andy_c said:
For the record, it turns out that my preferred view for understanding the fundamentals of EC is completely inline with andy_c's here, so I will not repeat it again. Andy and I have not talked about this offline just to be clear. Yes, it's true that some gain is required after the diff amp when trimmed to only extract the error, this was obvious to me.
It is clear that Andy has taken this further as it sounds as if he's working on an actual implementation. He points out the interesting issue of the EC signal becoming large at very high currents. I would have the goal for the EC circuit to saturate in such a way for it to at least not increase the uncorrected distortion significantly.
I don't know if I'll ever use EC, but I'll certainly keep it in mind for future designs.
Pete B.
janneman said:
Thanks Jan! You saved me a trip to the library.
BTW, what do you mean by "around 1994 I think" ? (italics by me)
If you had read the article you know that it was published in Oct. 1994 🙂
Cheers, Edmond.
Real World Implementations
My challenge still stands, I offer this simply because it demonstrates the difference in real world implementations. Show me a modern amplifier design using "traditional" negative feedback only that provides a null control.
jcx offers this, but it is a power supply not an amplifier, uses tubes not BJT or MOSFET, and positive feedback which he himself calls, how did he put it antiquated IIRC:
http://www.diyaudio.com/forums/showthread.php?postid=1309327#post1309327
This does shed some light on their point, one can have a null control in a system that uses both positive and negative feedback, however I would not call this traditional in the sense of a classical negative feedback only topology. The positive feedback offers a way of viewing the circuit as having infinite gain, show me a real world implementation with infinite gain that does not use positive feedback.
The next abstract transformation offered is to eliminate the positive feedback which then however requires the abstract infinite gain block. Next challenge: Show me a circuit with no positive feedback that provides infinite gain. This is what I mean by abstract, and more importantly that HEC is different.
One thing to keep in mind from an implemenation standpoint, is that while it is clear that infinite gain would be required to provide the same results with traditional feedback, Bob's HEC implementation works by simply adding a handful of low cost transistors and passive components. It also has the advantage from an implementation standpoint that it is a short loop local to the output stage.
Another challenge for the opposing view would be to show a simple traditional, local, negative feedback loop around the output stage that provides the same/similar performance. If it is all so simple, then show me an implementation, the solution is not obvious as it would be commonly used. The CFP output stage is perhaps an obvious NFB example, however many here seem to dismiss it, I do not but I've not seen one with near zero distortion.
Before anyone takes a cheap shot let me mention that I offer these challenges to open peoples eyes to the implementation differences not because I need to learn from them.
Pete B.
My challenge still stands, I offer this simply because it demonstrates the difference in real world implementations. Show me a modern amplifier design using "traditional" negative feedback only that provides a null control.
jcx offers this, but it is a power supply not an amplifier, uses tubes not BJT or MOSFET, and positive feedback which he himself calls, how did he put it antiquated IIRC:
http://www.diyaudio.com/forums/showthread.php?postid=1309327#post1309327
This does shed some light on their point, one can have a null control in a system that uses both positive and negative feedback, however I would not call this traditional in the sense of a classical negative feedback only topology. The positive feedback offers a way of viewing the circuit as having infinite gain, show me a real world implementation with infinite gain that does not use positive feedback.
The next abstract transformation offered is to eliminate the positive feedback which then however requires the abstract infinite gain block. Next challenge: Show me a circuit with no positive feedback that provides infinite gain. This is what I mean by abstract, and more importantly that HEC is different.
One thing to keep in mind from an implemenation standpoint, is that while it is clear that infinite gain would be required to provide the same results with traditional feedback, Bob's HEC implementation works by simply adding a handful of low cost transistors and passive components. It also has the advantage from an implementation standpoint that it is a short loop local to the output stage.
Another challenge for the opposing view would be to show a simple traditional, local, negative feedback loop around the output stage that provides the same/similar performance. If it is all so simple, then show me an implementation, the solution is not obvious as it would be commonly used. The CFP output stage is perhaps an obvious NFB example, however many here seem to dismiss it, I do not but I've not seen one with near zero distortion.
Before anyone takes a cheap shot let me mention that I offer these challenges to open peoples eyes to the implementation differences not because I need to learn from them.
Pete B.
Other Considerations
I'll also offer the question: Are we perhaps overlooking an issue with traditional Lin topology amplifiers that limits the distortion reduction possible with traditional NFB? Early effect due to the common mode signal in the diff pair for example as is mentioned as far back as this article:
"Distortion in low-noise amplifiers" by Eric F. Taylor, Wireless World August, 1977 and also the September part 2 issue?
