I'm glad we can discuss it civil, thanks!
You are correct I have no spice installed. I'm basing this stuff to very basic circuit analysis, and circuit analysis must be true otherwise our world wouldn't look the same as it looks today, electronic systems wouldn't exists. Adding an inductor in front of driver is same as if the driver inductance varied with excursion, it's just easier to do it like that, toggle the inductor in and out to see what varying inductance does to acoustic output. And it holds true per circuit analysis, everything in series. So only difference in the vituixcad sim is the circuit current held stable by high output impedance of the ideal source, making it approximate ideal current source. There is no need to go inside the amplifier, just replace the amplifier with ideal voltage or current source (Norton or Thevenin equivalent) and do circuit analysis, or simple vituixcad sim, to demonstrate.
If you have resources that negate this I'd gladly read and I'm ready to change my view on this, although all resources I've red favor mine so, perhaps spice can rock the boat.
Perhaps mikets42 could do it as he has simulator ready with the project?
Also, pardon me with the communication. It's always in a hurry when I write this stuff and it's hard to make it as civil as possible.
You are correct I have no spice installed. I'm basing this stuff to very basic circuit analysis, and circuit analysis must be true otherwise our world wouldn't look the same as it looks today, electronic systems wouldn't exists. Adding an inductor in front of driver is same as if the driver inductance varied with excursion, it's just easier to do it like that, toggle the inductor in and out to see what varying inductance does to acoustic output. And it holds true per circuit analysis, everything in series. So only difference in the vituixcad sim is the circuit current held stable by high output impedance of the ideal source, making it approximate ideal current source. There is no need to go inside the amplifier, just replace the amplifier with ideal voltage or current source (Norton or Thevenin equivalent) and do circuit analysis, or simple vituixcad sim, to demonstrate.
If you have resources that negate this I'd gladly read and I'm ready to change my view on this, although all resources I've red favor mine so, perhaps spice can rock the boat.
Perhaps mikets42 could do it as he has simulator ready with the project?
Also, pardon me with the communication. It's always in a hurry when I write this stuff and it's hard to make it as civil as possible.
Interesting, because there most be some miscommunication here somewhere 😀
But all the stability and runaway issues are real.
Can also very easily be tested with an actual driver and heating it up with a hairdryer.
With voltage drive, you see the SPL compressing (as it should), with current drive you see the SPL expanding.
This follows exactly what you see from the spice simulation as well from the literature.
(I am trying to find the Audioxpress article, can't remember when that was published)
As for norton/thevenin equivalents, the series resistor/inductor doesn't quite do the same thing.
It's an approximation, and only to some extend in particular in certain behavior.
In particular with complex impedances.
Even this can be simulated quite well.
I am not going to discuss how well those simulations work.
They are a 100% proven concept for absolutely decades (if not longer)
As long as we both don't let it get personal and just talk about the subject, I am happy with however you communicate 😀 😀
Even if that is rushed, blunt, rude or grumpy, I think we are old enough to ignore that part 🙂
Yes this is correct that thermal runaway is issue with current drive. On voltage drive heat in voice coil makes resistance rise, which reduces current through voice coil so heating reduces and acoustic output is compressing with the reducing current, as you state there.
With current drive current stays constant even if the resistance goes up, so heat keeps on rising but acoustic output stays the same, because current stays the constant, until voice coil burns out. Current drive keeps the acoustic output same, not expanding or compressing. Acoustic output is directly related to current.
There is more to this, jump resonance and all kinds of phenomenon, for both voltage and current drive. I think it's better to have voltage drive at drivers resonance, and soon above turn current drive by manipulating circuit impedance. I think either amp could be fine, just manipulate the passive network between the amp and driver to achieve any drive 😉 One could make "voltage drive" for driver resonance with current amp using a shunt circuit providing low impedance circuit for the driver electrical damping to function. As voice coil heats up more and more of the current amplifier delivers goes through the shunt instead through the voice coil. Also frequency response would look the same as with voltage drive as backEMF functions through the shunt (almost) like with low output impedance voltage amp. With voltage amp one could use series inductor to maintain the low impedance on driver resonance and start propping it up higher than that. Or what ever is the goal for the system.
