Folks, virtually everything that we have considered is 'feedback' in some form, BUT there are different types of feedback. One is series, another parallel, and a third group is global, which can be either series or parallel.
There are lessons to be learned from this:
For example, local feedback gone wild, can be called a follower. This is because, in order to get any meaningful output from a device with a strongly degenerated Gm, you have to use it as a follower. However, in principle, if you put a cascode stage on top, and then an active load in the collector, you could, in theory, have an extremely linear voltage gain stage. Noisy yes, hi Z output, yes, but real voltage gain, and real linearity too!
This implies that a follower is really just a 'series feedback' stage. However it will behave differently than a parallel feedback stage or a global feedback connection. This is what we want to keep clear. It is not necessarily 'feedback' that is the problem, but how it is implemented. I realize that many will insist that we should not have separated followers from fed-back common emitter pairs, because they use some similar basic mechanism, but they do not behave exactly in the same way. This is important to serious audio designers.
There are lessons to be learned from this:
For example, local feedback gone wild, can be called a follower. This is because, in order to get any meaningful output from a device with a strongly degenerated Gm, you have to use it as a follower. However, in principle, if you put a cascode stage on top, and then an active load in the collector, you could, in theory, have an extremely linear voltage gain stage. Noisy yes, hi Z output, yes, but real voltage gain, and real linearity too!
This implies that a follower is really just a 'series feedback' stage. However it will behave differently than a parallel feedback stage or a global feedback connection. This is what we want to keep clear. It is not necessarily 'feedback' that is the problem, but how it is implemented. I realize that many will insist that we should not have separated followers from fed-back common emitter pairs, because they use some similar basic mechanism, but they do not behave exactly in the same way. This is important to serious audio designers.
john curl said:
No, John, "global" isn't a third group of feedback with respect to series or parallel.
There are two types of feedback. Degenerative and regenerative.
That's it.
After that it's simply a matter of implementation.
It can be implemented as a series element, a parallel element or a combination.
It can be implemented locally in a single stage, less locally across two or more stages, or globally around the entire circuit.
But it's all still feedback.
How, specifically, does it behave differently?
Whether series or parallel, the end result is that you're feeding all or part of your output back to the input.
They use the EXACT SAME MECHANISM. It's called degenerative feedback.
Degenerative feedback behaves just like degenerative feedback.
It's important to serious audio designers who make their living selling amplifiers to a market which thinks "feedback" is a four letter word.
Which is why we have all this song and dance trying to make out that feedback is something other than feedback.
se
EF & series feedback
Ohm's law?! Really! That's interesting that you can describe a circuit with a non-linear *active* device, as "Ohm's law"! Your resistive divider is an oversimplification. The series feedback isn't obvious, but it's there.
A resistive divider is nothing like an EF. In a resistive divider, if one resistor is replaced with one of lower value, due to tolerance, both resistors will carry an increased current, as a result. There is no feedback mechanism to correct this. In an EF, if the transistor is replaced with one having higher forward current gain, 3 times higher, the overall voltage gain of the stage changes very little, remaining just under one. The current gain of the transistor triples, yet the stage voltage gain remains at one. The same is true with 5 times the original current gain, or 10 for that matter. A zero feedback circuit, such as a common emitter stage, will exhibit gain which varies directly with transistor current gain. Another way to view it is as follows. The reflected input resistance increases with increasing forward current gain. An hfe of 100, will reflect the emitter equivalent resistance (emitter resistor in parallel with load) by a factor of 101 (referred to base or input). If another transistor has hfe = 200, the emitter resistance gets multiplied by 201. In the first case, a current flows in the base side, with a gain of 101. In the second case, we have half the current, but a gain of 201. The circuit exhibits nearly constant gain despite large variations in transistor characteristics. With a common emitter circuit, such is not the case. In a CE circuit, there is no inherent feedback (except for the small inherent Re), and the stage gain varies directly with transistor parameters. In the EF, the series feedback, and its Re multiplying action, is what gives high input impedance, a desirable property for a buffer. A resistive divider simply presents the series combination of the two resistors as its input impedance. There is no buffering or multiplication. A resistive divider and an emitter follower are as far removed as night and day. How can you even compare the two?
