See the link I posted previously. The point is that if the open-loop distortion is low enough, and for even moderate amounts of feedback for the square-law device, the feedback-generated distortion components are lower in amplitude than what one will see for any real-world open-loop amplifier that allegedly lacks these components to begin with.
But this "open-loop distortion is low enough" requirement is not met by SET power amps, so one might be better off without feedback in that case.
Stated another way, for the square-law open-loop case that Baxandall looked at, one must have high open-loop distortion for the feedback-generated higher-order products to exceed what one would see in a real-world open-loop amplifier.
Last edited:
I don't have Baxandall's article in front of me. My understanding is that he analyzed a single-tone with a second order non-linearity. So if you are stating that the novel distortion products arising in a 2nd order non-linear system excited by a single tone with NFB are negligible when the OL distortion is moderate by current practical standards then I have no reason to disagree.
It´s a valid point IMO, but, a flat response until 2 or 3 octaves over 20kHz, with low distortion behaviour, must be enough, IMO too. I think it can be achieved using proper design and good,,but no soo "fast" devices at output.
regards
Last edited:
Baxandall's article can be found at Jan's site here. It is part 5 of the article (page 21 of the PDF). The BJT stuff is in part 6. Sounds like we are at least somewhat in agreement. Maybe it would be better to pick this up in the Cordell feedback thread? This is becoming OT to the original post about output devices.
This is not ever true. Many classic low Ft bipolar devices have lower internal capacitances than modern faster devices. Regrettably capacitances, Ft and speed are not so straightforwardly related among them. Judging internal behaviour of seminconductors on the assumptions that their equivalent circuits (with resistor, capacitors, dependent generators and so on) are "true" circuits may be heavily misleading.
The only thing we can safely assume about higher Ft devices is that they can still operate at frequencies where other devices are become just "dead loads" appended to the drivers. But if this assure better linearity and stability or not is a matter that should be evaluated only on the field. Mosfet are surely better provided than bipolars in high frequency features but this usually worsen stability problems instead of alleviating them (at least in physical circuits, where layout issues should be taken in account).
Hi
Piercarlo
Member
Joined 2009
Paid Member
How good does the open-loop linearity have to be in order for low levels of NFB to be 'safe' ?
Baxadall left me with the perception that low to medium levels of NFB are the worse kind for harmonic generation because the feedback is enough to increase higher order harmonics relative to the lower order harmonics but is not enough to lower overall THD to unobjectionable levels. I know this is oversimplifying but anyhow....
It would be a useful guideline to know how good the open-loop linearity should be - and I note you mention for square-law so I assume we are talking about a level of H2 distortion that would be 'acceptable', a level of H2 that when NFB is added generates a H3, H4 etc. at very low levels that we wouldn't find objectionable.
I might apply this guideline to my next design.
Member
Joined 2009
Paid Member
Please spill the beans and let us have a recommendation - especially if you have also tried it and listened to it ?
I'm still planning to use an all-BJT output in my next project, but perhaps I should consider alternatives if there is a good reason.
I'm not sure there's a standard answer to this.
A plot identical to Baxandall's for 10 percent open-loop distortion (second-order only) is shown here.
The one for 1 percent open-loop distortion (second-order only, not shown by Baxandall) is shown here.
And for 0.1 percent open-loop distortion (second-order only, also not shown by Baxandall), the plot is shown here.
For 0.1 percent open-loop distortion (second-order only), any feedback level could be considered "safe" (all distortion components 3rd-order and higher < -120dB).
Member
Joined 2009
Paid Member
Thanks Andy - interesting plots.
I've read that the one to worry about is the 5th, and this seems well controlled even at 1%H2 open loop. The 10%H2 open loop looks quite worrying though.
I need to check my math here - how many dB (I'm thinking Spice simulation now) of H2 is 1%...
-20 log (0.01) = -40dB ?
I've read that the one to worry about is the 5th, and this seems well controlled even at 1%H2 open loop. The 10%H2 open loop looks quite worrying though.
