You'd better ask him yourself - that would be a long and boring story for 99.9% of the diyaudio audience. To add insult to injury, there are a few "gurus" around which would rather have a tooth pulled than admitting they were wrong about the magic properties of TMC so, in the interest of all peaceful minds around, there's no need to light this TMC fuse again.
Actually you don't even need to calculate anything: for the same unity gain frequency, the TPC values are exactly the same as those of the "TMC" network.
All you need do is calculate the TPC component values for you chosen unity gain frequency and simply connect the resitor to the output. Voila!
Guys, cut out the sniping. Now. Next one to stray off the technical path goes into read-only mode.
Hi David,
Here's my comment on his star-delta transform:
http://www.diyaudio.com/forums/soli...erview-negative-feedback-331.html#post2420886
Cheers,
E.
Thank you, in all the enormous pile of posts on the subject I had missed that one.
Your square wave results seem a bit odd.
If we assume the amp is not in slew rate limited overload then Fourier tells us that the square wave results should be a linear combination of the sine waves.
How can yours be so apparently different?
Best wishes
David
Sometimes I think its Mike’s mission in life to prove Edmond and me wrong in regard to TMC vs TPC.
There are (at least) three matters at issue in the TPC vs TMC argument, and each has its own set of answers and caveats.
1) Are TPC and TMC the same?
2) Are the optimum values of C1, C2 and R1 the same for the two techniques?
3) is one technique better than the other?
Those who answer yes to these questions, as I think Mike would have you do, are cheating themselves out of an alternative to TPC that may allow them to make better amplifiers.
First of all, let’s agree that an apple-apples comparison should be based on the same degree of feedback stability. This includes both the phase and gain margin of the main global feedback loop and that of the feedback path that encloses the output stage. It has been properly pointed out that this latter loop evaluation is quite important; if ignored, it could lead to one erroneously thinking that overly aggressive TMC compensation could yield a better amplifier. Both TPC and TMC implementations must provide adequate stability for the total loop gain enclosing the output stage.
Are TPC and TMC different? Of course they are. They are different in numerous ways. Bear in mind, this alone does not mean that one is better than the other. Perhaps the most obvious difference is the frequency shape of the global feedback loop gain. With TPC, the global loop gain is of obvious second order, with a pole-zero pair in the band below the unity gain frequency. For TPC, the global loop gain is a first-order function that rolls off at 6dB/octave, just like ordinary Miller compensation, as long as parasitics are ignored. To argue that TPC and TMC are the same, or that they always behave the same is plainly naïve.
Indeed Mike has himself asserted that they are different. In arguing for TPC superiority, Mike rightly pointed out that TPC encircles the input stage with its added loop gain, while TMC does not. Right here we have a clue that TPC and TMC are different, and that each may perform better in different situations. If one has a perfect output stage and a poor input stage, TMC lends little net advantage, and Mike is right in advocating TPC. If, on the other hand, one has a very linear input stage and a nonlinear output stage, there will be little benefit to TPC’s enclosure of the input stage (but do not conclude from this that TMC can never be better than TPC).
So they are indeed different. It follows that it is unlikely that they will always behave the same and/or that the optimum values of the compensation components will be the same.
Every stage in an amplifier contributes excess phase shift. Indeed, TPC’s enclosure of the input stage may actually be a disadvantage here. Consider, for example, the pole formed at the input of the input stage where the feedback network impedance meets the capacitance of the input stage. This is but one example.
The phase shift seen by the output stage in the total round-trip gain about it will be different between TPC and TMC, and this can make a difference in the optimum unity-gain frequency of the net amount of negative feedback as seen by the output stage. The TMC loop around the output stage is obviously tighter than the TPC loop around the output stage.
The distortion of most well-designed amplifiers is ultimately established by the output stage. That is the remaining big elephant in the room, at least in my experience. Although it would be an overly-broad statement to make, it is my experience for simulation and real amplifiers that optimized TMC is more adept at dealing with output-stage-dominated distortion than optimized TPC. This is not to say that TMC is universally better than TPC. YMMV, depending on the particulars of the amplifier. Similarly, those who insist on blindly using the same component values for both TPC and TMC are depriving themselves of optimizing one or the other. It would be silly to assume that the same component values are optimum for these two different compensation schemes.
More power to those who get better performance from TPC in their own amplifier designs.
Cheers,
Bob
. There are (at least) three matters at issue in the TPC vs TMC argument, and each has its own set of answers and caveats.
1) Are TPC and TMC the same?
2) Are the optimum values of C1, C2 and R1 the same for the two techniques?
3) is one technique better than the other?
Those who answer yes to these questions, as I think Mike would have you do, are cheating themselves out of an alternative to TPC that may allow them to make better amplifiers.
