Like yourself, I have no idea what the debate points in this thread are about. It appears to me as more like an ego fight in which everybody is defending it's own trench and throwing grenades over the separation wall.
There seem to be a large amount of misunderstandings and confusion spread among various posters (about what "CFA" and "VFA" are, about small vs. large signal behavior, about models vs. a particular implementation, etc...).
I don't see anybody denying the existence of "CFAs" and their special small and large signal properties. The naming of "CFA" can, more or less academically, be rightfully debated, but ultimately who cares if it was already accepted by the industry? CFAs are here to stay, disregarding a rather confusing name chosen.
Personally, the only surprise was the original prof. Franco stance in defending CFAs (like they wold need any defense!).
What??? I thought the two-port analysis technique determines the feedback type, since the abstract two-port analysis doesn't make any assumptions about a specific implementation. But then, I'm currently way off the academic track, so perhaps he knows something about "feedback types" that I'm missing in the last 40 yeas or so. Anyway, I think I was able to mathematically re-construct the CFA known (and special) small signal properties by using the same input series, shunt output, two-port formalism that can be used in analyzing your favorite long tail pair VFA, and without any extra assumptions. This must tell something about the nature of CFAs.
Oh, and nothing is more "paper-and-pencil" prone than an abstract two-port analysis. Spice doesn't apply!
Last edited:
No it doesn't. It was in reflection that an RC time constant exists at the Tz node that doesn't exist for input stimulus voltage feeding into the Thevenin equivalent resistance, this as connected to the opposing end to the V Thevenin output. The output, and therefore V Thevenin output, changes in response to the charging of C at the Tz node.
If we nevertheless agree that no time/phase delay exists is it thereupon also irrelevant for audio frequencies? Is it your position that there is no point in examining audio frequencies either, or anything beyond DC? Are you stating that the Middlebrook analysis is irrelevant in concluding the nature of a CFA or VFA, at frequencies beyond DC? Why are all the graphs included?
I like the expression "house of cards" as oftentimes a single irreconcilable seed can do that... even taking down ones entire "being" in advance of perhaps ones reconstruction from the quantum of consciousness to correct.
Not at all... my argument begins from DC conditions. The irreconcilable seed that I perceive to exist is that the direction of output current from the current mirrors must match the direction of output current from the In- terminal and into R Thevenin.
Looking at the schematics, I only understand that lower left part cannot work.
The voltage source in the feedback loop sees an infinite impedance from the current source, so that inverting input sees nothing.
I cannot see how to conclude anything of this scheme.
I am sorry but no idea what your point is here. Surely the Tz RC-time exists for any signal at the Tz node? And of course, the Tz node being the output of a current source, the voltage changes exponentially according to the RC? Am I missing something here?
You are confusing things again. Of course there is phase shift in an amp, nobody in his right mind would deny that. And as has been stated repeatedly, the signal delay through the amp is in the order of a nanosec which means that the feedback is instantaneous at audio frequencies. And it is precisely the facts that phase shift exists and feedback is instantaneous which leads to instability at higher frequencies. (Depending on conditions of course; no need for you or other wise guys to explain that it depends on gain and phase margin. Stay focused).
So again, not sure what your point is here.
BTW I get that 'phase delay' expression. I think it is an oxymoron, and confusing as hell, but accept that it exists....
Jan
Last edited:
Which assertions do you consider to be “all crap”? 😛
Eg, You know that if the feedback loop contains a time delay then recycling of out of date info is a thing. Are you declaring this to be insignificant in audio or are you denying the whole recycling notion?
I tip my hat to you - you are correct! To remedy the situation, I added a 1 Gohm resistor across the current sources. Please see the updated attachments.
Middlebrook supplies the conclusion: there is more current feedback than voltage feedback in this circuit. And yet we've used a VFA as the gain block! How can this be?
It is because of what I have said before: To a good approximation for high loop gains, the ratio of current to voltage feedback at the inverting input of a circuit is a ratio of impedances: that of the feedback network (1Gohm in this case) to that of the inverting input (less than 1Gohm.)
Middlebrook does not reveal circuit topology; it only presents the voltage and current loop gains from which we can infer the feedbacks as described in the article I referred to.
Attachments
A demonstration of how the distributed nature of a real device can mimic a delay. This is a 5 stage R/C lowpass (1K and 2000pF) compared to 5K and 10000pF. This is grossly exaggerated a transistor like this would not even have 100kHz ft. If you saw this on a scope you would call it a delay but all ten devices have at least some current flowing instantaneously with applied step.