Series feedback is not that uncommon. It is often used on inputs to increase input impedance. A SIPO (series input parallel output) feedback topology achieves this.
To summarize, any increase in hfe, due to part to part device variation, will try to force greater current into the emitter resistance. This tends to raise the emitter voltage wrt ground. This forces the voltage across the base resistor to decrease, and the base-emitter junction voltage to decrease, which results in a lower current. Any attempt to increase current is cancelled by Re. Also, if hie, the base-emitter input resistance changes, its effect is neutralized as well by Re. I'm tired, it's late. Best regards.
Ohm's law?! Really! That's interesting that you can describe a circuit with a non-linear *active* device, as "Ohm's law"! Your resistive divider is an oversimplification. The series feedback isn't obvious, but it's there.
A resistive divider is nothing like an EF. In a resistive divider, if one resistor is replaced with one of lower value, due to tolerance, both resistors will carry an increased current, as a result. There is no feedback mechanism to correct this. In an EF, if the transistor is replaced with one having higher forward current gain, 3 times higher, the overall voltage gain of the stage changes very little, remaining just under one. The current gain of the transistor triples, yet the stage voltage gain remains at one. The same is true with 5 times the original current gain, or 10 for that matter. A zero feedback circuit, such as a common emitter stage, will exhibit gain which varies directly with transistor current gain. Another way to view it is as follows. The reflected input resistance increases with increasing forward current gain. An hfe of 100, will reflect the emitter equivalent resistance (emitter resistor in parallel with load) by a factor of 101 (referred to base or input). If another transistor has hfe = 200, the emitter resistance gets multiplied by 201. In the first case, a current flows in the base side, with a gain of 101. In the second case, we have half the current, but a gain of 201. The circuit exhibits nearly constant gain despite large variations in transistor characteristics. With a common emitter circuit, such is not the case. In a CE circuit, there is no inherent feedback (except for the small inherent Re), and the stage gain varies directly with transistor parameters. In the EF, the series feedback, and its Re multiplying action, is what gives high input impedance, a desirable property for a buffer. A resistive divider simply presents the series combination of the two resistors as its input impedance. There is no buffering or multiplication. A resistive divider and an emitter follower are as far removed as night and day. How can you even compare the two?
Series feedback is not that uncommon. It is often used on inputs to increase input impedance. A SIPO (series input parallel output) feedback topology achieves this.
To summarize, any increase in hfe, due to part to part device variation, will try to force greater current into the emitter resistance. This tends to raise the emitter voltage wrt ground. This forces the voltage across the base resistor to decrease, and the base-emitter junction voltage to decrease, which results in a lower current. Any attempt to increase current is cancelled by Re. Also, if hie, the base-emitter input resistance changes, its effect is neutralized as well by Re. I'm tired, it's late. Best regards.
i don't know about the pros here but it does say something that the discussion of what constitutes negative feedback could go on so long without a mention of the foudation documents:
US patent 2102671
"Wave Translation System"
Harold Black
"Network Analysis and
Feedback Amplifier Design"
HENDRIK W. BODE
or that both are available on the web:
http://www.uspto.gov
http://mimosa1.incubator.uiuc.edu/jba/
for a more modern approach check out Wai-Kai Chen's 1980 "Active Network and Feedback Amplifier Theory" it takes 200 + pages to show how to get to Bode's definitions of F and S and is billed as a graduate level text - so much of the confusion here could be excused
US patent 2102671
"Wave Translation System"
Harold Black
"Network Analysis and
Feedback Amplifier Design"
HENDRIK W. BODE
or that both are available on the web:
http://www.uspto.gov
http://mimosa1.incubator.uiuc.edu/jba/
for a more modern approach check out Wai-Kai Chen's 1980 "Active Network and Feedback Amplifier Theory" it takes 200 + pages to show how to get to Bode's definitions of F and S and is billed as a graduate level text - so much of the confusion here could be excused
Claude,
I'm sorry I'm not making my point clearly enough. By the way, Ohm's Law applies to non-linear circuits too.
