a prediction of the "negative fedback" view of Bob's version of HEC is that stability is exactly the same as other high loop gain approaches, the outer feedback loop gain may appear to better behaved but proper measurement inside of all of the negative and positive feedback loops around the output transistors shows that Bob's HEC pays the same price in the trade-off of reduced phase margin/higher gain intercept for greater "error correction"
an interesting posibility is that the inner positive feedback in this approach could contribute to better nonlinear stability as the gain boost from the positive feedback goes away when the output clips - a Lurie style "nonlinear compensation" built into the topology
but my sims show that the total gain around the output devices has a higher "ulg" frequency than shown by simply cutting the outer feedback loop
an interesting posibility is that the inner positive feedback in this approach could contribute to better nonlinear stability as the gain boost from the positive feedback goes away when the output clips - a Lurie style "nonlinear compensation" built into the topology
but my sims show that the total gain around the output devices has a higher "ulg" frequency than shown by simply cutting the outer feedback loop
HEC vs NFB
Don't worry, I wasn't referring to you in particular and you didn't claim HEC is better, you only gave that impression by saying:
For an HEC stage, the bandwidth has no such constraint. It's instead constrained to be approximately the same as without HEC
Indeed, a PITA.
Not only Brian has shown that, I did that too, see: http://www.diyaudio.com/forums/showthread.php?postid=1360102#post1360102
This OPS has ~5.5MHz BW. If I increase the compensation caps C1 and C2 from 47pF to 47nF, it has still the same BW. With even larger values, this OPS simply behaves like an ordinary emitter/source follower configuration.
Last but not least, there are even more DOFs:
1. The amount of NFB @ DC.
2. The roll-off frequency of the feedback loop.
3. The shape of roll-off (think of TMC or TPC).
4. The closed loop gain.
Bye-bye HEC.
andy_c said:
Don't worry, I wasn't referring to you in particular and you didn't claim HEC is better, you only gave that impression by saying:
For an HEC stage, the bandwidth has no such constraint. It's instead constrained to be approximately the same as without HEC
Indeed, a PITA.
Not only Brian has shown that, I did that too, see: http://www.diyaudio.com/forums/showthread.php?postid=1360102#post1360102
This OPS has ~5.5MHz BW. If I increase the compensation caps C1 and C2 from 47pF to 47nF, it has still the same BW. With even larger values, this OPS simply behaves like an ordinary emitter/source follower configuration.
Last but not least, there are even more DOFs:
1. The amount of NFB @ DC.
2. The roll-off frequency of the feedback loop.
3. The shape of roll-off (think of TMC or TPC).
4. The closed loop gain.
Bye-bye HEC.
Re: Re: HEC vs NFB
I fully agree with that.
Regarding the DC loop gain, I think that 80dB is the limit for a practical NFB-OPS, while HEC, using 0.1% resistors, will stick at say 60dB. Anyhow, these figures are not really important, as loop gain at higher frequencies is what really counts.
traderbam said:
I fully agree with that.
Regarding the DC loop gain, I think that 80dB is the limit for a practical NFB-OPS, while HEC, using 0.1% resistors, will stick at say 60dB. Anyhow, these figures are not really important, as loop gain at higher frequencies is what really counts.
syn08 said:
First, I'm not sure whether breadboarding is a appropriate way to ***** the stability of HEC vs. NFB.
Second, the fair comparison of the two topologies is not possible when using at same two totally different output devices: lateral vs. vertical. The latter is faster and more prone to oscillations.
syn08 said:........
It's fast, isn't it?The slew rate is around 270V/uS (non slewing, of course) at full output. Talking about the diamond buffer being slow
Sure, its pretty fast, almost as fast as Stochino's amp (300V/us). But don't think it's due to that exotic diamond buffer, rather C34, that remedies the inherent shortcomings of a diamond buffer. Without that cap the performance falls below mediocrity: rising SR=28V/us, falling SR=13V/us (approximately).
Edmond Stuart said:
If there is another way to experiment with let me know.
If you read the PGP website you'll find out the I have tried HEC with both laterals and verticals. My HEC PCB can accomodate either. The decision for laterals was based on their self protecting nature. Today I would do it different (and I'll probably give it a shot).
Edmond Stuart said:
Are you changing your opinins quickly, aren't you? From http://www.diyaudio.com/forums/showthread.php?postid=1602815#post1602815
Using a diamond topology (highly unusual in a high power amp) as remedy solves only halve of the problems (indeed fast turn-off but now slow turn-on)
How is C34 helping the turn on one of those deep mysteries
And BTW, you are the last entitled to criticize unusual topologies Regarding the whole HEC vs. NFB... You guys seem to forget that ultimately FB theory is an analysis instrument. On the same note, you can argue until the cows come home if the emitter follower is NFB or not. Does it really matter? The analysis result and circuit behaviour will be exactly the same whatever way you are looking at.
