Re: Re: Re: sophisticated amplifiers
Glen, you are such a nit-picker 🙂. Of course I was not referring to a Class A output stage. The Class-AB output stage is almost always where the big challenge lies in achieving low distortion. So, yes, Class A is "cheating" in the context of this discussion.
As far as duplicating Self's Class A results with "icky" MOSFETs in the output stage, probably no sweat at all. Run MOSFETs Class A and there is virtually no transconductance droop issue.
Cheers,
Bob
G.Kleinschmidt said:
Hi Bob.
I don’t think that we need to get all that sophisticated to achieve these numbers.
For instance, in a low power amp we can cheat a bit by using class A. I’ve been fiddling with 1,10, & 20kHz notch filters based on some fancy opamps and as far as I’m able to resolve, the THD-20 of my amp here…..
http://users.picknowl.com.au/~glenk/12W.HTM
…..is <0.001% THD at 20kHz and rated Pout into 4R.
Don’t forget, D. Self did ~0.003% THD-20 with 2MHz fT BJT’s in an quasi-complementary class A amp with standard miller compensation! I’d like to see someone replicate those results with icky MOSFET’s in the output stage!
Cheers,
Glen
Glen, you are such a nit-picker 🙂. Of course I was not referring to a Class A output stage. The Class-AB output stage is almost always where the big challenge lies in achieving low distortion. So, yes, Class A is "cheating" in the context of this discussion.
As far as duplicating Self's Class A results with "icky" MOSFETs in the output stage, probably no sweat at all. Run MOSFETs Class A and there is virtually no transconductance droop issue.
Cheers,
Bob
Re: sophisticated amplifiers and simulation
Hi Edmund,
First, no offense taken.
Second, I realize that when I say SPICE simulation results don't count, there are many caveats. You are pointing out situations where they DO count, and I agree with that. They especially count in moving one in the desired direction and giving perspective on what works better.
However, I just stop short of believing a bottom-line result on distortion in an absolute sense because there are so many factors that can influence that bottom line that may not be adequately modeled. SPICE is a great tool, and I love it, but most would agree that we recognize its limitations and don't depend on it exclusively. Those who throw it out completely because it doesn't do everything are no wiser than those who depend on it exclusively. I'm sure you agree that the middle ground is where it is at.
Cheers,
Bob
estuart said:
Hi Edmund,
First, no offense taken.
Second, I realize that when I say SPICE simulation results don't count, there are many caveats. You are pointing out situations where they DO count, and I agree with that. They especially count in moving one in the desired direction and giving perspective on what works better.
However, I just stop short of believing a bottom-line result on distortion in an absolute sense because there are so many factors that can influence that bottom line that may not be adequately modeled. SPICE is a great tool, and I love it, but most would agree that we recognize its limitations and don't depend on it exclusively. Those who throw it out completely because it doesn't do everything are no wiser than those who depend on it exclusively. I'm sure you agree that the middle ground is where it is at.
Cheers,
Bob
PMA said:
This is a good question, and I have sometimes wondered as well. However, I have seen numerous cases where I would have expected much lower distortions in this test, so maybe there is something to it.
Cheers,
Bob
Re: Re: Re: Re: Re: sophisticated amplifiers
Well, sure, but if you bias a typical class A amp that does <0.001% THD-20 in to class A/B operation it will still measure not far from your spec. At lest in terms of THD-20 at full rated power where crossover distortion has less significance.
Hmmmm.....but you still have to contend with a much lower transconductance and significantly worse large signal linearity.
Cheers,
Glen
Well, sure, but if you bias a typical class A amp that does <0.001% THD-20 in to class A/B operation it will still measure not far from your spec. At lest in terms of THD-20 at full rated power where crossover distortion has less significance.
Hmmmm.....but you still have to contend with a much lower transconductance and significantly worse large signal linearity.
Cheers,
Glen
Re: Re: Re: Re: sophisticated amplifiers
Hi Bob,
Actually, not so impressive, as I was cheating you a little bit.
At lower power levels I got 2ppm, so I can reproduce any figure between 2ppm and 48%, just by varying signal level. 😀
Hi Bob,
It is a fully complementary CFB amplifier, which slightly resembles the design of Mark Alexander, but on the following points it's completely different:
It uses a discrete fully bootstrapped input stage, a nested differential feedback loop (NDFL), (current) EC and "output stage inclusive Miller compensation" (OSIMC).
