“My understanding is that for a reasonably linear circuit lower order harmonics mostly are produced without global nfb and the degree is dependent on this intrinsic linearity.
As GNFB is applied these lower order distortion harmonics are reduced in amplitude as is the goal.
As GNFB is further increased lower order harmonics are further decreased but higher order harmonics increase/emerge from the noise floor and can become the subjectively dominant harmonics according to nfb level.
At high nfb levels the 'spray' of higher order harmonics can be programme dependent and the ear is very sensitive to/protests these unnatural dynamic changes in the system noise floor.
Lower levels of nfb allow operation without this dynamic generation of higher order harmonics and can be 'monotonic' and lower order only which is more 'natural' sounding.”
—
This whole line of thinking comes from an experiment originally carried out by Baxandall. He took s very simple single stage amp and looked at the distortion spectra as feedback was increased. As it’s increased, lower order harmonics are ‘folded’ into higher order ones and indeed the total distortion goes down. There is a region of between 6 and 15 dB (IIRC) where the folding phenomena makes the higher order harmonics quite prominent and objectionable.
However, the test circuit was exactly that - a highly compromised (highly non- linear to start off with in the open loop condition) circuit and not something you’d use normally. Secondly, in modern amps, you start off with a circuit that is already very low distortion in the open loop condition. Consider an amp that has 70 dB feedback at 20 kHz (perhaps using TPC or TMC) with a 20 kHz distortion of 2-3 ppm (quite a few designs on this forum). The open loop distortion of such an amp at full power is under 0.01%.
So, the trick is to make it linear in the open loop condition and only then to close the loop. As Bruno says ‘don’t be a whimp’
Feedback: A Short History
Baxandall articles on amplifier design Audio Power Amplifier Design – Peter J Baxandall
As GNFB is applied these lower order distortion harmonics are reduced in amplitude as is the goal.
As GNFB is further increased lower order harmonics are further decreased but higher order harmonics increase/emerge from the noise floor and can become the subjectively dominant harmonics according to nfb level.
At high nfb levels the 'spray' of higher order harmonics can be programme dependent and the ear is very sensitive to/protests these unnatural dynamic changes in the system noise floor.
Lower levels of nfb allow operation without this dynamic generation of higher order harmonics and can be 'monotonic' and lower order only which is more 'natural' sounding.”
—
This whole line of thinking comes from an experiment originally carried out by Baxandall. He took s very simple single stage amp and looked at the distortion spectra as feedback was increased. As it’s increased, lower order harmonics are ‘folded’ into higher order ones and indeed the total distortion goes down. There is a region of between 6 and 15 dB (IIRC) where the folding phenomena makes the higher order harmonics quite prominent and objectionable.
However, the test circuit was exactly that - a highly compromised (highly non- linear to start off with in the open loop condition) circuit and not something you’d use normally. Secondly, in modern amps, you start off with a circuit that is already very low distortion in the open loop condition. Consider an amp that has 70 dB feedback at 20 kHz (perhaps using TPC or TMC) with a 20 kHz distortion of 2-3 ppm (quite a few designs on this forum). The open loop distortion of such an amp at full power is under 0.01%.
So, the trick is to make it linear in the open loop condition and only then to close the loop. As Bruno says ‘don’t be a whimp’
Feedback: A Short History
Baxandall articles on amplifier design Audio Power Amplifier Design – Peter J Baxandall
Last edited:
Didn’t know truth can be surrealistic 😀. I’ve certainly learned something today 😀.
The rest of this ad hominem is not worth discussing, much to do about nothing. You could use your energy in much more helpful ways, like describing a simple test plan for anybody to try.
You are joking of course, Mark, Jakob and Merrill have already explained how this has no value to them, and so it follows, should have no value to anyone else.
Always glad to be corrected, thanks.
If OL THD is 'under 0.01%' does it then really matter, and what are the other distortions and do they matter ?.
IME open loop can sound good, really good.
What problems does high degree of FB solve that mid or low FB does not solve or what problems does mid or low FB introduce ?.
Dan.
