I was very disappointed by that article, expect better from Pass
Nelson had to admit he exaggerated the IM generation situation when I pointed out that no real high feedback amplifier has the same signal level across the stages
the signal level, signal to bias ratio decreases from that seen in the output stage as you move towards the input stage in typical high loop gain global negative feedback amps
this means the distortion is lower in early stages, IMD being a product is lower by power factors of the order of the IMD
in real amplifiers there is no "Tsunami of IMD" - pure poetry on Nelson's part - supported by "distorted" math to pander to a anti-feedback audience's preconceptions
for a more realistic view I highly recommend: (article tests the developed estimation math against actual Hardware)
Cherry’s “ESTIMATES OF NONLINEAR DISTORTION IN FEEDBACK AMPLIFIERS” JAES V48#4 2000 p299-313 provides a method of calculating individual component distortion mechanisms contribution to overall distortion of a amplifier, more importantly it gives a intellectual framework for reasoning about distortion and feedback in a feedback amplifier.
Nelson had to admit he exaggerated the IM generation situation when I pointed out that no real high feedback amplifier has the same signal level across the stages
the signal level, signal to bias ratio decreases from that seen in the output stage as you move towards the input stage in typical high loop gain global negative feedback amps
this means the distortion is lower in early stages, IMD being a product is lower by power factors of the order of the IMD
in real amplifiers there is no "Tsunami of IMD" - pure poetry on Nelson's part - supported by "distorted" math to pander to a anti-feedback audience's preconceptions
for a more realistic view I highly recommend: (article tests the developed estimation math against actual Hardware)
Cherry’s “ESTIMATES OF NONLINEAR DISTORTION IN FEEDBACK AMPLIFIERS” JAES V48#4 2000 p299-313 provides a method of calculating individual component distortion mechanisms contribution to overall distortion of a amplifier, more importantly it gives a intellectual framework for reasoning about distortion and feedback in a feedback amplifier.
Last edited:
Interesting comment. I thought it was well-established (by experiment) that the ear is almost entirely insensitive to waveform shape, which is exactly what the filter-bank model of the ear would predict. Could someone put either me or Richard straight on this?RNMarsh said:
In any case, the FFT preserves all the information so if you are interested in wave shape you can get this. Wave shape is a time concept though, so why do an FFT if you want a time concept?
Why is that somewhat confused paper by Pass so often cited?zinsula said:
No, that's my error. The section in Bob's book actually is about the AB output stage, and that's what the graph shows. My bad, sorry.
Anyway, it not at all contradictory. The single-stage FET has a nice, curved transfer curve and as the math predicts, in the low feedback cases you get increasing harmonics with increasing feedback. In the class AB output stage the transfer curve is pretty flat to begin with so there the effect is absent.
jan
Last edited:
Thanks for the link Zin. I have not completely read the article, found this interesting.
I believe this comment needs expanded upon.
1) Once the signal is tainted, it cannot be brought back to pristine. That is why the circuit design, layout, and parts quality are so important. The amount of degradation is relative to the design and/or part(s) quality.
2) Many, on forums, use the term "synergy". First, "synergy" is not an absolute term. If one improves his system, the "synergy" has improved, thus it is a variable.
I have seen some postings that simply matching the "right" components, even average to poor quality components, will result in great "synergy". Well, that is somewhat false, as Nelson comment suggests, because the more a component degrades the pristine musical signal, the more is lost and cannot be perfectly retrieved.
In otherwards, an "average" component will degrade the music more than a "good" quality component,,, with even greater trouble bringing back the music. And no amount of matching can bring back the musical signal to pristine condition.
Thus it is imperative to find the most flawless components to begin with in order to maintain the music as close to pristine as possible.
Cheers.
I think Nelson's article is pretty good at explaining feedback and related issues to those who don't have an engineering/math training. Even if it uses literary freedom here orthere, it does a good job of explaining.
For those looking for a more advanced text I would recommend Bruno Putzeys' article in Linear Audio Vol 1.
("The F-word - or, why there is no such thing as too much feedback.")
jan
Hi John
I saw this video this morning and thought of you, was wondering if you’re in it?
New and Hot Video: Premiere: Grateful Dead Play 'Going Down the Road' at 1974 Show | Rolling Stone
Also, I have been meaning to bounce something off you and being a similar age, figured you would remember this test and wondered if you ever tried it on the electronics you work on.
At work, we used a generation loss test to evaluate loudspeakers, both ours and our competitors and figured while loudspeakers are by far the least faithful to the signal of all, it might be something of use here.
What we did was hoist loudspeakers up on a tower with a precision measurement mic (to remove room effects and reflected sound leaving only the loudspeaker). Using a 24/96 stereo recorder, we played music through the speaker and recorded the mic signal.
The mic signal was played back and re-recorded. AT the same time, the music track was also re-recorded so that the AD/DA transmogrification process was captured at the same time for reference.
While loudspeakers generally only go one or just a couple generations before sounding bad and electronics are much more faithful to the signal, I thought maybe this approach from the recording tape era might be useful for revealing the “sound “ of various caps and electronic parts etc.
Best,
Tom Danley
Danley Sound Labs
I saw this video this morning and thought of you, was wondering if you’re in it?
New and Hot Video: Premiere: Grateful Dead Play 'Going Down the Road' at 1974 Show | Rolling Stone
Also, I have been meaning to bounce something off you and being a similar age, figured you would remember this test and wondered if you ever tried it on the electronics you work on.
At work, we used a generation loss test to evaluate loudspeakers, both ours and our competitors and figured while loudspeakers are by far the least faithful to the signal of all, it might be something of use here.
What we did was hoist loudspeakers up on a tower with a precision measurement mic (to remove room effects and reflected sound leaving only the loudspeaker). Using a 24/96 stereo recorder, we played music through the speaker and recorded the mic signal.
The mic signal was played back and re-recorded. AT the same time, the music track was also re-recorded so that the AD/DA transmogrification process was captured at the same time for reference.
While loudspeakers generally only go one or just a couple generations before sounding bad and electronics are much more faithful to the signal, I thought maybe this approach from the recording tape era might be useful for revealing the “sound “ of various caps and electronic parts etc.
Best,
Tom Danley
Danley Sound Labs
Yes, blind vias in pad are another important tool, especilly for fine pitch BGAs. I just got a quote the other day for a board with blind vias in pad vs normal. Increased cost of the bare board was 90% (almost double), but you gotta do what you gotta do. BTW this 90% increase is on top of what is normally an $800 board (18 layers in Megtron 6 with 2500 components). BTW, another interesting thing we do is put 1uF 0402 ceramic bypass capacitors right in the BGA array. Up until a couple years ago, I didn't even know that you could buy 1uF caps in a size as small as 0402.
Cheers,
Bob
It should ne noted that the discussed distortion is specifically caused by nonlinear active devices. Although (passive) component quality and layout are important, the effects are secondary to the 'main' culprits.
Also, feedback 'linearises' the whole amp, including the effects of passive components and layout issues.
You could also debate whether 'once the distortion is there it cannot be undone' is true; - in fact, using feedback, it IS undone! An amp that has 1% distortion wrapped within a 20dB feedback loop will now have only 0.1% distortion...
jan
Last edited:
Yes, some parts of the Pass article are fine. The problem is that people don't realise when Pass is teaching basic engineering, and when he is offering his own somewhat idiosyncratic view of audio. They then quote Pass as a guru, and wonder why some of us smile and refuse to bow.
Last edited:
Hi Jan,
I have to respectfully disagree. There are many more parts than active devices, and their sonic contribution can easily over bear that of the active device. That and frequencies as low as 5khz can be "transmitted" from wires and parts to adjacent wires/parts, so layout is very important to keep channel to channel separation.
Cheers.
... and I am sure you can link to some stuff that backs this up, not being the anecdotal 'everybody knows...'?
jan
My own decades of research, John Curl's, Thompson's, Crump's, Picking capacitors by Richard Marsh/Walter Jung. I suppose there is more but cannot remember.
I am interested in what research have you performed Jan since you are questioning me?
Please excuse the off topic posts.
Cheers.
Last edited:
What research? What are the conclusions? I know the reports from these respected persons. Some showed measureable effects from component linearities, others don't get farther then 'because I say so'.
My statement that component nonlinearity is secondary can be easily verified. A typical FET or BJT can easily have anywhere from 1 to 10% nonlinearity - just check the data sheet curves.