My point is to consider if the common Lin topology, NFB implementations are overlooking something that causes significant deviation from the theoretical NFB distortion reduction model.
Pete B.
I'll also offer the question: Are we perhaps overlooking an issue with traditional Lin topology amplifiers that limits the distortion reduction possible with traditional NFB? Early effect due to the common mode signal in the diff pair for example as is mentioned as far back as this article:
"Distortion in low-noise amplifiers" by Eric F. Taylor, Wireless World August, 1977 and also the September part 2 issue?
My point is to consider if the common Lin topology, NFB implementations are overlooking something that causes significant deviation from the theoretical NFB distortion reduction model.
Pete B.
Re: Real World Implementations
Hi Pete,
You have challenged me and my answer is: it is possible and very simple. Just one additional R and C. Similar performance and stability.
Cheers, Edmond.
edit: And indeed, the solution is not obvious, as it is not commonly used.
PB2 said:
Hi Pete,
You have challenged me and my answer is: it is possible and very simple. Just one additional R and C. Similar performance and stability.
Cheers, Edmond.
edit: And indeed, the solution is not obvious, as it is not commonly used.
PMA wrote:
I have never seen any NFB circuit with a null control, if by null you mean error elimination.
Show me a real world implemenatation with infinite gain, at all.
Bob's circuit does not use infinite gain. It uses a moderate amount of loop gain that would be easy to provide in a different NFB topology. Bob's circuit has about 30dB loop gain at 20kHz.
How much loop gain does you CFP provide? What do you mean by near zero distortion?
To me the important question is what ways the HEC topology has implementation advantages and disadvantages as a NFB loop.
Edmond Stuart said:
I was too lazy to look up the date when I posted. Turns out I was right, so why are you complaining?? 😉
Jan Didden
Re: Real World Implementations
The choice of what output stage gain (with what load/ output current) is also addressed by Hawksford, but I haven't seen it implemented yet.
Note that HEC as we discuss it concerns itself with correcting voltage levels. There is an additional part, described by Hawksford, that also corrects for the output stage input current (and thus indirectly for the load current). This would allow you to select a particular output gain with a particular load (even resistive) to set the K1 mentioned above, and then in real operation the EC also corrects for those variations. You still would have the issue of dynamic range of course.
Jan Didden
PB2 said:
The choice of what output stage gain (with what load/ output current) is also addressed by Hawksford, but I haven't seen it implemented yet.
Note that HEC as we discuss it concerns itself with correcting voltage levels. There is an additional part, described by Hawksford, that also corrects for the output stage input current (and thus indirectly for the load current). This would allow you to select a particular output gain with a particular load (even resistive) to set the K1 mentioned above, and then in real operation the EC also corrects for those variations. You still would have the issue of dynamic range of course.
Jan Didden
janneman said:
Me?...Complaining?
I only tried to demonstrate that you are lazy. 😉Cheers, Edmond.
Re: Real World Implementations
Hi Pete,
These are all very good points and relevant questions. I like your "show me" approach.
The problem with CFP is that the two separate tight feedback loops don't really enclose the whole Class-AB output stage. In particular, the local feedback in a CFP arrangement does not mitigate the crossover transition distortion from the top to bottom elements, leaving things like gm doubling distortion right out in the open.
Cheers,
Bob
PB2 said:
Hi Pete,
These are all very good points and relevant questions. I like your "show me" approach.
The problem with CFP is that the two separate tight feedback loops don't really enclose the whole Class-AB output stage. In particular, the local feedback in a CFP arrangement does not mitigate the crossover transition distortion from the top to bottom elements, leaving things like gm doubling distortion right out in the open.
Cheers,
Bob
Re: Real World Implementations
The only view I oppose:
that I should be required to abandon a basic pattern of mathematical proof: show by legal transformations that one system can be expressed in many logically equivalent forms, and that an implication/property of one of those forms applies to all other transformed representations of the same system
These are equivalent assumptions: perfect summers and infinite gain - given one you can construct a block diagram system that gives the other
Therefore the assertion that a real world implementation of EC can be perfect is to me a statement that you gan build a real world infinite gain block - since I reject the latter I reject the inital assumtion - no feedback EC circuit can be perfect - Exactly as no negative feedback cirucit can
as for other local negative feedback around the output stage examples you can look at Cherry's articles in the JAES:
http://www.aes.org/journal/journal_search.cfm
type in Cherry in Author
PB2 said:
The only view I oppose:
that I should be required to abandon a basic pattern of mathematical proof: show by legal transformations that one system can be expressed in many logically equivalent forms, and that an implication/property of one of those forms applies to all other transformed representations of the same system
These are equivalent assumptions: perfect summers and infinite gain - given one you can construct a block diagram system that gives the other
Therefore the assertion that a real world implementation of EC can be perfect is to me a statement that you gan build a real world infinite gain block - since I reject the latter I reject the inital assumtion - no feedback EC circuit can be perfect - Exactly as no negative feedback cirucit can
as for other local negative feedback around the output stage examples you can look at Cherry's articles in the JAES:
http://www.aes.org/journal/journal_search.cfm
type in Cherry in Author
I'm intrigued by the Halcro dm88 owners manual.