Last edited:
If you've got the time, I'm curious to understand your opinion on this - it appears to me that many/most of these issues can be avoided or worked around, especially in DIY if using DSP and creating filters/curves beforehand during testing itself.
For starters, I am assuming using 1 (current) amp per driver - I think it would be helpful to avoid funky impedance effects (be it from a crossover or from another driver).
Thermal runaway - I would assume that a thermometer (or a non-contact IR sensor/camera) could easily detect this. A low-tech solution would be to have a max voltage limiter. Though this may result in clipping, however nothing will hopefully get damaged.
Impedance changes around Fs - sure, it would, if the amplifier was "dumb". But if you program it such (perhaps a simple EQ could fix 80% of this, or maybe more advanced curves/tuning), it appears (imo) definitely feasible when using DSP.
I think I haven't fully understood some aspects of how impedance changes with cone excursion that you mentioned, though it seems to me that those would likely be quite repeatable - so if a driver is "calibrated" initially for all frequencies and such a curve/filter is made, it would function when listening too.
I do agree that current amps make little sense on commercial products where you're not testing/tweaking them yourself, or that many channels of amplification & DSP may not be cheap - but it seems to me that the reduction in distortion makes it a worthwhile effort if you're trying to go from say 95% to 97%. It appears that if you test and calibrate all aspects of the driver/system beforehand, you would be able to move and control the driver to a high level of precision later on.
Voltage drive is a much more robust traditional approach, the same with Class A/B amplifiers and analog passive crossovers. It took a very long time for Class D amps to break in, especially for Bruno Putzeys self-oscillating amps. He went through lots of contempt, neglect, denials, ostracization, etc despite hard-data proofs that he was right, till his invention became acknowledged and properly valued. Now Class D amps are mainstream, they are everywhere and dirt cheap. The adoption of DSP is still slow. The chips themselves are pennies, they are self-containing controllers, and the boards with them shall also cost pennies and work as a Swiss clock - but we are not there yet.
The peak-to-average ratio for music is 20...25dB, which means that the average amplifier power used is well below 1W. What thermal runaway?
The low frequencies may also be dealt with by a circuitry (yet unproven) like this:
I am of opinion that an amp should be integrated with a driver, and have a digital input... but we are not there yet.
F = Bl * I, where F= force, B = magnetic field, l=wire length, I = current.
The peak-to-average ratio for music is 20...25dB, which means that the average amplifier power used is well below 1W. What thermal runaway?
The low frequencies may also be dealt with by a circuitry (yet unproven) like this:
I am of opinion that an amp should be integrated with a driver, and have a digital input... but we are not there yet.
F = Bl * I, where F= force, B = magnetic field, l=wire length, I = current.
as well as @tmuikku
I will get back to this a bit later when I do have some more time to show what is going on. 🙂
Currently rather busy with a new design that was basically already finalized, but all of a sudden a manufacturer made a super major oopsy.
Meaning we had to do the whole juggling and tango dance all over again 🙁
The cabinets were almost already on order to make things worse (lucky that could be postponed for now)
There are a few brands that are boutique or tube oriented. But yes, measuring everything by sight, or in terms of a THD metric, and second-guessing your hearing to the point of masochism, where something worse-sounding must be endured due to an overriding belief that a particular measurement or theory of operation means it ought to sound better but your unreliable ears are merely fooling you. These are all popular traps. As is operating in a silo.
You must be new 😉
Re: thermal runaway, unlikely to be an issue unless output impedance is high right down to DC.
In response to several follow-up posts, I stand by my suggestion of mixed-mode feedback as a good compromise, at least as a concept. A backloaded horn could have a very messy impedance in the bass, unfixable with an RLC or active filter nested in the feedback loop, and would be difficult to EQ.
Velocity feedback in the bass could be interesting, but if it includes significant hardware, it becomes that much more difficult than a couple of extra Rs and Cs.
