traderbam said:
I think this may be where there is some difference of looking at things. Whenever you use the term "hunt", you seem to be analogizing to something slowly trying to figure something out. This is like the people who get hung up on NFB going around the loop and not being there soon enough to reduce the error. This is a poor analogy to apply to electronics.
By the same token, it is also wrong to assume that things happen instantaneously. But there the matter is one of degree. Real circuits are not instantaneous, but they can be lightning fast compared to the signal at hand.
I think you have unwittingly put your finger on why Speed is King in the EC circuit. The higher the speed of the circuit compared to the signals being considered, the closer that circuit comes to the ideal - but we all agree that it never quite gets there. It gets a lot closer at 1 kHz than at 20 kHz, for sure. But, nevertheless, it gets VERY nicely close at 20 kHz, enough to yield a partial error cancellation by a factor of about 30 times. Perhaps this way of looking at it can reconcile the differing views.
Cheers,
Bob
Jan wrote:
Here's how. The loop gain is the change in output voltage for a change in feedback voltage (where the feedback connects to the output). Vp and Vq in my diagram. This ratio is the feedback loop gain. This determines everything about the correction power and stability of the amplifer.
Until you stop lauding the greatness of your novel invention and understand how it works you will overlook opportunities to make it work better.
I'm not envious. I'm all for creativity. I just see you making some bad assumptions that will trip you up in the future.
Yes.
If you are designing an error correction system and building amps based on it and then taking them to shows you sure better figure out what your real feedback loop gain is.
Here's how. The loop gain is the change in output voltage for a change in feedback voltage (where the feedback connects to the output). Vp and Vq in my diagram. This ratio is the feedback loop gain. This determines everything about the correction power and stability of the amplifer.
Until you stop lauding the greatness of your novel invention and understand how it works you will overlook opportunities to make it work better.
Bob wrote:
Hi Bob. "Hunting" is a technical term. It says nothing about speed. It is a commonplace descriptor for both electronic and mechanical feedback control systems of any speed. Speed is relative, right? So to a photon even your output stage is working in slow motion.
Yes. The correction power of any feedback system depends on stability and loop gain and loop phase (speed). Same for all feedback systems.
traderbam said:
I think this is the crux of our difference of opinion. I understand this that you consider breaking the connection between the output and the feedback summer as 'breaking the feedback loop'.
I consider breaking the connection from the feedback summer to the input summer as breaking the feedback loop.
Is that a fair summary of the discussion?
Jan Didden
That would explain it.
Find the node in the circuit whose voltage you are trying to control. This node is the centre of your error correction world. You need to measure the sensitivity of the voltage at this node to changes in the feedback signal from this node (which is like making the amplifer see an error at its output), with the input voltage held constant. This is the feedback loop transfer function. From this you can measure the feedback gain and phase and also determine the stability margin. This tells you how much the distortion of the amplifier will be reduced at different frequencies.
Find the node in the circuit whose voltage you are trying to control. This node is the centre of your error correction world. You need to measure the sensitivity of the voltage at this node to changes in the feedback signal from this node (which is like making the amplifer see an error at its output), with the input voltage held constant. This is the feedback loop transfer function. From this you can measure the feedback gain and phase and also determine the stability margin. This tells you how much the distortion of the amplifier will be reduced at different frequencies.
janneman said:
Hi Jan,
"Is that a fair summary of the discussion"? No, as you ignore the far reaching implications of breaking the loop at the right place.
As for "If you want to break the loop, do it where it effects the system", I did just that, that is, where it effects the system the most.
Cheers, Edmond.
traderbam said:Bob wrote:
Hi Bob. "Hunting" is a technical term. It says nothing about speed. It is a commonplace descriptor for both electronic and mechanical feedback control systems of any speed. Speed is relative, right? So to a photon even your output stage is working in slow motion.
Yes. The correction power of any feedback system depends on stability and loop gain and loop phase (speed). Same for all feedback systems.
In my day job I have to work with guys involved in optics quite a bit. Many of them have a physics background. As an EE, I don't have nearly as much formal training in optics as they do, so I sometimes look at a problem differently than they do. I've seen this in other inter-disciplinary endeavors as well, and it often explains differences in understandings. This, by the way, is not necessarily a bad thing - sometimes it promotes needed looking at multiple angles of a problem. I'm curious, what area is your formal training in?