1) I cannot find any mention of the output impedance
2) They limit the peak output current to 15A
3) They claim >500W into 4 ohms resistive load
Stereophile says the dm38 output Z is about 0.1 ohms at 20kHz and has 120kHz bandwidth.
Many designs boast peak output currents of 30A or more these days and output Z much less than 0.1 ohms. Halcro are advertising a 500W product that has the sort of output specs you would normally associate with a 50W or less integrated amplifier. 15A peak is not enough to generate 500W avg into a 4 ohm resistor...it is short by 50W.
Does anyone else find this interesting?
1) I cannot find any mention of the output impedance
2) They limit the peak output current to 15A
3) They claim >500W into 4 ohms resistive load
Stereophile says the dm38 output Z is about 0.1 ohms at 20kHz and has 120kHz bandwidth.
Many designs boast peak output currents of 30A or more these days and output Z much less than 0.1 ohms. Halcro are advertising a 500W product that has the sort of output specs you would normally associate with a 50W or less integrated amplifier. 15A peak is not enough to generate 500W avg into a 4 ohm resistor...it is short by 50W.
Does anyone else find this interesting?
Re: Re: Real World Implementations
I agree, neither is perfect.
I also agree that, in theory, the assumptions of a perfect feedforward summer and the existence of an infinite gain block, however realized, are largely equivalent.
I do think it is interesting to see, in the real world, how close to perfect each comes.
Cheers,
Bob
jcx said:
I agree, neither is perfect.
I also agree that, in theory, the assumptions of a perfect feedforward summer and the existence of an infinite gain block, however realized, are largely equivalent.
I do think it is interesting to see, in the real world, how close to perfect each comes.
Cheers,
Bob
traderbam said:
I agree, Brian. Although there are certainly things that the Halcro appears to do right, there are others that are at least puzzling.
I'm guessing the Zout of 0.1 ohms at 20 kHz is largely a reflection of their output inductor. In the EC design that I did, I used a 0.5 uH coil in parallel with a 0.5 ohm resistor, and I think that resulted in a DF of about 125 at 20 kHz. Halcro's 0.1 ohm would correspond to a DF of about 80, so that sort of checks out. I'm guessing their Zout at lower frequencies is quite a bit lower.
Their current limiting is also puzzling, but it is unclear whether their current limiting just applies to long-term bursts or to short term bursts as well. If I had to make a wild guess, I'd blame it on their switching power supply, but I'll be the first to admit that is just a wild guess. It could just as easily be a conscious decision on their part to be part of their protection scheme.
Either way, their continuous sinewave performance into 2 ohms is disappointing. John Curl's JC-1 beats them hands-down in that department. I would like to have seen 2-cycle 20 Hz burst capability of the Halcro into 2 ohms, however.
Cheers,
Bob
Re: Re: Real World Implementations
Thanks Bob, yes, I agree with your points about CFP with regard to the local feedback.
Pete B.
Bob Cordell said:
Thanks Bob, yes, I agree with your points about CFP with regard to the local feedback.
Pete B.
Re: Re: Real World Implementations
I don't require you to abandon valid system transformations, simply the view that a specific and different implementation such as EC can be dismissed as equivalent to traditional negative feedback. Perhaps we should just agree to disagree.
I've tried to be very careful with my wording not to assert that EC is perfect, obviously nothing is perfect. I disagree with your claim that the non-ideal characteristics of EC are the same as the impossiblity of infinite gain. Obviously, no circuit or design on paper can be implemented exactly, since all real world designs are subject to tolerances, finite bandwidth etc. you certainly know this. Thus, for any real world implementation we might ask how close does this come to the mathematical model.