However messy is the horn-loaded driver impedance, it would be a piece of cake to fix when you have well-working DSP boards for a reasonable price. System identification of such an ARMAX system is a simple exercise straight from the Lennart Ljung "theory for a user" book... I mean, "simple" for a few experts. Once it's done, it can be reused by others as a black box. Otherwise, it's 10s of years of learning which renders it meaningless.
I agree. Velocity feedback on the accelerometer - yes, tricky as any nested feedback system. Just preventing it from going nuts on last-stage-saturation requires a PhD in rocket science. Again, it's most suitable to be done in DSP where you have full visibility of what happens inside the feedback loop.
Here, amplifier voltage gain on HiVi D5.4:
Same driver, on voltage drive, 2nd and 3rd harmonics for outputs varying from 65 to 85 dB SPL.
Same driver, on current drive:
and here is the comparison of the 3rd harmonic on voltage (red+yellow) vs current (green+cyan) drive, for outputs 85 and 65 dB SPL.
I agree. Velocity feedback on the accelerometer - yes, tricky as any nested feedback system. Just preventing it from going nuts on last-stage-saturation requires a PhD in rocket science. Again, it's most suitable to be done in DSP where you have full visibility of what happens inside the feedback loop.
Here, amplifier voltage gain on HiVi D5.4:
Same driver, on voltage drive, 2nd and 3rd harmonics for outputs varying from 65 to 85 dB SPL.
Same driver, on current drive:
and here is the comparison of the 3rd harmonic on voltage (red+yellow) vs current (green+cyan) drive, for outputs 85 and 65 dB SPL.
As I hoped, Barkhousen's noise is gone with the current drive. These noises are the short spikes, like Dirac's delta functions, during loud passages, corresponding to magnetic domain flops. Being delta-function alike, they have a very wide spectrum and therefore are not maskable. For my ear, they are especially annoying on piano.
Here are the non-linear residuals of playing a Mozart's 20th piano concerto's tutti through HiVi D5.4 on
red: RMS 75 dB SPL , voltage drive
yellow: RMS 75 dB SPL, current drive, distortions are much lower.
green: RMS 80 dB SPL, voltage drive: Barkhousen noise is prominent, emf/Zspk, when Zout (of amplifier) = 0.
cyan: RMS 80 dB SPL, current drive: Barkhousen noise is gone, emf/(Zspk+Zout), when Zout->inf
Here are the non-linear residuals of playing a Mozart's 20th piano concerto's tutti through HiVi D5.4 on
red: RMS 75 dB SPL , voltage drive
yellow: RMS 75 dB SPL, current drive, distortions are much lower.
green: RMS 80 dB SPL, voltage drive: Barkhousen noise is prominent, emf/Zspk, when Zout (of amplifier) = 0.
cyan: RMS 80 dB SPL, current drive: Barkhousen noise is gone, emf/(Zspk+Zout), when Zout->inf
Why over-complicate things?
This reminds me of one of the Hawksford papers, where an extremely elaborate current amplifier was developed, which included velocity feedback and soft-clipping for the bass, de-facto replicating almost exactly what voltage control (in the bass) would have done in the first place. First they remove thermal compression / defeat any built-in mechanical limiting, and then add fake compression via the soft clipping.
I am of opinion that an amp can be integrated with a driver, and have a digital input. Yes we are there.
Current drive is a hypothetical construct. Think of a transconducatance amplifier. Also think current feedback.
Current drive is a novelty less new than obscure. We are not going to see “current amplifiers” on the shelf at the hifi shop any time soon.
Drivers are complex machines with variable impedance. They transform voltage input into acoustic output. As the driver impedance changes so does the current. Current becomes the force that moves the driver cone. Driver designers are improving the acoustic output of drivers to reduce the “need” for inefficient current drive amplifiers.
I expect we are going to see less inefficient brute force current drive and more software DSP voltage control. Programed voltage controlled output is also controlled current output operating into a driver with known parameters. Think of the JBL M2 speaker packaged with DSP programing and a Crown voltage output amplifier. Also think transfer function.
