Re: MOSFET Cgs
I understand the possible issues beyond a strong Cgs(Vgs) dependency. A nonlinear Cgs characteristic can generate intermodulation distortion at the input of the circuit. The distorted signal is amplified by the active device, affecting the overall linearity of the circuit.
I'm not sure where you got the assumption on the Ggs variation, but the basic Meyer model for MOSFETs (also included in SPICE) is not predicting such a strong dependency. See http://nina.ecse.rpi.edu/shur/Ch5/sld020.htm and the following slides. The equivalent Cgs is the sum of a gate-source overlap capacitance (which is basically independent of the bias conditions Vgs and Vds) and a periphery diffusion capacitance which is nonlinear, it's dependency on Vgs and Vds changes with the MOSFET operation region (subthreshold, linear, saturation).
The situation is much better than for bipolars; usually the periphery capacitance is much smaller than the overlap cpacitance, even in strong inversion. Though, this nonlinear Cgs dependency is a major concern for microwave MOSFETs build on compound semiconductors; such devices have extremely small overlap capacitances (partly due to very shallow junctions) and therefore the nonlinear part of Cgs is a major contributor to the overall capacity.
For power devices, the nonlinear Cgs contribution would be significant for high voltage devices; due to a ligthly doped substrate and deep junctions, the nonlinear Cgs contribution would increase. But then nobody is using such MOSFETs for linear applications.
I don't think that the nonlinear Cgs and Cgd should be a major concern in audio applications.
Bob Cordell said:
I understand the possible issues beyond a strong Cgs(Vgs) dependency. A nonlinear Cgs characteristic can generate intermodulation distortion at the input of the circuit. The distorted signal is amplified by the active device, affecting the overall linearity of the circuit.
I'm not sure where you got the assumption on the Ggs variation, but the basic Meyer model for MOSFETs (also included in SPICE) is not predicting such a strong dependency. See http://nina.ecse.rpi.edu/shur/Ch5/sld020.htm and the following slides. The equivalent Cgs is the sum of a gate-source overlap capacitance (which is basically independent of the bias conditions Vgs and Vds) and a periphery diffusion capacitance which is nonlinear, it's dependency on Vgs and Vds changes with the MOSFET operation region (subthreshold, linear, saturation).
The situation is much better than for bipolars; usually the periphery capacitance is much smaller than the overlap cpacitance, even in strong inversion. Though, this nonlinear Cgs dependency is a major concern for microwave MOSFETs build on compound semiconductors; such devices have extremely small overlap capacitances (partly due to very shallow junctions) and therefore the nonlinear part of Cgs is a major contributor to the overall capacity.
For power devices, the nonlinear Cgs contribution would be significant for high voltage devices; due to a ligthly doped substrate and deep junctions, the nonlinear Cgs contribution would increase. But then nobody is using such MOSFETs for linear applications.
I don't think that the nonlinear Cgs and Cgd should be a major concern in audio applications.
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: I repeat my Request
And this is exactly the reply I was expecting.
I never said that the thing was completelly up and running, and I never said that I had a complete EC + bootstrapped 500W class A stage operational. If I did, I would be posting photos and measured results instead of revealing my schematics as I develop the final circuit blocks. This is a big project and it takes time. I have however (way back when we discussed this last) built and tested both an EC and and an opamp-linearised class A/AB output stage with bootstrapped rails and they worked fine, remarkably without oscillating.
I have experimented with these circuit blocks and topologies to know what I am doing (regardless of presumptions) and to have enough confidence in the final design to blow about 3k building it.
If anyone thinks my claims for the potential performance of the scheme I have outlined thus far are off the mark, then I'd welcome some good technical arguments as to why or what I am doing wrong.
Cheers,
Glen
PS.
Since this is deviating from the BJT vs MOSFET debate, I'll start a new thread for this later in the week. This will make a good excuse for developing some Sanken BJT SPICE models.
Bob Cordell said:
And this is exactly the reply I was expecting.
