Difference bewteen ultralinear & triode mode?

[cotdt]

Thanks for posting that data. It makes perfect sense based on how negative feedback works. That said, what rongon says is correct if there is clipping involved.

[DF96]

Quite small amounts of feedback, whether local or global, will create higher order terms. In most cases these will still be lower than the pre-existing higher order terms. People who are worried about higher order terms resulting from feedback should never use small amounts of feedback; use either none or enough to suppress distortion.

Clipping is a separate issue.

[artosalo]

Originally Posted by DF96

Quite small amounts of feedback, whether local or global, will create higher order terms....

...when semiconductors are used, but with vacuum tubes this is not the case.

[DF96]

It is the case. Whether these extra terms are noticeable or not is a separate issue, but they are definitely there - unless somehow tubes manage to avoid using conventional mathematics.

[artosalo]

How this result can be explained if it were impossible ?

[cotdt]

Originally Posted by DF96

Quite small amounts of feedback, whether local or global, will create higher order terms. In most cases these will still be lower than the pre-existing higher order terms. People who are worried about higher order terms resulting from feedback should never use small amounts of feedback; use either none or enough to suppress distortion.

Clipping is a separate issue.

How is this so when the measurements show that all levels of harmonic distortion are decreased as negative feedback starts being added? I don't see the mechanism by which negative feedback can add distortion in the audio band, except in the case of clipping.

[rongon]

There is a short discussion of this in the 4th Edition of M. Jones "Valve Amplifiers." I don't have that here, but I'll look it up this evening, if nobody else has access to it at the moment. The gist of it (as best I can remember) is that when you put gNFB around an amplifier with distortion, the lower order distortions create a tiny amount of higher order harmonics. His general argument is that adding gNFB to an amp with high distortion won't result in a wonderful low-distortion amplifier, but that creating an amp with low distortion open loop and then adding NFB is the way to go.

As usual, I suspect that the audio folklore is created from worst-case scenarios, then erroneously assumed to apply in *all* cases.

That said, a 7 watt per channel amplifier is likely to spend a lot of its time close to clipping. If the odd, higher order harmonics jump up when the amp clips, then I think that would be significant.

Also, 6AG7 is a pentode. Do things work the same with a PP triode output stage?

[DF96]

Originally Posted by artosalo

How this result can be explained if it were impossible ?

I don't understand your question. The graph you show does not contradict what I said. It appears to show significant amounts of high-order already present, either from the valve or from the input signal. Therefore the higher-order stuff produced by the feedback is buried under what is already there. The fact that you can't see it doesn't mean it isn't there! In some cases the FB-induced terms will add to existing terms, sometimes (less frequently) they will be opposite phase and cancel.

Also bear in mind that FFT can sometimes show spurious harmonic terms; it depends on things like the windowing function and floating-point rounding errors.

Originally Posted by cotdt

How is this so when the measurements show that all levels of harmonic distortion are decreased as negative feedback starts being added? I don't see the mechanism by which negative feedback can add distortion in the audio band, except in the case of clipping.

Feedback does two things:
1. create extra terms.
2. reduces everything in the output which is not in the input, including the extra terms it created itself.
You only see 1 when the input signal and the device are very clean (say, 2nd order only) but it is still happening whether you can see it or not.

The mechanism is quite simple: multiplication. An amplifier produces distortion by multiplying the signal by itself, perhaps an infinite number of times. Adding feedback means the whole thing gets squared. Take a 3rd order polynomial (i.e. terms uo to x^3). Square it. You get a 6th order polynomial (up to x^6). Then a bit of trigonometry shows how multiplying sine waves together produces sum and difference products or harmonics.

In order to understand how feedback affect distortion, you first have to understand how distortion arises.

[artosalo]

Originally Posted by DF96

....It appears to show significant amounts of high-order already present, either from the valve or from the input signal. Therefore the higher-order stuff produced by the feedback is buried under what is already there. The fact that you can't see it doesn't mean it isn't there!.....

The harmonic content of my test amplifier at the used power level was actually very small, not significant, since the circuit were fully fine tuned.

If the harmonics were already in the input signal, no improvement due to NFB at later stages would take place.

Negative feedback is typically used to lower the distortion that is generated inside the amplifier, mostly at the output stage.
As my results show, all harmonics are attenuated when NFB is added in practical tube amplifier.
Therefore I hope that such understanding, that small amount of GNFB will increase higher order harmonics in practical tube amplifiers, should be avoided.

The input signal I used with these tests had 3rd and upper harmonics at -95 dBc or lower.

[DF96]

Originally Posted by artosalo

Therefore I hope that such understanding, that small amount of GNFB will increase higher order harmonics in practical tube amplifiers, should be avoided.

That does not conflict with what I said. NFB will create higher-order terms. NFB will reduce distortion. The net result, with almost all active devices, is a reduction in distortion.

I said:

Originally Posted by DF96

People who are worried about higher order terms resulting from feedback should never use small amounts of feedback; use either none or enough to suppress distortion.

This is true. Your graphs show that people need not be worried about higher order terms resulting from feedback. This is also true! My point was that people who worry about feedback should not use a small amount of it, as that will do more of what they want to avoid. They should either use none or lots. Their 'intuition' leads them astray.

People who understand feedback, and so don't worry about it, will use whatever level of feedback is appropriate to the situation.