You seem very competent about models. So why don't you publish your own simulations with *your* so accurate models ?I see at the start of this thread some distortion simulations, but the FET models used seem to be practically useless for this.
Has anyone done distortion sims with accurate models? I would prefer not to have to search the whole thread
(It can seems disagreeable, but it is really boring this endless models controversy, never justified, whatever the ones some can try...)
Last edited:
If I can simulate an actual unit accurately then I can test new improvements with more confidence that they will work in reality.
Do you have a link to these measured actual distortion values?
That would be very useful.
I plan to do so, but obviously I don't want to duplicate work that has already been done.
It is also important to be polite and credit earlier works.
There is a difference between minor model tweaks and models that are seriously faulty.
But if you prefer impressive but incorrect claims that is your choice, you would not be alone.
I will certainly share my results with you.
That thread is excellent, thank you for the link.
Best wishes
David
Last edited:
Where is *the file* with the good models you are talking of ? Constantly completed and updated ?
With the prove that the models inside are better than other ones ?
Where is the web site where we can download this single and unique file everybody will find and use ?
How can exist "seriously faulty other ones" ?
What the hell are doing the so calling model "specialists" (and devices manufacturers) if they don't share this work in a single and unique place, working together for the community ?
I cannot understand this LTSPICE mess, this schizophrenic policy about models and those endless controversies. Pure nonsense. Waste of time.
And seriously bored by those critics and remarks about models, each time one publish a simulation with never any data to justify such comments...
Last edited:
Dave,
I assumed that Lazy Cat would have done actual distortion measurements on his VSSA amplifier after he was satisfied with the final results of his listening tests? Perhaps I am wrong on that score but there must be hundreds of his amplifiers out there and to think nobody has tested for the end result I would be surprised. I know that Eperado has those amplifiers himself and wonder if he has done any testing. How about it Christophe?
I assumed that Lazy Cat would have done actual distortion measurements on his VSSA amplifier after he was satisfied with the final results of his listening tests? Perhaps I am wrong on that score but there must be hundreds of his amplifiers out there and to think nobody has tested for the end result I would be surprised. I know that Eperado has those amplifiers himself and wonder if he has done any testing. How about it Christophe?
I have only an old LEA distortiometer, and the limit was under its thereshold (~ 0.001% @ 1000Hz).
Not so different than the results i had in simulation, using the Bob Cordell Models (seriously faulty ?).
Last edited:
No, the Bob Cordell models are the best I know.
The problem was that the simulations at the start of the thread are not done with Bob's models but with the seriously faulty, for this application, NMOS/PMOS models.
That was exactly my point, which you seem to have missed completely.
I asked to find the better simulations so I didn't have to redo them or search the >3000 posts in the thread.
Nice to have this clear now.
So where are simulations done with Bob's models?
Also do you have actual measurements that are more informative?
ITU-R standard IMD at 19 kHz+ 20 kHz or, still useful, 20 kHz THD?
Best wishes
David
Last edited:
You can make your own simul using my files here:
http://www.esperado.fr/temp/VSSA/vssa-vs-vfa.html
or
www.esperado.fr - VSSA with Diamond input and DC servo
and
http://www.esperado.fr/vssa-diamond/dvssa.php
Thank you for the link. Unfortunately your simulation files do NOT use the Cordell models of the FETs.
They use exactly the same type of PMOS/NMOS FET model that is known to be inaccurate for distortion.
The output FET is probably the dominant source of distortion so none of your distortion results is likely to be accurate.
Why did you not use the Cordell models?
Best wishes
David
I do use the Cordell models. the file 'Cordell-Models.txt' is even provided in the ZIP.
I'm NOT interested in simulated distortion results, and i don't believe in them anyway.
You are free, on your side, to use the models you want and publish your results.
You can even draw the circuit yourself. I don't even understand what you're looking for.
I'm NOT interested in simulated distortion results, and i don't believe in them anyway.
You are free, on your side, to use the models you want and publish your results.
You can even draw the circuit yourself. I don't even understand what you're looking for.
Last edited:
You provided the file Cordel-Models.Txt but you have not used it for the FET models.
Your ASC specifies 2SJ162 and 2SK1058 but they do not exist in Cordell-Models.Txt.
There are Cordell models for 2SJ162C and and 2SK1056C
So the models you have used are the ones on the ASC and not the Cordell models.
Your articles include many distortion results, you seem to be interested in them when you think they are in your favour.
So I re-simulated some of your work with Cordell models and your results are not as inaccurate as I expected.
I had previously seen errors of 50 or 100 times too low distortion.
Here it was only a factor of around 2 on the few I checked, not too bad really.
This has helped me learn why some circuits are more sensitive to model faults.
Best wishes
David
There are a few other points I noticed on your ASC that I can help you with, if you actually want accurate simulations.
Last edited:
http://www.esperado.fr/vssa-diamond/dvssa.php ???
"Nb: It has been reported my models for the OPS were not accurate. With Cordell models and +-45V, the distortion, at 1V input, goes up to: 0.0006% Not a big deal."
What do mean "in your favour" ? Am-i selling anything ?
I'm not interested in absolute values of distortion numbers ( in simulations), but to compare them. This is clearly indicated in my articles.
