Graham,
do you have graphs of measurements of amp+dummy load vs amp+real speaker? Say, typical AB topologies, or well known amps, e.g., Pass's Alephs, Leachs, Brystons etc? I followed some threads here, for instance the 'distortion microscope' thread, but most data on this forum come from simulations.
BTW the DoZ is not Nelson's - it is Rod Elliott's design he came up with as his "answer" to the Pass Labs "Zen" design...
do you have graphs of measurements of amp+dummy load vs amp+real speaker? Say, typical AB topologies, or well known amps, e.g., Pass's Alephs, Leachs, Brystons etc? I followed some threads here, for instance the 'distortion microscope' thread, but most data on this forum come from simulations.
BTW the DoZ is not Nelson's - it is Rod Elliott's design he came up with as his "answer" to the Pass Labs "Zen" design...
The crucial two words that are is needed are "minimum phase"
The bottom line with speakers is that anything that involves the Theil Small model of the speaker, including all the stored energy in both the moving diaphragm and compressed air in the box is captured in an equivalent RLC model.
What is not captured is anything that involves a time constant or delay (as opposed to a phase delay which is captured.) So effects like thermal changes of resistance of coils, and reflected energy from the room transducting back into the system is not covered.
As has been pointed out, you get quite desperately poor transduction efficiency, unless perhaps you manage to contrive things so you excite some sort of coupled room/speaker resonant mode - which one would imagine would be the last thing you would do in a real world situation.
If you measure the speaker's resistance vs frequency, which is both easy and commonly done in many speaker reviews, and you assume a minimum phase system, it is a reasonably trivial transform to create the complex impedance of the speaker. Once you have that, you have all you need to construct an equivalent RLC model. Indeed the, now quite old, speaker design software CALSOD did this as a matter of course.
Another really nice example of looking at energy storage in speakers is on Linkwitz's web site: http://www.linkwitzlab.com/frontiers_2.htm
The next question is to apply this to a feedback amplifier. I think it could be quite illuminating to do so. It may strike a blow for one of the fundamental fallacies in amplifier specmanship. The idea that damping factor has anything to do with amplifier quality, and most especially, anything to do with the quality or depth of bass. Since it doesn't. But damping factor, all other things being equal is a pretty good metric of feedback factor. (Well either that or a metric of the number of seriously biased output devices.)
The idea that the speaker will induce some sort of cycling feedback related effect goes by another name usually. Oscillation. It would suffice to show that your speaker can make the amplifier at least conditionally stable at some point to prove your thesis. If you can't do that, then you will need to look rather deeper.
The bottom line with speakers is that anything that involves the Theil Small model of the speaker, including all the stored energy in both the moving diaphragm and compressed air in the box is captured in an equivalent RLC model.
What is not captured is anything that involves a time constant or delay (as opposed to a phase delay which is captured.) So effects like thermal changes of resistance of coils, and reflected energy from the room transducting back into the system is not covered.
As has been pointed out, you get quite desperately poor transduction efficiency, unless perhaps you manage to contrive things so you excite some sort of coupled room/speaker resonant mode - which one would imagine would be the last thing you would do in a real world situation.
If you measure the speaker's resistance vs frequency, which is both easy and commonly done in many speaker reviews, and you assume a minimum phase system, it is a reasonably trivial transform to create the complex impedance of the speaker. Once you have that, you have all you need to construct an equivalent RLC model. Indeed the, now quite old, speaker design software CALSOD did this as a matter of course.
Another really nice example of looking at energy storage in speakers is on Linkwitz's web site: http://www.linkwitzlab.com/frontiers_2.htm
The next question is to apply this to a feedback amplifier. I think it could be quite illuminating to do so. It may strike a blow for one of the fundamental fallacies in amplifier specmanship. The idea that damping factor has anything to do with amplifier quality, and most especially, anything to do with the quality or depth of bass. Since it doesn't. But damping factor, all other things being equal is a pretty good metric of feedback factor. (Well either that or a metric of the number of seriously biased output devices.)
The idea that the speaker will induce some sort of cycling feedback related effect goes by another name usually. Oscillation. It would suffice to show that your speaker can make the amplifier at least conditionally stable at some point to prove your thesis. If you can't do that, then you will need to look rather deeper.
I think it's difficult to get that graph. With single tone or monotonic tones, speakers will not show it's worse nature. Nasty EMF will come from real music. And picturing a music is like watching a green band on scopes 😀
Yes, it would be very interesting to include the voltage produced by speaker cone's movement (which are not part of the music signal) in the feedback behavior. Surely it's not good to "feedback" voltage which is random (compared to source signal)
I'm confused with FM Accoustic power amp. In the brochure, it has 1000's of damping factor, yet in the specsheet it stated it has no Feedback, no Error Correction, no FeedForward, has 7 gain stages. I think it's impossible to marry the 2 facts. It's difficult to even control DC offset, if what they claim is true.