First of all, let’s agree that an apple-apples comparison should be based on the same degree of feedback stability. This includes both the phase and gain margin of the main global feedback loop and that of the feedback path that encloses the output stage. It has been properly pointed out that this latter loop evaluation is quite important; if ignored, it could lead to one erroneously thinking that overly aggressive TMC compensation could yield a better amplifier. Both TPC and TMC implementations must provide adequate stability for the total loop gain enclosing the output stage.
Are TPC and TMC different? Of course they are. They are different in numerous ways. Bear in mind, this alone does not mean that one is better than the other. Perhaps the most obvious difference is the frequency shape of the global feedback loop gain. With TPC, the global loop gain is of obvious second order, with a pole-zero pair in the band below the unity gain frequency. For TPC, the global loop gain is a first-order function that rolls off at 6dB/octave, just like ordinary Miller compensation, as long as parasitics are ignored. To argue that TPC and TMC are the same, or that they always behave the same is plainly naïve.
Indeed Mike has himself asserted that they are different. In arguing for TPC superiority, Mike rightly pointed out that TPC encircles the input stage with its added loop gain, while TMC does not. Right here we have a clue that TPC and TMC are different, and that each may perform better in different situations. If one has a perfect output stage and a poor input stage, TMC lends little net advantage, and Mike is right in advocating TPC. If, on the other hand, one has a very linear input stage and a nonlinear output stage, there will be little benefit to TPC’s enclosure of the input stage (but do not conclude from this that TMC can never be better than TPC).
So they are indeed different. It follows that it is unlikely that they will always behave the same and/or that the optimum values of the compensation components will be the same.
Every stage in an amplifier contributes excess phase shift. Indeed, TPC’s enclosure of the input stage may actually be a disadvantage here. Consider, for example, the pole formed at the input of the input stage where the feedback network impedance meets the capacitance of the input stage. This is but one example.
The phase shift seen by the output stage in the total round-trip gain about it will be different between TPC and TMC, and this can make a difference in the optimum unity-gain frequency of the net amount of negative feedback as seen by the output stage. The TMC loop around the output stage is obviously tighter than the TPC loop around the output stage.
The distortion of most well-designed amplifiers is ultimately established by the output stage. That is the remaining big elephant in the room, at least in my experience. Although it would be an overly-broad statement to make, it is my experience for simulation and real amplifiers that optimized TMC is more adept at dealing with output-stage-dominated distortion than optimized TPC. This is not to say that TMC is universally better than TPC. YMMV, depending on the particulars of the amplifier. Similarly, those who insist on blindly using the same component values for both TPC and TMC are depriving themselves of optimizing one or the other. It would be silly to assume that the same component values are optimum for these two different compensation schemes.
More power to those who get better performance from TPC in their own amplifier designs.
Cheers,
Bob
square figures
Odd?
Apart from square waves are odd anyhow (do you ever have seen a wave that is square? ), what's odd about my figures?
Cheers,
E.
Odd?
Apart from square waves are odd anyhow (do you ever have seen a wave that is square?
), what's odd about my figures?Cheers,
E.
Thanks to Bob for a nicely balanced summary.
It raises the question of the best metric to compare individually optimised amplifiers. If there is one loop then we can sensibly compare loop stability.
And if there are multiple loops in each amplifier?
Some minimum metric of any loop in the amp?
Phase by itself is not comprehensive.
Bode's work hints at some stability integral - what?
Or if no "ivory tower" satisfactory theory then perhaps some heuristics.
I have a few ideas here, in Jan Didden's next opus if lucky, but would appreciate more input from anyone with actual content.
Best wishes
David.
It raises the question of the best metric to compare individually optimised amplifiers. If there is one loop then we can sensibly compare loop stability.
And if there are multiple loops in each amplifier?
Some minimum metric of any loop in the amp?
Phase by itself is not comprehensive.
Bode's work hints at some stability integral - what?
Or if no "ivory tower" satisfactory theory then perhaps some heuristics.
I have a few ideas here, in Jan Didden's next opus if lucky, but would appreciate more input from anyone with actual content.
Best wishes
David.
Last edited:
For your "typical" sine wave @ 20kHz the TPC IPS output is about 1/20 of the TMC IPS output.
Yet for the square wave the TPC and TMC IPS currents are about the same.
In a linear system, Fourier says the square wave result should be the linear sum of the sine components.
If the sine components of the TPC are 20 times less then does that not seem odd?
Similarly for the VAS output currents, there seem to be anomalies
Best wishes
David.
I was often surprised to read in this forum so many wrong technical assertions and analyses, even, sometimes under well known signature. This lead often to boring flame wars between members, like the one under about TPC vs TMC.
Once again Bob Cordell gives an example of smart, easy to understand, balanced and complete analyze of a (not so dramatic ;-) question.
It is difficult, indeed to decide in absolute the best way to stabilize a closed loop amplifier. As far i'm concerned, looking at square waves behaviors both with high and little signals and the overall frequency curve help to figure-out.
Whatever the solution i decide, i try to set a flat bandwidth, with no peak at HF, by the compensation network, then filter the input signal by a low pass filter in, order to remove any overshoot on HF square waves. This prevent from TIM as well.
