First, I always regarded you as an experts on EC. So your question highly amazed me.
Anyhow, this is NOT a 'marvellous' circuit, as the distortion of Q20...Q27 is not reduced or eliminated, only the one from the output devices.
BTW, we have already discussed this 'feature' of the HEC circuit, so one more reason to be amazed by your question.
Anyhow, this is NOT a 'marvellous' circuit, as the distortion of Q20...Q27 is not reduced or eliminated, only the one from the output devices.
BTW, we have already discussed this 'feature' of the HEC circuit, so one more reason to be amazed by your question.
Sigh. Edmond, I know that Q20 distortion is not reduced. I didn't say that. And I thought that, since the signal at the base of Q24 can be considered, at a first approximation, to be equal to the signal at its emitter, I would be allowed to say the control node is connected to the ec input, that you would see through that simplification.
I really don't wish to continue this type of discussion, Edmond.
Jan Didden.
I really don't wish to continue this type of discussion, Edmond.
Jan Didden.
If you don't wish to continue this type of discussion, then don't put obvious questions like this one:
>the result of that is that the signal at R34/Q24 is no longer a faithfull copy of the Vas Vout: iow, the input signal to the ec loop is distorted by that loop itself. Am I reasoning correctly here?
Yes, of course that's correct! So, what's your point? Only this one???
>the result of that is that the signal at R34/Q24 is no longer a faithfull copy of the Vas Vout: iow, the input signal to the ec loop is distorted by that loop itself. Am I reasoning correctly here?
Yes, of course that's correct! So, what's your point? Only this one???
janneman said:
Hi Jan,
You are exactly right. In theory, the EC circuit predistorts the signal at R34/Q24 by just the amount needed to make the output signal close to a true copy of the signal at the output of the VAS. Of course, we all know that the accuracy of this process is a matter of degree, and that the need to compensate the (perhaps non-traditional) feedback loop formed by the EC tends to degrade the accuracy of this process at high frequencies.
If one were to probe this point with a distortion analyzer, one would see considerable distortion at this point. It is akin to the way that ordinary global negative feedback essentially pre-distorts the VAS output in such a way as to get the final output to be a faithful version of the amplifier input. If one probes the output of the VAS of an ordinary feedback amplifier, one will see quite a bit of distortion there, reflective of the distortion of the output stage.
Cheers,
Bob
Happy New Year Bob. You answered Jan's query in such an adroit way that I feel compelled to agree. Nice. 😎
Bob Cordell said:
Thanks Bob.
I guess the thing that still intrigues me is the following.
In the classical nfb case, in the ideal case (open loop gain infinite), one subtracts the original signal from the Vin to have only the distortion left. In that ideal case, the forward loop pretty much only amplifies the distortion, in appropriate phase, which then comes out as an undistorted signal because it is a complement of the forward gain non-linearity, so to speak.
With ec, there is no infinite gain in the forward loop (prior to application of ec), so the effective input signal needs to contain an appreciable level of the original signal (depending on the actual forward gain), superimposed on which is then the 'pre-distortion'. This 'pre-distortion' must be exactly right to make ec effective.
Now I know all this is not exactly a breakthrough reasoning. I know it works, I can do the sums and the equations come out right, so on an intellectual level I Get It. Still, I can't get my *intuitive* head around how ec gets it just right. Maybe my personal blind spot. Thanks for listening anyway.
Jan Didden
janneman said:
Hi Jan ! How've you be doing !!
If, without loss o generality one considers the case of an exactly unity gain output stage to be corrected, then the visualization of how it's done is:
The first comparator as seen going from output backwards, extracts only and exactly (ideally) the distortion introduced by the output stage.
This is the essential distinction that makes EC conceptually different if not functionally different from NFB in the end, at least in conventional realizations. You had it sorted out clearly with the idea of "feedback on demand". This is the way "ec gets it just right".
The rest is simply to handle this error information in an appropriate way.
Rodolfo
Ohh ohhh... 😀
I had been carefull not to open that particular pandora's box...
What I mean is this. If you look at the ec output, to be applied to the input summer, it consists of partly original signal, partly distortion. It does NOT contain the full distortion signal as a nfb would. Because the ec sort of bootstraps its own input, it only contains 'part of' the distortion. So there will always remain part of the distortion between the ec input and output when it is in balance, so to say. How much? Well, just enough to compensate for the forward path non-linearity. This is different from nfb where you try to re-insert, in the input, the full, complete distortion signal.
I'm not sure I'm expressing myself clear enough here. Sort of thinking out loud.
Jan Didden
I had been carefull not to open that particular pandora's box...
What I mean is this. If you look at the ec output, to be applied to the input summer, it consists of partly original signal, partly distortion. It does NOT contain the full distortion signal as a nfb would. Because the ec sort of bootstraps its own input, it only contains 'part of' the distortion. So there will always remain part of the distortion between the ec input and output when it is in balance, so to say. How much? Well, just enough to compensate for the forward path non-linearity. This is different from nfb where you try to re-insert, in the input, the full, complete distortion signal.
I'm not sure I'm expressing myself clear enough here. Sort of thinking out loud.
Jan Didden
Hi Rodolfo
Conceptually different? If you like so.
Functionally different? NO!
See: http://www.diyaudio.com/forums/showthread.php?postid=1360102#post1360102 (Vanderkooy and Lipshitz)
Cheers,
Edmond.
Conceptually different? If you like so.
Functionally different? NO!
See: http://www.diyaudio.com/forums/showthread.php?postid=1360102#post1360102 (Vanderkooy and Lipshitz)
Cheers,
Edmond.
janneman said:
All right, you did it, you popped the lid now brace yourself !!!!!
Seriously, probably your quandry stems from the fact you take a non-unity gain output stage and substract output from input without scaling accordingly to match signal levels.
While not intuitively obvious, the operation for non unity gais is exactly the same as the case of unity gain as can be shown, but muddies the underlying concept.
I should rather break the problem starting with a simple easily to visualize case - unity gain - and then extend to other situations as corollaries.
Rodolfo
Edmond Stuart said:
Hi Edmond!
Note I was carefull to to spell "if not functionally different from NFB in the end, at least in conventional realizations".
This is the curx of the matter in my view and the root of some old bitter disputes about what EC is in the end.
In other words, one thing is to extract exactly the error - as EC starts doing - and a different matter is how this error is used.
If it is added at the input as is usually the case, then we have an infinite gain driver stage stemming from a unity gain positive feedback inner loop, and global negative feedback for the complete contraption.
In reality, for a non-infinite bandwith summing node, we get at best a close to ideal integrator thus attaining near perfect correction at DC and decreasing perfomance with increasing frequency.
What I should like to stress is that the concept of extracting the net error not necessarily imply its application as an added correction term directly.
Rodolfo
Hi Rodolfo,
Agreed. But let's assume that the error signal IS added to the input (and consequently we have an infinite gain driver stage), then what are your conclusions (from a practical point of view) in terms of pros, cons and caveats WRT ordinary NFB?
Cheers,
Edmond.
Agreed. But let's assume that the error signal IS added to the input (and consequently we have an infinite gain driver stage), then what are your conclusions (from a practical point of view) in terms of pros, cons and caveats WRT ordinary NFB?
Cheers,
Edmond.
Edmond Stuart said:
At worst, uninteresting. At best, with some very clever, novel design for the summing node (*), the possibility of a better end result.
In a different scenario, not directly summing the error but processing it in a way that does not imply automatically an outer global feedback loop, things begin to turn really interesting.
Rodolfo
(*) To be more explicit, note that the summing node is required to be fast (large bandwidth), linear, but only unity gain or so. These are different design goals than the usual high gain intended for NFB. Of course it makes no sense to use NFB in it for this goal !!! (back to square one).
Edmond Stuart said:
I wish I could !!!
I am not in position to devote the required time to this, and should be perfectly happy to see someone doing it.
The idea goes more or less like this.
Take the net error as extracted, and analyze it toghether with the input signal. Here you have a complete description of the stage to be corrected, only there is no need to blindly attempt to do it in real time at submicrosecond rate. Instead, we can resort to the wondefull trove of digital processing resources readily at hand to deconstruct the transfer kernel both in amplitude dependent and frequency dependent elements averaged on longer time spans - may be miliseconds to tens of miliseconds as a minimum window - and use this information to precorrect the input.
I am not sure how far can this go, that is, if it is really possible to take a longer term average description to attain a substantial improvement, given the fact that averaging necessarily implies loss of information.
Thermal effects probably can readily be handled by this approach. Static nonlinearities equally, but nonlinar capacitances probably not, just to name some issues. A very good working knowledge of nonlinar circuit theory is a prerequisite.
But this way the outer loop is entirely suppressed and with it any reminiscences of classical NFB.
Probably this approach is better suited also in a digital input - switch mode power stage, where the output sample is taken directly from the actual zero crossings plus some more information about power rails, edge slopes and whatever may be important to take into account, while the input is readily available in digital code.
I don't think this concept may be surprising or novel. The proof anyway will be in the pudding - can it be worked out as a practical implementation?
Rodolfo
Hi Rodolfo,
I think I understand what you mean. So you will need a fast and accurate DAC, ADC and DSP. But can it compete (spec-wise and $-wise) with a top end analog design (THD20<1ppm)?
In the past I did something similar with my PC-based distortion analyzer in order to compensate for the distortion of the DAC and ADC itself. But it was difficult to get the measurement floor below -130dB, as the distortion also depends on temperature, amplitude, aging etc.
Cheers,
Edmond.
I think I understand what you mean. So you will need a fast and accurate DAC, ADC and DSP. But can it compete (spec-wise and $-wise) with a top end analog design (THD20<1ppm)?
In the past I did something similar with my PC-based distortion analyzer in order to compensate for the distortion of the DAC and ADC itself. But it was difficult to get the measurement floor below -130dB, as the distortion also depends on temperature, amplitude, aging etc.
Cheers,
Edmond.
janneman said:
Hi Jan,
Bear in mind that the EC circuit views any departure, even linear, from unity gain, as an error to be corrected.
Thus, if the un-corrected gain of the output stage is 0.9, the EC signal fed back will contain a significant amount of the real signal in addition to the nonlinear part.
Cheers,
Bob
Edmond Stuart said:
Hi Edmond,
Peace.
The three ways of looking at EC,
1) conventional NFB with very large loop gain from inner PFB loop,
2) error correction as Hawksford described it,
3) NFB on demand,
are useful ways of looking at the beast to different people in different circumstances. Analyzed properly, all three will give the same right answer. Some will find views 2 and 3 worthless, and they are welcome to stick soley with view #1. Some will view view #2 as a fraudulent description for employing the term "error correction". So be it. Just semantics. I get offended by other's choices of semantics in certain situations as well.
Back at Bell Labs in the '70s, some of us linear IC designers argued about whether the function of a transistor in a circuit should be viewed as one of current gain or as one of transconductance. Obviously, both views are equally correct when applied properly. Some designers were more comfortable with the Beta view, while others were more comfortable with the transconductance view. I'm a transconductance kind of guy.
BTW, TPC was employed by Bell Labs in operational amplifiers since the early '70's, where it was called T compensation. These op amps were used in high-performance active filters where the extra in-band open-loop gain was important (especially with early-generation op amps with limited GBW).
I think that you have shown that alternative approaches to distortion reduction that are more like conventional NFB, like high levels of tight local output stage NFB and TMC, can be just as effective as HEC. That is a good thing. The more ways we have of attacking the problem, the better.
Peace,
Bob