PMA said:I think we have forgotten one important example when exploring NFB here. Let's have a nonlinear element with pure 2nd harmonic nonlinearity. Let's close it into NFB loop and review the resulting spectrum. Maybe jcx has already done it before?
I'm not sure we have "forgotten" that. Also, the Baxandall JFET example, which we verified, was probably as close to pure second order nonlinearity as we need to get.
The question is, what is our further objective here? I ask this in a positive way, to better focus on what rocks still need to be turned over.
I think we have already verified that with virtually any real device, the application of NFB around it under the right set of conditions can introduce new distortion spectra. I don't think anyone is disputing that. I think that the important thing is to keep that phenomena in perspective by looking at the actual numbers in real situations.
For example, we demonstrated that a CE BJT stage with a reasonable amount of degeneration will almost always have all of its distortion spectra reduced by the application of any amount of feedback around it.
We also demonstrated, interestingly, that emitter degeneration behaves in this regard no differently than any other kind of NFB. So those feedback-averse folks who think emitter degeneration is OK and that explicit local feedback is not, are simply fooling themselves if they are worrying about the creation of new spectra by NFB.
I think we also showed that the application of NFB around an output stage with crossover distortion reduces all elements of the spectra.
Let's try to identify the questions or controversies that remain. If we can find situations where NFB is a bad guy, we can at least identify the governing conditions that need to be avoided. For example, I think we have learned that one should avoid putting small amounts of NFB around an un-degenerated stage.
I'm game for more experiments.
Cheers,
Bob
Hi Bob,
well, a practical example. I have an output stage with local FB and very good linearity, and I have improved it recently. For usual listening levels it has only 2nd and 3rd harmonic in spectrum measured with 125 - 130 dB dynamic. Distortion far bellow 0.01%. Output impedance is less than 0.07 ohm. It runs hot, as expected. To close it into global NFB or not? 😉
Cheers,
Pavel
well, a practical example. I have an output stage with local FB and very good linearity, and I have improved it recently. For usual listening levels it has only 2nd and 3rd harmonic in spectrum measured with 125 - 130 dB dynamic. Distortion far bellow 0.01%. Output impedance is less than 0.07 ohm. It runs hot, as expected. To close it into global NFB or not? 😉
Cheers,
Pavel
Re: Re: East Coast Burning Amp
An amp meet might just motivate me to build something new, so I'd be interested also.
Pete B.
Bob Cordell said:
In girls at Skids? Of course there is. Many of us at RPI dated them.
OOPS!! 🙂, you meant amplifiers!!
I could definitely be interested.
Cheers,
Bob
An amp meet might just motivate me to build something new, so I'd be interested also.
Pete B.
PMA said:Hi Bob,
well, a practical example. I have an output stage with local FB and very good linearity, and I have improved it recently. For usual listening levels it has only 2nd and 3rd harmonic in spectrum measured with 125 - 130 dB dynamic. Distortion far bellow 0.01%. Output impedance is less than 0.07 ohm. It runs hot, as expected. To close it into global NFB or not? 😉
Cheers,
Pavel
Sounds like a nice design, and sounds like the local feedback is doing a very good job. Of course, how local is local? For example, if you think of error correction as a negative feedback loop with an embedded positive feedback loop with a closed loop gain of infinity, is the EC really a local feedback loop? Of course, I don't know if you are using error correction or something similar.
Do you by chance have a spectral plot of 19 kHz + 20 kHz CCIF IM at near full power? That would be interesting to take a peek at.
Best,
Bob
Bob Cordell said:
Sounds like a nice design, and sounds like the local feedback is doing a very good job. Of course, how local is local? For example, if you think of error correction as a negative feedback loop with an embedded positive feedback loop with a closed loop gain of infinity, is the EC really a local feedback loop? Of course, I don't know if you are using error correction or something similar.
Do you by chance have a spectral plot of 19 kHz + 20 kHz CCIF IM at near full power? That would be interesting to take a peek at.