I never said that the thing was completelly up and running, and I never said that I had a complete EC + bootstrapped 500W class A stage operational. If I did, I would be posting photos and measured results instead of revealing my schematics as I develop the final circuit blocks. This is a big project and it takes time. I have however (way back when we discussed this last) built and tested both an EC and and an opamp-linearised class A/AB output stage with bootstrapped rails and they worked fine, remarkably without oscillating.
I have experimented with these circuit blocks and topologies to know what I am doing (regardless of presumptions) and to have enough confidence in the final design to blow about 3k building it.
If anyone thinks my claims for the potential performance of the scheme I have outlined thus far are off the mark, then I'd welcome some good technical arguments as to why or what I am doing wrong.
Cheers,
Glen
PS.
Since this is deviating from the BJT vs MOSFET debate, I'll start a new thread for this later in the week. This will make a good excuse for developing some Sanken BJT SPICE models.
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: I repeat my Request
Hi Glen,
In post # 1652, you stated, "From my experience in building/prototyping the various blocks of my rail-tracking 500W class A with 60 MHz/35 MHz Sanken BJT's, the output impedance and distortion figures quoted by you and Nelson look lame, and that is before I apply EC."
That is a pretty bold claim.
You failed to be forthcoming about the conditions of the claim and caveats after being questioned. Even now, you have still not said what distortion number you achieved that made Nelson's number look lame. This is why you have lost a lot of credibility with me.
Your amplifier looks like a very interesting and ambitious project, and I agree that it deserves its own thread. I hope you will not feel the need to raise up that accomplishment by putting down the accomplishments of others. I'm sure it will stand on its own.
It would be great if you develop some good SPICE models for the Sanken devices. Check with Andy_c, as he has done a very good job of putting together some good SPICE models of the ThermalTrak transistors, which will likely have some similar behavior and issues.
Bob
G.Kleinschmidt said:
Hi Glen,
In post # 1652, you stated, "From my experience in building/prototyping the various blocks of my rail-tracking 500W class A with 60 MHz/35 MHz Sanken BJT's, the output impedance and distortion figures quoted by you and Nelson look lame, and that is before I apply EC."
That is a pretty bold claim.
You failed to be forthcoming about the conditions of the claim and caveats after being questioned. Even now, you have still not said what distortion number you achieved that made Nelson's number look lame. This is why you have lost a lot of credibility with me.
Your amplifier looks like a very interesting and ambitious project, and I agree that it deserves its own thread. I hope you will not feel the need to raise up that accomplishment by putting down the accomplishments of others. I'm sure it will stand on its own.
It would be great if you develop some good SPICE models for the Sanken devices. Check with Andy_c, as he has done a very good job of putting together some good SPICE models of the ThermalTrak transistors, which will likely have some similar behavior and issues.
Bob
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: I repeat my Request
Bob,
It was also a general claim.
Your hypothetical/computed THD figure was 0.5%. For a 100W class A stage, that is lame (sorry).
Nelsons THD figures were presented without any mention of load impedance, power output, specific frequency, output stage topology (single-ended / bridged?) or number of output devices used, or bias current. All that is completely clear to me about Nelson's output stage in question is that it uses complementary P/N MOSFET's.
Yet, apparently, as demanded by you, all these things must be explicitly detailed by me, lest my credibility suffers.
I can assume, for the sake of the argument, that his Class A output stage with 0.1 and 0.06% THD is identical in every other respect to the hypothetical 100W class A stage put forth by you, but that would seem a coincidence. For this reason, I did explicitly ask Nelson if this was the case (for the purpose of making a more accurate comparison - particularly with regards to your hypothetical 0.5% THD figure) but I was ignored.
My comments on the THD figures presented were to express my opinion that they do not represent the performance limit of bipolar class A output stages.
Finally, as far as I can tell, ultra-low THD isn’t one of Nelson’s design aims, and this hasn't hindered his business successes. Now I could be wrong, but I doubt that he really cares if I don’t like his THD figures. We are each entitled to design to a specification that we personally see fit.
I not sure who's accomplishments exactly I’m guilty of putting down here.
Bob Cordell said:
Bob,
It was also a general claim.
Your hypothetical/computed THD figure was 0.5%. For a 100W class A stage, that is lame (sorry).
