not if I get to place the gain probe where it actually measures the "right" loop gain for predicting stability
my recent analysis has shown (again) that there is a slight advantage to "TMC" due to VAS/TIS loading differences - but the fundamental stability issues, having to design with the "inner" loop gain 2-pole curve's added phase is necessary for safe application of TMC
TMC designed to the same properly measured stability margin won't beat "global" 2-pole by more than single digit dB
my recent analysis has shown (again) that there is a slight advantage to "TMC" due to VAS/TIS loading differences - but the fundamental stability issues, having to design with the "inner" loop gain 2-pole curve's added phase is necessary for safe application of TMC
TMC designed to the same properly measured stability margin won't beat "global" 2-pole by more than single digit dB
Last edited:
It's really the other way round. It is usually assumed that the feed-forward zero is of theoretical interest only, turns out it can have a practical impact, so worth a check if one tries experimental circuitry. Not needed by people who just clone (or rip-off if, as often, uncredited) Mr Self's work.
Exactly what I think! I wrote "TMC can work better" in post #358.
Nice that we have reached accord.
Not what I wrote. I said "can" quite deliberately.
If you want to put it in quotation marks then you should reference whoever this is quoted from, then address your dissent to them.
Exactly. There is a metric for the maximum amount of feedback possible, based on Bodes's Gain-Phase conserved quantities.
Approximately the product of the forward transfer unity gain frequency and some function of the steepness of the slope around that frequency, for a minimum phase system.
I think TMC performs a bit better on this metric, consistent with JCX's posts.
I have not yet worked out the details, an exact quantification would require the phase from DC to infinity, which is a bit excessive.
An approximation should be informative anyway. Any constructive comments welcomed. A practical reference to the discrete Hilbert transform would be nice.
Best wishes
David
again I'm not so concerned about a few dB - I think too many don't get that far before jumping on TMC as a "free lunch", yet denying its 2nd order loop gain and stability requirements
some are making much a bigger conceptual mistake by thinking TMC is "as safe" as conventional Miller because of the apparent single pole loop gain seen just cutting the output feedback loop
designing TMC with the mindset that it is just like conventional dominant pole Miller compensation is likely to bite someone with the lack of stability margin evidenced in my added delay sim
some are making much a bigger conceptual mistake by thinking TMC is "as safe" as conventional Miller because of the apparent single pole loop gain seen just cutting the output feedback loop
designing TMC with the mindset that it is just like conventional dominant pole Miller compensation is likely to bite someone with the lack of stability margin evidenced in my added delay sim
there's still the higher level Bode Integral argument - if you can get the same shape of total loop gain around the output stage with TPC as TMC, albeit with different component values, then we do expect "the same" reduction in distortions inside that(those nested) loop(s)
I did a sim a while back with pretty much one of Bob's book example circuits
http://www.diyaudio.com/forums/soli...lls-power-amplifier-book-144.html#post2445393
with the slow MJL21192/3 power Q models from Cordell, 4 Ohms simulated load - the two properly measured gain curves matching for TPC vs TMC
I get 20 Vrms, 20 kHz ftt harmonics within fractions of a dB
the delay really slows my sim - you may want to cut them out for the distortion sims
I'm pretty sure those TPC, TMC component values were hand tweaked for matching the gain curves so there is the chance that neither is "optimal" for their respective compensation topology
so the question is does "optimum" alignment of both TPC, TMC with essentially identical high frequency intercept and phase give a few dB advantage to TMC at lower frequency?
can it be seen with double ef output (triple, HEC)..., or does it require lower loading on the VAS/TIS - hence Edmond's search for different TIS and output buffer stage to "get the most from TMC"
I did a sim a while back with pretty much one of Bob's book example circuits
http://www.diyaudio.com/forums/soli...lls-power-amplifier-book-144.html#post2445393
with the slow MJL21192/3 power Q models from Cordell, 4 Ohms simulated load - the two properly measured gain curves matching for TPC vs TMC
I get 20 Vrms, 20 kHz ftt harmonics within fractions of a dB
the delay really slows my sim - you may want to cut them out for the distortion sims
I'm pretty sure those TPC, TMC component values were hand tweaked for matching the gain curves so there is the chance that neither is "optimal" for their respective compensation topology
so the question is does "optimum" alignment of both TPC, TMC with essentially identical high frequency intercept and phase give a few dB advantage to TMC at lower frequency?
can it be seen with double ef output (triple, HEC)..., or does it require lower loading on the VAS/TIS - hence Edmond's search for different TIS and output buffer stage to "get the most from TMC"
Last edited:
Emerging or causing. You can't really consider the VAS in total isolation from the output stage it is supposed to drive. Since generally the VAS provides most of the voltage gain, the VAS output node is almost always very high impedance. Any non-linear input current into the output stage will ruin the VAS linearity.