Anyone has seen FM Accoustics schematic? What is inside it?
It should make no difference - indeed one might argue that if anything is going back the less correlation the better - since correlated noise will produce correlated products that are audible. Indeed the whole pint of feedback is that the error component is pretty much anything.
But the real point is that feedback, in principle, does not care. One could think of the output impedance (i.e. the reciprocal of the damping factor) as the metric of how good a voltage source the amplifier is. If the amplifier really does have that output impedance across all useful frequencies, you have a limit on the error that the speaker can create.
But the question is not about that. But rather if a real implementation of a feedback amplifier is subject to some additional error (distortion) mechanisms due the the complex load.
On the whole I would be surprised if it were, but I am quite prepared to be surprised.
More likely a carefully crafted set of weasel words. You can't make an amplifier that complex without killing lots of gain somewhere. One suspects it is the usual "no global feedback" claim. Could easily have lots of local feedback at each stage, and for a damping factor that huge, an output stage with lots of local feedback.
In ideal electronic behavior, indeed that is true. 😀 Anything can be considered as "error tobe eliminated", even random/non corelated condition at output point.
The problem is, feedback (error correction) does not work perfectly. A simple differential is a good example. If differential works perfectly 100%, we will never heard of 2nd-3rd-higher harmonic. All will be eliminated by a differential. We can make 0 harmonic power amp. But it is not what happened in real cct. A differental cannot eliminate all error, so producing 3rd--higher order harmonics from residual 2nd that transfered to output stage, re-enters feedback, making artifacts, transfered to output stage, re-enters feedback and so on.
I've been thinking about this. Mr.Maynard himself also stressed on how important this is to audible quality of power amp.
Shift less than 2deg at 20khz could be a strong indication. (classD selfoscilating amp can produce 20deg)
For "minimum phase" result, one possible way to achieve it is with a very fast amplifier.
And a current feedback power amp is known to be very fast.
Does a current feedback power amp produces better (less) shift at 20khz compared to ordianary voltage feedback? Does it sounds better than voltage feedback?
This isn't what minimum phase means. You can have heaps of phase angle and have a minimum phase system. The simplest way I can think of is to describe it as the system that has the smallest possible phase angle and yet still does what the transfer function defines it to do. (This implies that there are a (possibly infinite) set of systems that also implement the transfer function that have greater phase - something that is in general true.) However a minimum phase system is causal and stable. The causal bit is useful here, and comes, in part, from the lack of time delay in a speaker.
The other issue is the idea that if feedback is perfect it will correct all errors. This is not true either. Feedback reduces the error, by the feedback factor. This is from the basic definitions of how it works. A perfect feedback amplifier with an open loop distortion of 1% (or some arbitrary measure) and a closed loop feedback ratio of 10 will (in principle) have 0.1% of our mooted distortion. An op-amp is often designed with what is almost infinite open loop gain (well silly numbers anyway) and so, in principle, we could have arbitrary reduction of distortion. However the gain starts to drop with frequency (in any amplifier) and so quite quickly we cease to have arbitrary reduction. Further, the phase angle gets bigger, so eventually we get into trouble with oscillation. We must ensure the gain is less than one before this point is reached. This is the Nyquist stability criterion. This is all pretty basic (junior EE school) stuff.
Further, when the distortion products are reduced in magnitude, it is accepted that even this reduction cannot be perfect, and will result, not in an even attenuation of the error, but an uneven change in the transfer function that, as a rule, includes higher order effects. The question that is perhaps not satisfactorily answered is; what is the audible nature of this change in transfer function, and is it better or worse than what we had before? Some argue that it is possible to push all the distortion below detectability, even with the issues above. Others argue that the analysis is poor, and flawed.
The question being asked here is: given a real world load, are there effects that have been hitherto unnoticed? Especially perhaps those involved with perhaps either transient phenomena, or multi tone phenomena.
Personally I will not be surprised to see some interesting fun come out of this. However, whether the current simulation tools have enough resolution, and enough accuracy in their models, that is likely to be an argument that will continue.