Take one second to think about. The purpose was, each time, to compare various input stages, everything else equal. Less the simulation takes into account the distortion of the output stage, better the comparison focuses on the input stage. It is the purpose, in the 3 articles of the comparison, reason why i stay with the models i used. I'm too lazy to redo images all those graphs with other models, it takes a lot of time for no benefit.
Dave, again, why don't you *publish* your results instead of making abstract criticism to other's work ? I mean, less blabla, more real cooperative work ?
You will see that the numbers changes a little, but they will lead to the exact same conclusions.
The *real* thing should be to make *real* measurements IRL that i cannot for the moment. At this point, and at this point only, you will have demonstrated witch model is closer to reality. Until you do this work, you have not made any evidence of anything.
I had build those circuit and listened to them. They all measure under the limit of my distortion meter. I lie on my ears to discriminate them And my subjective conclusions correlate with simulations (about distortion).
Last edited:
Thanks Christophe, I was going to answer for you but thought you could say it better. I assumed we were comparing simulated distortion numbers and as you say that is all they are and good for comparison and not much more. The final arbitrator is always actual physical measurements or subjective testing, simulations just tell us if we are working in the wrong direction.
I have seen this in advanced composite manufacturing in aerospace. Predictions of the shape of a part after molding never shows reality once the real part is made. You may know that you are in the ballpark or find something that wouldn't be obvious but your predictions in sim are just going to get you close. Until the delta between the sims and reality become much smaller there is no way around that conclusion.
I have seen this in advanced composite manufacturing in aerospace. Predictions of the shape of a part after molding never shows reality once the real part is made. You may know that you are in the ballpark or find something that wouldn't be obvious but your predictions in sim are just going to get you close. Until the delta between the sims and reality become much smaller there is no way around that conclusion.
Thanks a lot, Steven.
Yes, you understood well. My purpose was to share the little experiences i had with my favorite analog technology: CFAs.
Give tracks to help people to understand the way they work, and how to optimize them on a theoretical and practical point of view. And underline the pros and cons of various choices around this feedback topology.
It was the purpose of this thread, before it was moved to the Vendor's Bazaar.
For me, simulation programs are just super "slide rules". Save time during the design process, help to validate an idea, a schematic...
I am not more convinced by the superiority of Cordell models on others, comparing distortions with real world measurements. With CFAs, and mr Marsh can confirm this, LTSpice pretend distortions sometimes worse than real ones.
Well, i believe my work, all along with L.C., Mr Marsh, ostripper, et some others had helped the DYier community to understand CFAs were not fake or audiophile snake oil...
Two years back, it was impossible to talk about CFAs without to be instantly flamed by an army of VFAs and long tailed pair ayatollah ;-)
Yes, you understood well. My purpose was to share the little experiences i had with my favorite analog technology: CFAs.
Give tracks to help people to understand the way they work, and how to optimize them on a theoretical and practical point of view. And underline the pros and cons of various choices around this feedback topology.
It was the purpose of this thread, before it was moved to the Vendor's Bazaar.
For me, simulation programs are just super "slide rules". Save time during the design process, help to validate an idea, a schematic...
I am not more convinced by the superiority of Cordell models on others, comparing distortions with real world measurements. With CFAs, and mr Marsh can confirm this, LTSpice pretend distortions sometimes worse than real ones.
Well, i believe my work, all along with L.C., Mr Marsh, ostripper, et some others had helped the DYier community to understand CFAs were not fake or audiophile snake oil...
Two years back, it was impossible to talk about CFAs without to be instantly flamed by an army of VFAs and long tailed pair ayatollah ;-)
Last edited:
Of course. But this strikes me as a poor excuse not to do the best simulation that one can.
What I recommend is precisely to reduce the delta as much as possible.
When the simulation does not match the reality then one response is to say "oh well it's just a simulation".
This misses a real opportunity, the fact that they don't match has a reason.. This tells you that you have failed to model correctly some important element.
It is a chance to learn that there is some effect that you didn't know about.
LTspice has a potential numerical accuracy of more than 16 decimal places, vastly more than anyone can measure.
So there is no practical limit inherent in the simulator, only bad simulations.
Best wishes
David
We are still waiting for your accurate simulations. Why talking about round and round instead of doing-it ?
Last edited:
Yes, I read that. What I do not see is any evidence that the Cordell models have actually been used.
When I asked about this you pointed me to an ASC that does not use them.
When I questioned this you just repeated the incorrect claim.
So I see no evidence that you know how to use them correctly.
When it confirms your already formed opinions.
I have already published 2 articles in Jan Didden's excellent Linear Audio.
I am currently at work on the next couple.
And you have published? Please don't include Tweets, Facebook or other "selfie" publications but I am keen to read refereed work.
I think you misunderstand.
I commented on the problems in your work to help you improve it.
If you consider that criticism then that's your problem too, I will save my time.
I have co-operated with Toni "ASTX" on *real* measurements.
I have helped work on models and simulations and Toni has used his excellent equipment to check this.
His distortion measurement capability is between 10 and 100 times better than your number, and my claims have been confirmed.
You can read his thread for details.
Freudian slip
As we discussed previously, until your test are unbiased i.e. "Blind" then your ears probably do "lie"
Best wishes
David
the more incomprehensible is that 80% of the work was done: you got the ASC files.
It looks like you are looking only for a controversy ? This does not interest me.
Last edited: