Misplaced Pages

Talk:Negative feedback: Difference between revisions

Article snapshot taken from Wikipedia with creative commons attribution-sharealike license. Give it a read and then ask your questions in the chat. We can research this topic together.
Browse history interactively← Previous editNext edit →Content deleted Content addedVisualWikitext
Revision as of 22:36, 13 October 2014 editBrews ohare (talk | contribs)47,831 edits Similarities between the negative feedback amplifier and error-controlled regulation← Previous edit Revision as of 22:40, 13 October 2014 edit undoBrews ohare (talk | contribs)47,831 edits Similarities between the negative feedback amplifier and error-controlled regulationNext edit →
Line 965: Line 965:
::2 To clarify: It is my position that Black's amplifier ''is'' a form of error-controlled regulation. It is ''not'' my position that Black's amplifier is a form of error-controlled regulation ''that opposes ΔA''. As you have correctly pointed out, nothing opposes ΔA. So it follows that I ''don't'' think it matters. Do ''you'' think it matters? ] (]) 19:50, 13 October 2014 (UTC) ::2 To clarify: It is my position that Black's amplifier ''is'' a form of error-controlled regulation. It is ''not'' my position that Black's amplifier is a form of error-controlled regulation ''that opposes ΔA''. As you have correctly pointed out, nothing opposes ΔA. So it follows that I ''don't'' think it matters. Do ''you'' think it matters? ] (]) 19:50, 13 October 2014 (UTC)


{{od}}OK, you propose that &beta;V<sub>out</sub> is an essential variable of the open loop amplifier, suggesting that the open-loop system maintains its normal operation by comparing the value of &beta;V<sub>out</sub> with some set point value, say V<sub>in</sub>, and forcing this variable toward its set point. That is how it operates. However, that is its normal operation, and there is no regulation going on here, no overcoming of a disturbance from normal operation. This point of view does nothing to explain why one would choose to use feedback in this way rather than simply to directly amplify the input signal with an open-loop amplifier of gain 1/&beta; and be done with it. Any explanation? ] (]) 21:51, 13 October 2014 (UTC) {{od}}OK, you propose that &beta;V<sub>out</sub> is an essential variable of the open loop amplifier, suggesting that the open-loop system maintains its normal operation by comparing the value of &beta;V<sub>out</sub> with some set point value, say V<sub>in</sub>, and forcing this variable toward its set point. That ''is'' how it operates. However, that is its normal operation, and there is no regulation going on here, no overcoming of a disturbance from normal operation. This point of view does nothing to explain why one would choose to use feedback in this way rather than simply to directly amplify the input signal with an open-loop amplifier of gain 1/&beta; and be done with it. Any explanation? ] (]) 21:51, 13 October 2014 (UTC)


Given that one could simply use an open loop amplifier, why introduce feedback? Recognized explanations for using feedback include benefits other than desensitivity to gain variations, such as increased bandwidth, greater linearity and so forth. But these last have nothing to do with coping with disturbances provide no opening for the notion that feedback is introduced to provide error-controlled regulation. Rather, feedback is introduced for these other benefits. ] (]) 22:12, 13 October 2014 (UTC) Given that one could simply use an open loop amplifier, why introduce feedback? Recognized explanations for using feedback include benefits other than desensitivity to gain variations, such as increased bandwidth, greater linearity and so forth. But these last have nothing to do with coping with disturbances and provide no opening for the justification that feedback is introduced to provide error-controlled regulation. Rather, feedback is introduced for these other benefits unrelated to regulation. ] (]) 22:12, 13 October 2014 (UTC)


Historically, Black was very concerned with gain variations, and they were a major impetus for his use of feedback. Perhaps you will recognize that a comparison of how feedback deals with disturbance by desensitization is a departure from the approach of homeostasis? ] (]) 22:31, 13 October 2014 (UTC) Historically, Black was very concerned with gain variations, and they were a major impetus for his use of feedback. Perhaps you will recognize that a comparison of how feedback deals with disturbance by desensitization is a departure from the approach of homeostasis? ] (]) 22:31, 13 October 2014 (UTC)

Revision as of 22:40, 13 October 2014

This is the talk page for discussing improvements to the Negative feedback article.
This is not a forum for general discussion of the article's subject.
Article policies
Find sources: Google (books · news · scholar · free images · WP refs· FENS · JSTOR · TWL
Archives: 1, 2, 3, 4Auto-archiving period: 30 days 
This article has not yet been rated on Misplaced Pages's content assessment scale.
It is of interest to the following WikiProjects:
Template:Wikiproject MCB Please add the quality rating to the {{WikiProject banner shell}} template instead of this project banner. See WP:PIQA for details.
WikiProject iconSystems: Cybernetics High‑importance
WikiProject iconThis article is within the scope of WikiProject Systems, which collaborates on articles related to systems and systems science.SystemsWikipedia:WikiProject SystemsTemplate:WikiProject SystemsSystems
HighThis article has been rated as High-importance on the project's importance scale.
Taskforce icon
This article is within the field of Cybernetics.

This is the talk page for discussing improvements to the Negative feedback article.
This is not a forum for general discussion of the article's subject.
Article policies
Find sources: Google (books · news · scholar · free images · WP refs· FENS · JSTOR · TWL
Archives: 1, 2, 3, 4Auto-archiving period: 30 days 

Lead sentence/definition (only) thread

Definition of negative feedback

The following thread is intended for use only for discussion of the lead sentence (any off-topic discussion will be moved somewhere else on this page without mercy or warning.)

I note that Encyclopedia Britannica defines negative feedback in 'coupling amplifiers' to be:

"In negative feedback, the feedback signal is of a sense opposite to the signal present at the point in the circuit where the feedback signal is applied."

It's clumsy, but so far as I can tell that's nearly a general definition that applies to DC behavior, as well as AC behavior, although it's restricted by the language to circuits.

There's a somewhat similar definition used at hyperphysics:

"...part of the output signal is taken back to the input with a negative sign."

I think the concept of a 'signal' is very general, and could be applied to other systems than circuits, but the 'taken back' bit is too vague.

For example, something like this seems to get the gist of it:

"In negative feedback, an output signal from a system feeds back so as to reduce an input signal of that system."

(I noticed that sometimes negative feedback involves subtraction, but other times it's attenuation; but in both cases the term 'reduce' covers it and it solves the 'taken back' vagueness.)

It 'feels' right, and it would also appear to cover geophysical things like cloud formation, where the cloud forms due to heat of the sun, rises up and condenses and blocks out the sun.

Is that considered better than what we have at the moment? Can anyone improve it or have a clearly better definition?GliderMaven (talk) 01:51, 4 September 2014 (UTC)

It's never attenuation; always subtraction (unless the signals are levels, as in an automatic gain control, but that's best recast in terms of subtraction of levels, I think). I got a lot of pushback when proposing a definition based on input and output. It's one common type of definition, when talking about amplifiers, but is not quite general enough as a complete definition. Dicklyon (talk) 02:08, 4 September 2014 (UTC)
I think attenuation does count, but either way AGC doesn't seem to be a counterexample. If there's a counterexample, then we need it, so we can try to generalise to cover it.GliderMaven (talk) 02:19, 4 September 2014 (UTC)

Earlier proposal

An earlier proposal, well sourced:

Negative feedback is feedback that opposes change.
Sources
  1. Annabel Beerel (2009). Leadership and Change Management. SAGE Publications Ltd. p. 52. ISBN 9781446205655. A negative or self-correcting feedback loop describes system behavior that opposes change
  2. Helen E. Allison, Richard J. Hobbs (2006). Science and Policy in Natural Resource Management: Understanding System Complexity. Cambridge University Press. p. 205. ISBN 9781139458603. Balancing or negative feedback counteracts and opposes change
  3. Jack Andrew Morton (1971). Organizing for innovation: a systems approach to technical management. McGraw-Hill. p. 13. Negative feedback occurs when a change in input or action of the system is opposed by the output fed back
  4. Santiram Kal (2009). Basic Electronics: Devices, Circuits and IT Fundamentals. PHI Learning Pvt. Ltd. p. 191. ISBN 9788120319523. If the feedback signal reduces the input signal, i.e. it is out of phase with the input , it is called negative feedback.

Brews ohare (talk) 02:30, 4 September 2014 (UTC)

I was OK with that one proposed by Trevithj. Unfortunately it seems to have been put in with kilobytes of extra baggage, so it got caught up in a big revert. Dicklyon (talk) 02:39, 4 September 2014 (UTC)
I quite liked that one too. :) Glidermaven's suggestion of a 'signal' seems similar to the idea of "information about the gap" from Ramaprasad. Frustrating that there doesn't seem to be a free online source for his paper - it is very good. [
Arkalgud Ramaprasad, On The Definition of Feedback, Behavioral Science, Volume 28, Issue 1. 1983. Accessed on 16-03-2012.
] (talk) 02:58, 4 September 2014 (UTC)
Anyway, a refinement of Glidermaven's proposition based on Ramaprasad's definition could be:
In negative feedback, information about a gap (between a system parameter and a reference value) is used to reduce the gap.
Is that broadly applicable? To touch on Dicklyon's point, I feel we should get away from control-of-output-by-adjusting-input cases. There are alternatives, especially in non-electronic systems. The control of water level in a hydroelectric plant's reservoir is a possible example: an internal state is being regulated by altering one of the outputs. Trevithj (talk) 20:57, 4 September 2014 (UTC)
Negative feedback is a form of feedback, so the primary purpose of the definition is to demarcate negative feedback from feedback in general. Feedback does not necessarily take the form of a signal, unless that term is broadly understood to include such things as a glance, a warning beep, or even a detailed recommendation. Also, negative feedback is not always of a form that can be 'subtracted' from something else in a quantitative fashion. The link to feedback and the notion of 'opposition to change' avoids the need to go into such details. Brews ohare (talk) 13:00, 4 September 2014 (UTC)
That 'definition' "Negative feedback is feedback that opposes change" is both awful and wrong. It's awful because 'change' is ridiculously vague, and it's wrong because negative feedback doesn't characteristically oppose change. On the contrary, it actively creates change to try to bring things back to an equilibrium state.
For example, if the room temperature is 10C and the thermostat is set at 20C, negative feedback DOES NOT keep the room at 10C- it warms it up to 20C. In what sense is negative feedback opposing change there??? The answer is, it isn't.
Either a thermostat is not using negative feedback or that definition is deeply, deeply flawed. If you try to claim that thermostats don't use negative feedback in that situation, I will laugh at you.
Nobody who has understood that 'definition' and really understands negative feedback can think that that is correct or any good. It's worse than simply being wrong, because it's super vague as well as wrong. Let us never mention it again.GliderMaven (talk) 01:41, 5 September 2014 (UTC)
Huh? The negative feedback loop there would be in the reservoir control system where the input is the level measurement and the output is the valve setting. The internal state that is controlled would often be the difference between a function of the spill valve setting and the level input. And it's not necessarily an electronic control system, it could be mechanical or hydraulic; a 'signal' is a very general concept that includes shafts turning, water overflowing into a bucket, virtually anything, provided it indicates something to the mechanism, causes it to do something.GliderMaven (talk) 02:00, 5 September 2014 (UTC)
GliderMaven: There is a problem with your attitude here: while most are searching for a solution, you are busy portraying everyone but yourself as lacking all understanding. That is not conducive to progress.
Another problem is your lofty stance that the proposed definition is 'wrong' , despite its adoption by many, many published authors, of which four are cited and quoted verbatim. Even supposing you were correct about this, WP could report this popular view, and if there were those, like you, who felt there were better alternatives, then those views would be reported too, and sourced. If a balanced view of all published definitions suggested some were preferable to others, possibly the published arguments pro and con could be summarized.
The objective here is to present an accurate portrayal of published opinion, not the most persuasive and compelling opinion devised by the minds of Talk-page editors.
To turn to the substance of your objections, it is sufficient to request that you support them with sources. As matters rest, your description, that feedback does not oppose change, but creates it, is both unsourced and inadequately verbalized. For example, in an error-correcting regulator, the cause of a departure of a system from it's status quo is a disturbance (by definition) and the departure instigates the feedback that corrects for the disturbance, not creating change, but nullifying it. Brews ohare (talk) 03:28, 5 September 2014 (UTC)
Look I'm not the one making the positive claim that this is correct; you are. I'm simply stating I don't understand it, that it at least seems to be quite incorrect, that I cannot explain how this can really be valid. And the first source you are relying on is "Leadership and Change Management". I don't consider that to be in any way, shape or form, a quality source for negative feedback. It's a flimsy 'management theory' pile of shit. This is primarily a hard-technical article, not sociology.
I repeat my question, precisely change from what, when? From an equilibrium? If so you must say that. For if it's not that, if I have an object that is constantly being perturbed, is negative feedback really correcting each of these changes? It isn't. If something moves the temperature down 5 degrees, and something else moves it up 2 degrees, is the feedback really opposing the second change? No, the feedback is still in the same direction and it is NOT opposing it. In fact because feedback can be non linear, the feedback signal may be completely unchanged by the second change; a thermostat is either on or off, and it's on throughout the second change. It's not reacting to the change, it's reacting to the displacement from equilibrium. The definition is wrong.
You said: For example, in an error-correcting regulator, the cause of a departure of a system from it's status quo is a disturbance (by definition) and the departure instigates the feedback that corrects for the disturbance, not creating change, but nullifying it.
But crucially, the definition you favour does not say that it 'corrects disturbances', it says opposes changes. That's not the same thing at all. 'Disturbance' can imply things have an optimum position and it's been moved away (disturbed) from that, whereas a change by definition, unless otherwise specified, is simply any difference, and has no such implication. Different words connote different meanings, otherwise they wouldn't be different words. The definition as it is stated is clearly wrong.GliderMaven (talk) 04:58, 5 September 2014 (UTC)
It is my bad to use the term 'corrects disturbance', which is a presumption that Ashby shows is theoretically impossible. Feedback only opposes the disturbance.
Let's debate sources, not opinions. We know that four sources quoted above use the 'opposes change' description of negative feedback. The other proposals are unsourced. Brews ohare (talk) 05:15, 5 September 2014 (UTC)
Yes, opposing the disturbance is better. Whether or not the other proposals are unsourced, but that particular use of those sources is untrustworthy. We cannot in good faith put material into Misplaced Pages as true, if we know it to be false.GliderMaven (talk) 10:41, 5 September 2014 (UTC)
Re the reservoir example: yes, I understand what the feedback loop is. My point is that the input to the control system is not the output of the hydroelectric system (outflow water and electricity) but an internal parameter. And the output of the control system does not influence the input to the hydroelectric system. The exact wording was "reduce an input signal of that system." If anything, it would increase an input signal (open the spill valve more). Trevithj (talk) 07:57, 5 September 2014 (UTC)
Yes, I don't think that the dam itself is a feedback loop. The level control system clearly is or can contain a feedback loop; but I don't think the dam is actually inside the loop. There's no control of an input to the dam. The dam and hydroelectric system overall contains a feedback loop to control the spill valve, but it's a fallacy of composition to think that the dam is part of the feedback loop. It's the other way around, the feedback loop is part of the dam. Likewise, just because my car contains feedback loops, doesn't mean my car is a feedback loop. You have to draw the (sub)system boundaries correctly to be able to identify the loop. My car, and the dam is certainly affected by the behavior of feedback loops, speed and level respectively.GliderMaven (talk) 10:41, 5 September 2014 (UTC)

GliderMaven: Regarding the sourced definition that negative feedback is that form of feedback that "opposes change", you say:

"We cannot in good faith put material into Misplaced Pages as true, if we know it to be false"

Playing with words, to say that 'opposing a disturbance' is not 'opposing a change' it is 'creating a change' is just silly word play. Now, where is the source that says this definition is "false"? Opinion of WP editors is not a basis.

Sources supporting the definition are:

Brews ohare (talk) 14:31, 5 September 2014 (UTC)

I don't have to come up with specific references to refute individual references, particularly if they're low quality ones that nobody in their right mind would critique. It's not playing with words to point out that your references are poor quality, nor to point out that the definition that you've pulled from those references is inconsistent with the other definitions.
This ultimately goes to your poor judgement. Why would we as a group allow you to edit the article when you continue to show such poor judgement? Referencing a work by management consultants in this context????GliderMaven (talk) 16:01, 5 September 2014 (UTC)
GliderMaven: Your opinion of these references is based upon your own parochial view of the subject that precludes some very common subject areas like education and management. They also include an electronics text, and all refer their definitions to control system usage. You have provided, actually, no references of your own. Brews ohare (talk) 17:11, 5 September 2014 (UTC)
I actually did previously check the electronics source, although it had somewhat similar wording, you have not accurately summarised it. In any case it wouldn't matter if they all used the same wording, that doesn't mean it agrees with the references in the generality.GliderMaven (talk) 17:34, 5 September 2014 (UTC)

So the candidates that aren't obviously wrong right now seem to be:

  • In negative feedback, an output signal from a system feeds back so as to reduce an input signal of that system.
  • Negative feedback is feedback that opposes disturbances.
  • In negative feedback, information about a gap (between a system parameter and a reference value) is used to reduce the gap.

The problem I have with the second one is that it's not entirely mechanistic; it's phenomenological. If there's a feedback loop that isn't a negative feedback loop opposes disturbances then the definition is faulty. Since in the generality feedback includes chaotic systems, it's not clear that it's correct.GliderMaven (talk) 17:34, 5 September 2014 (UTC)

GliderMaven (talk) 17:34, 5 September 2014 (UTC)

Your objection to the second definition, which you have significantly distorted by replacing "change" with "disturbances" thereby eliminating the negative feedback amplifier as an example of negative feedback, is that perhaps an exception can be found - a feedback system that opposes change but isn't negative feedback. Of course, we can come up with other approaches to combat change, like feed-forward systems. However, we are limiting ourselves to distinguishing between feedback systems. The huge number of authors that seem to think the distinction is the contrast between opposition to change versus reinforcement of change, seems to indicate that if there are some odd examples where positive feedback reduces change, they are sufficiently uncommon that the 'opposing-change' definition is not misleading except possibly in some unidentified peculiar circumstances, which could be earmarked if they ever show up. So I think this objection based upon a hypothetical exception is not important.

More references can be added, as Dicklyon has shown. Here are three:

  • "Balancing or negative feedback counteracts and opposes change" 1
  • "Most environmental systems are dominated by negative feedback, which opposes change." 2
  • "Negative feedback is a circular chain of effects that opposes change" 3

Brews ohare (talk) 17:48, 5 September 2014 (UTC)

The second isn't technically incorrect, negative feedback certainly can oppose changes in the context of environmental systems, but I don't read that as a definition of NF. The 3rd of those, is taken it out of context. The next but one sentence defines what they mean by 'change' as "When part of a system changes too much from what it should be..." without that context it's too vague, and as I have already shown, wrong; because it implies something that is not true, that negative feedback opposes all changes, which it certainly doesn't, that's damping, not negative feedback. If you include all the sentences from the third definition, it broadly seems to be correct, although I still have some minor qualms.GliderMaven (talk) 20:41, 5 September 2014 (UTC)

If there is a valid criticism of this definition it is that most of the sources using it are thinking about homeostatic uses of feedback, restoration of status quo, while the negative feedback amplifier is not at all about counteracting disturbances, but about letting disturbances take their course and making the system indifferent to their presence. The definition still applies, but only rather technically speaking, as an opposition to the input signal variation, albeit with a different design motivation than restoration of equilibrium by nulling the input signal. Brews ohare (talk) 18:05, 5 September 2014 (UTC)

I certainly wouldn't say that negative feedback amplifiers aren't about counteracting disturbances, indeed they are specifically used to do that very thing, for example they're used to largely remove cross-over distortion in audio amplifiers, which very much is a type of disturbance.GliderMaven (talk) 20:41, 5 September 2014 (UTC)
GliderMaven: The 'opposes-change' formulation could be reworded to make clear that it's intention is to distinguish between the two types of feedback, positive and negative. It is not suggesting that it opposes all or any change, but that if you are having trouble deciding if the feedback is positive or negative, just ask: "is the feedback reinforcing or opposing the change?"
As for removal of cross-over distortion using the negative feedback amplifier, if a signal exhibiting cross-over distortion is fed as the input signal to the circuit of the lede, that distortion will remain in the output, which ideally is simply an amplified replica of the input.
And as always, I await any presentation of a sourced opinion. Brews ohare (talk) 02:28, 6 September 2014 (UTC)

part 2

Re the reservoir example. It seems the reference for input/output is the control system, rather than the system being controlled. Hmm. Applying the proposed definition to the reservoir control system gives us something like:

  • If "an output signal from a system feeds back so as to reduce an input signal of that system"
  • and if "the input is the level measurement and the output is the valve setting",
  • then the valve setting signal feeds back so as to reduce the level measurement.

