Valery,
The ratio of 1M and 27k gives 37, around 31dB. Most SS amps with global fb have at least 60dB, so this fb snippet from Guillermo reveals we have reduced the 1st and 2nd stage distortion and reduced the Zout, permitting us to drive the output stage with more accuracy even if the lower global fb around the entire amp brings the OPS distortion higher. You eloquently explained this just now, so I'm preaching to the choir.
Jan,
The issues are not to do with the coarse THD figures. I deplore THD, it is not telling us what we need to know. I examine the harmonics, study the FFT. We can do this easily in LTSpice, a very useful design tool, and in practice with a good spectrum analyser as you know. You find that the nested fb regime does change the harmonic profile and in fact increases the H2-H5 levels whilst reducing the H7 and beyond. This is where the THD increases - the profile is changed. This has to sound better, even if the THD is higher because H2+H3+H4 is higher. It turns out that the first three harmonics are not machine tones, they are actually musical, so this is not a bad thing...... NP and Hansen understand all this, I'm quite sure of it.
The tube guys have it right. They know that THD for tube amps is absurd; it does not tell us what sounds better. Pete Millett in his clever designs examines most of the harmonics for his amps, and it is informative. He proposes large pentodes as voltage amplifiers, unconventional! THD 1% for a 300B SET and yet people drool over the sound?? WOW! They know that the harmonics are important, and in fact most tube amps are producing H2 at -45dB down - huge THD, yet the SS guys gloss over this. They regard the 'good sonics' as euphonic nonsense, dribble from the audiophools who don't know the engineering, yet people continue to buy these amps even in the digital era. Are they all fools? Can they all be ignored because they do not understand the solid math of a Taylor Series and feedback theory? Hmmm, perhaps engineering lower and lower THD figures is chasing rainbows!
Finally, what is one of great benefits of a tube amp? I would say it is the depth of image, this brings us closer to reality. Close your eyes and you can place the instruments in space, and when you hear clapping you can assess the size AND DEPTH of the crowd. Can global feedback SS amps do this? That is the question...... depth is difficult, but width is easy.
Hugh
The ratio of 1M and 27k gives 37, around 31dB. Most SS amps with global fb have at least 60dB, so this fb snippet from Guillermo reveals we have reduced the 1st and 2nd stage distortion and reduced the Zout, permitting us to drive the output stage with more accuracy even if the lower global fb around the entire amp brings the OPS distortion higher. You eloquently explained this just now, so I'm preaching to the choir.
Jan,
The issues are not to do with the coarse THD figures. I deplore THD, it is not telling us what we need to know. I examine the harmonics, study the FFT. We can do this easily in LTSpice, a very useful design tool, and in practice with a good spectrum analyser as you know. You find that the nested fb regime does change the harmonic profile and in fact increases the H2-H5 levels whilst reducing the H7 and beyond. This is where the THD increases - the profile is changed. This has to sound better, even if the THD is higher because H2+H3+H4 is higher. It turns out that the first three harmonics are not machine tones, they are actually musical, so this is not a bad thing...... NP and Hansen understand all this, I'm quite sure of it.
The tube guys have it right. They know that THD for tube amps is absurd; it does not tell us what sounds better. Pete Millett in his clever designs examines most of the harmonics for his amps, and it is informative. He proposes large pentodes as voltage amplifiers, unconventional! THD 1% for a 300B SET and yet people drool over the sound?? WOW! They know that the harmonics are important, and in fact most tube amps are producing H2 at -45dB down - huge THD, yet the SS guys gloss over this. They regard the 'good sonics' as euphonic nonsense, dribble from the audiophools who don't know the engineering, yet people continue to buy these amps even in the digital era. Are they all fools? Can they all be ignored because they do not understand the solid math of a Taylor Series and feedback theory? Hmmm, perhaps engineering lower and lower THD figures is chasing rainbows!
Finally, what is one of great benefits of a tube amp? I would say it is the depth of image, this brings us closer to reality. Close your eyes and you can place the instruments in space, and when you hear clapping you can assess the size AND DEPTH of the crowd. Can global feedback SS amps do this? That is the question...... depth is difficult, but width is easy.
Hugh
Last edited:
Very clever, Guillermo!
When you introduce nested fb with a LTP input stage, you skew the output offset and it's difficult to correct it. A small ceramic cap in series with the nested fb is effective, and I would suggest 10nF. This is NOT a phase lead cap, however. It's a DC blocker, nothing more. A phase lead, around 10pF, would be across the 1M resistor, if needed. This approach restores the offset at the output node, keeps the LTP nice and balanced, yet introduces the nested fb. Significantly, at very low frequency the nested 'cap' does not pass much AC current and the amp operates at full global feedback, very nice for driving bass signals. The cap comes in at 1khz and higher, delivering the benefits.......