Cheers,
Bob
Bob,
I've sent you an email. I'm an EE too, but educated in the UK so the terminology may be different. I couldn't agree more with your comments about the value of cross-disciplinary insight. My degree focussed on electronics, both analogue and digital, but also included mechanical systems, electrical machines, control theory, structural engineering, information theory and some other stuff. Being able to draw analogies among disciplines, particularly between mechanical and electrical systems has been really valuable. One begins to see commonalities. I think the most valuable learning is how to go about analyzing a problem with open mind and trying to avoid making bad assumptions: question everything.
My passion is to make the best sounding audio with available technology. In my experience, the audio challenge is as much a specification problem as an implementation one. That is, the audio industry is riddled with clever designers whose focus is on misplaced objectives, often based on good theories for closely related areas which comprise assumptions that are not correct for audio. Much grasping at straws is the result. Even denial of the ability of people to judge sound quality! The near absence of decent measurement methods is a serious problem.
A regular change of perspective is invaluable.
Brian
I've sent you an email. I'm an EE too, but educated in the UK so the terminology may be different. I couldn't agree more with your comments about the value of cross-disciplinary insight. My degree focussed on electronics, both analogue and digital, but also included mechanical systems, electrical machines, control theory, structural engineering, information theory and some other stuff. Being able to draw analogies among disciplines, particularly between mechanical and electrical systems has been really valuable. One begins to see commonalities. I think the most valuable learning is how to go about analyzing a problem with open mind and trying to avoid making bad assumptions: question everything.
My passion is to make the best sounding audio with available technology. In my experience, the audio challenge is as much a specification problem as an implementation one. That is, the audio industry is riddled with clever designers whose focus is on misplaced objectives, often based on good theories for closely related areas which comprise assumptions that are not correct for audio. Much grasping at straws is the result. Even denial of the ability of people to judge sound quality! The near absence of decent measurement methods is a serious problem.
A regular change of perspective is invaluable.
Brian
Pavel,
Yes, the sign of 'a' will flip the sign of the feedback as you say. The size of 'a' will also affect the sign:
Vp = N{ -a.Vq / (1 - a) }
[I have to make an assumption about N (the amplifer characteristic we are trying to correct) because it could be anything. Let's suppose that it is fairly linear so that the sign of N{x} is the same as the sign of x.]
If Vp is negative with respect to Vq you have negative feedback and vice-versa. You'll see that when a < 1, Vp is negative. When a > 1 Vp becomes positive. When a = 1 the gain becomes infinite.
Remember this is an idealistic math model. Real "EC" circuits are only an approximation to this to prevent oscillation.
Yes, the sign of 'a' will flip the sign of the feedback as you say. The size of 'a' will also affect the sign:
Vp = N{ -a.Vq / (1 - a) }
[I have to make an assumption about N (the amplifer characteristic we are trying to correct) because it could be anything. Let's suppose that it is fairly linear so that the sign of N{x} is the same as the sign of x.]
If Vp is negative with respect to Vq you have negative feedback and vice-versa. You'll see that when a < 1, Vp is negative. When a > 1 Vp becomes positive. When a = 1 the gain becomes infinite.
Remember this is an idealistic math model. Real "EC" circuits are only an approximation to this to prevent oscillation.
Brian,
Let me pick your brain a bit more, if you allow me. In your post 1281(!) you posted the circuit with the fb loop broken where I think it should be broken: between the output of the fb network and the input summer. You label the former Vx, the latter Vy. You provided an equation for the amp transfer function.
If I calculate in this situation the loop transfer function, I get: Vx=a.(Vn-N.Vn) where Vn = Vy, so we get:
Vx=a.Vy.(1-N). Can you agree to this? If we set a=1 than it degenerates to Vx=Vy.(1-N), right?
Jan Didden
Let me pick your brain a bit more, if you allow me. In your post 1281(!) you posted the circuit with the fb loop broken where I think it should be broken: between the output of the fb network and the input summer. You label the former Vx, the latter Vy. You provided an equation for the amp transfer function.
If I calculate in this situation the loop transfer function, I get: Vx=a.(Vn-N.Vn) where Vn = Vy, so we get:
Vx=a.Vy.(1-N). Can you agree to this? If we set a=1 than it degenerates to Vx=Vy.(1-N), right?
Jan Didden
Nit-picking is good.
The notation is going a bit awry in all this which doesn't help.
If you define N(Vy) = k.Vy where k is a constant then N is a linear function.
Also, N(x + y) = N(x) + N is only true if N is a linear function.
