It's interesting to see this carry on, thought the acrimony is disappointing.
For a slightly different angle let me digress. Duality is like some kind of universal appealing principle, symmetry, good/bad, yin/yang, KVL/KCL, Norton/Thevenin, series/parallel, series/shunt, etc. It seems natural that at the beginning when circuit blocks rarely went beyond a single active device a mathematical framework would develop around reducing the computational formality to the simplest representations and be a useful analysis tool.
This leads to two port formalism with input and output from blocks described by a 2 X 2 matrix of complex values. Voltage and current and their inverses yields four matrix representations, H, Y, Z, and G. Same goes for feedback series/shunt two ports, four types.
Classic two port analysis is still useful for Rf work involving T-lines, waveguides, and single transistor gain stages. Does anyone remember flow diagrams?
In my day this was all covered in all its traditional glory, but the fact remains that there are several ways of posing the problem such that the solution is unique, and complete. We moved immediately, for circuits of any complexity, to a matrix form derived from a set of simultaneous equations based on KVL and KCL at each circuit node. The kind of feedback never figures into anything. To make that clear you set up N equations in N unknowns and solve, noting the "type" of feedback is completely irrelevant to the problem, an academic curiosity so to speak. Another example of a classical concept often inappropriately applied is noise figure and match terminated systems. The general instrumentation problem (including audio) does not lend itself to broadband impedance matching and maximum power transfer.
@CPaul - I can no longer find any new line of reasoning, treating transistors as pure transconductances is perfectly valid and has no meaningful bearing on any of these arguments.
For a slightly different angle let me digress. Duality is like some kind of universal appealing principle, symmetry, good/bad, yin/yang, KVL/KCL, Norton/Thevenin, series/parallel, series/shunt, etc. It seems natural that at the beginning when circuit blocks rarely went beyond a single active device a mathematical framework would develop around reducing the computational formality to the simplest representations and be a useful analysis tool.
This leads to two port formalism with input and output from blocks described by a 2 X 2 matrix of complex values. Voltage and current and their inverses yields four matrix representations, H, Y, Z, and G. Same goes for feedback series/shunt two ports, four types.
Classic two port analysis is still useful for Rf work involving T-lines, waveguides, and single transistor gain stages. Does anyone remember flow diagrams?
In my day this was all covered in all its traditional glory, but the fact remains that there are several ways of posing the problem such that the solution is unique, and complete. We moved immediately, for circuits of any complexity, to a matrix form derived from a set of simultaneous equations based on KVL and KCL at each circuit node. The kind of feedback never figures into anything. To make that clear you set up N equations in N unknowns and solve, noting the "type" of feedback is completely irrelevant to the problem, an academic curiosity so to speak. Another example of a classical concept often inappropriately applied is noise figure and match terminated systems. The general instrumentation problem (including audio) does not lend itself to broadband impedance matching and maximum power transfer.
@CPaul - I can no longer find any new line of reasoning, treating transistors as pure transconductances is perfectly valid and has no meaningful bearing on any of these arguments.
Last edited:
This line of reasoning is wrong and doomed to produce bad conclusions. There is no actual delay because a true delay is non-minimum phase and in general this is not the case.
Remember I put a sim result up that clearly showed gain-bandwidth independence and pointed it out to you?
See the EDN article of his. It’s more of a side comment than anything formal.
Last edited:
So did in mine, therefore the two port approach I posted. And yes, I painfully recall flow diagrams, graphs and the Mason rule.
The "new" approach of solving linear system by brute force is IMO the result of the computer advent. As usual, the results are correct, but the insight may be lost in the process of crunching numbers. But then I have to admit the former is what really matters today, to hell with analytic understanding and integrating in common the body of knowledge.
Sure Scott. I have edited this error out of my previous post as per below.
The complete process control of a CFA can be described as input stimulus "over-current" followed by "feedback current limiting”. This is to state that input stimulus precedes feedback taking place. It is upon this basis that there exists faulty reasoning existing in the document:
https://www.edn.com/design/analog/44...back-amplifier
This begins on page 1 after Figure 3 (prior to Middlebrook). It states:
"Since the input buffer keeps Vn = 0, RG draws no current, so we must have In = (0 – Vf)/RF = –(1/RF)Vf, indicating that what comes back from Vf is only current and no voltage. It stands to reason to refer to this type of feedback as current feedback."
Figure 3 in the above website is reproduced as Figure 1. This can be converted to a Thevenin equivalent form of Figure 2 as to generate the same In currents. In Fig. 2 the input Vf is left to show how the value of V Thevenin is calculated as replaceable of Vf. Figure 3 transposes V Thevenin to the non-inverting terminal with R Thevenin connected to ground. The conclusion is that there is no difference between Figure 1 and Figure 3 in terms of behaviour of the inverting terminal.
