CFA Topology Audio Amplifiers

Don't worry it is techincally measured, but it is usually covered by major measurements, because it just a very small phase most in damping timing. That is why it depend on overall circuit.

Bonsai made a little summary a few post earlier and all we can conclude
is that unless i missed something the caracteristic of thoses CFBs without
exception is to have way more limiting factors than eventualy exploitable strengths ,
all drawbacks that are more or less successfully circumvented by using more complexity
wich end being often greater than for the decried VFAs and still this is not enough to display
more than average perfs.
 
Bonsai made a little summary a few post earlier and all we can conclude
is that unless i missed something the caracteristic of thoses CFBs without
exception is to have way more limiting factors than eventualy exploitable strengths ,
all drawbacks that are more or less successfully circumvented by using more complexity
wich end being often greater than for the decried VFAs and still this is not enough to display
more than average perfs.


I tried to give a brief technical summary of what makes a CFA a CFA. You seem intent on subverting the discussion on this thread to an opinion slinging session with no technical content whatsoever.

You can discuss why you think VFA is superior in the VFA thread which I opened for you. Why not also consider designs without feedback, tube amplifiers, and class D. I am sure there is plenty to go at over there. You can add blameless vs fully balanced too.
 
For playing the numbers - CMAmps

FYI -- I made a little challenge months ago --- make a circuit as simple as possible, with no more than 8 transistors max (less is best) and have thd well below .001% into low Z. Might look at it and the excellent discussions there as well... overlapping info - You can go to the Marsh Headphone Amp discussion to see it and the others FFT on it and many good designers commnets and construction. It needs no dc servo, either.

The CMA topology looks like this and an article was published in detail in Linear Audio #3. More info in the forum and Linear Audio than i can put up here.

Thx-RNMarsh

RM-CMA-2.jpg

And for background and details beyond the scope here:

Some of my High-End foundations stuff:

Hi-End Reading.jpg
 
Last edited:
Bonsai,
It doesn't matter that we want to have a discussion about the good and bad aspects of the cfa topology there will always be those who will knock whatever you say or show. It is to bad but let's try and keep the subject on track anyway. The best we can do is ignore most of the naysayers.
 
Towards a SOTA design/Circuit/Topology -

I think the same... any attempt to talk about VFA or bashing etc here is not to be responded to... totally ignored. Invisible to the eye. not there.

My hope is that after flushing out the CM topology/characteristics and with the best attributes of VFA and CFA together, we will be able to melt into an optimimum design and SOTA.... Perhaps for decades to come. And, knowing the design trade-offs for special apps.

[but, its going to have to be really, really, really good] Stay focused. Stay tight. And, enjoy the journey.


Thx-RNMarsh
 
Last edited:
I think the same... any attempt to talk about VFA or bashing etc here is not to be responded to... totally ignored. Invisible to the eye. not there.

My hope is that after flushing out the CM topology/characteristics and with the best attributes of VFA and CFA together, we will be able to melt into an optimimum design and SOTA.... Perhaps for decades to come. And, knowing the design trade-offs for special apps.

[but, its going to have to be really, really, really good] Stay focused. Stay tight. And, enjoy the journey.


Thx-RNMarsh
This is what I've been trying to do for years already. Trying to pull together the best elements from various topologies and develop my own addition to get the most out of those elements. The result is the amp surrounding my VAS, where my VAS is a key element in the much improved performance.

I ended up with a high-current single LTP, the currentmode driven VAS which is an actual pure current-to-current amplifier. It's only the the mirrors that allow the cascode devices to perform the actual voltage output. This VAS is not limited by supply rails and has no limit on input current, be it for the pushing or pulling input. That's exactly (With an ideal output stage) why my VFA Inverting LTP / VAS combo achieves speeds similar to CFA.

I tried CFAs, I tried blameless type topologies (with the CCS VAS, non-push-pull), fully symmetric topologies and what not more and eventually decided "the best from all worlds" is to have a single LTP in inverting configuration, a fully symmetric/complementary current gain stage with cancelling properties and a push-pull voltage output, where this current gain stage input has its DC operating point exactly in the middle of the LTP push-pull output to allow for maximum current/voltage swing without clipping and without saturating the VAS current gain stage.