The 'a' block is linear. So in 'function' notation we can say A(x + y) = A(x) + A. We have defined A(x) = a.x and so A(Vy - N{Vy}) = A(Vy) + A(-N{Vy}) = a.Vy - a.N{Vy}. No brackets needed.
Sorry, I just noticed I made an error in post 2132.
Vx = a.Vy - a.N{Vy}
The notation is going a bit awry in all this which doesn't help.
If you define N(Vy) = k.Vy where k is a constant then N is a linear function.
Also, N(x + y) = N(x) + N
is only true if N is a linear function.The 'a' block is linear. So in 'function' notation we can say A(x + y) = A(x) + A
. We have defined A(x) = a.x and so A(Vy - N{Vy}) = A(Vy) + A(-N{Vy}) = a.Vy - a.N{Vy}. No brackets needed. Sorry, I just noticed I made an error in post 2132.Vx = a.Vy - a.N{Vy}
traderbam said:Nit-picking is good.
The notation is going a bit awry in all this which doesn't help.
If you define N(Vy) = k.Vy where k is a constant then N is a linear function.
Also, N(x + y) = N(x) + N
The 'a' block is linear. So in 'function' notation we can say A(x + y) = A(x) + A
Vx = a.Vy - a.N{Vy}
But I don't define N(Vy)=k.Vy. Vx is the dependent variable, and Vx=a.[Vy-N.Vy] So, if N is nonlinear, Vx becomes a non-linear result, no? Or else, I just delete the a-block, and I get Vx=Vy-N.Vy, no?
Jan Didden
traderbam said:
Thanks, Brian.
I agree with your comments about the audio industry. I suppose that is part of what makes it fascinating and frustrating at the same time. There are an enormous number of claims out there that beg to be questioned. Some of those claims may have merit in what otherwise might be characterized as a sea of snake oil. Sorting out which claims and approaches have merit is truly a challenge. Add to that the fickle finger of psychology and you have quite a mix to deal with.
Perhaps you could elaborate more on your reference to a near-absence of decent measurement methods.
Cheers,
Bob
Bob,
There ought to be a permanent thread entitled "measuring sound quality".
In my flippant remark I didn't mean to say that all measurements are poor. Of course there are lots of very sophisticated measurements of specific things, like THD and TIM and slew rate and so on, that are very good as measurments. What I have observed is that each time a new measure is invented there is a big design effort to optimise that measure and you get a bunch of new products being reviewed in Stereophile. However, the correlation between these new products and improved sound quality is loose, in general. The measures are not hitting the nail on the head.
What do you think?
There ought to be a permanent thread entitled "measuring sound quality".
In my flippant remark I didn't mean to say that all measurements are poor. Of course there are lots of very sophisticated measurements of specific things, like THD and TIM and slew rate and so on, that are very good as measurments. What I have observed is that each time a new measure is invented there is a big design effort to optimise that measure and you get a bunch of new products being reviewed in Stereophile. However, the correlation between these new products and improved sound quality is loose, in general. The measures are not hitting the nail on the head.
What do you think?
traderbam said:
Yes it's a non-linear function. Brian, I understand your remark about the a.N=k.N, I'm just struggling with why it would be applicable here. I mean, if in a real amp I break the loop at x-y, and insert a signal at Vy, I get a certain Vx which is a non-linear function of Vy, times a. Now you seem to say that the math is illegal and it couldn't happen. If the a bothers you, delete the a, replace by a wire. Don't tell me that then Vx=Vy-N.Vy is illegal!
Earlier you said about an equation that 'its not implementable in practise'. Yet many have implemented it in practise. That doesn't seem to jive.
You also made a statement that pfb leads to infinite gain. It only does so if there is loop gain to support it. I don't need to tell you that an oscillator only oscillates with a loop gain of one or more, and that the stabilization of oscillation amplitude occurs at the point where the loop gain stabilizes at exactly one.
My point is that if you break the loop at x-y you end up with a loop gain of 1-N, which can only lead to pfb if N<1, but then the loop gain is too low to support any 'infinite gain'.
You also said earlier: if N>1 it's nfb, with N<1 its pfb, with N=1 its infinite gain. I agree with 2 out of 3. The math as well as practise shows clearly that when N=1 there's no fb at all because Vx=Vy.
I really, really would like to know where my reasoning goes wrong, if it does.
Edit: Brian, wwe cross-posted, but I think it comes out allright. For a change
Jan Didden