Though it goes against convention of a separate amplifier and speaker it seems to me and has for a long time seemed that until you include the speaker into the feedback loop we are missing a very important way to lower speaker distortion. There is no mechanical or electrical way to counter the back EMF from the speaker in isolation. After reading this thread and the recent discussion on the Blowtorch thread and discussions with the use of the Slewmaster amplifiers it seems like this is the missing link.
I have still not found out what would change inside the speaker itself besides changes in impedance to use a current drive effectively. Increasing the impedance of the voicecoil by adding winding length would change the BL curves and except for the added mass I don't see it being a problem to achieve this change in the speaker. Are there any other changes that anyone would see as a requirement to use current drive effectively?
I will be watching this thread carefully to see how to comprehensively create a self powered speaker taking advantage of current drive with a CFA type of amplifier and all the assorted requirements to make it all work. Great work and thread, I appreciate the technical discussion by those who have the knowledge to post real results and any compromises that may be required.
I have still not found out what would change inside the speaker itself besides changes in impedance to use a current drive effectively. Increasing the impedance of the voicecoil by adding winding length would change the BL curves and except for the added mass I don't see it being a problem to achieve this change in the speaker. Are there any other changes that anyone would see as a requirement to use current drive effectively?
I will be watching this thread carefully to see how to comprehensively create a self powered speaker taking advantage of current drive with a CFA type of amplifier and all the assorted requirements to make it all work. Great work and thread, I appreciate the technical discussion by those who have the knowledge to post real results and any compromises that may be required.
Powered Integrated Super Sub technology
It is entirely possible for an "amplifier" to know what the speaker is doing.
It needs to sense the speaker current. An amp which twiddled its Output Z using both current & voltage feedback does this. Speakers act as accurate microphones (sense the actual sound) if operated into Low Z.
See "Loudspeakers as Microphones" - Peter Baxandall special lecture London AES (early 80s, late 70s?)
If operated into High Z, then the voltage at the terminals is a measure of cone velocity.
Both these mechanisms obey superposition & Thevenin so if you're clever, you can look at this while the amp is giving zillion volts and amps to the speaker. But non est tantum facile.
There are several tried & tested methods of using this "controlled output Z" or "current + voltage feedback" or "actual sound & motion feedback" (different descriptions of the same thing) if you incorporate the amplifier design in the speaker. Some of these are in the zanier incarnations of my Powered Integrated Super Sub technology.
The simplest is the negative output R that Fons mentions.
More sophisticated but similar (??!) is ACE technology by Erik Stahl which was used by Audio Pro, Sweden for subs. Unfortunately, since he left, there isn't anyone there who understands it. Anyone have a contact for Erik? Or a clean copy of his original AES preprint?
These methods have the distortion reduction and dynamic overload protection features discussed in Mills & Hawksford. However, they are badly affected by heating of the voice coil.
David Birt (?) did an excellent IoA paper at Windermere where he arranged speaker and amp in a bridge so he could measure and compensate for heating on the fly. Anyone have an email for him?
These are the most elegant methods and they can be analysed from many viewpoints. Some of these viewpoints don't show up the distortion reduction advantages clearly.
I'm contemptuous of methods which rely on extra transducers or extra windings (like Mills) or zillion point DSP EQ especially if they don't give ALL the advantages of the elegant methods.
A brute force zillion point approach possible today is measure accurately speaker Z (not that easy) and tailor the output of a High Z amp to suit. This would give some but not all the advantages of the above systems cos it wouldn't "know" what the speaker is up to.
Both papers are available from the AES
They realised that to get flat response, the speaker had to be mass controlled and driven by a voltage source
Few Thiele-Small pseudo gurus have R&K's insights on bass response.
There is an equivalent serendipitous relation for current drive of electrostatic speakers but none for a practical electrodynamic speaker.
No present electrostatic speaker takes advantage of this.
Chester Rice did more important work on electrostatic speakers but nothing as good as the 1924 invention. Kellog left audio and founded his cornflake empire
>For what it is worth, I think this is impossible. It seems to violate the cause and effect principle. How could the amplifier "know" what the speaker is up to ? Short of sensing the actual motion of the drivers or the actual sound... this has to be sort of by guess and by golly, it seems to me.
It is entirely possible for an "amplifier" to know what the speaker is doing.
It needs to sense the speaker current. An amp which twiddled its Output Z using both current & voltage feedback does this. Speakers act as accurate microphones (sense the actual sound) if operated into Low Z.