Edmond Stuart said:
Hi Edmond,
I guess I understand your frustration to some extent because I also get frustrated on some different matters that I have felt were put to rest at the expense of a great deal of effort and discussion.
Here is where I think we can agree, or agree to disagree. First, depending very heaviliy on the details and specific implementation, I don't think it wise in general for anyone to make the general assertion that HEC is better than certain implementations of NFB, such as your version of local NFB around the output stage, or even well-implemented TMC.
However, I don't think that some of the HEC matters were put to rest in the way that you would like. While I and others acknowledge that HEC is a non-traditional form of NFB, we do not support the notion that other ways of looking at HEC are wrong or without value. We also don't suggest that HEC is something that is for NFB-haters. We also do not think of the HEC circuit as some inelegant kludge. In fact, I think it is a very elegant and effective circuit.
The supposed need for the trimpot is also misunderstood by many. While it is true that the use of a trim pot can get you the ultimate amount of distortion reduction at lower audio frequencies, anything better than 1% fixed resistors (once their value choice is optimized in initial design) is usually a waste as far as high-frequency distortion goes, such as THD-20. That is because at those higher frequencies the effectiveness of the EC is limited more by the necessity of compensating the EC than by static resistor tolerance. In my EC amp, I got 30 dB reduction in EC output stage THD-20; to first order, that corresponds to a 3% resistor error in rough terms; in fact, the reduction was limited by the compensation elements. NFB would also be exceptionally good at HF were it not for the need for frequency compensation. Indeed, feedforward error correction would also be exceptionally good at HF were it not for the realities of component matching and phase matching.
I think we can agree that the Devil is in the details. EC and local output stage NFB both have the potential to provide very good performance, but how each is implemented, and the degree of robustness of each to real-world realities, strays, and second-order effects may distinguish them.
For example, the way in I compensate the EC is such that its rolloff of the effect of EC at high frequencies does not introduce additional phase lag into the global loop. The R-C network I use actually moves in the direction of a feedforward short circuit at high frequencies. Not all other implementations necessarily have this property.
The proof is also in the pudding. To the best of my knowledge, my EC MOSFET amplifier held the world record for THD-20 (0.0006% in a 200 kHz measurement bandwidth) for over ten years, having been built with 1982 technology and unmatched first-generation HEXFETs. It was surpassed only in the 1990's by the Halcro, which was based on the same circuit, right down to the way I compensated the HEC.
Cheers,
Bob
jcx said:
What you have said here is largely true. If you view HEC as a particular way of getting additional local NFB around the output stage, it is clear that the total amount of NFB around the output stage is quite high at high frequencies, and that is another way of understanding that HEC needs its own frequency compensation. HEC is a great technique, but no matter how you slice it, there is no free lunch. If you view it as a variant of NFB, the HEC topology (at least the one I use) is a very tight, fast local negative feedback loop around the output stage that happens to have implementation and performance advantages over many alternatives, but not necessarily over all alternatives.
Cheers,
Bob
Bob Cordell said:
To be fair, I think that one of the reasons why you got such a high reduction in distortion is because you used a Hexfet output stage with both rather high intrinsic non-linearity and very high bandwidth.
It is very much harder to get a similar reduction without precise trimming from an output stage with significantly lower intrinsic distortion (the non-linearity of the EC circuity plays a more dominant role) and less bandwidth.
I’ve now given up on HEC on class-A RET output stages, for example. In such cases other tricks such as TMC can yield much better results with a great deal less complication.
Cheers,
Glen
G.Kleinschmidt said:
This is a fair observation, although I don't know if it is true. If amount of distortion reduction were, say, proportional to the amount of initial output stage nonlinearity, then that would suggest that the EC allows one to use crappier devices and get the same end result. Not a terrible position to be in. It would also suggest that one could reduce the idle bias below the 150 mA number I used and end up with the same amount of crossover distortion. The truth probably is somewhere in between.
I think the points I made about the trimpot are still relatively valid. Three-percent resistor error will still limit static EC distortion reduction to about 30 dB, not too shabby.
I would not normally use HEC on a class-A BJT design anyway. I think HEC is most applicable to vertical MOSFETs because they are very fast and because their transconductance droop puts them in need of something like HEC to achieve very low distortion performance with reasonable amounts of bias. That is basically the reason I chose to employ HEC when building a very high performance MOSFET power amplifier.
The bottom line is that HEC allowed me to achieve <0.0006% THD-20 with only 150 mA of bias current using first-generation HEXFETs 25 years ago.
I do agree, by the way, that TMC also has the potential of yielding very high performance. I have built vertical MOSFET power amplifiers without HEC, but have not yet tried one using TMC. I hope to do so soon.
Cheers,
Bob
Re: Re: Re: HEC vs NFB
It is clear to me that you use your implementation measurements, upon which your pregnant pride depends, to justify your erroneous, pseudo-science claims about how your circuit functions.