That seems another way of saying "opening the spill valve lowers the water level". And it doesn't say anything about stopping the spill once the level is low enough. If we say "so as to reduce the gap between the level measurement and the desired level" then it seems OK. Trevithj (talk) 21:44, 5 September 2014 (UTC)

iIt seems that you object to the 'opposes change' method of identifying negative feedback because you aren't satisfied with mere identification, but want to go further into description. If the only form of negative feedback were error-controlled regulation, the 'gap' approach would work. But now you have the problem of showing that in fact every example of negative feedback is an instance of error-controlled regulation. You may think, despite what I think is conclusive evidence that there is no reduction of any so-called 'gap' in the negative feedback amplifier, that despite the 'error' being I / (1+BA) and no attempt made to push it to zero, this is still a case of error-controlled regulation. However convoluted the argument for such classification in this example, it pales when we look at examples from education, management and psychology. Brews ohare (talk) 04:41, 6 September 2014 (UTC)
The point has been raised that "change" is ambiguous. I'm hoping that identifying a reference point removes that ambiguity. A reference point seems fundamental to the act of measuring anything.
I agree that error-controlled regulation can be described in terms of a gap. It doesn't follow that other forms of regulation cannot be so described. What is a change but a gap between a past measurement and a present one? No gap, no change. Trevithj (talk) 05:53, 6 September 2014 (UTC)
'Change' is not ambiguous, and does not imply a gap. It means that the present is different from the past. Opposition to change means trying to keep the present like the past. So I wear a raincoat that'll work for any forecast; I stay dry. I design for a gain of 1/B now and forever, using a design that doesn't depend on any knowledge of whatever strange A someone sticks in the circuit. I don't need to measure some gap from 1/B and marshal a response to close the gap tailored specifically to what A is doing to the system right now- I don't care what A is doing. I don't need to measure the gain: I trust my theoretical analysis that my design will produce 1/B, and build in no checks. My strategy is fixed and defends against all comers, regardless of their behavior. I stay dry. Brews ohare (talk) 15:08, 6 September 2014 (UTC)
I stand corrected. "Change" is not ambiguous, it is vague. It needs to be clarified by giving it a reference point. For example, your "opposition to change" describes comparing the present value of a parameter of interest to a value it held in the past (the reference point.) If we ignore the past value, how do we know the value has changed? Trevithj (talk) 05:15, 7 September 2014 (UTC)

Two definitions of negative feedback?

A different approach to definition would be to introduce two definitions.

One of these definitions refers to feedback in error-correction, which can take two major forms:

  • Homeostasis: Systems where there is no output, where feedback serves only to setting the essential variables of the system to their set point. Examples might be blood pressure, or body temperature where the only 'output' is a reaction that reduces the departure of the system from its proper condition.
  • Process control: Systems engaged in manufacturing or producing something, where the factory is kept 'in spec' using feedback, either by setting the essential variables of the system to their set point, or possibly, by adjusting the production system so the monitored quality of the output is held within specs for the product.1 This last approach is much more sophisticated, because the process might vary remarkably to keep the product in spec, for example, in response to changes in the input materials.

A second definition involves feedback in systems that do not monitor system departures from desired operation, and do not attempt to minimize departures by controlling the disturbance. Instead feedback is used to render the system indifferent to disturbances, which have no effect on operation regardless of their presence or exact nature. The poster child for this approach is the negative feedback amplifier. Here is a verbal description of its operation that was once presented on WP:3

"An amplifier with too large an open-loop gain, in a specific frequency range, will additionally produce too large a feedback signal in that same range. This feedback signal, when subtracted from the original input, will act to reduce the original input, also by "too large" an amount. This "too small" input will be amplified again by the "too-large" open-loop gain, creating a signal that is "just right". The net result is a flattening of the amplifier's gain (desensitizing)"

I under-appreciated this description when I first encountered it. On revisiting it, I believe it provides without math a verbal understanding of how the gain 1/B is achieved regardless of open-loop gain differences at different frequencies. The same explanation could be generalized to explain desensitization to variations in open-loop gain of any kind. Of course, there are restrictions upon how well this adjustment works that must be left to mathematical analysis. (For example, step response.) Reference to desensitization as the goal of the feedback in the negative feedback amplifier are readily located. Brews ohare (talk) 21:55, 5 September 2014 (UTC)

The description needs to be appreciated less. It's absurd. Dicklyon (talk) 02:15, 6 September 2014 (UTC)
Well, Dick it's a WP discussion that made it past WP review into print! My initial reaction was like yours, but I think it's accurate. If you follow a signal around the loop, the large open-loop gain amplifies a signal reduced by BO, a larger reduction for larger A, which reduces the input into the open-loop amplifier to a level that is in fact "just right" for amplification by A to obtain O. I think that's what the description says. Care to try wording it better? Brews ohare (talk) 02:42, 6 September 2014 (UTC)
Well, nothing is "too small" or "too large", so the whole basis for the discussion seems silly. It came in here by an anon, years ago, and lasted until you took out some of it here. Dicklyon (talk) 03:53, 6 September 2014 (UTC)
For me the description suggests verbally what happens tracing the feedback loop, and that helps to understand that the feedback is a different operation than one based upon error control. The is no measurement of a departure from 'normal' operation, and no minimizing of such a departure by a regulator. Of course, the mathematical analysis doesn't use those ideas either, but this may be a more intuitive explanation. It seems that grasping the 'design for immunity' use of feedback seen here is hard for some here to separate from the 'fix departures from the norm' use for feedback found in error-controlled regulation. Brews ohare (talk) 04:20, 6 September 2014 (UTC)
To be strictly accurate that is a rough description of how an underdamped negative feedback loop will often operate. But not a critically, or overdamped one.GliderMaven (talk) 12:26, 6 September 2014 (UTC)
'It seems that grasping the 'design for immunity' use of feedback seen here is hard for some here to separate from the 'fix departures from the norm' use for feedback found in error-controlled regulation.' Yeah. Just because you've stuck different labels on the same damn circuit, or because the blocks are implemented differently internally, or if the signals going around the loop are volts instead of water or shafts turning, doesn't mean they're actually logically different if they have the same function in the context of the loop. This topic is not about the words we use to describe it, it's about the mathematical and logical relationships; negative feedback is an identifiable pattern, not something necessarily designed for any human purpose at all, many, many natural processes involve negative feedback.GliderMaven (talk) 12:26, 6 September 2014 (UTC)

Maybe you can explain what exactly is the departure error that is measured in the feedback circuit of the lede? Then you can explain what is the regulator function in this circuit that reduces this error. With sources, preferably. There are neither of these elements present. The circuit simply is designed to produce a gain of 1/B and the circuit is impervious to variations in open-loop gain. Somewhat like GliderMaven does not address arguments, but bypasses them. Brews ohare (talk) 15:22, 6 September 2014 (UTC)

Another try

I had a go at composing a definition for the opening sentence:

"Negative feedback occurs when the output of a system, process or mechanism is fed back into the input in such a way as to reduce or eliminate changes in the output."

And GliderMaven immediately slapped the 'dubious' tag back on it. Could GliderMaven please suggest a wording that he'd be happy with, rather than just tagging or opposing anything anyone else proposes?

Let's see if we can build up a consensus step by step?

1. "Negative feedback when some function of the output of..." - Any alternative suggestions to that?

2. "... of a system, process or mechanism..." - Any alternative suggestions to that?

3. "...is fed back into the input..." - Any alternative suggestions to that?

4. "...in such a way as to reduce or eliminate changes in the output." - Any alternative suggestions to that? DaveApter (talk) 16:10, 6 September 2014 (UTC)

Dave: Happy to have a new voice here. The input/ output formulation seems adapted to process control, but how can it be worded so it is clear how it applies to homeostasis where there is no input/output but only an internal response to defend against some disturbance? Brews ohare (talk) 17:05, 6 September 2014 (UTC)
Another thing to chew on: in the negative feedback amplifier the output is not what is fed back, but a fraction B of the output. And not with the aim of reducing or eliminating changes, but of making the output 1/B times the input, that is, the output is an enlarged version of the input. None of this fits well with the beginning sentence, so what is to be done? Brews ohare (talk) 17:15, 6 September 2014 (UTC)
Brews, since your continued dominance of this talk page makes it impossible for people to hear each other, why don't you try stepping back again and see where this goes? Dicklyon (talk) 17:20, 6 September 2014 (UTC)
OK. I see no evidence of my domination; anyway, I have had no influence in bringing sources to bear. Editors here simply avoid my attempt to discuss sources, and instead present their personal formulations divorced from sources. We have already a very succinct and general definition for distinguishing between positive and negative feedback, one you have endorsed, and one with myriads of sources, but editors want to ornament it like congress attaches pork to straightforward bills. Brews ohare (talk) 17:39, 6 September 2014 (UTC)
The evidence shows that you have made 66% of the edits and have created 60% of the text on this talk page since its inception, almost all very recently. Most of the rest, not much over one-third of the total, is all other editors pushing back against you. Dicklyon (talk) 22:07, 6 September 2014 (UTC)
You measure dominance by number of edits, but if we judge by success I've got nowhere at all. Dick, you are not alone in preferring your unsourced opinion based upon personal intuition to finding what sources say. And where sources have been pointed out that don't agree with you, you are not alone in simply saying the sources got it wrong, or are lacking in some respect, without any source to supports those claims! Brews ohare (talk) 02:14, 7 September 2014 (UTC)
Right, your dominance has proved counterproductive, yet you keep it up. Give it a break, as you said you would. Dicklyon (talk) 02:37, 7 September 2014 (UTC)

Thanks for your questions Brews. The way I see it is that in a homeostatic process, such as temperature regulation, the input is the temperature and the output is the bodily activity to heat it (eg shivering or raising hair fibres to thicken the coat) or cool it (eg perspiration). That output is fed back to the input in either the raising or lowering of temperature to restore it to normal. Do you think a revision of the above wording would help to make this clear?

As for the second point, whether it is all of the output or a fraction of it is covered by my phrasing "... some function of the output". And yes, actually the gain is reduced by the action of the feedback. An op-amp might have an open-loop gain of maybe 10,000, and this is reduced to a gain of 10 if you had for example a 9k and 1k resistor divider network in the feedback loop. DaveApter (talk) 22:00, 6 September 2014 (UTC)

I agree on the amplifier that this works fine; I don't understand Brews's objection. On the homeostasis, I'd take temperature as the output, and heat as the input. If the environment is putting in too much heat, the organism senses increasing temperature and feeds that back to sweat or pant or whatever to let heat out or reduce the excess heat input to reduce the temperature fluctuation. Dicklyon (talk) 22:11, 6 September 2014 (UTC)
Sorry to intrude, but sources could help with vocabulary and concepts. Reading Ashby might help. An 'essential variable' is one that must be held in range for things to work properly. Temperature is an example. An essential variable is compared with a 'setpoint'. It's departure from setpoint is measured as an 'error' and goes to a 'controller' that in turn instructs a 'regulator' to take action. In the case of thermostatic control, the furnace adjusts heat to correct the error and, as Dick says, the heat is input, some of which is corrective feedback. So one might take the 'error' as one form of 'output' and the action taken by the regulator as contributing to 'input'. There is no need to reinvent the wheel here: the vocabulary can be sourced. In the case of the negative feedback amplifier, no source uses this vocabulary. The terms 'error' (I–BO), 'input' (I), and 'output' (OI/B) have different meanings, and regulation doesn't come up. Brews ohare (talk) 23:20, 6 September 2014 (UTC)
Hi DaveApter. Your proposed definition may work, provided that:
  1. the "system, process or mechanism" referred to is actually the control system, and not the system being controlled, and
  2. the changes in the output are defined with respect to some explicit reference.
Point 1 is somewhat un-intuitive. I would expect the "system" in the definition to refer to the system-being-controlled. But this produces some confusion and awkwardness.
May I present a possible counter-example: using spill-valves to control water level in a reservoir. Glidermaven and I discussed it (above). If the system is the reservoir then the control is via an internal measurement (water level) being fed "back" to the output (spill valve). That is awkward, but it does work if "system" refers to the control process (see above). A similar point holds with the temperature regulation example: what counts as "input" depends on which system you are talking about. For the control system, input=temperature. For the body system (Dicklyon's point, I believe) input=heat.
Also, Glidermaven made the point that not every change is opposed. After all, a correction is a change too. It would be a pretty poor regulator if it opposed its own corrections. So for that reason, "change" needs to be with respect to a reference point to make this clear. The temperature is compared to a desired internal temperature. The water level is compared to a safe maximum. It is the difference (the "gap" between measures) that is reduced/eliminated, not the absolute level of the output.
Thank you for the fresh input - you have helped me clarify my thinking. Trevithj (talk) 06:11, 7 September 2014 (UTC)
I don't understand why/how you would look at input and output backwards that way, with respect to the thing in the feedback path instead of the intended overall system input/output. Dicklyon (talk) 06:17, 7 September 2014 (UTC)
It is un-intuitive to do so. But take the example of the reservoir: the overall system input is a river. This is not controlled as far as the reservoir is concerned. The overall system output is the spillway. This is not measured. So if the "system" is the reservoir, the proposed definition doesn't fit this example. But it does fit if the "system" is the control system. Input=level measurement, output=signal to control valve. I'm not sure if I personally like it, but it does work. Trevithj (talk) 19:22, 7 September 2014 (UTC)
I don't understand what you meant when you said "Glidermaven made the point that not every change is opposed" - can you give an example? Thanks. DaveApter (talk) 12:35, 7 September 2014 (UTC)
It's because it's wrong. When every change is opposed, a system is subject to damping, not negative feedback. For example a dashpot opposes all changes. If I move a damper down, and then up, the damper opposes it on the up; that's not what negative feedback does. A negative feedback system will help you move it back up.
Consider a room being maintained at 20C by an electric fire/air conditioner on a thermostat. If I open the window, then the room may drop to 15C and the heater will turn on due to negative feedback; the negative feedback is very definitely opposing that particular change. But this is not a general thing, for if I then light the waste paper basket on fire; the room may warm up to 17C, but the negative feedback doesn't oppose that in any way; the thermostat is on, and stays on, since it's trying to push the temperature back to 20C.
So it's not that all changes are opposed by negative feedback, it's that changes from the equilibrium point are opposed by the negative feedback loop; the loop is active when the temperature is away from that equilibrium.GliderMaven (talk) 15:36, 7 September 2014 (UTC)
GliderMaven ignores the point indicated to him before that the definition here is intended only to separate negative from positive feedback. So the "opposes change" is to be contrasted with "reinforces change" in trying to decide what type of feedback is present. There is no intention to make "opposes change" a complete description of negative feedback separating it from all other phenomena found in the universe. It is obvious, for instance, as was also pointed out to GliderMaven before, that feed-forward can oppose change, but is not a form of negative feedback. Brews ohare (talk) 20:00, 7 September 2014 (UTC)
What you're doing is exactly like trying to define cats as being lions. You can say "well, I'm just trying to distinguish cats and dogs", but no, what you're doing is failing to define cats correctly.
If there were only two types of feedback, or even one type of negative feedback, then you might be correct. But as I have pointed out several times now, a feedback system that opposes change is the specific definition of a damping feedback system. Damping can be considered a particular form of negative feedback that attenuates the velocity/state rate change vector. However, most forms of negative feedback are not based on opposing change, thus the definition is too narrow for a general work such as Misplaced Pages. Note that at least one of three references you rely on further specifies what constitutes a 'change' but you have incorrectly summarised it, and the main one you've used is largely or completely unreliable.GliderMaven (talk) 10:23, 8 September 2014 (UTC)

Not all changes are opposed, and not all that opposes is negative feedback. But we can agree that negative feedback always opposes changes from the reference/equilibrium. True? Trevithj (talk) 20:41, 7 September 2014 (UTC)

In the feedback amplifier 'equilibrium' is not really a consideration. The feedback is negative because it is subtracted from the input to the open-loop amplifier, I–BO, and the 'changes' involved are due to the signal being amplified, which are not normally referred to as disturbances from equilibrium. The opposition in this case refers to the reduction of the input to the open-loop amplifier, I–BO, which reduction brings the output variation down to 1/B from A. Brews ohare (talk) 23:00, 7 September 2014 (UTC)

The case of homeostasis

Having thought a bit more about it, in the example of homeostatic body temperature I would say that the muscular action / sweating etc constitute the mechanism, the output is the actual temperature of the blood, and the inputs are the temperature sensory organs - which I think are within the brainstem and the adrenal glands (as best I understand it; I am not an expert in physiology). The classic reference is Cannon. The conclusion of that chapter states:

"If conditions are such that there is a tendency to tip the organism in one direction, a series of processes are at once set at work which oppose the tendency. And if an opposite tendency develops, another series of processes promptly oppose it. Thus quite automatically the remarkable uniformity of the internal environment is preserved, in opposition to both internal and external disturbing conditions." DaveApter (talk) 12:27, 7 September 2014 (UTC)
Walter B. Cannon; The Wisdom of the Body, 1932, pp 177-201
Dave: Your process here, invoking a source, is a most salutary development that has so far been little used on this page. However, the discussion is enlarging beyond the subject here, which is finding a definition that distinguishes negative feedback from positive feedback. Negative feedback does play a role in homeostasis, (see this), so the definition has to work for that instance of negative feedback. But a complete understanding of homeostasis or even process control goes into complexities beyond what is needed to establish how negative feedback is fundamentally different from positive feedback. Perhaps this or this, which must reconcile with this? Brews ohare (talk) 14:34, 7 September 2014 (UTC)
Here is an approach I hadn't seen before due to Li Qiu:
“When feedback modifies an event/phenomenon, the modification will subsequently influence the feedback signal in one of three ways:
1. -the feedback signal increases, leading to more modification. This is known as positive feedback.
2. -the feedback signal decreases, leading to less modification. This is known as negative feedback.
3. -the feedback signal does not change, indicating the phenomenon is in equilibrium.
Note that an increase or decrease of the feedback signal here refers to the magnitude of the signal's absolute value, without regard to the polarity or sign of the signal.”
Brews ohare (talk) 15:53, 7 September 2014 (UTC)
That's lame; case 3 about equilibrium contradicts the premise that feedback modifies an event/phenomenon. Both positive and negative feedback systems can approach equilibrium, and this tripartite scheme doesn't help understand that. No wonder you haven't seen it before. It's also very vague about what the various increases and decreases actually mean. Dicklyon (talk) 16:55, 7 September 2014 (UTC)
I would need convincing that it is true that "Both positive and negative feedback systems can approach equilibrium," - can you give an example? I should have thought that positive feedback never produces equilibrium except in the degenerate case where the system is driven into saturation (eg an amplifier output is driven to one power rail). DaveApter (talk) 10:33, 8 September 2014 (UTC)
Dick: I don't think we need part 3 as worded. I think the author meant something more along the lines that if feedback is not activated, the system is in equilibrium. In any case, parts 1&2 are one possible solution to separating positive from negative feedback in widely applicable terms. Brews ohare (talk) 17:05, 7 September 2014 (UTC)
As for not spelling out what an increase or decrease in feedback signal means, what is the confusion? Given a system and a feedback signal, won't it be obvious what constitutes an increase/decrease in that feedback signal? Got a confusing example in mind? Brews ohare (talk) 17:16, 7 September 2014 (UTC)
Your remark that positive and negative feedback can both lead to equilibrium does not address the distinction posed, which is that positive feedback leads to increasing modification, and negative feedback to decreasing modification. It fits error-controlled regulation, where negative feedback reduces a measured departure from setpoint, and to the amplifier where a portion B of the output is subtracted, reducing the signal to the open-loop amplifier. This view is very commonly expressed. Do you disagree with it? Brews ohare (talk) 17:32, 7 September 2014 (UTC)
It is my understanding that case 3 may refer to situations where the control has to be continuous in order to keep the parameter in equilibrium. Perturbations/disturbances are not always only momentary. Thoughts? Trevithj (talk) 20:48, 7 September 2014 (UTC)
My guess, only a guess, is that case 3 was meant to read that if feedback wasn't activated (not responding), the system was in equilibrium. But in any event, case 3 is beside the point as interest is in case 1 & 2. Brews ohare (talk) 22:51, 7 September 2014 (UTC)
Let's try fitting the homeostasis example into the proposed definition:
Negative feedback occurs when the output of the mechanism is fed back into the input in such a way as to reduce or eliminate changes in the output.
Making allowances for the wording, it sort of works. However I would like to offer an alternative, as a basis for comparison:
Negative feedback occurs when information about a difference between the actual level and a reference level of a system parameter is used to reduce the difference.
This is a variation on an earlier suggestion. Exactly how the information is used doesn't need to be specified. Does it fit the homeostasis case? "System parameter"=blood temperature, "reference level"≈37°C. Trevithj (talk) 06:55, 8 September 2014 (UTC)

OK, it seems to me that we won't get a concise one-sentence definition that covers all cases. But it should be possible to agree on a summary to go into the lead that will give a clear sense of the meaning that will be fleshed out in more detail in the body of the article, with the help of various specific examples and how the principles apply in those cases.

Are we agreed on the clauses 1-3 in my suggestions at the start of this ("Another try") section above? Would replacing clause 4: "...in such a way as to reduce or eliminate changes in the output." with "...with a polarity which tends to reduce the effect of changes in the input"? I'll put that into the article now but leave the 'dubious' tag there. Could GliderMaven please remove the tag if he's now satisfied, or suggest a wording that would suit if not?

btw, I'm not happy with forms of wording suggested above which refer to "subtracting from the input signal" because I'm not convinced that this is always the case - for one example, where the polarity inversion occurs in the amplifying stage rather than in the feedback loop? DaveApter (talk) 10:34, 8 September 2014 (UTC)

Trevithj: To be sure, this subsection is about homeostasis, but the big issue is to find a very general way to distinguish between negative and positive feedback, one that works for homeostasis of course, but includes all cases. For the case of homeostasis, an example of error-controlled regulation, you have presented a couple of ways to fill in the formula:
If feedback modifies a phenomenon, and that modification causes the feedback signal to decrease, leading to less modification, that form of feedback is known as negative feedback.
The purpose of such an exercise is to see if the formula does fit homeostasis. The goal is not to see if some alternative more explicit wording can be found that fits homeostasis.
The original issue was that the lead sentence was focused too narrowly upon error-controlled regulation and did not fit well with different examples, most particularly the negative feedback amplifier.
Dave's present formulation doesn't solve this problem. It requires a non obvious understanding of what is 'input/output', which may have nothing to do with (say) the 'input/output' of a process (like manufacturing cars, or amplifying an electrical signal) and can refer instead to the departure of some essential variable of a monitoring system from its setpoint followed by commands to a regulator. The italicized formulation just above does solve it. Brews ohare (talk) 10:38, 8 September 2014 (UTC)
The widely adopted criterion that negative feedback is feedback that opposes change, while positive feedback reinforces it, seems the best way to go IMO and the main objections to it are that it is limited to the purpose of distinguishing the two forms, and doesn't spell out more specifics. That is like saying a hammer can't saw wood. Brews ohare (talk) 10:53, 8 September 2014 (UTC)
Another source supporting the view that "if increases in the output cause the feedback to decrease the output, the feedback is called negative feedback". Brews ohare (talk) 17:47, 17 September 2014 (UTC)

Another quote

From Wiener, Cybernetics p. 97

"We thus have examples of negative feedbacks to stabilize temperature and negative feedbacks to stablilize velocity...also to stabilize position... The information fed back to the controller tends to oppose the departure of the controlled from the controlling quatity, but it may depend in widely different ways on this departure".

