CFA Topology Audio Amplifiers

Nothing to do with audio applications. "Higher frequencies" means 1 MHz and over. E.g. for DSL drivers (where HF distortions are critical and gain is < 10), CFBs almost have exclusivity.

Actually it does have to do with audio, Texas further explains that CFAs dont suffer from TIM whithout mentioning the word audio.
In terms of integrated ICs the 1 Mhz bandwidth might apply but not in the real world with discrete circuits. Here VFAs have to make use of advanced compensation methods to be able to match and better CFAs performance from around 1 Khz. This implies more complexity although not in terms of active devices.

See Texas TPA6120A2 datasheet and apllication notes.
 
Maybe I'm envisioning a different topology, but it would seem for the "standard" CF the asymmetry of P and N would account for it. If we're talking about voltage feedback and differential pairs, yes, if common-mode distortion is minimized then I agree we null even-order. I don't see how one achieves that with current feedback configurations of the typical sort, unless you go with a balanced pair of "diamond" input stages and preserve balanced throughout.

I do like the ideas around the very simple circuit with the independent feedback to each transistor, as that has the potential of a trim of certain components for ~nulling even-order. It suffers from a variety of other issues though.

Thinking about all of this, particularly the notion that the CFA is inherently compromised for distortion performance at lower frequencies compared to the VFA, I have some ideas I may present soon, that I haven't seen before but of course may well exist elsewhere. As I say I came to the thread late.

Balanced pairs of diamond buffer inputs are long in production by opamp manufacturers and marantz use the topology in their top of the range amps and opamps.

The diamond buffer is inherently more linear than a LTP. Ive shown documentation regarding the point and Edmund has shown this with sims. The diamond buffer displays the same 2nd harmonic distortion as a LTP but when one compares 3 rd harmonics the CFA is about a 1000 x better. The diamond buffer in CFAs is compromised for THD at lower frequencies only due to the topology having lower loop gain and nothing to do with the linearity of the input stage used. Its as a result of the better input stage that CFAs have very good performance despite its lower loop gain.

Looking forward to seeing new ideas if they involve CFAs. There are many that opamp manufacturers use to obtain performance seen with parts like TPA6120 and LM49743. Im not sure whether its allowed to show them here on the forum although Scott Wurcer seems to think its ok as many of the patents have expired.
 
"TIM" IS an IMD product - in my 2-tone fft plots I look at IMD sum and difference products

any IMD product could be the sum of the "TIM" component and "AM" type IMD - but their isn't cancellation - they are in quadrature

so the unknown amount of "TIM" can't exceed the amplitude of the fft plotted IMD magnitude

with 3 decades of the Otala inspired controversy there are no controlled listening demonstrations that "TIM" IMD is any more audibly objectionable than "AM" IMD

so I claim IMD product distortion plots below human hearing threshold in quiet is a strong indication that "TIM" is under control – despite the high, “sloping loop gain” of 2-pole compensated (VFA) amps


and some may have missed my input Q current comparison - at the peak of the 19+20 kHz envelope my VFA example has ~16 uApp out of 2.5 mA bias - vs the VSSA 150 uA out of 1.8 mA bias

high loop gain linearizes the LTP thru having smaller diff signal – 10x less signal/bias ratio with a dominant 3rd order nonlinearity nicely makes up the factor of 1000 some want to credit CFA input with - but I prefer to use a lot more that just one decade extra loop gain to beyond 20 kHz than Otala inspired amps



patents are public documents - may be freely reproduced, discussed freely - that is the point of the "trade" - patents are granted to encourage teaching, spread of the techniques
 
Last edited:
JXC i would accept your sims if you didnt use the baxandall pair, the VFA probably has higher open loop gain or did you equalize. How about showing the CFA using the same. I note youre a fan of yamaha s favoured VFA topology.

