Because I have the experience and the reputation. It is rather pointless for me to convince you that this is important. I am also relying on the experience of my fellow designers, Chas Hansen and Nelson Pass. We have had active competitions with each other and the rest of the world, so we have some idea as to what works and what doesn't.
Many here would like to find their own way, and ignore our past experience. That's OK with me.
Many here would like to find their own way, and ignore our past experience. That's OK with me.
john curl said:
But the difference is that Nelson is a gentleman and doesn't tell others he has better ears than anyone else, patronize, or chastise anyone who doesn't agree with him. One the other hand, he is genuinely helpful to DIYers, rather than his own ego, and when he has an opinion of the sound, he simply says he likes this or that better. John, you remain clueless to the fact that most of the criticism of you is not to your knowledge or skill, but to your demeanor.
Johan Potgieter said:
I have learned more from the last few pages of posts then of all the other posts in the last few years! This has been absolutely fascinating! Thank you all.
BUT I have the same uneasy feeling as Johan. The math says that if you have harmonics, and you put them through a feedback loop back into the same non-linear forward path, they intermodulate. You get sum- and difference harmonics; this is not new of course.
Now on Baxandall's plots, and as confirmed by Bobs work, I can understand that the various amplitude- and phase of the individual harmonics mix and cancel or add to various degrees and give rise to the ripples in the harmonic amplitude vs feedback factor we saw.
In some of Andy's work these riples have been absent; if I understand correctly because the zero nfb really wasn't zero nfb, there was, for instance, (explicit or implicit) emitter degeneration. That would mean that the 'ol gain' (with the degeneration) is rather linear to begin with; the math would then indicate that the additional harmonics due to the feedback circulating the initial harmonics would be quite low. Still, I would expect SOME evidence of it in Andy's graphs.
Am I missing something?
Jan Didden
john curl said:'Starved', is a relative thing, Bob. Your amps are 'starved' compared to my best ones, but my cheaper ones may have about the same bias as your first effort. Is 'current challenged' OK?
Actually, "current-challenged" is not OK, either, since that would for most people make them think of peak output current capability, which has little to do with Idle bias.
Bob
Johan Potgieter said:
Hi Johnan,
The Class-AB crossover distortion simulations that I did were done on an output stage with +/- 50V rails, which would be typical of a 100W amplifier into 8 ohms. The signal level at which I ran the sims was 10 V p-p, or about 3.6 V rms. This all corresponds to about 1.6 watts into the 8-ohm load.
Crossover distortion and its harmonic structure can be a function of level, and I do plan at some point to look at it at some other levels.
Since distortion is correlated to the program, we can not always take comfort in the fact that it may be below the noise floor.
Cheers,
Bob
I know that NP has 7 patents. From this 7 patents, 4 of them are related to output stage.
3,995,228 is sliding bias.
4,752,745 is opto bias.
5,343,166 is efficient classA output stage
5,710,522 is Aleph output stage
the others are
4,107,219 is the Statis patent
4,899,387 is acoustic patent
5,376,899 is SuSy patent
Just from counting the purpose of each patent, I should know that the most problematic is the output stage
3,995,228 is sliding bias.
4,752,745 is opto bias.
5,343,166 is efficient classA output stage
5,710,522 is Aleph output stage
the others are
4,107,219 is the Statis patent
4,899,387 is acoustic patent
5,376,899 is SuSy patent
Just from counting the purpose of each patent, I should know that the most problematic is the output stage
janneman said:
Me too. I just wanted to let everybody know that the idea of trying to replicate, then extend Baxandall's work on distortion using SPICE was Bob's alone. I was aware of what he was doing beforehand because he emailed me last week with his idea. I think it's a fantastic idea. This was Bob's first foray into SPICE distortion analysis, and he was having some troubles getting the residual distortion down. LTSpice is a PITA in that regard. I just showed him the usual tricks for getting LTSpice to cooperate. I hope I didn't give the impression that any of this was my idea, as it was not.
