• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

Feedback and distortion in Single Ended amplifiers

Mark said I would guess that at least of a portion of what happens on the secondary OPT side will be reflected on the primary side. As such the OPT is not outside the loop completely.
Good point. I think his main point is that you avoid the output transformer phase shift which makes it easier to apply a hefty feedback factor.
I send your paper to Menno and he tells me he is impressed 😎

Jan
Norman Crowhurst in the first article on his 'Twin Coupled Amp' makes the same case. However he does point out that only applies to the lower frequencies, leakage reactance in the OPT gets in the way. Putting the HF thru the leakage reactance is like pushing a rope up a hill.
 
I think you may be very right about some things, but I also think the HD is a bit of a red herring...

I've been hunting down research papers (like the one attached, and a couple of others if I can find them) to get to the bottom of things.
As can be seen, controlling the current tends to provide more linear results when measuring and validating the end result with a microphone.
Due to the traditionally high output resistance of tube amps, it seems the designers have been doing it right all along, but the higher distortion was harder to explain away, so "nice sounding distortion" became a myth that just won't die.

The same mechanisms that create HD unavoidably also create IMD when they have the opportunity, so to me the idea of the sound being somehow 'sweetened' with special effects makes no sense at all.

What's really missing in the equation is a 3rd factor: the speaker load.
I think it's a really bad habit for some people to attempt to treat an audio system as a collection of "independent components". They are not independent. Unless careful steps are taken to deliberately avoid it, the speaker almost invariably changes the amplifier's purported internal THD rating. Even a vastly over-simplified LCR circuit for the speaker can blow up an amplifier's THD by a factor of 10 or more in simulations. And that's with no speaker distortion, only a phase shift that sabotages the otherwise ideal performance of the negative feedback. And that's where I think you've got it spot on by avoiding unbuffered negative feedback on the output stage. In real life it's even worse, because it's not just benign phase shifts, but also non-linear inductance appearing in series, mechanical reverb, and chaotic glitches from the permanent magnet.

On a related note....
Step response graphs ring alarm bells for me: a steady voltage is applied, and what happens then? The cone traverses a bizarre scenic route, usually oscillating back and forth many times before settling. What seems to happen is the bass is fairly close to 'critically' damped, while at a much higher frequency, the voice coil -- in spite of a high damping factor -- is mysteriously unable to absorb the reverberating energy from the cone. All it takes is that the reverberations shift the mechanical phase 180 degrees, and negative feedback becomes positive feedback, and actually reduces stability, making cone break-ups decay longer than they need to.
Hi Abstract,

I fully agree that the output stages' OPT, power supply and complex speaker load are critical for how this will sound. Your point about the simulation therefore having limitations is valid, and I pointed that out throughout the document. I do believe its good enough to demonstrate the effect on the complex noise floor exists (but I deliberately didn't quantify it because of the above). Naturally, I still need to build the amp and validate the effect in real life, with the actual power supply, a real OPT and indeed a complex load. I will! Those measurements however will also have limitations, as they will only be proven for the OPT I'm using, my specific power supply and my specific speakers.

Hope that makes sense,
Mark
 
That was also L. V. Viddeleer's argument for using feedback from the transformer primary in the 1950's, see pages 449 ... 457 of the October 1954 issue of Radio Electronica,

http://nvhrbiblio.nl/biblio/tijdschrift/Radio Electronica/1954/Radio Electronica 1954-10-OCR.pdf

Viddeleer discussed his tone control more thoroughly in the August 1953 issue, pages 29 ... 32 and 64, see http://nvhrbiblio.nl/biblio/tijdschrift/Radio Electronica/1953/Radio Electronica 1953-08-OCR.pdf and in the September 1953 issue, pages 19 ... 22 and 38, see http://nvhrbiblio.nl/biblio/tijdschrift/Radio Electronica/1953/Radio Electronica 1953-09-OCR.pdf
Tricky for our international readers as it's in Dutch, but amazing these have been digitized and preserved. Very nice and thanks for sharing! So tapping from the primary has been done before. Could be interesting to see if this type of amp has a similar impact on the noise floor...
 
Norman Crowhurst in the first article on his 'Twin Coupled Amp' makes the same case. However he does point out that only applies to the lower frequencies, leakage reactance in the OPT gets in the way. Putting the HF thru the leakage reactance is like pushing a rope up a hill.
Jhstewart9, do you have a link to this article or a downloadable copy?

Edit - found it ...

Jan
 
Hi Abstract,

I fully agree that the output stages' OPT, power supply and complex speaker load are critical for how this will sound. Your point about the simulation therefore having limitations is valid, and I pointed that out throughout the document. I do believe its good enough to demonstrate the effect on the complex noise floor exists (but I deliberately didn't quantify it because of the above). Naturally, I still need to build the amp and validate the effect in real life, with the actual power supply, a real OPT and indeed a complex load. I will! Those measurements however will also have limitations, as they will only be proven for the OPT I'm using, my specific power supply and my specific speakers.