My model of a transistor is a "small signal" model. This is standard technique for analyzing circuit behaviour. For small signals a transistor behaves very much like a resistor with current gain. The base-emitter resistance is usually called Rpi and the transconductance gm. This is quite the same as a resistor with current gain at a particular operating point. When large signals are involved the non-linearities need to be considered explicitly and things get more complicated, but this is not necessary to demonstrate the potential divider equivalence.
You seem to be arguing that the fact that varying the beta doesn't vary the voltage gain demonstrates feedback. I disagree. A transistor's gm is largely set by Ic, not beta. Besides, Rpi will vary with beta (Rpi is effectively in parallel with the transconductance) and will cause some change in voltage gain. The fact that the transconductance does not vary linearly with beta is not a negative feedback effect. The fact that a transistors' small signal model is a resistor with current gain does not imply any negative feedback either.
Are you disagreeing with my small-signal model or the gain equations I published earlier? If they are correct then do you not agree that the mathmatics can't tell the difference betwen an EF and a resistive divider?
I'm sorry I'm not making my point clearly enough. By the way, Ohm's Law applies to non-linear circuits too.
My model of a transistor is a "small signal" model. This is standard technique for analyzing circuit behaviour. For small signals a transistor behaves very much like a resistor with current gain. The base-emitter resistance is usually called Rpi and the transconductance gm. This is quite the same as a resistor with current gain at a particular operating point. When large signals are involved the non-linearities need to be considered explicitly and things get more complicated, but this is not necessary to demonstrate the potential divider equivalence.
You seem to be arguing that the fact that varying the beta doesn't vary the voltage gain demonstrates feedback. I disagree. A transistor's gm is largely set by Ic, not beta. Besides, Rpi will vary with beta (Rpi is effectively in parallel with the transconductance) and will cause some change in voltage gain. The fact that the transconductance does not vary linearly with beta is not a negative feedback effect. The fact that a transistors' small signal model is a resistor with current gain does not imply any negative feedback either.
True. But this isn't due to negative feedback. It is due to a transistor behaving like a resistor with current gain.
Are you disagreeing with my small-signal model or the gain equations I published earlier? If they are correct then do you not agree that the mathmatics can't tell the difference betwen an EF and a resistive divider?
jcx,
Thanks for the patent links. I had a quick look and realised that it will take me more time than I can afford to read through all this info. What I seek is a succinct definition of feedback.
This succint definition can then be applied to circuit topologies to reveal whether they should be avoided, based on the popular assertion that feedback is bad for sound quality.
Thanks for the patent links. I had a quick look and realised that it will take me more time than I can afford to read through all this info. What I seek is a succinct definition of feedback.
This succint definition can then be applied to circuit topologies to reveal whether they should be avoided, based on the popular assertion that feedback is bad for sound quality.
Housekeeping note: the latest installment of The Bickersons has been moved to Off Topic. This thread is for discussions of feedback theory, not motivations, personalities, history, and employment.
I think I am accounting for that. The current flow through the transistor will be (Vb-Ve)/Rpi + (Vb-Ve)*gm. The gm is what you are refering to: the transconductance resulting from the current source. These are like two resistances in parallel. In my diagram I didn't show these separately and just assumed Gm included both in the Vo=(Vb-Vo).gm.RL.
In a small-signal model it doesn't matter where the collector is connected - that's why I left it floating; the circuit doesn't care where the current loop is - an ideal current source has infinite impedance. If it helps, connect the collector to ground.
In a small-signal model it doesn't matter where the collector is connected - that's why I left it floating; the circuit doesn't care where the current loop is - an ideal current source has infinite impedance. If it helps, connect the collector to ground.
I am happy to see that you guys can make a discussion so
wonderfully confused without my help.
Watching this show from the row, not on-stage, I get the impression
that some of the confusion might come from not separating certain
issues at a higher level. Mathematics doesn't know about feedback
at all, so in that sense traderbam is right saying that "the mathematics cannot distinguish between ..."