See "Loudspeakers as Microphones" - Peter Baxandall special lecture London AES (early 80s, late 70s?)
If operated into High Z, then the voltage at the terminals is a measure of cone velocity.
Both these mechanisms obey superposition & Thevenin so if you're clever, you can look at this while the amp is giving zillion volts and amps to the speaker. But non est tantum facile.
There are several tried & tested methods of using this "controlled output Z" or "current + voltage feedback" or "actual sound & motion feedback" (different descriptions of the same thing) if you incorporate the amplifier design in the speaker. Some of these are in the zanier incarnations of my Powered Integrated Super Sub technology.
The simplest is the negative output R that Fons mentions.
More sophisticated but similar (??!) is ACE technology by Erik Stahl which was used by Audio Pro, Sweden for subs. Unfortunately, since he left, there isn't anyone there who understands it. Anyone have a contact for Erik? Or a clean copy of his original AES preprint?
These methods have the distortion reduction and dynamic overload protection features discussed in Mills & Hawksford. However, they are badly affected by heating of the voice coil.
David Birt (?) did an excellent IoA paper at Windermere where he arranged speaker and amp in a bridge so he could measure and compensate for heating on the fly. Anyone have an email for him?
These are the most elegant methods and they can be analysed from many viewpoints. Some of these viewpoints don't show up the distortion reduction advantages clearly.
I'm contemptuous of methods which rely on extra transducers or extra windings (like Mills) or zillion point DSP EQ especially if they don't give ALL the advantages of the elegant methods.
A brute force zillion point approach possible today is measure accurately speaker Z (not that easy) and tailor the output of a High Z amp to suit. This would give some but not all the advantages of the above systems cos it wouldn't "know" what the speaker is up to.
Both papers are available from the AES
There are good reasons why Rice & Kellog's 1924 invention is still pre-dominant from the cheapest speakers to the very highest quality.
They realised that to get flat response, the speaker had to be mass controlled and driven by a voltage source
Few Thiele-Small pseudo gurus have R&K's insights on bass response.
There is an equivalent serendipitous relation for current drive of electrostatic speakers but none for a practical electrodynamic speaker.
No present electrostatic speaker takes advantage of this.
Chester Rice did more important work on electrostatic speakers but nothing as good as the 1924 invention. Kellog left audio and founded his cornflake empire
Last edited:
This publication covers all the possible methods of reduction of speaker distortion
http://cogsys.imm.dtu.dk/nonlincomp/Klippel.pdf
(Active Compensation of Transducer Nonlinearities )
I believe in DSP approach and modeling of speaker nonlinearities.
http://cogsys.imm.dtu.dk/nonlincomp/Klippel.pdf
(Active Compensation of Transducer Nonlinearities )
I believe in DSP approach and modeling of speaker nonlinearities.
Thank you for the suggestion, but since the beginning of the year 2012, audio has been just a hobby for me, and I am offering no kits or PCBs.
I have only built a functional sample of the current output amplifier, based on simple modification of the usual voltage output GNFB amplifier to V/I converter, as shown on my webpage. It works very well, and distortion of output current, feeding the speaker as a load, is about 0.002% in most of the audio band, not exceeding 0.008%. The only care to be done is to compensate for inductive load at high frequencies, which is easy achievable with R-C net similar to Zobel, in parallel wit speaker. This amplifier is enough for measurement purposes and I believe that the same concept may be used for a real life amplifier with current output.
This is excellent as is most of Klippel's stuff.
I need to get my patent lawyers to see him as part of my 'previous life' expertise was combining amp/speaker/linear & non-linear/analogue/digital stuff for better sound.
Certainly the digital stuff he's doing now was pie in the sky when I disappeared from the civilised world.The one big thing he acknowledges (that practically no one else does) is that speakers are neither Linear or Time Invariant. It's only convenient for us to treat them as LTI.
However, he still treats speakers as 1D transmission channels. They are in fact 1D input with 3D output. Because of this, improving the 1D transfer function can only give limited improvement to the sound.
What you need to do is to improve the Room Interface Profile.
AES E-Library Absolute Listening Tests-Further Progress shows the 1D transfer function was already good enough in 1980.
BTW, my previous post covers nearly every requirement in Klippel's paper. Read David Birt's patent for the early digital stuff.
Last edited:
Pavel,
I thin it was you or George who recommended a book by Esa Merilainen on Current driving of loudspeakers. Unless that book and the conclusions are pure drivel it does seem that current drive with the correct implementation is a way to reduce distortion. He does show methods to include the loudspeaker in the feedback loop so I would have to take what Kgrlee is saying with a grain of salt even though I really respect his views on most subjects.