Hope that makes sense,
Mark
Definitely. Most of my rant was more "shouting at the world" and not aimed at your work specifically. The numbers look very impressive and I'm looking forward to finding out how it proves itself. I have a collection of projects in various stages of starting, stalling and disrepair, and only occasionally finishing. And whenever I learn something compelling, there's often a strong urge to share the word.

There is also some good info on "jump distortion", where resonances in the frequency domain are tilted, and an interesting counter argument against too-high output resistance, at least/especially for edge cases like a speaker's bass resonance.

I have not heard this distortion before (that I'm aware of), but I imagine it would occasionally manifest as bright rattles or buzzing effects affecting specific tones, like a music box with a loose mechanical connection.

Lech
 

Attachments

Last edited:
I now repeated some simulations using a mu-follower instead of an SRPP for the 12AX7 stage and landed on very similar THD and IM products after some optimization. There was a positive effect on the phase shift, which due to increased bandwidth lowered by 1.5 degree at 20 kHz. All in all... I'll stick to the design documented already. The added complexity doesn't seem worth it.

Cheers,
Mark
 
Given that the paper is in it's foundation a subjective exercise, I think this should be clearly stated early in the paper to make sure no one starts to comment that a conveyor of information (e.g. an amp) is only perfect when distorsion is zero - anything else and it becomes a part of the "orchestra" that wasn't there to begin with (production and not reproduction). It should also be noted that it is impossible to "listen" solely to an amp as it takes a complete system to hear something. The last aspect, if accepted, leads to that all components in the system needs a similar analysis and the net result, the system output, need to be understood as well and then, and only then, might one dare to say something about and individual contribution of sound of a component.

//
 
Given that the paper is in it's foundation a subjective exercise, I think this should be clearly stated early in the paper to make sure no one starts to comment that a conveyor of information (e.g. an amp) is only perfect when distorsion is zero - anything else and it becomes a part of the "orchestra" that wasn't there to begin with (production and not reproduction). It should also be noted that it is impossible to "listen" solely to an amp as it takes a complete system to hear something. The last aspect, if accepted, leads to that all components in the system needs a similar analysis and the net result, the system output, need to be understood as well and then, and only then, might one dare to say something about and individual contribution of sound of a component.

//
By that logic also the recording process, the room, the passage listened to and the listener should be taken into account. That would in fact render all analysis of parts in the change impossible. Is that your point?
 
Pos feedback? Don't you mean negative feedback, that normally effects lower Zout because it makes the Vout less dependent on load.

Jan
No I do not mean that. In the example given in post #13, the feedback is derived from the output current. If it was applied as negative feedback it would increase Zout but as it is applied as positive feedback it reduces Zout. The negative feedback that you are referring to that reduces Zout has to be derived from the output voltage in order to reduce Zout.

Cheers

Ian
 
Correct except for the listener which is not effected for the same reason that the orchestra is not included.

//
The last portion is only true if you evaluate taking a specific listener in mind, again... making it subjective. The hearing system as well as preferences divert between listeners. I would argue that, if from this perspective, all evaluations are subjective, then why would I need to state that?
No I do not mean that. In the example given in post #13, the feedback is derived from the output current. If it was applied as negative feedback it would increase Zout but as it is applied as positive feedback it reduces Zout. The negative feedback that you are referring to that reduces Zout has to be derived from the output voltage in order to reduce Zout.

Cheers

Ian
I didn't understand that also. Thanks for the clarification.

Cheers,
Mark
 
No I do not mean that. In the example given in post #13, the feedback is derived from the output current. If it was applied as negative feedback it would increase Zout but as it is applied as positive feedback it reduces Zout. The negative feedback that you are referring to that reduces Zout has to be derived from the output voltage in order to reduce Zout.

Cheers

Ian
I think there is some confusion here. You can feedback output voltage or output current. In either case, you can have positive feedback or negative feedback.
Negative feedback makes the quantity that you feed back more independent from other factors, so negative voltage feedback makes Vout more constant with varying loads, so lowers Zout.
Negative current feedback makes the output current more independent of any load, it tends to keep de output current constant with varying loads, so increases Zout.

If you have positive feedback, voltage or current, your system is unstable and your amp starts to oscillate or slam into a rail.

You can have a combination of negative and positive feedback and it can be stable as long as the negative feedback dominates. This has been done long ago in tube amps.
The positive feedback increases the gain, and when unchecked would make it unstable, but the negative feedback kept it under control and could use the additional gain from the positive feedback to add to the negative feedback factor to lower gain.
Like here.