The latter may seem dangerous, however, the current gain of the two VAS's is reduced to 1, i.e. they are just current mirrors (and,to reassure you, with such low gain they don't have effective "arms" to "fight each other" 🙂).
The output stage comprises three pairs of MOSFET's (2SJ201/2SK1530), each biased at 120mA. Curiously, increasing the bias, even into class-heat, does not improve the performance.
Concerning EC, only the current component of the error is compensated for, as the voltage component is already minimized by means of OSIMC. The EC circuit is also the weakest point of this design, as without EC the distortion rises some 200 times! In other words, the gain of EC circuit has to set very accurately, within 0.2% or so.
In the mean time, I did a two tone test, 19/20kHz, and got 16ppb at 3/4 of max. power. Am I correct when I take the RMS of the components from 1 to 18kHz and divided it by the amplitude of one of the test signals? I mean, is this in accordance with the 'official' procedure?
You are right, I didn't provide much information before, only the input stage, which you can found here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=101468&perpage=10&highlight=&pagenumber=75 post#748
Cheers, Edmond.
Bob Cordell said:
Hi Bob,
Actually, not so impressive, as I was cheating you a little bit.
At lower power levels I got 2ppm, so I can reproduce any figure between 2ppm and 48%, just by varying signal level. 😀
Bob Cordell said:
Hi Bob,
It is a fully complementary CFB amplifier, which slightly resembles the design of Mark Alexander, but on the following points it's completely different:
It uses a discrete fully bootstrapped input stage, a nested differential feedback loop (NDFL), (current) EC and "output stage inclusive Miller compensation" (OSIMC).
The latter may seem dangerous, however, the current gain of the two VAS's is reduced to 1, i.e. they are just current mirrors (and,to reassure you, with such low gain they don't have effective "arms" to "fight each other" 🙂).
The output stage comprises three pairs of MOSFET's (2SJ201/2SK1530), each biased at 120mA. Curiously, increasing the bias, even into class-heat, does not improve the performance.
Concerning EC, only the current component of the error is compensated for, as the voltage component is already minimized by means of OSIMC. The EC circuit is also the weakest point of this design, as without EC the distortion rises some 200 times! In other words, the gain of EC circuit has to set very accurately, within 0.2% or so.
In the mean time, I did a two tone test, 19/20kHz, and got 16ppb at 3/4 of max. power. Am I correct when I take the RMS of the components from 1 to 18kHz and divided it by the amplitude of one of the test signals? I mean, is this in accordance with the 'official' procedure?
You are right, I didn't provide much information before, only the input stage, which you can found here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=101468&perpage=10&highlight=&pagenumber=75 post#748
Cheers, Edmond.
AndrewT said:
Hi Andrew,
I'm not sure, as I can't remember I've said something like that.
Cheers, Edmond.
Re: Re: Re: Re: Re: Re: sophisticated amplifiers
I don't think so. Do the math 🙂.
Cheers,
Bob
G.Kleinschmidt said:
I don't think so. Do the math 🙂.
Cheers,
Bob
Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
Care to eleborate a bit? The non linearity of Vgs vs Id causes significant large signal distortion.
"The use of power FETs in output stages is often advocated. However, after much investigation, I have found the conclusion inescapable that FETs suffer not only from poor basic linearity, due to low gm, but also a crossover region that is inherently more jagged than BJTs. It is not possible to explore this in detail here, but see [7],[8] "............
7] Self, D "Audio Power Amplifier Design Handbook." Newnes 1996, p231. ISBN 0-7506-2788-3 (poor FET linearity)
8] Self, D "FETs vs BJTs- the linearity competition." Electronics & Wireless World, May 1995 p387. (poor FET linearity)
Above from here:
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm#R
Any charitable soul out there with a copy of ref #8 to send to Bob Cordell? He's a poor misguided FET user who needs educating.
Cheers,
Glen
Bob Cordell said:
Care to eleborate a bit? The non linearity of Vgs vs Id causes significant large signal distortion.
"The use of power FETs in output stages is often advocated. However, after much investigation, I have found the conclusion inescapable that FETs suffer not only from poor basic linearity, due to low gm, but also a crossover region that is inherently more jagged than BJTs. It is not possible to explore this in detail here, but see [7],[8] "............
