Since this thread is about feedback, I wanted to ask a few questions about VFA and CFA topology.
I note some of the high end amps (e.g. Krell) use CFA whilst a lot of others use VFA.
There has been a lot of discussion about slew rates, Cdom, LTP currents, rise times, TIM, SID etc along the way in this thread either directly or via links to other sources. Along th e way there has been a robust discussion about zero f/back vs global feeback. Some folks have commented that the phase shift around the loop in a feedback amplifier is th e cause of the 'muddied' SS feedback amp sound. So, a low frequency signal with a fast transient or high frequency modulation is affected by this phase shift in a dynamic way. One particular thing I've noted about CFA is the very low phase shift input to output - and from this a assume (maybe incorrectly) that phase shift around the complete loop is also low. (NOTE: I talk in all cases here DC to 20KHz - so the classic audio bandwidth. I did some simulations, and depending on some of the variables I mentioned above, and th e feedback factor, a typical VFA has around 2 to 5 or 6 degrees, whilst a CFA has virtually zero.
It seems to me that CFA solve a lot of these problems, albeit introducing a few others - like power supply rejection in the front end. But maybe thats an easier problem to solve in th e big scheme of things.
Does anybody have a vew on this and why do some designers go for VFA and some for CFA?
Is this 'dynamic phase coherence' thing really a requirement for high end reproduction?
I note some of the high end amps (e.g. Krell) use CFA whilst a lot of others use VFA.
There has been a lot of discussion about slew rates, Cdom, LTP currents, rise times, TIM, SID etc along the way in this thread either directly or via links to other sources. Along th e way there has been a robust discussion about zero f/back vs global feeback. Some folks have commented that the phase shift around the loop in a feedback amplifier is th e cause of the 'muddied' SS feedback amp sound. So, a low frequency signal with a fast transient or high frequency modulation is affected by this phase shift in a dynamic way. One particular thing I've noted about CFA is the very low phase shift input to output - and from this a assume (maybe incorrectly) that phase shift around the complete loop is also low. (NOTE: I talk in all cases here DC to 20KHz - so the classic audio bandwidth. I did some simulations, and depending on some of the variables I mentioned above, and th e feedback factor, a typical VFA has around 2 to 5 or 6 degrees, whilst a CFA has virtually zero.
It seems to me that CFA solve a lot of these problems, albeit introducing a few others - like power supply rejection in the front end. But maybe thats an easier problem to solve in th e big scheme of things.
Does anybody have a vew on this and why do some designers go for VFA and some for CFA?
Is this 'dynamic phase coherence' thing really a requirement for high end reproduction?
Jan, it seems that you could listen to a fantastic sounding system.PMA said:
I think it is clear that such a quality is only possible if every component, every circuit detail, every parameter etc. has been evaluated for best music reproduction.
With ALL these factors optimized, everyone might be able to hear a (huge?) difference between say the first prototype and the final thing. Perhaps even in double blind tests.
How in the world could the developer of that equipment have reached such a quality level? Unfortunately, as every engineer knows, one change at a time is mandatory with a structured development approach!
With dozens to hundreds of single decisions regarding parameters, component choices, etc. etc. to be made?
Lucky strike? Seems rather improbable.
Measuring? No way, the differences are way to subtle for many permutations.
Listening? Sure. And here we are again: For sure very small differences when one single parameter has been changed.
You need to listen and discern very subtle differences.
Do you really believe that such an optimizing process would have been possible with double blind tests, when people is reporting that with such tests it is not possible to tell a difference between say an NE5532 and an OPA627??
I have not very good listening abilities. If there are subtle differences, I need to listen for several days to configuration A, and then switch to config B. If B sounds worse, it is very clear to me, immediately. If not, I have to live with B for a while, and then back to A. I could never be a professional developer of audio gear, I would need a lifetime just to evaluate all capacitors needed.. ;-)
Tino
zinsula said:
Clear? Why? Possible? Yes.
Maybe, maybe not. You don't know that. You are speculating.
Yes, I agree.
Agree. Of course, many of these hundreds of choices are connected to each other. For instance, if you choose supply voltage and bias current that immediately gives you a lot of values for resistors, for example.
No. You are again speculating, that the differences are too subtle for measuring. And of course measuremments are much, much more efficient, effective and repeatable if you have a large number of permutations.
Again, you ASSUME that there are subtle differences and therefore, you say, we need to listen. This is circular reasoning, and illogical and without any proof.
No, of course not. You can't optimize if you don't hear a difference. But again, you ASSUME that there was an optimization, you ASSUME that it was done by listening, you ASSUME that there is a difference between opamps, and from all this you conclude something. What value has that conclusion?