The other issue is the idea that if feedback is perfect it will correct all errors. This is not true either. Feedback reduces the error, by the feedback factor. This is from the basic definitions of how it works. A perfect feedback amplifier with an open loop distortion of 1% (or some arbitrary measure) and a closed loop feedback ratio of 10 will (in principle) have 0.1% of our mooted distortion. An op-amp is often designed with what is almost infinite open loop gain (well silly numbers anyway) and so, in principle, we could have arbitrary reduction of distortion. However the gain starts to drop with frequency (in any amplifier) and so quite quickly we cease to have arbitrary reduction. Further, the phase angle gets bigger, so eventually we get into trouble with oscillation. We must ensure the gain is less than one before this point is reached. This is the Nyquist stability criterion. This is all pretty basic (junior EE school) stuff.
Further, when the distortion products are reduced in magnitude, it is accepted that even this reduction cannot be perfect, and will result, not in an even attenuation of the error, but an uneven change in the transfer function that, as a rule, includes higher order effects. The question that is perhaps not satisfactorily answered is; what is the audible nature of this change in transfer function, and is it better or worse than what we had before? Some argue that it is possible to push all the distortion below detectability, even with the issues above. Others argue that the analysis is poor, and flawed.
The question being asked here is: given a real world load, are there effects that have been hitherto unnoticed? Especially perhaps those involved with perhaps either transient phenomena, or multi tone phenomena.
Personally I will not be surprised to see some interesting fun come out of this. However, whether the current simulation tools have enough resolution, and enough accuracy in their models, that is likely to be an argument that will continue.
Perhaps I have misunderstood one of the points that was raised regarding real world speakers and feedback?
I thought one of the ideas presented is that a real speaker will have both time delayed signal presented as "back EMF" due to phase angle changes from the reactive elements of the speaker and xover AND due to what is essentially "acoustic echo" - the speaker acting like its own microphone?
Yes, no?
Does the "long(er) delayed echo" component exist, and if it does will that signal matter?
Point 2... in general are we saying that an amplifier *without* global loop feedback (for example an output stage of multiple devices, well biased and with loop feedback from the driver back?) will be likely to perform *best* or *better* than most (or all??) global feedback amps are likely to do?
Question (3)... how do we feel something like the Halcro design fits into this picture (feedforward/feedback, multiple loop design) as compared to the topologies we've brought up so far??
_-_-bear
I thought one of the ideas presented is that a real speaker will have both time delayed signal presented as "back EMF" due to phase angle changes from the reactive elements of the speaker and xover AND due to what is essentially "acoustic echo" - the speaker acting like its own microphone?
Yes, no?
Does the "long(er) delayed echo" component exist, and if it does will that signal matter?
Point 2... in general are we saying that an amplifier *without* global loop feedback (for example an output stage of multiple devices, well biased and with loop feedback from the driver back?) will be likely to perform *best* or *better* than most (or all??) global feedback amps are likely to do?
Question (3)... how do we feel something like the Halcro design fits into this picture (feedforward/feedback, multiple loop design) as compared to the topologies we've brought up so far??
_-_-bear
MBK said:
More accurately, DOZ is a variation on the JLH. It is a bipolar
with 3 stages, and the Zen is a Mosfet with 1 stage.
I have curves of various amps into speakers versus resistive
loads, and they don't reveal anything you wouldn't expect. The
amplifier's distortion is often slightly smaller due to higher
impedances and reactive phase angle (yes, reactance often
reduces distortion) but you can also detect the distortion due to
the driver as reflected in the finite output impedance. In short,
no easy answers there.
Nelson,
so, in your experience and opinion, why do some people hear differences between standard, modern amps, when total and even higher order amp HD and IMD typically measure way below speaker driver distortion (in other words when amp distortion should be diluted to a neglegible addition to total distortion)?
I should add that of course not all people believe audible differences even exist below a certain THD under ABX conditions, as evidenced by the ubiquitous "all amps sound the same" posts. Where do you stand on this flavor of opinion?
so, in your experience and opinion, why do some people hear differences between standard, modern amps, when total and even higher order amp HD and IMD typically measure way below speaker driver distortion (in other words when amp distortion should be diluted to a neglegible addition to total distortion)?
I should add that of course not all people believe audible differences even exist below a certain THD under ABX conditions, as evidenced by the ubiquitous "all amps sound the same" posts. Where do you stand on this flavor of opinion?
Hi, Francis,
You explain it very well. I dont understand all the basics, I'm not EE.😀
Francis,
What makes a different sound when feedback is taken from VAS and feedback taken from output (the output is well biased)?
And what do you think of current feedback power amp? Does it better than ordinary voltage feedback?
You explain it very well. I dont understand all the basics, I'm not EE.😀
Francis,
What makes a different sound when feedback is taken from VAS and feedback taken from output (the output is well biased)?
And what do you think of current feedback power amp? Does it better than ordinary voltage feedback?
