Is the CFB topology superior, and why?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
What would be easier and more accurate ? Design by ears ?..:rolleyes:

I have to gain experience in relating simulation with reality (accuracy is one thing, but relationship between parameters and perceived sound is another). Some things I understand, more things I don't.

Just like the feedback story. Which one is better, none, a little, or much of it? And not everything translate itself to THD. Take for example a mosfet amp, you can make a low THD amp with less current drive. Where does this current drive translate itself in the global concept of performance?

Another example is the basic bootstrap of Lin topology. We can arrange the bootstrap so that the distortion spectrum falls monotonically or make the odd order becomes dominant without changing the THD considerably, but the sounds are different.

Then there are current mirror and cascade. These two will give us very low THD. So why not everyone use it in every chance (like you?)

Processors with roughly one billion transistors are first simulated
and they work wich is the prove that simulators are accurate enough
to model real world behaviour of very complex electric devices,

No, processors are not that complex at all, if you think about it.

Indeed , had the sims displayed a superiority of CFB that its
promoters would have gladly embraced the conclusions
without even checking but alas , nothing of the sort did
happen , so what is left is to brand the simulations as not
significant....

Who knows, may be Elvee is looking for the answer himself. My prediction is that the root cause is the phase. May be something is wrong with what people think about group delay.

I think simulations are significant. It is just not that easy to interpret the results.
 
Wahab,

Esperado mentions Rod Elliott's article about IM. From that article you can learn that if circuit has symetrical distortion and input is fed with symetrical waveform the whole truth about IM will not be told. Conclusion is that we do not measure IM properly.

In order to measure it properly we need at least asymetrical signal at input. The questions:

1. Does your signal generator has option to switch from symetrical waveforms to asymetrical?
2. Does simulator have suitable algorhythms to predict for both possibilities?

If answer to these questions is no, than our standard measurements are not able to cope with all tasks and to explain all the differencies we hear in amps. I am amateur that does not have even the simplest CR scope, let alone digital one, and I still do not use simulator, though I installed one on my PC. I expect answers from real experts and dogmatism does not help us to reach them. May be that we really need huge bandwidth and slew rate?
 
anecdote is not data - is not actionable engineering information

I don't see any reason to accept the assertion that CFA has any predictable "sound" vs VFA

in fact I recall a few years ago you couldn't mention CFA here without one "Guru" interjecting that they sound awful for audio


where are the controlled listening tests? - how do you explain the Carver Stereophile Challenge result that professional audio reviewers - thousands of hours of listening experience as their paid day job - couldn't tell a tweaked $600 SS amp from their selection of a "SOTA" tube amp, in their offices, with their speakers, source in blind listening?

do you expect the CFA vs VFA "sound difference" to be as big?


please point to "error of our ways" in modern audio power amp design - what human audibly significant errors are we missing when we can use MHz 20-bit ADC, DAC to measure?

anyone read Cabot's articles on audio test signals? - citing Rod Elliot is laughably lame - just go read the current AP analyzer manual, cruise pull-down menus

http://www.google.com/#hl=en&sugexp...ad04a2e2a7787&bpcl=37189454&biw=1920&bih=1069

everyone points to "accuracy", "the wire with gain" as the endpoint for audio amps - can you clearly articulate where the best are falling short of human audio thresholds?
 
Last edited:
Processors with roughly one billion transistors are first simulated
and they work wich is the prove that simulators are accurate enough
to model real world behaviour of very complex electric devices , but still , here , there s lot of people that think that it s up to them to predict
the behavior of basic circuits using a handfull devices.


Our foundry spends millions squid a year to get accurate behavioral models for their analog building blocks, and you claim a few free device models found on the Internet are supposed to deliver the same absolute performance metric?

