audability of mp3 compression is very valid. The mere fact that with good understanding of hearing perception we can remove as much as 90% of the data and still have very listenable results is good food for thought. There are many factors to take into account when designing a loudspeaker and NLD happens to be one of the easiest to push down on the priority list of major concerns.
/Anders
/Anders
Please post 128kbit/s MP3's without audible artefacts and I'll run them thru ABX to prove you wrong. 🙂
By no means I'm against MP3, I use to "defend" MP3 as being good enough in many situations and more or less transparent in some cases.
It was some time ago I made MP3 testing and possibly better algos are available now.
Well isn't that the point? It's easy to end up in a situation where you push the speaker to audible distortion and then you either need a better design for the size or a bigger speaker. I think that pretty much is end of discussion on that point since obviously linearity matters. Also the Klippel test is in contrast to your position.
By no means I'm against MP3, I use to "defend" MP3 as being good enough in many situations and more or less transparent in some cases.
It was some time ago I made MP3 testing and possibly better algos are available now.
Well isn't that the point? It's easy to end up in a situation where you push the speaker to audible distortion and then you either need a better design for the size or a bigger speaker. I think that pretty much is end of discussion on that point since obviously linearity matters. Also the Klippel test is in contrast to your position.
- Status
- Not open for further replies.