MBK said:
If you do not mind me answering, I will try. I believe it is not necessarily THD and IMD what annoys us. They are more likely digital audio troubles (jitter, for example, and HF D/A residuals) that are the issues. And non-harmonic distortions, that are difficult to find by steady-state sinus measurements. Harmonic analysis hides more than discovers.
There are two issues. If you take feedback from after the VAS you are living with whatever the transfer function of the output stage is - whereas if you include the output stage in the loop you convolve its transfer function with the feedback. Thus yielding a different transfer function. We have seen that this new transfer function will, in general, have many more higher order artefacts. The conventional wisdom is that a well designed output stage - and here this typically means class A (since you really don't want to contemplate crossover distortion in its unattenuated gory glory) has a reasonably benign transfer function, one that only includes low order harmonic components, indeed a mosfet stage might end up with little more than second order components, which are generally agreed to be, at worst, pleasant sounding, and at best inaudible. (However this is the issue we came into this thread with - a perfect second order harmonic is, curiously, not what the ear prefers as an octave, and not what some instruments create as their second harmonic - possibly leading to subtle but real perceived distortion effects. Certainly "roughness" of perceived sound.)
I really have little to say. Current amplifiers have bandwidth in their favour, but you can't drive a conventional loudspeaker with a current source. It simply doesn't work.
We seem to be breaking this down into the two components. The "and" bit about echo is pretty hard to quantify, and, as was pointed out earlier, likely to be a very very low level. You would need to multiply the acoustic energy in the room by transduction efficiency of the speaker to get the level of energy reflected back - and in general that is going to be very small. Not zero however. But before getting too enthusiastic one should actually do the numbers. Of course this component is not minimum phase, and any analysis in terms of THD etc. is impossible. You could get a rough idea of magnitude of the effect however, and then think about convolving it with the feedback system to see what it might look like.
Lordy, not shy of the hard ones are we? 🙂 I think this argument has raged for a decade odd. In a sense maybe it is an answer to this that we seek here. In looking at the nature of perceived distortion we might get a better handle on the answer here. There are a couple of issues. Is more but lower order better than less but higher order? (And what do we mean by "more" and "less".) What about transient effects that cannot be captured by THD and IM measurements? The latter question seems to have hardly been touched on in the question about amplifier topology.
I really wonder if anyone understands the implications of the Halcro design. Heck, I live in Adelaide, and work 5 minutes drive from Halcro. I have been very impressed with the sound, but have not spoken to anyone at Halcro (well apart from one of the guys at Minelab who fund Halcro.) I have build a couple of Cherry's designs, and they sound very very good. I find it odd that his work seems to be pretty much ignored. I see a range of multi-loop topologies about, but few NDFL, and scant attention paid to his careful analysis of loop stability. I also think feed-forward has a lot of potential, but is similarly ignored.
But before one gets too worried, one needs to have some criteria to put such designs up against. THD and IM are IMHO useful as sanity checks that an amplifier has no significant design flaws, but after that of limited utility. Listening is great, but is very hard to use as a scientific tool, since it lacks easily used metrics. It can be very hard to nail down the cause/effect. An amplifier is a very complex device. You cannot know that a change you make to address one issue, which changes the sound, is actually the issue that made the change in sound. You might have a good hunch, but it gets hard quickly. Analysis from first principles of a circuit is good, but hard.
I think perhaps there is more difference between class A and B output stages than just 'crossover distortion' ...
When lumanauw asked about 'current feedback' amps I believe he was referring to the topology similar to CFB opamps (seems I was just reading a thread discussing the semantics of this - I don't want to go there) and not to 'current amps' with high ouput impedance.
A CFB style amp has a low impedance inverting input - an example would be a Hiraga le Monster, or the JHL where the feedback goes to the emmiter of the input transistor(s). You can argue whether this is really current feedback but it is different than a diff pair. Is it better? - I have not built and listened to both (or measured) and therfore have no opinion - I suspect that both styles maybe done well or badly.
When lumanauw asked about 'current feedback' amps I believe he was referring to the topology similar to CFB opamps (seems I was just reading a thread discussing the semantics of this - I don't want to go there) and not to 'current amps' with high ouput impedance.
A CFB style amp has a low impedance inverting input - an example would be a Hiraga le Monster, or the JHL where the feedback goes to the emmiter of the input transistor(s). You can argue whether this is really current feedback but it is different than a diff pair. Is it better? - I have not built and listened to both (or measured) and therfore have no opinion - I suspect that both styles maybe done well or badly.