Last edited:
This may be "obvious" but it is not always true. In much of engineering, including audio electronics, ordinary good commercial quality is sufficient for the task so spending more money achieves nothing in technical terms although it may do wonders for marketing.mountainman bob said:
You are misunderstanding what people say. The issue is not "regardless of quality" but 'good enough really is good enough and good enough in most cases is cheap'. A big exception is signal transformers, such as microphone transformers and valve output transformers; with wound components you usually do need to spend more money to get better results. This is because wound components are about the most non-ideal of anything used in audio - apart from transducers.
No, not in electronics.
Yes, enough. Less than this is not enough. More than this requires more forward gain which then has to be provided and paid for and may require engineering compromises.RNMarsh said:
That is true. PM/GM are an attempt to quantify stability for those who need simple numbers. To get the true picture you need to plot the locus of loop gain on the complex plane. I assume this is what you do in your design?johnego said:
This is what a lot of people seem to imagine, but it is not true. The re-entrant distortion does not begin to emerge at high levels of feedback; it occurs at low levels of feedback, and then needs higher levels of feedback to suppress it. However, in many cases the re-entrant distortion is lower in level than the intrinsic distortion in the forward path so it can be a diversion from what matters.Max Headroom said:
Not correcting - just highlighting historical information that you perhaps were not aware of.
I have no beef about absolute distortion levels below 0.1 or 0.01% since under controlled listening, most (which not the same as saying all listeners) listeners will not be able to tell them apart. The beef starts when folks claim zero feedback amplifiers are better than ones the employ feedback and then cite completely erroneous facts/ideas to support their claims. The article I linked to earlier explores some of these issues.
As Cordell has pointed out in his book, designing good exemplars of either type requires some serious engineering - on the one hand how to linearize each amplifier stage in order to end up with a total zero feedback distortion of <0.1% and on the other, the same again plus an understanding of how to apply feedback successfully.
No; the gain and phase margin numbers are defined precisely in the context of the Nyquist plot, and are reflecting exactly the stability margins, according to the most general stability criteria.
A Nyquist plot is a rather simple transformation of a Bode plot, through:
x=real(M*exp(j*P))
y=imag(M*exp(j*P))
where M and P are the magnitude and phase of the Bode plot (the reciprocal transformation is also simple). Therefore, the gain and phase margins in a Bode plot, as we use them, are an exact metric of the system stability. What can be questioned (and is usually a very good idea to do so) is how stable are these margins against the real world (components tolerances, temperature, large signal parametric variations, etc...). Nothing that a competent engineer cannot address in the design phase.
There is the special case of a multi loop system, where peeling off feedback loops, one by one, and making sure each loop is stable, is a sufficient but not a necessary condition for global stability. That is, some feedback loops can be unstable, but the global system still stay stable. A more general stability criteria for multi loop systems is that the Nyquist locus trajectory encompasses the -1+j*0 Nyquist point an even number of times.
To directly (and ultimately) check the stability of such a system you would have to calculate the eigenvalues of the state transition matrix. As long as eigenvalues magnitude is less than 1 or real part less than zero, respectively, then your system is stable.
Last edited:
I don't dispute that they are defined precisely, and never said otherwise. Do they define stability? Can two numbers exactly define the locus of the loop gain and determine unambiguously how many times it loops around the critical point? That I doubt, although I am open to correction as I know that analytic functions can have unexpected properties. I suspect that the two numbers are fine provided that certain assumptions are true, which assumptions will be true in almost all normal audio design.syn08 said:
My main point was to challenge johnego, to see if he really was going further than the two numbers when I suspected that he was doing less.
Low feedback introduces re-entrant distortion, which may or may not be significant. More feedback suppresses this distortion. High feedback suppresses it even more. Hence the right amounts of feedback, with respect to this problem, are none or enough. If the forward path is already low in distortion then re-entrant distortion will not be a problem and you can safely use low levels of feedback to set gain or reduce output impedance.Max Headroom said:
This term might be unhappy, as it evokes popular audiophile belief that feedback "goes round and round". I think you might rather say that feedback may create distortion components not present in the input signal, and the resulting spectrum may depend on loopgain curve and OL transfer function.
Perfectly right. And it can be easily understood. There is several poles in an amplifier. That we can look as a delay between the signal and its correction to simplify our way of thinking. Faster is the amp, more rapid is the correction, but higher frequency is the distortion (delay between the original signal and the moment where the correction is applied).
