Claude Abraham,
you talk here about a schematic.
I have searched this thread for such a posted schematic.
You know what schematic you refer to.
Please link or repost that referenced circuit.
Thanks.
Claude Abraham,
I judge your point of view more credible than our MichaelKiwanuka's view.
Because you seem to be able to back up your view with some facts, numbers.
A schematic, a picture, would make us all a little bit wiser, I think.
Attached. It was attached earlier this month as well. Let me know if you need further explanation. Thanks for the vote of confidence. Best regards.
Claude
Did some more thinking about this.
In a conventional VFB opamp the feedback forces the input voltage on the inv input to zero (with respect to Vin of course). So, due to the high internal gain, the Vout will be whatever is required to make that error voltage zero. So far, so good.
In a CFA the feedback forces the inv input node with its almost zero Zin to zero current! The high internal current gain will drive Vout (across Tz) to what ever is needed to make the (error) current into the inv input zero.
Initailly I was thrown off with that very low actual error current I measured like 470nA. But the whole crux of the thing is to drive that current to zero, just as in a VFB the idea is to drive the input voltage to zero.
So it seems clear that the inv input of the CFA reacts to error current rather than error voltage.
The fact that, as Mike mentioned, you can adjust the gain of a CFA fb amp by manipulating the feedback voltage attenuation ratio (which seems to contradict the current sensitivity) works because of the very high internal gain. This means that even a very small deviation of the error current, which will not appreciably load the fb network, will be enough to deliver the required Vout.
We still have the unfortunate situation that 'current feedback' is still being taught as 'feeding back (a sample of) the output current' but that's a different issue and appears unresolvable - the term 'transimpedance amp' will not take the market by storm, if only because most people do understand 'current' but not 'transimpedance'...
Edit: Claude's linked schematic above illustrates it neatly.
jan
In a conventional VFB opamp the feedback forces the input voltage on the inv input to zero (with respect to Vin of course). So, due to the high internal gain, the Vout will be whatever is required to make that error voltage zero. So far, so good.
In a CFA the feedback forces the inv input node with its almost zero Zin to zero current! The high internal current gain will drive Vout (across Tz) to what ever is needed to make the (error) current into the inv input zero.
Initailly I was thrown off with that very low actual error current I measured like 470nA. But the whole crux of the thing is to drive that current to zero, just as in a VFB the idea is to drive the input voltage to zero.
So it seems clear that the inv input of the CFA reacts to error current rather than error voltage.
The fact that, as Mike mentioned, you can adjust the gain of a CFA fb amp by manipulating the feedback voltage attenuation ratio (which seems to contradict the current sensitivity) works because of the very high internal gain. This means that even a very small deviation of the error current, which will not appreciably load the fb network, will be enough to deliver the required Vout.
We still have the unfortunate situation that 'current feedback' is still being taught as 'feeding back (a sample of) the output current' but that's a different issue and appears unresolvable - the term 'transimpedance amp' will not take the market by storm, if only because most people do understand 'current' but not 'transimpedance'...
Edit: Claude's linked schematic above illustrates it neatly.
jan
Last edited:
I fear feedback never acts to force the error signal to zero, because the error is required to drive the forward path of the amplifier.
If the ideal situation were to drive the error signal to zero, it would imply a forward path of infinite gain. i.e. the amplifier would require no input to saturate to the supply rails; an amplifier with infinite forward path gain is clearly not required.
Very good Jan. Regarding the fact that the output voltage is the sampled quantity, I've been saying that since the thread begain around 2004. When a vendor classifies their op amp they do so with regard to the error signal, not the sampled output quantity. I've stated many times that the phrase "VFA" implies that said amp processes an error "voltage", not an error current. The sampled output quantity is up to the user. Most of the time, a VFA topology uses the output voltage as the sampled quantity that is fed back.
But, Jan, I'm sure you have seen networks where a motor, or loudspeaker, or other device is driven from an amp with a low value resistor in series w/ the load on the ground side. The voltage across said resistor is fed back to a VFA. Here the sampled output quantity is current, not voltage. But the op amp is and always will be a VFA, because it develops an error signal in the form of a voltage, not current.
You seem to have a very good understanding of the CFA as well as the VFA error signal. That is good. But you and others have a tendency to perpetually remind us that the sampled output that is fed back is voltage. I have, as have others, already conceded that output voltage is USUALLY the sampled quantity fed back. But that is up to the user. It does not have to be that way.