Or here; and I quote:

Feedback in the control system is defined as a way of providing output (or a part of the output) back to the input. The signal can be either current or voltage depending on the operation

Negative feedback also referred as degenerative feedback is a widely used type of feedback in the control system. Here the signal at the output which is out of phase with respect to the input is fed back to the input. Thus, the two signals at the input of the system get subtracted and the difference of these two signals further drives the system.

Positive feedback or regenerative feedback is the one that takes the output signal which is in phase with the applied input and fed it back to the reference input. This facilitates adding the feedback signal with the reference input and the added signal further acts as the controlling signal for the system in which the feedback loop is incorporated.


Jan
 
There is no confusion. When you want to realize by feedback an output impedance that neither tends to zero nor to infinity with increasing loop gain, you can use combinations of voltage-to-voltage and voltage-to-current feedback. When you give one of them the opposite polarity, you can indeed realize negative impedances. You can get an oscillator when the positive feedback is too strong and becomes dominant, but that's a matter of incorrect design then.

The same holds for input impedances. You can realize input impedances that neither tend to zero nor to infinity by combining voltage-to-something and current-to-something feedback at the input. An example is the electrically cold resistance/synthesized loading/whatever you want to call it technique that is often used in LNAs of receivers and sometimes in phono amplifiers. That dates back to 1939, as far as I know.
 
If you have positive feedback, voltage or current, your system is unstable and your amp starts to oscillate or slam into a rail.

If positive Fdbk is used, the system does not immediately become unstable.

For the voltage case one can increase the gain of a stage.

For the current derived case, one can produce a negative impedance (to sum with some hopefully greater positive, conventional, impedance.)

In the case of the Patent scheme, positive Fdbk is used to generate enough neg. impedance to cancel the OT primary resistance, Essentially, it feeds back enough extra drive V to the driver stage to produce the extra output voltage that the sensed current would drop (V wise) across the OT winding resistance. Leaving just the original desired signal to pass on to the secondary. Since magnetizing current is also sensed, sufficient extra drive is also developed to overcome the primary V loss that magn. current would ordinarily cause in the winding resistance. Effectively removing the magnetizing current effects. (well there is still the tube Ra, which can be cancelled too with a little more Pos. Fdbk.)

Since the OT winding resistance is removed, the output impedance will fall, and removing the tube Ra will lower output Z even further. If one were to try to do more Neg. impedance to fix the OT secondary resistance too, it would unfortunately introduce inverted magnetizing current drop, since that dist. does not appear across the secondary resistance to cancel out with. Same for speaker R. Although those can be dealt with a separate Pos Fdbk loop at the secondary side (usually back to the 1st stage). Too much neg. resistance however and you get an oscillator (when net impedance goes negative). Tube Ra changes some with current (so only the minimum Ra can be cancelled at most). OT resistance is fairly stable. Speaker resistance is all over the place. Practically, one would adjust the pos. current Fdbk for lowest distortion and lowest Zout without introducing instability under any conditions.

Note: the Hawksford type error correction circuit (usually for SS use) uses an inherent positive V Fdbk loop to produce (virtual) high gain for correcting errors. It was originally patented by Llewellyn for tube circuits in a specific form way back. Virtually unheard of in the tube world though.
 
Last edited:
Pls post an example of such an audio amp that you have built with some measurement data. Looks interesting.

The only example I have with valves doesn't exactly match the description, because it uses the mu of triode V1 to set the voltage-to-voltage gain. Then again, it's really a triode-connected pentode, so you could say it uses screen grid feedback. It's the left part of the attached schematic, see Linear Audio volume 4 for all details. R5A and R5B provide the current-to-voltage feedback.

There are two clearer examples with transistors and op-amps in Electronics World October 2003, see https://worldradiohistory.com/UK/Wireless-World/00s/Electronics-World-2003-10-S-OCR.pdf pages 38 ... 43.

All three are amplifiers for moving-magnet cartridges that make the +47 kohm input resistance with feedback techniques to reduce noise. For the attached circuit, the RIAA- and A-weighted averages of the equivalent input noise voltage and current densities were 10.26 nV/√Hz and 0.2534 pA/√Hz, with a more conventional design using a normal 47 kohm termination and a triode-connected EF86 they would have been around 10.26 nV/√Hz and 0.5869 pA/√Hz. The lower current noise has its largest effect for high-inductance cartridges, but even then it tends to be masked by record surface noise whenever there is a record playing. So all in all, the noise in between records is a bit lower than with the usual 47 kohm termination.

The neon lamps are meant as voltage limiting devices. For them to respond quickly to overvoltage, there has to be some light shining on them. As I wasn't aware of that when I built it, there is no light shining on them in my build. Nonetheless, it still works fine after ten years.

Figure3_tekstgroot.jpg
 
Last edited: