Guys, I'm sorry to poop your party, but some of you don't get it. As JCX already suggested in his rather cryptic style, CFA or VFA, there's no secret sauce for better distortion performance in the topology.
CFAs are VFAs that allow the open loop gain to be controlled by the feedback network. As I showed elsewhere, the CFA equations can be deduced using strictly the VFA hij parameters formalism, and CFAs degenerate in standard VFAs at high closed loop gains). There are two main features that make a CFA topology "special":
a) Possible very high slew rates, independent on the ULGF. This is a direct consequence of the "current on demand" property of the standard CFA input stage.
b) CFAs can always be made unconditionally stable. It is very hard (if possible at all) to build a VFA with high loop gain at HF, which is also unity gain stable.
You may or may not like these features (I do, in particular b) above), but otherwise CFAs and VFAs are strictly equivalent. A long time ago, a guy named Cherry showed that there are exactly three ways to reduce an amplifier distortion: a) increasing the bias, so that active devices are in a more linear region (that is, dynamic variations of the bias points are smaller) , b) use distortion cancellation (that is, use circuit symmetry wherever possible, to cancel the even harmonics) and c) use negative feedback (including local feedback).
Perhaps a diamond buffer (has local feedback) can be made with lower distortions compared to a differential pair loaded with a current mirror (has local feedback and cancels the even harmonics), but there is no magic property of any of these topologies. They can arbitrarily be made with lower relative distortions, by increasing the local feedback (c) and/or running the devices hot (a).
Regarding the global NFB loop, there's again no magic in the CFA. Whatever the topology, and given a phase margin, the available loop gain ultimately depends only on the GBW and the compensation network order. Whatever some renowned designers are claiming, the fact that regular CFAs trades open loop gain for open loop bandwidth doesn't provide any advantage when the feedback loop is closed. The claim that somehow CFA's could provide more loop gain compared to VFAs is false. There are a large number of CFA op amps with stunning distortion performances (like ppm in the MHz range), but that's not because they are designed as CFAs (which is only because of the speed requirements) but because these op amps are built in processes providing a GBW in the GHz range.
CFAs are VFAs that allow the open loop gain to be controlled by the feedback network. As I showed elsewhere, the CFA equations can be deduced using strictly the VFA hij parameters formalism, and CFAs degenerate in standard VFAs at high closed loop gains). There are two main features that make a CFA topology "special":
a) Possible very high slew rates, independent on the ULGF. This is a direct consequence of the "current on demand" property of the standard CFA input stage.
b) CFAs can always be made unconditionally stable. It is very hard (if possible at all) to build a VFA with high loop gain at HF, which is also unity gain stable.
You may or may not like these features (I do, in particular b) above), but otherwise CFAs and VFAs are strictly equivalent. A long time ago, a guy named Cherry showed that there are exactly three ways to reduce an amplifier distortion: a) increasing the bias, so that active devices are in a more linear region (that is, dynamic variations of the bias points are smaller) , b) use distortion cancellation (that is, use circuit symmetry wherever possible, to cancel the even harmonics) and c) use negative feedback (including local feedback).
Perhaps a diamond buffer (has local feedback) can be made with lower distortions compared to a differential pair loaded with a current mirror (has local feedback and cancels the even harmonics), but there is no magic property of any of these topologies. They can arbitrarily be made with lower relative distortions, by increasing the local feedback (c) and/or running the devices hot (a).
Regarding the global NFB loop, there's again no magic in the CFA. Whatever the topology, and given a phase margin, the available loop gain ultimately depends only on the GBW and the compensation network order. Whatever some renowned designers are claiming, the fact that regular CFAs trades open loop gain for open loop bandwidth doesn't provide any advantage when the feedback loop is closed. The claim that somehow CFA's could provide more loop gain compared to VFAs is false. There are a large number of CFA op amps with stunning distortion performances (like ppm in the MHz range), but that's not because they are designed as CFAs (which is only because of the speed requirements) but because these op amps are built in processes providing a GBW in the GHz range.
I checked, the Cyrus 3 (50W output power) uses the same strategy, one pair of 20MHz devices (2SD1047/2SB817) and brutal shunt compensation (here, 300pF).
The Marantz PM94 is VFB.
Hi manso,
A complex monster?
Are you talking about the 1st or the 2nd schematic in post 1387?
Cheers, E.
Both, to be able to take advantage of such a inputstage the outputstage needs similar attention. Those inputstages gain very little if anything at all when the output is the main distortion contributer. The ones you show on your site are elegant and good case study but the complexity for a build is not to my liking, can you obtain the same distortion figure with just a triple outputstage ?? My guess one needs the other. The VFB input circuit is in the ballpark though.
