lumanauw said:
David,
Feedback works 100% according to theory. I challenge you to prove this is not the case.
What we have here is gross simplifications and ignorance of the theory. The theory DOES NOT predict that feedback removes 100% of the distortion. Theory DOES NOT predict that a fb amp has zero Zout. Theory DOES NOT predict that an amp should sound the same whether you take the fb from output, or from Vas, or anywhere else. In fact, theory DOES predict that the measured performance in these cases will be different. Of course, whether that is audible depends on many other factors.
Jan Didden
lumanauw said:
David,
The situation you describe is not feedback, but is trying to linearise the loop linearity by matching devices with complementary transfer characteristic. This is fundamentally different from feedback, where you manipulate the SIGNAL to make it have a complementary linearity (or call it pre-distortion) so that the output from the non-linear device transfer characteristic gives you the wanted, original non-distorted signal. These are two things that are non-related, although a smart designer may try to combine the two techniques.
Jan Didden
mauropenasa said:
But I think you lose sight of the objective. You throw out the baby with the bath-water. You can use transimpedance amps to suppress effects of EMF *to the amp*, but that means that the EMF effect is fully remaining at the output and will necessarily intermodulate and interfere with the wanted output signal. What do you want, low EMF suseptibility or good sound?
Jan Didden
Hi, Janneman
. That's why in the first sentence, I wrote "I'm not EE". Maybe someone here (where's Swedish Chef?) can answer (or strenghten) Janneman's challange?
To some (including me) I can hear pleasant and not pleasant power amp reproduction, and I can hear difference if the output stage is included or not included in feedback.
If we look at a differential, the left base is feeded by input signal, the right base is feeded with feedback signal+backEMF.
Can the differential eliminate the back EMF? If the left differential also has the same back EMF, then back EMF can be eliminated. That's the nice thing about NP's SUSY.
Hi, Amplifierguru,
I have a question. If I take the feedback point, one from output stage, and one from VAS (with the same amp), does these 2 will have different damping factor?
I cannot answer your challange
. That's why in the first sentence, I wrote "I'm not EE". Maybe someone here (where's Swedish Chef?) can answer (or strenghten) Janneman's challange? So....what is the feedback theory saying about?
How does Mr. Jean Hiraga know that 0.008% of high order distortion will be noted by ear? The more stages in feedback amp will give more and higher harmonic artifacts. In simulator, I cannot see harmonics more than 7th order. Is the "high order artifacts which is not pleasant to human ear" is a hoax only?
To some (including me) I can hear pleasant and not pleasant power amp reproduction, and I can hear difference if the output stage is included or not included in feedback.
You're right, I'm typing wrong there. Like previously stated in pages before, exponential needs exponential device to perform differential stage, not (1/exponential) device.
If we look at a differential, the left base is feeded by input signal, the right base is feeded with feedback signal+backEMF.
Can the differential eliminate the back EMF? If the left differential also has the same back EMF, then back EMF can be eliminated. That's the nice thing about NP's SUSY.
Hi, Amplifierguru,
I have a question. If I take the feedback point, one from output stage, and one from VAS (with the same amp), does these 2 will have different damping factor?
lumanauw said:
Lets take an amp with open loop gain "A" and a feedback factor beta.
The closed loop amplifier gain equation is: CL gain = A/ ( 1+ (beta*A)). So, if the open loop gain A * beta is very much larger than 1, you can replace (1+beta*A) with beta*A. The equation then *approaches* (but never reaches) CL gain = 1/beta. Now, if it WAS 1/beta, it would mean that whatever A does (non-linearity, drift, offset) would no longer be part of the output. So, it really depends on the product of A (open loop gain) and beta (feedback factor) how strong the reduction of non-linearities really is. Edit: That is of course the reason why we try to get large open loop gain, so we have a lot of gain to *waste* through the beta, and have a large reduction of non-linearities.
You also mentioned taking feedback from the Vas. In that case, the "A" in the equation is only the amplifier up to the Vas. The output stage would then be added on to the feedback amp and work open loop. That would mean that in that case the feedback does not reduce the output stage non-linearities.
All this is a first order approximation leaving out freq dependent effects. However, these are easily included by replacing "A" with A *(1+s), where s is the complex operator. But the gain equation doesn't change.
More theory here: http://www.numberwatch.co.uk/feedback.htm .