Incidentally I don't subscribe to the view that feedback in an amplifier circuit differs in principle from in a control system or a living organism; the intention of applying feedback in say an audio amplifier is to ensure that the output signal is a more faithful reproduction of the input signal. DaveApter (talk) 16:39, 8 September 2014 (UTC)

Dave: Well, this quote fits error-controlled regulation well, speaking of departures of an essential variable from its setpoint, but is not applicable to amplifiers. Your earlier formulation was good for process control and amplifiers, but not for this.
Your notion of an audio amplifier (or an amplifier in general) is inaccurate. The function of an amplifier is to produce an enlarged exact copy of its input, including any distortion that may happen to be present. See this definition. That includes the negative feedback amplifier, where the output is simply A/(1+BA) times the input, an exact copy of the input, but of different size. Cleaning up a signal is a different function, which can be accomplished in many ways of course, such as filtering. Brews ohare (talk) 17:26, 8 September 2014 (UTC)
Of course, the negative feedback amplifier can introduce distortion, because it isn't an ideal amplifier. See, for example, the article on step response. Brews ohare (talk) 17:36, 8 September 2014 (UTC)

I probably didn't express myself clearly enough. When I said "faithful reproduction of the input signal", I was including the case of it being an enlarged reproduction of the input. And when I said "more faithful" I meant more faithful than it would have been in the absence of negative feedback (ie with a reduced amount of any distortion that would otherwise have been added by the amplifying stage), not that it would have magically eliminated any distortion presented to the input of the amplifier! DaveApter (talk) 19:23, 8 September 2014 (UTC)

I don't know if I have mistaken your meaning. Perhaps we agree that the feedback does not make the negative feedback amplifier 'improve' upon the fidelity of the input signal. The output as noted is AI/(1+BA) and exhibits every flaw present in I. Of course the feedback amplifier introduces less distortion than the open-loop amplifier because it is independent of A, if A is large, and A exhibits more sources of distortion than 1/B.
As you say, the intent of feedback is to reduce the role of unpredictable variations in function, but perhaps you also agree that the way this is done in the amplifier does not rely upon measuring a departure of an essential variable and commanding a regulator to take action to reduce this departure. There is no measured 'error' and no attempt to reduce such an error. Rather than maintain performance by combating the effects of variations in A, the gain simply doesn't depend on A (assuming A is sufficiently large),1 and no monitoring is necessary. Brews ohare (talk) 21:54, 8 September 2014 (UTC)
To date, Dave, the other editors on this page refuse to believe Kal's analysis of the above paragraph captures the role of feedback in the negative feedback amplifier. That opinion is voiced without sources for backup. Yet Kal's is the standard analysis of every electronics text of the last 40 years. Your approach using sources is the only way to cut out unsupported assertions of editor intuition. Brews ohare (talk) 14:47, 9 September 2014 (UTC)

Dave: On my talk page, rather than respond here, you have suggested my participation is counterproductive. You remark that you "did not get the impression that had attentively read what had written, or that had thought through own argument carefully before rushing to the keyboard." That reaction upsets me, as I felt I responded in detail and most cogently to your remarks regarding the role of feedback in the negative feedback amplifier, suggested that perhaps mine coincided with your own, and provided a source 1 to elucidate its analysis. You also suggest that I "Make specific suggestions about changes to the article, rather than maintaining a running debate." I have done so repeatedly, proposing a variety of approaches to the lede. Here is one and here is another.

Much of the unproductive discussion on this Talk page revolves around this original suggestion regarding the lede. You have made some changes to improve this situation.

The prolonged Talk page debate about the lede can be traced directly to the refusal by most participants to discuss sources instead of their opinions, suggesting that textbook treatments of the negative feedback amplifier are somehow incomplete, as they aren't consistent with these editors' personal and unsourced intuitions that every example of negative feedback is a particular implementation of error- controlled regulation.

At this point, I've exhausted all attempts to bring editors to consider sources. I've provided the sources, presented the sources, explained the sources, and yet editors' persist in the notion that their beliefs, if sufficiently lucidly presented, should prevail over published commentary, and have even suggested that sources don't matter.

So, as you propose and as I have proposed myself above, I will withdraw, with no optimism that sources will prevail. Brews ohare (talk) 17:44, 9 September 2014 (UTC)


Re an earlier comment by DaveApter: it may be a bit hasty to say that there isn't a way to summarize negative feedback in a generic way. I believe there is, but it is difficult to do so in a way where people don't feel an urge to delve into domain-specific details.
I'm beginning to wonder if we should split the article into two pages. I hate the idea, but the discussions in the past do seem to imply there is some sort of basic conceptual conflict. It seems to be an old problem:
From William Ross Ashby, Introduction to Cybernetics (1957) pp 53-54

"On the one side stand those ... whose aim is to get an understanding of the principles behind the multitudinous special mechanisms that exhibit them. To such workers 'feedback' exists between two parts when each affects the other... On the other side stand the practical experimenters and constructors, who want to use the word to refer, when some forward effect from P to R can be taken for granted, to the deliberate condition of some effect back from R to P by some connexion that is physically or materially evident."

Trevithj (talk) 21:32, 9 September 2014 (UTC)
I must admit, I don't really understand what precisely is meant by this, and the link is not loading for me at the moment to find the context.
Anyway, aren't we trying to define what a 'feedback loop' is, not 'feedback' itself? Is it not the case that a positive feedback loop is a loop where when you follow cause-and-effect around the loop, that that cause and effect chain has tended to increase the size of the signal at that point, whereas a negative feedback loop, is when it has tended to decrease it?GliderMaven (talk) 00:34, 11 September 2014 (UTC)
Everything that GM has written about this topic is clear, correct, and helpful. Almost everything else is either unclear, incorrect, or unhelpful. Yes, this is self-referential, and it could lead to feedback instability. Johnuniq (talk) 00:58, 11 September 2014 (UTC)
Hi GliderMaven. Do you mean 'feedback loop' as opposed to 'feedback' the signal? Yes I agree we should define negative feedback with respect to the action of the entire loop. Talking in terms of circular-causality seems more generic than in terms of output-to-input.
Regards the Ashby quote: I'm trying to understand why quite good definitions in the lede seem to get completely rewritten on a regular basis. There does seem to be a debate between theoretical approaches and practical examples - a search of this talk page for "mathematical" may clarify what I mean.
BTW: I understood that you disliked the use of "change" in the existing definition. Is that not true? Trevithj (talk) 01:49, 11 September 2014 (UTC)
While the loop-based view is a good way to distinguish negative from positive, it's not great at making clear what feedback means, which is conceptually tied to the word itself, feeding a signal from an output or sensed state back to an input (a "target" and "error" being optional parts of that, typically included in the feedback part of the loop). Analysis almost always includes a separation between a forward part of the loop and a feedback part of the loop, in both the control and amplifier applications. Dicklyon (talk) 15:13, 11 September 2014 (UTC)
Good point, especially re analysis. Agreed that the output-to-input (R→P) view is a more intuitive way of expressing feedback. On the other hand, it has problems being applied to cases of control where the output of a process is not what is being measured, and/or the input is not what is being influenced. And if an example process involves several components, then it may not always be obvious which of several inputs and outputs are being referred to. The loop-based view (P⇔R) doesn't require these details, and so is easier to apply to other cases. Trevithj (talk) 21:23, 11 September 2014 (UTC)
Typically people will just define what's being measured as an output, and all influences feeding back as inputs. With modern (c. 1970) vector matrix state-space methods, it's all covered. Any state that is "observable" is observed via outputs, and anything that is "controllable" is controlled via inputs. See or . Dicklyon (talk) 00:33, 12 September 2014 (UTC)
I think that the first few sentences in the lead is not really the place to be doing any analysis. We're not trying to define feedback; we're simply trying to identify the topic, to state what a negative feedback loop is.
That seems to be the primary thing, but any initial definition certainly shouldn't preclude the definition or relevant examples of feedforward and feedback mechanisms later in the lead or the article, but they don't seem to be primary thing anyway, nor really are inputs and outputs.GliderMaven (talk) 00:47, 12 September 2014 (UTC)

Good feedback definition

The 1947 book Theory of Servomechanisms, of the MIT Rad Lab series, which I just found in the stack by my bed, has a good definition:

2-21. Characterization of Feedback Systems.—A mechanical or electrical system with feedback is one in which the output of some part of the system is used as an input to the system at a point where this can affect its own value. A servosystem is a feedback system in which the actual output is compared with the input, which is the desired output, and the driving element is activated by the difference of these quantities.

Sure, it's a little narrow with "mechanical or electrical", and "driving element" and such concepts, but it does capture the general case of things (which he calls outputs) being fed back in where they can affect their own value, and the specialization to systems with targets and errors. We need something of this sort, not necessarily these words. We'd have to add some about negative, which that book does not do (since it treats feedback and stability more generally, where in real systems phase often tends to change from negative to positive as frequency increases). Either that or use a much simpler definition of negative feedback, like the one in this book: Negative feedback is a circular change of effects that opposes change. Dicklyon (talk) 05:04, 12 September 2014 (UTC)

This book says In the language of cybernetics, if increases in the output cause the feedback to decrease the output, the feedback is called negative feedback. Something like that combined with something like the servomechanisms characterization seems like it should satisfy all interests here. Dicklyon (talk) 05:13, 12 September 2014 (UTC)

Perhaps something like this would be a good lead:

A system with feedback is one in which the output of some part of the system is used as an input to the system at a point where this output can affect its own value. The feedback is called negative feedback if increases in the output cause the feedback to decrease the output. A common use of negative feedback is in servosystems, or control systems, in which the actual output is compared with an input representing the desired output, and the difference or error signal is used to drive the output in the direction that reduces the error toward zero.

Comments? Dicklyon (talk) 05:22, 12 September 2014 (UTC)

Maybe with less emphasis on "input/output". And perhaps we can join the first two sentences, and use less passive voice. Something like:
A system with negative feedback is one in which a measurement of some part of the system influences the system in a way that keeps the measurement within certain limits.
The servomsystems case looks useful. A nod to the history of the term might not be a bad idea either.
The term originated with the negative feedback amplifier, where an amplifier was controlled by using some of the output to cancel or "negate" its input. The term is also applied to servosystems, or control systems, where a measured output is compared with a reference value representing the desired output, and the difference or error signal is used to drive the output in the direction that reduces the error toward zero.
Hmm. Too wordy? Trevithj (talk) 06:02, 12 September 2014 (UTC)
It's not about useful definitions. We need a very, very general definition. Inputs and outputs only confuses things. How about this:
A negative feedback loop is a closed loop of cause and effect where the overall effect of the loop is to tend to reduce the measurable quantity effecting any point in the loop. GliderMaven (talk) 12:41, 12 September 2014 (UTC)
And a negative feedback signal is the part of the loop that connects a measurement to a point of influence. Trevithj (talk) 20:08, 12 September 2014 (UTC)
"in a way that keeps the measurement within certain limits" is not correct in general, and as far as I know has no basis in sources. And I don't understand the objection to input and output, which is common in sources. That seems to me to be the most common and intuitive way that feedback is defined. And I don't know think "The term originated with the negative feedback amplifier" is correct, but I could be wrong; do we know? Dicklyon (talk) 03:36, 13 September 2014 (UTC)
It's not that we have any problem with inputs and outputs, it's that a feedback loop doesn't necessarily have to have anything that you would normally call an input or an output, and you can add or specify the existence of inputs and outputs willy-nilly, and it doesn't change the essence of it, it was a feedback loop both before and after you added any number of them. For example, an example of a feedback loop with only outputs is called: an oscillator; if you take the output off, it's still an oscillator and a feedback loop.
Inputs and outputs are, practically speaking, very important, but they don't define what a negative/positive feedback loop is, and it's clear that the cause and effect loop is really the real heart of the topic.GliderMaven (talk) 14:17, 13 September 2014 (UTC)
I would argue rather that an oscillator may or may not be built using a feedback loop (as this book supports), and that if it is, then one can identify the path from an output to an input that makes it so (like here). You're right of course that it will continue to oscillate whether or not we identify such an output and input. So what? Dicklyon (talk) 14:25, 13 September 2014 (UTC)
And using "measurements" for outputs and "points of influence" for inputs doesn't really change things, just uses less familiar words for the concepts. Dicklyon (talk) 14:38, 13 September 2014 (UTC)
You don't seem to be quite getting what I'm saying. I'm saying you can build a feedback loop with no signal inputs or outputs at all, and it will do something, oscillate or behave chaotic, or go to zero, depending on the loop. Of course, you can always potentially measure something (add an output on a measurable quantity), but it's still a feedback loop when you're not doing that. So the inputs and outputs are a complete red-herring; they're completely unnecessary to the definition.GliderMaven (talk) 22:39, 13 September 2014 (UTC)
I do understand what you're saying: that it's all about the loop, and the sign of the loop gain. But it's very hard to appreciate what the loop might be, or why it's called feedback, without the notion of "back" meaning from output to input, via a path outside the system that the feedback is "around". But let's try it your way. What lead definition are you proposing? You've proposed a definition of the loop, above, in terms of its effect, as "A negative feedback loop is a closed loop of cause and effect where the overall effect of the loop is to tend to reduce the measurable quantity effecting any point in the loop." Here I assume you mean "affecting", not "effecting", but it's hard to understand what "the measurable quantity affecting any point in the loop" might refer to. Dicklyon (talk) 03:27, 14 September 2014 (UTC)
Re the last point: "familiar" is a relative term. It may be familiar to readers with (eg) an electronics background, but we aren't necessarily aiming the lede at readers who are familiar with that (or any) discipline. Trevithj (talk) 23:59, 13 September 2014 (UTC)
Nevertheless, I think we can probably agree that making up new terms makes the concepts less familiar, in general. Dicklyon (talk) 03:29, 14 September 2014 (UTC)
OK, I'm not trying to make up terms. How about: "A negative feedback loop is a closed loop of cause and effect which tends to reduce the quantity acting at each point in the loop.".GliderMaven (talk) 20:34, 14 September 2014 (UTC)
I think "the quantity acting" is not a standard term of art, nor easy to interpret in any way that makes sense. And opposing change is not really related to reducing a quantity. Dicklyon (talk) 21:07, 14 September 2014 (UTC)
By the way, the Britannica definition you quoted kind of misses the mark. It says: In negative feedback, the feedback signal is of a sense opposite to the signal present at the point in the circuit where the feedback signal is applied. The point is rather that the output is fed back to an input in such a way as to oppose changes in the output; there need not be any other (input) signal present at that point of application, so applying it in the opposite sense is not possible in general. Review the quotes from books at the start of this subsection. Also, the hyperphysics definition that you provided did have input and output, and you didn't seem to object at that point. Dicklyon (talk) 00:04, 15 September 2014 (UTC)

As GliderMaven points out, we're not trying to define 'feedback', which has its own article, but negative feedback. It might also help to use sources to supply this distinction, inasmuch as WP is supposed to report what sources say, not WP editors. Brews ohare (talk) 00:22, 14 September 2014 (UTC)

Another try at a lead

Here is another attempt at a synthesis of definitions, focusing on negative, not needing inputs, and defining an output as some measurement of the system's state:

  • A system with negative feedback is one in which an output (that is, some measurement of the system's state) is fed back into the system at a point where this output affects its own value in such a way as to oppose changes in the output.

I'm not loving it, but I think it meets most of the objections. Dicklyon (talk) 00:22, 15 September 2014 (UTC)

Doesn't attack 'negative feedback' directly, but as an aspect of a system. Could a system be eg a legal system, an educational or management system, a philosophical system, a notational system? Brews ohare (talk) 05:27, 15 September 2014 (UTC)
Hmm. All measurements are outputs, not all outputs are measurements. I'm unclear why we keep saying "output" when "measurement" is the intended meaning.
  • Negative feedback is feedback in which a measurement of a system influences that system in a way that tends to prevent or reduce variations in the measurement.
I would rather say something like "relative variations", since there are many cases where the measurement is not absolute. Trevithj (talk) 08:32, 15 September 2014 (UTC)
Measurements do not take place in the negative feedback amplifier, where a fixed fraction of the output is subtracted from the input signal, regardless of its value, but nothing is measured. Measurement does occur in error-controlled regulation, where the value of an essential variable (e.g. temperature) is measured to determine its departure from its set point to find the error. Brews ohare (talk) 11:43, 15 September 2014 (UTC)
How about using sources? Brews ohare (talk) 11:46, 15 September 2014 (UTC)
In a broader context, little of this discussion applies. Brews ohare (talk) 13:14, 15 September 2014 (UTC)
I guess my objection to "output" comes from a systems analysis point of view, where an output is a flow that crosses the system boundary. If it feeds back externally, that is outside the context of the "system" and so irrelevant to analysis. If it feeds back internally, then the fact that the flow crosses the boundary becomes irrelevant. So "output" kind of grates. Trevithj (talk) 15:17, 15 September 2014 (UTC)
Trevithj: The idea of a system involves drawing a boundary, as you point out, but the placement of the boundary is dependent upon one's purpose. Is the thermostat inside or outside, for example. If the monitor feeds information about the 'system' back to the 'system', the entire feedback 'system' lies outside the system. If the monitor feeds information about the product of the system back to the system, then the information isn't necessarily about the status of the system making the product. For example, the information may be a market analysis - customers want different features. I think this discussion is too narrowly drawn. Brews ohare (talk) 16:01, 15 September 2014 (UTC)
I'm completely with Trevithj on this. This very, very definitely is a systems topic. If you don't define 'input' and 'output' wrt a system boundary, then all inputs are from outputs and all outputs go to inputs anyway. It's inherently redundant to even use the terms 'output' and 'input'. And there's nothing that says you need to take a system output to create a feedback loop; plenty of feedback loops go from some point in the middle of a system and go back to control a system input or even go to some place else in the system that isn't even a system input, but is part of a control loop.
I mean, what are we trying to tell the reader with the use of term 'output' here? Where else would you get a signal from? By specifically mentioning it, you're implying that it's necessarily a system output, but that just isn't true.GliderMaven (talk) 00:42, 16 September 2014 (UTC)
The notion of feedback, in which a signal is "fed back" into the system needs to get it out first. There are at least two relevant systems, or subsystems here: the forward system, which produces the signal that gets fed back, and the closed-loop system that you get by feeding that signal back in where it can affect itself. So, yes, it's a system problem. The "output", or "measurement" if you prefer, is from the forward system. Feeding it back in (often via a feedback loop filter, or controller) makes the closed loop system. The feedback path is outside of the forward subsystem. I think we all understand these things, just getting hung up on words to describe it. I'm almost OK with "Negative feedback is feedback in which a measurement of a system influences that system in a way that tends to prevent or reduce variations in the measurement" except that "measurement of a system" seems awkward or not quite meaningful. Maybe "measurement on a system" or "measurement of a system's state" would be a little more clear, if not less awkward. Anyway, calling it a measurement doesn't make it different than an output; both come outside the system and then get fed back in. Dicklyon (talk) 03:08, 16 September 2014 (UTC)

An 'output' might be measurable or not, and it might measurable and measured or not. Measurement and output are not the same thing. And measurement does not occur in the negative feedback amplifier. Dick, if you think measurement occurs, what is measured and how is it measured? Brews ohare (talk) 03:33, 16 September 2014 (UTC)

How's this?

Negative feedback is an opposing response of a system or entity that is triggered by a perturbation or disturbance of that system or entity.

Camazine, p. 16

It is sourced. Brews ohare (talk) 19:27, 16 September 2014 (UTC)

The dilemma seems to be getting clearer. We want to describe negative feedback in an intuitive and tangible way, one in common use - which means the output-to-input view. We also want to cover as many examples as possible - which means the circular-causality view. Attempts to reach a compromise between the views seem mutually unsatisfying.
Maybe a historical approach would work. That is hopefully less controversial.

Negative feedback originally described the practice of using the output signal of an amplifier to partially cancel (or negate) the input signal. In general, negative feedback loops are any balancing or self-correcting circle of cause-and-effect.