Attachments
Sorry not to get back earlier Hans... so much to think about. This was in reference to the EDN's article in I believe Figure 3, suggesting the Vf was the feedback voltage of a non-inverting amplifier as feeding Rf and Rg being shown in the feedback path. The problem was that Vf was 180 degrees out of phase with what would have appeared as Vf output. Being inverted, Vf could only translate to input current stimulus of an inverting amplifier. The article claimed this represented current feedback when in fact it was showing input stimulus current as a proof this represented current feedback.
Sorry for the confusion... these PDF's weren't intended to be necessarily connected. My 35 year background in creating 100's of electronic devices and networks never involved explaining anything to anyone, hence I can sporadically switch gears in some understood connection as knowing to myself, yet be completely confusing to others trying to follow along.
In the brevity Figure 4 is only intended to identify what a feedback system will have to contend with... nothing more. I hope to provide more detail as time goes on. Thanks Hans
"almost exactly equal" : excellent !
This is absolutly not a proof of current feedback but a simple application of Kirchhoff law on In- node.
Current feedback cannot explain the ratio between "gm*Vbe" and "Vce/Ro" current in the input device. Voltage feedback does.
I wrote almost the same thing in that thread before knowing the existence of MDIT.
You wrote "electron flow"
See attached.
What are the criteria ? Who established it ? Who accepted it ?
See attached. Current on demand is not inherent in CFA.
Constant bandwidth is easy to obtain with VFA, as current on demand.
Attachments
I hardly think your circuit is representative. Would you compensate a practical amplifier like that - VFA or CFA?
Its not surprising you have 'COD' like behaviour. You simply shunted some of the current mirror output to ground via C1 (4.7nF!?), so the front end is driving hard to try to recover the shunted current via Rf. If you do that to a classic VFA you get the same behaviour but the output of the VAS slews because the LTP tail current is fixed i.e. one half of the LTP is likely to turn off. What you have demonstrated here is that if you **** the compensation, you can get 'COD'.
There is no argument that there are tweaks to the VFA topology that will give it some gain bandwidth independence and high slew rates. This is why we should be clear about what we are discussing. There is certainly a big overlap in the two topologies wrt to behaviour as the (open) loop gain is raised. But, to try to sort this whole CFA vs VFA thing out by arguing about the behaviour in the overlap area to my mind seems a waste of time.
But, the question remains, is your circuit a VFA or a CFA?
Its not surprising you have 'COD' like behaviour. You simply shunted some of the current mirror output to ground via C1 (4.7nF!?), so the front end is driving hard to try to recover the shunted current via Rf. If you do that to a classic VFA you get the same behaviour but the output of the VAS slews because the LTP tail current is fixed i.e. one half of the LTP is likely to turn off. What you have demonstrated here is that if you **** the compensation, you can get 'COD'.
There is no argument that there are tweaks to the VFA topology that will give it some gain bandwidth independence and high slew rates. This is why we should be clear about what we are discussing. There is certainly a big overlap in the two topologies wrt to behaviour as the (open) loop gain is raised. But, to try to sort this whole CFA vs VFA thing out by arguing about the behaviour in the overlap area to my mind seems a waste of time.
But, the question remains, is your circuit a VFA or a CFA?
Last edited:
It could behave, small signal, like a CFA, but for lower closed loop gains (that is, lower R3/R2 ratio). Try R3=150 ohm and you'll get it.
For the R3/R2 as in the schematic it will likely behave (small signal) as a VFA.
The compensation is indeed poorly chosen; I would rather connect C1 between the Q2 collector and the Q1 emitter, resulting in a Faroudja compensation schema, but of course, more analysis is required. If C1 is connected as Faroudja compensation (and the amp is stable), then the (large signal) "current on demand" feature is guaranteed.
Or at least add a resistor in series with C1 to make a lead-lag compensation.
Of course it not representative of a realistic amplifier.
This caricatural scheme is to demonstrate that "current on demand" is not inherent to so called CFA but only to those which have a diamond (or similar) input.
Of course yes.
Gain bandwidth independance is obtained by playing with Rg resistor.
You have the same result by playing with the emitter resistor in a LTP or a double diamond.
Double diamond VFA allows same slew rate as so called CFA.
This caricatural scheme is to demonstrate that "current on demand" is not inherent to so called CFA but only to those which have a diamond (or similar) input.
Of course yes.
Gain bandwidth independance is obtained by playing with Rg resistor.
You have the same result by playing with the emitter resistor in a LTP or a double diamond.
Double diamond VFA allows same slew rate as so called CFA.
Attachments
My question as to whether is was CFA or VFA was rhetorical 😉.
Agree with your comp proposal - I used it on the sx and nx amps. For high loop gain CFA I use other approaches.
Agree with your comp proposal - I used it on the sx and nx amps. For high loop gain CFA I use other approaches.