gl said:
Perhaps it will amuse you to hear the story, and it's not too off
topic as it has to do with an ancient EC design.
You may recall Michael Dayton Wright as the designer of the XG8
electrostatic loudspeakers that used Sulfur Hexaflouride gas to
raise their stator potentials to 10 KV.
Mike also designed a high power Error Correction amp which
featured a 1000 watt Class AB amplifier with a single-ended
output which had an inverting feedback loop, and thus a
virtual ground at the minus input where he could sense the output
error. This voltage was amplified by a high quality Class A
amplifier whose output looked at the other side of the speaker
load, creating an opportunity for cancellation.
At least in theory.
Dayton Wright the company had been acquired by Leigh
Industries, all this around Toronto, and someday Madrigal president
Mark Glazier and future Krell designer Dan D'Agostino were
employed in sales and marketing. Apparently Mike Wright was no
longer associated with the operation.
Through the efforts of Dan, I was flown up to Mississauga to take
a look at the amplifier which didn't work. Legend had it that the
amplifier had worked for a short period of time and then failed.
It was large and handsome (in a sort of ugly way) and the heat
sinks looked like the hot water radiators used in New York
apartments for heat. It had an incredible number of parts inside.
I was agreeable to doing something with it, and was ushered
into the office of the president of Leigh Industries, who proceeded
to say, "Forget that. How about if you sell me your company?
You give me the company for free, and I'll give you an ongoing
paycheck."
This was not the offer I was looking for, so Dan and Mark took
Rene and I out on the town to experience the night life of
Mississauga, with as much of Leigh's money as Dan could spend
in one night.
The next day we flew home, and to my knowledge nobody ever
fixed that amplifier. I have a photo of it somewhere, and if I ever
find it, I'll post it. D'Agostino, Glazier and I went on to have
fruitful relationships for some years afterword until they both
found the dark side
How's that for a nice bedtime story?
traderbam said:
Bob, are you some sort of comedian????
One requires gallons of NFB and the other uses none at all and you call these merely "caveats"!!!
Good grief!
I'm perfectly serious.
Unfortunately, you don't get it. You are completely blinded by your myopic focus on only one way of looking at error correction. I'm sorry I can't help you, but others here can grasp the point.
Bob
Nelson,
Thank you for the story. It fills in some gaps. I had my XG8's updated by Leigh during that period and had a good experience with them.
It also places Mark Glazier in Canada at that point in time which raises a question. I knew a Mark Glazier who owned and operated a high end audio store in Montreal in the late '70's and was the local DW dealer. Could they be one and the same?
The deal you were offered by Leigh is what I've always heard called the "magic bean deal" where someone offers you a bag of magic beans for your cow. I find it interesting that the largest promoter of those types of deals in my (direct) experience during the '80's and '90's was DynaTech. I don't know the details of the final Threshold sale but there seems to be a certain irony here.
Graeme
Thank you for the story. It fills in some gaps. I had my XG8's updated by Leigh during that period and had a good experience with them.
It also places Mark Glazier in Canada at that point in time which raises a question. I knew a Mark Glazier who owned and operated a high end audio store in Montreal in the late '70's and was the local DW dealer. Could they be one and the same?
The deal you were offered by Leigh is what I've always heard called the "magic bean deal" where someone offers you a bag of magic beans for your cow. I find it interesting that the largest promoter of those types of deals in my (direct) experience during the '80's and '90's was DynaTech. I don't know the details of the final Threshold sale but there seems to be a certain irony here.
Graeme
ingrast said:
Hi Rodolfo,
The only reason I did it that way was to demonstrate a fair apples-to-apples comparison between error feedback (HEC) and error feedforward. In other words, I think it is very significant that if the bandwidth limitations are the same in the error-derivation path, then the feedback and feedforward approaches demonstrate the same error correction behavior.
This demonstration bolsters the point about the error correction view of HEC being legitimate and in fact sharing many things in common with error feedforward.
Now, of course, this is a demonstration in simulation using ideal summing elements.
In practice, there is absolutely no constraint on placing that bandwidth restriction in the error derivation path of the feedforward architecture. This, in principle, would lead one to believe that feedforward error correction can do better. Indeed, I could easily demonstrate that in simulation by removing that filtering in the feedforward case.
