The "What's your reasoning?" and not "What's your belief?" thread got my interest to skim the Otala, Cordell, and Gilbert papers and I thought I'd offer a few comments for discussion:
It's interesting that Otala's assumptions 1.1 - 1.3 do not include frequency dependent nonlinearities, or even worse, time variant nonlinearities in the open loop model. This would be fine if it were even approximately true, but it usually is not. It seems that Cordell's findings also suggest not, and this is why most real world amplifiers have PIM in the forward path, and as is seen in his article feedback reduces PIM when it is present in the forward path.
Very interesting that Cordell only tests OP amps in the inverting configuration since it is well known that their high frequency performance is much better than in the non-inverting case. The non-inverting configuration is much more widely used and there's no excuse for him choosing inverting. The inverting configuration has both diff amp inputs at ground, or virtual ground, and thus there is no common mode signal at these inputs. The worst case unity gain non-inverting configuration has both diff amp inputs swing the same range as the output. 10V out means a 10V diff amp common mode signal which modulates Vcb and the Miller capacitance. Refer to "Distortion in low-noise amplifiers" by Eric F. Taylor, Wireless World August, 1977 for a detailed discussion.
Seems if Otala expected to get a paper published he should have done some verification of his assumptions to determine if they offered a "good enough" model of what we commonly see in typical amplifier topologies. Cordell seems to be determined to prove the other side and weakens his position by using the less common inverting configuration.
It's interesting that often a certain configuration, such as amplifers with a Cdom cap (or Diff amp driving the VAS as an integrator) are assumed in a paper and we have to keep in mind that the analysis and conclusions are only valid for that configuration. This is seen in the Gilbert paper and Self's analysis with regard to Cdom. Some amplifiers such as the Bryston and Tigers do not fit this topology.
I always liked the Hafler XL power amp null test for those who claim there is an "unknown" distortion or design issue where one can use real source material, real loads, and hear or view the unmasked distortion(s), even types that have no name. Obviously, linear errors such as frequency response and phase shift result in the presence of non-distortion components in the subtraction but it is still worthwhile.
Comments please?
Pete B.
It's interesting that Otala's assumptions 1.1 - 1.3 do not include frequency dependent nonlinearities, or even worse, time variant nonlinearities in the open loop model. This would be fine if it were even approximately true, but it usually is not. It seems that Cordell's findings also suggest not, and this is why most real world amplifiers have PIM in the forward path, and as is seen in his article feedback reduces PIM when it is present in the forward path.
Very interesting that Cordell only tests OP amps in the inverting configuration since it is well known that their high frequency performance is much better than in the non-inverting case. The non-inverting configuration is much more widely used and there's no excuse for him choosing inverting. The inverting configuration has both diff amp inputs at ground, or virtual ground, and thus there is no common mode signal at these inputs. The worst case unity gain non-inverting configuration has both diff amp inputs swing the same range as the output. 10V out means a 10V diff amp common mode signal which modulates Vcb and the Miller capacitance. Refer to "Distortion in low-noise amplifiers" by Eric F. Taylor, Wireless World August, 1977 for a detailed discussion.
Seems if Otala expected to get a paper published he should have done some verification of his assumptions to determine if they offered a "good enough" model of what we commonly see in typical amplifier topologies. Cordell seems to be determined to prove the other side and weakens his position by using the less common inverting configuration.
It's interesting that often a certain configuration, such as amplifers with a Cdom cap (or Diff amp driving the VAS as an integrator) are assumed in a paper and we have to keep in mind that the analysis and conclusions are only valid for that configuration. This is seen in the Gilbert paper and Self's analysis with regard to Cdom. Some amplifiers such as the Bryston and Tigers do not fit this topology.
I always liked the Hafler XL power amp null test for those who claim there is an "unknown" distortion or design issue where one can use real source material, real loads, and hear or view the unmasked distortion(s), even types that have no name. Obviously, linear errors such as frequency response and phase shift result in the presence of non-distortion components in the subtraction but it is still worthwhile.
Comments please?
Pete B.
PB2 said:
I agree this should be an ultimate means of determining and characterizing deviations from the proverbial wire with gain. It seems though there is no much interest in pursuing the issue further, check here and here, and given the lack of interest, I will be doing some more research on the subject in connection with a very high performance project I am currently pursuing.
If results warrant, I will be posting details.