This challenges the conclusion that "It stands to reason to refer to this type of feedback as current feedback." In the case of Figure 3 the output current is a function of V Thevenin / R Thevenin, the current being caused by a voltage being imposed on R Thevenin by V Thevenin. From this perspective this is voltage controlled current.
Without input stimulus in Figure 1 or Figure 2, feedback has nothing to respond to... hence a Vf signal can't exist in reality to render the conclusion of current feedback being true. This does not apply to Figure 3 for reasons that the CFA is responding to input stimulus voltage on R Thevenin connected to ground.
In conclusion a time delay exists between input stimulus current flowing through the Thevenin equivalent resistance and the delay of the Thevenin equivalent voltage rising in the feedback loop. The feedback process is of current, yet the action is of "current limiting"
TIS - trans impedance stage
TAS - transadmittance stage
TPC - two pole compensation
TMC - transitional Miller compensation
OIC - output inclusive compensation
OPS - output stage
MIC - Miller inclusive compensation
ULGF - unity loop gain frequency
TAS - transadmittance stage
TPC - two pole compensation
TMC - transitional Miller compensation
OIC - output inclusive compensation
OPS - output stage
MIC - Miller inclusive compensation
ULGF - unity loop gain frequency
That's the crux of the issue. There are those that simulate until good, an unfortunate result of having the tools out there for everyone to use blindly. The simple reduced model of CFA vs VFA (not so brute) lends itself to pencil and paper analysis so the question of insight remains, where does one have that moment when realizing a feedback amplifier with BW invariant over a range of closed-loop gains is useful? To me the immediate next step is to explore what do you lose, but that is another matter.
It is important to remember that virtually all classical analysis is based on small signal steady state, no large signal effects are modeled. Volterra series methods are an exception but they are a special case and must be treated separately.
Last edited:
I'm interested in hearing from those that understand that there must exist a time delay between an input stimulus and feedback correction, and can offer any support or criticism of my submission, in whole or in part.
This is disappointing. Never though that someone so erudite as you would hold these naive, absolutely wrong ideas.
Jan
The Fig 3 you are referring to has no feedback, the circuit is in open loop just as my test in #1837 and only getting a stimulus at In-, so feedback correction is impossible.
When offering a signal in a closed loop system to the In+ input , you will have the effect that you describe, so you must confuse one with the other.
In the meantime, with the link in your posting it seems we have driven EDN to madness by closing this entry.
Hans
I think the EDN link is closed because Michael Kiwanuka, Moo Koo, or whetever he calls himself these days (at one time he used the handle 'Security Update' or something like that) keeps on sending emails to the participants.
Those participants have repeatedly asked to be taken of the distrib list but he does not respect that, the man is a loose cannon.
Jan
Those participants have repeatedly asked to be taken of the distrib list but he does not respect that, the man is a loose cannon.
Jan
Originally Posted by Hierfi View Post
I'm interested in hearing from those that understand that there must exist a time delay between an input stimulus and feedback correction, and can offer any support or criticism of my submission, in whole or in part.
———-
The signal ‘flight time’ in an audio amplifier using feedback is measured in nano-seconds worst case. Similar by the way for non- feedback ampifiers also.
I would be very wary about invoking feedback ‘delays’ or latency to try to understand how feedback works. There is essentially zero (nada, none, SFA) delay.
Phase shift is not a delay but is a property of all circuits that posses reactance, which an amplifier does (compensation capacitor).
(Tube amps have much larger phase angles between the input signal and the feedback signal because of the transformers - I’ve heard of folks claiming that tube amps sound better because there is less ‘delay’ around the loop)
I'm interested in hearing from those that understand that there must exist a time delay between an input stimulus and feedback correction, and can offer any support or criticism of my submission, in whole or in part.
———-
The signal ‘flight time’ in an audio amplifier using feedback is measured in nano-seconds worst case. Similar by the way for non- feedback ampifiers also.
I would be very wary about invoking feedback ‘delays’ or latency to try to understand how feedback works. There is essentially zero (nada, none, SFA) delay.
Phase shift is not a delay but is a property of all circuits that posses reactance, which an amplifier does (compensation capacitor).
(Tube amps have much larger phase angles between the input signal and the feedback signal because of the transformers - I’ve heard of folks claiming that tube amps sound better because there is less ‘delay’ around the loop)
I’ll try to find it - I plotted the curves, printed it out and annotated it.
(I am not going to start an argument over what an H bridge is. There are more important things to think about 🙂
It had the intended effect of causing a member of the v.f. camp to at least partially rethink the understanding of how the circuit works. The discussion has since moved on, and this effect has not been mentioned for some time.
But since you've raised the matter again, please be so kind as to explain how a transistor in which the magnitudes of gm vbe and vce / ro are comparable can be considered to be a pure transconductor.