traderbam said:
Well, you've chosen the operating point to be Vout=Vin=0. At this operating point, the derivative of Vout with respect to Vin of the open-loop device is zero. So small-signal-wise, its gain is zero. That's why I called it a frequency multiplier. But it's not the name that's important. It's the behavior, and whether that behavior is representative of real-world situations.
I'm not sure if y'all were aware of the posts on this subject that went on in the Bob Cordell negative feedback thread. It started with a discussion of Baxandall's paper in Wireless World, where, for a fixed output voltage amplitude, he took the open-loop amplifier to be a square law device (a JFET) and plotted the amplitude of the harmonic distortion components vs. the amount of feedback.
Bob got the idea of duplicating his results with SPICE, and was able to do so, both for Baxandall's JFET and BJT analysis. But then he also posed an interesting question. What happens with crossover distortion of a BJT class AB output stage? Do some values of feedback increase higher-order harmonics in the same way as what occurs with a square law device such as a JFET? It turns out that it doesn't. It reduces them pretty uniformly. So while the JFET case shows some interesting behavior with varying feedback, it's incorrect to generalize the results of that case. Bob's post is number 1278 on this page
These results cause me to conclude that the so-called "SPLIF" concept is without merit.
But, Bob only shows that these high order harmonics also get reduced by feedback, a known fact.
It's about the fact that without closing the feedback loop these higher order harmonics might have been much lower. You can't reduce feedback to zero, you can only cut the loop.
This mechanism creating new harmonics is activated by closing the loop and is independent of the amount of feedback, more feedback simply reduces these harmonics at the same ratio as feedback gets increased. That's exactly what Bobs diagram shows.
Mike
It's about the fact that without closing the feedback loop these higher order harmonics might have been much lower. You can't reduce feedback to zero, you can only cut the loop.
This mechanism creating new harmonics is activated by closing the loop and is independent of the amount of feedback, more feedback simply reduces these harmonics at the same ratio as feedback gets increased. That's exactly what Bobs diagram shows.
Mike
MikeB said:
This has already been demonstrated in this post and the one following. The leftmost data point is the case of no feedback (which was treated separately).
Edit: In this case there is no "creation of new harmonics". The original point was to determine what happens when these higher-order harmonics are already there in significant amounts - such as output stage crossover distortion.
So, for "0db" feedback the loop was opened, and the data connected with lines to the first measuring at feedback = ~14db?
If yes, you at least see that the spectrum changed by closing/opening the loop, by extrapolating the lines to 0db.
Extrapolated: (change of distortion by closing the loop)
2nd: -6db
3rd: -2db
4th: 0db
5th: +2db
6th: +6db
7th: +3db
Your diagram also shows the effect, low order distortions get reduced and high order increased until feedback is large enough and the reduction outweighs the initial increase.
Also, -60db is a quite low open loop distortion, this effect increases heavily with open loop distortion.
Mike
If yes, you at least see that the spectrum changed by closing/opening the loop, by extrapolating the lines to 0db.
Extrapolated: (change of distortion by closing the loop)
2nd: -6db
3rd: -2db
4th: 0db
5th: +2db
6th: +6db
7th: +3db
Your diagram also shows the effect, low order distortions get reduced and high order increased until feedback is large enough and the reduction outweighs the initial increase.
Also, -60db is a quite low open loop distortion, this effect increases heavily with open loop distortion.
Mike
The extrapolated changes you show are not meaningful.
For integrator feedback (which is what the simulation was), the 14 dB feedback factor at 20 kHz corresponds to a 100 kHz unity loop gain frequency. It is already well below what would ever be used in practice. All of this is explained in the post I referenced, so please read it carefully.
My graph shows no increase in higher order distortion at all.
Edit: Oh, and the -60 dB without feedback is not unrealistically low either. Look at the Stereophile graphs of distortion of Charles Hansen's no-feedback amp.
For integrator feedback (which is what the simulation was), the 14 dB feedback factor at 20 kHz corresponds to a 100 kHz unity loop gain frequency. It is already well below what would ever be used in practice. All of this is explained in the post I referenced, so please read it carefully.
