Surely it's as you have stated in the last phrase. Effectively NFB has historically suffered of many bad "synchronism" with eletronic tecnologies available in different moments. Before solid state era problems associated with existing output transformers surely have not contribute to build a good opinion of NFB because of stability issues involved... and after tube era happen the same thing but triggered by different causes: once set free from trasformers, designers thrown itself on deep diving in heavy NFB.... with not exactly enthusiastical result.
In my opinion exist again nowadays some lacks of insight about the true nature of audio amplifier (preamplifier or power don't count much, altrhough, of course, power amplifiers cope with heavily worsened handicaps than preamplifiers): these are WIDEBAND AMPLIFIERS that covers well THREE decades of frequency. This, in my opinion, is the only serious flaw of the so called "op-amp" structure: that his intimate nature, as originally conceived, is really those of a NARROW BAND amplifier where changes in voltage gain due to compensation exigences are not really important as in wideband ampliiers, where open loop falling with frequency reveal ineluctably the intimate nature of high feedback, heavily compensated amplifier, which, pheraps, if originally should be regardes as linear amplifiers, they were turned out into a first order integrator by cdom - what need to be linearized by the same negative feedback that introduced this before INEXISTENT problem.
Actually, as designers strive for reaching in their amplifiers stunning performances yet in open loop state, the true need of heavy NFB loop gain stem uniquely from the exigence of avoiding (for purely economical reasons) stabilized power supplies and substitute them with an amplifier which, at some extent, stabilize "by itself" (or better: PROTECTS by itself) its own power rails. In low end commercial amplifiers this economical movent may hold again. But otherwise, in the so called (high priced) "hi-end amplifiers" should be existing enough "budget room" for investing in a good, regulated power supply that avoid the need of heavy NFB loop gain which conduct, unavoidably, to an heavy compensated amplifier, i.e. the most unsuitable for coping with "wideband" exigences of audio amplifiers.
I agree!
Even from my viewpoint the best amplifier are REALLY good "no-feedback" design with a reasonable amount of NFB that "flatten out" the residual amount of non-linearity - Here for "reasonable amount" I intend comprised between 40-50 dB; a Baxandall graph that I had the opportunity of view, show clearly that so called "low-feedback" design with loop gain below 35-40 dB really WORSEN the distortion behaviour - especially from spectral distribution viewpoint - if confronted with open loop condition. A quirk from which can be derived ad interesting consideration: once we decide to use NFB, we can maximize its advantage ONLY maximizing its entity, with the only limits of not boost its value until a level where other problems rise out above.Happy new year to you and all!
Piercarlo
Last edited:
Bode's "Optimal Feedback" compensation easily gives 60 dB loop gain @ 20 KHz with < 2 MHz loop intercept - not a problem with Mosfet or BJT with Ft >20 MHz with "unconditional stability"
additionaly, there is good reason to ignore the tube era definition of unconditional stability with semi's that can achive full BW operating points within small fractions of the period of potential oscillations for conditionally stable compensations
to learn lots more about how to handle really high loop gains check out BJ Lurie:
(oops his site seems to be offline today, here's a page from archive.org)
Classical Feedback Control, Chapter 4: Shaping the Loop Frequency Response
note that the loop gain is flat in the "working frequency" range
http://www.amazon.com/Classical-Feedback-Control-MATLAB-Engineering/dp/0824703707
http://www.ieeecss.org/columns/Feb2007/PeopleFeb07.pdf
Mitchell also has some good stuff on using higher levels of feedback
http://www.personal.reading.ac.uk/~shsmchlr/selected.htm
Last edited:
Surely it's as you have stated in the last phrase. Effectively NFB has historically suffered of many bad "synchronism" with eletronic tecnologies available in different moments. Before solid state era problems associated with existing output transformers surely have not contribute to build a good opinion of NFB because of stability issues involved... and after tube era happen the same thing but triggered by different causes: once set free from trasformers, designers thrown itself on deep diving in heavy NFB.... with not exactly enthusiastical result.
In my opinion exist again nowadays some lacks of insight about the true nature of audio amplifier (preamplifier or power don't count much, altrhough, of course, power amplifiers cope with heavily worsened handicaps than preamplifiers): these are WIDEBAND AMPLIFIERS that covers well THREE decades of frequency. This, in my opinion, is the only serious flaw of the so called "op-amp" structure: that his intimate nature, as originally conceived, is really those of a NARROW BAND amplifier where changes in voltage gain due to compensation exigences are not really important as in wideband ampliiers, where open loop falling with frequency reveal ineluctably the intimate nature of high feedback, heavily compensated amplifier, which, pheraps, if originally should be regardes as linear amplifiers, they were turned out into a first order integrator by cdom - what need to be linearized by the same negative feedback that introduced this before INEXISTENT problem.
Actually, as designers strive for reaching in their amplifiers stunning performances yet in open loop state, the true need of heavy NFB loop gain stem uniquely from the exigence of avoiding (for purely economical reasons) stabilized power supplies and substitute them with an amplifier which, at some extent, stabilize "by itself" (or better: PROTECTS by itself) its own power rails. In low end commercial amplifiers this economical movent may hold again. But otherwise, in the so called (high priced) "hi-end amplifiers" should be existing enough "budget room" for investing in a good, regulated power supply that avoid the need of heavy NFB loop gain which conduct, unavoidably, to an heavy compensated amplifier, i.e. the most unsuitable for coping with "wideband" exigences of audio amplifiers.
I agree!
Happy new year to you and all!
