"Aquarius"
"When the Moon is in the Seventh House,
And Jupiter aligns with Mars"
. . . Then, all these thread's schemes work perfectly.
Check your amplifier's output phase (or more accurately its group delay), when it is driving most real world loudspeakers.
That output node is where you have to take the sample for various negative feedback schemes (but only if you want the output to be perfectly correct when connected to your loudspeaker).
But do not lose heart . . .
Most schemes if implemented properly, work very very well.
And, that also applies to non-negative feedback amplifiers that are implemented properly.
I am stealing an Audio conference / show statement: "Shut up and Listen!"
And from a good online magazine: "Enjoy the Music."
Just my opinions and theory.
"When the Moon is in the Seventh House,
And Jupiter aligns with Mars"
. . . Then, all these thread's schemes work perfectly.
Check your amplifier's output phase (or more accurately its group delay), when it is driving most real world loudspeakers.
That output node is where you have to take the sample for various negative feedback schemes (but only if you want the output to be perfectly correct when connected to your loudspeaker).
But do not lose heart . . .
Most schemes if implemented properly, work very very well.
And, that also applies to non-negative feedback amplifiers that are implemented properly.
I am stealing an Audio conference / show statement: "Shut up and Listen!"
And from a good online magazine: "Enjoy the Music."
Just my opinions and theory.
Last edited:
Off topic, but that reminds me of Pa Pinkelman, a Dutch cartoon character who could do magic, but only when the Moon was in the right phase and the wind came from the right direction.
Thanks Jan, yes that's exactly what I've done in my last PP amps and what I'm implementing in a SE.
The advantage I've found is that the nfb can be applied to gnfb-less amps without affecting the sensitivity, you can reach class D-ish Zout and you can adjust it while playing music. I liked this solution alot, that's why I'm asking. Thanks for your reply!
What you suggest is similar to what is called pre-distortion that attempts to cancel distortion by predicting the amplifier distortion. But the predistortion is always an approximation and the results are never perfect.
Negative feedback is similar to the math where you keep adding a= a +x(b-a), (x<1). You never reach a=b. You cannot feedback just the distortion, although "current dumping" attempts it. The difference between the input and feedback cannot be zero, or there would be no input. So if the OLG is 100 times the feedback, the gain is actually 99% of the feedback ratio.
"Perfect is the Enemy of Good"
I hope all of you are testing your perfect amplifiers using a simulated loudspeaker load.
Then, graduate your tests to include your loudspeaker (going to have to do some burst testing to be able to test at high power levels, and at the same time keep the average RMS power within all your speaker drivers safe area.
Be sure to wear a set of 'gunfire earmuffs'.
I used to have a pair of ADS loudspeakers, they had really troublesome multiple alternating from highly reactive low impedance, to low resistance, and beck to highly reactive low impedance, across the range from 5kHz all the way to over 20kHz.
Those were very hard on many global negative feedback amplifiers.
I hope all of you are testing your perfect amplifiers using a simulated loudspeaker load.
Then, graduate your tests to include your loudspeaker (going to have to do some burst testing to be able to test at high power levels, and at the same time keep the average RMS power within all your speaker drivers safe area.
Be sure to wear a set of 'gunfire earmuffs'.
I used to have a pair of ADS loudspeakers, they had really troublesome multiple alternating from highly reactive low impedance, to low resistance, and beck to highly reactive low impedance, across the range from 5kHz all the way to over 20kHz.
Those were very hard on many global negative feedback amplifiers.
Yes, it can be applied without changing the amp sensitivity, because the excess gain that you need for feedback is generated in the feedback loop.
It can even be done around a power output stage with a gain lower than what the final gain should be. Even with a gain = 1 power follower.
It gives you much more freedom also with respect to stability and compensation because it all can happen at low level.
Jan
I actually use multiple feedback pathes joined in a single point, plus a fifth in my SE: a voltage divider from anode to grid of the ouput tube (local negative to triodize output tube curves, still being connected in pentode more) then going to the cathode of the driver (positive over two stages to improve sensitivity) then an unbypassed cathode resistor (local negative in current to linearize the stage) going to an opamp instead of ground to apply feedback (gnfb with distortion only, ideal for low level signals being supplied at +-15V).
The output tube is cathode driven through a pmosfet à la UNSET, and I use this low impedance to bootstrap the laod of the driver to improve linearity.
Yes, indeed the OPAMP has been a simple way to avoid one coupling cap (I apply feedback directly to the cathode of the driver) and to modify “on the fly” the amount of feedback to the system.
May I share with you the concept through PM, to get your opinion?
Thanks in advance,
Roberto
jan.didden,
I was only trying to be fair when I said Theory.
Because I only owned a single pair of loudspeaker models that had really nasty impedance curves, I did not have a chance to make tests of amplifier problems on Even Worse Loudspeakers.
Without more testing to verify the results of an amplifier driving an even worse speaker, I just said that the bad speakers I had, and the measurement results I got, might have not been fair (Verify using an independent and separate setup and different speaker)
Without more tests of nasty loudspeakers to verify the principle, it is just "Theory".
Your Theoretical Results May Vary versus Mine.
I hope that explains it.
I was only trying to be fair when I said Theory.