Feedback is a conceptual thing, which
we use as an aid to understand systems. As soon as we translate
the system to mathematics, the feedback is "lost". Maybe I am wrong
here, but I think the concept of feedback is always based on some
assumption of causality, since we humans tend to like to think of
things in terms of causal realtionships, or at least those of us who
are trained as engineers or natural scientists. This seem related
to the "controversy" between natural scientists and philosophers.
Natural scientists are usually satisfied if they can find a mathematical
model with good predictive power of a phenomenon, while many
philosophers will argue that this does not mean we really understand
the phenomenon and they will ask for or search for other types of
models that "explain" the phenomenon.
The other thing to note is it is always good to try having a precise
terminology for things to avoid confusions and misunderstandings.
Unfortunately, it is very common that no universally agreed upon
terminology exists. We have seen plenty of such examples in
previous threads, the two (or even more) fundamentally different
uses of the term "current feedback" being a good example.
From a practical engineering point of view, at least, it seems to me
(and it seemss some of our famous designer think so too)
that it is useful to distinguish between various types of feedback,
even if it is not clear if they are fundamentally different. I think
this once again is related to distinguishing between the conceptual
level and the mathematics. Even if two different topologies result
in identical sets of equations, they may look different to us and
it may be sensible to have different explanations for how they
work in causal terms. Mathematics can only describe, it cannot
explain.
wonderfully confused without my help.
Watching this show from the row, not on-stage, I get the impression
that some of the confusion might come from not separating certain
issues at a higher level. Mathematics doesn't know about feedback
at all, so in that sense traderbam is right saying that "the mathematics cannot distinguish between ..."
Feedback is a conceptual thing, which
we use as an aid to understand systems. As soon as we translate
the system to mathematics, the feedback is "lost". Maybe I am wrong
here, but I think the concept of feedback is always based on some
assumption of causality, since we humans tend to like to think of
things in terms of causal realtionships, or at least those of us who
are trained as engineers or natural scientists. This seem related
to the "controversy" between natural scientists and philosophers.
Natural scientists are usually satisfied if they can find a mathematical
model with good predictive power of a phenomenon, while many
philosophers will argue that this does not mean we really understand
the phenomenon and they will ask for or search for other types of
models that "explain" the phenomenon.
The other thing to note is it is always good to try having a precise
terminology for things to avoid confusions and misunderstandings.
Unfortunately, it is very common that no universally agreed upon
terminology exists. We have seen plenty of such examples in
previous threads, the two (or even more) fundamentally different
uses of the term "current feedback" being a good example.
From a practical engineering point of view, at least, it seems to me
(and it seemss some of our famous designer think so too)
that it is useful to distinguish between various types of feedback,
even if it is not clear if they are fundamentally different. I think
this once again is related to distinguishing between the conceptual
level and the mathematics. Even if two different topologies result
in identical sets of equations, they may look different to us and
it may be sensible to have different explanations for how they
work in causal terms. Mathematics can only describe, it cannot
explain.
Christer,
I agree with your overview. Andyc was arguing that if the maths fits a feedback diagram then the circuit must involve feedback. I have demonstrated this is not the case. So as you say the equations do not dictate the topology.
What may be interesting is that the electrons will always obey the mathematics (and vice-versa). So, ultimately, the equations are king. Therefore, whether something is named as feedback or not is not of itself important EXCEPT in so far as someone asserts that is is being used or not used in a particular circuit or that it is detrimental to circuit performance.
I've been encouraging thinking in this regard by challenging assumptions about the definition of feedback, which as you say is a conceptual term, and challenging the assertion that feedback is bad. My aim is to break the perceived connection between the identification of the concept in a circuit and actual circuit performance. I think it is misleading to tell people that feedback is bad or that they should try to remove it from their circuits.
I have observed that the concept of feedback is ill defined in this thread and this has lead to ordinary circuit behaviour being attributed to feedback. Thus implying that everything uses feedback - even a simple resistor - and therefore implying that either feedback has not been defined correctly or that feedback itself has nothing to do with sound quality.