Jan Didden
Yes, when taking feedback from the output the resistance from the output stage is reduced by the overall feedback, when using feedback from the VAS stage you'll get a completely different behavior.
Why? First of all the VAS stage output resistance is reduced by the feedback factor. Second the resistance from the output stage
Re + intrinsic emitter resistance of the power BJT + 1/beta * (intrinsic base resistance + output resistance of the previous driver stage)
is not inside the feedback loop (with a MOSFET it is similar). Third the interaction between VAS and loading the VAS by the output stage is different. Under typical conditions the output resistance of the VAS is about 100k ohms. In the case of using feedback from the output a signal controlled CURRENT source (approximation) is loaded with an EF. In the second case the VAS output is approximately a signal controlled VOLTAGE source which is loaded by the output stage.
This will not only affect the damping factor it will also affect your distortion spectrum because of totally different loading mechanism of the VAS.
bocka said:
Bocka,
If he takes the feedback from the Vas, ideally the Vas loading becomes unimportant, because, as you mentioned, the Vas will approach a voltage source. So, Vas non-linearity and loading effects are minimised. But of course the full unreduced output stage non-linearity (Gm doubling, xover distortion, beta droop) remains.
Jan Didden
Hi, Janneman,
Thanks for the explenation. In college (I major in Mechanical Engineering), there was a similiar lesson about "System and Control". Studying Bode plot and stability for Mechanical engineering (like for heat boiler or water system)
But it's like headed with "Chicken or Egg first" question.
The "approach" will always be correct, because it only analyze a "black box" where it compares output to input, no matter what process in the "black box". So, the feedback theory will always be right.
In making an audio power amp, are we making mathematical analysis first (then draw schematic) or draw schematic first (and study its behavior later)? For me, I built the amp first and try to understand why eachone sound different. Is there a "mathcad" or analyzing program like SIM that can predict how a design will sound like? It will save many time and money. Mauro's work can be about it, "picturing the sound".
Hi, Bocka,
Thanks for the explenation Which do you personally use?
Thanks for the explenation. In college (I major in Mechanical Engineering), there was a similiar lesson about "System and Control". Studying Bode plot and stability for Mechanical engineering (like for heat boiler or water system)
But it's like headed with "Chicken or Egg first" question.
The "approach" will always be correct, because it only analyze a "black box" where it compares output to input, no matter what process in the "black box". So, the feedback theory will always be right.
In making an audio power amp, are we making mathematical analysis first (then draw schematic) or draw schematic first (and study its behavior later)? For me, I built the amp first and try to understand why eachone sound different. Is there a "mathcad" or analyzing program like SIM that can predict how a design will sound like? It will save many time and money. Mauro's work can be about it, "picturing the sound".
Many things are made their mathematical approach (like any not-chaotic curve can be made its mathematical approximation), but then again it is only approximation, not the exact behavior. If you combine many non-exact parameter to equation, it will drift quite some to the actual result.
Like a transistor will always have capacitance. Where there is AC signal and capacitance, there will always be sin/cos/i(imaginary number) involved. This frequency and phase shift is the one who makes high-order harmonic, doesn't it? If so, trying to minimize using transistor(which always have capacitance)=less stages amp=can lead to less high order harmonics with feedback system?
How does this corelates with Mr.Otala who wrote his famous paper about "not so high OL gain + large OL bandwith" for better sound?
Hi, Bocka,
Thanks for the explenation
Which do you personally use?
The reason feedback increases higher order harmonics is that the distorted feedback signal is distorted again when fed back to the input. For example, an amplifier which produces 2nd harmonics will have a fundamental and 2nd harmonic fed back to it. This second harmonic feedback will then itself be distorted, resulting in 4th harmonic, then again for 8th etc. Even an amp with perfect frequency/phase response would suffer the same effect.lumanauw said:
Yes, but not for the reason you're thinking of. As with feedback adding harmonics of harmonics, so multiple stages in series will distort the distortions of the previous stages. This happens regardless of feedback. With feedback it can go either way: More stages means more distortion, but also higher gain, which reduces distortion. Enough feedback will reduce even the higher order harmonics, so whether more or less stages is better depends on how much distortion they add relative to their gain.lumanauw said:
lumanauw said:
Well, David, obviously if you want a reasonble speaker damping, no nasty xover distortion, and supression of EMF, nfb is a simple, well tried out and effective solution. Of course, you need to know what you are doing. And, as you referred to your courses, nfb is not just something for audio. I said it before, if nfb didn't exist, we would all be dead within the next 4 minutes. Our breathing reflex is a prime example of nfb. I haven't heard anybody say that he wanted to breath open loop. You probably would either suffocate or hyperventilate, neither of those would be comfortable in the long run.
