There is one point of concern about some of the CFAs that we have seen. This is certainly not a show-stopper, but is rather something to be aware of.
The feedback resistor for CFAs is often quite a bit lower in value than that for a VFA. For example, if we have a 40-ohm feedback shunt resistor, we will use a 1k feedback resistor to obtain a closed loop gain on the order of 25. Power dissipation in the 1k feedback resistor can be substantial. Consider a 100-watt amplifier with 40V peak output voltage. Feedback resistor current will be 40mA peak, and peak power dissipation will be about 1.6 watts. This may cause thermal modulation in the feedback resistor if it is not way over-sized.
I have always like low-z feedback networks for my VFAs to keep noise down and to minimize the effects of the summing node pole at the input to the input stage LTP, but I've never gone this low. In my MOSFET power amplifier I went down to about 250 ohms for the shunt resistor and ended up using a 2-watt metal film resistor for the feedback resistor to keep LF distortion down.
Cheers,
Bob
The feedback resistor for CFAs is often quite a bit lower in value than that for a VFA. For example, if we have a 40-ohm feedback shunt resistor, we will use a 1k feedback resistor to obtain a closed loop gain on the order of 25. Power dissipation in the 1k feedback resistor can be substantial. Consider a 100-watt amplifier with 40V peak output voltage. Feedback resistor current will be 40mA peak, and peak power dissipation will be about 1.6 watts. This may cause thermal modulation in the feedback resistor if it is not way over-sized.
I have always like low-z feedback networks for my VFAs to keep noise down and to minimize the effects of the summing node pole at the input to the input stage LTP, but I've never gone this low. In my MOSFET power amplifier I went down to about 250 ohms for the shunt resistor and ended up using a 2-watt metal film resistor for the feedback resistor to keep LF distortion down.
Cheers,
Bob
There is your definition for you Waly, a CFA has poorer PSRR... you defined it for yourself.
Huh? Did I say a CFA has poorer PSRR? I only said that the common diamond buffer input stage has a lower PSRR than the long tail pair input stage.
That is the fundamental error. Feedback theory is essentially about black boxes (aka "plants" in control theory) with some external circuit called "feedback network", that's what it is since Black.
I've mentioned this before, call your baby "current controlled amplifier" (CCA) and it would be fine by me, just don't pretend CFAs are defining a new magic "current feedback" type. Such CCAs are, from a control theory perspective, VFAs with some special properties (controlled by the interaction between the "black box" and the feedback network).
We discussed the feeback resistor here http://www.diyaudio.com/forums/solid-state/243481-200w-mosfet-cfa-amp-7.html#post3656329.
Maybe you can take a look in that thread if our soulution is good enogh, there were differen proposals for that.
BR Damir
If you apply the two tests, it will cater for BJTs, FETs, single ended and balanced input stages of the diamond or any other input structure and give you the right answer almost every time.
When I looked into 'what is a CFA?' I got lots of math explanations, a lot of stuff that left me no wiser. That's how I ended up with the two tests.
I don't think to simply say as long as the feedback signal is applied to the low impedance inverting port without any interposing active stage it's a CFA. An H bridge input structure breaks that definition, but it's still a CFA. Likewise, if you have a classic CFA diamond input structure but the feedback resistor value is made very high, the gain bandwidth independence of CFA is no longer met, and what you actually have is a very sub optimal VFA. In this case, test no. 1 would indicate, correctly, a VFA because CLG -3db BW would be dependent on gain, despite the fact that the TIS peak input current would be set by Rf.
IMV what defines a CFA is much more to do with how it performs/behaves and less about the way the parts are connected.
Youre incorrect, it breaks the definition and thats why industry no longer considers it a CFA, but a VFA. Refer to the numerous literature by Texas, Analog devices and if I remember correctly Walt Jung mentions it as well.
It is considered a VFA build with CFA building blocks.
Hopefully you dont share Waly`s stubbornes.
Looks like I've allowed myself to be suckered into another dead end discussion with you . . .
No offense taken at all, but I hope you don't take offense by me pointing out again that the input to a so called CFA circuit in situ is not between the plus and minus pins on a so called black box but from the plus input to ground through the feedback network, i.e. the voltage across the plus and minus pins is not the input the current flowing caused by this voltage is. This current charges the comp cap and makes the voltage at the gain node.
EDIT - I would hope we can stay out of being offended one way or the other and just have a friendly arguement.
Last edited:
Hi Richard,
Anyhow, RE2 is chosen in such a way that gm is neither compressive, nor expansive. IOW, this will give the lowest distortion.
).
Cheers, E.
PS: I hope I have made clear in my previous post, that the noise from a CCS injected into one leg of the input stage, also leaks into the other leg. By choosing the right RE1 and RE2, they have the same magnitude and are cancelled in the next stage (provided that the top and bottom TIS of the next stage have the same gain).
This is because of the Early and Cob effect, which, opposed to a LTP-VFA, has not been cancelled. In case of LTP-VFA, both legs of the LTP are subject to these effects and are (largely) cancelled by the action of the current mirror.Hear! Hear!
[..]
Would the true gurus please tell us why CFAs have poorer PSRR?
I'm not sure which RE you mean, RE1 or RE2?
Anyhow, RE2 is chosen in such a way that gm is neither compressive, nor expansive. IOW, this will give the lowest distortion.
VT is the so called 'thermal voltage' = kT/q, where k is the Bolzmann constant, T is the abs. temp. and q is the electron charge. The optimal RE also applies to a non diamond complementary CFA, of course (hope I'm not too pedantic
).Cheers, E.
PS: I hope I have made clear in my previous post, that the noise from a CCS injected into one leg of the input stage, also leaks into the other leg. By choosing the right RE1 and RE2, they have the same magnitude and are cancelled in the next stage (provided that the top and bottom TIS of the next stage have the same gain).
Last edited:
how about calling the error signal from the "CFA" front end the difference between the plus input V and the Thevenin V of the feedback network
with the additional degree of freedom changing equivalent input gm by scaling the feedback Thevenin Z seen at the "CFA" negative input
the Thevenin V is of course the Vout times divider ratio - and the definition works for open loop gain too - giving Scott's gnd and series R
with the additional degree of freedom changing equivalent input gm by scaling the feedback Thevenin Z seen at the "CFA" negative input
the Thevenin V is of course the Vout times divider ratio - and the definition works for open loop gain too - giving Scott's gnd and series R
Last edited:
You pointed an important aspect to take care of, while choosing our CFA components.
Those resistances are determinant for the sonic quality of the amp. Need to be over sized, non magnetic, non inductive, low temperature coheficient.
Thin metal Film ones ?
Notice that, if we want to achieve the same kind of bandwidth/phase in a VFA, and because the parasitic capacitance of the base, we will face the same low impedance/power problem. Even worse, if the base parasitic capacitances are higher than the emitter's ones. Not to talk about FETs.
yes, this is certainly true -- CFA used in preamps that is not any concern... where Wattage is low enough and voltage swings low also. But maybe if using smd, you can get into trouble.
-RNM
Well, now, we have a perfect description of the theory of operation of a CFA.
I'm a little disappointed, because it do not describe in a simple and short way the topology of CFAs.
I mean that, looking at a schematic, without the values, it is impossible to say if it fulfill the described requirements. And that will not help the understanding of unwashed (at CFAs) people.
One thing is for sure, all the aspects of CFA behaviors has been described in this thread, and this is a VERY positive reference.
Thanks to Edmond to entered deeply in the input stage analyses (noise and PSRR), although I must note he is as lazy as all of us, preferring simulation than pure mathematics ;-)
I'm a little disappointed, because it do not describe in a simple and short way the topology of CFAs.
I mean that, looking at a schematic, without the values, it is impossible to say if it fulfill the described requirements. And that will not help the understanding of unwashed (at CFAs) people.
One thing is for sure, all the aspects of CFA behaviors has been described in this thread, and this is a VERY positive reference.
Thanks to Edmond to entered deeply in the input stage analyses (noise and PSRR), although I must note he is as lazy as all of us, preferring simulation than pure mathematics ;-)
Last edited:
Happily, the peak/average power ratio in music and resistance's thermal inertias are high enough.
Yes, the point remains that the power dissipation of the feedback resistor must be increased significantly in some way. There are obviously many ways to do this. The idea expressed in that thread of possibly allowing higher peak dissipation in the hope that the feedback resistor and the shunt resistor would track their dynamic temperature changes seems rather hopeless.
If I had to use a 2W metal film 4k feedback resistor to eliminate dynamic thermal distortion in a 50W amplifier, then one might argue that a similar level of conservatism for a 200W amplifier might require a 16W 1k resistor. That is quite non-trivial. The mention of using a 50W non-inductive power resistor in a TO-220 package is interesting, however.
Cheers,
Bob
Lazy
>Thanks to Edmond to entered deeply in the input stage analyses (noise and PSRR), although I must note he is as lazy as all of us, preferring simulation than pure mathematics ;-)
Hi Christophe,
I wonder if a pure mathematical approach (read analytical) is feasible anyhow. Because of the exponential functions involved, things get too complicated and you have to revert to numerical methods. That means using a computer anyway for finding the optimal resistor values.
Cheers, E.
>Thanks to Edmond to entered deeply in the input stage analyses (noise and PSRR), although I must note he is as lazy as all of us, preferring simulation than pure mathematics ;-)
Hi Christophe,
I wonder if a pure mathematical approach (read analytical) is feasible anyhow. Because of the exponential functions involved, things get too complicated and you have to revert to numerical methods. That means using a computer anyway for finding the optimal resistor values.
Cheers, E.
It is an interesting point about the impedance level in the feedback network. When I used a 215-ohm shunt resistance in my MOSFET amplifier, I deemed it low enough, and that going lower was strongly subject to the law of diminishing returns. However, I was using a JFET input stage. So, a fair question for me is whether going lower than 215 ohms in a BJT front-ended amplifier would improve things very much as far as HF phase lag.
BTW, for a power amplifier, anything better than about 10 nV/rt Hz input noise is gravy, so going lower than the approximate 2nV/rt Hz noise of a 215 ohm resistor is not important. I usually achieve something on the order of 5-7nV/rt Hz in my amplifiers.
Cheers,
Bob
The high slew rate being achieved in the CFA designs is an interesting matter. Besides the so-called current-on-demand action, there is obviously the issue of how the feedback is compensated.
Take a look at the compensation OS used in his excellent design. I would argue that it is a form of Miller Input Compensation (MIC). I achieved 300V/us with MIC in my 50-watt MOSFET power amplifier. So I don't think high slew rate is exclusive to the realm of CFAs.
Cheers,
Bob
Take a look at the compensation OS used in his excellent design. I would argue that it is a form of Miller Input Compensation (MIC). I achieved 300V/us with MIC in my 50-watt MOSFET power amplifier. So I don't think high slew rate is exclusive to the realm of CFAs.
Cheers,
Bob