Some EC circuits do cause non-linear loads on the VAS, unless there's a buffer in between.
So if a non-linearity turns up with an EC, it might well be that the EC actually causes it, even if it makes the output stage more linear.
Jan
Never write such things about a person from whom you don't know the slightest thing about the experience of building she or he has.
Currently, there precisely is on my bench a push-pull VAS after a Samuel Groner's schematics (not the one published in Linear Audio), mainly built for curiosity, some months ago.
Ok, I'll bite. Please show a Miller compensated circuit where the RHP zero is of a practical importance (that is, significantly affecting the stability).
For the rest of the TMC vs. TPC, jcx said it all above. Think Bode integrals and let go. A few % difference favoring TMC can result, for a particular amplifier topology, e.g. by mechanical Wye-Delta transforming the TMC network and calculating the new TPC values. This is not necessary an optimal TPC (at least since some simplifications are required during the transformation). However, the Bode integrals are saying that another optimal TPC does exist, even for something like Edmond's TIS. How is that TPC looking theory doesn't say, though. Interesting enough, TMC is not even necessary always optimal, providing the Maximum Feedback, among all possible second order compensation methods.
Yes.
More than a chance I think.
My basic principle is that the lower load from the TMC should push up the Gain-Bandwidth-Product (or similar measure, as in previous post) that can potentially be used to lower the distortion, or put in (more of) a Bode Step to improve the stability or both. At worst the extra gain can be burned off with shunt compensation equivalent to TPC.
I continue to consider your earlier post and look for an idea.
Just feels like there should be some Middlebrook-style clever approximation that makes it all obvious. Unfortunately my earlier idea wasn't it!
Best wishes
David
This is the plot that started the discussion.HERE.
Whether the loss of GM counts as "of a practical importance" is somewhat a matter of opinion, so I expect I can predict yours!
But it's sufficiently close to make me concerned that a fairly conventional circuit can show obvious non-minimum phase behaviour, which was my point in the OP.
Best wishes
David
Actually, now I look at it after a year, it is not obvious that the zero is from feed-forward.
Also that I didn't check the inner loop. That may show worse effects, since the ULGF would be increased.
Last edited:
Actually it's not optimal. That was one of the points in my second Linear Audio article.
But it's a reasonable approximation of Bode Maximum Available Feedback.
The peak near the lower frequency poles, seen on the Bode plot of most TPC schemes also has potential to improve the approximation to BodeMAF.
It would be nice to combine that low frequency peak with the reduced load of TMC up near the ULGF.
It is hard to approximate sufficiently to make the problem understandable and yet still retain the essentials.
Harry Dymond simplified to make his JAES article very clear, but lost the essentials of that peak, unfortunately. So I want to build on that.
Best wishes
David
Regarding my opinion, you got that right, it makes me anything but concerned. You plot shows that for all practical purposes, audio amplifiers are minimum phase systems.
Hi Jan,
That's exactly what I have observed with Bob's HEC OPS driven by the SuperTIS. See (again) this page at the bottom (fig.17). An additional buffer solved it.
Cheers, E.
The plot shows at least 6dB of deterioration in outer loop GM from non-minimum phase behaviour.
This is borderline acceptable, in one more-or-less randomly selected sample.
To make any assertion about "audio amplifiers" as a body, and "for all practical purposes" is a total non sequitur.
I actually trained first as a statistician so I can safely say that as an example of inference, yours wouldn't pass.
It is a mystery to me why your desire to quibble or to try to point-score leads you to make statements that you would rubbish if written by anyone else.
Best wishes
David
At frequencies at which there are many other practical things to worry about (wiring, capacitors ESL, parasitic capacitances, etc...).
Am I correct that you never posted the schematic for which you got those plots for? If so, please do. I would be curious to analyze that zero that, it seem not to be due to Miller. That's because it's at a frequency that's much lower than a common Miller loop ULGF.
I see, Q21, 22. It is interesting, isn't it - to effect the EC, the corrector circuit has to modulate the VAS (I mean TIS 🙂) output, therefor needs a certain VAS Zout to work against. But unless the VAS Zout is linear, the modulation also generates additional VAS errors that cannot be corrected by the EC because the EC of course doesn't see that.
Took me some time to find that one!
Jan
- Status
- Not open for further replies.