Re: andy_c...
mikek,
Thanks for that link. It ended up pointing to this schematic:
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
which looks promising. I'm still scratching my head over some of the aspects of the design of that amp.
mikek said:
mikek,
Thanks for that link. It ended up pointing to this schematic:
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
which looks promising. I'm still scratching my head over some of the aspects of the design of that amp.
>>>"What is the point in tying to cascode the VAS stage anyway?"
In Slone's case, it seems to me that it is necessary in order to implement his soft clip circuit. He biases the inner transistors with a voltage diovider that sets the bias at 1/2 the difference between the output signal boltage and the rails. You can see this in the "'OPTI-MOS" design posted on www.sealelectronics.com. You may also notice that in this design he, too, abondonds the current mirrors.
I, too, tried modeling something like the original poster (used SIMetrix) and find it troublesome. My conclusion so far is that it doesn't provide enough current to the dominate pole capacitors at higher frequencies and reducing the resistor values beyond some point results in IS oscillations.
Andy C may be on the right treack as it looks like something is needed to get greater current out of the IS.
These are only my very amatuer guesses.
It should be noted that a number of people here have built the OPT-MOS design without repoting any problems. I don't think the design *exactly* as posted was ever presented as a complete amp in any of Slones books so it may be going a bit far to dump on Slone for a design didn't propose.
As a further note, the OPT-MOS design seems to be going into commercial production (ZUS Electronics) and Slone is wrapping up his on-line business. He says he may continue to sell the remaining OPTI-MOS pcbs but that's abot it.
In Slone's case, it seems to me that it is necessary in order to implement his soft clip circuit. He biases the inner transistors with a voltage diovider that sets the bias at 1/2 the difference between the output signal boltage and the rails. You can see this in the "'OPTI-MOS" design posted on www.sealelectronics.com. You may also notice that in this design he, too, abondonds the current mirrors.
I, too, tried modeling something like the original poster (used SIMetrix) and find it troublesome. My conclusion so far is that it doesn't provide enough current to the dominate pole capacitors at higher frequencies and reducing the resistor values beyond some point results in IS oscillations.
Andy C may be on the right treack as it looks like something is needed to get greater current out of the IS.
These are only my very amatuer guesses.
It should be noted that a number of people here have built the OPT-MOS design without repoting any problems. I don't think the design *exactly* as posted was ever presented as a complete amp in any of Slones books so it may be going a bit far to dump on Slone for a design didn't propose.
As a further note, the OPT-MOS design seems to be going into commercial production (ZUS Electronics) and Slone is wrapping up his on-line business. He says he may continue to sell the remaining OPTI-MOS pcbs but that's abot it.
andy_c said:
Oh, that one. Yes, I see it now. I didn't remember he ever
used buffering between input and VA stages in any of the
designs, although he discusses the possibility.
Well, I don't have time to see what he writes about it now, since
I really ought to get my suitcases packed to go away for
a week. I was just so puzzled since I didn't remeber him
using this design, that I had to ask you right away.
To Andy C, et al,
I just opened the Slone book to Fig 11.12, page 351. My vision may be clouded with age, but I think that the cicuit you presented to start this thread is similar but NOT exactly the same as Slone's. I suggest you run a simulation on the exact circuit first, then make one mod at a time.
I have tried the same thing: take a piece from one example that I though looked good and tried to match it up with another piece withth idea of getting the best of everything; then I find they just don't work together! Very frustrating. Both Self's and Slone's books make it seem like designing a unique amp cicuit should be possible with a little cut and paste plus a SPICE simulator. In fact I'm comming to the realization that it's actually damn hard. I get the feeling that both of these guys have a lot of experience in their heads that doesn't come out in the books. Probably can't. It may be the only way to get from here to there is to be willing to try a lot of things and blow up a lot of transistors. (Done enough of that that my wife made me buy a CO2 fire extinguisher to keep by the bench.)
I just opened the Slone book to Fig 11.12, page 351. My vision may be clouded with age, but I think that the cicuit you presented to start this thread is similar but NOT exactly the same as Slone's. I suggest you run a simulation on the exact circuit first, then make one mod at a time.