Trevithj (talk) 06:00, 17 September 2014 (UTC)
Trevithj: The history of negative feedback is not as described. The use of negative feedback in steering of ships (the PID controller), an example of error-controlled regulation, predates the negative feedback amplifier, as you well know. As a reminder, see Sperry's (1911) ship-steering mechanism "that incorporated PID control and automatic gain adjustment to compensate for the disturbances when the sea conditions changed" and the 1922 presentation by Minorsky of a "clear analysis of the control involved in position control and formulation a control law that we now refer to as three-term or PID control." (from Stuart Bennett)
The 'dilemma' is not resolved by an historical approach, which involves the same issues of separation between error-control and other forms of negative feedback found on this Talk page. The failure is rather the failure to use sourced discussion, and also one of generality of definition. On one hand, the definition of Camazine, p. 16 views negative feedback as a form of response by an entity. That also is the view of the those sources that use the 'opposition to change' formulation. This general approach based upon an opposing response does not go into specifics about the implementation of the response, nor the triggering of the response, nor even the objective of the response, which detail enjoyed by those engaged here automatically leads to controversy because such details tend to overly restrict the concept and rule out some examples in favor of other examples of negative feedback. Brews ohare (talk) 14:37, 17 September 2014 (UTC)
In this regard, as you well know, your first definition of negative feedback above is too restrictive, limiting itself to amplifiers. The second form is far too loose, as the mechanism described includes many forms of response, even feed forward, that have nothing to do with negative feedback. Brews ohare (talk) 14:49, 17 September 2014 (UTC)
And, again, reliance upon sources about negative feedback would help focus discussion upon what sources say. Discussion about what published sources say is likely to be less controversial than discussion of what WP editors think. Brews ohare (talk) 15:13, 17 September 2014 (UTC)

An approach

Here is an approach:

Negative feedback is an opposing response of a system or entity that is triggered by a perturbation or disturbance. This opposing response involves a variety of different specifics as to its implementation, triggering, and objectives, all of which details vary from one instance to another. For example, ...

Camazine, p. 16, Beerel, Hobbs, Kal Morton

Brews ohare (talk) 16:29, 17 September 2014 (UTC)

The details of applications are provided in the subsections devoted to particular cases. Brews ohare (talk) 17:05, 17 September 2014 (UTC)

To clarify my earlier post, the term "negative feedback" was first used somewhere around the 1920s, by Bell technicians working with electronic amplifiers. (Eg: Harald_T._Friis and A. G. Jensen. "High Frequency Amplifiers." Bell System Technical Journal, April 1924.) It was popularized by Harold Black in the 1930s. According to Mindell, it wasn't until WWII that the term began to be used more broadly to describe control systems.
It is certainly true that control mechanisms were in use long before 1920, but as far as I can tell they weren't called "negative feedback". James Clerk Maxwell never used the term, for example. Neither does the PID control article mentioned above. Trevithj (talk) 01:13, 18 September 2014 (UTC)
So you say the 'term' was coined in 1924 but the technique was in use as described by Bennett as early as 1911, and as a Google search of "PID control"&"negative feedback" shows, these two are often discussed in conjunction. Here is a particular example. The Mindell source you probably refer to is Between Human and Machine, which agrees with Bennett on the early theoretical role of Minorsky and the early practical role of Sperry.
All of which has no bearing upon the proposal immediately above. Brews ohare (talk) 01:33, 18 September 2014 (UTC)
It has a very strong bearing on the proposal above. The last two of the five sources quoted use output-to-input terminology. The first two talk in terms of feedback loops, while the middle source and the newer example given here talk in terms of the overall "balancing" effect. All use the term "negative feedback", which was first coined circa 1920.
The proposal itself aims at a compromise definition. Reasonable, but this hasn't worked very well in the past, even when cited sources are used. (eg:) People have strong opinions on the subject.
By beginning with an acknowledgement of the output-to-input origins of the term, and then describing the more generic later usage, I hope to better satisfy both sides. Trevithj (talk) 23:12, 18 September 2014 (UTC)
Trevithj: This definition covers the whole subject, and details that vary can be supplied by examples or reference to subsections. I don't think an historical approach will work as easily. So far you haven't provided one that does. Brews ohare (talk) 23:35, 18 September 2014 (UTC)

I like Trevithj's idea, in the sense that it gets the original sense of the term in there, in terms of output being fed back to input. It can then be generalized to systems with loops, whether any point is considered an input or an output or not. Dicklyon (talk) 03:47, 19 September 2014 (UTC)

Maybe this way:

Negative feedback is any balancing or self-correcting circle of cause and effect. The term negative feedback originated in the electronics field, where it described the practice of feeding the output signal of an amplifier back into the input with a sign inversion, with the effect that the net gain of the amplifier was reduced.

"Negative feedback is any balancing or self-correcting circle of cause and effect" is inaccurate. It is too general and includes forms of self-correction beyond negative feedback. Use sources. Brews ohare (talk) 05:24, 19 September 2014 (UTC)
I'm unclear on the objection. The quoted passage seems to fit the Camazine and Beerel sources already mentioned above. Then there's Sterman (2000) "negative loops counteract and oppose change" and "processes that seek balance and equilibrium" Trevithj (talk) 10:05, 19 September 2014 (UTC)
Trevithj: If indeed this statement agreed with Camazine and Beerel, the initial proposal that does indeed fit would be preferable as it can be referred to the language of these sources. However, these authors do not suggest all forms of balancing or self-correction fall under 'negative feedback'. Brews ohare (talk) 14:05, 19 September 2014 (UTC)
To this objection might be added that such a description of the use of negative feedback in an amplifier is tortured and muddled at best. Brews ohare (talk) 14:09, 19 September 2014 (UTC)
And additionally, the article already has a substantial history subsection. Brews ohare (talk) 15:13, 20 September 2014 (UTC)

Removing 'dubious' tag

So far as I can see, no one has proposed an alternative to the definition I put in the lead on 8th September, so I am removing the 'dubious' tag that was appended immediately afterwards. If anyone thinks it should be re-applied please could they explain why here first before re-tagging it? And suggest a proposed wording, preferably using the four numbered clauses in my suggestion above (in the section 'Another try'). Thanks. DaveApter (talk) 12:31, 19 September 2014 (UTC)

It could be improved by adding a line that explains 'input/output' can be construed in the normal sense of inventory input/output or in the sense of monitoring (status output) and regulation (status adjustment via input). Brews ohare (talk) 14:24, 19 September 2014 (UTC)
And sources might help to avoid another 4-month Talk-page embroglio down the road, although there is next to no sensitivity to sources exhibited so far. Brews ohare (talk) 14:41, 19 September 2014 (UTC)
I actually think this is among the best lead proposals to date, though I have made a few edits just now that I think help clarify it a bit. The existence of equalibrium, and exponential decay toward it, is not a property that distinguishes positive from negative feedback systems: just stable from unstable ones. We don't need sources referenced in the lead, as long as the lead summarizes what's in the article and there are good sources there. So we need to review and make sure this is indeed a good summary. Maybe augment it with something that will make trevithj happier. Dicklyon (talk) 15:10, 19 September 2014 (UTC)

Thanks, @Dicklyon, and I think your tweaks to the wording are a definite improvement. DaveApter (talk) 16:32, 19 September 2014 (UTC)

The sentences claiming a distinction between positive and negative feedback based upon stability considerations are simply mistaken, ignoring entirely the Nyquist stability criterion and the concepts of gain margin and phase margin. The change in wording from 'negative feedback can lead to equilibrium' to 'negative feedback tends to promote a settling to equilibrium' is a backward step inasmuch as amplifiers have nothing to do with equilibrium. Brews ohare (talk) 15:35, 19 September 2014 (UTC)
While dubious isn't quite the right word, IMO, I must say that I do support the 'dubious' tag. The existing definition isn't a definition. Rather, it is an example, focused on the electronics/control theory POV. It is a good example, and I agree that negative feedback does occur in the case described. However, I do not agree that all cases of negative feedback involve some function of the output of a system, process, or mechanism that is fed back into the input with a polarity that tends to reduce the fluctuations in the output. For example, the following statement is also an example:
  • Negative feedback occurs when some function of the state of a system, process, or mechanism is fed to the output with a polarity that tends to reduce the fluctuations in the state.
This describes several biological examples (eg, regulating body temperature) although "polarity" is an odd choice in that context. BTW: given that at least as many (and possibly more) biology/social pages link to this article as electronic/control pages, maybe an alternative word is in order.
Also, I respectfully suggest that several alternatives have been proposed. I offer one, and as requested, I'll use the four numbered clauses:
  1. "Negative feedback when some measured quality..." (not necessarily 'the' output)
  2. "... of a process or mechanism..." ('system' has confusions around boundaries)
  3. "...is used to influence" (not necessarily 'the' input))
  4. "...in a way that maintains the measurement within a relatively narrow range"
Having said all that, re-tagging the lede would be counter-productive to the obvious good will shown here towards reaching a consensus. Please pardon the long post. I didn't have time to make it shorter. Trevithj (talk) 01:54, 20 September 2014 (UTC)
Your concerns are at least partially addressed by this previous proposal of mine: "A system with negative feedback is one in which an output (that is, some measurement of the system's state) is fed back into the system at a point where this output affects its own value in such a way as to oppose changes in the output." Maybe the use of "system" here can be cleaned up. Feel free to propose a new lead sentence or two. Please omit "within a relatively narrow range" as that's too unclear; relative to what? Within a narrower range than the system would have without feedback maybe. Dicklyon (talk) 02:10, 20 September 2014 (UTC)
I agree with your last point. I've suggested in the past using "reduces the gap between the measured value and a reference value", but didn't think that was well received. Main objection is that some cases don't have an explicit reference value. Also, as you pointed out, it is a bit of a circular definition.
As regards the rest - I feel compromise is the wrong way to go. It is a fact that negative feedback originally described using the output of a simple inline process to alter the input. It is also a fact that negative feedback loops are a more generic idea. I feel that we can say that in a succinct way, and are more likely to get consensus. So something like this...
  • Negative feedback originally described the practice of feeding some function of the output of a process or mechanism back into the process at a point where this output affects its own value in such a way as to oppose changes in the output. In subsequent usage, negative feedback loops refer to the more general idea of balancing or self-correcting circles of cause-and-effect.
...only perhaps shorter. That's more definite, and there's less compromise in the terminology.Trevithj (talk) 09:48, 20 September 2014 (UTC)

Trevithj, the idea of 'consensus' is not merely 'agreement among a majority of WP editors present on this Talk page at the moment'. To be a meaningful consensus, it should be agreement upon what sources say and it's clear presentation. We have already several well-sourced versions, for example: #An approach, #Earlier proposal and #Proposed change to lede, which you choose to ignore. Apparently you find fault with these sources because they disagree with your own ideas that are not about distinguishing negative feedback in its full generality, but instead attempt to characterize it in terms of a narrow view tied to the 'gap' of error-controlled regulation. Feedback has its own article where your views are prominent. Brews ohare (talk) 12:42, 20 September 2014 (UTC)

If you look at the various subsections of this article, particularly Negative feedback#Negative feedback amplifier and Negative feedback#Error-controlled regulation, the lede should be compatible with both discussions, and also with sources describing management, educational, and psychological uses of the term 'negative feedback'. So far compatibility with the text has been asserted, but not actively investigated. Brews ohare (talk) 12:56, 20 September 2014 (UTC)

Sourcing Dave's formulation

The formulation in question is:

Negative feedback occurs when some function of the output of a system, process, or mechanism is fed back into the input with a polarity that tends to reduce the fluctuations in the output, whether caused by changes in the input or by other disturbances.

There are some details of wording that could be adjusted. For example, the idea that the definition of 'negative feedback' involves some event or occurrence seems awkward. Maybe a few tweaks like these below would be better:

Negative feedback is the form of feedback in which some function indicative of the performance of a behavior, process, or mechanism is fed back into its operation in a manner that tends to reduce the effect of disturbances upon its output, whatever the origin of these disturbances.

In any case, in the interest of reassuring the reader that this is not a WP invention, but reflective of sources, some sources would be helpful. It seems unlikely that a source that contains this exact wording can be found. A way around this problem would be to present the exact quotes from several sources and indicate how they relate to this WP definition.

Is there any interest in sourcing this lead definition? Brews ohare (talk) 15:23, 19 September 2014 (UTC)

The canard mentioned by Dicklyon: "We don't need sources referenced in the lead, as long as the lead summarizes what's in the article and there are good sources there", however trite, is simply a way to duck this issue of a supportable lead definition. The subsections following in the text illustrate a variety of applications that differ substantially and a general definition that fits all of them is cause of all this brouhaha on the Talk page. Brews ohare (talk) 15:41, 19 September 2014 (UTC)

I've re-added the sentence about equilibrium in the lead, as I'm sure it's broadly accurate, relevant, and useful for giving readers a general sense of the concept. On a related point in the above discussion, I don't agree with the suggestion that there is an essential difference in the concept of negative feedback in the context of an amplifier than in the context of a control system or in biological homeostatic processes. I'll work this idea out in more detail (with refs) shortly. DaveApter (talk) 11:09, 23 September 2014 (UTC)
That there is no "essential difference" between negative feedback in the negative feedback amplifier and that in error-controlled regulation is, of course, not only your belief, Dave, but that of all editors present on this Talk page except for myself. Unfortunately, that consensus is at variance with 40 years of textbook analysis and has no basis in any source, as is made very clear in the subsection on this amplifier. That subsection points out that the so-called 'error signal' in this amplifier has no connection to a 'performance gap' based upon a setpoint, that no 'essential variable' is measured, no setpoint is supplied, and no 'gap' calculated. Instead, this (unrelated) 'error signal' is fixed by the three parameters β. AOL and the input signal amplitude I, and is in no way minimized or reduced during circuit operation, as is, of course, the entire object of error-controlled regulation, which is devoted to minimizing the 'gap'. Brews ohare (talk) 14:42, 23 September 2014 (UTC)
This operation means that the definition of feedback by Ramaprasad quoted here:
""Feedback is information about the gap between the actual level and the reference level of a system parameter which is used to alter the gap in some way",
does not apply to the negative feedback amplifier. Brews ohare (talk) 16:24, 23 September 2014 (UTC)
To this I add that there is no relation of any kind between the operation of the negative feedback amplifier and homeostasis, because the concept of 'equilibrium' and its maintenance is completely foreign to it. An amplifier's objective, of course, is amplification, not equilibrium. Brews ohare (talk) 14:52, 23 September 2014 (UTC)
It would be desirable that discussion of these points should involve 40 years of textbook analysis, which should not be disparaged (based upon no sources) as a mere 'shadow of what's really going on' and a discussion of 'a secondary effect', both of which claims are based upon personal prejudices and have no basis in fact. Brews ohare (talk) 15:18, 23 September 2014 (UTC)

Too much noise

Brews, on Aug. 31 you said, "OK folks. I'll leave it to you." Several times since then you've withdrawn again. Yet every time we start to discuss a proposal, and start to find some common ground, you jump in and torpedo the concept and pollute the discussion with excessive noise. You have continued to dominate the talk page, both by number of edits and amount of text, and have done nothing that moves toward convergence. I do my best to ignore you and reply to people with useful suggestions, but you distract them back into arguments. Please take a month or two off and see if things settle. Dicklyon (talk) 16:11, 19 September 2014 (UTC)

Dick, pointing out obvious errors, omissions, and contradictions with sources is 'noise' only to those who prefer their own ideas. Brews ohare (talk) 16:22, 19 September 2014 (UTC)
Speaking very generally, but also to you Dick, no-one here is interested in conveying what sources have to say, but simply restate their personal views without substantiation. You yourself, Dick, have switched back and forth from an amplifier input/output orientation to a stability and equilibrium error-based regulation standpoint without ever twigging that these two uses of feedback are fundamentally different. One bypasses fluctuations and the other combats them. A general definition has to include them both, not to mention the many other applications to things like educational and management theory that do not fit neatly into a engineering framework. Brews ohare (talk) 16:52, 19 September 2014 (UTC)
Trevith has thrown up his hands regarding sources:
"but this hasn't worked very well in the past, even when cited sources are used. (eg:) People have strong opinions on the subject."
That suggests that some discussion of sources has taken place, but discussion of these sources or any others, for that matter, has not happened, with the exception of Dave's useful reliance upon sources and your doubtful relegation, Dick, of the standard textbook treatment of the feedback amplifier to a secondary effect. 'Strong opinions' are fine if they are sourced, and if they differ, the different views can be summarized and sourced. Brews ohare (talk) 17:06, 19 September 2014 (UTC)
If you could throttle your input even just moderately, say to not more than all other contributors put together, you might have a better chance of making your point. I don't think anyone is reading your walls of rants. Dicklyon (talk) 20:30, 19 September 2014 (UTC)
Dick: My point is made: USE SOURCES. However, your point that no-one is paying attention is quite on target. Brews ohare (talk) 22:59, 19 September 2014 (UTC)
Your percentage of total text on this talk page has increased from 60% to 70% during the last three weeks, after you said you'd back off. You probably aren't using your words very effectively. Try being quiet for a while and see. There are four others here who are actually trying to converge. Dicklyon (talk) 02:17, 20 September 2014 (UTC)
I'll lay off again for a while. However, convergence is not what is happening. What is happening is the misdirected effort by each WP editor to show their own individual opinion, perhaps by virtue of its eloquence or its intuitive obviousness, should trump all published discussion that, as is asserted in various ways, supposedly is not needed anyway. Brews ohare (talk) 14:21, 20 September 2014 (UTC)

Deletion of erroneous material in lede

The sentences claiming a distinction between positive and negative feedback based upon stability considerations are simply mistaken, ignoring entirely the Nyquist stability criterion and the concepts of gain margin and phase margin. The change in wording from 'negative feedback can lead to equilibrium' to 'negative feedback tends to promote a settling to equilibrium' is a backward step inasmuch as amplifiers have nothing to do with settling to equilibrium. I have removed these errors. Brews ohare (talk) 16:34, 19 September 2014 (UTC)

I think ignoring entirely the Nyquist stability criterion in the lead is a good idea, and that "Whereas positive feedback tends to lead to instability via exponential growth or oscillation, negative feedback generally promotes stability" is a pretty fair characterization of their distinction. It doesn't mean all positive feedback systems are unstable, or that a negative feedback system can't go unstable at a frequency where its feedback loop gain is actually positive (Nyquist's point). I care less about the equilibrium part; I was trying to make it more meaningful, but leaving it out of the lead is OK, too. What do others think? Dicklyon (talk) 20:27, 19 September 2014 (UTC)
It possibly confounds the lede to introduce stability. Again, it means different things in different disciplines, especially related to oscillation. Some disciplines regard any oscillation as a sign of instability. Others are only concerned about stability of amplitude or frequency. Best to avoid the subject in the lede, IMO. Trevithj (talk) 01:04, 20 September 2014 (UTC)
OK by me. I just edited what was there to make it more sensible. Dicklyon (talk) 02:11, 20 September 2014 (UTC)

Unfortunate removal of subsection

Dicklyon first, and then Blackburne, have deleted the following addition to the article without discussion on this talk page, certain that their actions require no follow-up:

Psychology. management, and education

In psychology, the term 'negative feedback' is commonly used to label rejection or negative reaction to a subject's response. There is a connection here to the systems view of negative feedback in that such criticism is intended to limit a subject's response and bring it more in line with some objective. However, the term sometimes is used in a variety of organizational settings in a manner more closely related to its use in engineering, such as leadership and management, and education.