I took note of the current comparison but that was using the VSSA topology, I wonder whether the diamond will display the same. The VSSA may be possible for use but is compromised in many ways. Bias currents are not stable and vary with temp and other problems. AudioLabor was one company that used the topology in the 1980 s envisaged by JLH, that took measures to stabilize operating currents.
 
This is from Texas Instruments by Xavier Ramus:
Distortion Issues
• At lower frequencies, the lowest distortion will be given by voltage feedback amplifiers
– We believe this is due to a linearity floor set by the error sensing point in the CFB topology. The CFB inverting input linearity sets a floor to distortion much higher
than the best VFB designs. This is normally a 2nd harmonic term.
– CFB will, however, give relatively constant distortion vs. Gain setting and hold
better numbers to higher frequencies due to considerably more slew rate margin.
Tommy, where is this quote from? Do you have a link to the article?
Waly said:
Nothing to do with audio applications. "Higher frequencies" means 1 MHz and over. E.g. for DSL drivers (where HF distortions are critical and gain is < 10), CFBs almost have exclusivity.
Waly, my own naive experiments with simple circuits suggest this is true even at audio frequencies.

Brad, Waly & yus other gurus, is this what I surmise in #682? .. ie that the 'input emitter followers' in Diamond IPS, being 'out' of the feedback loop set a lower limit to THD? But for the 'symmetrical' topologies that CFA seems to demand, the remaining products should be odd.

Brad, does this mean you think the TI statement is just generalizing on the difficulty of matching PNP/NPN rather than anything inherent in CFA topologies?

I'd still like to know if the simple CFA IPS like VSSA or my #500 circuit has similar limitations.
_____________________

Waly, wahab, there is no need to indulge in the 'yus are idiots' stuff that the semantic pedants are prone too. Both your contributions are valued.

wahab, it would be nice if you showed some circuits. My small brain has problems with complicated verbal descriptions. If you have LTspice *.ASC models, so much the better so everyone can play.
_____________________

While I relish the sight of supa dupa VFA examples and am far too eager to discuss these, IMHO, the real gain in useful human knowledge will be better simple CFAs. I would really like those versed in the art like Waly & wahab to focus their skill & experience on making these.

Something with similar complexity (13 devices) to my #502 VFA example.

wahab, I concede that it might be easier to make supa complex VFAs. After all, my #499 design is 1ppm 20kHz 50W 8R with only slightly more compexity. (Double the size of C1/3 if you are unhappy with PM & GM. It's still 1ppm) I don't know how to take CFAs to that level. 😡

But to me, it looks like simple CFAs have an advantage over simple VFAs on all counts.

I'm laboriously trying DTMC en other supa techniques (courtesy Bonsai, Edmond etc) on my simple naive CFA topologies but its slow.

And I'm still waiting for some reason to use Diamond IPS. 🙂
 
jcx,

I expect with lower loop gains you will drive the input stage harder, hence the input stage p-p differences you see can be largely accounted for. However, if you put a step function in, you are likely to see a larger error signal across the VFA input pair. Now, if the VFA is decently comp'd you are ok, but in a CFA you don have these issues. And, as Richard Marsh has noted, the CFA input stage provides current on demand into the TIS. I did some sims recently and to push my CFA input into class B requires truly heroic input overload signals. In practice, you may be able to get a few ppm better distortion out of a VFA, but usually the CFA will win on bandwidth and slew rate - important aspects of overall performance in my book. And, generally they are simpler.
 
Brad, does this mean you think the TI statement is just generalizing on the difficulty of matching PNP/NPN rather than anything inherent in CFA topologies?
That's my simplified surmise, although there are all the other not-perfectly-symmetrical subcircuits in a typical topology which will complicate things. Perhaps one should at least simulate, or do a Gedanken experiment, to illustrate this. This would take two alternate polarity CB bipolar stages with I source/sink loading and drive them from equal source impedances. Comparing the collector outputs, are even harmonic components equal? Surely there is some opportunity for cancellation, the issue is how much.
This is crude to some extent as it sidesteps the mutual loading of a real complementary CB configuration.