Another example: strech your right arm outward to the right. Look straight ahead. Now take the decision to move your outstreched arm straight in front of you, AS FAST AS YOU CAN. Watch the tip of your fingers come into your field of vision if you do that. 99.99% of people will see the fingertips slightly overshoot a little to the left and then move back in front of you to the position you decided to move it in the first place. Watch carefully because you're faster then you think! What happened? NFB, that's what happened. because you decided to do it as fast as possible, you increase the brain-motor circuit loop gain to its max. Because there is an inherently physical delay in the way your muscles activate, there is a phase shift between the position of the arm and the commenad to move or stop it: you overshoot.
Now do the same at your leisure. No overshoot. of course, because doing it at your leisure requires much less loop gain, and even if the phase shift is still there, the loop is much more stable. Elementary control system theory as I am sure you learned.
Jan Didden
lumanauw said:
Hi David,
I don't know what you mean with chicken&egg situation?
Anyway, the black box thing is a simplification allowing us to analyse complex circuits without worrying about the circuit. Don't worry about what's in the box, that the very beauty of it!
If you have a black box with for instance an unloaded output voltage of 5V, and a short circuit current of 1 amps, that defines the circuit (let's talk DC for this example, although it holds for any frequency with some extensions).
What's in the box? Maybe it is an ideal voltage source of 5V DC with an output series resistor of 5 ohms. Or maybe it is a currentsource of 1 amps with a parallel resistor of 5 ohms. Or maybe it is a combination of 33 voltage sources, 6 current sources and 87 complex resistor networks. Who cares, because as far as DC analysis, and its use in a circuit is concerned, NO ENGINEER ON THIS EARTH CAN SEE ANY DIFFERENCE OR FIND OUT WHAT IT IS. NOBODY, because they are equivalent as long as the open voltage is 5V and the short circuit current is 1 amp.
In effect, that's how audio designers work also. They need a black box that accepts up to 1V rms between 20Hz and 20kHz, and drives 4 ohms without sagging, and has a gain of 28. What's in the box? Well, if you have two boxes that fullfill those requirements, nobody has any clue, and nobody knows whether the boxes are different or identical. And why would they want to know? But then the fun starts, because you would like to have as little in the box, and as simple, and as cheap as possible.
The problem with audio is that it also must look very good on a picture. So, you need to put a lot of shiny components, nicely aligned in rows, and lots of gold plated stuff in the box because otherwise it will look not as good as the competition in the magazine pictures. So, the requirements of what is in the box continue to expand, without having anything to do with how it sounds, and that drives up the prices. And it is YOU who will pay for this additional marketing.
Jan Didden,
lumanauw said:
Not a fair question.
You have the benefit of having seen lots of circuits, understand them, and then decide to built your own, based on what you have seen, conciously or not. But if you came from mars, without ever having seen an amp. circuit, and we gave you a bunch of parts & data sheets and say: built us an amp? No way you could get started without some sort of theoretical / mathematical analysis first!
Jan Didden
Trying to focus the controversy
Splendid Jan!!
Now, I am afraid the root of this controversy lies in the fact - with due respect - that those not familiar with electrical engineering and the (moderate) level of math implied, are working off heuristic approaches.
What I mean with this, is that as long as one's insight is mostly based on a part intuitive, part reasoned thinking, cannot gauge which effects are relevant and which are not.
Again this is with due respect to those hard working and dedicated diyers and fanatics who for different circumstances have not received this specialized form of education. Only they must acknowledge there is a point where they must accept other people is better qualified by profession, to whom they should deposit credibility as long as they demonstrate to be both well meant and proficient.
We do not want to imply others as ignorant, but only lacking a moderately complex set of skills we acquired through a substantial learning and working period of time.
Now:
Accepting there is no such thing as a perfect amplifier (wire with gain), one must settle for second best.
Second best is an amplifier with as low as practical introduced arctifacts. Better yet, start with a well tested and accepted set of thresholds below which it is accepted from the standpoint of human perception (up to golden ear standards) and for all possible program material.