One of my earlier plots (of the BJT) did not correlate to Baxandall's or Bob's plots. This ended up being due to the resistor attenuator in the base that I copied from Baxandall's article. Going to voltage drive gave much better correlation and showed the dip in the third harmonic at 3 dB feedback. The problem plot was in this post, and the corrected one was in this post. These show the ripples. Then Bob tried the same kind of simulation, but with a class AB crossover distortion mechanism here. Note there are no ripples. John had some concern that the overbias in Bob's class AB sim was artificially making the feedback look good, and also Pavel wanted to see the effect of frequency compensation. So I did the frequency compensated case and optimum bias here. Bob's and my results are similar with regard to the lack of ripple as the feedback is varied with a class AB distortion mechanism.
I'm not sure what the math would indicate. The only case in which the math is tractable is the FET case in which the open-loop distortion is only a second-order nonlinearity. Then the binomial expansion gives a closed-form expression for the entire power series. But for the BJT case, there is no closed-form expression for the entire power series. Instead, it must be evaluated term-by-term by taking derivatives. Baxandall gave up after the third-order term. For more complex cases, the math becomes completely intractable. If there are energy storage components (like frequency compensation), it gets even worse. The solution then is no longer a simple power series but a rather nasty Volterra series. In the Volterra series, the first term is a single integral, the second a double integral, the third a triple integral, and so on. At this point, assuming accurate models, SPICE seems the only viable alternative, and for me at least, I have no clue as to what the end result should look like.
andy_c said:
This article may help in an analytical attempt to analyze the distortions (unfortunately it's not free).
http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/10444/33166/01562224.pdf?arnumber=1562224
You may want to take a look at this one as well:
http://www.acse.shef.ac.uk/~dodd/papers/files/ifac2005d.pdf
syn08 said:
the 1st paper doesn't get as far as the last few days of sim, it includes only the 2nd order nonlinear term in the feedback example buffer circuit, nor is feedback level varied
likewise the 2nd paper seems not directly relevant, allowing for the math that I can't follow, it looks like they have just compared their Volterra model "predictions" with the data used to identify the system - not testing the predictive value by changing signal level
but both point to the considerable academic interest and activity of the past decade or so sparked by telcom/cell/dsl demands for low distortion amplifiers at high frequencies - distortion, esp. IMD eats into channel capacity in frequency domain multiplexed channels
which is an interesting return to the roots of negative feedback where the boys at Bell were faced with the same problem at the beginnings of transcontential voice lines
myhrrhleine said:
Very good question. Noise levels are always bandwidth related. I think to simply spread it over the audio band (BW = 20 kHz) is a too simple assumption in relation to what we can perceive audibly. We simply don’t know how our ears and brains process the mix of noise and correlated signals. IMO we can’t say the noise level is a border to what we can perceive Maybe if the noise level itself is below the treshold of hearing .... ?
Note the trained radio operator on a ship who can extract intelligibly information burried in the noise.
myhrrhleine said:
As Pjotr said, likewise (good remark).
I meant the Self quotation more as an image than an ideal. (I am not sure that he himself considered this enough.) I should perhaps have said the threshold of audibility - which, if the amplifier is "quiet" by hi-fi standards, is not too far below the noise floor.
Apology if this is old hat - I vaguely recall that it was somewhere in the 70s, when Peter Walker's discovery that his very good transistor amplifier nontheless caused listener fatigue, the business of "objectionableness" of high order harmonics was seriously studied. At least that was my first contact with the phenomenon. As I recall a Scandenavian study (aren't they often on the forefront) revealed that the brain might be sensitive to levels of certain strident tone combinations even below the threshold of audibility (such a test is obviously difficult to quantify). It was already known that the ear did not stop "detecting" at the threshold of audibility; we only stop sensing it as conscious sound. Brain activity was certainly observed as a result of such low stimuli. But apparently the rejection of an amplifier on no other grounds than that there were high order harmonics (compared to "clean" amplifiers), called listener fatigue for lack of a better definition, sensitized folks to this business of stridency at very low levels.