I am neither impressed by your 30dB of loop gain at 40kHz, nor by your lack of engineering rigour, both in your 1984 paper and in your posts 24 years later.
If you want to gain my respect for a claim, then show me the analysis.
This isn't kindergarten. I don't need to build something to understand its theory of operation. Besides, implementing it doesn't seem to have helped you.
Bob Cordell said:
It is clear to me that you use your implementation measurements, upon which your pregnant pride depends, to justify your erroneous, pseudo-science claims about how your circuit functions.
I am neither impressed by your 30dB of loop gain at 40kHz, nor by your lack of engineering rigour, both in your 1984 paper and in your posts 24 years later.
If you want to gain my respect for a claim, then show me the analysis.
This isn't kindergarten. I don't need to build something to understand its theory of operation. Besides, implementing it doesn't seem to have helped you.
Re: Re: Re: Re: HEC vs NFB
I think you have answered my question; you have not built a HEC amplifier.
From your last post, it seems clear that you may not understand its theory of operation. If I read your last post correctly, you were asserting that, in the negative feedback view of HEC, one can replace the unity-positive-feedback loop part of the architecture with an integrator. While there is some resemblance of behavior in this regard, that assumption is wrong. It is true that the effective gain of the positive feedback loop will fall with increasing frequency from a very large value to a small value as a result of the HEC loop compensation, but at higher frequencies the behavior differs from that of an integrator. The integrator will continue to introduce at least a 90 degree phase lag, while the positive feedback loop will drop to unity forward gain with close to zero degrees phase shift.
When you don't build something, or at least simulate it, you fall prey to the assumptions you make when you try to analyze it from your own theoretical point of view.
Have you SPICE simulated a HEC circuit?
What IS kindergarten here is when you try to make your technical point with personal insults.
People have legitimately differing ways of looking at circuit functions and I'm all for diversity here. I can't help it if you do not subscribe to the way I choose to describe a circuit. However, what you cannot deny in the end is the high level of performance that the circuit achieved.
Cheers,
Bob
traderbam said:
It is clear to me that you use your implementation measurements, upon which your pregnant pride depends, to justify your erroneous, pseudo-science claims about how your circuit functions.
I am neither impressed by your 30dB of loop gain at 40kHz, nor by your lack of engineering rigour, both in your 1984 paper and in your posts 24 years later.
If you want to gain my respect for a claim, then show me the analysis.
This isn't kindergarten. I don't need to build something to understand its theory of operation. Besides, implementing it doesn't seem to have helped you.
I think you have answered my question; you have not built a HEC amplifier.
From your last post, it seems clear that you may not understand its theory of operation. If I read your last post correctly, you were asserting that, in the negative feedback view of HEC, one can replace the unity-positive-feedback loop part of the architecture with an integrator. While there is some resemblance of behavior in this regard, that assumption is wrong. It is true that the effective gain of the positive feedback loop will fall with increasing frequency from a very large value to a small value as a result of the HEC loop compensation, but at higher frequencies the behavior differs from that of an integrator. The integrator will continue to introduce at least a 90 degree phase lag, while the positive feedback loop will drop to unity forward gain with close to zero degrees phase shift.
When you don't build something, or at least simulate it, you fall prey to the assumptions you make when you try to analyze it from your own theoretical point of view.
Have you SPICE simulated a HEC circuit?
What IS kindergarten here is when you try to make your technical point with personal insults.
People have legitimately differing ways of looking at circuit functions and I'm all for diversity here. I can't help it if you do not subscribe to the way I choose to describe a circuit. However, what you cannot deny in the end is the high level of performance that the circuit achieved.
Cheers,
Bob
Robert Cordell
traderbam must have his bad day, too
Interesting though, to read the different opinions
on different ways of corrective feedback techniques
we all have our own ways
especially when we have beeen into audio amplifier designs for a while.
Look at Nelson Pass -> he certainly have his own ways .. and is not too bad
YAP, syn08 + Edmond Stuart effort, will end up something extra.
Those guys are good! This we can tell, no doubt.
I would say, their amplifier projects are more Professional Audio, than Diy Audio.
Nevertheless, there is much to learn from these 'pros'.
And no doubt Lineup will learn. More.
He is one quick learner, this man.
Regards
traderbam must have his bad day, too
Interesting though, to read the different opinions
on different ways of corrective feedback techniques
we all have our own ways
especially when we have beeen into audio amplifier designs for a while.
Look at Nelson Pass -> he certainly have his own ways .. and is not too bad
YAP, syn08 + Edmond Stuart effort, will end up something extra.
Those guys are good! This we can tell, no doubt.
I would say, their amplifier projects are more Professional Audio, than Diy Audio.
Nevertheless, there is much to learn from these 'pros'.
And no doubt Lineup will learn. More.
He is one quick learner, this man.
Regards
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.