Jcx was more concise (and hopefully more credible than yours truly), about. I'm happy to see I'm not alone. Unfortunately, even for audio amplifier loop gain and stability margins, there is no free lunch.
Yes, JCX is absolutely correct about this.
What is interesting to me about schemes like these is whether they offer cheaper and/or more power-efficient means of achieving lots of feedback in the audio band.
You are right: there is no free lunch. But some lunches cost more than others. TPC and TMC come to mind.
Cheers,
Bob
what's wrong with TPC and TMC?Do you have an amplifier or compensation design that you've contributed to this forum? If so please could you direct me to it?
Nothing wrong with them, but I prefer TPC for sound technical reasons.
Your memories are short or lacking. We ("we" obviously not being limited to the present contenders) had a long debate about TMC vs. TPC here and in other threads, and my opinion about TMC vs. TPC (including TMC as an oversold concept) can be found by using the Search tool.
It took over a year for Mr. Cordell to reach from unconditionally stating that TMC is a largely superior compensation method (compared to TPC) to his current more balanced opinion, and the lunch metaphor (which he used before).
I thought I am entitled to post a small friendly joke, but apparently some of you guys are lacking any sense of humour. Good night.
No, and for good reasons (which you don't want to know about) it's not going to happen anytime soon, sorry. Good night.
I think the joke was too easily misread. I thought you were questioning the utility of both TPC and TMC.
I do agree with you and Mike that TMC was initially oversold. However, what I find intriguing is that most people who have tried the two in "real life" and measured the performance of both, report lower distortion with TMC.
Harry, just before going to bed, here's a disclosure:
I am specializing in mixed signal ASICs. As linear amplifier compensation is a topic of interest for me, and having the opportunity to run several projects through Europractice, I took the opportunity to evaluate the TMC vs. TPC compensation method, on small gain blocks (think mosfet opamps macrocells). I was able to confirm through reliable measurements in our university lab that TMC and TPC are exactly equivalent. This should be understood in the sense that given a TMC compensated amp, there is a TPC compensation that provides exactly the same loop gain and phase margin.
A Canadian friend of mine (to which I am mucho indebted) developed an analytical way to calculate the exact TPC equivalent compensation elements, starting from the TMC compensation values, and I was able to test this theoretical result (to an under 1% error). To me, this concludes the story, everything else is either anecdotal or de gustibus preference (which I accept as valid). What others measured on power amplifiers, I have no idea, but I am not under the impression this forum is anywhere focused on reliable lab measurements.
BTW, FWIW, me and Mike disagree on his point that TMC and TPC are optimized and interchangeable using the same component values. This is factually incorrect, and can easily be proven in Spice.
I also think it was oversold but TMC does seem a little superior to TPC when second order effects are considered. This does not violate JCX's (and other's) observations about Bode feedback integral conservation. The fact that both schemes are subject to the same limit does not mean they are equally effective.
More details in Linear Audio
The practical results reported tend to confirm this analysis.On the subject of practical results, well, simulations too, no one has commented on the actual evidence for non-minimum phase I posted, twice.
Harry, you seem to have avoided the "stupid ray" or what ever it is that affects people so they write abuse rather than actually look at any data.
Any comments on those Bode plots? Anyone?
Best wishes
David
I don't accept that lower distortion is obtainable with TMC compared with TPC. For a fair comparison, one has to compare apples and apples, which requires that the compensator component values be made identical in both cases.
I must have missed it, but for your information a non-minimum phase system is one with singularities in the RHP, be they zeros or poles.
That makes a TIS a non-minimum phase system.
Mike, this is quite wrong. In this instance, I must agree with Waly. The same component values are not necessarily optimum for TPC and TMC. Insisting on your version of an apples-apples comparison, where identical components are used, over-constrains the problem. Think about the ratio of C2/C1.
The fair comparison is which one can deliver lower amplifier distortion with a given level of amplifier stability. This of course depends on the details of the amplifier. As you have correctly pointed out in the past, TPC reduces input stage distortion while TMC does not. Amplifiers that suffer serious distortion in the input stage may benefit more from TPC. But most competent designers end up with amplifiers where the input stage is not nearly the dominant problem for distortion.
The nearly-identical loop gain around the output stage for some TPC/TMC designs (often C1 = C2, for example) can easily seduce one into thinking that TPC and TMC are largely the same. They are not. That they are different does not mean that one is ALWAYS better than the other.
There are also differences between the two that can become important in amplifiers where there is moderate phase lag in the output stage.
Keep your eyes open for the next LA.
Cheers,
Bob
I disagree. Profoundly.