Thus I would certainly not stop there. I try to go some 10dB lower, but that is my subjective call.
Pjotr!
The radio operator and noise - ingenious and a little naughty [
back to you]. Yes, but now this opens another interesting subject! The recognition of words there, yes - but tones, less probable. Again not to bore and digressing somewhat; I had experience of this in the design of radio tracking equipment for wild animals. A pulsing sound at a minumum repetition rate of about 1/second could be barely picked out at 6dB below wide band noise level, when you were expecting it. (We were working at about 8nV into 50 ohm rf signal!) If lower than this is audible ... please do not post here because then I did not earn my salary, having tried to study all on this at the time!Myhrrhleine, this is the scope of my take on this subject, if rather vague and lengthy. Perhaps others can broaden our horizons. The wealth of literature available at the CSIR is sadly no longer at my disposal after retirement, and the internet ... well, one has to pay for most info of value in this field.
Could it be that the auditory perceptual model based codec people have anything to say?
You could look it up yourself, but briefly the central principle of modern audio codecs is exactly that signal below the perceptual threshold ( taking into account critical bands, time, frequency masking in addition to the hearing threshold curve) can be cheerfully tossed overboard by not wasting bits coding those signals
http://www.mp3-tech.org/programmer/docs/index.php
It sure seems likely that we could claim distortions that would have only existed in the tossed bits due to the low levels of the distortion components, ie way below those perceptual thresholds wouldn’t be any easier to hear than the signal bits that were tossed
The success of audio codecs at tossing 75%+ of the “information” (say 320k AAC+, Ogg Vorbis or current best) on a CD resolution recording with a vanishingly small fraction of the population detecting the difference would seem to pose a big logical obstacle to too much speculation along the lines of ”infinite” human auditory resolution and sub-noise floor distortion detection.
Not that I'm advocating designing amps any where near that bad!, just pointing to a solid body of evidence of real perecptual limits
You could look it up yourself, but briefly the central principle of modern audio codecs is exactly that signal below the perceptual threshold ( taking into account critical bands, time, frequency masking in addition to the hearing threshold curve) can be cheerfully tossed overboard by not wasting bits coding those signals
http://www.mp3-tech.org/programmer/docs/index.php
It sure seems likely that we could claim distortions that would have only existed in the tossed bits due to the low levels of the distortion components, ie way below those perceptual thresholds wouldn’t be any easier to hear than the signal bits that were tossed
The success of audio codecs at tossing 75%+ of the “information” (say 320k AAC+, Ogg Vorbis or current best) on a CD resolution recording with a vanishingly small fraction of the population detecting the difference would seem to pose a big logical obstacle to too much speculation along the lines of ”infinite” human auditory resolution and sub-noise floor distortion detection.
Not that I'm advocating designing amps any where near that bad!, just pointing to a solid body of evidence of real perecptual limits
Pjotr said:
Very good question. Noise levels are always bandwidth related. I think to simply spread it over the audio band (BW = 20 kHz) is a too simple assumption in relation to what we can perceive audibly. We simply don’t know how our ears and brains process the mix of noise and correlated signals. IMO we can’t say the noise level is a border to what we can perceive Maybe if the noise level itself is below the treshold of hearing .... ?
Note the trained radio operator on a ship who can extract intelligibly information burried in the noise.
I have a book on these issues that describes various tests. In one, two harmonically related tones are produced, where one tone (the strongest and lowest in freq) actually masks the 2nd. Subjects cannot hear whether the 2nd tone is on or off. THEN they add random noise, and what do you know, all of a sudden the 2nd tone is clearly heard. So, yes, there's more to it than just the noise level itself.
Jan Didden