Hi Hifijim,
No, I am still at a loss. Theoretically, the right way is Principal Component Analysis (PCA) but I have no idea how to account for the hearing sensitivities and Barkhausen noise. Judging by the approach chosen by the recording industry in the 70s - the "wall of sound" - to make the music listenable on average consumer equipment, the major problems are dynamics and spectral gaps being filled with whatever discrepancies from the original. Thus, these aspects shall be accentuated by the proper approach to quantifying the drivers' distortions. Much easier to say than done.
Hi Boden,
I think that the current approach of "program control" when you feed voltage and do not care what comes out acoustically has dead-ended. To avoid Barkhausen and hysteresis discontinuities, we'll need to get rid of permanent magnets and create the magnetic field by driving induction coils with pure current, and driving the voice coil with pure current too while performing proper motion feedback instead of shorting rings. In other words, a bunch of nested wideband feedback loops. Efficiency shall jump up to be electric motor alike 85%, if class D is used throughout. It's hard to estimate how many GWH of power could be saved over the world. Current drivers are in reality voltage drivers with feedback on current sensor (0R1 or so). Because of the need for deep feedback, you need a lot of bandwidth, therefore everything must be very close together and ideally fully integrated. Just dreaming...
No, I am still at a loss. Theoretically, the right way is Principal Component Analysis (PCA) but I have no idea how to account for the hearing sensitivities and Barkhausen noise. Judging by the approach chosen by the recording industry in the 70s - the "wall of sound" - to make the music listenable on average consumer equipment, the major problems are dynamics and spectral gaps being filled with whatever discrepancies from the original. Thus, these aspects shall be accentuated by the proper approach to quantifying the drivers' distortions. Much easier to say than done.
Hi Boden,
I think that the current approach of "program control" when you feed voltage and do not care what comes out acoustically has dead-ended. To avoid Barkhausen and hysteresis discontinuities, we'll need to get rid of permanent magnets and create the magnetic field by driving induction coils with pure current, and driving the voice coil with pure current too while performing proper motion feedback instead of shorting rings. In other words, a bunch of nested wideband feedback loops. Efficiency shall jump up to be electric motor alike 85%, if class D is used throughout. It's hard to estimate how many GWH of power could be saved over the world. Current drivers are in reality voltage drivers with feedback on current sensor (0R1 or so). Because of the need for deep feedback, you need a lot of bandwidth, therefore everything must be very close together and ideally fully integrated. Just dreaming...
Field coil drivers are an old invention..
But most hifi loudspeakers have 3-4 different type of transducers. My AINOgradient's sub has ferrite, woofer neodymium, mid and tweeter are magnetoplanars with neo magnets...
I haven't grasped how mikets method handles different crossover types and imperfections.
But most hifi loudspeakers have 3-4 different type of transducers. My AINOgradient's sub has ferrite, woofer neodymium, mid and tweeter are magnetoplanars with neo magnets...
I haven't grasped how mikets method handles different crossover types and imperfections.
@ mikets42
These published LTI plots are very, very dense in terms of infomation, and as you state, one better "... avoid any detailed interpretation of my graphs due to the danger of introducing a personal bias ...". So these graphs may be good-looking, but as is, they seem definitively too dense for some in-depth precise differential analysis/diagnostics. As you say in German: Man sieht vor lauter Bäumen den Wald nicht mehr. One can't see the forest anymore because of these lots of trees.
Maybe there might be a need of diluting/sorting out all this density. I guess that the plots shown dispay a superposition of linear (LD) and non-linear distortion products (NLD). If so, then it might be useful to generate two plots instead of a single one, if the method allows to do so: A plot showing the LD only, and another plot selectively showing the NLD.
I guess this would certainly dramatically reduce the information density, especially for the NLD graph. By the way: Which is the typical LD/NLD ratio within a typical speaker driver? 20dB or so? I ask, because I have the impression that the display of the NLD gets swamped by the display of LD in your graphs. Anyway and as for me, I do interprete a lot of LD in them, maybe because these graphs seem so familiar to me, resembling plots which can be generated from traditional linear measuring methods. I might be wrong.
Maybe even a simple trick could do? A maybe naive analogy: Performing your LTI method, you already successfully subtracted [Mozart] from [Mozart+LD+NLD] = [LD+NLD]. So what about further subtracting [LD] from [LD+NLD] = [NLD]? Presuming, that NLD display/diagnostics in your graphs are intrinsically hampered by LD >> NLD.
