Could looking at this in completely the wrong way. But if you break the HEC loop the spike still happens. Add that cap back in series with R4 and the spike dissappears. The loop gains become equal from about 20Khz up with the HEC loop broken between R16/18 and C5 in my amp. Is this change in loop gain for Tzan significant?
This thread has created more questions than answers in my mind.
Better to try and have an understanding than to blindly move forward...
Paul
This thread has created more questions than answers in my mind.
Better to try and have an understanding than to blindly move forward...
Paul
Hi Paul
I have posted a version of your amp with Beta release versions of my common mode balun and also the auxiliary Tian Probe.
That can be used to demonstrate probe placement and the like.
Now that I have reasonable Betas, that should be more convenient for you than to remove my development versions from your work.
Best wishes
David
I have posted a version of your amp with Beta release versions of my common mode balun and also the auxiliary Tian Probe.
That can be used to demonstrate probe placement and the like.
Now that I have reasonable Betas, that should be more convenient for you than to remove my development versions from your work.
Best wishes
David
Last edited:
That point does not break the Feed-Forward that I believe causes the problem.
The FF occurs via R15 R16.
Best wishes
David
This is not the way I considered it but perhaps a different perspective.
I mentioned to Paul earlier that I liked the database concept of "normalization", that a value should not be determined in two places or two ways.
This idea prompted me to use MIC compensation in my own amp, so that there were not two compensations loops, one around each polarity of the push-pull VAS.
In the HEC case the two paths interact all the time, not just when they are put out of synchronisation by some event as you propose, so that part of your post probably needs to be reconsidered.
Best wishes
David
By the way, should I call you M. Box or ...?
Last edited:
The GNFB and the HEC loop, contrary to something low-speed as a DC servo or auto-bias.Please define "high-speed [feedback] loops".
I didn't mean any different, I understand that the loops interact all the time. I was trying to describe that under normal operating conditions the loops wouldn't work against each other, but could in the event of an exception to the normal operation, like that of coming out of clip.
As for my name, it's always been like that. Though here on DIY it might as well have been MusicBox. more appropriate. My real name is so dull
Anyways, I think I fixed my HEC on the coming out of clip issue
- Low bandwidth # "low speed".
- High bandwidth # "high speed".
- Auto bias is not low bandwidth (if it would, it would defeat the very purpose).
Hi M. Box
I understood your description and it is what one would expect but I think it is not correct, it seems HEC amplifiers act in an unexpected way.
I think the two loops do work opposed to each other under normal conditions, not just in an exception.
As the frequency is increased the dominant role passes from one to the other.
That occurs at the peak in the Tzan plot, notice that this is non-minimum phase, the phase drops even as the response peaks.
That sort of non-minimum phase behaviour is characteristic of two alternate paths in opposite phase.
When the two feedback paths cancel each other then the -ve feedback is reduced and so the gain peaks.
An ideal unilateral buffer in the HEC circuit makes the peak vanish.
Clip issues are another problem, I haven't even considered that yet, nice to read you solved yours.
Best wishes
David
P.S. I don't even know what is considered a dull name in the Netherlands, the equivalent of "John Smith"?
Last edited:
OK. while we wait for response on HEC.
I did choose my phrase with some care, here's the rationale.
We can abstract and simplify to produce different level models.
For instance, the most basic model of an op-amp has no output impedance, draws no input current, no limit to gain, has no poles, etc. etc.
But this is still a useful model of an op-amp, we can add a feedback network and it works - perfectly in fact.
Then we can add complexity to the model to capture reality better.
Include a Gain limit, finite input impedance, add a slew rate limit and so on.
These are secondary effects, not essential to "op-amp" behaviour.
Similarly, the most fundamental "CFA" model has zero output impedance and infinite +ve input impedance, just like an op-amp, but the -ve input impedance is zero.
For normal use the -ve input is used for feedback, hence my comment that 'the foundation of a "CFA" is lower impedance on the feedback node'.
The other properties you list are not the foundation.
Slew rate, for instance, is mainly coincidental with certain typical circuitry.
VFAs can be produced with no first order slew rate limits, as Stochino has illustrated.
I personally use a MIC scheme rather than Miller compensation and this also practically removes slew rate limits.
The fact that "CFA"s typically have lower gain is also only incidental.
With their typical application there is little need for more gain, it could certainly be done.
So, while these are important factors in a practical amplifier, they are not inherent in a VFA or "CFA", even if typical implementations conform.
Best wishes
David
I did choose my phrase with some care, here's the rationale.
We can abstract and simplify to produce different level models.
For instance, the most basic model of an op-amp has no output impedance, draws no input current, no limit to gain, has no poles, etc. etc.
But this is still a useful model of an op-amp, we can add a feedback network and it works - perfectly in fact.
Then we can add complexity to the model to capture reality better.
Include a Gain limit, finite input impedance, add a slew rate limit and so on.
These are secondary effects, not essential to "op-amp" behaviour.
Similarly, the most fundamental "CFA" model has zero output impedance and infinite +ve input impedance, just like an op-amp, but the -ve input impedance is zero.
For normal use the -ve input is used for feedback, hence my comment that 'the foundation of a "CFA" is lower impedance on the feedback node'.
The other properties you list are not the foundation.
Slew rate, for instance, is mainly coincidental with certain typical circuitry.
VFAs can be produced with no first order slew rate limits, as Stochino has illustrated.
I personally use a MIC scheme rather than Miller compensation and this also practically removes slew rate limits.
The fact that "CFA"s typically have lower gain is also only incidental.
With their typical application there is little need for more gain, it could certainly be done.