With a CFA, most applications sample the output voltage and feed it back in the form of a current. The op amp maker must classify said amp as CFA because the error signal is current. You know that well so we agree there.
The only point I wish to discuss and clarify with you is the following. Do you agree or disagree that which output quantity is sampled is up to the user. An op amp maker has no control over that. A VFA can be used with output voltage or output current as the sampled and fed back quantity. With current output sampled and fed back, a VFA erquires that said output current be translated into a voltage, which is the function provided by the low value sense resistor.
Otherwise, we seem to agree. Thanks for your feedback and feel free to ask questions or comment. BR.
Claude
This is incorrect. The type of amplifier is established by consideration of the transfer function of its forward path.
On the other hand, the type of feedback enjoyed by an amplifier is defined by by the feedback transfer function:
1) voltage feedback for an amplifier with a voltage amplifying forward path.
2) transimpedance feedback for an amplifier with a transadmittance forward path.
3) transadmittance feedback for an amplifier with a transimpedance forward path.
4) current feedback for an amplifier with a current amplifying forward path.
Now, a so-called "CFA" has a voltage amplifying forward path with voltage feedback: it is therefore a voltage feedback amplifier.
The so-called "CFA" does NOT "process" an error current.
You are right of course, but it is a very close approximation the assume that the error is zero.
In my AD844 example, the error current was 478nA. Not zero, but very, very small if you realise that the input Z is just a few ohms.
jan
That current really is not the error signal: I am certain the error signal is the difference in voltage between the non inverting and inverting inputs.
It is not my opinion that sampling output current has always been classified as current feedback - that's the original definition, and that is what I and countless others have learned in college, and what is still taught. I'm not trying to force my opinion on anyone, just to clarify a de facto situation.
You know about things like motional feedback - that's feedback where the sampled quantity which is fed back is derived from the motion. It can be used with a 'CFA' or a 'VFA' opamp as error amp but it is always 'motional feedback'.
In feedback and control theory (also in motional feedback) you can have acceleration feedback, velocity feedback - all denote the derived quantity that is fed back, NOT whether it is going into a CFA or a VFA.
The AD846 was presented in 1988 in the journal Analog Dialog from ADI by one of Scott's buddies as a 'transconductance opamp', which is also a correct identification and would avoid the conflict in terminology but it apparently didn't catch on with the general public.
jan
It is by definition the error signal because it drives, through the current mirrors, Vout.
Mike, look at it this way. Ever since Mr. Ohm we know that there's no voltage without current and vice versa.
So in any case, VFA or CFA, there's both an error current and an error voltage. I believe I've seen the light
that, determined by the topology of the inv node, it is the (error) current into that node that is conveyed up to the mirrors and back to the common emittor output stage into the Tz node, where it is converted into Vout and subsequently buffered.Edit: I checked the error voltage that goes with that 478nA error current. It's about 60uV, which gives a differential Zin for the inv node of around 120 ohms. I would have expected something much lower, but I'm a bit out of my dept here.
Scott?
jan
Last edited:
The current sunk or sourced by the inverting input of a so-called "CFA" is a function of the Thevenin impedance "seen" by that port to ground. It is not an error signal.
The error signal is the difference between the input voltage and the feedback voltage; this is then amplified by the input stage transistor(s) to produce changes in their collector currents which changes are level shifted by the current mirrors to the load.
Last edited:
Agreed. If the input voltage is changed, that causes a current difference into the collectors of the LTP, as the LTP is a voltagedriven current source.
The inv input is a common base stage from the point of view of the feedback network, so if the feedback voltage changes, it causes a current difference into the collectors etc bla bla.
I think you'd agree that looking into the noninv input you see a fundamentally different topology (common emitter) than looking into the inv input (common base).
jan
I agree, but the fact that the inverting port of the "CFA" sources and sinks current to and from the feedback network does not imply that this current is the error signal driving the amplifier. That's my point.
At this point, I don't understand what's the struggle about. I think I've proved that the "CFA" topology properties can be mathematically deduced by using the "VFA" h-parameters formalism (so exactly what Michael says, about sampling voltage at the output and providing the sample to the inverting input to create the error signal. The big difference is that in a CFA topology the feedback network loading at the input node can't be ignored, and it modifies the forward gain and forward UGF. I've also proved that for large closed loop gains (theoretical exercise, since it's kinda stupid to use a CFA for) the CFA topology degenerates to a typical VFA. So indeed a "CFA" category is theoretically not really required, although it can be somehow shoehorned using Mr. Didden et. al. arguments, I see no theoretical or practical damage in doing this.