I have tried the same thing: take a piece from one example that I though looked good and tried to match it up with another piece withth idea of getting the best of everything; then I find they just don't work together! Very frustrating. Both Self's and Slone's books make it seem like designing a unique amp cicuit should be possible with a little cut and paste plus a SPICE simulator. In fact I'm comming to the realization that it's actually damn hard. I get the feeling that both of these guys have a lot of experience in their heads that doesn't come out in the books. Probably can't. It may be the only way to get from here to there is to be willing to try a lot of things and blow up a lot of transistors. (Done enough of that that my wife made me buy a CO2 fire extinguisher to keep by the bench.)
sam9 said:
Right. I believe I mentioned that I made some superficial modifications (in the current sources for the input diff amp for example) to avoid possible copyright issues. The current values provided by these sources are very nearly equal to those in the book (one Vbe or diode drop divided by 120 Ohms). What I did capture was the essentials of the circuit required to analyze the bias stability of the VAS. That is, complementary diff amps with current mirror output feeding a VAS consisting of an emitter follower feeding a cascoded common emitter amp. I used single-pole compensation instead of the two-pole approach he uses. But this doesn't affect the VAS bias stability (or lack of same), which was the motivation for the original post.
traderbam said:
That's what I ended up doing. I'd like to figure out how to get it to work right with the current mirror, but I'm out of ideas at present.
Well, I think my explanation was somewhat flawed. I'm not a transistor physics guy, so maybe someone who is can jump in and explain it better. The way I look at it is to visualize the curve tracer display of common emitter vs. common base transistor characteristics. The latter are flat as a pancake, yet the emitter current, and therefore the base current, is changing at each step in the family of curves. The common emitter characteristics have varying slopes of Ic vs Vce at each Ib step, and they "curl up" for large Vce, almost like a very slow onset of breakdown. So it's clear to me by simple lab observation that the output impedance of a common base amp is both higher and more constant than a common emitter amp. But I do admit that I don't have a rigorous transistor physics explanation.
But it does not always follow that reducing the number of transistors reduces nonlinearity. I was once asked to design an RF limiting amplifer which, at a certain number of dB below the absolute limit level, was required to have third-order intermods less than a certain amount. I used a diff amp for this. I figured out how much emitter degeneration was required. My boss insisted that I replace the current source with a simple resistor. The intermod performance of the circuit was horrible, nowhere near as good as the calculated value. As soon as I used a current source, the intermods met spec, and were within a few tenths of a dB of the calculated values. I realized then that any AC tail current in the diff amp will cause it to act as a Gilbert cell mixer. It mixes its tail current with its difference mode voltage. So here was a case where increasing the transistor count reduced the distortion - by about 20 dB in this case. This was about twenty years ago, so I can't recall the exact details. But there are many other examples. The current mirror at the diff amp output has been shown by Self to reduce distortion. That's why I was interested in preserving it.
" I believe I mentioned that I made some superficial modifications (in the current sources for the input diff amp for example) to avoid possible copyright issues."
As I said, I've been down the that road and discovered that what I thought were "superficial modifications" were not so superficial after all. At least in the sense that Slone's circuit worked and mine was crap.
Part of my post was also prompted by another post that expressed some negative vibes about Slone publishing unworkable circuit's. While Slone can take care of himself, I wouldn't want a reader to avoid building one of Slone's amps just because of that one statement which was based on a mod of Slones circuit rather than the original.
Not long ago I took one of Slones circuits and put in a SPICE model, then more or less systematically varied the component values. There seem to be two categories of values: those that make little if any difference over a broad range and those where the slightest change degrads the simulated performance. I suspect Slone spent a lot of time with a simulator to fine tune the final published circuit.
As I said, I've been down the that road and discovered that what I thought were "superficial modifications" were not so superficial after all. At least in the sense that Slone's circuit worked and mine was crap.
Part of my post was also prompted by another post that expressed some negative vibes about Slone publishing unworkable circuit's. While Slone can take care of himself, I wouldn't want a reader to avoid building one of Slone's amps just because of that one statement which was based on a mod of Slones circuit rather than the original.
Not long ago I took one of Slones circuits and put in a SPICE model, then more or less systematically varied the component values. There seem to be two categories of values: those that make little if any difference over a broad range and those where the slightest change degrads the simulated performance. I suspect Slone spent a lot of time with a simulator to fine tune the final published circuit.
Andy,
Just some food for thought.
I was a little unclear. I meant that each transistor in the signal path adds distortion. A CCS adds some distortion because its effective impedance is non-linear but the idea is that it is so much higher than a simple resistor that the net effect is to greatly reduce the non-linearity of the diff pair as you say. I'm not so sure the same can be said of a current mirror, where signals are amplified. Some care is needed as you know. A well designed diff amp has extremely low distortion when dynamic Ic changes are small and so the benefit of a current mirror in reducing common-mode error may be countered by the non-linearities of the current mirror itself.
Another thought is about the output Z of the VAS. It appears to me that the transistor's impedance will be swamped by the miller cap feedback loop and, in addition, will be in parallel with the impedance of the output stage and loudspeaker. Worrying about the "Early effect" may be guilding the lilly as it were and may not be a good reason, on its own, for cascoding the VAS transistor and introducing another semiconductor into the signal path.
Just some food for thought.
I was a little unclear. I meant that each transistor in the signal path adds distortion. A CCS adds some distortion because its effective impedance is non-linear but the idea is that it is so much higher than a simple resistor that the net effect is to greatly reduce the non-linearity of the diff pair as you say. I'm not so sure the same can be said of a current mirror, where signals are amplified. Some care is needed as you know. A well designed diff amp has extremely low distortion when dynamic Ic changes are small and so the benefit of a current mirror in reducing common-mode error may be countered by the non-linearities of the current mirror itself.
Another thought is about the output Z of the VAS. It appears to me that the transistor's impedance will be swamped by the miller cap feedback loop and, in addition, will be in parallel with the impedance of the output stage and loudspeaker. Worrying about the "Early effect" may be guilding the lilly as it were and may not be a good reason, on its own, for cascoding the VAS transistor and introducing another semiconductor into the signal path.
sam9 said:
For all but the output stage currents (because they're normally adjustable), and given a good design, it should be possible using nothing more than a pen, paper and hand calculator to compute with very good accuracy the bias currents and voltages of a power amplifier. SPICE is not needed for this! All that's needed is an engineering understanding of how the biasing works. You're making the assumption that I'm a complete idiot and don't understand any of this. I do understand quite well however. Overall power amp design is complex, but biasing is not rocket science. Any degreed EE should know how to do this.
I've looked at some of the designs he has on the web. The OptiMOS design seems fine for example. The only thing I take issue with is the use of current mirrors in the input differential amplifiers of a fully complementary design as is used in design 10. It results in the current of the VAS being essentially an uncontrolled parameter. This is inadequate design, plain and simple. To be more specific, this problem occurs in the following designs:
Design 10
Design 11
Design 12
All others are okay. The non-complementary designs have the VAS current set by a constant current load whose value is clearly and obviously well controlled. The good complementary designs achieve stable VAS current by avoiding the use of a current mirror in the input diff amp. They also have an emitter resistor in the VAS, whose voltage is established by the voltage drop across the diff amp collector resistors.
That's about all I can say without repeating myself even further.
traderbam said:
Ahh, now I see what you're saying. I should probably go back and reread Self's distortion analysis. If a diff amp with no emitter degeneration is assumed, it seems reasonable that any distortion in the current mirror will be more than offset by the reduction in distortion by balancing the diff amp, as Self claims. That tanh() charactistic is mighty nonlinear. But suppose a lot of emitter degeneration is used (as I plan to do), making the diff amp itself much more linear. It may be that the balancing of the diff amp by the current mirror in that case causes such a small improvement in distortion that the effect may be more than offset by the distortion added by the current mirror itself. In fact, if you look at the order in which he did things, he first added the current mirror to the undegenerated diff amp, improving its distortion. Then he added degeneration to make the gm what it was before the current mirror was added. But now that I think about it, I'm not sure whether he subsequently went back to evaluate the distortion of the degenerated diff amp with and without the mirror. Details, details!
At any rate, I've gone with the resistive loading, and even with greatly reduced feedback to get good stability with capacitive loads and no output inductor, I'm getting low enough distortion in the simulation that I'm happy with the results. As an added bonus, the circuit is much simpler than before - definitely a plus.
Thanks for your reply.
CLASS AB???????
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
"....actually its a high-slew design, (~300V/uS), in which ALL three stages operate in class AB.....you may need to get a-hold of the full article to fully appreciate the concept...infact i think it's a three- part paper."
Will someone tell me how the first diff pair biased at 5 mA per transistor can be biased Class AB? Better yet can someone actually look at the reference it see if this claim is made and the rational stated for such a classification. This appears to be a very dubious claim that I would not even dare to make without the article sitting in front of me. But then again I try not to drop such logic bombs without references or explanation and then run away......
Wireless World/Electronics World Giovanni Stochino
Ultra fast Audio amplifier. #1 Apr 1997, p278
Ultra fast Audio amplifier #2 Aug 1998, p633
Doug Self and Dr. Alex Megann describe it as being a two part article not a three part.
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_amp.html
http://www.dself.dsl.pipex.com/ampins/library/ampartew.htm
I can image the reaction if I made such a controversial claim without any supporting evidence ..........
Thanks,
Diogenes of Sinope
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_circuit.jpg
"....actually its a high-slew design, (~300V/uS), in which ALL three stages operate in class AB.....you may need to get a-hold of the full article to fully appreciate the concept...infact i think it's a three- part paper."
Will someone tell me how the first diff pair biased at 5 mA per transistor can be biased Class AB? Better yet can someone actually look at the reference it see if this claim is made and the rational stated for such a classification. This appears to be a very dubious claim that I would not even dare to make without the article sitting in front of me. But then again I try not to drop such logic bombs without references or explanation and then run away......
Wireless World/Electronics World Giovanni Stochino
Ultra fast Audio amplifier. #1 Apr 1997, p278
Ultra fast Audio amplifier #2 Aug 1998, p633
Doug Self and Dr. Alex Megann describe it as being a two part article not a three part.
http://www.soton.ac.uk/~apm3/diyaudio/Stochino_amp.html
http://www.dself.dsl.pipex.com/ampins/library/ampartew.htm
I can image the reaction if I made such a controversial claim without any supporting evidence ..........
Thanks,
Diogenes of Sinope
"Will someone tell me how the first diff pair biased at 5 mA per transistor can be biased Class AB?"
When you drive either the P or N diff pair beyond the limit set by their respective current sources they go into class B mode and the extra current goes through the 200R resistor cross connecting the two current sources.
When you drive either the P or N diff pair beyond the limit set by their respective current sources they go into class B mode and the extra current goes through the 200R resistor cross connecting the two current sources.
Electronics World: march 1996, april 1997, august 1998. 
may post the whole shooting march...if no copy right issues..?
It's bad enough that one has to endure class AB operation in the output stage...to extend it to the small power stages is, at least for audio amps., unnecessary to say the least....
may post the whole shooting march...if no copy right issues..?
It's bad enough that one has to endure class AB operation in the output stage...to extend it to the small power stages is, at least for audio amps., unnecessary to say the least....
"It's bad enough that one has to endure class AB operation in the output stage...to extend it to the small power stages is, at least for audio amps., unnecessary to say the least...."
So you would rather listen to it overload and clip rather than smoothly transition into class B on the peaks?
So you would rather listen to it overload and clip rather than smoothly transition into class B on the peaks?
djk said:
Hi djk,
The amp. will not go into 'overload and clip' exclusively because its voltage gain block, (i.e the 'small signal' stages), is NOT biased in class-AB.
The only reason G. Stochino opted for class-AB operation for the first two stages of his amp., was to facilitate the generation of higher, (read non-linear), slew rates than are otherwise possible with normal 'small-signal', (a wretchedly inaccurate term!!.....i prefer small-power), class A operation, (~50V/us with this design).
A Class-AB voltage gain block allows his design to slew at over 300V/uS....There are no other benefits to this approach as far as i can tell.
As i have shown else where....
http://diyaudio.com/forums/showthread.php?s=&threadid=14482
...obtaining high slew rates does not require the complexity and suboptimal compromises inherent in running your voltage gain block in class-AB.
complexity and suboptimal compromises
Something we can agree on without reservation...... One half of the diff pair running out of current is called current starving and I have never seen refered to as Class AB. I still would like to hear under what conditions this would occur other than clipping and just quoting the relevent section should not constitute copyright issues. This proves my point about many of the amps designed by academics with the aim of optimizing some particular measured paramter at the expense of others........
"So you would rather listen to it overload and clip rather than smoothly transition into class B on the peaks?"
Two entirely different animals. Transition into Class B occurs when the output current demand exceeds the quiescent bias current of the output stage. Clipping occurs as the output voltage peak amplitude approaches the power supply rail.
Limiting signal swing in an amplifier so that it clips before the output stage goes into saturation, is an approach often used. It allows the amp to recover and restore loop feedback much more quickly and with considerably less nasty sonic consequences. This can very beneficial for small amps and is also can one of the rationales for amps with no negative feedback around the output stage.
Something we can agree on without reservation...... One half of the diff pair running out of current is called current starving and I have never seen refered to as Class AB. I still would like to hear under what conditions this would occur other than clipping and just quoting the relevent section should not constitute copyright issues. This proves my point about many of the amps designed by academics with the aim of optimizing some particular measured paramter at the expense of others........
"So you would rather listen to it overload and clip rather than smoothly transition into class B on the peaks?"
Two entirely different animals. Transition into Class B occurs when the output current demand exceeds the quiescent bias current of the output stage. Clipping occurs as the output voltage peak amplitude approaches the power supply rail.
Limiting signal swing in an amplifier so that it clips before the output stage goes into saturation, is an approach often used. It allows the amp to recover and restore loop feedback much more quickly and with considerably less nasty sonic consequences. This can very beneficial for small amps and is also can one of the rationales for amps with no negative feedback around the output stage.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.