Nelsons THD figures were presented without any mention of load impedance, power output, specific frequency, output stage topology (single-ended / bridged?) or number of output devices used, or bias current. All that is completely clear to me about Nelson's output stage in question is that it uses complementary P/N MOSFET's.
Yet, apparently, as demanded by you, all these things must be explicitly detailed by me, lest my credibility suffers.
I can assume, for the sake of the argument, that his Class A output stage with 0.1 and 0.06% THD is identical in every other respect to the hypothetical 100W class A stage put forth by you, but that would seem a coincidence. For this reason, I did explicitly ask Nelson if this was the case (for the purpose of making a more accurate comparison - particularly with regards to your hypothetical 0.5% THD figure) but I was ignored.
My comments on the THD figures presented were to express my opinion that they do not represent the performance limit of bipolar class A output stages.
Finally, as far as I can tell, ultra-low THD isn’t one of Nelson’s design aims, and this hasn't hindered his business successes. Now I could be wrong, but I doubt that he really cares if I don’t like his THD figures. We are each entitled to design to a specification that we personally see fit.
I not sure who's accomplishments exactly I’m guilty of putting down here.
Toshiba 2SJ201 & 2SK1530
Does anybody have a good source for the Toshiba power MOSFETs 2SJ201 and 2SK150?
A source where there is a low probability of counterfeit devices?
Digikey lists them and prices them, but appears to have none on hand.
BTW, I contacted Toshiba re SPICE models for them, and of course they don't have them. Anyone got some?
In some applications these Toshiba MOSFETs may be more attractive than the IRF-like devices because they appear to have a somewhat lower turn-on voltage.
Also, they appear to spec a higher 100V, 10 ms SOA of about 5.5A than the number for the IRF devices, but this might just be a reflection of some greater conservatism on the part of IR.
Thanks!
Bob
Does anybody have a good source for the Toshiba power MOSFETs 2SJ201 and 2SK150?
A source where there is a low probability of counterfeit devices?
Digikey lists them and prices them, but appears to have none on hand.
BTW, I contacted Toshiba re SPICE models for them, and of course they don't have them. Anyone got some?
In some applications these Toshiba MOSFETs may be more attractive than the IRF-like devices because they appear to have a somewhat lower turn-on voltage.
Also, they appear to spec a higher 100V, 10 ms SOA of about 5.5A than the number for the IRF devices, but this might just be a reflection of some greater conservatism on the part of IR.
Thanks!
Bob
john curl said:
Hi John,
Your comment reminds me of an old computer book I bought once: it was called "Real Men Use DOS"
.Seriously, very good amplifiers can be made without resort to SPICE. Indeed, I did not use SPICE in doing my MOSFET amplifier with error correction. This includes development of the compensation scheme necessary to make HEC practical. So it certainly can be done.
As I have said many times before, SPICE is an excellent tool that can play a very positive role in design exploration and in optimizing a design. It would be unfair to suggest that those who use SPICE over-depend on it.
Bob
john curl said:
I use SPICE to good effect, but I don't rely on it to give very
accurate results for non-linear circuits. I have to smile if someone
quotes parts-per-million distortion figures from a simulation.
john curl said:
You may have been there at AES when Lipshitz and Van der Kooy
presented their (masterful) analysis on the Quad current dumping
amplifier. When asked by a member of the audience how it
sounded, one of them said (and I paraphrase here), "I don't
know, we didn't listen to it."
Re: Toshiba 2SJ201 & 2SK1530
Hi Bob,
Did you mean the 2SK1530? The models for these devices were discussed earlier in this thread here . I haven't done extensive verification of these, but I have done some capacitance sims. Also, in private email to Edmond, who came up with the original models, I sent him some plots of Cgd vs Vgd. These are in Excel spreadsheets attached to this post.
Bob Cordell said:
Hi Bob,
Did you mean the 2SK1530? The models for these devices were discussed earlier in this thread here . I haven't done extensive verification of these, but I have done some capacitance sims. Also, in private email to Edmond, who came up with the original models, I sent him some plots of Cgd vs Vgd. These are in Excel spreadsheets attached to this post.
I think people in the business of selling mysticism are prone to use the anti-SPICE argument as a political platform to create a false dichotomy between "the mystical" and "the superficial".