(Don't fire-me, please, It is a question of phases, I know ;-)
The mysterious question is why our ears can be sensitive to theses added parasitic signals that are out of the frequency range of our ears.
More than this, how those very high frequencies can have an influence on the way speakers are moving, while they are filtered by the inertia of the tweeters ? More than this, in a good amplifier, the level of those high order distortions is so low, nowadays, that I can hardly believe it is audible. And yet !
Right too, but, now, the question could be seen another way: the feedback increase the "Damping" factor of an amplifier. Is-it the source of the difference we can hear ?
To figure out this question, one could do a listening comparison between high GNFB and low one, taking care of keeping the damping factor unchanged between the two.
[edit] I had seen the correction of PMA, that is perfectly right, reason why i said "don't fire-me". But I'm too lazy to explain the difference between a delay and a phase turn: at a given frequency it is the same. ;-)
Last edited:
An even more mysterious question is why people keep perpetrating this “feedback delay” aka “feedback goes round-and-round” BS. I personally find it not even intuitive, although intuition is sometimes called to compensate for the lack of knowledge/understanding.
Our messages are crossing. I took care to precise. I know the difference. and made an edit to avoid your answer ;-)
Well, about this the difference is, indeed, related to the open loop of the amplifier. I tend to prefer CFA for the treble, with as minus poles a possible. And this for a very simple reason, it is easier to design a CFA that has an open loop flat to 10KHz. While, VFA allows higher feedback ratios, but lower open loop, if you keep high impedance of the feedback path high enough in order to keep the LTP balanced.
The difference is that VFAs will have less distortions at low frequencies (higher feedback ratios) but CFA will produce a constant feedback ratio up to higher frequencies.
We all will be happy with both of them when the active devices will all be fast enough to provide an open loop bandwidth from 20 to 20 000Hz when assembled.
It can look strange that, for audio, with active devices that goes , nowadays, up to hundreds of MHz, they are not fast enough for our 20-20 000Hz target, but it is the way it is.
An other question that remain mysterious to me, is why the brunch objectivists of this forum are constantly insulting people. Yes, syn08, it is addressed to you. I have probably designed more commercially successful electronic gears than you since more years than you and I don't need our advices on matters I had learned and experienced probably more than you. I don't addressed your knowledge, on my side.
Each time i open one of the message of those few who are in my ignore list , I always read the same monotonic inputs. Or an insult, or the same words: "ABX", "proof", "sighted", "flawed", "level matched", "provide measurements" in loops. It is just BORING and disagreeable.
If you started by learning the civilized way of communicating between people of good education?
And, now, i will give-you the response at your technico/psychological question. For us, humans at a given instant and a given frequency, it is easier to figure out a delay, that is a natural effect we experience every time in our lives, that a "phase", that is a word that has no meaning for much of the people. And, at a given moment and a given frequency, it is the same result.
We don't see the electrons, we are free to build our own mental images are long are they are not leading-you to false conclusions. I have worked enough on servos to know how work feedbacks. Including servoed speakers that was my first interest in the 70 years.
Last edited:
I prefer to talk about harmonic spectra 'folding' up into higher order harmonics as feedback is applied precisely because re-entrant gets interpreted as 'feedback going round and round' - from the (in)famous Colloms article 'A future without feedback' (1998, Stereophile) - the Stereophile website seems to have an issue today so I cannot link it.
Ok daddy, can I go and play now?
If you admit you are confusing "delay" and "phase" in your #21522 then there's nothing more to add. Otherwise, the "feedback goes round-and-round" model you described is still <ahem> flat wrong, aka <ahem> BS.
P.S. I don't have any plan to entertain you, and "boring" is in the eyes of the beholder. I could myself claim boredom when reading about "large open loop bandwidth is good", "feedback goes round and round", "high feedback is bad", "40 years ago", "3 legs good, 8 legs bad", "current feedback is better", etc...
Last edited:
I’ve been down the ‘good enough’ path enough times to know different.
I usually find a happy place somewhere in the middle.
I find it interesting that you can think for me.......I know,I know somebody has to right. 😀
- Status
- Not open for further replies.