BTW, what is your idea of using 47k shunt from VAS collector down to ground? I think I know what you want here, but not quite sure. Valery has suggested this too, I know what it does, is this your intention?
Hugh
When you introduce nested fb with a LTP input stage, you skew the output offset and it's difficult to correct it. A small ceramic cap in series with the nested fb is effective, and I would suggest 10nF. This is NOT a phase lead cap, however. It's a DC blocker, nothing more. A phase lead, around 10pF, would be across the 1M resistor, if needed. This approach restores the offset at the output node, keeps the LTP nice and balanced, yet introduces the nested fb. Significantly, at very low frequency the nested 'cap' does not pass much AC current and the amp operates at full global feedback, very nice for driving bass signals. The cap comes in at 1khz and higher, delivering the benefits.......
BTW, what is your idea of using 47k shunt from VAS collector down to ground? I think I know what you want here, but not quite sure. Valery has suggested this too, I know what it does, is this your intention?
Hugh
Last edited:
Hugh, doesn't that mean that the ratio between the feedback from the two loops is 31dB, rather than the total feedback being 31dB?
The total feedback is the ratio of the // value of the two values (1M & 27k IIRC) to the value of the R to ground (1k) so the total feedback is something like 28.4 dB.
Jan
Still thinking about this issue or harmonic profiling so to say. If you could make a Vas stage with a very low Zout, the non-linearities of the Vas load ((pre)driver, output stage) would not lead to distortions on the Vas signal.
That would be the ideal situation.
Thus the remaining engineering problem is to build a Vas with low Zout to make it insensitive to what comes after it.
Question is, what is the best method. Putting an R to ground at its output does do that, but seems to me a brute force and not well thought out way to solve it.
You also could just place a cap from the Vas output to ground. Make the impedance the same as the R you would use at say 20kHz and the effect is the same.
Let's take an example: suppose you would use a 10k R as Vas load. To get this 10k at 20kHz from a cap needs 1/(2*pi*20kHz*10k) = about 650pF.
So far so good.
BUT at 2kHz (where you don't need the // R because the output stage is much more linear), that cap represents a load of not 10k but 100k, so all of a sudden the amp has 20dB more gain to reduce non-linearities with.
Sounds like having your cake and eat it too?
Comments?
Edit: Then again, putting a lowly emitter- or source follower after the Vas should fix it too. Wideband, no extra issues with stability and improving linearity a 100-fold.
Jan
That would be the ideal situation.
Thus the remaining engineering problem is to build a Vas with low Zout to make it insensitive to what comes after it.
Question is, what is the best method. Putting an R to ground at its output does do that, but seems to me a brute force and not well thought out way to solve it.
You also could just place a cap from the Vas output to ground. Make the impedance the same as the R you would use at say 20kHz and the effect is the same.
Let's take an example: suppose you would use a 10k R as Vas load. To get this 10k at 20kHz from a cap needs 1/(2*pi*20kHz*10k) = about 650pF.
So far so good.
BUT at 2kHz (where you don't need the // R because the output stage is much more linear), that cap represents a load of not 10k but 100k, so all of a sudden the amp has 20dB more gain to reduce non-linearities with.
Sounds like having your cake and eat it too?
Comments?
Edit: Then again, putting a lowly emitter- or source follower after the Vas should fix it too. Wideband, no extra issues with stability and improving linearity a 100-fold.
Jan
Last edited:
This is something I have noticed in sims. The effect is present whether you have plain resistive shunts or nested feedback from the VAS outputs.
Been thinking that there is no reason to favour resistive shunts over returning the resistors back to the LTP negative input. You get two benefits this way. The first is you still load the VAS output to GND via the feedback shunt resistor. The second is that you get nested feedback for the IPS. This is getting dangerously close to MIC compensation?
From what I have read, you actually want the resistors taken from the VAS output and not from the (pre-)drivers. This was a bad idea from my part. VAS loading is a pre-requisite for this to work. Unlike normal compensation where
taking the pick off point to the next stage after the VAS can be an advantage.
Is my understanding ok?
Paul
Hi Jan,
Yes, that is what I mean, the gain (OLG) is very close to 31dB but it is not the same. We have agreed the series fb resistor is 1k754 (taking account of the nfb resistor of 68k), and the shunt is 68R. Therefore the gain is (1.754+0.068)/0.068 = OLG =26.8 ie 28.56dB. Now, when we examine the comparison between global fb and nested fb, we have (1.8+0.068)/0.068 for global gain, and (68+ 0.068)/0.068. Clearly the ratio drops out the 0.068 shunt fb resistor (and the shunt cap) so the ratio of global v. nested is (68+0.068)/(1.8+0.068) = 36.44. If we ignore the shunt resistor, as it is small, we introduce an error (3.67% in fact) and if we use 68k/1k8 for 37.77, we are a little higher.