Rodolfo
PB2 said:
Hi Pete B.
Wellcome, to the club of the "null test fans" !
For me, is the only measure ,that coincide with "subjective " sound quality of amplifiers and is a invaluable toll in amplifier design.
See my post # 17 in
http://www.diyaudio.com/forums/showthread.php?s=&threadid=12752&perpage=10&pagenumber=2
And my post # 8 in
http://www.diyaudio.com/forums/showthread.php?s=&threadid=12752&perpage=10&pagenumber=2
Yes ...but the opposite is also true . If the null residual is only thermal noise (as in one of my amp design ) we have a amplifier that is the proverbial "wire with gain".
Cheers
Cortez said:
Hi Cortez
Sorry!!...The second link must be post #8 at
http://www.diyaudio.com/forums/showthread.php?s=&threadid=13415
Cheers
Cortez said:
Hi Cortez,
I don't think the papers are online, I got them from another member. I'll forward them if you email me with your email address as it seems you don't accept email from this site.
Hi Rodolfo,
I agree about the wire with gain, but I do prefer some filtering just to eliminate out of band garbage.
Hi Jorge,
Oh I didn't know there was a club, ... thanks! Yes it's possible to minimize freq response and phase errors as Hafler did in the XL by peaking up the ultrasonic response to improve the inband response. Did you use another method? Is your amp, bipolar, FET, tube, or?
Pete
PB2 said:
Hi Peter.
No I don't use any form of peaking . For a good null you must use the maximum overall feedback that you can .
The more the feedback the better the null.
All bipolar DC coupled.
No way to make a tube amplifier with a good null!!...To much phase shift components,as capacitors in the signal path and the output transformer...
some words to support null test. Seems the best technique – to find the objective parameters of well designed amplifier. Accuracy which is enough for all purpose and with arbitrary signal - not a close loop trajectory as in harmonic analyses (very important). I've made a successful improvement of such a measurement and even with 16 bit onboard ADC high accuracy level (not more need
of transfer characteristic measurement is reachable. It is on my sight http://amator2001.nm.ru/Voltage_measurement/measurement.htm
( in Russian yet though
Even better result will be with the delay line or programmable delay of input signal, but not for all cases it is need.
This was about HOW and WHAT to measure - my opinion is a GedLee approach
http://www.gedlee.com/distortion_perception.htm
Reducing a thermal distortions is also possible - http://amator2001.nm.ru/Voltage_Amp_pII/Voltage_Amp_pII.htm - is a corrected Hawksford correction.
of transfer characteristic measurement is reachable. It is on my sight http://amator2001.nm.ru/Voltage_measurement/measurement.htm( in Russian yet though
Even better result will be with the delay line or programmable delay of input signal, but not for all cases it is need.
This was about HOW and WHAT to measure - my opinion is a GedLee approach
http://www.gedlee.com/distortion_perception.htm
Reducing a thermal distortions is also possible - http://amator2001.nm.ru/Voltage_Amp_pII/Voltage_Amp_pII.htm - is a corrected Hawksford correction
.
PB2 said:
This is going to be a bit awkward, but here goes. This actually started out as an email conversation between Pete and myself, and continued for a bit before he posted it here. Below, I'll paste in an edited version of my original reply to him, and I've asked him to paste in an edited version of his reply to that afterwards. Anyway, here was my original reply to Pete...
Hi Pete,
To tell the truth, I didn't even notice that the non-inverting op-amp case wasn't covered by Cordell. I was mainly interested in looking at his power amp example showing the effect of feedback on PIM. It does seem odd that the non-inverting case wasn't covered. It makes me wonder whether the additional distortion due to the non-zero common-mode input voltage shows up as an AM-to-AM or AM-to-PM type of distortion.
It also appears that Otala's assumptions are fundamentally flawed. It's suspicious that he expresses the nonlinearity as a product, when in reality what occurs with memoryless non-linearities apporximates function composition. IOW, given two memoryless nonlinearities f(x) and g(x) in cascade that don't interact with each other, the combined nonlinearity is g(f(x)), not some kind of product. He also makes use of the multi-variable Laplace transform as a transfer function F(x, s). But for this to be valid, the input-output description in the time domain involving x and t must be a linear partial differential equation. I don't think it can be expressed in that form. Also, in computing F(x, s) he seems to be assuming that some kind of commutative property of nonlinear transfer functions applies. But nonlinear input-output descriptions of cascaded networks do not commute. As an example, the behavior of the circuit changes markedly depending on where the low-pass function is. His model apparently neglects this. There are so many holes in his analysis that I'm not surprised it never became a full-up paper. And as you mentioned, frequency-dependent nonlinearities aren't included either. But in all fairness, to treat that case requires Volterra series analysis, which is a big mess. So I'm not surprised he didn't include it. Doing otherwise is mathematically pretty intractable.