Sources
"One type of instruction that may markedly distort individual responses is criticism or negative feedback...The interrogator may make the adapt himself to the expectations reflected in the interrogator's manner and style of questioning." Gisli H. Gudjonsson, p. 347
"When systems try to change, certain factors support that change (positive feedback) and certain factors counter that change (negative feedback)." Annabel Beerel, p. 52
"egative or corrective feedback is in some instances a necessary condition for second language acquisition." Chun Lai, p. 338

Comments

  • In these fields, 'negative feedback' sometimes is used informally to mean "a critical, derogatory, or otherwise negative response", but it also is used more carefully to describe a form of guidance towards an end, much like error-controlled regulation guides a variable to its setpoint. In defense of this addition, no claim is made that it is the best or most complete treatment of this aspect of negative feedback, nor that these sources are the most authoritative or lucid expressions of the role of negative feedback in psychology, management, or education respectively. However the claim is made that these fields do make use of the term 'negative feedback' and frequently cite engineering or cybernetics as its origin in what is unquestionably the topic of this article.
These published views thereby contradict decisively the in-line comment of Blackburne that this contribution "has nothing to do with the topic of this article". Likewise, it seems clear that Dicklyon's in-line comment that this "section exhibits very poor synthesis, including complete misinterpretation of the last source", if merited, should lead to constructive comments about how this addition could be rewritten or better sourced, and should not lead to its summary removal disdaining further comment. Brews ohare (talk) 04:51, 21 September 2014 (UTC)
My view is that it was entirely appropriate to remove that section. It is nothing to do with the topic of this page and it confuses the issue. @Brews, if you really think that the usage in that sense merits inclusion in Misplaced Pages, then I suggest you create a separate article an put up a disambiguation page. DaveApter (talk) 12:25, 22 September 2014 (UTC)
Dave, I wonder why you think this use of "negative feedback" has nothing to do with article of that name? Two of the cited sources think there is a connection to the systems definition. I believe WP:UNDUE is satisfied by this brief subsection. Brews ohare (talk) 15:16, 22 September 2014 (UTC)
For example Annabel Beerel, p. 52 has an entire section on p. 52 connecting the use of feedback for learning to "system dynamics", apparently drawing upon John D Sterman (2000). Business Dynamics: Systems Thinking and Modeling for a Complex World., a source I do not have access to.
On the other hand, Chun Lai, p. 337 in his article The role of feedback and technology in SLA means by 'negative feedback' the pointing out of student errors in order to alert them to their mistakes and correct them. Personally I don't think that is very different from detecting a 'gap' as a departure of system behavior from its setpoint and using that to correct the system behavior. It might be of interest to note that this concept of negative feedback in learning also is applied in engineering to 'teach' neural networks (also see Bak & Chialvo), with wider applications.
I'll look for more sources, if you think that necessary. Brews ohare (talk) 15:32, 22 September 2014 (UTC)
Your personal odd interpretation of Lai is part of what my edit summary was complaining about. That was clear, was it not? Dicklyon (talk) 17:49, 22 September 2014 (UTC)
Dick: Your opinion is clear, but not your reasoning. Brews ohare (talk) 18:24, 22 September 2014 (UTC)
My reasoning is this: when you say stupid things, it is impossible to talk you out of your odd way of thinking, so I give up. Dicklyon (talk) 19:11, 22 September 2014 (UTC)
Dick: It seems you are too angry to try to make sense. Brews ohare (talk) 19:24, 22 September 2014 (UTC)
No, I'm perfectly fine with how things are going. Dicklyon (talk) 20:02, 22 September 2014 (UTC)
Goodbye, Dick. Brews ohare (talk) 20:11, 22 September 2014 (UTC)
Yes, the form of feedback and the structure of the loop is the same as with negative feedback. But it still isn't negative feedback!
In fact, whole societies were attempted to be constructed based on the idea that this is negative feedback, and they always failed! That's how the hippy community tried to work. They all disbanded, because repeated criticism is, essentially, bullying somebody until they do what you want.
The problem with it isn't the shape of the feedback loop, or the nature of the signal, it's the nature of how the feedback is processed by the humans receiving what is, in the end, criticism.
It turns out that in many cases, humans receiving criticism, overall, get worse at their jobs. That's the opposite of negative feedback in machinery.
And this is not unusual. Management theory is, in many cases, cargo cult science. It takes the forms of science, but very often hasn't been properly tested and very often doesn't work.
Source: All Watched Over by Machines of Loving Grace (TV series)#The_Use_and_Abuse_of_Vegetational_Concepts (the real documentary not the article).GliderMaven (talk) 20:13, 22 September 2014 (UTC)
GliderMaven: You raise the point that although formally a system may follow what looks like negative feedback in determining a performance gap and have in place a system that in theory should reduce the gap, the system may not succeed. I m sure that does happen sometimes. This discussion should appear in this subsection. Brews ohare (talk) 21:32, 22 September 2014 (UTC)
But I'm not convinced it should even be in the article. Most of the references to 'negative feedback' in management theory I've found are solely about criticism, versus 'positive feedback' is about complimenting people. Whereas, in the context of this article, negative feedback should be a mixture of criticism and compliments- in other words they are using 'negative feedback' in a different sense than we are taking it in the rest of the article.
So I would be much happier with a 'for...' at the top to deal with this.GliderMaven (talk) 01:00, 23 September 2014 (UTC)
GliderMaven: Perhaps many times negative feedback is informally used, but this reversion is unwarranted. Brews ohare (talk) 01:06, 23 September 2014 (UTC)

Please cool it

@Brews, I am sure you mean well, and some of your suggestions definitely are useful, but your style is combative rather than collaborative and the sheer volume of your input is overwhelming and counter-productive. You have made about 1500 edits running to thousands of words in the last few months, almost all of them here or in related discussions. And to be frank, some of your ideas are misconceived and some of them are downright eccentric.

When I suggested recently that you step back for a bit your feelings were hurt (and I'm genuinely sorry for that, I had no intention of being offensive), but you said you would give it a break; this lasted a few days and then you were back, and with the same patterns of behaviour. Please, please, leave this topic alone for a few weeks. Maybe edit a bunch of completely different subjects, or just get some exercise. But try to cultivate a less combative style and a more succinct expression when you do. DaveApter (talk) 10:57, 23 September 2014 (UTC)

Dave: In almost all editors' contributions, your own being an exception, sources do not come up, and the introduction of them is considered offensive. For example, the actions above by Dicklyon & GliderMaven, and I regret, yourself, indicate an amazing disparagement of the widely published arenas of social dynamics and complex adaptive systems, which includes the application of feedback in education, management, and psychology, described above by GliderMaven as cargo cult science, and according to Dicklyon my desire to mention it is 'a stupid thing'. The sources cited, including even Wiener's cybernetics, and scientific application to neural networks, are not paid any attention. You have suggested that the mere mention of this work is so far off-topic (despite published opinion to the contrary) that it should be made a separate article and omitted here entirely. This shared hostility to a recognized field is contrary to WP policy and common sense.
I have raised the issue of using sources repeatedly here, and so far the only discussion of sources beyond my own is to ignore or to disparage them, and direct attention elsewhere, mostly as complaints that I am interfering with consensus. Whatever consensus that might be, it has nothing to do with sources.
This intolerance of sources and obsession with personal beliefs is inadvisable. Brews ohare (talk) 12:56, 23 September 2014 (UTC)
This kind of continuing misrepresentation of my attitudes and positions is one reason I am unable to participate productively in discussions with you, Brews. Anyone who knows my editing style knows that I very much seek out, value, and respect good references. You just have no idea what is going on with you. Dicklyon (talk) 17:31, 23 September 2014 (UTC)

Misrepresentation?

While I agree that negative feedback is important both in systems with identifiable setpoints and errors and in systems without, I do not agree that the kind of parameter insensitivity that those guys talk about is the first-order way to look at negative feedback; it's a secondary effect, and applies to both types system with and without setpoints and errors. You have jumped to an odd interpretation of those sources that makes it impossible for you to converge with others. And I though you said you were going to go away and let it get worked out... Dicklyon (talk) 05:25, 2 September 2014 (UTC)
My reasoning is this: when you say stupid things, it is impossible to talk you out of your odd way of thinking, so I give up. Dicklyon (talk) 19:11, 22 September 2014 (UTC)

Anything else? Brews ohare (talk) 17:53, 23 September 2014 (UTC)

Of course, this point, obvious to me, is no less obvious than the impossibility of conveying it to those assembled. I understand your difficulty to some degree, as my thought originally was like yours that error-controlled regulation was the paradigm. That meant that the negative feedback amplifier was outside the paradigm unless the concept was widened. Unfortunately, instead of that widening, you all have decided that willy-nilly the negative feedback amplifier will be forced into the Procrustean bed of error-correction, even if no gospel can be found advocating this faith.
That is why I changed my attention to other fields of scholarship that employ the term 'negative feedback'. This topic is refused as well, not on the basis of sources, but as before, based upon the personal belief of those assembled. The belief in this case is that these other uses of the term 'negative feedback' are either like cargo cult science, or far off-topic using 'negative feedback' only in the loose everyday sense of unwanted rejection. That belief is erroneous, and readily disproven by reading the sources instead of assuming what they say. Brews ohare (talk) 21:02, 23 September 2014 (UTC)
I have never taken the position you ascribe to all of us here. I disagree with both Trevithj and GliderMaven on various points. Yet we have several times been close to converging on a compromise, until you torpedoed it. Go away, as you said you would, and maybe we'll work it out. Dicklyon (talk) 02:16, 24 September 2014 (UTC)

Back to amplifiers

I have moved contributions by Trevith j above that were injected into the above thread that concern the subject of the negative feedback amplifier.

Please don't be concerned about giving offense, Trevith. I do agree that when the textbook analysis does not mention error correction in discussing the negative feedback amplifier, that is not to say that it is absent. However, as their discussion does not use this idea, we can definitely say that their discussion of negative feedback does not need this idea. Accordingly, if the negative feedback amplifier in fact does incorporate some form of 'gap' that is reduced, this point is not part of the standard discussion. We therefore need some basis for bringing it up in this context, which obviously is not supplied by stating that it is the consensus of editors here present, sans sources. There is no logical fallacy in this argument that I can see. Brews ohare (talk) 21:02, 23 September 2014 (UTC)
I believe you can't see it. But it is there. "Textbook analysis" either refers to a standard and widely-held analysis of the subject, or it refers to your analysis of two textbooks. Then there is that false premise: "does not mention error correction". Actually, Bhattacharya devotes much of the introductory chapter to exactly that. Trevithj (talk) 02:18, 24 September 2014 (UTC)
Trevith, Amplifiers are not control systems, but Battacharya's book is about control systems. Its relevance to the negative feedback amplifier is found in his discussion of the effect of feedback upon sensitivity, which is for our purposes indistinguishable from that of Kal, who addresses directly the negative feedback amplifier. So, for example, Bhattacharya's introductory chapter is about control systems, not the negative feedback amplifier, and he describes things like comparators and set points typical of control systems that have nothing to do with the negative feedback amplifier, and refers to comparators, set points, and errors that are found only in control systems.
Naturally the way to get to the bottom of these things is to examine the sources carefully and identify exactly what portions we might interpret differently. The sensitivity analysis of both Bhattacharya and Kal are basically identical and do not require any of these ideas that describe control systems. Both are based upon the figure in the lede (mathematically identical to this figure in the feedback amplifier section), while the figure for a system involving set points, 'gaps' and regulators is like the figure found in the error-control section. If you wish to assert that the negative feedback amplifier, whatever one cares to say about its sensitivity, is at bottom a control system like that of this figure, then you will have to support that assertion with a source. No source making such an assertion will be found. Brews ohare (talk) 04:29, 24 September 2014 (UTC)
Negative feedback amplifier using ideal unilateral elements. The feedback is negative if βAOL> 0.
What do you regard as evidence that you are wrong? Bhattacharya includes the negative feedback amplifier in a book about control systems. Trevithj (talk) 04:53, 24 September 2014 (UTC)
Argument from authority is the only VALID form for Brews Ohare; even though argument from authority is a well-known fallacy. Mere logic based on facts need not apply. I'm with the WP:SHUN camp. If Brews can't sort his head out, we certainly can't.GliderMaven (talk) 13:27, 24 September 2014 (UTC)

My argument is that the figure at the right is used by both Kal and Battacharya to establish the traditional argument as to why feedback renders this circuit insensitive to variations in the open loop gain AOL, and this argument does not make use of any ideas from control theory about measurement of the value of an essential variable, measurement of a 'gap', use of a comparator, or regulator, or minimizing a 'gap'. So, I believe, you two have at least two possible positions to take, namely that this block diagram does not represent a negative feedback amplifier, or the traditional analysis of this circuit is mistaken. You have both made version of these arguments in the past, and you both have not provided any source that suggests either of these possibilities is held to be valid. Showing comendable creativity, you two have now converged upon a different kind of objection:1,2:

"So let me get this straight, you are claiming that a negative feedback amplifier is not an example of an error-controlled regulator, and you are basing this on something that a textbook did not say, and then writing it into Misplaced Pages as true?."GliderMaven (talk) 17:33, 30 August 2014 (UTC)

I believe that the logic of this argument as a rebuttal to my own is of the same ilk as saying just because a theorem about the angles of a triangle doesn't involve pentagrams, doesn't mean that pentagrams are unimportant to all theorems about triangles. Now that is a true statement, but it doesn't dispute the statement that the theorem in question has no need of pentagrams. It does, however, leave open the possibility that some theorems about triangles employ pentagrams.

With this context in mind, it is your responsibility to show that some aspects of feedback in the negative feedback amplifier cannot be demonstrated using this figure, and the reality of the negative feedback amplifier transcends this figure. (Dicklyon also supports this view, also without sources to date.) If you can do that, it then remains to flesh out whatever these newly discovered sources have to offer about feedback that is missing from that based upon the traditional argument. Until you succeed, you are in the wilderness of personal assertions and beliefs that, so far as is known here, have no basis in published work.

I add that, whatever new may be discovered about feedback in the negative feedback amplifier during your researches, it remains the case that the traditional argument succeeds in demonstrating feedback provides insensitivity without any need for these newly discovered features, whatever the nature of the amazing new features you may find for it.

Do you agree? Brews ohare (talk) 15:29, 24 September 2014 (UTC)

You seem to be repeatedly arguing that negative feedback amplifiers aren't feedback loops and don't use negative feedback.GliderMaven (talk) 15:43, 24 September 2014 (UTC)
"It is your responsibility": No, WP does not work like that. It works by consensus. Consensus is against you, has been against you again and again. Repeatedly ignoring it is simply disruptive, and was a long time ago. Just stop.--JohnBlackburnedeeds 16:26, 24 September 2014 (UTC)
GliderMaven: I have not argued any point of view other than that expressed by the traditional argument. I have suggested that if you and Trevithj wish to supplement that argument in some way, you should source it and not rely upon your own opinions. Brews ohare (talk) 17:20, 24 September 2014 (UTC)
Blackburne: Reporting what sources say is indeed a responsibility of WP editors, although you may not think so by looking at this talk page. Consensus among WP editors concerning their personal opinions about content is meaningless without the support of sources, unless that consensus is about how sourced material is to be organized and summarized; matters of presentation are entirely within editor discretion and subject to their consensus. This sensible use of consensus is not what is seen at the moment. Brews ohare (talk) 17:20, 24 September 2014 (UTC)

I have removed some of my comments, since I object to them being placed in a context other than the original one. Trevithj (talk) 19:14, 24 September 2014 (UTC)

Regards the rest of this section: I agree with the other editors that the burden of proof falls on Brews to show that the negative feedback amplifier is not a control system, when both the primary authors he has cited clearly treat it as though it is a control system. If the conclusion is true, this shouldn't be much of a burden. Trevithj (talk) 19:28, 24 September 2014 (UTC)
Tevithj: The diagram used in the traditional argument is not that of a control system. Should it devolve that you consider that it is the diagram of a control system, that implies that some control systems do not use a 'gap', a set point, and neither do they attempt to regulate by minimizing a 'gap'. On the other hand, there are control systems that do all those things, and their analysis is based upon a different block diagram that incorporates those features.
We are now back at square one, with the words "control system" expanded to include the negative feedback amplifier, but with this new member of the set "control system" exhibiting insensitivity according to an analysis without employing a 'gap', a set point, or regulation by minimizing an error. So where does that leave you? Nowhere. The traditional argument stands and does not use any of these features of feedback you wish to invoke. Brews ohare (talk) 20:15, 24 September 2014 (UTC)
No, we're back further than that. We're already reverting most of your edits, but I think we're pretty much at the stage where we'll have to go through the entire history of the feedback articles and remove all of your edits, one by one, or at least look at them. We just can't trust somebody in this much denial about simple, basic facts about the topic.GliderMaven (talk) 20:50, 24 September 2014 (UTC)
What do you regard as evidence that you are wrong? We are back to square one because of the basic form of your arguments. The diagram referred to is clearly that of a control system. Trevithj (talk) 21:07, 24 September 2014 (UTC)
Your link to the table of contents of Battacharaya's contribution to CONTROL SYSTEMS, ROBOTICS, AND AUTOMATION is insufficiently detailed to establish anything. If you look at his Figure 2: A feedback control system, you will notice 'reference inputs' that provide set points, 'disturbances' that represent upsets from status quo, and a controller that compares the monitored plant variables with the reference inputs and feeds back a regulatory signal to the plant. On the other hand, his Figure 4: A feedback system depicts a feedback loop, which is only a simplified part of a control per se, the same loop he uses in the traditional argument. There is nothing there to suggest he thinks this represents a complete control system.
None of this matters of course, Whether you call the negative feedback amplifier a 'control system' or not, the traditional argument is based upon this figure and that analysis does not depend in any way upon a set point, a measured 'essential variable', or regulation by minimizing a 'gap'. Don't you agree with that much? The analysis of the figure makes no use of the elements found in Figure 2 and missing in Figure 4.
I fail to see any 'circular reasoning' here, or in the presentations of Kal or Bhattacharya, or in the summary of their argument. If there is circularity, please identify it rather than refer to what is, for me, a chimera. Brews ohare (talk) 23:36, 24 September 2014 (UTC)

The traditional argument based upon the block diagram of the lead remains an accurate analysis of the mathematical implications of this figure, which makes no use of a 'performance gap' or regulation by reduction of such a 'gap'. So far editors on this page refuse to acknowledge the connection between the diagram and the math associated with it described in the traditional argument, preferring to drag up red herrings to avoid direct discussion of these sources.

By the 'traditional argument' linked here, in case there is any confusion, is meant simply the unadorned summary of the Kal - Bhattacharya analysis, see Kal and Bhattacharya. The original sources are linked, so anyone dissatisfied with this summary can read the original arguments for themselves. Brews ohare (talk) 12:45, 25 September 2014 (UTC)

Discussion between Dicklyon and brews_ohare

I don't think anyone has expressed any doubt about the advantages of feedback as expressed by those two refs. Why do you keep bringing them up as if someone was disagreeing with them? Dicklyon (talk h) 04:06, 26 September 2014 (UTC)
A regulator R adjusts the input to a system T so the monitored essential variables E are held to set-point values S that result in the desired system output despite disturbances D.
Dick: I take it that you would agree that this summary is an accurate description of the textbook approach to the analysis of amplifier desensitivity. If so, you and I are pretty much on the same page about the textbook approach, but you seem to entertain the idea that any real-world amplifier, and not some idealization like that of the figure, enjoys aspects of feedback not described by this analysis, that are in fact more significant than the factors involved in this insensitivity analysis. GliderMaven and also Trevithj go even further to suggest (but do not attempt to substantiate) that this summary is a defective presentation on my part that doesn't capture the work of Kal and Bhattacharya at all, and according to Trevithj is in essence a ploy on my part to gain acceptance of my own weird ideas. Where do you actually stand, Dick? Brews ohare (talk) 14:45, 26 September 2014 (UTC)
For example, we have this remark by GliderMaven that this summary is about a stability point. Further, both editors hold the view 1 2 that the fact that this summary doesn't need any ideas like a 'performance gap', a set point, or regulation to minimize a 'gap' doesn't mean that these ideas are not pertinent to this analysis, even though not employed by it. These two also seem to think that the standard diagram does not uniquely determine the Kal-Battacharya analysis based upon it, but that the Kal-Battacharya analysis is related to some emasculated view of this diagram that ignores its 'deeper' meaning that is buried in a realistic understanding of what these blocks actually contain and stand for, a point of view they have yet to substantiate. Brews ohare (talk) 15:02, 26 September 2014 (UTC)
In essence, they believe that a more refined understanding of the feedback amplifier diagram would demonstrate its equivalence to some version of the diagram for an error-controlled regulator, perhaps the one shown with this comment. For example, Trevithj claims, his second point, that he can see no essential difference between them. Brews ohare (talk) 15:10, 26 September 2014 (UTC)
Among the differences between these diagrams that makes it impossible to reconcile them is the fact that the error-controlled regulator deals with external disturbances D, while the negative feedback analysis deals only with internal disturbances generated within the open-loop amplifier, that the error-controlled regulator uses an externally supplied set point, not found in the negative feedback amplifier, and that the error-controlled regulator employs a measurement of some internal variable E and a comparator that determines a 'performance gap' or error that is used to calculate the feedback response intended to drive the error to zero, none of which elements appear in the negative feedback amplifier. Brews ohare (talk) 16:49, 26 September 2014 (UTC)
To state my own position about these matters in terms of these diagrams, I'd agree that the desensitivity analysis is not everything that is going on, that some more complicated diagram would be more accurate and could account for other kinds of disturbances. But what the idealized Kal-Battacharya analysis does show is that immunity to internal disturbances can be achieved using feedback without the apparatus of error-controlled regulation, by simpler means. That means the definition of negative feedback used in this article should be extended to go beyond any limitation to error-controlled regulation, because the negative feedback amplifier demonstrates that alternatives are available. Brews ohare (talk) 17:42, 26 September 2014 (UTC)
My agreement with the sources, and maybe even with your analysis of the "primary advantage of negative feedback", should not be taken as an indication that we "are on the same page" about this. Go back a few pages in those sources and you'll find out what feedback is; advantages, even primary ones, are secondary to the fact that negative feedback reduces gain, or output fluctuation per input fluctuation. Dicklyon (talk) 18:05, 26 September 2014 (UTC)
Dick: I assume your reference to gain reduction is the reduction from AOL to 1/β. Of course, that means that any input variation δI produces the output variation δI /β. But of course also the input signal I also produces an output reduced to I /β, so the noise is the same percentage as before. The same result is achieved by simply dropping the feedback and using an open-loop amplifier with a reduced gain to start with. Maybe you can clarify? Brews ohare (talk) 18:46, 26 September 2014 (UTC)
Kal, p. 194 makes the same point. Brews ohare (talk) 18:58, 26 September 2014 (UTC)
BTW, this is not "my analysis". It is the desensitivity analysis of the negative feediback amplifier found in every electronics text of the last 40 years. Here are some of them: Kandaswamy, Chen Choudhury. The most famous of them, Sedra & Smith, § 8.2.1 Gain Desensitivity, p.795 and Gray & Meyer, §8.2 Gain Sensitivity are not on-line accessible. There seems to be a bootleg copy of S&S here, and possibly of G&M here. Brews ohare (talk) 20:53, 26 September 2014 (UTC) Brews ohare (talk) 19:08, 26 September 2014 (UTC)
That's right. Dicklyon (talk) 23:50, 26 September 2014 (UTC)

Cute respose, Dick, but unhelpful. So now tell me why a lower gain is an advantage for input noise when it doesn't improve the signal-to-noise ratio? Please. Brews ohare (talk) 00:47, 27 September 2014 (UTC)