Consider this in a conventional amp: Normally we have a very high OLG from the input+VAS stage of around 90dB. If we set CLG at 28.5dB (26.7), then the usual figure of global feedback (loop gain) is (90 - 28.5) = 61.5dB, a usual figure at 1KHz. If, however, we impose some nested feedback with a 68k/1k8/68R network, we reduce the OLG open to the entire amp, though it is available to the 1st and 2nd stage, and therefore delivers a superior performance, fidelity, THD and output impedance, to the output stage. But the negative feedback available to reduce the output stage is now much reduced; the factor of ratio of the nested ratio to the CLG.
Figuring out these relationships if you ignore the output stage decompose to the standard relationships of any amp with feedback since the output stage has approximately unity. I always meant this, but was trying to cut down the math. But we agree and are not talking at cross purposes.
Yes, Paul, I believe we agree. If we set VAS collector with resistive loading to ground, we reduce the loop gain. This worsens THD, but has some benefits, although driving impedance of the drivers is probably not one of them. Correspondingly, if we set lower loop gain by loading the collector VAS back to the fb node, we can achieve the same thing, but now we have the benefit of lower THD at the output of the VAS, and lower Zout to drive the output stage. Equally, the loop gain could be the same. Question: Which sounds the best - without regard to the total THD amp figure? I will leave that to you.......
Hugh
Yes, that is what I mean, the gain (OLG) is very close to 31dB but it is not the same. We have agreed the series fb resistor is 1k754 (taking account of the nfb resistor of 68k), and the shunt is 68R. Therefore the gain is (1.754+0.068)/0.068 = OLG =26.8 ie 28.56dB. Now, when we examine the comparison between global fb and nested fb, we have (1.8+0.068)/0.068 for global gain, and (68+ 0.068)/0.068. Clearly the ratio drops out the 0.068 shunt fb resistor (and the shunt cap) so the ratio of global v. nested is (68+0.068)/(1.8+0.068) = 36.44. If we ignore the shunt resistor, as it is small, we introduce an error (3.67% in fact) and if we use 68k/1k8 for 37.77, we are a little higher.
Consider this in a conventional amp: Normally we have a very high OLG from the input+VAS stage of around 90dB. If we set CLG at 28.5dB (26.7), then the usual figure of global feedback (loop gain) is (90 - 28.5) = 61.5dB, a usual figure at 1KHz. If, however, we impose some nested feedback with a 68k/1k8/68R network, we reduce the OLG open to the entire amp, though it is available to the 1st and 2nd stage, and therefore delivers a superior performance, fidelity, THD and output impedance, to the output stage. But the negative feedback available to reduce the output stage is now much reduced; the factor of ratio of the nested ratio to the CLG.
Figuring out these relationships if you ignore the output stage decompose to the standard relationships of any amp with feedback since the output stage has approximately unity. I always meant this, but was trying to cut down the math. But we agree and are not talking at cross purposes.
Yes, Paul, I believe we agree. If we set VAS collector with resistive loading to ground, we reduce the loop gain. This worsens THD, but has some benefits, although driving impedance of the drivers is probably not one of them. Correspondingly, if we set lower loop gain by loading the collector VAS back to the fb node, we can achieve the same thing, but now we have the benefit of lower THD at the output of the VAS, and lower Zout to drive the output stage. Equally, the loop gain could be the same. Question: Which sounds the best - without regard to the total THD amp figure? I will leave that to you.......
Hugh
Hi Hugh,
This I shall discover soon. Fortunately, my latest amp's PCB is ideally laid out, by chance, to enable easy modification between the two options. My technical mind tells me that returning the resistors back to the FB node is best but reality may be different...
Another variable in all this is LTP degeneration. One of my observations is that THD is lower is you reduce the loop gain via loading of the VAS output rather than use excessive LTP degeneration. This is obviously done while keeping loop gains around the OPS equivalent to make it a fairer test. Always wondered why and now it is beginning to make some sense. Think I am observing the benefits of the lower VAS Zout etc?
Paul
Yes, I fully agree with the calculation & method and such.
I would however question the desirability of allocating more loop gain to the 1st and 2nd stage to the detriment of the output stage. As see it, it is the output stage that is the limiting factor of the overall feedback so I would rather allocate more loop gain to the output stage.
Jan
Yes, Jan, I know you don't agree; but put it this way: You are so sure that only your normal, zero THD paradigm is correct that you would even travel to the other side of the planet and reject utterly even LISTENING to this proposition!! Is that not a completely closed mind on the matter?
You and many thousands of good engineers all over the world agree with you. But most of you would not even consider it at all - this is the reason the present technology is standing still after sixty years.... and why there are so many polarised views on this and other many other issues. Once you consider the psychoacoustics and musical perception, you will realise that the first three low order harmonics have very little difference to the musical perception, and this is why tube amps sound so good regardless of their dreadful THD figures. This is an intellectual paradigm, like the trick of two faces facing each other, or, a second later, the silhouette of a Grecian urn!
