Not really.
Several people have already commented on the fact that a phase shift that is not exactly proportional to frequency does not correspond to a genuine delay, but I think that's not really the point.
As the fed back signal is getting more and more out of phase with the input signal, their difference is getting larger and the feedback is becoming less effective. That's correct.
You can also describe it in terms of dropping loop gain, the conclusion remains that a 50 kHz loop bandwidth is rather small when you want to reduce the distortion at 20 kHz. There is a reason why audio op-amps typically have gain-bandwidth products in the 3 MHz...100 MHz range.
So an op amp that has gain bandwidth products in the 3 MHz to 100 MHz frequency range has a gain of 1 (unity)somewhere between from 3 MHz to 100 MHz.
I think most commercial vacuum tube Hi Fi / Stereo power amplifiers with an output transformer, do not have unity gain at 10 MHz, and certainly not at 30 MHz or 100 MHz.
Those who are interested in those kind of bandwidth products, can 'cross over' to the solid state threads for a discussion of gain bandwidth products from 10 MHz to 100 MHz,
or . . .
perhaps most tube amplifiers that have that much gain bandwidth typically are OTL amplifiers.
Everything is what it seems to be, until you find out it is different.
I think most commercial vacuum tube Hi Fi / Stereo power amplifiers with an output transformer, do not have unity gain at 10 MHz, and certainly not at 30 MHz or 100 MHz.
Those who are interested in those kind of bandwidth products, can 'cross over' to the solid state threads for a discussion of gain bandwidth products from 10 MHz to 100 MHz,
or . . .
perhaps most tube amplifiers that have that much gain bandwidth typically are OTL amplifiers.
Everything is what it seems to be, until you find out it is different.
Yes, which for example corresponds to a bandwidth between 300 kHz and 10 MHz when you set the gain to 10.
Indeed. The transformer will make it very difficult to keep such an amplifier stable.
I think this is one of the main reasons why negative feedback valve amplifiers usually have more distortion than negative feedback transistor amplifiers. The non-linearity of a valve is much more benign than that of the voltage-to-current transfer of a bipolar transistor, but applying strong overall feedback is much more difficult.
I don't know if there are any large-gain-bandwidth-product negative-feedback valve amplifiers, but it should indeed be easier to design such a thing as an OTL design than as a design with a transformer in the loop.
Baxandall, and more recently Putzeys, made the point that there can be an "excluded middle" in effectiveness of feedback, at least in a pathological case of large forward path distortion. Small amounts of feedback, even over a single sufficiently non-linear stage, can increase the amount of distortion products when weighted for our perception.
Vacuum valves can easily be operated hot enough to have monotonically decreasing distortion with decreasing signal level, and are (best case, but not difficult) similar in distortion levels to practical loudspeakers. (Yes, exotic loudspeakers can be better, but so can exotic amplifiers. But they're playing in the same ballpark.) Loudspeaker distortions tend to be monotonic with level also, fortunately because that would be tough to fix. Amplifiers that are not monotonic have a whole extra issue when playing music, with its 60dB plus dynamic range and our perception.
The greatly missed DF96 once tried to explain through my thick skull how re-entrant distortion could work without the Black model being incorrect, but left us before I could catch on. So I remain ignorant.
All good fortune,
Chris
Vacuum valves can easily be operated hot enough to have monotonically decreasing distortion with decreasing signal level, and are (best case, but not difficult) similar in distortion levels to practical loudspeakers. (Yes, exotic loudspeakers can be better, but so can exotic amplifiers. But they're playing in the same ballpark.) Loudspeaker distortions tend to be monotonic with level also, fortunately because that would be tough to fix. Amplifiers that are not monotonic have a whole extra issue when playing music, with its 60dB plus dynamic range and our perception.
The greatly missed DF96 once tried to explain through my thick skull how re-entrant distortion could work without the Black model being incorrect, but left us before I could catch on. So I remain ignorant.
All good fortune,
Chris
Agree on DF96 being missed.
Chris are you saying that re-entrant distortion cannot be explained by the usual feedback math?
I hope you say no!
Jan
Chris are you saying that re-entrant distortion cannot be explained by the usual feedback math?
I hope you say no!
Jan
Not so much "no", but more "I don't see how". The Black model seems to assume that distortion only happens once in the forward path. Putzeys' is more complete, but the re-entrant distortion products still seem to appear from nowhere. I'm sure it's just a problem in my understanding, but I'm not smart enough to work through it myself.