In response to the question in your edit summary, yes, it is my considered opinion that if would withdraw we could work this out. But if you keep trying to put words into my mouth, I'm not going to engage. With respect to your query re advantages and SNR, I didn't bring those up; why did you? It just shows your inability to listen to what is actually being said. Dicklyon (talk) 01:12, 27 September 2014 (UTC)
Dick: You said:
"advantages, even primary ones, are secondary to the fact that negative feedback reduces gain, or output fluctuation per input fluctuation"
Now I interpreted this as saying the output fluctuation is reduced if the gain is reduced compared to what it would be if the gain weren't reduced. That seems like the only thing you could mean. However, because the signal is also decreased the S/N ratio is not improved. My observation, as you say, not yours. So the query is, what is the advantage of reducing the output fluctuation if the output signal is reduced proportionally? I see no inability to listen on my part. I simply saw that this conclusion of mine would render your comment meaningless, and so I asked for an interpretation that would provide some sense. Brews ohare (talk) 01:41, 27 September 2014 (UTC)
Perhaps you were not referring to input noise, but to internal noise? That is what the insensitivity analysis is about, and reduction of internal noise according to the standard analysis has to do with βA >> 1. If this is your meaning, the standard analysis covers the situation and you are not identifying new factors, but those already handled in the standard analysis. Brews ohare (talk) 01:45, 27 September 2014 (UTC)
I didn't bring up noise. I thought the point of my comment was clear, not meaningless: the primary effect of negative feedback is to reduce the gain. That's what I meant when I said "advantages, even primary ones, are secondary to the fact that negative feedback reduces gain". I clarified gain as "output fluctuation per input fluctuation", which as you noted doesn't depend on any distinction of signal from noise. I complained that in trying to say what feedback is, you jumped over that to advantages, focusing on pages in sources that are several pages past where feedback and its primary effects are explained. That's what I meant by "advantages, even primary ones, are secondary ...". I don't know a way to express myself more clearly, or get my content to break through your filter that interprets everything through your idiosyncratic lens. And that's why I usually don't try. It's a waste of time. Dicklyon (talk) 02:22, 27 September 2014 (UTC)
OK, so I guess you are saying noise (and fluctuations) are not the point; the point is that a lower gain is desirable in itself. One might ask what feedback has to do with that, as low gain is easy to find by many methods. So, I'll surmise further. One situation where high gain is a problem is when the output signal, due to high gain, is so large that it drives the amplifier into saturation. Now, Dick, you haven't brought this up either. So as part of this guessing game about what exactly is on your mind, is this it? And if it isn't amplifier saturation, and it isn't noise, perhaps you could actually come right out and say why low gain is an advantage and why achieving low gain by using feedback is the method of choice? Brews ohare (talk) 03:14, 27 September 2014 (UTC)
Listen to yourself! Putting words in my mouth again. I didn't express any opinion about what's desirable, or why feedback might be chosen. Nothing. Nada. Dicklyon (talk) 03:21, 27 September 2014 (UTC)
Well, Dick you say you have nothing to say and that is what I have got from these attempts to draw out what you want to contribute to this discussion. It's pretty clear that I want to understand your views, and you don't really want to explain them. You blame my lack of listening, but that seems to be an excuse to avoid being pinned down to anything specific about the role of feedback in the negative feedback amplifier. Brews ohare (talk) 03:36, 27 September 2014 (UTC)
I'm sorry you refuse to hear what I have to say, which is that we need to start with what negative feedback is, before we get to discussing its potential advantages. Your confusion on that point seems to be driving your walls of text on this page. I do not disagree with you or with the sources about what the advantages typically are (in amplifiers at least). Dicklyon (talk) 03:41, 27 September 2014 (UTC)

I thought the subject was the idea of GliderMaven and Trevithj that the negative feedback amplifier is not fundamentally different from an error-controlled regulator. In my opinion, the implementation of negative feedback in the circuit for the negative feedback amplifier and the analysis of that circuit is completely different from its implementation in the circuit for the error-controlled regulator and the analysis of that circuit. One is not a version, or simplification, or whatever of the other. The circuits for these two types of circuit are known, and it really is an issue of how the circuits and their analyses relate to one another. Have you anything to say about this subject? Brews ohare (talk) 04:03, 27 September 2014 (UTC)

The underlying problem is that if the two circuits are fundamentally the same, then any definition of negative feedback that fits the error-controlled regulator applies to both. If they are different, the definition has to be broader in order to include them both. Brews ohare (talk) 04:07, 27 September 2014 (UTC)

I think I don't really know what their sticking points are because whenever I get something started with them you jump in and squash it. I do agree that they are "fundamentally" the same, but that doesn't mean it's easy to find a definition that applies equally well to both. Dicklyon (talk) 04:15, 27 September 2014 (UTC)
One doesn't have to 'opine' that the two circuits are 'fundamentally the same'. One has to compare the analyses for the two only to find they use different concepts. Or one can look at the two circuits and see they involve different inputs and different components and incompatible topologies. If instead one relies upon intuition or vague verbalizations, one can go round and round as has been happening. Brews ohare (talk) 04:59, 27 September 2014 (UTC)
Typically both amplifiers and control systems are analyzed as linear systems (at least approximately), with transfer function equal to the forward gain divided by one minus the loop gain. Slightly different arrangements of components around the loop, or labels on components, or what path you take to be "forward" don't make a fundamental difference. This is the sense in which (linear) feedback systems are all "fundamentally the same". I'm not denying that the differences in viewpoints are also important, which is why it's not so easy to find a definition that applies well to both. In particular I agree that a definition involving a "gap" or a "reference" or "target" is probably too control specific, even though it is easy enough to augment an amplifier diagram to treat input as target (appropriately scaled), or to add a zero target and treat the input as a disturbance. It's not a matter of fundamentals, but of finding a best way to treat the multiple viewpoints. For nonlinear systems, it's more complicated. Log amplifiers (op amps with diode in feedback, for example) have the feedback path more nonlinear than the forward path, unlike your "standard" amplifier analysis; nonlinear control systems do all kinds of things. In many cases, however, they still get analyzed as linearized about an operating point, so that their dynamics can be described conventionally at least for small deviations. Maybe we need to start by describing what negative feedback is just in respect to a loop, with no inputs, outputs, targets, gaps, etc., and then explain separately how that applies to the two main cases. But we've also had some simple definitions that might be OK to cover both, I think, except that we didn't far enough to see what others thought because you already shot them down with a wall of text before anyone had a chance. Dicklyon (talk) 16:13, 27 September 2014 (UTC)
Happy to see signs of discussion from you Dick. The simple definitions are OK, and can be specialized to illustrate application to error control and to the negative feedback amplifier. Already tried in an earlier attempt proposing the 'opposition-to-change' formulation widely used in the literature and examples using error control and the negative feedback amplifier. Maybe GliderMaven and Trevithj are ready to consider this now? I'll watch developments. Brews ohare (talk) 19:08, 27 September 2014 (UTC)
Or you could go back and support one proposed by someone else. Dicklyon (talk) 00:35, 28 September 2014 (UTC)
Remember this?. Brews ohare (talk) 03:03, 28 September 2014 (UTC)

New examples section

I've added a new section at the top with some brief descriptions of various examples of negative feedback in practice. I've call this new section 'Examples', and re-named the one which was previously called that. I feel this will give the reader a quick general idea of the concept before moving on to the more detailed exposition. DaveApter (talk) 14:52, 30 September 2014 (UTC)

Those are nice concise understandable examples, spanning considerable ground. Thanks. Dicklyon (talk) 17:45, 30 September 2014 (UTC)
I like the text, and the examples you've chosen, but I have some reservations about the structure. You've created a section called 'examples' but there's already a section called 'Some specific implementations' which is an examples section in all but name.GliderMaven (talk) 19:32, 30 September 2014 (UTC)
These examples are good - they avoid details of implementation, and simply talk about the effect. Nice. While I have some reservations, they aren't about the specific examples. My concern is that the focus of the examples is on the term "negative feedback", rather than the concept.
Thinking about the examples with the overall concept of negative feedback in mind, they do seem to have a strong emphasis on electrical engineering. Understandable given the origin of the term, but the concept is of course much older: James_Clerk_Maxwell's work on governors (1868); Adam Smith's "invisible hand" (1776); Cornelis_Drebbel's thermostat (1600). We could go right back to the Golden_Rule (1700 BCE)!
I believe there is a basic confusion in this subject between term and concept. It may make sense to treat this page as term-centric, and focus on electrical engineering. But if not, the examples should perhaps be given a historical sequence, and have more non-electronic examples. Trevithj (talk) 19:56, 30 September 2014 (UTC)
Never, never, never base an encyclopedia article on terms.GliderMaven (talk) 21:10, 30 September 2014 (UTC)

I did think I was illustrating the concept by means of these, broadly drawn, examples; that was the intention at any rate. The objective is to get the reader rapidly to a general intuitive understanding by these examples in various different fields, so that the rest of the article becomes accessible to them. No problem about adding more, or changing the order. I put the electronic ones first because that's the area where the wording was first commonly used (or so I think), or maybe it's just because that's the area I'm personally most acquainted with! DaveApter (talk) 21:44, 30 September 2014 (UTC)

Sure, electronics was my introduction to the concept too. The wording/term "negative feedback" was popularized by Black, according to several sources. (There doesn't seem any usage prior to 1920.) I suggest the historical sequence, simply because it is (hopefully) a less controversial framework. And as per GliderMaven's hint, it avoids putting focus on the term.
Mind you, we don't want to clutter the section with lots of examples either: perhaps 'classical' examples by key authors? Trevithj (talk) 22:39, 30 September 2014 (UTC)
Okay, I have added a few mechanical examples, and reordered in (rough) historical sequence. I've tried to keep the wording similar to the existing examples. Trevithj (talk) 01:09, 1 October 2014 (UTC)
Perhaps I wasn't totally clear. Right now the article actually has two different examples sections. IMO this is a fairly horrible idea. The reason it probably seems to work is because the history section is so bad. If the history section was better, the examples would be the historical ones and the reader would be more naturally drawn into the topic. At least, that's what I think.GliderMaven (talk) 01:51, 1 October 2014 (UTC)
Okay, historical examples are kind of where I was headed with these. History does give a natural progression to the development of the concept, so should we look at merging examples with history?
BTW, GliderMaven - not sure if you reverted the right version. I have undone change because there were a group of other edits involved. I am not sold on the content though, so please critique. Trevithj (talk) 06:04, 1 October 2014 (UTC)

OK, I think the article is much improved - now someone who is new to the subject can get a pretty good idea of what it's about by the time they've read the lead and the first section, and they can go further in to build more detail. I'm somewhat with GliderMaven's reservations about the fact that we now have two "examples" sections. The old one (now re-named "Some specific implementations" - and I'm open to other suggestions for the title) was too detailed to serve to give a quick general understanding, but it has plenty of useful information. Any suggestions for a more comfortable structure? DaveApter (talk) 17:42, 1 October 2014 (UTC)

The Reinforcement page takes an interesting approach to dealing with confusion of terms. Perhaps we could follow suit: rename Examples to Brief History, and move the History section to a sub-heading History of the term, where origin/mis-use can be directly discussed. Trevithj (talk) 22:55, 2 October 2014 (UTC)
It would be easy to recast the Examples section as Historical examples by adding a date for the introduction of each example. History could be renamed Background. Brews ohare (talk) 04:30, 3 October 2014 (UTC)

Mindell's "confusion"

We have in the text, "Nyquist and Bode built on Black’s work to develop a theory of amplifier stability, but chose to define "negative" as applying to the polarity of the loop (rather than the effect on the gain), which gave rise to some confusion over basic definitions." But I don't understand this. Aren't these essentially equivalent? Does anyone other than Mindell find a confusion or difference between the definitions that these guys used? If this concept is unique to Mindell, perhaps we should dismiss it? Dicklyon (talk) 06:09, 1 October 2014 (UTC)

I think the definitions are operating in the same direction in that particular case. The value being controlled is assumed to be higher than the desired value, so control is always reducing the value. That's a bit like steering a car around a circular track: to stay in the lane, you only have to turn one way. But Black's version doesn't generalize to the case where the actual value is lower than the desired value. Likewise, you can't control a car on a straight road by only turning one way.
Ramaprasad mentions at least one other author who uses a definition like Black's. Berrien (1968,1976) apparently talks about thermostats as using positive feedback if the heat is increasing! Trevithj (talk) 07:59, 1 October 2014 (UTC)
OK, I can see that you are confused; but that's because you're not understanding Black's definition. The negative loop gain doesn't affect what sign of error can be corrected. What's lowered is the gain from input to output, the close-loop gain, relative to the open-loop gain (e.g. from disturbance to effect, without and with the negative feedback loop closed). The loop gain is the forward gain divided by one minus the loop gain, which is less than the forward gain when the loop gain is negative. That what negative feedback is, and does. The polarity of the loop and the effect on the gain are strongly linked this way, and it doesn't matter which one you take to be defining (at least to the extent that the loop can be treated in an approximately linearized way). Dicklyon (talk) 18:04, 1 October 2014 (UTC)
I see. The troublesome text seems to be:

... the feed-back referred to as positive feed-back or negative feed-back according as the absolute value of 1/(1-μβ) is greater or less than unity. Positive feed-back increases the gain of the amplifier; negative feed-back reduces it.

(from H.S. Black, "Stabilized feed-back amplifiers") It isn't immediately clear if Black called the feedback "negative" because it reduced the gain, or saying that it reduced the gain because it is negative. Trevithj (talk) 18:29, 1 October 2014 (UTC)
There's no real distinction of cause and effect between the loop gain μβ and the closed-loop gain 1/(1-μβ); they are just algebraically equivalent descriptions of the feedback loop. When μβ is negative, 1/(1-μβ) is less than unity. No conflict or confusion here that I can see. I think Black understood this, but chose to focus on the gain, since that was what he wanted good control of. But when he says, "Positive feed-back increases the gain of the amplifier; negative feed-back reduces it", I presume he means "Positive μβ increases 1/(1-μβ); negative μβ reduces it ." Dicklyon (talk) 04:19, 2 October 2014 (UTC)
Well, that's the confusion - we have to presume his meaning. I suspect Black pushed the gain aspect because he had such trouble with the patents. He doesn't seem to use "negative feedback" as a term until some of the later submissions (around 1934). And when he does, he either talks in terms of gain-reduction, or refers to the "prior art". It is easy to assume that his definition is "negative = reduction". I'll re-read Mindell, to see the exact quote. (Mindell 2002). Trevithj (talk) 07:53, 2 October 2014 (UTC)

Mindell says:

In Black’s time, however, the definition of this specific-sounding term, “negative feedback,” had yet to be settled. The idea of positive feedback had become current in the 1920s with the introduction of the regenerative amplifier. Positive feedback, or regeneration, in a radio amplifier increased the sensitivity of a receiving tube by sending a wave back through an amplifier many times. Black insisted that his negative feedback referred to the opposite of regeneration: gain was reduced, not increased. Yet, to return to the analogy of the steam engine governor, Black’s use of “negative” means the energy required to spin the balls reduces the energy output of the engine, not that the balls trigger an action that slows it—hardly a significant effect for a steam engine. In their 1924 paper Friis and Jensen had made the same distinction Black had between positive feedback and negative feedback, that is, distinguishing one from the other not by the sign of the feedback itself but rather by its effect on the amplifier’s gain.39 In contrast, Nyquist and Bode, when they built on Black’s work, referred to negative feedback as that with the sign reversed. Black had trouble convincing others of the utility of his invention in part because confusion existed over basic matters of definition.

But this is just Mindell. Personally, I don't understand his point, since the two concepts (e.g. Black's versus Bode's) are completely equivalent. It didn't bother Nyquist and Bode, so why does it bother Mindell? His going off on "energy" here strongly suggests that Mindell is not reasoning clearly about these engineering concepts. This "confusion" is Mindell's invention, unsupported by anyone else, as far as I know. Hence we have a primary source issue in assigning this novel thesis any weight in the article. Dicklyon (talk) 05:12, 3 October 2014 (UTC)

I looked up some of the citing works. Aström and Murray (2008) make this point in their introduction:

The term feedback refers to a situation in which two (or more) dynamical systems are connected together such that each system influences the other and their dynamics are thus strongly coupled. Simple causal reasoning about a feedback system is difficult because the first system influences the second and the second system influences the first, leading to a circular argument.

— KJ Aström, RM Murray., Feedback Systems: An Introduction for Scientists and Engineers
That's the confusion - circular definitions. Is it negative feedback because it reduces the gain, or does it reduce the gain because it is negative feedback? Black's description is ambiguous, but strongly implies the former. The views aren't equivalent: Black defined the term based on its effect; Nyquist defined it as a cause. I think Mindell makes a valid point, and other authors echo it. Perhaps it could be better worded in the article though. Trevithj (talk) 08:05, 3 October 2014 (UTC)
The real confusion here is using words of vague content to replace mathematical argument. While some philosophers support ideas like emergence on the basis that our present theory of complex feedback and feed-forward systems is inadequate, for a simple case like the negative feedback amplifier the math is perfectly clear and logical, and makes the quote from Aström important if interpreted (as a reading of his work suggests) as saying causal reasoning is inapplicable and circular here. The notion of 'causality' is not used by the math, and is completely irrelevant to an understanding of this amplifier. Brews ohare (talk) 13:43, 3 October 2014 (UTC)
On the one hand, we could avoid circular definitions by using a mathematical approach. On the other hand, we could keep the article broadly accessible by using a narrative approach. How do we keep the article broadly accessible while avoiding circular definitions?
I suspect it might be true that Black wasn't defining negative feedback, so much as describing its effect. If so, it seems to have confused Mindell (among others) that the guy who popularized the term didn't define it. Does that imply there is an earlier source to be found? Trevithj (talk) 00:25, 5 October 2014 (UTC)
It would be helpful to define the objective here. At the moment the sentence under discussion occurs in the History subsection:
"Harold Stephen Black detailed the use of negative feedback in electronic amplifiers in 1934, where he defined negative feedback as a type of coupling that reduced the gain of the amplifier, in the process greatly increasing its stability and bandwidth. Nyquist and Bode built on Black’s work to develop a theory of amplifier stability, but chose to define "negative" as applying to the polarity of the loop (rather than the effect on the gain), which gave rise to some confusion over basic definitions."
This statement is an attempt to summarize remarks by Mindell, p. 121. The point of Mindell's discussion is to say that the term 'negative feedback' evolved over time. Personally I find that an unremarkable observation and also that Mindell has not made its historical background clear, but muddy.
One of Mindell's problems is that he suggests that "Black's sense of 'negative' means that the power required to spin the balls reduces the power output of the engine." One problem with this idea is that it equates an erroneous idea about how the governor works to Black's conception of feedback, which proposes mistakes Black would never make.
And Mindell does not understand that the steam-engine governor is a form of error-controlled regulation, which is unrelated to the idea behind the negative feedback amplifier, and has nothing to do with gain. The spinning balls of the governor control the throttle just like a cruise control keeps a car's speed constant. The negative feedback amplifier doesn't work that way - the circuit does not use feedback to 'regulate' the gain by adjusting the operation of the open-loop amplifier, in the manner of how a throttle control adjusts the power input to an engine. If one thinks of the input to the open-loop amplifier as an analog of fuel input to an engine, we then have the odd situation where the engine (open-loop amplifier) runs exactly the same way regardless of how much fuel is sent to it. Unless Mindell's ridiculous assessment of Black's concepts can be traced directly and verbatim to Black, I'd call this a gross misconception by Mindell.
Given that Mindell has introduced a mistaken comparison without adequate sourcing, why elevate this confusion to the level of a metaphysical argument over cause and effect unrelated to the accuracy of Mindell's historical account? Brews ohare (talk) 04:44, 5 October 2014 (UTC)
Mindell's unsupported historical assertion is that "Black had trouble convincing others of the utility of his invention in part because confusion existed over basic matters of definition" I am with Dicklyon on this one: if there was any historical confusion, Mindell has failed to identify it, and has failed to establish any historical importance for it. Brews ohare (talk) 05:42, 5 October 2014 (UTC)
The present text in the WP article suggests that the 'confusion' is exemplified by the difference in approaches of Black and the later work by Nyquist and Bode. However, this later work had no historical role in the acceptance of Black's ideas, and is not indicative of an earlier confusion. The Nyquist/Bode analysis was of amplifier stability in the sense of immediate response to electrical inputs, unlike Black's objective which was the long-term service stability of the amplifier over its service life to aging of components and power supplies. As Dicklyon says, there is no conflict between their views, only a matter of emphasis. Brews ohare (talk) 15:17, 5 October 2014 (UTC)
While I agree the Mindell's mention of the centrifugal governor is muddled, one can hardly conclude that "Mindell does not understand that the steam-engine governor is a form of error-controlled regulation." I think your analysis here goes way too far, but I'll agree that we should just leave out mention of the "confusion" that seems to only be discussed my Mindell, without much support or impact. Dicklyon (talk) 22:56, 5 October 2014 (UTC)
Going back to an earlier point, there is an earlier source that defines negative feedback:

Use of a common grid battery ... introduces a small feed-back from the second stage to the first. This feed-back may be either positive or negative, depending upon the phase relations in the intermediate transformer and may be eliminated by placing a condenser across the grid battery terminals.