If it was you who suggested this book thank you. It is a heavy read for me as I am not an EE but I am slowly reading the book and think I will learn much from this work.
I thin it was you or George who recommended a book by Esa Merilainen on Current driving of loudspeakers. Unless that book and the conclusions are pure drivel it does seem that current drive with the correct implementation is a way to reduce distortion. He does show methods to include the loudspeaker in the feedback loop so I would have to take what Kgrlee is saying with a grain of salt even though I really respect his views on most subjects.
If it was you who suggested this book thank you. It is a heavy read for me as I am not an EE but I am slowly reading the book and think I will learn much from this work.
Steven,
I was one of those who mentioned Esa Merilainen's book.
I have also tried to show how easy, in principle, is to build an amplifier that has output current proportional to input voltage.
Current drive of speakers and speaker distortion
In fact, it is a trivial task.
Current drive is just one necessary condition how to decrease speaker nonlinear distortion, however, it is not any satisfactory condition. You will measure linear speaker current with current drive, but you are interested in acoustical output. You will linearize Le(x) nonlinearity effect, but Cms(x) and Bl(x), plus other sources of nonlinearity would remain unsolved. You will have to check every speaker, at acoustical side, if there is any improvement with current drive. And, you have to pre-equalize input signal to keep flat amplitude response.
I was one of those who mentioned Esa Merilainen's book.
I have also tried to show how easy, in principle, is to build an amplifier that has output current proportional to input voltage.
Current drive of speakers and speaker distortion
In fact, it is a trivial task.
Current drive is just one necessary condition how to decrease speaker nonlinear distortion, however, it is not any satisfactory condition. You will measure linear speaker current with current drive, but you are interested in acoustical output. You will linearize Le(x) nonlinearity effect, but Cms(x) and Bl(x), plus other sources of nonlinearity would remain unsolved. You will have to check every speaker, at acoustical side, if there is any improvement with current drive. And, you have to pre-equalize input signal to keep flat amplitude response.
Last edited:
Thank you Pavel. I do understand there are other parameters that are not linear in reality. I have removed one of those which is the BL non-linearity by using very high MGOe Neo magnets with an extremely long gap, four times the voicecoil length in fact. Yes I have to give up efficiency to do that but with modern amplifier power I don't see that as a problem really.
The area that most intrigues me right now is the spider, surround non-linear actions. I have been contemplating as few experiments but haven't had the time recently as I have been involved designing some new products for a new group who has invited me in by offering 20% of the company for doing all the design aspects they are trying to create. I now have a cell phone add on device I am working on that is taking a lot of my time. It is taking all of my varied skills to do what they want including industrial design, packaging, material selection and selection of vendors. They do drive me nuts though with the ever changing end requirements but that is why they hired me, to keep things on track and actually something that can be made. There is nothing like getting an industrial design from a designer who has no idea how the device has to function or be made, at least I get to control the entire process.
The area that most intrigues me right now is the spider, surround non-linear actions. I have been contemplating as few experiments but haven't had the time recently as I have been involved designing some new products for a new group who has invited me in by offering 20% of the company for doing all the design aspects they are trying to create. I now have a cell phone add on device I am working on that is taking a lot of my time. It is taking all of my varied skills to do what they want including industrial design, packaging, material selection and selection of vendors. They do drive me nuts though with the ever changing end requirements but that is why they hired me, to keep things on track and actually something that can be made. There is nothing like getting an industrial design from a designer who has no idea how the device has to function or be made, at least I get to control the entire process.
Pavel,
On that schematic you just posted I see the RC network on the speaker terminals to correct the impedance rise. Now a friend of mine insists that a RCL network is what he would use to do this rather than a simple RC network. What difference is there between the two in end results? Does the RCL tank circuit have an advantage over the RC network or are they truly equivalent? Also I assume you are flattening the general rise from lowest to highest frequency range and that you are not attempting to kill the resonance peak at speaker Fs, is this correct. Would there be a problem in trying to correct both functions, would one filter shift the other or are these two separate issues that can both be tackled?
On that schematic you just posted I see the RC network on the speaker terminals to correct the impedance rise. Now a friend of mine insists that a RCL network is what he would use to do this rather than a simple RC network. What difference is there between the two in end results? Does the RCL tank circuit have an advantage over the RC network or are they truly equivalent? Also I assume you are flattening the general rise from lowest to highest frequency range and that you are not attempting to kill the resonance peak at speaker Fs, is this correct. Would there be a problem in trying to correct both functions, would one filter shift the other or are these two separate issues that can both be tackled?
You should always take what beach bums say with a huge measure of salt
but what bit are you most suspicious off?All that I wrote earlier, I've tried & tested in my previous life .. some of it going back to the late 70's as the AES papers show.
Without wishing to go too far off-topic, I wanted to show there are other ways of achieving equivalent or better distortion reduction as current drive .. without some of its disadvantages.
Not all of Merilainen is drivel but some of it is. IMHO, his methods come under "fudging". There's nothing wrong with that if you have the correct tools ... unless you are interested in quantity production.
Stahl's method allows design
With current drive, one of the biggest faults is that certain variable resonances (eg the cabinet wall resonances that John Atkinson of Stereophile is so obsessed with) appear on the Impedance curve and are exacerbated by current drive.
The Q of these makes the Digital method I propose the only sensible way to ameliorate these .. but they aren't Time Invariant
Even Klippel's method with microphone feedback will have problems dynamically adapting to these as there are inherent limits due to time sampling & other DSP mumbo jumbo.
________________________
There's also an important caveat about modelling Cms non-linearity digitally. This interacts with Bl non-linearity and is inherently unstable. The yuckiest result of this is that with a long voice coil and short gap, as the coil moved out of the pole, its inductance reduces and instantaneous current increases and the coil tries even harder to jump out with Voltage drive. Here a suspension which deliberately limits can REDUCE distortion over a large range of frequencies & outputs.
Current Drive mostly avoids this evil.
As I neber lern 2 kont or reed en rite, I can't work out whether Klippel's stuff correctly models and deals with this.
___________________
Another 'cure' is short coil & long gap but this gives the most advantage with Voltage Drive.
If you are using Current Drive, you might as well have a long (perhaps 4 layer) coil and short gap as the evils these introduce are directly addressed by Current Drive. A 4 layer coil would vastly increase the thermal power handling.
But your short coil/long gap method shouldn't sacrifice any efficiency if designed properly. It only costs more.
Last edited:
Kgrlee,
I will readily accept that you have more experience and knowledge than I do without doubt. I think if I remember correctly you worked for KEF or a similar speaker firm. I am still learning and have far to go. I have a much better grasp of the mechanical side of things.
If I could I would also be a beach bum, perhaps not a bum but I would love to move to the Islands. your Island would be a bit to large for me.
I will readily accept that you have more experience and knowledge than I do without doubt. I think if I remember correctly you worked for KEF or a similar speaker firm. I am still learning and have far to go. I have a much better grasp of the mechanical side of things.
If I could I would also be a beach bum, perhaps not a bum but I would love to move to the Islands. your Island would be a bit to large for me.
Kindhornman, this beach bum is really in awe of the efforts you have made to make better speakers.
I'm only sorry that I've not been able to help you make an inexpensive current drive amp from lm3886 et al. The obstacle is that there are no SPICE models of 3886 which address the stability issues that a good current amp would encounter.
This leaves soldering iron stuff which I can't really do at present.
_________________________
But to get back to current drive speakers. I've listed the problems of current driven speakers and also how I think 21st century tech. can alleviate them to some extent.
This would require an inexpensive DSP platform which could process a music signal before it is sent to the Current Output amp.
Today, this would be a laptop running some sort of DSP plugin capable of at least a 8192pt FFT in 'real time' in series with the digital stream.
I would be happy to help anyone wanting to do the necessary measurements and design the Digital Filters. You would need a measurement mike .. an ECM8000 will do and something like woofertester to measure Impedance & Response.
I offered this early on in the thread with no takers.
I'm only sorry that I've not been able to help you make an inexpensive current drive amp from lm3886 et al. The obstacle is that there are no SPICE models of 3886 which address the stability issues that a good current amp would encounter.
This leaves soldering iron stuff which I can't really do at present.
_________________________
But to get back to current drive speakers. I've listed the problems of current driven speakers and also how I think 21st century tech. can alleviate them to some extent.
This would require an inexpensive DSP platform which could process a music signal before it is sent to the Current Output amp.
Today, this would be a laptop running some sort of DSP plugin capable of at least a 8192pt FFT in 'real time' in series with the digital stream.
I would be happy to help anyone wanting to do the necessary measurements and design the Digital Filters. You would need a measurement mike .. an ECM8000 will do and something like woofertester to measure Impedance & Response.
I offered this early on in the thread with no takers.