These published LTI plots are very, very dense in terms of infomation, and as you state, one better "... avoid any detailed interpretation of my graphs due to the danger of introducing a personal bias ...". So these graphs may be good-looking, but as is, they seem definitively too dense for some in-depth precise differential analysis/diagnostics. As you say in German: Man sieht vor lauter Bäumen den Wald nicht mehr. One can't see the forest anymore because of these lots of trees.
Maybe there might be a need of diluting/sorting out all this density. I guess that the plots shown dispay a superposition of linear (LD) and non-linear distortion products (NLD). If so, then it might be useful to generate two plots instead of a single one, if the method allows to do so: A plot showing the LD only, and another plot selectively showing the NLD.
I guess this would certainly dramatically reduce the information density, especially for the NLD graph. By the way: Which is the typical LD/NLD ratio within a typical speaker driver? 20dB or so? I ask, because I have the impression that the display of the NLD gets swamped by the display of LD in your graphs. Anyway and as for me, I do interprete a lot of LD in them, maybe because these graphs seem so familiar to me, resembling plots which can be generated from traditional linear measuring methods. I might be wrong.
Maybe even a simple trick could do? A maybe naive analogy: Performing your LTI method, you already successfully subtracted [Mozart] from [Mozart+LD+NLD] = [LD+NLD]. So what about further subtracting [LD] from [LD+NLD] = [NLD]? Presuming, that NLD display/diagnostics in your graphs are intrinsically hampered by LD >> NLD.
Hi Daihedz,
It is only NLD. LD graphs are omitted (they can be obtained in other, much simpler ways). Yes, it's complicated. If it were simple, the problem of loudspeaker distortions would have been solved ages ago.
Hi Juhazi,
My method is essentially System Identification by Regularized Recursive Least Squares, performed in a certain transform domain to constrain eigenvalue spectrum spread and lower the MIPS. It does not require any special input like MLS or sine wave sweep, the results of which are hard to extrapolate to the performance on music. It can use the music itself as input and show the NLD on this particular piece of music. It analyses the system as a Single Input Single Output (SISO) black box. It does not care if the system is sonar / radar / room / plate tectonics / semiconductor circuitry / etc physically. It does not distinguish between crossovers or loudspeakers, between motor or suspension, etc. Ok?
It is only NLD. LD graphs are omitted (they can be obtained in other, much simpler ways). Yes, it's complicated. If it were simple, the problem of loudspeaker distortions would have been solved ages ago.
Hi Juhazi,
My method is essentially System Identification by Regularized Recursive Least Squares, performed in a certain transform domain to constrain eigenvalue spectrum spread and lower the MIPS. It does not require any special input like MLS or sine wave sweep, the results of which are hard to extrapolate to the performance on music. It can use the music itself as input and show the NLD on this particular piece of music. It analyses the system as a Single Input Single Output (SISO) black box. It does not care if the system is sonar / radar / room / plate tectonics / semiconductor circuitry / etc physically. It does not distinguish between crossovers or loudspeakers, between motor or suspension, etc. Ok?
You are absolutely correct. It is not a silver bullet, at all. It only gives an observation. It does not pinpoint the source of the defect. That's what you can do yourself, by making incremental changes and comparing the results.
No, I have not. Actually, you do not need an anechoic chamber for using this method.
Hi mikets42,
Gee, I've been in the CN Tower transmitting room and looked out over Toronto through the top plates on the tower. It's really warm in the room, but the view is amazing!
I disagree with your stance that the various sine and other measurements are valueless. In the beginning, we were able to see frequency response and resonances. Useful? You can clearly see linearity using a sine wave and an oscilloscope. That's how I test loudspeakers for defects and how I've done it for decades. It works - well.
Not once have I ever said I use only simple single sine wave testing. I use every method available. And yes, if you can hear it, it is very easily measured down to levels far below our ability to sense it. Our human bodies only have so much sn ratio, and ambient noise restricts the dynamic range even more. Those are the realities of the human body and our environment. In special, quiet locations we can extend our dynamic range (as is commonly done), but instruments can still detect and quantify well below what can be heard. Period.