Under this circumstances, an amplifier meeting measured arctifact levels below threshold should be undistinguishable from a perfect one.
It should now be obvious actual amplifier topology is completely irrelevant as long as the required performance thresholds are met.
Just to throw some fuel into the furnance, my (very elemental) performance threshold should be:
Less than 0.001% THD 20-20KHz
Below 0.5 dB and 1 deg. gain flatness and phase shift in the same range.
This should hold with a fairly representative load (or actual load for a particular instance)
(dunking for cover)
Rodolfo
Splendid Jan!!
Now, I am afraid the root of this controversy lies in the fact - with due respect - that those not familiar with electrical engineering and the (moderate) level of math implied, are working off heuristic approaches.
What I mean with this, is that as long as one's insight is mostly based on a part intuitive, part reasoned thinking, cannot gauge which effects are relevant and which are not.
Again this is with due respect to those hard working and dedicated diyers and fanatics who for different circumstances have not received this specialized form of education. Only they must acknowledge there is a point where they must accept other people is better qualified by profession, to whom they should deposit credibility as long as they demonstrate to be both well meant and proficient.
We do not want to imply others as ignorant, but only lacking a moderately complex set of skills we acquired through a substantial learning and working period of time.
Now:
Accepting there is no such thing as a perfect amplifier (wire with gain), one must settle for second best.
Second best is an amplifier with as low as practical introduced arctifacts. Better yet, start with a well tested and accepted set of thresholds below which it is accepted from the standpoint of human perception (up to golden ear standards) and for all possible program material.
Under this circumstances, an amplifier meeting measured arctifact levels below threshold should be undistinguishable from a perfect one.
It should now be obvious actual amplifier topology is completely irrelevant as long as the required performance thresholds are met.
Just to throw some fuel into the furnance, my (very elemental) performance threshold should be:
Less than 0.001% THD 20-20KHz
Below 0.5 dB and 1 deg. gain flatness and phase shift in the same range.
This should hold with a fairly representative load (or actual load for a particular instance)
(dunking for cover)
Rodolfo
Hi lumanauw,
personally I prefer taking feedback from the output stage und using a triple EF (important). To me this configuration has the lowest distortion.
Yes correct. The only trade off is that we have to use very high feedback resistors otherwise the VAS loading by the feedback resistors will reduce the feedback factor.
To me it is more important that not only the non-linearity itself is minimised. To prevent high-order distortion the transfer characteristic must not have sharp bends. Similar like tubes. The best solution I have is using a triple EF, much better than CFP or two transistor EF. Best solution is class A of course but this will limit the power to some ten watts.
Using a lot of output transistors is also important, real devices show a beta drop at currents larger than about 4 amp. Biasing the output devices with about 0.3 to 0.5 amps seems a resonable current. Decreasing the emitter resistors as much as possible. Practically this is limited to about 0.2 ohm because lower resistors sometimes lead to thermal instability.
BTW nobody has mentioned that I've forgotten the component Vt * gm in the formula of the bipolar output resistance...
personally I prefer taking feedback from the output stage und using a triple EF (important). To me this configuration has the lowest distortion.
Yes correct. The only trade off is that we have to use very high feedback resistors otherwise the VAS loading by the feedback resistors will reduce the feedback factor.
To me it is more important that not only the non-linearity itself is minimised. To prevent high-order distortion the transfer characteristic must not have sharp bends. Similar like tubes. The best solution I have is using a triple EF, much better than CFP or two transistor EF. Best solution is class A of course but this will limit the power to some ten watts.
Using a lot of output transistors is also important, real devices show a beta drop at currents larger than about 4 amp. Biasing the output devices with about 0.3 to 0.5 amps seems a resonable current. Decreasing the emitter resistors as much as possible. Practically this is limited to about 0.2 ohm because lower resistors sometimes lead to thermal instability.
BTW nobody has mentioned that I've forgotten the component Vt * gm in the formula of the bipolar output resistance...
bocka said:
Bocka,
I agree with most of your post. But on the first para, I think it was either Leach or baxandall who prove that the loading of the output by the feedback network has no effect as long as the stage can drive the fb network in terms of current. In the final analysis, the effect was the same. Anyone remember more of this than I do?
Jan Didden
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.