janneman said:
Hi Jan,
Care to specify the components, so we can spice it?
Good luck at the festival!
Cheers, Edmond.
I also have posted an article I found, exploring the issues of output coils, zobels etc. Interesting reading:
http://www.linearaudio.nl/Documents/amp-spkr interface.pdf
Jan Didden
http://www.linearaudio.nl/Documents/amp-spkr interface.pdf
Jan Didden
G.Kleinschmidt wrote:
I still haven't found a common definition of "error correction" topology that differentiates it from "conventional" NFB. I've seen a description that says that the open loop gain of the amplifer before applying "EC" should be unity. But then when the "EC" circuit is show in situ it is clear that the forward path of the original circuit has been modified: it now incorporates a large forward gain block created using a positive feedback loop.
So you could say "EC" means taking a unity gain amplifer, inserting a PFB loop in it's forward path and then applying unity gain NFB.
But this is a narrow definition based on applying the "EC" to a unity gain emitter/source follower output stage.
If I were asked to make a general "EC" definition, based on the common threads within Hawksford and Cordell and Janneman and Rodolfo and other peoples' ideas, I would would say something like this: "it is a method of applying an overall NFB loop to an amplifier having first boosted the amplifiers gain by inserting a PFB loop into it".
The question then becomes "is it better to generate forward OL gain using gain devices or is it better to do it using a PFB loop?". I would concentrate on this question with your particular amplifer circuit in mind. Then you can be more creative about how you apply gain elements, NFB loops and PFB loops to optimise the performance. IOW think beyond the particular topology used for an emitter follower output stage when considering a complete amplifier.
I still haven't found a common definition of "error correction" topology that differentiates it from "conventional" NFB. I've seen a description that says that the open loop gain of the amplifer before applying "EC" should be unity. But then when the "EC" circuit is show in situ it is clear that the forward path of the original circuit has been modified: it now incorporates a large forward gain block created using a positive feedback loop.
So you could say "EC" means taking a unity gain amplifer, inserting a PFB loop in it's forward path and then applying unity gain NFB.
But this is a narrow definition based on applying the "EC" to a unity gain emitter/source follower output stage.
If I were asked to make a general "EC" definition, based on the common threads within Hawksford and Cordell and Janneman and Rodolfo and other peoples' ideas, I would would say something like this: "it is a method of applying an overall NFB loop to an amplifier having first boosted the amplifiers gain by inserting a PFB loop into it".
The question then becomes "is it better to generate forward OL gain using gain devices or is it better to do it using a PFB loop?". I would concentrate on this question with your particular amplifer circuit in mind. Then you can be more creative about how you apply gain elements, NFB loops and PFB loops to optimise the performance. IOW think beyond the particular topology used for an emitter follower output stage when considering a complete amplifier.
traderbam said:
I agree that it is really easy for us to get fouled up in the semantics of error correction, but of course the result is more important than what we call it. I agree that it can be argued that some license has been taken semantically in some cases as well.
I don't lose a lot of sleep over it, but my intuitive feel for what reasonably can be called error correction is a process that produces a moderately deep null in distortion as a parameter is changed, rather than an increasing reduction.
For example, feedforward error correction, where a correcting signal is somehow derived and then somehow "subtracted" from the output signal seems to very comfortably fit the definition of error correction.
Similarly, conventional negative feedback, where the error is reduced as the amount of NFB is increased, seems clearly not to be error correction.
Hawksford Error Correction is in the murky middle because it can be looked at in more than one way. Indeed each of those ways have their value in different analytical situations. Although one can certainly argue that it is just NFB with an internal PFB loop providing a gain of infinity, I am still comfortable calling it a form of error correction because you can twiddle a parameter and get a null. I still remember how excited I was the first time I built it and twiddled that pot!
Cheers,
Bob
Bob Cordell said:
Definitely a different kind of beast, fully agree.
Been there too !!!!
The fact that in abstract NFB and EC end indistiguishable as global transfers, does not mean they are the same.
I have suggested earlier a completely different architecture, namely a unity gain low level signal summing node optimized for linearity and bandwidth, working as the correction engine for the whole.
This leads to completely different toplogies in comparison with the cannonical Diff-VAS-Out Op Amp paradigm.
The EC summing node could even be implemented digitally as long as the bandwidth can be met, recall the correction factor (counterpart to the classic OL/CL gain ratio) is in this case determined to a good approximation by the ratio of EC engine 3dB cutoff to overall desired 3dB bandwidth.
Leaves some room to think creatively I assume ....
Rodolfo
Hi Bob,
At the risk of opening this can again...
It is not possible for a circuit to "cancel" its output error by means of subtracting its output from its input. Only reduction is possible. That's a law of mathematics if you like.
When you adjust that pot in your circuit you are not seeing a "null" in the sense of a cancellation of errors, rather you are seeing a minima. You are finding a minimum distortion setting. This is the setting where the PFB forward gain is at a maximum and thus the NFB loop gain is at a maximum. The fact that there is a minima rather than a monotonic change is just a side-effect of the topology and is of little consequence.