Hi Manso,
Perhaps my post #1387 wasn't clear enough. I was responding to Esperado who said: "many designers witch have tried-it in real world never will return back to VFA" Therefore I couldn't resist the temptation to tell him that I'm one of the (few???) designers who reverted from CFA to VFA again. Why? If you strive to low disortion, low noise, high slew rate, high PSRR, low offset and a symmetrical topology, then VFA offers better opportunities. Besides, the VFA front-end (fig.2) is less complex than the CFA counterpart (fig.1): 14 vs 18 trannies. Compared to the PGP front-end (42 trannies!) or Bob Cordell's HEC front-end (19 trannies), I would characterize my VFA front-end as a "complex monster".
>can you obtain the same distortion figure with just a triple outputstage ??
No! Of course not, but that is another story. My point was that CFA is certainly not better than VFA, at least for audio applications.
Cheers, E.
Perhaps my post #1387 wasn't clear enough. I was responding to Esperado who said: "many designers witch have tried-it in real world never will return back to VFA" Therefore I couldn't resist the temptation to tell him that I'm one of the (few???) designers who reverted from CFA to VFA again. Why? If you strive to low disortion, low noise, high slew rate, high PSRR, low offset and a symmetrical topology, then VFA offers better opportunities. Besides, the VFA front-end (fig.2) is less complex than the CFA counterpart (fig.1): 14 vs 18 trannies. Compared to the PGP front-end (42 trannies!) or Bob Cordell's HEC front-end (19 trannies), I would characterize my VFA front-end as a "complex monster".
>can you obtain the same distortion figure with just a triple outputstage ??
No! Of course not, but that is another story. My point was that CFA is certainly not better than VFA, at least for audio applications.
Cheers, E.
For what reason do you call that brutal. Its neccesary or you end up with too high ULGF . You can degenerate the diamond or reduce bias current but not advisable for noise. I use 330pf and have ULGF of near 7 Mhz and end up with a phase margin of 85 degress. In sim the cyrus is a little over 2 Mhz. That amp has LT simmed phase margin of 89 degrees. Take the cyrus use a beta enhanced vas and see THD20 drop to below .001 figures with similar margins.
In essence I agree with your other post although some factors werent entirely addressed in full but lets leave it at that.
Not the pm94, the PM9S2 or PM11S2.
CFA vs. VFA ... interesting. (like tube vs. silicon )
I left the option for the "brutal" shunts in DIYA's AB amp (badger). Either high Ft drivers and/or - the use of higher UGLF , the option exists (not too brutal).
Used on roughly half of the Japanese audio EF2 stages.
I'd would like to try a CFA amp , if I had to use a "brutal shunt" , so be it.
I like having just a 12 semi front end , easy to source .... I just might be a
VFA "fanboy" , as well.
OS
)I left the option for the "brutal" shunts in DIYA's AB amp (badger). Either high Ft drivers and/or - the use of higher UGLF , the option exists (not too brutal).
Used on roughly half of the Japanese audio EF2 stages.
I'd would like to try a CFA amp , if I had to use a "brutal shunt" , so be it.
I like having just a 12 semi front end , easy to source .... I just might be a
VFA "fanboy" , as well.
OS
To better understand why I call it "brutal" you need a lecture in Miller compensation vs. shunt compensation (pole splitting theory, etc...).
I'm afraid we are talking different languages, what you are saying doesn't make much sense. 7MHz ULGF with 330pF shunt compensation at the VAS/TIS is out of this world. Are you sure you are not confusing ULGF with UGF?
Good night, it's very late here.
Manso, your posts suggests you have some CFB circuits that are very simple and have good performance.
Can you post LTspice *.ASC of one of them? Please
______________
I too would like to see how you get 7MHz ULGF with 330p stuck on the VAS output ... grovel grovel
Last edited:
I counted the CSS's .A typical cfb front including a beta enhanced vas can be with only 8.
OS
Manso...
Can you post LTspice *.ASC of one of them? Please
I asked for even a few numbers and have had no response yet, but I live in hope.
And you haven't replied to my enquiry about the "low noise CFB" that you mentioned.
Best wishes
David
Could we just say that CFB has a set of nice technical virtues which combined with good subjective impression makes it a worth-while and relevant alternative for VFB? I am not suggesting that production of VFB amps should cease. Only that we deserve the alternative.
It's like PC monitors. It is impossible to find them with 4:3 ratio anymore. Only 16:9 which is more suited for multimedia and movies than for PC work. We are not watching movies all the time! I am not the only one that prefers 4:3 but the industry is foisting 16:9 and that's irritating.
It's like PC monitors. It is impossible to find them with 4:3 ratio anymore. Only 16:9 which is more suited for multimedia and movies than for PC work. We are not watching movies all the time! I am not the only one that prefers 4:3 but the industry is foisting 16:9 and that's irritating.
Hi Manso,
Are you saying that your boss asked you to build the super-TIS front-end?
If so, he didn't ask me permission to use my intellectual property.
Anyhow, thank you for refusing.
Cheers, E.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.