I agree with your overview. Andyc was arguing that if the maths fits a feedback diagram then the circuit must involve feedback. I have demonstrated this is not the case. So as you say the equations do not dictate the topology.
What may be interesting is that the electrons will always obey the mathematics (and vice-versa). So, ultimately, the equations are king. Therefore, whether something is named as feedback or not is not of itself important EXCEPT in so far as someone asserts that is is being used or not used in a particular circuit or that it is detrimental to circuit performance.
I've been encouraging thinking in this regard by challenging assumptions about the definition of feedback, which as you say is a conceptual term, and challenging the assertion that feedback is bad. My aim is to break the perceived connection between the identification of the concept in a circuit and actual circuit performance. I think it is misleading to tell people that feedback is bad or that they should try to remove it from their circuits.
I have observed that the concept of feedback is ill defined in this thread and this has lead to ordinary circuit behaviour being attributed to feedback. Thus implying that everything uses feedback - even a simple resistor - and therefore implying that either feedback has not been defined correctly or that feedback itself has nothing to do with sound quality.
It's not clear...
And from the interesting part of the thread that was removed...
...you prefer a topology not used in your power amplifier designs. Further they use lots of global feedback, and therefore there is no attempt to mimimize its use.
Maybe you mean that amplifiers you own and prefer to listen to minimize or don't use global feedback. Otherwise, feedback is necessary to sell to the general consumer's desire for low distortion specifications. This kind of makes sense. Is that what you mean? (It just seems to me that if no global feedback was so good, it would sell itself.)
JF
john curl said:I would say that feedback is sometimes necessary, but I would prefer not to use it, if possible. I feel the same way about coffee.
And from the interesting part of the thread that was removed...
john curl said:
...you prefer a topology not used in your power amplifier designs. Further they use lots of global feedback, and therefore there is no attempt to mimimize its use.
Maybe you mean that amplifiers you own and prefer to listen to minimize or don't use global feedback. Otherwise, feedback is necessary to sell to the general consumer's desire for low distortion specifications. This kind of makes sense. Is that what you mean? (It just seems to me that if no global feedback was so good, it would sell itself.)
JF
Please remember that John Curl works for two different companies; CTC and Parasound. My understanding is that the Parasound amps use feedback because that is the cheapest way to meet the distortion specs mandated by THX. (I believe that the Parasound directors believe that THX approval makes it easier to sell their products.)
On the other hand, I believe that every quote you pulled out of context was referring to either the CTC products, or else Curl's previous "state-of-the-art" company, Vendetta.
It is my understanding that John used minimal feedback in the past and is headed in the direction of zero feedback. I think his current preamp is zero feedback, but that the current CTC power amp is simply a modified Parasound that has feedback (for the reasons stated above).
On the other hand, I believe that every quote you pulled out of context was referring to either the CTC products, or else Curl's previous "state-of-the-art" company, Vendetta.
It is my understanding that John used minimal feedback in the past and is headed in the direction of zero feedback. I think his current preamp is zero feedback, but that the current CTC power amp is simply a modified Parasound that has feedback (for the reasons stated above).
traderbam said:
I looked through your example, and the one with the resistor 1/gm in series with RL to ground indeed does have the same defining equations as the bottom block diagram of the two I posted earlier. I've also posted a similar picture along the same lines as yours that has the same equations as the top block diagram of the two I previously posted, provided B=1. I think you've made a good point here. And that is that the underlying equations of a system or circuit are not sufficient to establish the existence of feedback in that circuit. I think it would be correct to say that satisfying the equations is necessary but not sufficient. I suspect that some of the information jcx posted has the definitive answer, probably stated in a very formal, complex mathematical way. I don't think the picture I've posted below disqualifies the upper block diagram that I posted earlier from being a feedback system when B=1. Rather, it suggests that more information is necessary to establish whether or not it is truly a feedback system.
Like you, I'd be looking for a simple, concise definition of feedback, but I suspect that no such simple, concise definition exists. I'm not sure it's worth all the effort of precisely defining it anyway.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.