It's clear that traditional negative feedback systems do not even come close to infinite gain, especially at 20 KHz. Indeed, this is not even generally a goal due to the conflicting requirement for stability. EC on the other hand adds small 100 to 300 MHz transistors which provide huge margin over the 20 KHz bandwidth for audio, and therefore as Bob points out there is a deep null (obviously not a perfect null, but what we call in engineering a null) over the audio band. Contrast this with traditional negative feedback amplifiers which do not even come close to inifinite gain over the audio band. This is why I view the reduction argument offered to be obviously flawed. The details are in the implementation differences as I see it, and this is why the implementation details cannot be casually dismissed.
I'm somewhat familiar with Cherry's work, having been a member of the AES for over 20 years, however I've not read those papers probably since they were published. It's certainly possible to put negative feedback around the output stage. I don't dispute this, I would question if there is an implementation that does as good a job as EC with equal or lesser complexity.
Pete B.
jcx said:
I don't require you to abandon valid system transformations, simply the view that a specific and different implementation such as EC can be dismissed as equivalent to traditional negative feedback. Perhaps we should just agree to disagree.
I've tried to be very careful with my wording not to assert that EC is perfect, obviously nothing is perfect. I disagree with your claim that the non-ideal characteristics of EC are the same as the impossiblity of infinite gain. Obviously, no circuit or design on paper can be implemented exactly, since all real world designs are subject to tolerances, finite bandwidth etc. you certainly know this. Thus, for any real world implementation we might ask how close does this come to the mathematical model.
It's clear that traditional negative feedback systems do not even come close to infinite gain, especially at 20 KHz. Indeed, this is not even generally a goal due to the conflicting requirement for stability. EC on the other hand adds small 100 to 300 MHz transistors which provide huge margin over the 20 KHz bandwidth for audio, and therefore as Bob points out there is a deep null (obviously not a perfect null, but what we call in engineering a null) over the audio band. Contrast this with traditional negative feedback amplifiers which do not even come close to inifinite gain over the audio band. This is why I view the reduction argument offered to be obviously flawed. The details are in the implementation differences as I see it, and this is why the implementation details cannot be casually dismissed.
I'm somewhat familiar with Cherry's work, having been a member of the AES for over 20 years, however I've not read those papers probably since they were published. It's certainly possible to put negative feedback around the output stage. I don't dispute this, I would question if there is an implementation that does as good a job as EC with equal or lesser complexity.
Pete B.
Is the Earth round?
Recent arguments in favour of EC being different from and superior to NFB:
Objective
There is a simple way to insert a pot to adjust for a THD minima.
Subjective conjecture
1) It cancels the output error just like feed-forward rather than reducing it
2) It enables the use of faster transistors in the correction circuit
3) The system is more stable because it has an open loop gain of <1
4) NFB requires infinite gain to eliminate the error. EC does not.
5) EC topology results in a simpler circuit
6) Bob Cordell says it is and he should know
Engineering has objective analysis as a basic tenet.
This is to stop our powerful motivations to believe (or to claim) something from blinding us to reality.
Just ask Aristotle.
😎
Recent arguments in favour of EC being different from and superior to NFB:
Objective
There is a simple way to insert a pot to adjust for a THD minima.
Subjective conjecture
1) It cancels the output error just like feed-forward rather than reducing it
2) It enables the use of faster transistors in the correction circuit
3) The system is more stable because it has an open loop gain of <1
4) NFB requires infinite gain to eliminate the error. EC does not.
5) EC topology results in a simpler circuit
6) Bob Cordell says it is and he should know
Engineering has objective analysis as a basic tenet.
This is to stop our powerful motivations to believe (or to claim) something from blinding us to reality.
Just ask Aristotle.
😎
Bob wrote:
It think it is rather odd for a big amp to have a 15A peak limit. Usually, as you mention, there would be a much higher peak but a sustained level limit. In these cases a mfr is not likely to make such a bunch of excuses about it as Halcro do in their manual.
According to Stereophile the output Z is not much less than 0.1 across the audio band.
I wonder if it is more to do with their OS. It could be the SMPS if they don't use large caps on the output. It might be the OS devices they are using are weedy. It could also be that their feedback circuit, which is floating and has quite limited voltage headroom (10V is it?), may not be able to drive enough Vgs on the output FETs to get more than 15A peak.
It think it is rather odd for a big amp to have a 15A peak limit. Usually, as you mention, there would be a much higher peak but a sustained level limit. In these cases a mfr is not likely to make such a bunch of excuses about it as Halcro do in their manual.
That fits. I am concerned that such a high power amp would be so limited in peak current. Real speaker loads, especially high end speakers, are going to hit that peak quite easily wouldn't you think?