I need to check my math here - how many dB (I'm thinking Spice simulation now) of H2 is 1%...
-20 log (0.01) = -40dB ?
Exactly right. It's -20*log10(percent/100). Or you can just look at the H2 value at the far left of the plot (no feedback).
Member
Joined 2009
Paid Member
I can't help thinking we've got this thread cross-threaded with another thread 
Anyhow, for the first time I feel there is some science coming through the fog on this feedback subject - in that I am looking to reduce all the chit chat and complex theory to something simple that I can grasp. That's how science helps the common idiot like me.
So, feedback done badly is bad, done well it is excellent. In my simple thinking I divide the Baxandall result into 3 regimes:
1) Zero / Low global NFB: many good amplifiers in this category. It's now all about good open-loop linearity only, distortion will be what it is. It's an approach best for Class A since there's no NFB to take care of any cross-over distortion. Damping factors are limited so you have to match amp with speakers. Clipping behaviour is generally good. Choice of devices, topology and biassing will tell all; and it's easiest with a simple topology. Result is usually very good HF but depending on match with speaker can have varying bass performance. Can have onerous consequences for power supply design.
2) Medium global NFB: this is the danger zone where many amplifiers find themselves. Distortion products are typically slanted towards higher order harmonics and this is accepted as bad, plus there isn't enough NFB to push the higher orders out of earshot. To fix this you can purposely add low order harmonics so that the profile is at least one of decreasing magnitude as harmonics increase and rely on psycho-acoustics; this is a contentious solution. Another option is to avoid complex circuits with elements that have exponential transfer curves, try and stick with Triodes/JFETs so that it's mostly quadratic and have an open-loop distortion performance such that H2 is lower than 1% - quite a severe restriction that most middle of the road amplifiers fail to meet. Achieves reasonable damping factors. Clipping behaviour is degraded. Compromise between 1) and 3) with good bass and good HF ?
3) High global NFB: also a popular category, especially where marketing departments live. It pushes all distortion below audible limits, can use any kind of amplifying element including exponential transfer curves. BUT this approach places a huge burden on the designer to ensure the NFB is very accurate because all the devil's come out of the wood work if it isn't and usually hit the HF performance. Includes having to pay careful and particular attention to the error amplifier and gain-bandwidth so that there is enough gain at high freq plus minimal phase shift. This will be hard to achieve. Damping factor very high, clipping behaviour can be bad. PSRR is fantastic, takes pressure of psu design.
Understanding how accurate the NFB has to be to avoid creating problems for this last option would be useful.
Anyhow, for the first time I feel there is some science coming through the fog on this feedback subject - in that I am looking to reduce all the chit chat and complex theory to something simple that I can grasp. That's how science helps the common idiot like me.
So, feedback done badly is bad, done well it is excellent. In my simple thinking I divide the Baxandall result into 3 regimes:
1) Zero / Low global NFB: many good amplifiers in this category. It's now all about good open-loop linearity only, distortion will be what it is. It's an approach best for Class A since there's no NFB to take care of any cross-over distortion. Damping factors are limited so you have to match amp with speakers. Clipping behaviour is generally good. Choice of devices, topology and biassing will tell all; and it's easiest with a simple topology. Result is usually very good HF but depending on match with speaker can have varying bass performance. Can have onerous consequences for power supply design.
2) Medium global NFB: this is the danger zone where many amplifiers find themselves. Distortion products are typically slanted towards higher order harmonics and this is accepted as bad, plus there isn't enough NFB to push the higher orders out of earshot. To fix this you can purposely add low order harmonics so that the profile is at least one of decreasing magnitude as harmonics increase and rely on psycho-acoustics; this is a contentious solution. Another option is to avoid complex circuits with elements that have exponential transfer curves, try and stick with Triodes/JFETs so that it's mostly quadratic and have an open-loop distortion performance such that H2 is lower than 1% - quite a severe restriction that most middle of the road amplifiers fail to meet. Achieves reasonable damping factors. Clipping behaviour is degraded. Compromise between 1) and 3) with good bass and good HF ?