The idea of model development is to continually refine and improve the mathematical models of devices in order to get the most accurate simulations possible. Will they ever be perfect? No. But the better they get, the more ridiculous the false dichotomy becomes.
The idea of model development is to continually refine and improve the mathematical models of devices in order to get the most accurate simulations possible. Will they ever be perfect? No. But the better they get, the more ridiculous the false dichotomy becomes.
john curl said:
Are we talking rear engine Porsche or front engine? We won't bother with the mid engine 914 which really should have been a Volkswagen.
It would have to be rear I'd say for the time period that you are referring to and it actually had rather strange characteristics when the rear end broke loose. Very quirky and unnatural. After all it was the evolution of a Volkswagen, at least the conceptual design. Not to be picky or anything, just sayin' ...
"The rear-bias was always a problem to 911's handling. Any tail-heavy cars have a tendency to oversteer. If such oversteer is not adequately suppressed, lost of control may occur. "
From: http://www.autozine.org/911/911_5.htm
Porsche ... Controlled skid, I don't think so ...
And I can't believe I took the time to write this, LOL!
SPICE has it's purpose as does prototyping. The productive discussion would be to further the understanding of the strengths and weaknesses of each.
Pete B.
Remember this:
http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm
I wonder if Ivor T. uses SPICE, I seriously doubt it.
http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm
I wonder if Ivor T. uses SPICE, I seriously doubt it.
john curl said:
I owned two 944s, yeah I've driven them some.
The 924 should have been a Volkswagen, and probably the 944 also. The 911 series is what made Porsche the legendary company. You stated from the same time frame, and the 944 is certainly not from the same time frame, change your story often?:
http://en.wikipedia.org/wiki/Renault_Dauphine
And by the way, SPICE is not emulation, as you seem to repeatedly state, rather it is simulation:
http://en.wikipedia.org/wiki/Emulator
Quickturn is an emulation system where programmable logic is used typically, to emulate the behavior of a chip or system. My cache controller chip design for example worked first time in both emulation (Quickturn) and the real chip because I took the time to learn and use modern tools such as simulation. My NTSC Video encoder also worked in first pass silicon without any engineering changes. Modern tools are what make first pass success possible.
I hear that the first Pentium was emulated in a large array of Quickturn boxes. They learned their lesson using cut and try with the 286 which was seriously late to market, and never did work right in Protected mode.
Learn and grow you say? Seems I do that fairly regularly, what about you?
Let me repeat my position just in case it goes unnoticed:
"SPICE has it's purpose as does prototyping. The productive discussion would be to further the understanding of the strengths and weaknesses of each. "
I have suggested rapid prototyping on occassion when it made sense.
Pete B.
john curl said:
I got the point of your auto analogy from the start, and I do agree with having a good amount of design margin and handling misuse conditions gracefully. However, the war and cheap shots from some members of the camps are unfortunate. Makes constructive discussion difficult.
Just wanted to point out the Porsche thing, and the sweeping statements about SPICE users.
Using SPICE is not to be taken lightly, we often have a team of tools people for validation and to make sure that everything is in proper working order. I welcome having a SPICE Guru because it is a difficult task to validate models. If someone enjoys the technical challenge and contributes I welcome the input.
I also welcome your positive technical input regarding your experience John, and do appreciate it.
Pete B.
Slowly (assuming you were asking me) Though I did come up with a neat little cascode that cancels (yes cancels, not reduces)one more source of distortion.
I don't claim to a spice expert (or any other kind of expert), but I do want to point out the potential of spice. All the complaints people have with spice are really complaints about how people use it.
As an experiment for my own edification, I just did a Monte Carlo of an idealized LTP pair: just two transistors (2n3904s using On Semi's model), an ideal current source, an ideal differential current to voltage converter and an input stimulus such that the Ic swung 10%.
I kept one transistor constant and used spice to vary every parameter of the other +-20%. Across 50 runs the second harmonic varied between -50db and -81db, averaging -61db, with perfectly matched transistors it was -95db (common-mode distortion).
The 34db difference does show the need for balanced circuits to be intentionally imbalanced in simulation.