Perhaps you can use free simulators and free models to ballpark some topologies, or confirm performance trends, or do some rough comparisons, but absolute performance results, or comparing at the fourth decimal? In your dreams...
 
in the same vein Harmon, Toole spent $millions on decent blind audio testing, published in peer reviewed journals

why are we even considering personal anecdotes from people who may never have had two copies of the circuits under consideration in the same room at the same time, much less 0.1 dB level matched, output impedance padded to equal levels, testing double blind?

from perceptual psychology, psychoacoustics we are very certain naive “just listen” commentary is more useless, more unreliable than sims
 
Our foundry spends millions squid a year to get accurate behavioral models for their analog building blocks, and you claim a few free device models found on the Internet are supposed to deliver the same absolute performance metric?

Perhaps you can use free simulators and free models to ballpark some topologies, or confirm performance trends, or do some rough comparisons,


I know that foundries have a the worforce to design and extract
extremely accurate modelisations of their processes devices caracteristics
at a completely different level than the models we re using but at the
diyer level what we have in hands is enough to have an accurate
idea of a circuit behaviour with acceptable error margins.

but absolute performance results, or comparing at the fourth decimal? In your dreams...

Please spare us these kind of straw man , putting arguments in the
mouth of people and then drawing ill conceived conclusions....
 
1. Does your signal generator has option to switch from symetrical waveforms to asymetrical?
2. Does simulator have suitable algorhythms to predict for both possibilities?

If answer to these questions is no, than our standard measurements are not able to cope with all tasks and to explain all the differencies we hear in amps. I am amateur that does not have even the simplest CR scope, let alone digital one, and I still do not use simulator, though I installed one on my PC. I expect answers from real experts and dogmatism does not help us to reach them. May be that we really need huge bandwidth and slew rate?

Whatever wave form can be used but it will always be a sum
of sines by principle....

Depending of its convergences capabilities it should predict
any behaviour , the only limit as pointed by Waly is the accuracy
of the models.

As for Rod Elliott article , thank you , that s very interesting
but i m afraid that it will only give more weight to the high
NFB ratios camp.

Funnily , he noticed that :

the simulation and real life are remarkably close
 
(...)but at the diyer level what we have in hands is enough to have an accurate idea of a circuit behaviour with acceptable error margins.

Please spare us these kind of straw man , putting arguments in the mouth of people and then drawing ill conceived conclusions....

Unless you define "accurate" and "acceptable error margin", you may want to use your own advice above.
 
Unless you define "accurate" and "acceptable error margin"

This was discussed at length , including with you of course..

IIRC last time you questionned the validity of power mosfets
models that do not model the subthreshold region and then drawing
erroneous conclusions about the consequence in real world...

Now , since you pretend to be in the know just tell us what
is your estimations of the current models probable errors
margins when computing circuits caracteristics.

As for me , errors margin with 2 ratio error when computing
THD for exemple is already correct.
 
As for me , errors margin with 2 ratio error when computing THD for exemple is already correct.

Not sure I understand, but you think you can (using the free stock models) predict absolute THD simulation results within +/-1% precision?

That would be an extraordinary claim. The multi-million dollar models coming from our foundry are guaranteed to provide 5% accuracy against any valid configuration (though, for all practical purposes, they do better, probably around an average 2%). The whole art of designing analog circuits is to make those 5% irrelevant for the final result.

Sorry for the OT, I'm off my soapbox now.
 
Last edited:
Not sure I understand, but you think you can (using the free stock models) predict absolute THD simulation results within +/-1% precision? That would be an extraordinary claim.

Of course it s not +-1%.
I mean by a 2 ratio -50% to +100% error in THD evaluation.

Fortunately gain and phase response computation are way
more accurate for obvious reasons.

The multi-million dollar models coming from our foundry are guaranteed to provide 5% accuracy against any valid configuration (though, for all practical purposes, they do better, probably around an average 2%). The whole art of designing analog circuits is to make those 5% irrelevant for the final result.

Sorry for the OT, I'm off my soapbox now.

I had no idea about foundries models accuracy , to be honnest
i thought that it was better than the numbers you re quoting
given the relative complexity of V/ULSI circuits.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.