7] Self, D "Audio Power Amplifier Design Handbook." Newnes 1996, p231. ISBN 0-7506-2788-3 (poor FET linearity)
8] Self, D "FETs vs BJTs- the linearity competition." Electronics & Wireless World, May 1995 p387. (poor FET linearity)
Above from here:
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm#R
Any charitable soul out there with a copy of ref #8 to send to Bob Cordell? He's a poor misguided FET user who needs educating.
Cheers,
Glen
Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
Self was at least partially full of baloney on this one. At the very least, he had blinders on and did not tell the whole story. MOSFETs have about 10 times less gm at a given operating current than bipolars. If you get past this, they are every bit as linear. As a very rough approximation, the dynamic source resistance (1/gm) of a MOSFET is about 250/Id in mA.
Let's build a 100 W Class A amplifier from MOSFETS. It must support 40V pk and 5A pk into an 8 ohm load. Lets give it 44V main rails. The Class A idle bias must be 2.5 A. Lets use 5 pairs of MOSFET output devices, each carrying 500 mA of idle bias and each dissipating 22W at idle. The dynamic source impedance of each MOSFET is about 250/500 = 0.5 ohms. The five upper devices in parallel will yield a net upper Rs of 0.5/5 = 0.1 ohms. The bottom five devices will do the same, bringing the total open-loop output impedance at idle to 0.05 ohms.
At a peak current of 5 A, each of the upper transistors is at 1 amp with an Rs of 0.25 ohms. We have five in parallel, so we get an effective net Rs of 0.05 ohms. To first order, this is the same as at idle bias. Of course, there will be wiggles in between, and this approximation is only that, but the point is that the output stage net transconductance, variations of which will cause distortion, will actually vary very little with signal in this Class-A arrangement.
In any case, the variation in net 1/gm is certainly less than 0.04 ohms over signal. This is less than 0.5% of the 8 ohm load, suggesting in very rough terms open loop distortion on the order of less than 0.5%. Now we put 40 dB of NFB around it and we are at 0.005%, and actually probably quite a bit less.
Cheers,
Bob
G.Kleinschmidt said:
Care to eleborate a bit? The non linearity of Vgs vs Id causes significant large signal distortion.
"The use of power FETs in output stages is often advocated. However, after much investigation, I have found the conclusion inescapable that FETs suffer not only from poor basic linearity, due to low gm, but also a crossover region that is inherently more jagged than BJTs. It is not possible to explore this in detail here, but see [7],[8] "............
7] Self, D "Audio Power Amplifier Design Handbook." Newnes 1996, p231. ISBN 0-7506-2788-3 (poor FET linearity)
8] Self, D "FETs vs BJTs- the linearity competition." Electronics & Wireless World, May 1995 p387. (poor FET linearity)
Above from here:
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm#R
Any charitable soul out there with a copy of ref #8 to send to Bob Cordell? He's a poor misguided FET user who needs educating.
Cheers,
Glen
Self was at least partially full of baloney on this one. At the very least, he had blinders on and did not tell the whole story. MOSFETs have about 10 times less gm at a given operating current than bipolars. If you get past this, they are every bit as linear. As a very rough approximation, the dynamic source resistance (1/gm) of a MOSFET is about 250/Id in mA.
Let's build a 100 W Class A amplifier from MOSFETS. It must support 40V pk and 5A pk into an 8 ohm load. Lets give it 44V main rails. The Class A idle bias must be 2.5 A. Lets use 5 pairs of MOSFET output devices, each carrying 500 mA of idle bias and each dissipating 22W at idle. The dynamic source impedance of each MOSFET is about 250/500 = 0.5 ohms. The five upper devices in parallel will yield a net upper Rs of 0.5/5 = 0.1 ohms. The bottom five devices will do the same, bringing the total open-loop output impedance at idle to 0.05 ohms.
At a peak current of 5 A, each of the upper transistors is at 1 amp with an Rs of 0.25 ohms. We have five in parallel, so we get an effective net Rs of 0.05 ohms. To first order, this is the same as at idle bias. Of course, there will be wiggles in between, and this approximation is only that, but the point is that the output stage net transconductance, variations of which will cause distortion, will actually vary very little with signal in this Class-A arrangement.