While, in practical situations, i tend to prefer TMC, for the reasons bob explained ( the output stage is the often the more critical) in an ideal amplifier, where each stage is set faster than the previous one, TPC is supposed to give a little better results ?
At the end, ears are the judges, and, if no audible differences can be obviously noticed between the two compensation schemes, who care ?
Once again Bob Cordell gives an example of smart, easy to understand, balanced and complete analyze of a (not so dramatic ;-) question.
It is difficult, indeed to decide in absolute the best way to stabilize a closed loop amplifier. As far i'm concerned, looking at square waves behaviors both with high and little signals and the overall frequency curve help to figure-out.
Whatever the solution i decide, i try to set a flat bandwidth, with no peak at HF, by the compensation network, then filter the input signal by a low pass filter in, order to remove any overshoot on HF square waves. This prevent from TIM as well.
While, in practical situations, i tend to prefer TMC, for the reasons bob explained ( the output stage is the often the more critical) in an ideal amplifier, where each stage is set faster than the previous one, TPC is supposed to give a little better results ?
At the end, ears are the judges, and, if no audible differences can be obviously noticed between the two compensation schemes, who care ?
Last edited:
anomalies
Hi David,
It isn't that odd as one might think at first glance. The load (and current) at the output of the IPS rises with 6dB/oct in case of TMC. With TPS however, the the load rises with 12dB/oct.
As for the VAS output currents, I don't see anomalies.
Cheers,
E.
Hi David,
It isn't that odd as one might think at first glance. The load (and current) at the output of the IPS rises with 6dB/oct in case of TMC. With TPS however, the the load rises with 12dB/oct.
As for the VAS output currents, I don't see anomalies.
Cheers,
E.
Yes. I considered that. But it is still odd to compare the considerable difference between TPC and TMC at audio frequencies and the minimal square wave difference.
Eventually both approximate to CMC with one pole, at sufficient frequency.
The TPC and TMC square wave currents are about equal.
So the square wave result is almost entirely composed of harmonics at frequencies where TPC and TMC are practically identical? Is this a useful comparison?
At 20Khz the TPC current is 26dB lower. So if separation closes at 6dB an octave they will be similar above 400 kHz. The square wave result is dominated by harmonics above 400 kHz? Is that realistic?
Does this not still seem odd to you?
Best wishes
David
Obviously the answers to the questions above are three counts of "No".
It looks like a duck, but it doesn't quack... The aspect of the global loop gain phase for TMC has little to nothing to do with the global stability margin of the amplifier. Comparing the TPC second order global loop gain phase with the TMC global loop gain phase, for stability purposes, is not apples to apples. This was extensively discussed and I though you already agreed on that.
Regarding question 3) above, I do appreciate you changing your hard stance regarding the "superiority of TMC", as stated here and here.
I pontificate slightly on the subject at http://www.diyaudio.com/forums/anal...screte-opamp-open-design-209.html#post3243177
I'm assuming the design is competent in that device variations (with reasonably good layout, decoupling, earthing bla bla) will not degrade stability. Then the biggest 'stability' variation not under the control of the designer is the load. An amplifier should be UNCONDITIONALLY STABLE with ALL loads.
There are many Golden Pinnae amplifiers that exhibit small bursts of oscillation on parts of a sine wave cycle with real speaker loads ... often dependent on the thermal and signal history too. No wonder they sound different and are critical to match with speakers.
Check with ALL capacitive loads between 1n & 100n with and without a parallel resistor. Also a real speaker at various frequencies and power levels. A large guitar speaker is good for this.
Do this in SPICE but don't neglect doing this in real life too. (You'll see anomalies between LTspice AC analysis and Transient analysis too. I sorta prefer the transient plots for stability but I'm a SPICE newbie.)
If the amplifier is good under all these tests, it will be good under the theoretical metrics too. The converse isn't always true.
Also check stability and sensible behaviour under overload with various loads. http://www.diyaudio.com/forums/anal...screte-opamp-open-design-231.html#post3266142 The usual metrics won't tell you anything about this.
If it isn't obvious, I think the 'usual metrics' are OK if your load is fixed. All the methods, being argued over can be stabilized, either with maths, Cherry's matrix arithmetic or even finger in the air. When you allow ALL possible 'real life' loads and nasty things that users do like clipping your amp, things become much more complicated.
Last edited:
Apparently, a simulation reveals 'oddities' which are easily overlooked by humans.
That is where simulators are for.
Not sure I understand your point.
I think that the 20 KHz simulation result is relevant, it shows that TPC requires less IPS output current in the audio spectrum. This makes sense and provides one data point for the size of the reduction.
We know TPC and TMC are similar at sufficient frequencies.
The square wave simulation seems only to tell us that most of the simulated harmonic content is above that frequency. I don't see what it contributes.
Best wishes
David
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.