My graph shows no increase in higher order distortion at all.
Edit: Oh, and the -60 dB without feedback is not unrealistically low either. Look at the Stereophile graphs of distortion of Charles Hansen's no-feedback amp.
We might be talking 2 different things, i agree completely with your statement "There is no evidence of feedback increasing distortion as it is increased."
What i say, is that closing the loop initially increases high order harmonics until feedback is high enough to compensate.
I think, extrapolating is absolutely valid here, as it shows how the distortions would look at much lower feedback.
Edit: Your graph does show the increase in higher order distortion, you need to look at low feedback levels.
-60db open loop distortion is realistic, but most amps have much much higher.
I am quite a fan of negative feedback, SymAsym uses ~60db of feedback up to ~10khz. I only want to show that using low levels of feedback around high openloop distortion is evil.
Mike
What i say, is that closing the loop initially increases high order harmonics until feedback is high enough to compensate.
I think, extrapolating is absolutely valid here, as it shows how the distortions would look at much lower feedback.
Edit: Your graph does show the increase in higher order distortion, you need to look at low feedback levels.
-60db open loop distortion is realistic, but most amps have much much higher.
I am quite a fan of negative feedback, SymAsym uses ~60db of feedback up to ~10khz. I only want to show that using low levels of feedback around high openloop distortion is evil.
Mike
MikeB said:
Baxandall showed this was true when the open-loop amplifier had a large amount of only second-order distortion. But the result of discussions and simulations in the feedback thread show that this result is often incorrectly generalized. In other words, there may be cases in which generalizing his conclusions for that specific case are just flat wrong.
Regarding extrapolation, surely everyone knows that is problematic. I could, for example, extrapolate last week's stock market data and conclude that I'll be a very rich guy at this time next year 🙂.
I'll try to dig up that old sim and get more detailed data between 0 and 14 dB feedback.
andy_c said:
Yes, extrapolation can be very wrong, but i am quite sure that it continues as a line. It might be difficult to get below 6db feedback.
Mike
A feedback for you Michael.... a positive feedback
You are a very nice person.... an excelent professional and an excelent designer.
The negative feedback is that i feel pitty...so good designer and satisfied with only a very good amplifier....just one?
You can do more....better.... i feel sad you have retired so soon, now only watching, alike adviser, consultant (have competence)....but missing the designer, testing, feeling emotion with the results.
Hey man!..... return to the activity of creation.... give us this gift!
Observe... you have generated many other designs, clearly based over your ideas, and nice ones....let's go Michael!
Man!.... you were very close to produce a simple Class D!
regards,
Carlos
You are a very nice person.... an excelent professional and an excelent designer.
The negative feedback is that i feel pitty...so good designer and satisfied with only a very good amplifier....just one?
You can do more....better.... i feel sad you have retired so soon, now only watching, alike adviser, consultant (have competence)....but missing the designer, testing, feeling emotion with the results.
Hey man!..... return to the activity of creation.... give us this gift!
Observe... you have generated many other designs, clearly based over your ideas, and nice ones....let's go Michael!
Man!.... you were very close to produce a simple Class D!
regards,
Carlos
andy_c said:
Hi Andy, i stand corrected... Obviously never too old to learn new stuff.
I checked feedback behavior with different transfer functions, this time i used an exp() function (as opposed by bjts) which has a rich harmonic spectrum. Even with only 6db of feedback, feedback reduced high order distortions.
But... exp() is special... feedback transforms exp(x) to some exp(-x). Maybe feedback'd transfer function is always some exp(), giving rich harmonic content? (exp(-x) = 1/exp(x)) My mathematical skills end here...
Carlos, maybe i will find again the mood to design new amps...
Mike
MikeB said:
Thanks for checking that, Mike. I started to re-do my previous analysis with the integrator, then realized that for very low amounts of feedback, the bandwidth was very poor. This made it a real hassle to figure out what the input level should be to get the desired output - so I set it aside.
It's interesting that when the distortion spectrum of the open-loop amplifier is already rich in harmonics, things are different from the simple square law case vis-a-vis distortion vs. feedback. In the case of a class AB output stage with optimum bias, it is rich in harmonics, yet the levels of distortion aren't too bad with the optimum bias. Then feedback does a surprisingly good job even when there isn't much of it. That's why I'm skeptical of the "SPLIF" topology. This finding surprised me as well when it was first discussed in the Cordell feedback thread a while back.
MikeB said:
exp(x) = 1 + x +x^2/2 + x^3/3! + x^4/4! + x^5/5! + x^6/6! +..... ad infinitum.
Where n! = 1 * 2 * 3 * 4 *....* n
Example 6! = 1 * 2 *3 * 4 * 5 *6 = 720
! is called the "factorial" symbol
So the exponential function really has all orders of non-linearity at once!
You should see that the spectrum changes depending on the amplitude of the signal. As you increase the amplitude of the input the harmonics will increase in amplitude relative to the fundamental.
If I'm not mistaken, in a real transistor the base current goes as the logarithm of the voltage? 0.7V is just the "knee."
fizzard said:
Yes, that's why i chose the exp() function, it's taylor series show that all orders of harmonics are present...
Andy, i think that the crossover distortion might be the problem, as it should introduce dynamic phase shifts into the feedback loop.
Feedback itself seems to not produce real problems...
Mike
Andy & Mike,
Just to clarify, where are you now on whether NFB introduces spectral content that wasn't in the OL output?
Brian
Just to clarify, where are you now on whether NFB introduces spectral content that wasn't in the OL output?
Brian
MikeB said:
Exactly, it changes the transfer function, ergo the OLG module and phase, dependant on output amplitude. But it is not the only culprit - in output stages beta vs Ic not being constant also changes OLG. Introduce a speaker as a load instead of a perfect resistor, and you have a nonlinear, memory capable (mechanical hysteresis + acoustic delay line effects), resonant system, all of which show themselves as dynamic changes in impedance, so again affect OLG. Of course, how much which effect contributes depends on the actual topology and active element technology used.
What I am trying to say that more frequently than not these sorts of discussions are reduced to analysis of simplified models, which certainly tell a lot of the story, but like all 'simulations' only that part you are speciffically looking for. In a real amplifier there may be others.
I would also submit that discussing feedback is really a misnomer - thing always seem to work just fine where the OLG curve is flat and has negligible phase shift and feedback is constant.
The 'interesting' stuff happens when that's not so, so we are really discussing differences in OLG. I concur with MikeB that it's not about feedback itself being 'bad' (come on, how bad can at most two resistors and maybe a cap be compared to dozens of components making up the circuit that actually produces OLG!) but the character of the OLG whose excess it's using.
rdf said:
What are you talking about? I’m not arguing any such thing and I did not “dismiss” any technology or topology – SET design included.
A typical power triode pushing out it’s rated power into a loudspeaker via an impedance matching transformer in a typical circuit without GNFB will generate copious amounts of audible distortion which some individuals have a preference for – this says nothing for the intrinsic utility or ‘musicality’, 'goodness' or 'badness' of NFB - just for some peoples preferences WRT audible coloration.
I have a personal preference against such designs. Is that difficult to understand?
traderbam said:
Just for starters:
So an audio power amplifier which needs GNFB to establish it's DC operating point and set it's gain (thus relying on an intrinsic utility of NFB) is "ill-designed" in your view?
So are there actually any SET power amps having no GNFB, and that have very low distortion? This is a serious question - I'm completely ignorant of the vacuum tube world.
rdf said:
Thank you. If you could point out any contradictions between those pretty straightforward paragraphs that would be great.
It should be obvious now that SET design is one area where I think the spurning of global NFB has some degree of legitimacy, given the potential design aims.
andy_c said:
I guess that would depend on what you mean by "very low".
Your run-of-the-mill SET power amp may be pushing 10% THD or more a its rated power output, but you could conceivably parallel up some big power tubes and run them with high plate impedance to get the THD down (ie, lots of dissipation for very little output). Ideally, you would want tubes intended for not-too-high voltage operation at high currents, as there is a practical limitation to how high you can go with the impedance presented to the plate before the design of the impedance matching transformer becomes a headache.
Cheers,
Glen
- Status
- Not open for further replies.