But compare to an LTP with matched devices. No prob, and the particular transcendental function corrected with parasitic resistances does indeed have vanishing even order except for common mode effects.
 
"And I'm still waiting for some reason to use Diamond IPS. "

It's elegant, its very linear, it's fast and it supports wide band widths. What more do you want?
Alas, my naive efforts seem to show no advantage .. in fact worse performance than simple VSSA type IPS like #500

There are a number of subtleties that need to be watched. eg The Diamond needs to run each stage at 2x that of the simple CFA IPS so takes 4x the current for similar input gm. But even accounting for all that, the linearity is still worse.

Though I've said Diamond IPS is very linear, the simple CFA IPS is also very linear and I've taken advantage of this in very low noise stuff (not power amps though).
 
Last edited:
Have you degenerated your VFA input to ensure the input stage remains in its linear region with fast input signals? You will need to feed a square wave into the input and look at the the peak error signal acorss the diff pair. Might be on a stady state sinusoidal you get better from an LTP, but in most cases to avoid SID, you have to degen the input pair and so gm goes down. See page 23 onwards here http://hifisonix.com/wordpress/wp-content/uploads/2011/03/The_e-Amp_V2.03.pdf

As you point out, the input buffer pair in a classic CFA are not in the feedback loop, but you still get to within a few ppm (low signgle digit of the best VFA's with CFA. in my book, thats plnety good enough, and I for one will not be aggonizing over anything below 10ppm for a practical amp.. Maybe for mental gymnastics, but not for the practical stuff I like to do.
 
Have you degenerated your VFA input to ensure the input stage remains in its linear region with fast input signals? You will need to feed a square wave into the input and look at the the peak error signal acorss the diff pair. Might be on a stady state sinusoidal you get better from an LTP, but in most cases to avoid SID, you have to degen the input pair and so gm goes down. See page 23 onwards here http://hifisonix.com/wordpress/wp-content/uploads/2011/03/The_e-Amp_V2.03.pdf

As you point out, the input buffer pair in a classic CFA are not in the feedback loop, but you still get to within a few ppm (low single digit of the best VFA's with CFA. in my book, thats plenty good enough, and I for one will not be agonizing over anything below 10ppm for a practical amp.. Maybe for mental gymnastics, but not for the practical stuff I like to do.
Andrew, I'm not comparing Diamond IPS with LTP. I'm comparing it with the simple CFA IPS in VSSA and my #500 example.

I'm still not seeing any advantage for Diamond IPS over 'simple' for Power Amps. There are several disadvantages including ..
  • Worse noise.
  • 4x current for the same input gm
  • What appears to be worse linearity but I'm not sure I'm the person to pontificate on this.
I don't think we've settled this important point in deciding what is a 'typical CFA' topology. I like stuff that's simpler .. yet has better performance. 🙂
 
That why I just stick with my VFA. 750ppb @ 20K, fine enough stability, fine enough slewing, at maximum output. And this delicious single digit ppb at 1KHz/1W. With numbers like that, it's all about PCB layout.

This thread is too much about fighting, rather than addressing the CFA topology itself. Like I said in an earlier post, where are the posts that describe a CFA's workings? How to calculate resistor values? You know, the stuff that helps people understand and develop their own CFAs? Don't just go "sim it" as I've already done that and have my answers. I didn't start this thread so I don't feel particularly inclined to take this on me.
 
Last edited:
That why I just stick with my VFA. 750ppb @ 20K, fine enough stability, fine enough slewing, at maximum output. And this delicious single digit ppb at 1KHz/1W. With numbers like that, it's all about PCB layout.

What I do is make, test, listen, compare, listen, compare, listen and then decide. 750 ppm means nothing in practice if simplest CFA outperforms it in practice. :yes:
 
Andrew, I'm not comparing Diamond IPS with LTP. I'm comparing it with the simple CFA IPS in VSSA and my #500 example.