Because I only owned a single pair of loudspeaker models that had really nasty impedance curves, I did not have a chance to make tests of amplifier problems on Even Worse Loudspeakers.
Without more testing to verify the results of an amplifier driving an even worse speaker, I just said that the bad speakers I had, and the measurement results I got, might have not been fair (Verify using an independent and separate setup and different speaker)
Without more tests of nasty loudspeakers to verify the principle, it is just "Theory".
Your Theoretical Results May Vary versus Mine.
I hope that explains it.
Last edited:
Exactly so.
Class D has a fundamental limitation: the sample rate.
This results in poor bandwidth and distortion.
Putzeys succeeds in hiding these issues from THD measurements by sleight of hand. But our ears are not fooled. Not are proper electrical measurements.
You really should read up on some more than 90 years old theory: https://en.wikipedia.org/wiki/Nyquist_stability_criterion
You may find Nyquist makes you feel confident, but what does this have to do with audio reproduction? Lol
That's why any sensible amp has a Zobel network and output inductor, so reactive loads are drivable. However no amp can be expected to function into impedances whose magnitude is less than they designed for, whether they use feedback or not. Typically an amp will current-clip or the protection circuit cut-out, or output devices simply blow up from secondary breakdown. Or if the amp lacks output-to-rail diodes it may over-voltage the output devices
I think its wiser to test with a selection of known fixed reactive loads, a specific loudspeaker model isn't readily reproducable. A programmable active load would be the most flexible means for testing amp capabilities, given a large enough budget.
It's the small-signal stability criterion for feedback systems, such as audio amplifiers.
For an audio amplifier, small signal stability is not enough. I decent amplifier is critically or over damped. And it must cleanly and quickly recover from overload, ie rail sticking. And it must respond linearly to back EMF, ie reactive loads. Much of the material on the subject is designed to impress academics which only serves to hide the important concepts.
If you don’t have small signal stability you’re dead in the water - you can’t reasonably expect unconditional large signal or parametric stability if the basic nyquist stability isn’t satisfied.
True, reactive loading is a bitch. It dumps all the stored energy into your output transistors, and usually degrades the phase margin. But it’s unrelated to actual “back EMF”, which tends to REDUCE the amount of current drawn by a lightly loaded electric motor. It’s in phase with the applied voltage, and opposite in sign.
True, reactive loading is a bitch. It dumps all the stored energy into your output transistors, and usually degrades the phase margin. But it’s unrelated to actual “back EMF”, which tends to REDUCE the amount of current drawn by a lightly loaded electric motor. It’s in phase with the applied voltage, and opposite in sign.
I couldn't agree with you more, small-signal stability is necessary but insufficient. In fact I can't think of any application of control theory where small-signal stability would be sufficient.
That's why I asked the designer of this amplifier with third-order compensation https://www.diyaudio.com/community/threads/three-pole-compensated-blameless-clone.397740/ to check recovery from clipping. It recovers beautifully. On the other hand, I've seen small-signal-stable amplifiers with simple single-dominant-pole compensation burst into large-signal oscillations at start-up and never recover.
I think there is a very pragmatic reason why small-signal stability gets more attention than large-signal stability in electronics engineering education: the mathematics to check large-signal stability are just too complicated. A pragmatic way to deal with that is to design for small-signal stability, keep your fingers crossed and do some sanity checks (start up, recovery from clipping) in a simulator and/or a real life model to check large-signal stability.
Off topic: I wonder how this is handled in other engineering disciplines, and particularly how it was handled before simulators became available. You can't do experiments like that on a nuclear power plant.
Simulation?
In the old days, engineers did it by long hand, hard work, good analysis, experience, etc.
Did you ever consider how Western Electric drew all those curves of the 300B?
There were no vacuum tube curve tracers back then.
The best example of a design project I know of had 3 engineers involved.
2 said it will work, 1 said it will not work.
The "it" was the Tacoma Narrows Bridge.
1 engineer was right.
I do not plan to drive every loudspeaker model in the world.
I do not use a Zobel network on the output of my amplifiers.
I do not need a Zobel network. You go figure it out.
In the old days, engineers did it by long hand, hard work, good analysis, experience, etc.
Did you ever consider how Western Electric drew all those curves of the 300B?
There were no vacuum tube curve tracers back then.
The best example of a design project I know of had 3 engineers involved.
2 said it will work, 1 said it will not work.
The "it" was the Tacoma Narrows Bridge.
1 engineer was right.
I do not plan to drive every loudspeaker model in the world.
I do not use a Zobel network on the output of my amplifiers.
I do not need a Zobel network. You go figure it out.
Last edited:
Small signal stability analysis is given such attention in EE education because it is the basis of all large signal stability analysis. The math isn’t THAT hard to analyze LS stability. It just requires an infinite number of permutations where small signal stability must be maintained in order to rigorously analyze. It’s the same pole zero analysis - it’s just that they all
MOVE, dependent on pretty much everything. You have to pick some sort of reasonable subset of excitation and load conditions, run until you’re out of computer power and patience, and call it good.
MOVE, dependent on pretty much everything. You have to pick some sort of reasonable subset of excitation and load conditions, run until you’re out of computer power and patience, and call it good.