Transitional Miller Compensation (TMC) vs. Two Pole Compensation (TPC)

Jump to Latest
#25
WOW! TPC is great. All these years I'd overlooked it; that was a mistake.

I took a stereo TMC amp, and swapped one channel to TPC. The compensation networks were set up to produce nearly identical loopgain bode plots on the sim.

TPC sounds clearer, less congested, with better separation between instruments. Complex sounds are startlingly accurate.

What could explain this? It's not likely to be the amount of feedback around the OPS which is identical (and ample!) in each case.

My guess is that TPC and TMC load the input stage differently. To the IPS, at audio frequencies, a TMC network looks exactly like a plain old Miller cap: it must be charged and discharged as the output slews, demanding currents up to 10s of microamps.

TPC draws orders of magnitude less current from the IPS.

Given all of Self's work showing the importance of precise current balance in the input stage pair to reduce distortion, it's intuitive that we should load the IPS with high impedance to draw as little current as we can. Miller compensation and TMC fail equally at this.

If that's the problem, why does the simulator report ultralow distortion for TMC amps? Is the simulator blind to the distortion effects of an unbalanced IPS? More research is needed...

It's possible that the TMC version of this amp was just marginal in some way, and its TPC incarnation happened to be healthier. Maybe a healthy TMC amp would sound identical to its TPC doppelganger.
 
#26
Another benefit of TPC: it can be used in a preamp gain stage with only a VAS and no separate OPS, where TMC is impractical. Again, the value is in presenting an easier load to the IPS than miller compensation.



Another benefit of TPC: if the compensation network capacitors are nonlinear, TMC makes the IPS drive that nonlinear capacitance and TPC doesn't. I modeled a nonlinear cap in spice, used it in the compensation networks. It severely impacts the distortion performance of a TMC amp and doesn't touch the performance of a TPC amp at all.



Maybe that's academic; C0G/NP0 caps are supposed to be very linear over a wide voltage range. But if you'd rather not rely on that, or if you only have crappy X7R caps available, go for TPC.


For that matter, ceramics are microphonic aren't they? It might be a good idea to spec a film cap facing the IPS for either TMC or TPC. Tiny error currents at the IPS output can corrupt the whole amp performance.
 
#27
The benefit of TPC for the input stage (which is not present in TMC), is due to TPC increasing the overall open loop gain of the amplifier (A_OL and not LG is what matters here) in the frequency region before the zero (i.e. in the flat gain and two pole region). This increased gain reduces the differencial signal across the input stage VD=VOUT/A_OL which keeps it in the linear range of the input stage for a larger frequency range.

Does the effect equate to lower loading of the IPS? It does, but that is a consequence rather than the main intent of TPC. So I would not think about TPC this way.

Regarding the benefit of TPC in the pre-amp you are correct, with no OPS you cannot do TMC.

Finally, if you use crappy capacitors for compensation you will exhbit distortion since the voltage signal is essentially changing the loop gain of the amplifier over frequency due to the voltage coefficient of the capacitor. Is TPC more forgiving? yes. Is TPC not affected at all? No. I think in your case, something else now dominates the distortion.

Distortion is an interesting since it sums up in an RSS fassion (SQRT(x1^2 + X2^2+...)). So it is a winner takes all type deal, what you see is the largest one, so if the capacitor distortion fell below the distortion of some other artifact, you won't see it.

"Tiny error currents at the IPS output can corrupt the whole amp performance". Every distortion artifact that is inherent to the amplifier design can be equated to a load current from the input stage, even cross-over distortion believe it or not.

BTW, "TPC sounds clearer, less congested, with better separation between instruments. Complex sounds are startlingly accurate." This is a subjectivist comment.
The engineering comment is: "TPC reduces distortion artifacts at both the input and output stages..." to which we can add "while only TMC only focuses on the OPS distortion artifacts".

Regards, Sandro
 
Last edited:
#28
Sandro, you're right. I've also been thinking about this terms of loop gain at the IPS. That's one of the starkest ways to describe it: would you rather have 30db or 60db of loopgain around the IPS at 20kHz? That's the difference between TMC and TPC.


TPC has roughly the same loop-gain plot regardless of where you probe. It's one loop. TMC has separate local and global loops, and you see different loopgain plots depending on where you probe. I used to think this complexity and modality was "the cost of doing business" if you wanted high feedback around the OPS but it's not.



The simulator says TMC can perform just as well. But the simulator thinks that the LTP transistors are identical; it thinks the degeneration resistors are perfectly matched. It thinks the current mirror is equally perfect. In a real circuit with "lumps and bumps" we can be happy to have the extra 30db of loopgain. The more we can reduce the signal excursions at IPS, the more linear it will be within its narrowed range of travel, probably.
 