In any case, the variation in net 1/gm is certainly less than 0.04 ohms over signal. This is less than 0.5% of the 8 ohm load, suggesting in very rough terms open loop distortion on the order of less than 0.5%. Now we put 40 dB of NFB around it and we are at 0.005%, and actually probably quite a bit less.
Cheers,
Bob
Re: Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
In an actual case with .47 ohm source resistors, I get bout .06%.
It is less with lower Source resistance. This lower figure, of
course, is due to push-pull cancellation and applies to the
Fairchild P channel parts - the IR P channel parts are about .1%
mid-band.
😎
Bob Cordell said:
In an actual case with .47 ohm source resistors, I get bout .06%.
It is less with lower Source resistance. This lower figure, of
course, is due to push-pull cancellation and applies to the
Fairchild P channel parts - the IR P channel parts are about .1%
mid-band.
😎
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
Thanks, Nelson. I sort of felt that my back-of-the-envelope was very wosrt-case. Glad to hear it is so much lower. Interesting that the IRF P channel problem causes it to nearly double. Did you notice the doubling of distortion at all frequencies, or was it different for different frequency bands?
Bob
Nelson Pass said:
Thanks, Nelson. I sort of felt that my back-of-the-envelope was very wosrt-case. Glad to hear it is so much lower. Interesting that the IRF P channel problem causes it to nearly double. Did you notice the doubling of distortion at all frequencies, or was it different for different frequency bands?
Bob
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
No argument, Glen. I have always said that transconductance droop in a Class-AB output stage with MOSFETs biased at fairly low values (e.g., about 150 mA) would put MOSFETs at a disadvantage. That is why I used EC, as I explained in my original paper. If you bias them hot, use a bunch of them, and run them in Class A, they will be totally competitive with bipolars without EC, and that is the situation I thought we were discussing.
Cheers,
Bob
G.Kleinschmidt said:
Yes, I know the D.Self treatise on MOSFET linearity wasn't as good as it could have been and raises many issues for debate(that's why I cited it in the cheeky manner I which I did 🙂 ). However, I still think that his point that the intrinsic linearity of MOSFET's as source followers is generally worse than that of BJT's as emitter followers is valid.
0.5% THD is still quite high, especially @ 100W into an 8 ohm load with 5 parallel MOSFET pairs in class A.
Bob, your 50W MOSFET design, without the EC, I believe, produced a THD-20 of 0.02% with about 40dB NFB.
As a comparison, D. Self's EF output 'blameless' does about 0.01% THD-20 with less NFB (miller compensation), using a single pair of 2MHz fT BJT's in a 30W design.
Cheers,
Glen
No argument, Glen. I have always said that transconductance droop in a Class-AB output stage with MOSFETs biased at fairly low values (e.g., about 150 mA) would put MOSFETs at a disadvantage. That is why I used EC, as I explained in my original paper. If you bias them hot, use a bunch of them, and run them in Class A, they will be totally competitive with bipolars without EC, and that is the situation I thought we were discussing.
Cheers,
Bob
Interestingly, the higher distortion is at mid-band, declining
somewhat toward the top end.
😎
somewhat toward the top end.
😎
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
OK. I'm just sayin’ that source followers aren’t as linear as emitter followers. I think that is still the case whether they are biased in class A or class AB.
Cheers,
Glen
Bob Cordell said:
OK. I'm just sayin’ that source followers aren’t as linear as emitter followers. I think that is still the case whether they are biased in class A or class AB.
Cheers,
Glen
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: sophisticated amplifiers
OK, but how does the amplifer producing those figures compare to Bob's example, namely:
"40V pk and 5A pk into an 8 ohm load. Lets give it 44V main rails. The Class A idle bias must be 2.5 A. Lets use 5 pairs of MOSFET output devices, each carrying 500 mA of idle bias and each dissipating 22W at idle."
I assume the "push-pull cancellation" you refer due to a balanced/bridged output stage?
Cheers,
Glen
Nelson Pass said:
OK, but how does the amplifer producing those figures compare to Bob's example, namely:
"40V pk and 5A pk into an 8 ohm load. Lets give it 44V main rails. The Class A idle bias must be 2.5 A. Lets use 5 pairs of MOSFET output devices, each carrying 500 mA of idle bias and each dissipating 22W at idle."
I assume the "push-pull cancellation" you refer due to a balanced/bridged output stage?
Cheers,
Glen