I'm still not seeing any advantage for Diamond IPS over 'simple' for Power Amps. There are several disadvantages including ..
  • Worse noise.
  • 4x current for the same input gm
  • What appears to be worse linearity but I'm not sure I'm the person to pontificate on this.
I don't think we've settled this important point in deciding what is a 'typical CFA' topology. I like stuff that's simpler .. yet has better performance. 🙂

One for your list: worse sound (before VSSA I experimented a lot with diamond buffer input stages) :yes:
 
What I do is make, test, listen, compare, listen, compare, listen and then decide. 750 ppm means nothing in practice if simplest CFA outperforms it in practice. :yes:
I build, test, listen and compare too. I just haven't built my current project, but I'm working on it. Amazing though that you can predict the simplest CFA outperforms something that has yet to be built. Sounds a bit like the "Brilliant Pebbles" you place atop your stereo set that supposedly help improve soundstage. In this case it would be like sticking a label "CFA" on your VFA amp and expect it to sound better immediately.
 
Last edited:
This thread is too much about fighting, rather than addressing the CFA topology itself. Like I said in an earlier post, where are the posts that describe a CFA's workings? How to calculate resistor values? You know, the stuff that helps people understand and develop their own CFAs?
Err.rh! I don't think this exists even for VFAs.

Look at the resistor values around the Triple & EF2 OPSs in both Self & Cordell. They illustrate very different priorities, even with the same topology. These priorities are only apparent to those versed in the art.

Difficult (??) to calculate resistor values until you decide on a topology. As I said earlier, there are at least 3 different CFA IPS topologies that have been presented.

I seem to be the only one to have voted and given reasons for my vote .. and shown the naive sims that led to this choice.

You don't have to agree with me but I'd like to know the REASONS for your choice. Some of the 'reasons' for VFA over CFA are obviously myth. eg CFA's noisier & have poor PSR.

But please focus on different CFA topologies rather than VFA vs CFA.

Trying to understand why Self & Cordell make different choices is an important 'advanced' exercise. I'd hoped this thread would do the same for better CFAs.
____________

Magic, what LC is saying is that he believes his simple VSSA design will sound better than complicated CFA or VFA designs because several have been built and compared in listening tests. In my previous life, I was a true Double Blind Listening Test bla bla guru. The small number of such tests I've done on amps (alas all VFA) suggest that for the same THD bla bla performance, the simpler amp is likely to sound 'better'.

I can quite believe its possible that VSSA sounds better than more complicated CFA or VFA designs .. including those with sub 1ppm THD.
 
Last edited:
@kgrlee:

I think LC's VSSA would be a good topology to start with. It's well, simple and contains the essence of what a CFA is. And then go from there.

As my choice for VFA: 90dB loopgain at 20K, 120dB at 1K. More feedback to erase nasties from the audio band 🙂
 
A little off-topic for kgrlee:

In your conducted blind listening tests, have you found any correlation between the rated power of amps v.s. sounding good at normal listening levels? I.e. 25W/8Ohm small amps sounding better than say 150W/8Ohm?
 
Err.rh! I don't think this exists even for VFAs.

I seem to be the only one to have voted and given reasons for my vote .. and shown the naive sims that led to this choice.

But please focus on different CFA topologies rather than VFA vs CFA.

Trying to understand why Self & Cordell make different choices is an important 'advanced' exercise. I'd hoped this thread would do the same for better CFAs.
____________

The small number of such tests I've done on amps (alas all VFA) suggest that for the same THD bla bla performance, the simpler amp is likely to sound 'better'.

I can quite believe its possible that VSSA sounds better than more complicated CFA or VFA designs .. including those with sub 1ppm THD.

😀😎

[BTW - your observations and conclusions seem to be very close to my own and for much the same reasons. hmmm...]

Thx-RNMarsh