I found the Gilbert article to be worth spending some time on. There are some typos and unexplained notation that makes it a bit difficult at first, but after wading through it I felt I understood the problem much better than before I started. What was kind of funny about all this was that the 3rd harmonic distortion ended up being a much more sensitive measure of the PIM problem than the phase change with output level. There's no "hidden distortion" in this scenario.
Pete?
PB2 said:
You may notice from my
post # 51 in the "New distortion measurement method for audio amplifiers" started by Pavel, that I advocate for taking full benefit of readily available processing resources like 24 bit/ 96 KHz high quality inexpensive hardware, high performance software like SpectraLab and math processing software like MathCad.
Putting all this toghether is no trivial task either, but I hope the end result is a more accurate characterization of audio amplifier chain perturbations and how they correlate with the listening experience.
The time waterfall spectral plot in particular is very good at providing a visualy intuitive presentation, with the capability to adjust frequency / time resolution tradeoffs to match even on a dynamic basis how the ear - brain system process stimulus to convey the listening experience.
Certainly all types of linear phase - frequency compensations can be easily (in concept, it is all hard work of course) included in dataset preprocessing.
On another subject, I have the Hawksford and Cordell papers, but not the (controversial) Otala ones, I should appreciate if you can send or link them to me, my e-mail is ingrast@adinet.com.uy.
Rodolfo
2 Cortez
what is present online-
about Otala approach:
http://home.online.no/~tsandstr/AudioStartsHere.htm
and about Hafler null test (p. 7):
http://www.hafler.com/techsupport/pdf/XL-280_amp_man.pdf
may be some other resources available?
what is present online-
about Otala approach:
http://home.online.no/~tsandstr/AudioStartsHere.htm
and about Hafler null test (p. 7):
http://www.hafler.com/techsupport/pdf/XL-280_amp_man.pdf
may be some other resources available?
Hi Andy thanks for joining in!
We did a lab in college showing some of the non-ideal characteristics of OP amps. We measured slew rate and power bandwidth by the first sign of distortion and triangle method and were told to take note of the difference between inverting and non-inverting. We looked at the 741, 748, 351 and others. There was also a popular OP amp that "snapped" to the other rail on clipping, can't remember which one that is? Made a lasting impression contrasting inverting and non-inverting cases.
I first read about the issues with non-zero common-mode inputs in an article by Eric F. Taylor Wireless World, 1977. Edit: I mentioned the August issue in the first post here, there was a second part in the September issue. Taylor states that the varying Vcb with a common mode signal the diff amp input impedance is non-linear. I believe that this is due to the non-linear Miller capacitance.
Edit: We should note that this means the performance of the amp will depend on the driving source impedance.
Just a quick thought about your AM-to-PM question, wouldn't it follow than any amplitude non-linearity (really transconductance non-linearity) turns into AM-to-PM in any Cdom compensated circuit since the integrator is a frequency dependent component? Edit: I think of this as an integrator where the resistor (diff amp transconductance) is amplitude, frequency, and/or time dependent.
I also notice that many such as Self assume a Cdom compensated circuit in their discussions, and I don't believe that the same conclusions are reached when Cdom is not used. I understand that it does apply, obviously, to most OP amps.
I understand your comments about Otala, and agree. Edit: I believe there is a subset of non-linearities that can be cascaded as g(x)*f(x) and agree that the general case is g(f(x)). I did not dig that deep or put a lot of thought into this issue.
My point was that when simplifying assumptions are made it's important to verify how much they deviate from what is seen in real circuits, but I also see your point that the theoretical analysis is much more complex. Have you seen Otala's 1973 IEEE Transactions paper?
The Gilbert paper looks good I'm just skimming it. However, this jumped out at me (pg 2, 3rd paragraph): "Because the dominant source of nonlinearity is in the input cell, the distortion will be lowest in the voltage-follower mode." Shouldn't this be: unity gain inverting amplifier here, and not voltage-follower mode? Edit: Perhaps he was only considering non-inverting? Gilbert's analysis is familiar to me, it brings back memories from semiconductor classes.
Pete B
We did a lab in college showing some of the non-ideal characteristics of OP amps. We measured slew rate and power bandwidth by the first sign of distortion and triangle method and were told to take note of the difference between inverting and non-inverting. We looked at the 741, 748, 351 and others. There was also a popular OP amp that "snapped" to the other rail on clipping, can't remember which one that is? Made a lasting impression contrasting inverting and non-inverting cases.
I first read about the issues with non-zero common-mode inputs in an article by Eric F. Taylor Wireless World, 1977. Edit: I mentioned the August issue in the first post here, there was a second part in the September issue. Taylor states that the varying Vcb with a common mode signal the diff amp input impedance is non-linear. I believe that this is due to the non-linear Miller capacitance.
Edit: We should note that this means the performance of the amp will depend on the driving source impedance.
Just a quick thought about your AM-to-PM question, wouldn't it follow than any amplitude non-linearity (really transconductance non-linearity) turns into AM-to-PM in any Cdom compensated circuit since the integrator is a frequency dependent component? Edit: I think of this as an integrator where the resistor (diff amp transconductance) is amplitude, frequency, and/or time dependent.
I also notice that many such as Self assume a Cdom compensated circuit in their discussions, and I don't believe that the same conclusions are reached when Cdom is not used. I understand that it does apply, obviously, to most OP amps.
I understand your comments about Otala, and agree. Edit: I believe there is a subset of non-linearities that can be cascaded as g(x)*f(x) and agree that the general case is g(f(x)). I did not dig that deep or put a lot of thought into this issue.
My point was that when simplifying assumptions are made it's important to verify how much they deviate from what is seen in real circuits, but I also see your point that the theoretical analysis is much more complex. Have you seen Otala's 1973 IEEE Transactions paper?
The Gilbert paper looks good I'm just skimming it. However, this jumped out at me (pg 2, 3rd paragraph): "Because the dominant source of nonlinearity is in the input cell, the distortion will be lowest in the voltage-follower mode." Shouldn't this be: unity gain inverting amplifier here, and not voltage-follower mode? Edit: Perhaps he was only considering non-inverting? Gilbert's analysis is familiar to me, it brings back memories from semiconductor classes.
Pete B
mikeks said:
Hi Mikeks,
Good to see you've joined in! Don't know if your familiar with the XL amps but the peaking was around 50 kHz IIRC and did not continue into the RF range. I don't like it but it worked. I'd probably null to prove that the distortion was low then remove the peaking since it's a minor change. Or just null to investigate a prototype, then remove the peaking.
Tube_Dude said:
I don't think that was the case Jorge, let's take the Leach amp as a typical example, if we look at the phase response it can be seen that even though the freq response is flat far above 20 kHz there is some phase shift in the audio range, not much but enough to reduce the depth of the null for frequencies above about 10 kHz. See Fig 3 at this page:
http://users.ece.gatech.edu/~mleach/lowtim/bckgrnd.html
Peaking up the amplitude response up around 370 kHz flattens the phase response within the audio band. The null will be worse up around 370 kHz but who cares what it does up there.
I pulled out the Audio Critic brief review of the XL-280, Issue 11 - 1988, where they noted and were very critical of the 3 to 7 dB peak (depending on trimmer setting) at around 370 kHz (OK - I was wrong above where I said 50 k from memory). I initially found "The Audio Critic" to be logical and scientific but later noticed that they (mainly Peter Aczel) seem to insist on certain requirements for ultimate sound reproduction that are not universally held. Anyway, they point out that this peaking causes ringing on square waves and therefore it cannot be right as far as a straight wire with gain - my interpretation. Let me point out that the ringing is probably near the peak frequency certainly over 100 kHz and I doubt there is any real world audio signal with fast enough rise time to produce ringing. I say so what - if it produces a better null within the audio band then so what? Of course it is not perfect, but it's performance within the audio band is "better". As I said I don't like it, I'd probably use it to unmask non-linear distortions then remove this linear change (or make it switchable) when finalizing the design. I believe "The Audio Critic" was nit picking and missing the point, since the amp produces a deep null it has no significant nonlinearities in the audio band or when used with audio signals. This is absolute proof of performance and it is significant.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.