(assuming you were asking me) Though I did come up with a neat little cascode that cancels (yes cancels, not reduces)one more source of distortion.I don't claim to a spice expert (or any other kind of expert), but I do want to point out the potential of spice. All the complaints people have with spice are really complaints about how people use it.
As an experiment for my own edification, I just did a Monte Carlo of an idealized LTP pair: just two transistors (2n3904s using On Semi's model), an ideal current source, an ideal differential current to voltage converter and an input stimulus such that the Ic swung 10%.
I kept one transistor constant and used spice to vary every parameter of the other +-20%. Across 50 runs the second harmonic varied between -50db and -81db, averaging -61db, with perfectly matched transistors it was -95db (common-mode distortion).
The 34db difference does show the need for balanced circuits to be intentionally imbalanced in simulation.
Tim__x said:Slowly
I don't claim to a spice expert (or any other kind of expert), but I do want to point out the potential of spice. All the complaints people have with spice are really complaints about how people use it.
As an experiment for my own edification, I just did a Monte Carlo of an idealized LTP pair: just two transistors (2n3904s using On Semi's model), an ideal current source, an ideal differential current to voltage converter and an input stimulus such that the Ic swung 10%.
I kept one transistor constant and used spice to vary every parameter of the other +-20%. Across 50 runs the second harmonic varied between -50db and -81db, averaging -61db, with perfectly matched transistors it was -95db (common-mode distortion).
The 34db difference does show the need for balanced circuits to be intentionally imbalanced in simulation.
Hi Tim,
This is a very interesting experience and a good example of how SPICE can be used to explore the design space and also the effect of real-world imperfections.
When you were varying all those parameters by +/- 20%, were the two collector currents kept identical? I'm usually under the impression that, because of the nature of the exponential Vbe characteristic, matching the collector currents is pretty much the most important thing in a diff pair, and am also under the impression that if the collector currents are matched, they are pretty tolerant of some other mismatches in parameters. What did you find?
Thanks,
Bob
PS: one area in which I find SPICE especially useful is in analysis of feedback compensation circuits and also analysis of circuitry that might be prone to local oscillations.
Tim__x said:Slowly
I don't claim to a spice expert (or any other kind of expert), but I do want to point out the potential of spice. All the complaints people have with spice are really complaints about how people use it.
As an experiment for my own edification, I just did a Monte Carlo of an idealized LTP pair: just two transistors (2n3904s using On Semi's model), an ideal current source, an ideal differential current to voltage converter and an input stimulus such that the Ic swung 10%.
I kept one transistor constant and used spice to vary every parameter of the other +-20%. Across 50 runs the second harmonic varied between -50db and -81db, averaging -61db, with perfectly matched transistors it was -95db (common-mode distortion).
The 34db difference does show the need for balanced circuits to be intentionally imbalanced in simulation.
Yes, I was asking. The cross quad looks odd to me, and I have
yet to wrap my head around it.
That is an interesting example of Monte Carlo, and is appreciated.
At the same time, I have several examples of simulations which
don't give correct results, and I have to admit that I'm not an
expert either.
My favorite is a MicroCap9 sim of an ordinary op amp in a
Sallen Key type 2 pole high pass with gain in the op amp. As
you raise the gain, the Q increases and a peak forms at the
knee, and as you raise the gain further, it turns into an
oscillator. Try as I might, I can't make the simulation oscillate,
although it behaves as expected up to that point.
Any ideas?
Thanks Nelson for the tip on Microcap. I have a very early version that works directly with a Mac and I have used it for almost 20 years. It does what I need, mostly. You know AC and transient analysis.
However, I still cannot get my Mac to Windows emulator working properly, and trying to install it the other day with the expert help of a friend, crashed my operating system and I am still recovering. I also misplaced my codes for the Windows operating program, so I have to find it first, before I can make the emulator try to make the Spice simulation work. In any case, thanks for the input.
However, I still cannot get my Mac to Windows emulator working properly, and trying to install it the other day with the expert help of a friend, crashed my operating system and I am still recovering. I also misplaced my codes for the Windows operating program, so I have to find it first, before I can make the emulator try to make the Spice simulation work. In any case, thanks for the input.