Unfortunately, feedforward error correction has other big problems of its own, most significantly how you add in the error correction AFTER the power output stage without other serious penalties.
Cheers,
Bob
andy_c said:
Sounds right!
I think you could make a good case that I'm overemphasizing the importance of maximizing the dynamic range of the EC. But the idea I had for my design was, in simulation, to minimize the distortion of the output stage itself under very high current conditions (say, 100A peak). So it was just a way to try to make the EC as robust as possible.
I like it. I think it is very worthwhile to be mindful of the dynamic range required of the EC circuit, and ways to mitigate that need.
Thanks,
Bob
Bob Cordell said:
Yes, this is the big issue, one can in principle extract and cleanly power up the error (we are talking orders of magnitude below nominal, so fractions of a watt), but it must be added passively while looking into a complex load and of course not applying further feedback (then it should be regular EC / NFB).
With respect to the X feedforward I thought yesterday on your same lines at first, was even about to post, when I realized it is in fact legitimate.
To make sense in the implementation, one extracts the error from one of the arms (output minus input) and amplifies it through the other arm.
Now, when one extracts a new error from this latter arm, the former error injected at input will be only slightly distorted (like the signal itself), meaning the related component in the new error to inject back in the first arm input will be highly attenuated, meaning the feedback loop gain is in fact extremely low and does not really qualify as such.
So in the end it is a clever implementation of feedforward with the only caveat the error itself is added by means of the same power amplifier. In my mind it does not look as a fatal flaw if one notes the error signal is only slightly degraded, which should normally mean orders of magnitude improvement against nothing at all.
Wish I had some spare time to simulate .... any takers?
Rodolfo
gl said:
Mark worked at a high end store prior to this episode (with Dan,
I believe), and joined Threshold in '77.
We sold Threshold to Dynatech in '87, however there was actual
money involved, and in addition they paid me a very large salary
for the following 4 years - no complaints about Dynatech.
ingrast said:
Yes, this is the big issue, one can in principle extract and cleanly power up the error (we are talking orders of magnitude below nominal, so fractions of a watt), but it must be added passively while looking into a complex load and of course not applying further feedback (then it should be regular EC / NFB).
With respect to the X feedforward I thought yesterday on your same lines at first, was even about to post, when I realized it is in fact legitimate.
To make sense in the implementation, one extracts the error from one of the arms (output minus input) and amplifies it through the other arm.
Now, when one extracts a new error from this latter arm, the former error injected at input will be only slightly distorted (like the signal itself), meaning the related component in the new error to inject back in the first arm input will be highly attenuated, meaning the feedback loop gain is in fact extremely low and does not really qualify as such.
So in the end it is a clever implementation of feedforward with the only caveat the error itself is added by means of the same power amplifier. In my mind it does not look as a fatal flaw if one notes the error signal is only slightly degraded, which should normally mean orders of magnitude improvement against nothing at all.
Wish I had some spare time to simulate .... any takers?
Rodolfo
Thanks for the insight. This sounds quite plausible. I wonder if someone can come up with a mathematical analysis that can confirm that is exact, or just a very good approximation
Cheers,
Bob
Re: Re: Some summary views on EC
Hi Rodolfo,
I apologize for missing this post earlier.
Yes, I agree that your diagram represents the third view of HEC, assuming that you are referring to Figure 3 and that we view the "feedback" as the signal emanating out of the block labeled "B". Summer S2 derives the error and Block B is a gain block that determines how much of that error is injected back into the input stream at Summer S1.
If we break the feedback loop between B and S1, we see a loop gain that is small and dependent on the loss in Block A. This loop gain can be positive or negative. It will normally be a smidge positive. I believe this view is like that of Jan. In his words, the varying loss in A automagically adjusts the loop gain in such a way as to render the desired right result.
This view is there only for completeness, and I think is an adequate articulation of Jan's view (please forgive me and correct me if I am wrong, Jan).
Cheers,
Bob
ingrast said:
Hi Rodolfo,
I apologize for missing this post earlier.
Yes, I agree that your diagram represents the third view of HEC, assuming that you are referring to Figure 3 and that we view the "feedback" as the signal emanating out of the block labeled "B". Summer S2 derives the error and Block B is a gain block that determines how much of that error is injected back into the input stream at Summer S1.