Piercarlo
The Baxandall distortion tests are very interesting stuff. I have duplicated his results and studied them further in simulation. It is important to realize that even the feedback inherent in emitter degeneration counts the same way as more global feedback in creating some newer harmonics. Most people don't realize this and think emitter degeneration is more benign. I've posted curves showing this in another thread. The good news is that the presence of local degeneration contributes to overcoming the first 15-20 dB of where NFB can in principle create new components. Thus, a stage with 20 dB of emitter degeneration will do just fine even if only another 10 dB of NFB is put around it in a more global fashion.
The Baxandall results were correct, but they were confined to a single stage with only a second-order nonlinearity. With BJTs and multiple stages, higher order nonlinearities exist from the get-go (not to mention crossover distortion). It is unfortunate that some have mis-understood the Baxandall results and latched onto them as a way of besmirching negative feedback.
Happy New Year,
Bob
That one is in my bookmarks
. It is here.Also, it should be noted that Baxandall's findings for the case of an open-loop amplifier with only second-order distortion assumed the distortion of the open-loop amplifier was 10 percent. This is quite pessimistic. Let's suppose this 10 percent value is the distortion of the open-loop amplifier and that we've computed the distortion components of the closed-loop amplifier for a given amount of feedback. Now suppose we reduce the open-loop distortion by a factor of K and keep the amount of feedback the same. What happens is that, relative to the case where the open-loop distortion is the original value of 10 percent, the closed loop distortion percentage will be reduced by a factor of K for the second harmonic, K2 for the third harmonic, K3 for the fourth harmonic, K4 for the fifth harmonic, and so on, generalizing to KN-1 for the Nth harmonic. Thus by decreasing the open-loop second-harmonic distortion, we get a disproportionate improvement in the closed-loop distortion components for harmonics higher than 2. This shows that reducing open-loop distortion in this system (whose nonlinearity is second-order only) has a very important effect on the amount of high-order distortion of the closed-loop amplifier.
This was summarized and verified in simulation at posts starting here.
Lumba, what is your goal for an ideal amp?
Eva makes an interesting point about speaker design. Since they are inherently far from perfect (current speakers at least), a designer who can control every aspect of his design still has to decide what tradeoffs to make (impulse response, flatness of frequency response, on-axis time domain coordination versus off-axis dispersion, etc). You really can't have it all with a speaker, so every speaker will have some coloration or some behavior inherent to it.
Obviously (at least to me) the SET guys aren't aiming for an amp with zero distortion or perfectly flat frequency response. They are after something (IMO, almost definitely ADDED by the amp), that's not to be found in precise transfer of the input signal. Since you don't believe in measurements, and I'm not sure you trust specifications either, I'm wondering if a lot of the argument here is about conflicting goals. In short, would you rather have an amplifier that perfectly amplified an input signal, or one that measured worse but sounded better to you because of something it was doing to that signal?
Eva makes an interesting point about speaker design. Since they are inherently far from perfect (current speakers at least), a designer who can control every aspect of his design still has to decide what tradeoffs to make (impulse response, flatness of frequency response, on-axis time domain coordination versus off-axis dispersion, etc). You really can't have it all with a speaker, so every speaker will have some coloration or some behavior inherent to it.
Obviously (at least to me) the SET guys aren't aiming for an amp with zero distortion or perfectly flat frequency response. They are after something (IMO, almost definitely ADDED by the amp), that's not to be found in precise transfer of the input signal. Since you don't believe in measurements, and I'm not sure you trust specifications either, I'm wondering if a lot of the argument here is about conflicting goals. In short, would you rather have an amplifier that perfectly amplified an input signal, or one that measured worse but sounded better to you because of something it was doing to that signal?
That one is in my bookmarks
Also, it should be noted that Baxandall's findings for the case of an open-loop amplifier with only second-order distortion assumed the distortion of the open-loop amplifier was 10 percent. This is quite pessimistic. Let's suppose this 10 percent value is the distortion of the open-loop amplifier and that we've computed the distortion components of the closed-loop amplifier for a given amount of feedback. Now suppose we reduce the open-loop distortion by a factor of K and keep the amount of feedback the same. What happens is that, relative to the case where the open-loop distortion is the original value of 10 percent, the closed loop distortion percentage will be reduced by a factor of K for the second harmonic, K2 for the third harmonic, K3 for the fourth harmonic, K4 for the fifth harmonic, and so on, generalizing to KN-1 for the Nth harmonic. Thus by decreasing the open-loop second-harmonic distortion, we get a disproportionate improvement in the closed-loop distortion components for harmonics higher than 2. This shows that reducing open-loop distortion in this system (whose nonlinearity is second-order only) has a very important effect on the amount of high-order distortion of the closed-loop amplifier.
This was summarized and verified in simulation at posts starting here.
Hi Andy,
These are very good points. Thanks for posting those bookmarks - I had forgotten where they were!
Happy New Year!
Bob
The 3904/3906 transistors (in whatever package) are actually pretty good SS transistors. They are used by the millions, er billions, thus they are nearly as cheap as the sand they are made from.Hello Bob
The Blameless amp with a single pair of Thermal trak Onsemi devices used in an emitter follower configuration achieves THD-20 of 0.02% at 100W into 8R . This is based on actual measurements.
Regards
Arthur
Thanks, Arthur. That is more like it. Those OnSemi ThermalTrak devices are probably an order of magnitude faster than the old MJ802. With output devices sporting ft of over 30 MHz, it is much easier to believe that Self achieved 0.02% out to 20 kHz. I suspect he was inspired to try the ThermalTrak devices by the discussions on this very Forum quite awhile back.
Cheers,
Bob