Thanks DT
Last edited:
Yeah Klippel Controlled Sound is one application. I imagine we could soon have AI that sniffs the driver and adjusts things automatically, a magic box to put into a system.
DSP can't solve everything, and especially not the chaotic and unpredictable voltage spikes from permanent magnets interacting with voice coils. See Barkhousen noise above.
Incidentally, I think the high efficiency class-D architecture naturally lends itself to transconductance operation. But all the class-D chips I've seen are basically locked down and there's no inverting input available.
E.g. the TPA3116, while interesting, is locked down and dumbed down with an feedback line that is hardwired between the voltage output and a hidden inverting input. Then again, maybe my searching is at fault and I should specifically look for class-D op-amps, not amplifiers.
I would like to see that noise demonstrated on an FFT of a speaker output sampled with a lab grade microphone.
Make your own inverting input, add an op-amp, make your own feedback loop.
Thanks DT
That's a ruddy good point! My brain is definitely in "slow" gear this week. Class D interests me enough to maybe build a bluetooth module -- nothing too serious --so an op-amp hack is probably the way to go.
A side quest while working on a power-hungry class A. (Nothing quite like working on an SMPS and pretending not to notice that it's basically a class D half-bridge outputting zero volts 😆 )
A side quest while working on a power-hungry class A. (Nothing quite like working on an SMPS and pretending not to notice that it's basically a class D half-bridge outputting zero volts 😆 )
Hello Michael,
Thanks for starting this article on diyAudio. Great work and thank for sharing your findings.
We are not there yet, perhaps not in audiophile passive speaker land let me switch out my amplifier, cable lifters, interconnects for fun, but elsewhere DSP is everywhere.
Can your software provide labels like dB, H2 and H3 instead of Sweep 2 and Sweep 3
And logarithmic x axis like 100Hz, 1Khz, 10KHz, 100Khz
The graphs are dense and data heavy and enjoyable to some of us pixel peepers, others find it challenging to decipher the non standard nomenclature, particularly first thing in morning. ☕
Looking forward to see how this evolves.
BR
@tmuikku
Great scientific communication.
Thanks for starting this article on diyAudio. Great work and thank for sharing your findings.
We are not there yet, perhaps not in audiophile passive speaker land let me switch out my amplifier, cable lifters, interconnects for fun, but elsewhere DSP is everywhere.
Can your software provide labels like dB, H2 and H3 instead of Sweep 2 and Sweep 3
And logarithmic x axis like 100Hz, 1Khz, 10KHz, 100Khz
The graphs are dense and data heavy and enjoyable to some of us pixel peepers, others find it challenging to decipher the non standard nomenclature, particularly first thing in morning. ☕
Looking forward to see how this evolves.
BR
@tmuikku
Great scientific communication.
Last edited:
Member
Joined 2003
I'm no amplifier expert, but I do own the "My-Ref" LM3886 implementation from Mauro Penasa, which claims to be a current source topology. More details here:
https://www.diyaudio.com/community/threads/my-audiophile-lm3886-approach.54571/
https://www.diyaudio.com/community/threads/my-audiophile-lm3886-approach.54571/
I agree. Alas, control systems tend to be much more complex than the objects they control. I was taught at university that the control loops should not be any more complex than strictly necessary - which comes only with a very good understanding of the automatic control theory. Many lack it, including those who teach it. Then, you have those awkward theories and pathetic designs, which may have worked at some moment, to a degree but are not reproducible or debugable when something goes wrong.
To DcibeL: On Mauro Penasa amplifier: it's quite complicated. I don't understand how it works - nor why it is advertised as a current source. I also live in BC ... could I see how well it lowers the driver's distortions?
To tktran303: thank you! I'll fix my graphs.
To abstract: BTW, self-oscillating Class D by Bruno Putzeys has a pretty wide bandwidth and it is quite nice, overall. Yet, it could be improved by adding a nested feedback loop - as it is, it's way too sensitive to the power supply's non-idealities, thus, in effect, you need 3 amps🙂.