mikeks said:
Mikeks,
I appreciate all of the work that you've done, on this. I have been trying to follow this part of the discussion, even though my knowledge of control theory and mathematics is a bit rusty. But, mainly, I have found it to be difficult to re-navigate the thread, well-enough, due to the number of messages involved, and also all of the links to previous posts, intervening and conflicting posts, corrections, repetitions, et al.
I don't like to be a bother. But, do you already have, or can you provide, a summary, or a single document, or, maybe easiest, a properly-sequenced list of links, that includes all of the really-relevant stuff, so far, "all in one place"? That would also make it much more accessible for everyone who tunes-in later.
If that's not practical or desirable, from your point of view, please feel free to just forget that I asked.
Thanks.
- Tom Gootee
janneman said:
Well, in electronics like in real life, brute force often has the edge. But playing with ec is quite interesting. Notwithstanding the discussions here that it is equivalent to nfb, the character, if I can call it that, of an ec amp is definitely different from a nfb amp.
You really have to think differently, there is no longer the overriding concern to get very high gain. What strikes me is the almost complete adsense of oscillatory tendencies, ec circuits (at least in my experience) tend to be very docile.
If you are looking to something different from the tired old ways, ec amps are worth a try!
Jan Didden
Jan, there are many ways to Rome....I look always for the shortest (corresponding to my knowledge)
Till today I have never seen EC design which beats NFB.
The proof of the pudding can be, to improve a classD stage with it´s "nice" filter at the end. Here end the way for NFB....
I dont´t try it till now and I´m not all too optimistic.
For the same purpose I did some simulations with feedforward EC.
This works quite good but it ends always with a far to big correction amp... at that time I also missed a good idea!
Regards
Heinz!
powerbecker said:
Heinz!,
Thanks so much for the encouraging words!
You are probably being way too kind. And I probably just got lucky, accidentally. But still, it's always great to hear some praise, especially since I have only been dabbling in electronic design for the last several years, and only seriously for the last two years, after doing no EE work at all for about 15 years, and knowing mostly only control theory and not too much about actual circuits, before that.
Thanks again! Back to the circuits...!
- Tom Gootee
Mikeks wrote:
The scheme you have described, Mike, is one where a new PFB loop is added to the original one so that if the gain generated by the original should decrease significantly (which it does when one of the summers does not have a gain of 1 - the infinite gain condition of your example circuit) then naturally having the extra PFB circuit will make up the difference. When the original has infinite gain the extra PFB doesn't add anything.
But the whole thing is a complete waste of transistors when you realise that the original PFB loop is redundant and can be replaced by a simple integrator that completely removes the dependence of the circuit on summer gain tolerance.
I stress once again that generating gain by means of a PFB loop MULTIPLIES the distortion of the components within that loop. Putting a PFB loop around the output stage, which is the source of the distortion one is trying to minimize, is counter productive.
For those of you who are new to control theory, please think about why Hawksford comes up with all these ideas. Think carefully about that.
The scheme you have described, Mike, is one where a new PFB loop is added to the original one so that if the gain generated by the original should decrease significantly (which it does when one of the summers does not have a gain of 1 - the infinite gain condition of your example circuit) then naturally having the extra PFB circuit will make up the difference. When the original has infinite gain the extra PFB doesn't add anything.
But the whole thing is a complete waste of transistors when you realise that the original PFB loop is redundant and can be replaced by a simple integrator that completely removes the dependence of the circuit on summer gain tolerance.
I stress once again that generating gain by means of a PFB loop MULTIPLIES the distortion of the components within that loop. Putting a PFB loop around the output stage, which is the source of the distortion one is trying to minimize, is counter productive.
Moderator's edit: The posts have been moved to the Studio amplifier thread
powerbecker said:
Jan, there are many ways to Rome....I look always for the shortest (corresponding to my knowledge)
Till today I have never seen EC design which beats NFB.
The proof of the pudding can be, to improve a classD stage with it´s "nice" filter at the end. Here end the way for NFB....
I dont´t try it till now and I´m not all too optimistic.
For the same purpose I did some simulations with feedforward EC.
This works quite good but it ends always with a far to big correction amp... at that time I also missed a good idea!
Regards
Heinz!
Heinz!,
Don't give up on it, quite yet!
Comparing to my last posted circuit (EC_Amp-tg3.asc), but with a slightly-modified EC topology, it LOOKS like I have just found a simple way to decrease the ec amp gain by a factor of more than 10, while retaining the same THD level (and also improving the steps/stability), even when not using an input filter!
Alternatively, I can keep the ec amp gain the same but get about 1/2 the THD, and not change the step response.
Even better, it looks like this effect "should" be able to be further "scaled", so that the improvements above could be increased by larger factors, which is what I am trying, right now.
I am still investigating how far I can take it, et al, and hope to be able to provide details, soon.
- Tom Gootee
As shown here, making K4=1 reduces THD roughly by half for a 10% misalignment in the gain of S1.
This is consistent with Hawksford's averment that sensitivity to such is reduced by 1/(1+K4).
Indeed, his recommended connection of a resistor Rs between the output of the power stage to the input of the error-cancellation loop necessarily sets K4=1.
While desirable in principal, this is clearly inadequate in practice: what is required is some means to effect K4>>1.
Discuss.
Text changed by request of Mikeks, Dec,9,2006.
This is consistent with Hawksford's averment that sensitivity to such is reduced by 1/(1+K4).
Indeed, his recommended connection of a resistor Rs between the output of the power stage to the input of the error-cancellation loop necessarily sets K4=1.
While desirable in principal, this is clearly inadequate in practice: what is required is some means to effect K4>>1.
Discuss.
Text changed by request of Mikeks, Dec,9,2006.
gootee said:
Hello Tom,
I never give up!
But there are some little things to :
Revenue waits for my monthly vat declaration
, my client wait for the specs from the next to last project, I had to check the layout from my own current work, I had to design a simple 500kHz 5kV source, friends are waiting for a call and..... and what do I :Sitting in front of PC and check out my last idea!!
Tom keep on, I´m curious about your results!
Regards
Heinz!
Hi Mikeks! Below picture says my thought. What do you think of it ?
Can we calculate distortion of T1 ?
I tried to simulate Error Correction schematic. Sim show that distortion still quite hight. Is THD of EC still high ? Are You, tradebarm, AndyC, powerbecker, gootee trying optimize more EC ?
thanks!
Can we calculate distortion of T1 ?
I tried to simulate Error Correction schematic. Sim show that distortion still quite hight. Is THD of EC still high ? Are You, tradebarm, AndyC, powerbecker, gootee trying optimize more EC ?
thanks!
Attachments
Hi, Thanh,
I think (not calculate) that EC capability of this CCT is depended on T1 and T2. The EC will be limited by the nonlinearity of T1-T2 itself.
Can you simulate, for example put darlington for T1-T2, does it change anything?
Or put a trick, to make T1-T2 heavily biased (to get linear operation on T1-T2)?
I think (not calculate
) that EC capability of this CCT is depended on T1 and T2. The EC will be limited by the nonlinearity of T1-T2 itself.Can you simulate, for example put darlington for T1-T2, does it change anything?
Or put a trick, to make T1-T2 heavily biased (to get linear operation on T1-T2)?