I also don't believe the linearity of a CFA topology is anywhere worse than a VFA. That's a little harder to prove mathematically, but the number of DSL CFAs with distortions in the ppm range at MHz is witnessing.
CFAs are bad for audio? I haven't heard any serious arguments, based on measurements or controlled subjective tests. I personally have one big argument pro-CFA, the unconditional stability. It is very hard (if possible at all) to build a VFA with high loop gain at HF, which is also unity gain stable. I also like the high speed and high slew rate, but that's my purely subjective opinion.
I also don't believe the linearity of a CFA topology is anywhere worse than a VFA. That's a little harder to prove mathematically, but the number of DSL CFAs with distortions in the ppm range at MHz is witnessing.
CFAs are bad for audio? I haven't heard any serious arguments, based on measurements or controlled subjective tests. I personally have one big argument pro-CFA, the unconditional stability. It is very hard (if possible at all) to build a VFA with high loop gain at HF, which is also unity gain stable. I also like the high speed and high slew rate, but that's my purely subjective opinion.
For a typical CFA the current flowing in/out of the inverting input is the difference between the +/- currents driving the high impedance gain node (subtle difference but important since each of these collector currents are more nonlinear than there difference).
This resulting current does impose an error on the feedback voltage as a measurement of the true output voltage.
Thanks
-Antonio
This resulting current does impose an error on the feedback voltage as a measurement of the true output voltage.
Thanks
-Antonio
Not necessarily, but one can do much better with traditional VFA design at audio frequencies.
We need to be clear what "HF" means: from my perspective it's not hard to build a VFA with relatively high loop gain at the top end of the audio band, siginicantly higher than that provided by a so-called "CFA" for comparable stability margins. Just use double pole compensation.
Clearly double pole compensation or even "TMC" cannot be used in the usual so-called "CFA" topology.
Don't these two phrases mean the same thing?
At any rate slew rate is not an issue with competently designed VFAs intended for the audio spectrum.
Last edited:
I found some interesting explanations of the CFA topology in several issues of Analog Dialogue. I wrote it up and linked to the pertinent pages from the 1987, 1988 and 1990 journals here (scroll to the bottom of the page).
jan
jan
One of those articles describes a source follower as a current feedback amplifier. This is utter nonsense and is on a par with describing a voltage follower as a current feedback amplifier.
A source follower is an example of 100% shunt derived series applied voltage feedback.
http://www.linearaudio.nl/linearaudio.nl/images/pdf/AD_23-3-89.pdf
A source follower is an example of 100% shunt derived series applied voltage feedback.
http://www.linearaudio.nl/linearaudio.nl/images/pdf/AD_23-3-89.pdf
Don't these two phrases mean the same thing?
No. IMMV, but to me the difference is like the small signal slew rate (aka "speed") vs. the large signal slew rate (aka "slew rate"). The former is a measure of the bandwidth, the latter is a measure of a non-linear input stage limiting effect. AS CFAs are essentially "current on demand", the "slew rate" can be very high (agreed, perhaps uselessly high for audio).
BTW, I suppose you are aware you can use a VFA in "CFA mode"? See for example the Alexander amp.
Last edited:
I wouldn't call it "utter nonsense", but a wrong example.
None of the four feedback types and canonical formalisms (hij, yij, zij, gij) are actually mandatory to analyze a feedback circuit. You may characterize a feedback circuit using any of these (or none at all
, but only Mason, or Kirchoff, or whatever fancies you) and get exactly the same results (after all, real circuits are deterministic). It's a matter of convention and convenience to use the appropriate formalism, based on the assumed properties of the forward gain amp (which are not always a clean cut, as for CFA topologies).
One of those articles describes a source follower as a current feedback amplifier. This is utter nonsense and is on a par with describing a voltage follower as a current feedback amplifier.
A source follower is an example of 100% shunt derived series applied voltage feedback.
http://www.linearaudio.nl/linearaudio.nl/images/pdf/AD_23-3-89.pdf
Yeah wasn't that a bad example? There's also a section where he says that current feedback is where you feed back a current that is subtracted from or added to an input current.
Funny how they wrestled with this - it's easy to see that everybody was releaved when they coined this 'CFA' term. As you know, acronyms can have 3 letters only
jan
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.