Jan Didden
...you ASSUME that I assume...janneman said:
...you ASSUME that I assume...janneman said:
Of course, thank you for reminding me that.janneman said:
I assume everything in life.
I assume that I feel better today than yesterday.
I assume that people like John Curl, PMA, Charles Hansen, Nelson Pass are better listeners than people like Jan Didden, Mr. Kleinschmidt or mikeks.
I assume that Tina Turner looks better than Patti Smith.
Thank you Jan.janneman said:
I think that was it. Brilliant.
Maybe a little presumptuous.
Ciao, Tino (I assume)
The simple A-B works (no problem that listeners do not know what is "A" and what is "B").
I say: "now you hear A", "now you hear B". Tell me when you want to switch back, tell me how long time you want to listen. Then please tell me what you hear.
I never try them to guess, like "I play something, is it A or B"? Such guessing stirs sense of the test.
I say: "now you hear A", "now you hear B". Tell me when you want to switch back, tell me how long time you want to listen. Then please tell me what you hear.
I never try them to guess, like "I play something, is it A or B"? Such guessing stirs sense of the test.
SY, don't think that blind testing is totally scientific, just because it is strict. To be truly scientific it has to count human psychology in large amounts. And being strict it stresses it. How can you discount the fact that you know more about your amp's performance after you have lived some time with it, connected many sources and speakers, listened to your beloved music relaxed in different mood swings, counting the known effects of different vintage wines?
salas said:
Salas,
I thought that the purpose of doing it 'scientifically' was to try to EXCLUDE human psychology as much as possble. Assuming it is ONLY the sound difference we are interested in, we would want to structure our test so that all other variables are excluded, no?
Jan Didden
Bonsai said:
Hi Bonsai,
You're right, current feedback solves a lot of problems. It provides for more bandwidth, higher slew rate and less phase shift. Therefore we used this technique in the second (NDFL) stage of the PGP amplifier.
Indeed, the downside is a lower PSRR, which means that you'll need a well regulated PSU. Using CFB in the input stage has another drawback: a high offset voltage, necessitating a servo to keep the offset at the output within reasonable limits.
In another preliminary design, see: http://www.diyaudio.com/forums/attachment.php?s=&postid=1367089&stamp=1196786670, the higher bandwidth is exploited by the Miller compensation between the VAS output and inverting input of the CFB stage (see: C6...8 and R17). Opposed to conventional Miller compensation, this scheme gives far less distortion, in particular at higher frequencies, as the input stage doesn't have to supply the current flowing through the Miller caps.
Not shown in this schematic is the clamping circuit to protect the amp from overdrive, but the higher bandwidth allows also feedback from the drivers (that is, when they are approaching the saturation point) back to the inverting input, resulting in very clean clipping and preventing overdrive of both the input stage and VAS. See: http://www.diyaudio.com/forums/attachment.php?s=&postid=1410411&stamp=1201261135
(blue is Vout, red is VAS current)
Cheers, Edmond.
PS: In a later version, C2, C3, R15, R16 are replaced by a series RC network (100pF, 150 Ohm) between VAS-output and GND.
''Salas,
I thought that the purpose of doing it 'scientifically' was to try to EXCLUDE human psychology as much as possble. Assuming it is ONLY the sound difference we are interested in, we would want to structure our test so that all other variables are excluded, no?
Jan Didden''
In my view, since the target is the human aural and soul pleasure, it is totally unscientific not to count for its adaptation period. That varies individually a lot.
I have seen that for me and others alike, having had many flight hours on consoles for live gigs, where evaluation and correction has to be scanning fast before performers or audience notice, is much easier to pick up flaws or differences in Hi-Fi quick, for instance. Many times I mention something early, and several days later a phone ring finally confirms what was initially strongly disputed.
''Don't you accept that it might be possible to make a change to your system and not hear a difference?
Jan Didden''
Yes, but not often. Almost never.
I thought that the purpose of doing it 'scientifically' was to try to EXCLUDE human psychology as much as possble. Assuming it is ONLY the sound difference we are interested in, we would want to structure our test so that all other variables are excluded, no?
Jan Didden''
In my view, since the target is the human aural and soul pleasure, it is totally unscientific not to count for its adaptation period. That varies individually a lot.
I have seen that for me and others alike, having had many flight hours on consoles for live gigs, where evaluation and correction has to be scanning fast before performers or audience notice, is much easier to pick up flaws or differences in Hi-Fi quick, for instance. Many times I mention something early, and several days later a phone ring finally confirms what was initially strongly disputed.
''Don't you accept that it might be possible to make a change to your system and not hear a difference?
Jan Didden''
Yes, but not often. Almost never.