I was leafing through Self's book and happened to come across the chapter about output inclusive compensation. It intrigued me, so I tried to try it on the amplifier I happened to have open in SPICE and it improved both phase margin and THD. The THD went from 0.00077% to 0.00035% (no, it's probably not an audible difference, but improving your numbers is fun!). More notably the decrease in THD was most pronounce on the third harmonic. This seems like evidence of Self's assertion than this semi-local feedback reduces cross-over distortion. All good stuff!
But what are the pitfalls? I can't remember seeing this technique being used that often. There must be a reason for that.
The compensation in question is made up of C6, C24 and R52. The amplifier is work in progress, so be nice. 🙂
The phase margin looks pretty nice!
Distortion without output inclusion. Notice the 3rd harmonic!
But what are the pitfalls? I can't remember seeing this technique being used that often. There must be a reason for that.
The compensation in question is made up of C6, C24 and R52. The amplifier is work in progress, so be nice. 🙂
The phase margin looks pretty nice!
Distortion without output inclusion. Notice the 3rd harmonic!
Last edited:
Check out similar compensation in this amp:
From my experience, 2-pole compensation will give lower Thd, and better stability (E.g. phase margin), but it's more difficult to come up
with correct values for C and R.
One variation of 2-pole compensation have R going to the ground instead to the output, but using output gives better results.
I also remember somewhere on this forum an amp with 3-pole compensation..
And here is a random image that I saved on my PC long time ago, apparently I thought it was important 🙂
Amp from the post https://www.diyaudio.com/community/threads/unusual-amp-from-1987.357369/post-7068145
has been built (1 channel so far). Works like a champ. the only correction was to change C19 from 5pF to 9pF.
One channel built and tested so far. Will build 2nd channel, and then test with real music.
Idle current 40-50mA per output fet, output DC voltage (with C20 installed): 0.4 mV. Without out - 160mV.
has been built (1 channel so far). Works like a champ. the only correction was to change C19 from 5pF to 9pF.
One channel built and tested so far. Will build 2nd channel, and then test with real music.
Idle current 40-50mA per output fet, output DC voltage (with C20 installed): 0.4 mV. Without out - 160mV.
From my experience, 2-pole compensation will give lower Thd, and better stability (E.g. phase margin), but it's more difficult to come up
with correct values for C and R.
One variation of 2-pole compensation have R going to the ground instead to the output, but using output gives better results.
I also remember somewhere on this forum an amp with 3-pole compensation..
And here is a random image that I saved on my PC long time ago, apparently I thought it was important 🙂
Might become one of mine too. The THD of this Blameless-like baby is starting to get in the “unnecessarily low” range. I love it! 😀
Right, it is TMC.
It could be more than 2 poles. Like this: https://www.diyaudio.com/community/threads/super-tmc.234846/post-3468584
In your schematic, C24 should be >> C6 to prevent unwanted frequence peaking, so that the equivalent Cdom is about the same before and after the "Transition".
It could be more than 2 poles. Like this: https://www.diyaudio.com/community/threads/super-tmc.234846/post-3468584
In your schematic, C24 should be >> C6 to prevent unwanted frequence peaking, so that the equivalent Cdom is about the same before and after the "Transition".
This is 3-pole comp that was actually built:
For me this rule did not always work (see schematic at post #2). With C24 bigger than C6 Phase Margin was not acceptable...
I see i had a similar solution in 1976 as in post 2. It was a resistor from output in series with a capacitor to the base of the VAS transistor and a usual miller cab C - B.
In october 1984 i used exactly the described compensation and many times later. I have never thought of it as something special. Just a thing to try when the miller compensation affects the performance.
I just wonder if i had the idea or if it was in a magazine. In that case probably Wireless World.
Both amplifiers had ordinary Blameless topology.
In october 1984 i used exactly the described compensation and many times later. I have never thought of it as something special. Just a thing to try when the miller compensation affects the performance.
I just wonder if i had the idea or if it was in a magazine. In that case probably Wireless World.
Both amplifiers had ordinary Blameless topology.
Self has a long discussion in his book about who came up with it. I’ll summarize it for you: It’s complicated. 😀Maybe it was you!
I thought that consensus was the other way round, i.e. C6 > C24. C24 is the dominant pole. C6 can be in the range of five to ten times C24 then. The additional resistor R52 can be connected either to ground, the supply rail or the output. Which connection makes most sense depends on the amplifier topology. Some connections are bad for PSRR. If an amplifier has good PSRR with the resistor to ground, it can be connected to the output as well.
In the circuit i described you will get much better distortion reduction with the resistor to the output.
With the resistor to ground or negative rail you can get a nasty phase shift so it is much more difficult to try out.
With the resistor to ground or negative rail you can get a nasty phase shift so it is much more difficult to try out.
Let me propose a better way.
C6 = C24.
There should be 2 resistors, 2K from the output, 2K from the ground.
PS: The single resistor version is just a simplified version. In this case of connecting from the output, the other resistor to the ground is infinity. Thus, C24 should be >> C6 to keep the voltage ratio about the same to avoid peaking at the frequency response. TMC should feedback about the same voltage amplitude before and after the "Transition".
Last edited:
I dont remember that i ever used so big miller capacitors. In a case like this i would begin trying with C6 = 10pF, C24= 100pF and R 10kohm.
A MOS input with much higher current usually makes the capacitors more equal to me.
If you replace T8 with a 2N7002 in cascode with Q12 you will get higher open loop gain. D3 stays of course. A resistor from D to E can function as a current limiter.
A MOS input with much higher current usually makes the capacitors more equal to me.
If you replace T8 with a 2N7002 in cascode with Q12 you will get higher open loop gain. D3 stays of course. A resistor from D to E can function as a current limiter.
OP, your OLG graph in post 1 is showing typical miller compensation, did you post wrong???
and here is a thread of interest https://www.diyaudio.com/community/...tion-tpc-for-a-power-amplifier-method.406145/
and here is a thread of interest https://www.diyaudio.com/community/...tion-tpc-for-a-power-amplifier-method.406145/
Equal capacitors seem to work best. Close loop gain plots.
C6=C24=100p, R52=1k
C6=22p, C24=100p, R52=10k
C6=C24=100p, R52=1k
C6=22p, C24=100p, R52=10k
TPC has inherent peaking at some frequency in the closed loop response.
Setting both capacitors equal value reduces that peak indeed, but makes the dual pole concept very ineffective.
In order to gain appreciable advantage from TPC, the ratio needs to be at least five.
If you mind the unavoidable closed loop gain peak, better abandon TPC altogether.
Meanwhile I had a look at one of my books. In Douglas Self's book Audio Power Amplifier Design (6th edition) figure 13.16, you can see a correct example application with the correct capacitor ratio.
In Self's example, Cdom=120pF, the TPC capacitor has 1nF and the resistor 2k2.
A second resistor across the dominant pole capacitor can be used to get rid of the open loop gain peak.
While the open loop gain peak is considered inconsequential, it is probably better to remove it nonetheless.
Self chose 1.4Meg for the resistor in the example.
I think that’s a different topology from what I’m attempting. Mine is a “output inclusive” or “transitional Miller Compnensation”, whereas the one you’re referring to is a two-pole compensation. I think Self sees them as two separate techniques. The latter doesn’t have any semi-local feedback/output inclusion.