Cheers,
Hugh
You and many thousands of good engineers all over the world agree with you. But most of you would not even consider it at all - this is the reason the present technology is standing still after sixty years.... and why there are so many polarised views on this and other many other issues. Once you consider the psychoacoustics and musical perception, you will realise that the first three low order harmonics have very little difference to the musical perception, and this is why tube amps sound so good regardless of their dreadful THD figures. This is an intellectual paradigm, like the trick of two faces facing each other, or, a second later, the silhouette of a Grecian urn!
Cheers,
Hugh
Last edited:
Very interesting discussion! Let me think a bit more with regards to the global/nested loop gains proportion ...
Guillermo - your calculation is right, it's just 1M looks very high at a first glance - my bad 🙄
Guillermo - your calculation is right, it's just 1M looks very high at a first glance - my bad 🙄
Gee, Valery, I've got you thinking now........
I am sure you will figure this out very quickly! Actually, you are 75% there already.
Hugh
I am sure you will figure this out very quickly! Actually, you are 75% there already.
Hugh
Very nice to do feedback on the VAS, but it has a limited drive capability.The feedback resistance snoops away a part of the drive for the output stage and is an extra load that make the gain of the VAS smaller allready.
Mona
Mona
True, but would you agree that the two bases of the complementary drivers load the VAS too? And a complex load too, rather nastier than a single resistor, and besides, you could load it with 1Megohm, as GEIrin has shown us........
Hugh
Hugh
Really? Have I ever 'utterly rejected even LISTENING to this proposition!!'? I spend considerable funds and time to take the trouble to visit you, because I respect you as an audio designer and wanted to meet you. I enjoyed our chats and lunch, but I was not expecting that this would be used as proof I have a closed mind. I normally don't use private conversations as a means to try to discredit my visitors.
We are discussing it right now, are we not? We both bring up arguments pro and con, are we not? Or is your idea of a closed mind anyone who has a difference of opinion with you?
Jan
Yes, we are discussing it, certainly, but in your recent post you stated that you find it very difficult to agree, giving me the impression this was the end of your discussion, to wit:
BTW, I do not agree with your assertion here. The primary influence on the sound quality of a well adjusted AB SS power amp is the VAS, and then second, the input stage. I included you in the majority of my engineer friends, most of whom humour me with these fantastical notions, polite fellows that they are. I do not insult you, but you need to remember that you never gave me at the time any interest in any of my amps and I assumed therefore that you did NOT regard me with any regard. In light of this your comment that you respect me as an audio designer surprises me and might actually be ingenuous! You preferred over lunch to discuss linearity, and THD, and your interesting mag and the people you have been in contact with, as recall, an interest of Paul's as well. That's fine, I still offered you my hospitality, so please do not arc up months, even years, after the meeting.
I would be happy to discuss this offline if you'd like to. Finally, after four years of relearning my language, I feel I can put my ideas in clear English, but this has taken a long time.
Cheers,
Hugh
BTW, I do not agree with your assertion here. The primary influence on the sound quality of a well adjusted AB SS power amp is the VAS, and then second, the input stage. I included you in the majority of my engineer friends, most of whom humour me with these fantastical notions, polite fellows that they are. I do not insult you, but you need to remember that you never gave me at the time any interest in any of my amps and I assumed therefore that you did NOT regard me with any regard. In light of this your comment that you respect me as an audio designer surprises me and might actually be ingenuous! You preferred over lunch to discuss linearity, and THD, and your interesting mag and the people you have been in contact with, as recall, an interest of Paul's as well. That's fine, I still offered you my hospitality, so please do not arc up months, even years, after the meeting.
I would be happy to discuss this offline if you'd like to. Finally, after four years of relearning my language, I feel I can put my ideas in clear English, but this has taken a long time.
Cheers,
Hugh
Last edited:
Wow...
You really need to make a technical difference of opinion personal don't you?
Please note that it was you that started to drag in our personal meeting.
You really need to make a technical difference of opinion personal don't you?
Please note that it was you that started to drag in our personal meeting.
Beg to differ. The Vas is much more linear than the output stage.
It can easily be checked.
As the feedback loop is closed so the output gets vanishingly low distorted *something* must compensate for that. IOW, the non-linear element gets driven by a 'pre-distorted' waveform so that the output, going through the non-linear element, is undistorted.
If you hang a distortion analyzer on internal nodes of the amp, be it the output of the diff pair or the output of the Vas, you will see a (heavily) distorted waveform; that's what is required to make the output linear.
By looking at the relative distortion at those points it is easy to see which part of the amp is the most distorting. A very revealing tests, and often the internal distortion is large enough to see on a scope if you don't have a disto analyzer handy.
Jan