The only fair comparison is one that guarantees the same loop gain about the output stage with TPC and TMC, and, simultaneously, gives virtually the same unity loop gain frequency and stability margins.
The above can only be guaranteed by using the same compensator component values for TPC and TMC as I have previously demonstrated.
I don't think the AFEC scheme is a free lunch, but if you are after lower distortion, it is a useful technique. And as pointed out, you get DC offset servoing as a byproduct (that's a free lunch in my book).
You could argue that TMC and TPC are simpler. On power amplifiers, my investigation into TMC seems to indicate about 5x reduction in 20THD is doable, but not much more than this, whereas AFEC easily gets you to 20 dB or more.
I think the equivalency with simply raising the overall loop gain using this method is accepted, an no one is claiming it is without its problems and will need further development. It should be seen against the back drop of some other distortion reduction schemes which in my book are much more complex
OTA means Operational Transconductance Amplifier.
The two first stages of most power amplifiers are Miller compensated OTA .
I was referencing these authors to start from a common understanding of the open loop Miller OTA transfer function and its well known approximations. If you take the generic transfer function as explained in let say Lundberg and if you simplify for widely separated poles and large Cmiller you can get:
To summarize (for large Cmiller and gm2):
gain bandwith : gm1/2piCmiller
freq of zero : gm2/2pCmiller
non dominant pole : gm2/2piCload this non dominant pole is created in the OTA, it is not a non dominant pole coming from the class AB buffer poles although the input capacitance of this buffer is included in Cload.
If Cmiller is >> Cload the zero is at lower frequency than the non dominant pole. From a phase point of view, a positive zero acts as a pole. To be able to stabilize such a circuit, this singularity should be at least 3 times the GBW frequency to have enough phase margin ( rule of thumb), this can be achieved with gm2>3 gm1
In this case we would have a Miller OTA showing all pass property around the zero. There are of course well known ways to avoid this trap like a resistor in serie to make the zero negative or making the capacitor path unilateral ( Ahuja) . Then the positive zero disapears .
If these tricks are not used, the OTA stage is non minimum phase because it has a positive zero and the amp with feedback is still non minimum phase because if I recall well a basic property of control theory: the zeros of the open loop transfer function are the zeros of the closed loop function.
Using Blacks formula: F(s)= G(s)/1+G(s) H(s) if G(s) = N(s)/D(s) then
F(s) = N(s)/(D(s) + N(s)H(s))
I am not posting to make a point or handwaving but to simply trying to understand what I found intriguing in Bob Cordell post about all pass.
All pass and minimum phase is semantic discussions but the existance of a positive zero and the parameters influencing it are relevant
Please remain educated in your reaction. I was asked to explain and I did it. I can be wrong and I will accept the demonstration but I accept a demonstration and not agressive assertions close to be insulting.
The two first stages of most power amplifiers are Miller compensated OTA .
I was referencing these authors to start from a common understanding of the open loop Miller OTA transfer function and its well known approximations. If you take the generic transfer function as explained in let say Lundberg and if you simplify for widely separated poles and large Cmiller you can get:
To summarize (for large Cmiller and gm2):
gain bandwith : gm1/2piCmiller
freq of zero : gm2/2pCmiller
non dominant pole : gm2/2piCload this non dominant pole is created in the OTA, it is not a non dominant pole coming from the class AB buffer poles although the input capacitance of this buffer is included in Cload.
If Cmiller is >> Cload the zero is at lower frequency than the non dominant pole. From a phase point of view, a positive zero acts as a pole. To be able to stabilize such a circuit, this singularity should be at least 3 times the GBW frequency to have enough phase margin ( rule of thumb), this can be achieved with gm2>3 gm1
In this case we would have a Miller OTA showing all pass property around the zero. There are of course well known ways to avoid this trap like a resistor in serie to make the zero negative or making the capacitor path unilateral ( Ahuja) . Then the positive zero disapears .
If these tricks are not used, the OTA stage is non minimum phase because it has a positive zero and the amp with feedback is still non minimum phase because if I recall well a basic property of control theory: the zeros of the open loop transfer function are the zeros of the closed loop function.
Using Blacks formula: F(s)= G(s)/1+G(s) H(s) if G(s) = N(s)/D(s) then
F(s) = N(s)/(D(s) + N(s)H(s))
I am not posting to make a point or handwaving but to simply trying to understand what I found intriguing in Bob Cordell post about all pass.
All pass and minimum phase is semantic discussions but the existance of a positive zero and the parameters influencing it are relevant
Please remain educated in your reaction. I was asked to explain and I did it. I can be wrong and I will accept the demonstration but I accept a demonstration and not agressive assertions close to be insulting.