I'm pretty sure you'll agree that the price of a speaker system does not guarantee any level of quality or performance, same for the electronics. Of course, if something doesn't have enough money invested in manufacture, material and design it cannot perform well. But greatly exceeding this expenditure absolutely does not make anything better. So this makes it very difficult for the average person to know they are getting what they paid for.
Basically, yes. People do listen as a final judge of quality. But there are so many excellent examples where this does not lead them to a good system. I have yet to hear anything designed without careful measurements that actually does sound good. People will think something does until they hear whatever it is that really does perform well. They will also measure well by the way. From my experience, you have to both measure and listen. I have always and consistently said this. Look back through my entire history here or ask anyone who does know me.
All I am going to say is that to build the best, you have to use every measurement method available to you. Some tell you the same information presented in different ways, but the tests are extremely valuable. To ignore testing (measurements) is to throw away incredibly useful information, and that isn't very intelligent. Yes, of course you listen to it as well.
Gee, I've been in the CN Tower transmitting room and looked out over Toronto through the top plates on the tower. It's really warm in the room, but the view is amazing!
I disagree with your stance that the various sine and other measurements are valueless. In the beginning, we were able to see frequency response and resonances. Useful? You can clearly see linearity using a sine wave and an oscilloscope. That's how I test loudspeakers for defects and how I've done it for decades. It works - well.
Not once have I ever said I use only simple single sine wave testing. I use every method available. And yes, if you can hear it, it is very easily measured down to levels far below our ability to sense it. Our human bodies only have so much sn ratio, and ambient noise restricts the dynamic range even more. Those are the realities of the human body and our environment. In special, quiet locations we can extend our dynamic range (as is commonly done), but instruments can still detect and quantify well below what can be heard. Period.
I'm pretty sure you'll agree that the price of a speaker system does not guarantee any level of quality or performance, same for the electronics. Of course, if something doesn't have enough money invested in manufacture, material and design it cannot perform well. But greatly exceeding this expenditure absolutely does not make anything better. So this makes it very difficult for the average person to know they are getting what they paid for.
Basically, yes. People do listen as a final judge of quality. But there are so many excellent examples where this does not lead them to a good system. I have yet to hear anything designed without careful measurements that actually does sound good. People will think something does until they hear whatever it is that really does perform well. They will also measure well by the way. From my experience, you have to both measure and listen. I have always and consistently said this. Look back through my entire history here or ask anyone who does know me.
All I am going to say is that to build the best, you have to use every measurement method available to you. Some tell you the same information presented in different ways, but the tests are extremely valuable. To ignore testing (measurements) is to throw away incredibly useful information, and that isn't very intelligent. Yes, of course you listen to it as well.
the best motor means nothing if the evil cone resonates madly as it usually does
most cones are evil aren't they?
RAD / Rival Acoustics had offered such nice and nearly flat FR yummies but went to digital dust somehow it's a crying shame :
https://rhythmaudiodesign.com/collections/speaker-woofers
Black Kevlar , with nearly flat impedance curve , 5inch & 7inch :
https://rhythmaudiodesign.com/colle...r-5-5-mid-woofer-4-ohm?variant=31459124281457
https://rhythmaudiodesign.com/cdn/s...-a385-3c908a82a192_1024x1024.jpg?v=1585082882
https://rhythmaudiodesign.com/cdn/shop/products/R176-KB-08_SPL-on-axis_1024x1024.jpg?v=1551418020
we are the coneheadz
most cones are evil aren't they?
RAD / Rival Acoustics had offered such nice and nearly flat FR yummies but went to digital dust somehow it's a crying shame :
https://rhythmaudiodesign.com/collections/speaker-woofers
Black Kevlar , with nearly flat impedance curve , 5inch & 7inch :
https://rhythmaudiodesign.com/colle...r-5-5-mid-woofer-4-ohm?variant=31459124281457
https://rhythmaudiodesign.com/cdn/s...-a385-3c908a82a192_1024x1024.jpg?v=1585082882
https://rhythmaudiodesign.com/cdn/shop/products/R176-KB-08_SPL-on-axis_1024x1024.jpg?v=1551418020
we are the coneheadz
Why complicated? It correspond with my measurements very well. Current drive takes advantage from around 2x Fs.