#29
PB2 said:
The authors are quite frank about the issues with an average home builder building the PGP, yes it is overly complex but it is just an example of what can be done. I believe that they learned a lot and have come a long way so that a simpler topology might come close enough to the PGP to make the complexity unnecessary. In fact they call their older designs outdated, still we can learn from their work.

Bob Cordell simulated a slightly modified "Blameless" design with TMC and obtained excellent, good enough IMO, results. This is what is of more interest to me. Bob's old MOSFET design with a bipolar output stage triple and TMC is also of interest to me - not too complicated.

Should have included a link to Bob's Blameless simulation comparing TMC and TPC: https://www.diyaudio.com/forums/sol...lls-power-amplifier-book-119.html#post2413197

Link to Bob's revision of his old MOSFET amp to employ TMC rather than error correction in the output stage and obtain similar results with less complexity: https://www.diyaudio.com/forums/sol...ells-power-amplifier-book-86.html#post2391212

I thought that I included these links in the first post of the thread.
Click to expand...
Does anyone know why the links in the above post no longer work, was there an "upgrade" to the forum?
They don't work for me at least, anyone know how to correct those links?
I tried searching for the first one but it is not easy to find.
 
#32
PB2 said:
Does anyone know why the links in the above post no longer work, was there an "upgrade" to the forum?
They don't work for me at least, anyone know how to correct those links?
I tried searching for the first one but it is not easy to find.
Click to expand...

I don't know if it is the same issue, but when you copy the address from the browser address bar rather than from the number at the top right of the post, where you end up when you click the link depends on the number of messages per page. See https://www.diyaudio.com/community/...or-links-report-them-here.380497/post-6878046
 
#33
#34
Compensation scheme depends on topology. Regulating the amount of high frequency feedback as used in TPC and TMC needs careful evaluation, cherry is most suitable between ltp and vas but even there careful evaluation is needed, yes the lm3886 uses cherry but those devices may have closely matched ft, again it all depends
 
#35
TPC+Cherry also named OITPC by Dadod displays potential for oscillation. Unless all parameters are fed into a simulator it won't catch some oscillations. Some of this instability is waiting for a cymbal strike or a metal track, when cymbals don't sound like cymbals or metal sound as metal. When a class AB amplifier sounds like a class D amplifier, then its likely its oscillating, the difference being that class D is controlled oscillation
 
#36
I fixed the links in the first post and noticed that the page numbers were half of the previous ones. It seems that the
default posts per page was doubled. After fixing all the links and providing the post number so that if it changes in
the future one can find the posts, I thought to look at preferences for the site and found posts per page as another
way to fix it. In fact, I believe that all old posts with links will no longer work, so that would be a handy way to make
them work again.
 
#37
Bonsai said:
Thanks for the analysis.

I settled on TPC on all my amps and get great distortion performance for the cost of 1 extra resistor and cap ( same of course for TMC).
Click to expand...
I regret to say, my opinion on this has changed. I settled on TMC not long after posting the above. The HF peaking you get with TPC requires heavier front end filtering if you want a perfect square wave response. Quite how Syn08 [RIP] or anyone else can claim they are equivalent is beyond me and especially given the comments in post #26 above. They both extend the loop gain bandwidth, but in practice, their behaviour is less than equivalent given TPC’s peaking problem.

Re post #26 above, what if the max current available to charge the VAS comp cap is not limited by the tail current as it is in a VFA?
 
Last edited:
  • Like
Reactions: PB2
#38
Thanks for the update.
I view this as look we can "play" with the math and make these look the same, but implementation matters AND
if I'm reading you correctly the forward path transfer function is different (less bandwidth) once you add a filter
to eliminate overshoot. I thought a lead cap in the global feedback was another way to solve this problem or
does that also result in the same forward bandwidth reduction?
 
  • Like
Reactions: Bonsai
#39
I’ve seen a few tricks to resolve the peaking issue - Bob discusses bridging the ‘T’ with a small cap in his book as an alternative to a front end BW limiting filter. However, with this approach, you end up losing 4-6 dB loop gain at HF.
It seems from Sandro’s analysis that at the frequencies of concern (assume here he means 20-20 kHz in-band signals) there is little to choose between the two techniques, but clearly at HF (so well above the audio band) they are not behaving the the same due to the peaking. I do think any analysis must consider how the comp network (TMC or TPC) is behaving wrt to main feedback path. I can’t comment on the lead comp approach since I have only designed CFA’s the last 10 yrs and lead comp caps across the feedback resistor is a big no-no and guaranteed to cause oscillation.
 
  • Like
Reactions: PB2
You must log in or register to reply here.
Top Bottom