— Robert W. King, Thermionic Vacuum Tubes and their Applications page58 (1923)
Clearly "negative" refers to out-of-phase signals "negating" each other.
BTW: I'm not sure if Mindell is referring to an "earlier" confusion - rather a subsequent one, especially when applying this term to the overall concept outside the context of electronic signals. Ashby and Ramaprasad mention this too. Trevithj (talk) 21:30, 5 October 2014 (UTC)
That is not my reading of the passage examined here. In any event, unless it can be demonstrated that this 'confusion' is important to history by finding some other sources that think so, it is not worth mentioning here. Brews ohare (talk) 02:44, 6 October 2014 (UTC)
The role of a 'confusion' of definition outside electronics is not germane to Mindell's claims about Black's difficulties 'selling' his idea. I believe that the primary evidence for the existence of any such difficulty is the nine years it took to get the patent granted. However, another explanation I have read is that the patent was enormously long and included a great many claims that had to be evaluated. Brews ohare (talk) 03:00, 6 October 2014 (UTC)
In any event, Black had lab prototypes that demonstrated the principle that negative feedback made overall gain independent of changes in the gain of the open-loop amplifier. This was not just a debate over terms. And Nyquist and Bode were entirely persuaded that Black's feedback circuit was working, regardless of any putative confusion over definitions. Brews ohare (talk) 03:11, 6 October 2014 (UTC)

Second paragraph of lede

The second paragraph of the lede presently reads:

Whereas positive feedback tends to lead to instability via exponential growth or oscillation, negative feedback generally promotes stability. Negative feedback tends to promote a settling to equilibrium, and reduces the effects of perturbations. Negative feedback loops in which just the right amount of correction is applied in the most timely manner can be very stable, accurate, and responsive.

There are some problems with this paragraph. The first sentence should be rewritten more accurately as:

Whereas positive feedback tends to lead to instability via exponential growth or oscillation, negative feedback can promote stability.

The words 'generally promotes' suggests that negative feedback as a general statement promotes stability. Of course, that is not the case, as negative feedback results in greater stability only under well-known circumstances.

The second sentence:

Negative feedback tends to promote a settling to equilibrium, and reduces the effects of perturbations.

makes sense for things like thermostatic control of house temperature (assuming an extended idea of 'equilibrium' to include 'steady-state'), but is inapplicable to the negative feedback amplifier (illustrated in the lede) where issues of "settling to equilibrium" play no role whatsoever. This sentence should be replaced by:

Negative feedback tends to reduce the effects of disturbances, in some cases external disturbances, and in other cases, internal.

Brews ohare (talk) 14:07, 3 October 2014 (UTC)

It is not unusual to look at amplifiers in terms of their step response. Negative feedback helps the output settle rapidly and smoothly to a new equilibrium after a step. Dicklyon (talk) 02:10, 4 October 2014 (UTC)
Oh baloney. It "is not unusual" to look at amplifiers in terms of lots of things. Brews ohare (talk) 02:46, 4 October 2014 (UTC)
The baloney was your statement that the second sentence "is inapplicable to the negative feedback amplifier (illustrated in the lede) where issues of 'settling to equilibrium' play no role whatsoever". I'm just pointing out that settling to equilibrium does play a role, sometimes. Surely you recall characterizing amplifiers by their step responses? Like here? Dicklyon (talk) 03:09, 4 October 2014 (UTC)
Dick, you are right, of course, that feedback does control some aspects of the step response of the negative feedback amplifier, although the settling time or damping is related to the open-loop amplifier time constants. I just doubt that will come to reader's mind when they read "Negative feedback tends to promote a settling to equilibrium". They are more likely to think of homeostasis, and it would be desirable for the reader to be nudged toward a broader view rather than reinforce a widely prevalent myopia. Brews ohare (talk) 15:35, 4 October 2014 (UTC)
Brews, you should realize that negative feedback will result in a corner frequency much higher than the one of the open-loop high-gain amplifier, no? That is, it promotes fast settling to a new equilibrium after an input step. It's hard to tell from your analysis in Step response, as that analysis assumes so much feedback that the second pole is what matters, rather than looking first at the simpler system with less feedback such that the forward amplifier's dominant pole is what matters; the feedback makes a faster closed-loop pole, and more feedback makes it even faster, up to the point that the second pole makes it start to ring, which is not where you want to be because that's where the net feedback is phase shifted enough to be positive it high frequencies; stop short of there, and negative feedback speeds up settling. However you look at, negative feedback usually promotes stability and fast settling, as in homeostasis and also in amplifiers. I see it as the opposite a positive feedback pushing toward ringing and singing and slow settling. Dicklyon (talk) 22:30, 5 October 2014 (UTC)
If you want to support your ideas about settling time, the place for that is Talk:Step response. I don't agree with you, and the analysis there doesn't either. However, here all this is digression and the point is: what is the likely inference of the general reader? I seriously doubt step response is on their short list.
The main problem throughout the various discussions on this Talk page is the unwavering belief that the negative feedback amplifier implements some variant of error-controlled regulation. This forced-fit is imbedded in the minds assembled here, and underlies your support for references to 'settling to equilibrium' as this choice of words satisfies your love for the simplicity of one universal approach. Brews ohare (talk) 02:37, 6 October 2014 (UTC)
I have updated Step_response#Step_response_of_feedback_amplifiers to include the simple case; please verify that I got it right. I'm not sure what you mean by "If you want to support your ideas about settling time"; I'm not interested in supporting any particular ideas here; I was just explaining why your claim is wrong, when you said that the statement about promoting settling is "inapplicable to the negative feedback amplifier (illustrated in the lede) where issues of 'settling to equilibrium' play no role whatsoever." Just trying to clarify the role so that you can get off this and do something else, perhaps more useful. Dicklyon (talk) 03:22, 6 October 2014 (UTC)
And just for the record, as I have said before, I am not at all in favor of an alleged "unwavering belief that the negative feedback amplifier implements some variant of error-controlled regulation". One does not need to believe that to see that negative feedback promotes stability and quick settling. Dicklyon (talk) 03:25, 6 October 2014 (UTC)
Dick: I've agreed with you that negative feedback affects step response; my reservation is simply that the wording involving equilibrium isn't likely to start the reader thinking about step response. Instead, they are going to think about homeostasis or the like. IMO that is too narrow a view of negative feedback. Brews ohare (talk) 04:45, 6 October 2014 (UTC)
I agree with your single-pole analysis. However, a reservation is that no real amplifier is single pole. If there are two poles then, for a normal amplifier design, the shorter time constant of the open-loop amplifier will determine the damping, and this will be in fact faster than τ/(1+βA) because stability and step response considerations force such a requirement. Brews ohare (talk) 04:58, 6 October 2014 (UTC)

Removal of unsubstantiated statements

The following paragraph:

"Harold Stephen Black detailed the use of negative feedback in electronic amplifiers in 1934, where he defined negative feedback as a type of coupling that reduced the gain of the amplifier, in the process greatly increasing its stability and bandwidth. Nyquist and Bode built on Black’s work to develop a theory of amplifier stability, but chose to define "negative" as applying to the polarity of the loop (rather than the effect on the gain), which gave rise to some confusion over basic definitions."

asserts that the attention given to 'polarity of the loop' by Nyquist and Bode, led to confusion over 'basic definitions' of the term negative feedback. This claim is unsupported and historically incorrect. The work of Nyquist (1932) and Bode (1940) led to the ideas of the Nyquist stability criterion and of gain margin that occurred much later than Black's idea of the negative feedback amplifier (1927). An extensive historical account is provided by Bennett, CHapter 3, pp. 70 ff. There is no evidence that the different emphasis, Black upon gain and Bode/Nyquist on polarity, caused any confusion. The Mindell source cited in support of this statement juxtaposes a sentence about Nyquist/Bode with a conclusion about 'confusion', but does not make the historically impossible claim of the WP sentence that this 'confusion' was occasioned by the differences between the Nyquist/Bode analysis and that of Black. I replaced the Mindell source with Bennettt, who supplies a more detailed history without Mindell's claims of confusion.

Accordingly, I have removed this last portion of the last sentence. Brews ohare (talk) 16:38, 5 October 2014 (UTC)

The lower sections involving Ashby and Skinner now don't follow on from the previous section.
Going back to an earlier point that seems to have been glossed over - when Black says "negative feed-back reduces ", is he defining negative feedback, or is he describing its effect? Trevithj (talk) 18:18, 5 October 2014 (UTC)
I suppose the interest in this question is historical: what does the record indicate Black thought was the definition? Evidence on this point might be found in his writings and patents. The possible significance of this point to this WP article has not been made clear. Brews ohare (talk) 01:00, 6 October 2014 (UTC)

This seems like it belongs in an earlier section; why the loss of context? Dicklyon (talk) 03:27, 6 October 2014 (UTC)

Inaccurate version of Black's definition of negative feedback

In the History subsection the sentence occurs:

" Harold Stephen Black came up with the idea of using negative feedback in electronic amplifiers in 1927, submitted a patent application in 1928, and detailed its use in his paper of 1934, where he defined negative feedback as a type of coupling that reduced the gain of the amplifier, in the process greatly increasing its stability and bandwidth. "

Black's BSTJ paper is the primary source for the assertion that Black's definition of negative feedback was a 'form of coupling that reduced the gain of the amplifier'. In actual fact, Black (p. 5) in the section titled Change of Gain Due to Feedback following his derivation of the gain as

G C F = 20 l o g 10 | 1 1 μ β | {\displaystyle G_{CF}=20log_{10}\left|{\frac {1}{1-\mu \beta }}\right|}

says:

"1/(1-μβ) will be used as a quantitative measure of the effect of feedback and the feedback referred to as positive feedback or negative feedback according as the absolute value of 1/(1-μβ) is greater or less than unity."

Now, of course, one implication of Black's definition is that negative feedback reduces the gain, but his definition is more general than this single implication, and implies many other matters, among them an increase in bandwidth and certain stability implications.

In view of the concern expressed on this page that Black's definition did not match Nyquist and Bode's interest in the sign of μβ:

"Nyquist and Bode built on Black’s work to develop a theory of amplifier stability, but chose to define "negative" as applying to the polarity of the loop (rather than the effect on the gain), which gave rise to some confusion over basic definitions" (See this).

the accurate statement of Black's definition assumes importance, to some WP editors at least. Brews ohare (talk) 15:11, 7 October 2014 (UTC)

It should be noted that Black does not refer to 'cause and effect' nor to the concept of error-controlled regulation where feedback reduces a measured performance discrepancy. Brews ohare (talk) 17:11, 7 October 2014 (UTC)

Where does this leave us?
If we can't talk about cause and effect, then how can we generalize Nyquist's "loop polarity" definition to examples that don't involve an electronic signal?
If we can't talk about measured performance discrepancy, then how can we generalize Black's "effect of feedback" to examples that don't involve reducing some "absolute" value?
Trevithj (talk) 22:19, 7 October 2014 (UTC)
Exactly: Black's approach is different, and does not use the ideas of error-controlled regulation, such as 'performance discrepancy'. It also does not involve 'cause and effect'. None of Black's works use these ideas, which at a minimum shows they aren't necessary to Black's work. Brews ohare (talk) 23:30, 7 October 2014 (UTC)
The suggested generalization that includes both approaches is 'opposition to change' discussed already . In the case of the amplifier opposition to change is reduction of output swing that reduces gain. In the homeostatic case opposition is to an increase in performance gap leading to a reduction in the gap. Brews ohare (talk) 23:34, 7 October 2014 (UTC)
More fundamentally, Black's approach lives with disturbances and works around them to render them uninfluential, while the other approache is to fight the disturbance and directly oppose it, thereby reducing its influence. It's the difference between avoiding flat tires by driving on roads free of debris and making tires tougher. Brews ohare (talk) 00:13, 8 October 2014 (UTC)
No, it's not. Those analogies make no sense at all. Dicklyon (talk) 03:59, 8 October 2014 (UTC)
Well. Dick, don't try to explain yourself. The analogy, whether it succeeds with you or not, has this going for it: it points out that Black's feedback circuit makes zero attempt to counteract variations internal to the open-loop amplifier (walks around the puddle), but makes the closed loop amplifier gain the same no matter what the open-loop amplifier does. In contrast, the error-controlled approach determines how far the open-loop amplifier has strayed and tries to correct this discrepancy by returning it to its 'normal' state (drains the puddle and walks straight through). Brews ohare (talk) 05:05, 8 October 2014 (UTC)
Your claim, or interpretation, that "Black's feedback circuit makes zero attempt to counteract variations internal to the open-loop amplifier" is nonsense. Negative feedback DOES counteract any such variations, by feeding some of the output back to the input with a negative sign. See Negative feedback. Dicklyon (talk) 06:09, 8 October 2014 (UTC)
Dick, it's not nonsense. Perhaps you have not understood my meaning? It's what Black's gain formula says, as pointed out repeatedly in discussion of Kal's gain desensitivity analysis. The open-loop gain can change by a relative amount ΔA/A and the open loop gain change is lowered in the closed-loop gain to (ΔA/ A) / (1+βA), by virtue of the circuit feedback. The change ΔA is not affected in any way; it is not counteracted or reduced. But it's influence on the closed-loop gain is small.
Error-controlled regulation, in contrast, confronts the variation ΔA head on and tries to reduce it to zero. Brews ohare (talk) 15:08, 8 October 2014 (UTC)
I understand that Black's gain formula only talks about the gain. But the same formula can be applied to show the the gain to a disturbance anywhere inside the forward path is reduced by the same factor. The fact that that fact is not explicit in one formula is not a reason to assert the nonsense that ""Black's feedback circuit makes zero attempt to counteract variations internal to the open-loop amplifier". Dicklyon (talk) 15:13, 8 October 2014 (UTC)
Apparently your goal is to entertain yourself by jeering rather than trying to understand what is said. It is very very clear that the internal workings of the open-loop amplifier play no role in the negative feedback amplifier. To attempt complete clarity, the focus is not upon the reduction of ΔA, not upon dealing with the fundamental origins of ΔA and fixing them. Historically that is the key point motivating Black who observed the frustrating failures of the technology of the day to succeed in countering the gain fluctuations of high-gain amplifiers. Black's solution was instead of fighting these fluctuations, to render them unimportant. The question of other noise sources inside the feedback loop (a topic not under discussion here) is treated exactly like those disturbances introduced by the open-loop amplifier, and in Kal's analysis, the closed-loop amplifier is desensitized in exactly the same way, by the desensitivity factor (1+βA). Brews ohare (talk) 15:35, 8 October 2014 (UTC)

It is unclear why this is relevant. The internal workings of the steam engine play no role in the centrifugal governor either. Nor does the workings of the furnace play a role in the thermostat. What exactly is the point of making the above distinction? Trevithj (talk) 18:35, 8 October 2014 (UTC)

I don't quite understand Brews either, as he points out. He seems to be arguing that since Black was more concerned about keeping the gain constant, it is no longer fair to note that the closed-loop ampifier "counteracts variations internal to the open-loop amplifier" or "reduces fluctuations in the output". This seems to be motivated by him wanting to prove that these concepts that apply to control systems do not also apply to negative feedback amplifiers. But they do. So he's wrong, and going off on an odd tangent for an odd reason, I think. Dicklyon (talk) 19:31, 8 October 2014 (UTC)
I guess this all really doesn't matter. The sections on error-controlled regulation and on the negative feedback amplifier are pretty clear and nobody (more accurately, not everybody) who reads them will confuse these different applications of feedback as being somehow the same. The leading sentence is not wrong. The second paragraph is a bit screwy, but WP has worse problems to deal with.
I do regret being unable to get through to you two, but unreasoned convictions are hard to overcome, and even appeal to the objective of reporting what sources say carries no weight.. Brews ohare (talk) 20:47, 8 October 2014 (UTC)
I do agree with his last sentence, in a pot-to-kettle sort of way. Trevithj (talk) 01:44, 9 October 2014 (UTC)
Exactly. From a person with a long history of idiosyncratic interpretations of sources. Dicklyon (talk) 02:26, 9 October 2014 (UTC)
Thanks for including me out of this love fest. It would help WP if talk pages weren't used for personal innuendo and instead focussed on presentation of sources. I have taken that route. Brews ohare (talk) 02:46, 9 October 2014 (UTC)
In reply to Trevith's remark that 'internal workings' play no role in error-controlled regulation, it may be noted that such regulation is predicated upon assessing the internal state of the system by monitoring selected 'essential variables', for example, temperature, or angular velocity. In the negative feedback amplifier no essential variable is selected to monitor the status of the open-loop amplifier. Perhaps Trevith regards the gain A of this amplifier as such a variable, but in the negative feedback amplifier the open-loop gain is not monitored, nor is it compared with a desired value or set point, nor is any attempt made to adjust its value in any way. Rather, the closed-loop gain is set at 1/β independent of A, so long as A is large. There is no candidate for an essential variable other than A, because this variable is the only role for the open-loop amplifier in the analysis.
It is doubtful that Trevith or Dick are unaware of these points, and their misunderstanding and their distorted summary of them is due, not to the error of these arguments, but to their impatience with them. This hostility toward obvious facts is born largely of their personal prejudice that this argument cannot possibly be correct if it concludes error-controlled regulation actually differs from what happens in the negative feedback amplifier. Brews ohare (talk) 15:36, 9 October 2014 (UTC)
Because directly confronting the obvious objections above to their prejudices is uncomfortable, Trevith and Dick resort to repackaging these objections in wrong or very muddy versions that they then can mock. This approach to protection of an idée fixe is common. Brews ohare (talk) 15:45, 9 October 2014 (UTC)
It is certainly common to this idea.
For the record, the "muddy" version is that a negative feedback amplifier monitors the output amplitude, compares it to the input amplitude, and processes the difference. It therefore is a control system. Trevithj (talk) 23:30, 9 October 2014 (UTC)
The actual statement of the linked source is
""A fraction of φo is fed back to the input and added with externally applied input signal φS"
There is no mention of a 'monitor', of 'comparing', or of a 'control system'. I believe the main connection between amplifiers and control systems is the use of op amps in control system comparators, which is not a role for the negative feedback amplifier.
Apparently, for you, this linked statement means the negative feedback amplifier is a 'control system'. I guess you think every control system is an error-controlled regulator, so QED? Else, if control systems come in various flavors, e.g. op amps, regulators, servos, whatever, then this linked statement with your extrapolation to 'control systems' doesn't support this contention of yours that the negative feedback amplifier works the same way as an error-controlled regulator. Both fire hoses and fire extinguishers are fire equipment, but they don't work the same way. Or, perhaps you would like to restate your contention about the similarities? Brews ohare (talk) 00:17, 10 October 2014 (UTC)
In the amplifier, the "essential variable" that is monitored and fed back is the output. To me, that doesn't make it a control system, but with minor contortions it can be cast as one. I'm not sure why Trevithj wants to do that, but it doesn't make what you've been saying any more sensible. Dicklyon (talk) 04:26, 10 October 2014 (UTC)
To be facetious, I don't think Trevith or you are trying to make more sense of these views. More seriously, if the output serves as the 'essential variable' here, how does it fit the definition of an essential variable? To what set point is it compared? What is the performance gap? How is this gap reduced? The lack of any answers to all these questions means 'output' cannot play this part. Brews ohare (talk) 04:43, 10 October 2014 (UTC)
Maybe a contender for 'essential variable' is the closed-loop gain? One might argue that the feedback loop sets β, which is the desired value of the closed-loop gain. If the output is not the input/β, the closed-loop gain is in error. Somehow one has to argue that the process of feeding back βO is some analog of regulation based upon the performance gap in the closed-loop gain. Could that be made plausible, at least? Brews ohare (talk) 04:59, 10 October 2014 (UTC)

Similarities between the negative feedback amplifier and error-controlled regulation

Taking the earlier questions in turn:

  1. It depends on what definition of 'essential variable' you are referring to. Vout is a variable, and it is essential to the operation of an amplifier, since it is the obvious sign that the amplifier is working.
  2. The set point is Vin.
  3. The performance gap is (of course) the difference between actual and desired Vout.
  4. The signal to the amplifier element is modified as required.

Maybe "control system" isn't exact terminology, but negative feedback involves some sort of comparison/difference. That is its salient point, IMO. Trevithj (talk) 05:08, 10 October 2014 (UTC)

These points reflect several misunderstandings.
1. An essential variable is a variable that has to be kept within assigned limits to achieve a particular goal.1
2. Vin cannot serve as a set point because the output is Vin/β, which is by design not at this set point.
3. There could be a performance gap of this kind, that is: error = Vout–Vin
4. The signal to the open-loop amplifier is modified as I–βO. However, it is uncertain just how this modification is achieved as a regulation of the error, although perhaps some argument can be devised to make the analogy.
A remaining issue is that this is all original research as no source has been found that uses this approach. Brews ohare (talk) 05:28, 10 October 2014 (UTC)
If β=1 then this would be error-regulation. Are you arguing that an amplifier can't be error-regulation because β<1? Trevithj (talk) 06:31, 10 October 2014 (UTC)

OK, given the strict definition of "essential variable", it would be best to say that doesn't apply in the case of the amplifier. When I said it was the output, I was speaking by analogy; it's the thing being monitored and fed back, not that it needs to be kept within certain bounds or compared to a reference. I also don't agree with Trevithj that negative feedback necessarily requires "some sort of comparison/difference", though it can often be cast that way, as with an op-amp which is a lot like a comparator. I prefer to just think of linear systems with adders combining input and feedback paths, where the notion of comparison or difference does not come up. Casting control systems into this form is more natural (to me) than going the other way. It's all in your POV, I think. As your assertion, Brews, that "Vin cannot serve as a set point", I think that's just wrong; control systems very often have indirect comparisons, so that the setpoint doesn't have to be specified in the physical domain of the output; there is usually an analog, or analogical, relationship between the setpoint signal and the actual monitored variable, is there not? If you made a system to compare by taking the difference Vout/beta - Vin, perhaps with another negative on Vin, and you amplified that error and used it to drive Vout to close the gap, you'd have a feedback amplifier, wouldn't you? Not that that's the only kind of feedback amplifier. But making up weird shit like trying to make the gain be the essential variable isn't making any useful point. Dicklyon (talk) 06:46, 10 October 2014 (UTC)

I'm afraid that Brews seems to have an obsession with the completely erroneous notion that there is some kind of fundamental distinction between the concept of negative feedback in the instance of a control system and the concept in the instance of, say, an audio amplifier. Let's take three examples: 1) a room thermostat; 2) the power steering system in a car; 3) an audio amplifier.
  1. For the thermostat the system input is the desired temperature setting and is static (of course it may be adjusted from time to time), and the output is the room temperature.
  2. For the steering system, the system input is the position of the steering wheel and it changes slowly, and the output is the direction of the car wheels.
  3. For the amplifier, the system input is an audio signal (eg from a microphone) changing rapidly at up to 70 radians/sec, and the output is generally a larger (more "ample") replica of that waveform (eg to drive a loudspeaker).