All good fortune,
Chris
All good fortune,
Chris
With very few exceptions in the nuvistor range, tube amps have low gain to start with...ecc83 allows for max 100x voltage gain , their distortions can't get too high anyway unless bad design is present...there's less left for correction.Tubes are current devices too, just the impedances are so high that we judge them as voltage devices and they don't exhibit avalanche current amplification similar to transistors due to their nature so their slew rate limitation help if negative feedback is not very efficient or quick enough depending of course on tube.More difficult task with e810f than with ecc83, so good designs emploied some short path feedback technique like bootstrap, mu follower, white cathode follower...If we starve the anode with current we see all distortion rise.They need minimum current to work linear too.
I guess this is the issue that feedback turns a forward path curve with a quadratic term into a square-root-like error signal. There is an article from 1938 about that on Jan's other site, https://linearaudio.nl/sites/linearaudio.net/files/Farren feedback1938.pdf It has a reference to an even older article.
It's in section 7. The older article is reference 10, Feldtkeller's article in German.
Thanks very much for this. He sees the same problem with Black's model that puzzled me, and uses the term "secondary distortion" for what we call re-entrant distortion. I'll try to track down Feldtkeller, although my German is more than a half century old. It looks like he may have described the harmonic series first.
Very much thanks,
Chris
Very much thanks,
Chris
Some of the Dynaco amplifiers had a small cap from one output tube plate summing into the global N Fdbk path. That would cause the N Fdbk to transition at some HF from global to local (taking the OT out of loop), allowing tube distortion to still be corrected to high frequencies, and the output Z from the tubes to be kept low. (it would work better with two caps back to a differential stage)
The OT distortion comes from OT magnetizing current interacting with the tube output impedance and OT winding resistance. Low output Z from the tubes fixes most of that, but the primary winding resistance still allows some error thru (Verror = Ierror x Wresistance). An output tube current sensing positive Fdbk can fix that last bit by increasing tube drive by just enough to fix the winding resistive loss. (essentially a negative resistance cancelling the primary resistance. Magnetizing current not effecting the secondary resistance.) So one is left with secondary winding resistance affecting the damping factor, which only matters at low freq. anyway, where global N Fdbk fixes it.
All this talk of frequencies just confuses matters for N Fdbk discussions. The signal has a voltage and a derivative at any point in time. That's it. Frequency is a derived concept, measured over some extent in time. The cap across the N Fdbk resistor just predictively aligns the N Feedback with the incoming signal time wise, much like a PID controller uses for process control. (time alignment. or phase alignment in frequency parlance) The signal arriving when the N Fdbk gets applied is just the slightly earlier (by nSecs) signal + k times the derivative. k depending on reactive elements in the amplifier + delay time. Of course stability of the amplifier is quite important here too. The derivative can only predict accurately over well less than the internal amplifier time components. (reactances and delay) and the signal variation time. So I don't see any issue with N Fdbk delay as long as the amplifier stays well stable with sufficient bandwidth.
Each recursive loop of error thru the amplifier gets time aligned with the signal + amplifier delay. The remaining error due to insufficient loop gain ( to completely fix the error the 1st time thru) gets progressively fixed better each time around. So the output error should stepwise decrease, asymptotically approaching perfection. The problems arise when the signal is changing faster than the system can fix them. So either more gain (with stability) or more amplifier bandwidth (less loop correction time allowing more recursions) are needed.
The OT distortion comes from OT magnetizing current interacting with the tube output impedance and OT winding resistance. Low output Z from the tubes fixes most of that, but the primary winding resistance still allows some error thru (Verror = Ierror x Wresistance). An output tube current sensing positive Fdbk can fix that last bit by increasing tube drive by just enough to fix the winding resistive loss. (essentially a negative resistance cancelling the primary resistance. Magnetizing current not effecting the secondary resistance.) So one is left with secondary winding resistance affecting the damping factor, which only matters at low freq. anyway, where global N Fdbk fixes it.
All this talk of frequencies just confuses matters for N Fdbk discussions. The signal has a voltage and a derivative at any point in time. That's it. Frequency is a derived concept, measured over some extent in time. The cap across the N Fdbk resistor just predictively aligns the N Feedback with the incoming signal time wise, much like a PID controller uses for process control. (time alignment. or phase alignment in frequency parlance) The signal arriving when the N Fdbk gets applied is just the slightly earlier (by nSecs) signal + k times the derivative. k depending on reactive elements in the amplifier + delay time. Of course stability of the amplifier is quite important here too. The derivative can only predict accurately over well less than the internal amplifier time components. (reactances and delay) and the signal variation time. So I don't see any issue with N Fdbk delay as long as the amplifier stays well stable with sufficient bandwidth.