Result? A VFA amp with CFA-like performance (in speed/bandwidth), though with all the benefits of VFA. Closed loop bandwith of this IPS/VAS is similar to that of a CFA, so is the slew-rate. And due to this topology, there are no practical limits to compensation schemes to chose from.

Speed was a factor (see my signature), but ofcourse this is not the only parameter what makes an amp. The IPS/VAS I developed provides all the speed I'd need while exploiting the strength of each sub circuit.

I'm very happy with it and a prototype will be in the works some time soon. Theoretical sub-ppm @ 20K, well over 130dB loop gain up to 1KHz, 90dB at 20KHz.

If CFA was the 'best' of many worlds combined, I'm sure I would have ended up with it - but I didn't.

In conclusion, I think you'll be hard pressed to succeed creating a CFA based amp that has the best and best distributed parameters combined.
 
Last edited:
FYI -- I made a little challenge months ago --- make a circuit as simple as possible, with no more than 8 transistors max (less is best) and have thd well below .001% into low Z. Might look at it and the excellent discussions there as well... overlapping info - You can go to the Marsh Headphone Amp discussion to see it and the others FFT on it and many good designers commnets and construction. It needs no dc servo, either.

The CMA topology looks like this and an article was published in detail in Linear Audio #3. More info in the forum and Linear Audio than i can put up here.

Thx-RNMarsh

View attachment 366143

And for background and details beyond the scope here:

Some of my High-End foundations stuff:

View attachment 366144
How are JFETs CMA considering they're voltage driven devices?
 
Hi -- I hope more minds will try doing the same thing and see what developes. Many people will learn about CMA design in the process.

Myself -- in 1980 (pre Comlinear/Elantec). I had much less than --100dB THD+N with only 20dB of GNFB. Not cascoded, either. What would it do today with more GNFB?

THx-RNMarsh
 
Last edited:
Different approach - same low distortion -

How are JFETs CMA considering they're voltage driven devices?

Originally, it was developed for BJT and my original 1980 is all bipolar. This one just biases everything off the front end without seperate ccs/biasing circuits to simplify. But the distortion is extreamly low for such a simple circuit And rock stable temp/drift-wise. Note too -- 25Khz open-loop bw. Only 62dB open-loop gain.

Very low distortion via a different approach and few parts. Very stable.

[But only designed for headphone amp.]


Thx-RNMarsh
 
Last edited:
FYI -- I made a little challenge months ago --- make a circuit as simple as possible, with no more than 8 transistors max (less is best) and have thd well below .001% into low Z. Might look at it and the excellent discussions there as well... overlapping info - You can go to the Marsh Headphone Amp discussion to see it and the others FFT on it and many good designers commnets and construction. It needs no dc servo, either.

The CMA topology looks like this and an article was published in detail in Linear Audio #3. More info in the forum and Linear Audio than i can put up here.

Thx-RNMarsh

View attachment 366143

And for background and details beyond the scope here:

Some of my High-End foundations stuff:

View attachment 366144
What do you know, a Jfet FE.
 
Hi -- I hope more minds will try doing the same thing and see what developes. Many people will learn about CMA design in the process.

Myself -- in 1980 (pre Comlinear/Elantec). I had much less than --100dB THD+N with only 20dB of GNFB. Not cascoded, either. What would it do today with more GNFB?

THx-RNMarsh
Recreate the circuit with modern devices, in LTspice and see.. 🙂 Always nice to learn about circuit variants.
 
That the point of all the controversies in audio: Nobody knows exactly.

I believe (how to prove ?) our ears are VERY sensitive to energy.

Well, nowadays, we are able to sample before/after musical signals and print the differences. But this gives no informations to how and why.

That is all the question of this topic. We are several to find differences in our listening experiences between CFA and VFA. And something constant in presentation.

All the interest of this thread is to figure out the differences, and try to correlate those with our listening experiences.

Lets hear what Wahab had to say:
Bonsai made a little summary a few post earlier and all we can conclude is that unless i missed something the caracteristic of thoses CFBs without
exception is to have way more limiting factors than eventualy exploitable strengths, all drawbacks that are more or less successfully circumvented by using more complexity
wich end being often greater than for the decried VFAs and still this is not enough to display more than average perfs.

Problem here is some people like Christophe can observe audible performance and see that numbers are too complicated to make a perfect correlation between the two.