But i have another question. If these measurements derived from real music signal, what actually they show? What point in time do they show? Or are these kind of peak hold measurement? For sweep measurement signal is fixed in amplitude. So you get one curve.And here amplitude is changing constantly. So your thd is changing together with music signal as it is amplitude dependant. So infinite number of curves or as much as your FFT lenght alow. Could you explain?
Hi Baffless,
They show the distortions... actually, they produce something you can listen to. I am not making any assumptions about the nature of these distortions. They may be harmonic, or subharmonic, or totally unrelated to the music played. Here is what they are. Make your own judgment.
My judgment is that the concept of THD is completely flawed and obsolete, something from the Stone Age. I dug some history and concluded that it appeared, once upon a time, because it was measurable - not because it was of much meaning. You don't have to agree.
I am interested: how much distortions have you managed to suppress? What was the loop gain curve?
Hi Anatach,
I agree with almost everything you wrote. To declare that we still disagree on something, I'd like to understand you better first. It well may be that we use different words while meaning the same things. How did you measure Barkhausen noise and dynamic distortions, when the music goes instantly from piano to forte? They are surely audible - but how would you measure them with either sine sweep, chirp, MLS, or multitone?
They show the distortions... actually, they produce something you can listen to. I am not making any assumptions about the nature of these distortions. They may be harmonic, or subharmonic, or totally unrelated to the music played. Here is what they are. Make your own judgment.
My judgment is that the concept of THD is completely flawed and obsolete, something from the Stone Age. I dug some history and concluded that it appeared, once upon a time, because it was measurable - not because it was of much meaning. You don't have to agree.
I am interested: how much distortions have you managed to suppress? What was the loop gain curve?
Hi Anatach,
I agree with almost everything you wrote. To declare that we still disagree on something, I'd like to understand you better first. It well may be that we use different words while meaning the same things. How did you measure Barkhausen noise and dynamic distortions, when the music goes instantly from piano to forte? They are surely audible - but how would you measure them with either sine sweep, chirp, MLS, or multitone?
Why do they still sell us (HiFi) drivers with rubbish surrounds that are doubtfull in lasting longer than 10 years under above average exposition of critical factors!
Suspicious PURIFY surrounds - will deteriorate even faster with that much more friction in the complex structure/shape!
Death to all rubber surrounds
Suspicious PURIFY surrounds - will deteriorate even faster with that much more friction in the complex structure/shape!
Death to all rubber surrounds
Last edited:
It seems there has been a misunderstanding.
Some posts ago, I asked for an option to spit this very dense kind of graph into two graphs:
In my post I meant with "LD" == what you call "harmonic", and with "NLD" == what you call "subharmonic or totally unrelated" distortions. Maybe misleading you by my maybe unsharp terms. And your answer was ...
I was a bit puzzled by this maybe mislead info, having (retropectvly rightly) had the impression that still there is a lot of info about harmonic distortions shown in the graphs, but decided to let it be ... After your clarification now, may I once again reiterate my following initial and basic question about splitting the infos into selective graphs? Hopefully now using the appropriate terms:
Is there a way to filter out the harmonic distortion products such as to only display the distortions what you call " ... subharmonic, or totally unrelated to the music played ...". And the other way, display only what you call the "... harmonic ..." distortions. Two distinct graphs, each with different, distinct informations. And for now one could even imagine one splitting-up step further, now. How about splitting the info into three graphs? One graph selectively displaying the harmonic, another displaying the subharmonic, and another one displaying the totally unrelated to the music played distortions? Does your method provide this option?
Some posts ago, I asked for an option to spit this very dense kind of graph into two graphs:
In my post I meant with "LD" == what you call "harmonic", and with "NLD" == what you call "subharmonic or totally unrelated" distortions. Maybe misleading you by my maybe unsharp terms. And your answer was ...
I was a bit puzzled by this maybe mislead info, having (retropectvly rightly) had the impression that still there is a lot of info about harmonic distortions shown in the graphs, but decided to let it be ... After your clarification now, may I once again reiterate my following initial and basic question about splitting the infos into selective graphs? Hopefully now using the appropriate terms:
Is there a way to filter out the harmonic distortion products such as to only display the distortions what you call " ... subharmonic, or totally unrelated to the music played ...". And the other way, display only what you call the "... harmonic ..." distortions. Two distinct graphs, each with different, distinct informations. And for now one could even imagine one splitting-up step further, now. How about splitting the info into three graphs? One graph selectively displaying the harmonic, another displaying the subharmonic, and another one displaying the totally unrelated to the music played distortions? Does your method provide this option?