3) High global NFB: also a popular category, especially where marketing departments live. It pushes all distortion below audible limits, can use any kind of amplifying element including exponential transfer curves. BUT this approach places a huge burden on the designer to ensure the NFB is very accurate because all the devil's come out of the wood work if it isn't and usually hit the HF performance. Includes having to pay careful and particular attention to the error amplifier and gain-bandwidth so that there is enough gain at high freq plus minimal phase shift. This will be hard to achieve. Damping factor very high, clipping behaviour can be bad. PSRR is fantastic, takes pressure of psu design.
Understanding how accurate the NFB has to be to avoid creating problems for this last option would be useful.
Last edited:
just propose, we ll see..
as front end,i did test a VFB LIN, a CFB Profet , and i finally
sticked with a symetrical differential front end and hitachi s
mosfets for the output, with no miller caps on the road..
per channel, 6 bjts 2SA872A/2SC1775A for the front end
and 6 power devices 2SJ48/2SK133 with a +-50V rail...
compensation is made with current limitation in the vas,
so the gates capacitance are used as shunt compensation,
thus, there s no need for any Cdom attached to the 2 vas..
I'm glad you've found the Baxandall feedback concepts to be useful. They're certainly worth the effort to ponder in my opinion.
I think I know where you got this "accuracy of the NFB" idea, and I consider it to be a concept that's not well-formed. If you mean the integrity of the feedback error signal formed by the difference between the voltages at the non-inverting and inverting inputs (thinking of it like an op-amp), this can be thought of in terms of input stage distortion. In a properly-designed class AB feedback power amp, the output stage distortion will dominate, followed by VAS distortion, and lastly by input stage distortion. By "properly-designed", I'm assuming the input stage, if it's a BJT diff amp, has emitter degeneration. If it doesn't, it could well exceed the VAS in its distortion contribution. 100 Ohms here is a good rule of thumb.
Another interesting read on this subject is an interview with tube amp designer Scott Frankland. At first I was skeptical of this by how it starts out, but he does nail many of the issues without getting into arcane technical arguments. The interview is here.
Last edited:
Member
Joined 2009
Paid Member
I think I know where you think I got this idea and it was certainly a contributing factor. But when I say accuracy of nfb I was firstly & primarily thinking about the issue of phase. I noticed in Sims that distortion was worse at simulation frequencies where the phase was not a flat line in the small signal analysis plot (the one where the phase is a dotted line). My main concern is that the nfb can't possibly accurate if there is a phase difference between the two inputs to the error amplifier. It should be possible to relate this phase error to subsequent harmonic distortion profile and establish an accuracy criteria. Has this been done and published ??
As for the error amplifier itself. I recognize that it is possible to design a good one whether an LTP, Singleton or other. My feeling is that the LTP has it's strength when used with split rail amplifiers where dc offset needs to be controlled and with ICs where near perfect matching and thermal coupling is relatively easy. I like the concept of the Singleton for a DIY amp with single rails and I would like to build one to see for myself.
Interesting article, rather like a re-run of my thought processes this past day or so. However, it really only deals with option 1) and 3) from my list above - in that it seems to say zero feedback is one solution or lots of nfb with ultra-wide bandwidth is the other. I am still intrigued by the middle option, open loop dominated by H2 which itself should be below 1% and then adding nfb such that the harmonic products produced are still below -90dB or better.
i suppose that it use four 2SK1058 in a quasi
complementary scheme..
it has been proposed in a french site, and the
author claim almost perfect mo,otonic deacrease of
harmonics products..
the site is very intersting, with many links..
there s also a comparison of amps with simulations..
the passe s son of zen show awfull harmonic balance..
francis.audio
the quasi complementary amp :
http://pagesperso-orange.fr/francis.audio2/C07_FB1a.gif
schematics and review of manys designs
http://pagesperso-orange.fr/francis.audio2/Concours_Conception_Ampli.pdf
have some fun with the latter link..