It turns out that adjustment first increases the PFB gain and then past a certain point it reduces it again. Thus giving a minimum distortion setting. This should not be confused with a cancellation. Indeed, you said yourself that you measured the distortion reduction at this minima setting was about 30dB and I showed in simulation that the PFB gain was very close to 30dB at its peak.
In a conventional NFB arrangement one doesn't see this "minima" feature because no one bothers to adjust the forward gain up and down. It serves no useful purpose. Instead, the effort goes into crafting the forward gain roll-off. Indeed, compensation is usually intended to allow as much forward gain as possible to maximize the NFB.
One could define "EC" as being "a NFB loop around an amplifer whose forward gain is boosted by means of an adjustable PFB loop" if that would capture it better.
At the risk of opening this can again...
It is not possible for a circuit to "cancel" its output error by means of subtracting its output from its input. Only reduction is possible. That's a law of mathematics if you like.
When you adjust that pot in your circuit you are not seeing a "null" in the sense of a cancellation of errors, rather you are seeing a minima. You are finding a minimum distortion setting. This is the setting where the PFB forward gain is at a maximum and thus the NFB loop gain is at a maximum. The fact that there is a minima rather than a monotonic change is just a side-effect of the topology and is of little consequence.
It turns out that adjustment first increases the PFB gain and then past a certain point it reduces it again. Thus giving a minimum distortion setting. This should not be confused with a cancellation. Indeed, you said yourself that you measured the distortion reduction at this minima setting was about 30dB and I showed in simulation that the PFB gain was very close to 30dB at its peak.
In a conventional NFB arrangement one doesn't see this "minima" feature because no one bothers to adjust the forward gain up and down. It serves no useful purpose. Instead, the effort goes into crafting the forward gain roll-off. Indeed, compensation is usually intended to allow as much forward gain as possible to maximize the NFB.
One could define "EC" as being "a NFB loop around an amplifer whose forward gain is boosted by means of an adjustable PFB loop" if that would capture it better.
traderbam said:
What you are saying here is certainly true, and has been said many times before. Like I said, I don't get hung up over semantics.
BTW, the null I mentioned was at 20 kHz, where phase imperfections are already starting to take their toll. The null is much, much deeper at, say, 1 kHz. No real circuit, even a "legitimate" feedforward error correction circuit, will have a truly perfect null. If you want to fiddle with the difference in wording between "null" and "minima" I'm OK with that.
Cheers,
Bob
In Brian's World feedback systems reduce errors by continuously hunting for a minimum error equilibrium whereas feedforward systems correct errors by cancellation. I agree that neither is perfect at eliminating errors in the analogue world.
I classify the "Hawksford EC" as a feeback system. The idea of using PFB to generate gain is inventive. It's an old idea that predates Hawksford. I think the real advantage of it is in implemenation convenience in the case of symmetrical follower output stages, which is no small advantage and so it's well worth understanding the thing.
If any of you are thing of extending "EC" to other types of circuit I would consider two less obvious characteristics:
1) In the exact opposite way that a NFB loop reduces the distortion caused by the components within its loop, a PFB loop increases it. So it is vital to make the parts within the PFB loop as linear as possible.
2) A PFB loop is inherently unstable. So more care has to be exercised with the frequency compensation than one might expect. I found in my simulations of an output stage using "EC" that it has poorer reactive load tolerance than a NFB circuit using a conventional gain block.
Brian
I classify the "Hawksford EC" as a feeback system. The idea of using PFB to generate gain is inventive. It's an old idea that predates Hawksford. I think the real advantage of it is in implemenation convenience in the case of symmetrical follower output stages, which is no small advantage and so it's well worth understanding the thing.
If any of you are thing of extending "EC" to other types of circuit I would consider two less obvious characteristics:
1) In the exact opposite way that a NFB loop reduces the distortion caused by the components within its loop, a PFB loop increases it. So it is vital to make the parts within the PFB loop as linear as possible.
2) A PFB loop is inherently unstable. So more care has to be exercised with the frequency compensation than one might expect. I found in my simulations of an output stage using "EC" that it has poorer reactive load tolerance than a NFB circuit using a conventional gain block.
Brian
Hi Brian,
When you complete the loop with feedback, it is very similar to electrical-mechanical systems. The servo (your amplifier with feedback) will not tolerate zero error. It will create an error so it doesn't hunt (assuming here it's stable). A mechanical system is the same, just slower. A servo (or feedback system) will not operate without some kind of error signal.
This isn't bad depending on the magnitude of the error. It's the devil you know and will keep the error to some value or below. Amplifiers can not operate without some error. Even chopper stabilized DC amps need some system to null the DC output error.
-Chris
From my own experience, I agree 100%.
When you complete the loop with feedback, it is very similar to electrical-mechanical systems. The servo (your amplifier with feedback) will not tolerate zero error. It will create an error so it doesn't hunt (assuming here it's stable). A mechanical system is the same, just slower. A servo (or feedback system) will not operate without some kind of error signal.
This isn't bad depending on the magnitude of the error. It's the devil you know and will keep the error to some value or below. Amplifiers can not operate without some error. Even chopper stabilized DC amps need some system to null the DC output error.
-Chris
traderbam said:
Well, I don't know about your mathematics, but if I you do the sums in ec it actually comes out at zero. It may be a 'law of practise' that you can't really quite get there, but in the maths you can.
I also have to take exception to 'ec is just pfb in disguise'. I have tried to make the case that in any feedback system you need to look at the effective feedback factor. That's what you learn in 'control systems 101' as well: manipulate a complex feedback scheme until you have a single expression that gives you insight in how the system behaves.
When you do that with ec, you will get a fb factor that actually has 'A', the forward gain, in it. So the effective fb factor varies with the ol gain. It can become pfb or nfb as is required to bring the cl gain to the design value, but it is not by definition pfb and certainly not huge (assuming your ol gain is reasonably close to the cl gain, as in a follower stage). In an earlier post we had a calculation for ec around a gain of 0.95 stage which showed a fractional dB fb factor.
Jan Didden
Only if you have excessive loop gain! As I stated before, the fb factor in an ec loop can turn either to pfb or nfb depending on the ol gain versus the cl gain. In case the fb factor turns pos it is because the ol gain is less than the design gain. There is no excess gain, there is no instability. Bob's and many other amps are living proof.
Think about the situation with classic nfb. Why do we want to roll off the gain to below 1 before the dreaded 180-degree phase shift? Because, if we would still have loop gain at 180 deg shift (which of course is just pfb) we get instability. But, that pfb is still there, the phaseshift that makes nfb into pfb is still there, but because we rolled of the gain, we don't get oscillations.
The same with ec. If the effective fb factor turns pos, it is because the loop gain is less than the design gain. No excess gain, no instability.
Jan Didden
Holy Grail anyone?
I wrote:
Hawksford's algebraic manipulation is incorrect unless infinite gain is assumed. Unfortunately, he fails to explain this which is rather naughty of him. Perhaps he thought it was obvious.
I wrote:
Jan, There is no stronger argument I can make. In a feedback system, zero error is only possible with infinite, stable loop gain. If you do not accepted this then I am not the right person to have a debate with.
Hawksford's algebraic manipulation is incorrect unless infinite gain is assumed. Unfortunately, he fails to explain this which is rather naughty of him. Perhaps he thought it was obvious.
Chris wrote:
The common op-amp is designed with integration in mind. The gain vs f is an exponentional curve for most of the useable frequency range. They don't have inifinite gain at dc but it's pretty big.
Yes, the same behaviours are seen in mechanical systems. And I agree that there will always be some error as the control system attepts to track changes in the input and output and minimize them. I should just point out that the output error can reduce to zero (ignoring noise) in theory if the system includes a "pure" integator and the input is "steady state". IOW if your input is unchanging then the hunting will decay with time until there is immeasurable error at the output.
The common op-amp is designed with integration in mind. The gain vs f is an exponentional curve for most of the useable frequency range. They don't have inifinite gain at dc but it's pretty big.
Just different approach to Error Correction. It does not necessarily has to be the Hawskford one.
Oh sure. This is a great skill of John Curl. He seems to be quite The Man at finding ways to cancel out non-linearities by cleverly combining transistors. Eg: If you had an OL gain y=x^2 then you could insert a new gain block with gain y = 1/x to linearise it. Not easy to do in practice but the best way to fix things.
Re: Holy Grail anyone?
But I do agree that with 'classic' nfb zero distortion is not possible in theory, even with ideal components in the fb loop. Lipschitz and Vanderkooy showed that many years ago. However, I do think that with ec zero distortion in theory is possible.
Huh? That's news for me. I always thought that he explicitly assumed non-infinite gain, even gain close to the design cl gain. Care to elaborate?
Jan Didden
traderbam said:
But I do agree that with 'classic' nfb zero distortion is not possible in theory, even with ideal components in the fb loop. Lipschitz and Vanderkooy showed that many years ago. However, I do think that with ec zero distortion in theory is possible.
traderbam said:
Huh? That's news for me. I always thought that he explicitly assumed non-infinite gain, even gain close to the design cl gain. Care to elaborate?
Jan Didden
janneman said:
Hi Jan.,
I'm sorry to say so, but may I remind you about our discussion last month:
http://www.diyaudio.com/forums/showthread.php?postid=1282575#post1282575
I thought I had clearly pointed out that the loop gain depends on how it is defined, that is, where the loop has been broken. In order to examine stability issues, the only correct place is right at the output, case #3 in my diagram, where the loop gain reaches infinity. So I disagree with your statement: "There is no excess gain, there is no instability". If that were the case, why has Bob applied frequency compensation (C6,C7,R36,R37) in his EC stage? Without compensation this stage is definitely unstable, even without front-end and/or the NFB loop (R12). Looking at the loop in "my way" makes perfectly clear that an EC circuit isn't so "innocent" as you might think.
Also, what is this "effective feedback factor"? Open loop gain divided by close loop gain? As far as concerning stability issues, it's a completely useless metric, as it is not the same as the loop gain and phase response of a feedback path.
Cheers, Edmond.