If we break the feedback loop between B and S1, we see a loop gain that is small and dependent on the loss in Block A. This loop gain can be positive or negative. It will normally be a smidge positive. I believe this view is like that of Jan. In his words, the varying loss in A automagically adjusts the loop gain in such a way as to render the desired right result.
This view is there only for completeness, and I think is an adequate articulation of Jan's view (please forgive me and correct me if I am wrong, Jan).
Cheers,
Bob
Re: Re: Re: Some summary views on EC
True, and Jan coined the term "feedback on demand" which adequately describes the error signal injected at the system input. I like to look - from this viewpoint - EC sort of "big brother" wielding a large hammer. As long as the amplifier behaves, it sits quietly there, but the slightest provocation makes it step in with a vengeance.
Seriously now, bottom line is the overall system characterization and results.
It is to be expected for any system that proposes a near perfect correction, that there must be at hand some large asset and we are used to identify this asset with gain. Eq. [3] undersores what others have loudly voiced and I did never deny, that EC is equivalently a form of NFB where gain is attained by alternate means, namely positive feedback. What I have found irritating is the insistence in qualifying positive feedback as a dead end not worth further consideration.
In my mind there is nothing wrong pursuing this way as long as workings are clearly understood, i.e. don't try to make a unity gain summing node with an OpAmp for that way one is erasing with one hand what is being written with the other.
Rodolfo
Bob Cordell said:
True, and Jan coined the term "feedback on demand" which adequately describes the error signal injected at the system input. I like to look - from this viewpoint - EC sort of "big brother" wielding a large hammer. As long as the amplifier behaves, it sits quietly there, but the slightest provocation makes it step in with a vengeance.
Seriously now, bottom line is the overall system characterization and results.
It is to be expected for any system that proposes a near perfect correction, that there must be at hand some large asset and we are used to identify this asset with gain. Eq. [3] undersores what others have loudly voiced and I did never deny, that EC is equivalently a form of NFB where gain is attained by alternate means, namely positive feedback. What I have found irritating is the insistence in qualifying positive feedback as a dead end not worth further consideration.
In my mind there is nothing wrong pursuing this way as long as workings are clearly understood, i.e. don't try to make a unity gain summing node with an OpAmp for that way one is erasing with one hand what is being written with the other.
Rodolfo
Further reflections
How does one reconcile the fact there is an infinite loop gain under ideal conditions, yet nothing in the math explicitely points to singularities? How is it that actual amplifiers are built and not only do not explode but a nice "distortion nulling" pot allows smooth adjustment?
This can be better understood if one looks for a moment only at the loop formed by S1-S-S2-B. Under ideal and cancellation conditions, the path gain is exactly unity and real. It is equivalent to a ball set on a flat floor, as long as it does not hit constraints (walls or supply rails for the circuit) it stays put or keeps happily rolling, no black holes here.
In reality one cannot make a unity gain loop without some form of active device in between, which automatically places at least one pole in the transfer. I represented this fact with amplifier S, and if one makes the numbers for an arbitrary first order model, the equivalent transfer from S1 + input to S output turns out to be an ideal integrator with the same gain-bandwidth product as the bare amplifier S. From there on, it is bussiness as usual from the standpoint of correction, stability and so forth.
Rodolfo
How does one reconcile the fact there is an infinite loop gain under ideal conditions, yet nothing in the math explicitely points to singularities? How is it that actual amplifiers are built and not only do not explode but a nice "distortion nulling" pot allows smooth adjustment?
This can be better understood if one looks for a moment only at the loop formed by S1-S-S2-B. Under ideal and cancellation conditions, the path gain is exactly unity and real. It is equivalent to a ball set on a flat floor, as long as it does not hit constraints (walls or supply rails for the circuit) it stays put or keeps happily rolling, no black holes here.
In reality one cannot make a unity gain loop without some form of active device in between, which automatically places at least one pole in the transfer. I represented this fact with amplifier S, and if one makes the numbers for an arbitrary first order model, the equivalent transfer from S1 + input to S output turns out to be an ideal integrator with the same gain-bandwidth product as the bare amplifier S. From there on, it is bussiness as usual from the standpoint of correction, stability and so forth.
Rodolfo
Re: Re: Re: Re: Some summary views on EC
I agree, and I have also never denied the validity and value of the high feedback view of HEC. The important thing is that one is open to the other useful views and that the value of the different insights they offer is exploited. This who fear the very term "positive feedback" are cheating themselves out of parts of the universe of possibilities. They also do not understand it fully, in my opinion. It is not unlike those who harbor a nonsensical fear of negative feedback because some people have made some bad amplifiers by mis-applying negative feedback. Any weapon will bite you in the butt if you do not understand it, respect it, and use it wisely.
Cheers,
Bob
ingrast said:
I agree, and I have also never denied the validity and value of the high feedback view of HEC. The important thing is that one is open to the other useful views and that the value of the different insights they offer is exploited. This who fear the very term "positive feedback" are cheating themselves out of parts of the universe of possibilities. They also do not understand it fully, in my opinion. It is not unlike those who harbor a nonsensical fear of negative feedback because some people have made some bad amplifiers by mis-applying negative feedback. Any weapon will bite you in the butt if you do not understand it, respect it, and use it wisely.
Cheers,
Bob
Hi Rodolfo.
You wrote:
I would be surprised if several real circuits have not exploded (or at least demonstrated marginal stability) as designers have tried to get their systems working on the bench. Real circuits do not explode because they are gain and frequency compensated such that the PFB loop never attains infinite gain AND the phase margin of the NFB loop is large enough. [Bob Cordell's circuit is an example: his circuit is only a shadow of the HEC block diagram, it slugs the forward gain, slows the ciruit down and incorporates an output L. And it loses phase margin into reactive loads. Which is Edmond's point]. Otherwise they would indeed oscillate. Just like any other NFB loop. There is no "feedback on demand" as far as the plant (N) is concerned...it is held with an iron grip at all times. Just like any other NFB loop.
The diagram shows another possible implementation topology. The middle, unity gain subtractor needs to have very wide bandwidth and very low distortion as both degrade proportionately to the synthetic gain = -1/(1-a).
By the way, do you have any recommendations for suitable subtractor/buffer ICs?
You wrote:
The system equations clearly show the infinite gain condition in the limit as 'a' tends towards 1 (in the HEC diagram) as far as I am concerned. The amount of NFB loop gain tends towards infinity and is effectively independent of output error. In other words, the system is under the tight grip of feedback no matter whether the plant (N) is linear or not. The feedback loop gain is not error dependent, just like a more conventional NFB system.
I would be surprised if several real circuits have not exploded (or at least demonstrated marginal stability) as designers have tried to get their systems working on the bench. Real circuits do not explode because they are gain and frequency compensated such that the PFB loop never attains infinite gain AND the phase margin of the NFB loop is large enough. [Bob Cordell's circuit is an example: his circuit is only a shadow of the HEC block diagram, it slugs the forward gain, slows the ciruit down and incorporates an output L. And it loses phase margin into reactive loads. Which is Edmond's point]. Otherwise they would indeed oscillate. Just like any other NFB loop. There is no "feedback on demand" as far as the plant (N) is concerned...it is held with an iron grip at all times. Just like any other NFB loop.
If the ball is meant to be Vdrv then the height of one side of the table is Vin and the other is Vout. The table is only level when the output exactly equals the input...the feedback error signal is zero. Of course, in any NFB system if you force the error signal to zero the plant input is free to assume any value...free to roll along the level surface. What is special here?
Absolutely. As the PFB loop gain approaches unity the bandwidth of the synthetic forward gain block containing the summers and S decreases as its gain increases as we've agreed before. This is like reducing the 1st pole frequency of an op-amp in OL mode. We have also agreed that the decision as to whether a synthetic gain block is superior to a conventional gain block is dependent on summers having an inherently superior GBP and distortion characteristic and upon circuit topology conveniences.
The diagram shows another possible implementation topology. The middle, unity gain subtractor needs to have very wide bandwidth and very low distortion as both degrade proportionately to the synthetic gain = -1/(1-a).
By the way, do you have any recommendations for suitable subtractor/buffer ICs?
Attachments
traderbam said:
The proof is in the pudding. It makes a wonderful amplifier. So much so, that Halcro decided to essentially copy it. What's your problem with that?
It has plenty of phase margin into an arbitrary reactive load with a mere 0.5 uH coil. What's your problem with that?
You're getting desperate, Brian. You are unable to get others to subscribe to your one-dimensional view of HEC, so now you have to slander an implementation of it.
I detect a severe case of NIH here.
Cheers,
Bob