In each case any tendency of the output to deviate from the desired value generates a corrective signal which is fed back to the input in such a way as to restore it to an accurate tracking of the system input (as processed by the designed transfer function). Is this clear, or do we need to work through these examples in more detail to make the point? DaveApter (talk) 09:51, 10 October 2014 (UTC)

@Trevithj: You ask: "Are you arguing that an amplifier can't be error-regulation because β<1?" Yes, that is true for the negative feedback amplifier of Black. Because β=1 (the unity gain buffer) is a special case where the output and the input are the same, so the input can serve as a set point. See Friedland, Figure 1.2. This circuit can enforce a gain of β=1, but won't work when β<1 because then the input is no longer a set point. Brews ohare (talk) 15:45, 10 October 2014 (UTC)

A voltage amplifier using an operational amplifier to set the voltage across R1 to Vin.

@Dicklyon: Regarding the use of Vin as the set point of the negative feedback amplifier: it doesn't work because the output is Vin/β which is not Vin, and so Vin is not the output's set point. You go on to say:

"If you made a system to compare by taking the difference Vout/beta - Vin, perhaps with another negative on Vin, and you amplified that error and used it to drive Vout to close the gap, you'd have a feedback amplifier, wouldn't you? Not that that's the only kind of feedback amplifier. But making up weird shit like trying to make the gain be the essential variable isn't making any useful point."

Yes, the system you describe would be an error-controlled regulator. It might be the op-amp version in the figure. This op amp circuit takes the difference between its inputs 'Vin–βVout' and drives this error toward zero. It is not Black's negative feedback amplifier, and although it also amplifies with gain 1/β, it doesn't work like Black's amplifier, because the negative feedback amplifier does not take the difference 'Vout/β–Vin' and does not use this error to drive the error toward zero. Obviously, the discussion is not about whether different ways to make amplifiers can be found. The discussion is about how the way Black's amplifier works differs from the way you describe. You also say:

"But making up weird shit like trying to make the gain be the essential variable isn't making any useful point."

My purpose, unlike yours, was not to change the subject. It was to suggest a way of looking at Black's amplifier that resembles an error-controlled regulator. This suggestion is almost the same as your own, although you don't seem to realize it. The weak link is pointed out, that of finding the explanation of how this error is monitored and used in Black's amplifier, as it isn't at all clear that this happens. Brews ohare (talk) 14:50, 10 October 2014 (UTC)

@DaveApter: It isn't helpful to label the sourced views I am attempting to bring to your attention as an obsession. And the idea is that the same negative feedback applies to both amplifiers and error-controlled regulators. But, it is implemented in different ways. The governing 'negative feedback' is stated in your first sentence of the lede although, because of your second paragraph, I doubt you see the ambiguity in that first sentence that allows it to cover both applications. Your first two examples, the thermostat and the steering control are both well understood as instances of error-controlled regulation. Why you bring them up is unclear. The third example, an audio amplifier, could be implemented in many ways, and that makes the discussion vague because here the subject is Black's amplifier, not audio amplifiers in general. Your new definition of 'negative feedback' is a system in which any "tendency of the output to deviate from the desired value generates a corrective signal which is fed back to the input in such a way as to restore it to an accurate tracking of the system input" This new definition is inferior to the existing one in the lede, as it doesn't apply to Black's amplifier, where there is not present any attempt at "accurate tracking of the system input". The goal instead is the creation of an enlarged image of the input where the degree of enlargement is held constant despite disturbances.

@All: One missing item in all three of your expositions here is any specification of a disturbance that has to be minimized, which is the point of the feedback. In Black's work, the disturbance whose influence he is reducing is any undesirable variation in the open-loop gain. That goal has not come up in your presentations. Brews ohare (talk) 15:09, 10 October 2014 (UTC)

A few points arising from this latest batch of exchanges:
  1. I didn't explicitly mention resilience to "disturbances" because dealing with them is a natural by-product of the operation of the feedback that has been described.
  2. There is no difference in principle between disturbances (ie factors that would tend to move the output of the system away from the desired value) that arise externally and those that arise internally. Examples of external disturbances - for the three examples above - would be: a) heat leaving or entering the room being regulated; b) the road wheels being jolted by bumps and contours; c) the loading on the output of the amplifier being altered, eg by extra speakers being switched in or out. Examples of internal disturbances would be: a) The heating element becoming more or less efficient; b) the hydraulic steering pump becoming less efficient over time due to wear; c) changes in the open-loop gain of the amplifier stage due to, for example, changes in semiconductor characteristics because of temperature drift. In all of these cases the feedback operation will cause the desired output to be maintained regardless.
  3. My reference to the 'transfer function' above was intended to cover all cases, whether it is positive amplification, fractional amplification (normally called attenuation of course), unity gain, and whether inverting or non-inverting, as well as the implementation of other mathematical functions. The principles apply in all cases.
  4. I've been designing with op-amps for decades - for audio amplification, audio signal processing, control systems, and analog computation. I see no difference in principle between any of these. Since the original purpose of op-amps was as building blocks for modelling physical processes, I see the same basic ideas in play however they are implemented (and this is a key theme in Ashby's book). Prior to that, I built amplifiers with discrete transistors and earlier still with valves. I see no difference in the fundamentals in any of these. I'm not acquainted with Black's amplifier but I'll look inot it, but I'll be astonished if it deviates in any essential regard from the fundamentals that apply to all of these examples. DaveApter (talk) 17:06, 10 October 2014 (UTC)
A voltage amplifier with gain 1/β when the open-loop gain A is large
Dave: The idea that the op-amp amplifier of the figure immediately above and Black's amplifier are'fundamentally' the same is bound to be true at some level of abstraction. However, that is not at stake here, where the issue is how they differ. Obviously, they do differ. The difference is that the op amp circuit uses a performance gap and its reduction, while Black's amplifier does not, at least as its operation is described in textbooks like that of Kal. Brews ohare (talk) 17:28, 10 October 2014 (UTC)
The figure at right shows a voltage amplifier with gain A/(1+βA) ≈ 1/β for discussion. I think it's an example of Black's amplifier. The error is Vin–βVO. The circuit drives this error toward zero. These aspects of operation seem clear, and similar to error-controlled regulation. However, in error-controlled regulation the 'error' is an actual deviation of the disturbed system (i.e. A, which has become AA) from its 'normal' state, and the return of A to its normal state by regulation, by driving ΔA→ 0. That is not what happens here. Instead, the gain is set up by the feedback to be 1/β whether or not A remains at its deviant value AA. Brews ohare (talk) 18:52, 10 October 2014 (UTC)
In candor, my understanding has evolved, and I think this figure and the op-amp figure are basically identical. The key distinction of both circuits from error-controlled regulation is the last sentence above: neither of these circuits drive the disturbance ΔA→ 0. Brews ohare (talk) 19:01, 10 October 2014 (UTC)
That distinction brings to mind an earlier analogy. Is the process of keeping a car in the center of the lane the same if the road is straight as it is when the road is curved? One requires occasional correction, the other requires continuous correction. Does this matter?
BTW, looking at the diagram above - why can't βVout be an essential variable? Trevithj (talk) 01:24, 11 October 2014 (UTC)
Can you address the point raised that the difference to be noted is that ΔA is not driven to zero when Vin-βVout is driven to zero, but when a performance gap based upon an essential variable is driven to zero, the system governed by that variable is returned to its normal state that prevailed before the disturbance occurred? In Black's amplifier, the disturbance is ΔA in the open-loop amplifier, but this disturbance remains despite the feedback. Brews ohare (talk) 02:23, 11 October 2014 (UTC)
To me, this difference indicates different ways of using feedback operating differently. Brews ohare (talk) 03:12, 11 October 2014 (UTC)
I thought I was addressing the point. But since you ask: Vout is returned to its normal state (or near enough if A is large) that prevailed before the disturbance occurred. If the disturbance remains, that means continuous correction is required - like the car on the curved road. Does this matter?
Also, why can't βVout be an essential variable? Trevithj (talk) 04:09, 11 October 2014 (UTC)
Trevith: We don't understand the 'disturbance' the same way. Assuming A is large, in Black's amplifier the disturbance ΔA has no effect upon Vout. That is because ΔA changes the open-loop gain A but doesn't change the closed loop gain 1/β so Vout remains Vin/β whether ΔA takes place or it doesn't. That desensitivity results because the percentage change in closed loop gain is reduced from that in the open loop gain by the feedback factor (1+βA) which in principle is as large a reduction as one can imagine if A is large enough. Brews ohare (talk) 05:02, 11 October 2014 (UTC)
We also understand 'normal state' differently. A variable may have a normal value, but only a system (a complicated entity) has a normal state. This state may be identified by the values of certain essential variables. For example, the state of an ideal gas can be identified by specifying two of the variables pressure, temperature and volume, with the three interrelated by an equation of state. The state of an amplifier involves a myriad of transistor parameters and bias voltages and temperatures that are only hinted at by the single gain parameter A.
I am unclear what your interest is in deciding whether βVout is identifiable as an essential variable. Whether the answer is yes or no doesn't seem to play a role in the present discussion. In any event, the state of the open-loop amplifier, the system whose state variations cause ΔA, has nothing to do with βVout, which is not an essential variable of the open-loop amplifier and has no influence over ΔA. One type of change ΔA, one type of variation among those envisioned by Black, is caused by things like aging of the transistors or heating of components, or fading of battery voltages; nothing to do with the input signal or the feedback loop. Brews ohare (talk) 05:02, 11 October 2014 (UTC)
Black also was interested in reproducibility of manufacture, where he wanted to be able to make amplifiers with a specified gain 1/β despite poor reproducibility of high-gain amplifiers. (One application to telephony he was involved in incorporated the combined use of many amplifiers that had to have identical gains.) In this case ΔA is a manufacturing variation and despite this variation, the closed loop amplifiers regardless of their different values of A, all will have the specified gain 1/β. 1 Brews ohare (talk) 14:44, 11 October 2014 (UTC)

This is all very interesting, professor, but you still haven't addressed either of my questions.

  1. I assume that you believe that βVout can not be an essential variable. My question is: why not? It fits the definition you provided earlier.1
  2. I accept that Black's amplifier does not directly influence ΔA. My question is: does it matter? A car doesn't directly influence the curve of a road either, but stays in its lane. Trevithj (talk) 22:37, 12 October 2014 (UTC)
1. It makes no sense to suggest that βVout is an essential variable without specifying what system you are talking about for which it serves that purpose. If we are talking about the open-loop amplifier, I answered your question in detail: No. A possible exception would appear to be a signal so large as to saturate the amplifier, but that circumstance would not be an exception in fact, because it would drive the gain A to a small value, violating the assumption that A is large.
2. Does it matter that Black's amplifier does not affect ΔA? It does if you believe that Black's amplifier is a form of error-controlled regulation that opposes ΔA. And as far as I can determine, it is your belief that Black's circuit does employ error-controlled regulation. If this emphasis upon error control isn't your position, then we have nothing to discuss: Black's amplifier uses feedback in a different fashion than error controlled regulation. If error control is your position, then you must specify what performance gap it is that Black's circuit opposes; obviously it is not ΔA. Brews ohare (talk) 23:26, 12 October 2014 (UTC)
1 We agree that the open-loop amplifier doesn't include the feedback signal βVout. That is why it is called open-loop, I guess. I am talking about the closed-loop amplifier, since βVout is the feedback signal that forms the closed loop in question. So in this system (the one with negative feedback) can βVout be an essential variable?
2 To clarify: It is my position that Black's amplifier is a form of error-controlled regulation. It is not my position that Black's amplifier is a form of error-controlled regulation that opposes ΔA. As you have correctly pointed out, nothing opposes ΔA. So it follows that I don't think it matters. Do you think it matters? Trevithj (talk) 19:50, 13 October 2014 (UTC)

OK, you propose that βVout is an essential variable of the open loop amplifier, suggesting that the open-loop system maintains its normal operation by comparing the value of βVout with some set point value, say Vin, and forcing this variable toward its set point. That is how it operates. However, that is its normal operation, and there is no regulation going on here, no overcoming of a disturbance from normal operation. This point of view does nothing to explain why one would choose to use feedback in this way rather than simply to directly amplify the input signal with an open-loop amplifier of gain 1/β and be done with it. Any explanation? Brews ohare (talk) 21:51, 13 October 2014 (UTC)

Given that one could simply use an open loop amplifier, why introduce feedback? Recognized explanations for using feedback include benefits other than desensitivity to gain variations, such as increased bandwidth, greater linearity and so forth. But these last have nothing to do with coping with disturbances and provide no opening for the justification that feedback is introduced to provide error-controlled regulation. Rather, feedback is introduced for these other benefits unrelated to regulation. Brews ohare (talk) 22:12, 13 October 2014 (UTC)

Historically, Black was very concerned with gain variations, and they were a major impetus for his use of feedback. Perhaps you will recognize that a comparison of how feedback deals with disturbance by desensitization is a departure from the approach of homeostasis? Brews ohare (talk) 22:31, 13 October 2014 (UTC)

Black's amplifier compared

Thanks Brews for providing the link to Black's paper. The two things I notice immediately are: a) The equation he gives for the closed-loop gain is exactly the same as the equation for a standard op-amp implementation of a positive gain non-inverting amplifier; and b) The conceptual diagram he gives is exactly as we would draw the schematic for an op-amp implementation. This is even though op-amp modules did not exist as available building blocks at that time! This would seem to indicate that the underlying principles are common. Designing with op-amps is so much simpler because the signal paths and feedback paths can be dealt with conceptually, whereas in discrete valve or transistor designs, the feedback mechanism is often not obvious without a thorough circuit analysis. In some ways it's analagous to the difference between programming in say Java and in assembler - in Java you can just concentrate on the algorithm but in assembler you have to deal with the details of registers, opcodes, addressing modes etc.

The short conclusion is that I can't see that it's been established either that Black's original design differs in principle from modern op-amp implementations, or that either differ in principle from control-system feedback applications. And I can't see that any of the sources that Brews has cited say otherwise. Also that with about 35 refs to the article, that it is inadequately supported by citations. DaveApter (talk) 09:47, 13 October 2014 (UTC)

Hi Dave: I agree with you about the equivalence in principle between op-amp amplifiers and Black's circuit. The difference between them is the implicit summer in the op amp and the explicit summer in Black's circuit. In transistor amps realizing Black's circuit, this summer is built in by such subtleties as the sign-flip of a single transistor stage, or feeding back to the emitter of the input transistor in a two-stage amplifier. In the op amp the summer is incorporated as a differential pair. These different ways of achieving the subtraction of the feedback from the input are not essential.
However, it does not follow that there is "no difference in principle from control-system feedback applications". First, "control systems" is a broad category that includes both error-controlled regulators and Black's circuits, and these two both are in this category but are not alike. Second, as pointed out above, the feature distinguishing between them is how they approach undesirable disturbances. In error-control the effect of the disturbance is countered by bringing the disturbed system back into normal operation. In Black's approach the disturbance ΔA is not corrected and the open- loop amplifier is not brought back. Instead the open-loop amplifier is allowed to do its thing, but the closed-loop amplifier is rendered insensitive to this behavior. This achievement of desensitivity to disturbances without their correction is very well documented by numerous sources linked above. Brews ohare (talk) 15:30, 13 October 2014 (UTC)
These two classifications of control systems use fundamentally different ways of approaching disturbances. Brews ohare (talk) 13:48, 13 October 2014 (UTC)
By this argument, steering a car along a straight road is "fundamentally different" to steering a car along a curved road. How is that a useful distinction? Trevithj (talk) 19:56, 13 October 2014 (UTC)
Trevith: Perhaps you could explain where the analogy is here that you find so compelling? Brews ohare (talk) 21:08, 13 October 2014 (UTC)

Reverted addition of new figure

Negative feedback amplifier using ideal unilateral elements
A negative feedback amplifier with gain 1/β when the open-loop gain A is large
A voltage amplifier using an operational amplifier to set the voltage across R1 to Vin.

I reverted this edit in which Brews added the figure above, with an amplifier and an abstract adder. It seems to obscure, rather than clarify, the conditions under which the op amp circuit behaves approximately ideally, since the abstract adder gives no indication of where current might flow, or what it adds even, so it's hard to interpret much from it that would be different from the op-amp idealization; just say the op-amp has a finite gain if that's the point. And it's not clear what the goal of another figure there would be anyway. Dicklyon (talk) 22:36, 12 October 2014 (UTC)

Hi Dick: It's my understanding that the adder combines the voltage inputs, just as it does in the ideal negative feedback amplifier circuit assuming voltage input and voltage output. In both, the adder combines its inputs, producing Vin–βVout. The only difference between these two circuits is that the lower one provides a voltage divider for the network that feeds back the voltage βVout.
Perhaps it needs to be clarified that a voltage amplifier is modeled here, as the general negative feedback amplifier circuit of the top diagram is vague on this point, and could have any combination of current and voltage inputs and outputs provided β incorporated the appropriate dimensions (V/V. V/A, A/A, A/V) to convert the output to the same form as the input.
The goal of the center figure is, of course, to clarify the connections between the negative feedback voltage amplifier and the op amp voltage amplifier (bottom diagram). I believe it is well-known that op-amps ideally admit no current, and that the summer likewise deals only with its voltage inputs. To clarify the confusions that plague you about this comparison, perhaps you could suggest some rewording? Brews ohare (talk) 23:58, 12 October 2014 (UTC)
In the event that the algebra behind the negative feedback amplifier circuit escapes you, for both negative amplifier circuits the output is given in terms of the input by:
V o u t = A ( V i n β V o u t ) {\displaystyle V_{out}=A\left(V_{in}-\beta V_{out}\right)}
V o u t ( 1 + β A ) = A V i n {\displaystyle V_{out}(1+\beta A)=AV_{in}}
V o u t V i n = A 1 + β A 1 β {\displaystyle {\frac {V_{out}}{V_{in}}}={\frac {A}{1+\beta A}}\approx {\frac {1}{\beta }}}
where, of course, in the lower circuit:
β = R 1 R 1 + R 2 . {\displaystyle \beta ={\frac {R_{1}}{R_{1}+R_{2}}}.}

Brews, the top and bottom pictures are clear and unambiguous. The top is abstract, and the bottom is a circuit with inputs and outputs being voltages. The one in the middle is a muddled mix, using an abstraction of a voltage adder and amplifier with a voltage divider circuit. With the op-amp circuit, one can talk about near-infinite input impedance not loading the voltage divider. With your new diagram, it's less clear what is circuit and what is abstract idealization, or how it is intended to help with the understanding. Dicklyon (talk) 00:49, 13 October 2014 (UTC)

Dick: I am not sure I understand the 'muddled mix' description. It's clear that voltages are used for input and output, so that's not an issue. My best guess is that the replacement of the β-block by a voltage divider has derailed you because this is a very particular example of such a β-block and not an abstract version that could be represented in many different ways. Is that the issue? Brews ohare (talk) 02:35, 13 October 2014 (UTC)
To my mind the op-amp circuit has done the same thing. What is the difference in your mind? Brews ohare (talk) 02:40, 13 October 2014 (UTC)
A second point you raise is that it is not clear that the summer is assumed to draw no current from the divider, which loading would alter the value of β from the simple resistor ratio. If that is a probable concern of readers, it is avoidable by simply pointing out the assumption that, like the ideal op amp, the summer draws no current. Brews ohare (talk) 17:21, 13 October 2014 (UTC)
As a real example, if I used this figure and replaced the dashed block enclosing the transistors with the triangular designation of this amplifier, that figure would be equivalent to the one I have used here. Brews ohare (talk) 02:58, 13 October 2014 (UTC)
As indicated in the discussion above with Dave, the only difference between the center and the bottom figures is making the summer explicit in the center circuit that is implicit in the op amp. This equivalence might seem more transparent if the summer were simply moved to the right and stuck inside the amplifier triangle? The center circuit is possibly more general in not assuming a particular implementation of the summer, while the op amp almost always uses a differential pair as the input stage that differences the signals at its two inputs. Of course, the op amp summer might be achieved differently. Brews ohare (talk) 14:01, 13 October 2014 (UTC)
I believe this discussion demonstrates your reservations about the center figure are groundless, and the discussion above with Dave shows the comparison between op-amp circuits and Black's serves a useful purpose in demonstrating their fundamental equivalence. However, the presentation of these points could benefit from your attention, so please look into that. Brews ohare (talk) 14:43, 13 October 2014 (UTC)
I can only agree with Dicklyon, the picture was a weird mishmash of a system diagram and a circuit diagram.GliderMaven (talk) 19:10, 13 October 2014 (UTC)
Your endorsement of Dicklyon's characterization contains nothing to explain your reasons for it. This figure, as has been carefully pointed out, is no more a 'mishmash' with its idealization of a summer that draws no current from the resistor divider than is the idealized op amp with its differencing at its input that acts exactly the same way. Without any comment on this point, GliderMaven, your support is simply a "me too" with all the weight that mindless effort deserves. Brews ohare (talk) 21:16, 13 October 2014 (UTC)
Categories:
Talk:Negative feedback: Difference between revisions Add topic