So, while these are important factors in a practical amplifier, they are not inherent in a VFA or "CFA", even if typical implementations conform.
Best wishes
David
Last edited:
Dave, I did (and always where required) qualify my statements on VFA with 'MC comp'd'. I agree more advanced VFA comp schemes deliver very high slew rates - MIC of course being the most popular example.
My input on this VFA / CFA discussion has been around clarifying what it is that separates the two - so I used the basic topologies to explore the differences and came up with the two tests and two pointers. Of course, we can easily design high loop gain CFA's ( many examples on this forum), but to do so means we then have to resort to comp schemes that are usually deployed to deal with higher loop gains and accumulated phase shift. The result is that the performance characteristics of a CFA then tend to emulate high loop gain VFA's.
I have designed and built both topologies. On the CFA, I showed a high loop gain example (somewhere on the forum) that simmed at single digit ppm at >400 W out into 8 ohms (20 kHz). The ULGF in that design was a respectable 2 MHz with c. 60 degrees PM
But, as I have accumulated more practical design and construction experience, my preference has increasingly lead me towards simpler designs.
My input on this VFA / CFA discussion has been around clarifying what it is that separates the two - so I used the basic topologies to explore the differences and came up with the two tests and two pointers. Of course, we can easily design high loop gain CFA's ( many examples on this forum), but to do so means we then have to resort to comp schemes that are usually deployed to deal with higher loop gains and accumulated phase shift. The result is that the performance characteristics of a CFA then tend to emulate high loop gain VFA's.
I have designed and built both topologies. On the CFA, I showed a high loop gain example (somewhere on the forum) that simmed at single digit ppm at >400 W out into 8 ohms (20 kHz). The ULGF in that design was a respectable 2 MHz with c. 60 degrees PM
But, as I have accumulated more practical design and construction experience, my preference has increasingly lead me towards simpler designs.
Last edited:
Hi Andrew
I appreciate that what you have is practically useful but I would call them four pointers. I think the real test is lower -ve input impedance.
This seems like a nice example of "normalization" that I mentioned earlier.
If you have two tests then there is the possibility that one is inconsistent with the other, better to have one test that is definitive.
Unless it's not, of course
Do you have an example where my definition wouldn't work? That sounds pretty impressive. Can you remember where or find the ASC?
I'd like to look at that and hopefully learn.
Yes, I also want simpler amps and better performance, and that requires a rethink of the basics.
So I don't want to just look at the usual circuits and that's why I try for a more universal definition of "CFA".
For typical circuits we are clearly in accord.
Best wishes
David
P.S. Have you looked at HEC? More complicated than I like but it's very educational to try to work out how to analyse it.
It seems that the Middlebrook GFT approach is useful here, so it's on topic.
Last edited:
Hi Andrew
I appreciate that what you have is practically useful but I would call them four pointers. I think the real test is lower -ve input impedance.
This seems like a nice example of "normalization" that I mentioned earlier.
If you have two tests then there is the possibility that one is inconsistent with the other, better to have one test that is definitive.
Unless it's not, of course
That sounds pretty impressive. Can you remember where or find the ASC?
I'd like to look at that and hopefully learn.
Yes, I also want simpler amps and better performance, and that requires a rethink of the basics.
So I don't want to just look at the usual circuits and that's why I try for a more universal definition of "CFA".
For typical circuits we are clearly in accord.
Best wishes
David
P.S. Have you looked at HEC? More complicated than I like but it's very educational to try to work out how to analyse it.
It seems that the Middlebrook GFT approach is useful here, so it's on topic.
I think it was on the CFA vs VFA thread somewhere. Otherwise, I'll have tomtry and dig it up on my computer.
Re HEC - no. I am looking at a practical amp using AFEC. I think for MOSFET amps HEC is well suited - for BIP, AFEC seems a better bet. I've updated the article and clarified a few points - managed also to generate the LF plots by the way. It's up on my website.
CFA are in fact dominant pole compensated - its just that CFA Ccomp is a internal shunt C, part of the CFA fundamental tz gain parameter rather than a Miller multiplied local feedback C
and open loop gain is determined by tz real part - open loop gains of >100k are possible - were offered in the AD846
and open loop gain is determined by tz real part - open loop gains of >100k are possible - were offered in the AD846
Last edited:
CFA models use transimpedance and 1/gm between the + and - inputs as basic parameters often tz or Z_t, and rx or R_in
raw open loop gain then is ~tz/rx before the feedback Z is added to rx
there are lots of different modeling papers, but you should be able to puzzle out the meaning after seeing the common pieces in a few
raw open loop gain then is ~tz/rx before the feedback Z is added to rx
there are lots of different modeling papers, but you should be able to puzzle out the meaning after seeing the common pieces in a few
lazy man's model - just a Spice gm up front http://www.diyaudio.com/forums/soli...terview-error-correction-250.html#post1328496
asc? http://www.diyaudio.com/forums/soli...nterview-error-correction-73.html#post1062892
Qmodels, some comments next page http://www.diyaudio.com/forums/soli...terview-error-correction-248.html#post1325081
asc? http://www.diyaudio.com/forums/soli...nterview-error-correction-73.html#post1062892
Qmodels, some comments next page http://www.diyaudio.com/forums/soli...terview-error-correction-248.html#post1325081
Excellent, thanks, did you ever do any more on this?
Paul's HEC amp looks like it has a fundamental "issue" and your plot of Bob's seemed to imply the same there, haven't reread the thread yet.
I am surprised no one else seems to have noticed this, or have I missed some reference?
Jan Didden's Pax amp uses a different technique for HEC and may not suffer the same concern, haven't looked closely yet.
Best wishes
David
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.