Last edited:
Imho:
LD = linear distortion, translates roughly into: same input level, various output levels (like any bandpass system shows) BUT same frequency.
NLD = nonlinear distortion, translates roughly into: every output signal that is NOT the same frequency as the input signal.
I refrain from phase behaviour here.
So harmonic distortion essentially is NLD.
While I think mikets42 presents interesting stuff, I still wonder if it's not pixel peeping. Waiting for the ABX proof here.
LD = linear distortion, translates roughly into: same input level, various output levels (like any bandpass system shows) BUT same frequency.
NLD = nonlinear distortion, translates roughly into: every output signal that is NOT the same frequency as the input signal.
I refrain from phase behaviour here.
So harmonic distortion essentially is NLD.
While I think mikets42 presents interesting stuff, I still wonder if it's not pixel peeping. Waiting for the ABX proof here.
Hi mikets42,
Possibly terminology.
There are bigger fish to fry than Barkhausen noise when you look at the entire speaker system. To combat that, I try to use the better quality drivers. Depends on the budget I have to work with. These days I don't design much in speaker systems as they are large and heavy. I no longer have that nice 6,000 sq ft shop to play in. At 65, I really don't want to lug that stuff around anymore.
You have to look at the entire picture. Focusing on one aspect will never lead you to a good product, although I guess it is a great advertising point. That allows you to differentiate your product from the rest of the market whether the issue is valid or not. Advertising has always worked this way. I'm not saying you are behind a commercial product, but the thinking has become common. If you can reduce Barkhausen noise in a driver while not negatively impacting the other characteristics, then you are onto something. But it is the entire performance that is important, and certainly some defects people are more sensitive to than others. You have to go after the more audible problem areas. If you are chasing something that can't be solved ... well Don Quixote comes to mind.
Dynamic distortion, you look at driver linearity. The same way you test a new or re-coned driver. You can drive at higher power levels, but the basic method remains the same. High efficiency drivers give you an edge as the vc temperature remains lower. If you're designing for high quality, it will be an active crossover, directly driven system. This also really helps you keep each driver in it's "happy" frequency range.
-Chris
Possibly terminology.
There are bigger fish to fry than Barkhausen noise when you look at the entire speaker system. To combat that, I try to use the better quality drivers. Depends on the budget I have to work with. These days I don't design much in speaker systems as they are large and heavy. I no longer have that nice 6,000 sq ft shop to play in. At 65, I really don't want to lug that stuff around anymore.
You have to look at the entire picture. Focusing on one aspect will never lead you to a good product, although I guess it is a great advertising point. That allows you to differentiate your product from the rest of the market whether the issue is valid or not. Advertising has always worked this way. I'm not saying you are behind a commercial product, but the thinking has become common. If you can reduce Barkhausen noise in a driver while not negatively impacting the other characteristics, then you are onto something. But it is the entire performance that is important, and certainly some defects people are more sensitive to than others. You have to go after the more audible problem areas. If you are chasing something that can't be solved ... well Don Quixote comes to mind.
Dynamic distortion, you look at driver linearity. The same way you test a new or re-coned driver. You can drive at higher power levels, but the basic method remains the same. High efficiency drivers give you an edge as the vc temperature remains lower. If you're designing for high quality, it will be an active crossover, directly driven system. This also really helps you keep each driver in it's "happy" frequency range.
-Chris
It is very hard to do, especially for DIY.
You can make N recordings of the same musical piece, and average them. The average will be what you call "LD" and the variations from the average will be "NLD". Easier to say than done. 1) Drivers are temperature dependant which means long-term history. 2) you'll need a very repeatable environment. If you record 15 sec, and a car drives by, you drop this recording. I drop ~60%. If you want to record 64 repetitions ... how are you going to organize the QC? It would become way too tedious, and not fun.
I myself would love to know how to do it more easily.
Hi Anatech,
Yes, we are on the same wave. No, I am not behind a commercial product. Moreover, I would keep as far away from commerce as possible. I have had too many very negative experiences with pro audio companies, and I am not looking towards another one. DIY perfectionists and academic researchers only.