Each recursive loop of error thru the amplifier gets time aligned with the signal + amplifier delay. The remaining error due to insufficient loop gain ( to completely fix the error the 1st time thru) gets progressively fixed better each time around. So the output error should stepwise decrease, asymptotically approaching perfection. The problems arise when the signal is changing faster than the system can fix them. So either more gain (with stability) or more amplifier bandwidth (less loop correction time allowing more recursions) are needed.
Last edited:
The stage where the distortion occurs affects the closed loop results.
Distortion sources closer to theinput output have more distortion reduction.
See figure 5.8 and the related text on page 188, Control System Design by Astrom, here:
Edited for errata at member's request.
Distortion sources closer to the
See figure 5.8 and the related text on page 188, Control System Design by Astrom, here:
Edited for errata at member's request.
Last edited by a moderator:
Oops, correction to my above post #34. A PID controller approaches perfection with more looping thru the controller due to the integrator element ( the I in PID, which stands for Proportional, Integral, Differential) but a typical amplifier lacks this. So the amp only approaches the low distortion allowed by the usual N Fdbk formula for a given loop gain.
I made some simple measurements on a 2 stage self inverting push pull tube amplifier tonight.
This is just a sample of one amplifier.
Your Mileage May Vary.
Measurement 1: Square wave generator rise time is 30 nano second, bandwidth is about 11.7MHz.
Measurement 2: Channel 2 and its probe, is leading Channel 1 and its probe, by 2 nano second.
Measurement 3: The amplifier output lags the input by 3.2 u second at zero crossing.
Correcting for the 2 nano second difference of the channels and their probes:
3.2 u sec - 2 nano sec = 3.198 u sec.
Measurement 4: The amplifier rise time = 5.4 u second (about 64.8 kHz bandwidth).
Rise time is measured from 10% to 90%.
It is not measured from 0% to 50%, or other combinations of %.
Now, suppose the amplifier has distortion at zero crossing.
If you use negative feedback from the output, the feedback is already lagging the input by 3.198 u second. (The input stage where we are applying the negative feedback to correct for the amplifier distortion).
The lag time at zero crossing does not equal the rise time of 5.4 u second.
But the lag does equal 3.198 u second, at zero crossing.
The lag time will be slightly more than 5.4 u second at the 90% amplitude.
(0% to 90% is more than 5.4 u second)
Suppose the amplifier waveform distorts at 90% amplitude; what happens to the negative feedback signal that is already lagging the amplifier input by more than 5.4 u second.
So at what point along the waveform do you want to use negative feedback to correct any waveform distortion?
For high frequency signals, it will be delayed, out of phase, and not ideal (unless you use Lead compensation for the negative feedback signal that goes back to the input stage that also is leading.
Try making the measurements on your own square wave generator; scope channels and probes; and your own amplifier.
Have fun measuring!
Just my opinions (and measurements).
Everything is what it seems to be, until you find out it is different.
This is just a sample of one amplifier.
Your Mileage May Vary.
Measurement 1: Square wave generator rise time is 30 nano second, bandwidth is about 11.7MHz.
Measurement 2: Channel 2 and its probe, is leading Channel 1 and its probe, by 2 nano second.
Measurement 3: The amplifier output lags the input by 3.2 u second at zero crossing.
Correcting for the 2 nano second difference of the channels and their probes:
3.2 u sec - 2 nano sec = 3.198 u sec.
Measurement 4: The amplifier rise time = 5.4 u second (about 64.8 kHz bandwidth).
Rise time is measured from 10% to 90%.
It is not measured from 0% to 50%, or other combinations of %.
Now, suppose the amplifier has distortion at zero crossing.
If you use negative feedback from the output, the feedback is already lagging the input by 3.198 u second. (The input stage where we are applying the negative feedback to correct for the amplifier distortion).
The lag time at zero crossing does not equal the rise time of 5.4 u second.
But the lag does equal 3.198 u second, at zero crossing.
The lag time will be slightly more than 5.4 u second at the 90% amplitude.
(0% to 90% is more than 5.4 u second)
Suppose the amplifier waveform distorts at 90% amplitude; what happens to the negative feedback signal that is already lagging the amplifier input by more than 5.4 u second.
So at what point along the waveform do you want to use negative feedback to correct any waveform distortion?
For high frequency signals, it will be delayed, out of phase, and not ideal (unless you use Lead compensation for the negative feedback signal that goes back to the input stage that also is leading.
Try making the measurements on your own square wave generator; scope channels and probes; and your own amplifier.
Have fun measuring!
Just my opinions (and measurements).
Everything is what it seems to be, until you find out it is different.
Last edited:
I think you just explained in the time domain why negative feedback is not as effective against crossover distortion as against smooth types of distortion.