OTOH, some people like Wahab cannot observe anything other than oscilloscope figures. Not to mention that oscilloscope figure is different than TrueARTA figure.

People like Christophe is actually open minded. If only Wahab had the ability to logically and technically prove or show where Christophe has gone wrong. But Wahab is not of that kind, and he is a pure engineer who cannot observe what Christophe can. So it is a deadlock.

Technically, I agree more with Wahab. But can he be more open mind and think if there is anything that he cannot observe?

For me, most if not all debates such as "which one is better A or B?" are stupid. A or B should be just tools or part of the bigger process.

May be I can simplify or break down this CFA vs VFA issue into smaller case such that it is easier to understand. It is now FB vs no FB amp.

Can you understand why some people prefer no feedback amp? Can you understand why feedback is necessary? Can you understand why it is a necessary evil? Do you know what can be improved by FB? Do you know what are the "evils"?

I see CFA as the middle solution between FBA and NFBA. In my view, the best implementation of CFA is in a situation where both FBA and NFBA are not optimal. It is in high power class B amp where we can accept a relatively high thd associated with "not enough feedback" amps.
 
FYI -- I made a little challenge months ago --- make a circuit as simple as possible, with no more than 8 transistors max (less is best) and have thd well below .001% into low Z.

I have made similar circuit but with bjt input ala vssa. 0.001% is a very impressive number but it is a jfet input pair so it must be simulation only. It is hard to predict jfet specification especially for matched pairs and this topology requires that. At least that the case with bjt input.

ADD:
How is the performance of the bjt version of yours?
 
Last edited:
I see CFA as the middle solution between FBA and NFBA.
I agree with this. Non feedback amplifiers or preamplifier's require, in general, for they sound optimal, a lot of non industrial processes. Best parts (expensive). Matching and pairing. Fine tuning. I'm thinking to the Blowtorch, as an example.
Feedback topologies are more industrial, as they allow the use of average components for most of the parts and lies more on the design. It seems that CFA need special care to its input stage (and feedback resistance's quality) somewhere between the two worlds.
The same feeling can be experienced, in a very subjective manner, about some "general character" in the way each topology reproduce music.

It was a time, in the audio world, when Feedback was (wrongly) considered as evil. Good writers had explained how TIM occurs and how to be rid of it. Yes, but it seems to me that something always remain even when you minimize an effect to an infinitesimal ratio. Homeopathy ?.
Need for speed in closed loop is, on my opinion too, the main key, and CFA is an easier tool for it than VFA.

Measurements numbers. Oh, Lord ! Who really know what they means ? Their real correlation with your listening experience ? Of course, an amplifier will never be better than its numbers and we all try to minimize them, as much as possible. But, as L.C. noticed, they are only tools to help you to improve your work.
It is like lenses for photography. The measurements tells-you few about their character. Looking at the photos tells you more. Jut measurements will help-you to avoid too bad ones.

I really hope this topic will help us all to better understand the influence of speed (slew rates), various harmonic and IM distortions etc. in our listening experience. If we knew perfectly this, a computer program would be perfect to design a near perfect amplifier.

The main difficulty is most of the measurements are done with constant signals, while music is made of transients. This quest for good transients, reproduction is something you cannot totally achieve with your oscilloscope. In this domain, CFA behave (in general) better than VFA. Better slew rates. This can explain why we are several, here, to prefer them most of the time. Impossible to understand, it seems, by some, the eyes stuck at their distortion numbers.
Well, i'm sure, at the end of this collaborative work, we will know better how to make better amps, chose the best topologies for what we are looking for, and be able to make VFA amplifiers witch sound as good as CFA, and CFA ones witch measure as good as VFA (or the contrary for those who think the contrary 🙂
 
Diamond buffer as subtraction node is not an option. Never like them anyway, tend to be 'nervous' at HF and the sound is consequently 'erased' like.
This remark, by example, interest-me a lot.
I'm curious to know better what are your feelings, L.C. and the reasons that have led you to this conclusion.
You know how it happens, you try some topology, it don't give you the expected listening results, you don't have the reflex to try it in your next work..
I had a similar listening experience, something empty in the reproduction, like Darlington, sometimes, so i'm not experienced a lot in Diamond buffers. They can measure very good. Where is the problem if any ?
One node in excess ?
 
Last edited: