Alfred wrote:
There appears to be argument about this definition within this thread even though the mathmatics is indisputable.
The argument appears to arise because certain NFB topologies use a positive feedback loop to mimmick a gain block in the forward path. It seems that some choose to call a NFB loop that uses a PFB loop in this way as "error correction". Some claim implementation benefits although these have not been explained yet.
My guess is that the misleading term "error correction" became popular in the 1980s after a professor from the University of Essex in the UK, Hawksford, published a paper outlining a universal feedback-feedforward mathmatical equation. He called his paper "distortion correction". When implemented node for node, his mathmatical model leads to a NFB system using a PFB gain block, even though it needn't use a PFB gain block. The emergeance of the PFB loop is caused by neglecting to reduce the mathmatical model to its simpler equivalence. I imagine the term "error correction" then emerged because it was a trendy term, from the digital world, that suggested a new idea in audio so as to attract interest.
So to be consistent one either has to define all NFB loops as "error correction", or only those that use a PFB gain block in their forward path as "error correction" (which seems a misnomer to me) or dismiss the term when applied to NFB altogether (which makes the most sense to me).
IMO the term "error correction" is more aptly applied to a feed-forward system.
It is usually about digital data correction. In this thread it is about the application of negative feedback to reduce the distortion of an analogue amplifier stage.
There appears to be argument about this definition within this thread even though the mathmatics is indisputable.
The argument appears to arise because certain NFB topologies use a positive feedback loop to mimmick a gain block in the forward path. It seems that some choose to call a NFB loop that uses a PFB loop in this way as "error correction". Some claim implementation benefits although these have not been explained yet.
My guess is that the misleading term "error correction" became popular in the 1980s after a professor from the University of Essex in the UK, Hawksford, published a paper outlining a universal feedback-feedforward mathmatical equation. He called his paper "distortion correction". When implemented node for node, his mathmatical model leads to a NFB system using a PFB gain block, even though it needn't use a PFB gain block. The emergeance of the PFB loop is caused by neglecting to reduce the mathmatical model to its simpler equivalence. I imagine the term "error correction" then emerged because it was a trendy term, from the digital world, that suggested a new idea in audio so as to attract interest.
So to be consistent one either has to define all NFB loops as "error correction", or only those that use a PFB gain block in their forward path as "error correction" (which seems a misnomer to me) or dismiss the term when applied to NFB altogether (which makes the most sense to me).
IMO the term "error correction" is more aptly applied to a feed-forward system.
Does anybody know if the term "Error correction" or the concept labelled so by Hawksford appear elsewhere in EE or related sciences like control theory or general systems theory?
I tried to google for an answer but got tired after a while, when finding mostly things like "error correction in the classroom" and "error correction in birdsong".
I tried to google for an answer but got tired after a while, when finding mostly things like "error correction in the classroom" and "error correction in birdsong".
what's in a name?
I have always taken "error correction" as descripive of a strategy, much more than an end.
We must acknowledge any correction strategy which relies on taking an action "after the fact" cannot fulfill its goal in entirety without violating the causality principle, more on this later.
Negative feedback can achieve its objective asyntotically with infinite gain.
Error correction can only succeed at DC, which also does not exist ( any power supply must have been turned on after the big bang at least).
Perfect correction can only be attained with forecknowledge of the distortion ahead, this is a provoking idea worth being considered now we have digital technology in principle capable of modeling and self adjustment, and is in my horizon as long as I can devote time and resoruces in this direction.
For the time being, I think the issue is whether the error correction strategy can bring practical benefits when compared with conventional negative feedback, given the same starting conditions and constraints. My oppinion is that it does, but I do not have more evidence than an interesting undescribed before theoretical insight and results obtained so far.
Rodolfo
I have always taken "error correction" as descripive of a strategy, much more than an end.
We must acknowledge any correction strategy which relies on taking an action "after the fact" cannot fulfill its goal in entirety without violating the causality principle, more on this later.
Negative feedback can achieve its objective asyntotically with infinite gain.
Error correction can only succeed at DC, which also does not exist ( any power supply must have been turned on after the big bang at least).
Perfect correction can only be attained with forecknowledge of the distortion ahead, this is a provoking idea worth being considered now we have digital technology in principle capable of modeling and self adjustment, and is in my horizon as long as I can devote time and resoruces in this direction.
For the time being, I think the issue is whether the error correction strategy can bring practical benefits when compared with conventional negative feedback, given the same starting conditions and constraints. My oppinion is that it does, but I do not have more evidence than an interesting undescribed before theoretical insight and results obtained so far.
Rodolfo
wwood said:
Speaking only for myself (I suspect Bob is more or less on the same line as per previous posts), I monitor the output spectrum and adjust for a minimum of spurious content. This minimum I found to be non-critical and stable.
Rodolfo
PS. It is kind of nice, to have a "minimum distortion adjustment".
ingrast said:
Speaking only for myself (I suspect Bob is more or less on the same line as per previous posts), I monitor the output spectrum and adjust for a minimum of spurious content. This minimum I found to be non-critical and stable.
Rodolfo
PS. It is kind of nice, to have a "minimum distortion adjustment".
If you have only a multimeter you can measure the DC between the output stage input and its output (assuming your ec is on the output stage) and null that. Ec nulls DC offset as well as dynamic distortion.
Jan Didden
traderbam said:
What Bob did in his original article was to show a linear system, with the distortion injected as an additive signal. You say that your system has no non-linear elements and I do not see the injection of any error term and therefore I cannot make sense out of it without making assumptions. You seem to be saying that I made some incorrect assumptions, however I see no way that your system is identical in behavior to Rodolfo's, for the non-linear case.
It is true that the linear part of the model can be simplified, however it makes sense for clarity to show the subtraction around the output stage and thus clearly show how the error feedback term is produced.
I believe that the key word/difference between traditional feedback and EC, is null. We null out the signal, in order to extract the error in isolation of the signal (Ideal case - with an excellent approximation in real hardware) which can then be fed back without this path setting the gain of the system. A traditional feedback system, feeds back signal + error, and therefore the closed loop gain is tied to the magnitude of distortion reduction. This is a fundamental difference between the two, and as several have stated, reducing the distortion to zero in a traditional system is impossible even in an ideal system (with error injection). The defining equations are obviously different.
Pete B.
janneman said:
Jan, this is true, and for a good reason.
It has been noted previously that for the case of a first order summing node model, the inner unity loop gain positive feedback block can be replaced by an ideal integrator with the same results. In fact it can be seen that for any summing node causal network, the block behaves as a generalized integrator, more precisely there is at least one pole at s=0.
This being the case, it follows that for its output to be bounded - the drive signal for the amplifier to be corrected -, then the average (DC) input must by 0. This is taken care of by the outer negative feedback branch.
Rodolfo
PS. Anyway I confess to be more comfortable adjusting with a spectral view.
Rodolfo,
I find that global feedback and error correction do interact. Agree that you cannot just remove the GNF, but you can move it to the vas and then do the adjustments.
With GNF around the EC, I find the distortion minimum to be somewhat broad and ill definded (depending on how much GNF there is) and without the GNF around the EC I find the distortion minimum to be better dfined ... not surprising.
If I try to minimize Zout with GNF around the EC, the EC gets grossly mis-adjusted. Minimizing Zout with no GNF around the EC tends to minimize distortion when GNF is re-connected around the EC, however the Zout will not necessarily be minimum after GNF is added.
I will have to try Jan's approach with and without GNF around the EC.
Bill
I find that global feedback and error correction do interact. Agree that you cannot just remove the GNF, but you can move it to the vas and then do the adjustments.
With GNF around the EC, I find the distortion minimum to be somewhat broad and ill definded (depending on how much GNF there is) and without the GNF around the EC I find the distortion minimum to be better dfined ... not surprising.
If I try to minimize Zout with GNF around the EC, the EC gets grossly mis-adjusted. Minimizing Zout with no GNF around the EC tends to minimize distortion when GNF is re-connected around the EC, however the Zout will not necessarily be minimum after GNF is added.
I will have to try Jan's approach with and without GNF around the EC.
Bill
ingrast said:
Rodolfo,
Acknowledging that our respective topologies are different, I can relate to the integrator model. In fact, if you would do a high-level simulation on the model in my earlier posted AES paper, you will see that the lower you go in frequency, the the deeper the distortion null. I remember some sims where I got reductions of several hundred dB's at fractional Herz frequencies, going down at 6dB/oct. This clearly shows the underlying integrator action even if it is not reachable in practical cases.
Jan Didden