A standard, cheap, metal film resistor has a non-linearity down into the -100 to -120dB or more - read Ed Simon's article in Linear Audio Vol 0. That's 0.00001 to 0.000001%.
Look, I'm not saying components are not important, I'm not saying there cannot be cases that they can contribute to audible differences. I'm only saying that their effect is secondary and that any controlled tests for these are very, very scarce!
If you believe otherwise we should agree to disagree. You have the last word.
jan
.
Bruno's article is necessarily brief but manages to touch on a variety of doctrines and does a pretty nice job of demolishing them. In particular he addresses the notion that having open loop gain cornering at a low frequency is prima facie a bad thing, which is a cornerstone of Otala's prescriptions.
What is not discussed, but surely something of which Putzeys is well aware, is the behavior of systems in overload. I remarked earlier that Miller and Zanfino suspected this occurs far more frequently than most people believe. Under those conditions a low-feedback (i.e., low loop gain) system may well handle the overloads better.
However --- if that is going to be a problem, surely the strategy should be to limit the signal peaks ahead of the overload-prone systems. And this can be done cleanly, if you are diligent.
By the way, that consideration also feeds into the supposed audibility of waveform shape. If a waveform has a high peak-to-average ratio compared to one that does not, even when both have identical spectra, obviously the first one will push your system into clipping at lower average levels. Presumably, again at high levels, the ear will respond similarly.
Brad
Hi Zinsula,
Perhaps I can clarify that graph a bit more. It is Figure 24.5 from my book "Designing Audio Power Amplifiers" from the chapter titled "The Negative Feedback Controversy".
The graph was arrived at by simulation of a BJT class AB output stage, and the results do not represent that from a whole amplifier (although by extension the results are also applicable to a whole amplifier whose distortion is dominated by the output stage). Basically, the simulation included an output stage circuit and ideal gain blocks to create/change/increase the amount of negative feedback around the output stage. So the only source of nonlinearity was the output stage.
The text description of the graph is as follows:
"Crossover distortion is usually the biggest and most audible distortion in a properly designed amplifier. It starts out being rich in high-order products. Figure 24.5 shows that the application of NFB to an amplifier reduces all orders of crossover distortion, right from the beginning. Although the Baxandall effect was an intriguing eye-opener, the audio community read too much into it, wrongly generalizing the results and asserting that negative feedback did not really reduce the net imperfection of the signal."
Cheers,
Bob
Fully agree - one of the reasons that a ss amp needs probably a 10x power spec than a comparable tube amp to sound 'nice'. Tube amps typically clip much more benign than high-feedback ss amps. All my own amps that I developed for the last 25 years have some form of soft clipping.
Edit: Here's Bruno's article for those who missed Linear Audio Vol 1
:http://www.linearaudio.nl/Documents/Volume_1_BP.pdf
jan
Last edited:
Although "tubes overload better" is a common mantra, in reality, tube amps have an overload recovery issue not present in solid state- blocking. A very few are designed to minimize this effect (e.g., Morgan's Crystal Palace, my Red Light District), but most aren't and unfortunately sound that way when driven by very dynamic music.
Hi Brad,
You are exactly correct. The crest factor of well-recorded music can be very large, and the numbers can get ugly fast. Take some measurements on Rickey Lee Jones' "Ghetto of my Mind" from her Flying Cowboys album. With 84-dB speakers in a hotel room at RMAF and at realistic but not annoyingly loud levels, we clipped a 250-wpc amplifier while the average power was on the order of 1-2 watts. The thwack of the snare drum was the prime culprit.
In fairness, and unfortunately, most music is not recorded with this much dynamic range. Nevertheless, many people are clipping their amplifiers more often than they think, and how an amplifier behaves in clipping does matter to the sound.
Controlled or soft clipping outside the feedback loop is an excellent approach, even if it cuts a bit into the impressiveness of the overall amplifier's THD measurements at high levels (particularly so for soft clipping as with the Klever Klipper I use on some of my amplifiers). The proper use of Baker clamps inside an NFB amplifier can also be very effective in eliminating undesired clipping waveshapes (like "sticking").
Cheers,
Bob
- Status
- Not open for further replies.