However it must surely be considered the dominant difference. The context was an amplifier where the output stage was outside the feedback loop. I don't care how carefully you bias the stage, if there is no correction for the transfer at the crossover it is going to sound bad. Clearly differences in output impedance and other bias related effects will also be important. But gobs of distortion are not a good start. Once it is inside the loop, well things become more interesting. But that wasn't the question.
Indeed. I don't have an opinion either. It does strike me as a slightly useless semantic war worrying about the terminology, so I stuck to the more conventional terminology of current output. Pure current mode is great for so many things anyway.
Francis_Vaughan said:
... while the Halcro is a commercial product, and he did get patents, the ideas and even the general topology isn't really "new." A friend of mine laughs at the Halcro's patents saying that he built a virtually similar amp about 15 years ago in college (he was an EE into audio). I assume everyone has a copy of their patents?
Interesting though...
Is it significant to note that they "do not recommend using the amp below a 4 ohm load"?? Hmmm...
As far as Cherry's work, my limited understanding is that his *topology* (the ones suggested in the paper) as presented represented an improvement in that day (and showed the way to subsequent understandings) but are flawed in a number of ways in light of the present - I seem to remember a thread here or in another forum outlining the specific issues?
_-_-bear
Hi, Francis,
What I mean with current feedback is like Cullingford means. It's a topology with feedback to emitors, like JLH or Accuphase or works of Peranders.
"if there is no correction........"
This discussion is getting more and more interesting 😀
What about this case. The output stage is low biased class AB (not class A), but with Hawksford Error Correction with it.
Can this be applied with good result in power amps that takes feedback excluding the output stage?
Low biased classAB surely have bad crossover distortion, high order artifacts.
On the other side, Hawksford EC is said tobe very fast, and can fix errors locally. Can Hawksford EC eliminate crossover distortion in low biased classAB, so we can take feedback from VAS without having to use hot classA?
What I mean with current feedback is like Cullingford means. It's a topology with feedback to emitors, like JLH or Accuphase or works of Peranders.
"if there is no correction........"
This discussion is getting more and more interesting 😀
What about this case. The output stage is low biased class AB (not class A), but with Hawksford Error Correction with it.
Can this be applied with good result in power amps that takes feedback excluding the output stage?
Low biased classAB surely have bad crossover distortion, high order artifacts.
On the other side, Hawksford EC is said tobe very fast, and can fix errors locally. Can Hawksford EC eliminate crossover distortion in low biased classAB, so we can take feedback from VAS without having to use hot classA?
Hi PMA,
you previously noticed better sound in your buffer when you changed configurations (low R load resistance, different op-amps) even though by your graphs the THD went, say, from -80 dB to -100 dB. Both should be equallly inaudibly low. Have you found any metric that would give a better explanation for the differences than the THD difference?
Francis,
I also find Linkwitz's distortion measurements very instructional and in some other thread I suggested one could use his bursts for amp testing as well (since they will produce harmonic and IMD products). I haven't found any data of burst measurements for electronics so far.
Linkwitz distinguishes between non-linear distorion (harmonic, IMD) and what he calls linear distortion, i.e. energy storage. Now, energy storage means resonance, and this information should show up in the FR curve as well. In other words, while the burst testing shows the data better, and presents a severe test, it does not by itself concern a different metric.
In other words, with FR, phase, and HD/IMD, ideally from multitone bursts, we should have a lot of information already. What is missing?
you previously noticed better sound in your buffer when you changed configurations (low R load resistance, different op-amps) even though by your graphs the THD went, say, from -80 dB to -100 dB. Both should be equallly inaudibly low. Have you found any metric that would give a better explanation for the differences than the THD difference?
Francis,
I also find Linkwitz's distortion measurements very instructional and in some other thread I suggested one could use his bursts for amp testing as well (since they will produce harmonic and IMD products). I haven't found any data of burst measurements for electronics so far.
Linkwitz distinguishes between non-linear distorion (harmonic, IMD) and what he calls linear distortion, i.e. energy storage. Now, energy storage means resonance, and this information should show up in the FR curve as well. In other words, while the burst testing shows the data better, and presents a severe test, it does not by itself concern a different metric.
In other words, with FR, phase, and HD/IMD, ideally from multitone bursts, we should have a lot of information already. What is missing?
Missing is time resolution. Transient response.MBK said:
Spectral analysis must change to wavelet transform to dig into that realm, imo.
Mistake is to bet your money that any reactive electro-mechanical system has fixed static transfer function. We barely have means to measure transfer function at single condition, and have no standard means to quantify signal triggered changes to it. Thus we don't even have much ground to discuss audibility of that.
Note that energy storage isn't necessarily resonance. Energy stored could be released in totally unrelated spectrum, thus FR shows very little about it directly.
- Status
- Not open for further replies.