Best,
Bob
This one does not use error correction. In principle, it looks like Walt Jung diamond buffer. Outputs are complementary feedback pairs, here the notice about local feedback. The output stage is biased at about 1.65A now. The structure is completely symmetrical, but not differential.
I am able to measure 13+14 kHz CCIF IM distortion, but have no current plot. I have plots made about 5 years ago, measured on previous version of this amplifier.
It makes significant sonic difference if I close that stage into global feedback or not. Though, the distortion of the output stage is only 2nd and 3rd harmonic, up to 0.004% at listening levels. Also CCIF IM, measured 5 years ago, was low. I believe we have to explore different methods when trying to correlate measured and perceived results.
Sincerely,
Pavel
PMA said:
This one does not use error correction. In principle, it looks like Walt Jung diamond buffer. Outputs are complementary feedback pairs, here the notice about local feedback. The output stage is biased at about 1.65A now. The structure is completely symmetrical, but not differential.
I am able to measure 13+14 kHz CCIF IM distortion, but have no current plot. I have plots made about 5 years ago, measured on previous version of this amplifier.
It makes significant sonic difference if I close that stage into global feedback or not. Though, the distortion of the output stage is only 2nd and 3rd harmonic, up to 0.004% at listening levels. Also CCIF IM, measured 5 years ago, was low. I believe we have to explore different methods when trying to correlate measured and perceived results.
Sincerely,
Pavel
Pavel,
Can you say anything about that difference in sound? Which one would you prefer, and why, sound-wise?
Jan Didden
Bob Cordell said:
Do you by chance have a spectral plot of 19 kHz + 20 kHz CCIF IM ?
My 1st attempt, simulator output:
(but simulared THD is different from measured, measured is about 2-3x lower. It horribly depends on transistor models)
Attachments
"sonic differences" may not be "interesting" if they result from easily predicted linear deviations between the global feedback vs open loop case
did the gain/listening level change by more than 0.1 dB?
did the output impedance change by enough to give 0.1 dB output diff? at any freq (such as load impedance dips)?
further the 10x reduction in sidelobes in the sim suggests a low-modest feedback factor that I think recent pages have shown can give complicated and suboptimal distortion spectrum changes
also the smaller change in 1 KHz diff product seems odd, I'd expect substantially higer reduction with typical single pole op amp gain
did the gain/listening level change by more than 0.1 dB?
did the output impedance change by enough to give 0.1 dB output diff? at any freq (such as load impedance dips)?
further the 10x reduction in sidelobes in the sim suggests a low-modest feedback factor that I think recent pages have shown can give complicated and suboptimal distortion spectrum changes
also the smaller change in 1 KHz diff product seems odd, I'd expect substantially higer reduction with typical single pole op amp gain
Feedback is 37dB at 13kHz and 59dB at 1kHz.
And it is ONLY simulation. As mentioned before simulation gives me 3x WORSE THD than measurement.
And it is ONLY simulation. As mentioned before simulation gives me 3x WORSE THD than measurement.
jcx said:"sonic differences" may not be "interesting" if they result from easily predicted linear deviations between the global feedback vs open loop case
Where is the lectured threshold of perceived distortion? Is 0.004% of 3rd assumed to be audible?
I am enclosing MEASURED result on a sample 5 years ago, output stage out of global NFB.
Attachments
jcx said:"sonic differences" may not be "interesting" if they result from easily predicted linear deviations between the global feedback vs open loop case
did the gain/listening level change by more than 0.1 dB?
did the output impedance change by enough to give 0.1 dB output diff? at any freq (such as load impedance dips)?
Do you often perform listening tests?
The sonic differences are very different from gain change. Output impedance was 0.07 ohm vs. miliohms (speaker wire has much higher impedance than both amps), gain might decrease about 1.5%.
If you done listening tests, you would know that 0.1dB in gain is inaudible. And does not change character of sound.
How would you explain sonic differences between AD797 and OPA627? By THD and IMD?
john curl said:What sonic differences have you found between the AD797 and the 627?
Good question 😀 . I would describe 797 as more detailed, more distinctive. 627 softer, pleasant, hiding a bit.
PMA said:
Good question 😀 . I would describe 797 as more detailed, more distinctive. 627 softer, pleasant, hiding a bit.
I think I better shut up now...😉
Jan Didden
pma,
I believe that meaningful listening tests are difficult
maybe you should think about what double blind testing does say (everyone jumps on what it can't tell you - proving the negative proposition that no one can hear x)
DBT testing shows that frequency response matching to within 0.1 dB is required to reduce the ability of a large number of listeners to discriminate with statistically significant reliability
http://www.provide.net/~djcarlst/abx_crit.htm
some indicate that differences of 0.1 dB over more than 1 octave are sufficient clue to discriminate between systems
So >0.1 dB frequency response differences can be discriminated and contribute to "voicing", some amp reviews have commented on a particular designer’s "signature" x dB lift/rolloff/"presence peak", ect. shaping of amplifier response
some people who make their living reporting subjective differences have had problems when amp responses are carefully nulled and listening levels matched
The Stereophile - Bob Carver Challenge (Bob wins!):
http://www.diyaudio.com/forums/show...2392#post152392
everyone participating in this thread should be sophisticated enough to know that psychoacoustics is an approximate and incomplete science which doesn’t have a answer to absolute distortion audibility - and that the math and experience show that there will never be a Simple answer
but an interesting data point is the “gold standard” of analog master tape recordings have very much larger 3 rd harmonic distortion than anyone has recently proposed as acceptable in an amp (a high quality analog tape example with a 75 dB s/n spec used 0 dB ref level defined as the level which gives 3% 3rd harmonic distortion!)
my questions are about the relevance of your amp’s sim/measurement to the thread’s topic – if the distortion didn’t decrease in rough proportion to the added loop gain around the output stage then your amp must have additional distortion in the earlier stage, which is a fine place to get to since improving lower level circuitry may be easier than fixing output stage distortion
I believe that meaningful listening tests are difficult
maybe you should think about what double blind testing does say (everyone jumps on what it can't tell you - proving the negative proposition that no one can hear x)
DBT testing shows that frequency response matching to within 0.1 dB is required to reduce the ability of a large number of listeners to discriminate with statistically significant reliability
http://www.provide.net/~djcarlst/abx_crit.htm
some indicate that differences of 0.1 dB over more than 1 octave are sufficient clue to discriminate between systems
So >0.1 dB frequency response differences can be discriminated and contribute to "voicing", some amp reviews have commented on a particular designer’s "signature" x dB lift/rolloff/"presence peak", ect. shaping of amplifier response
some people who make their living reporting subjective differences have had problems when amp responses are carefully nulled and listening levels matched
The Stereophile - Bob Carver Challenge (Bob wins!):
http://www.diyaudio.com/forums/show...2392#post152392
everyone participating in this thread should be sophisticated enough to know that psychoacoustics is an approximate and incomplete science which doesn’t have a answer to absolute distortion audibility - and that the math and experience show that there will never be a Simple answer
but an interesting data point is the “gold standard” of analog master tape recordings have very much larger 3 rd harmonic distortion than anyone has recently proposed as acceptable in an amp (a high quality analog tape example with a 75 dB s/n spec used 0 dB ref level defined as the level which gives 3% 3rd harmonic distortion!)
my questions are about the relevance of your amp’s sim/measurement to the thread’s topic – if the distortion didn’t decrease in rough proportion to the added loop gain around the output stage then your amp must have additional distortion in the earlier stage, which is a fine place to get to since improving lower level circuitry may be easier than fixing output stage distortion
OPA134 is quite very good opamp. IMHO not the absolute top class. I would prefer 797, and also, depending on application 627. I also did like AD844, in certain circuit designs.
- Home
- Amplifiers
- Solid State
- Bob Cordell Interview: Negative Feedback