A while back I looked at Monticelli outputs to improve the efficiency of an audio power amp.
It's a common emitter circuit so it doesn't waste all the Vbe drops of the usual emitter followers.
This complicates the bias so the result is a little more complex.
The extra complexity is a reasonable trade off in low power op-amps and it's widely used there.
Probably not worth the pain in a conventional audio amp, where the few volts saved are a small fraction of the ~50 to ~70 volt rails.
Instead I looked to improve the efficiency with outputs that are floated by a switch mode power supply.
A kind of switch mode Class H, and it simulates very nicely.
But now the outputs only see ~5 to ~10 V, just like a modern, low rail V op amp.
So we are back in Monticelli territory.
If you can save 3 Vbe on each rail then they can drop from say, 7 V to 5 V, plus and minus.
The conventional circuit is ~40% worse power consumption, at the same bias.
And maybe the quiescent bias can be reduced too, for further improvement.
Which is the question for this post.
Anyone understand what limits the quiescent bias of a Monticelli output?
For the conventional emitter follower we have the Oliver criterion, 26 mV across the emitter resistors (more or less).
But for Monticelli there's no emitter resistors (more power saved!), and it's not clear to me how it's optimised.
Anyone have any ideas so I know what to try in Spice?
I have included the patent sketches to start the discussion, pretty obvious really
Bonus points for the sharp eyed who spot an anomaly.
David
It's a common emitter circuit so it doesn't waste all the Vbe drops of the usual emitter followers.
This complicates the bias so the result is a little more complex.
The extra complexity is a reasonable trade off in low power op-amps and it's widely used there.
Probably not worth the pain in a conventional audio amp, where the few volts saved are a small fraction of the ~50 to ~70 volt rails.
Instead I looked to improve the efficiency with outputs that are floated by a switch mode power supply.
A kind of switch mode Class H, and it simulates very nicely.
But now the outputs only see ~5 to ~10 V, just like a modern, low rail V op amp.
So we are back in Monticelli territory.
If you can save 3 Vbe on each rail then they can drop from say, 7 V to 5 V, plus and minus.
The conventional circuit is ~40% worse power consumption, at the same bias.
And maybe the quiescent bias can be reduced too, for further improvement.
Which is the question for this post.
Anyone understand what limits the quiescent bias of a Monticelli output?
For the conventional emitter follower we have the Oliver criterion, 26 mV across the emitter resistors (more or less).
But for Monticelli there's no emitter resistors (more power saved!), and it's not clear to me how it's optimised.
Anyone have any ideas so I know what to try in Spice?
I have included the patent sketches to start the discussion, pretty obvious really
Bonus points for the sharp eyed who spot an anomaly.
David
you just throw up a convoluted circuit with no explanation, what do you expect?
Hi Jan
This is not some obscure circuit, known only to me, that I refuse to explain.
It's nearly 40 years old, the patent is readily available (I even provided the number), discussed in the "Art of Electronics" and in more detail in the "X chapters", used in a lot of operational amplifiers, and universities teach it to Elect.Es as a standard circuit block.
I expected that there would be some discussion because there are multiple explanations already on-line, both the patent and the AoE X chapter.
I hoped that someone would understand it better than I do, and explain the subtleties to me.
But it's not that convoluted, the first patent picture shows it's basically just 2 output transistors and a level shift system of two more transistors, with bias sources.
Do you want me to try to explain it?
If anyone is actually interested I could do that, it would probably help me to clarify my own ideas, There's certainly aspects the patent skips over, and AoE too.
Best wishes
David
Hi Jan
This is not some obscure circuit, known only to me, that I refuse to explain.
It's nearly 40 years old, the patent is readily available (I even provided the number), discussed in the "Art of Electronics" and in more detail in the "X chapters", used in a lot of operational amplifiers, and universities teach it to Elect.Es as a standard circuit block.
I expected that there would be some discussion because there are multiple explanations already on-line, both the patent and the AoE X chapter.
I hoped that someone would understand it better than I do, and explain the subtleties to me.
But it's not that convoluted, the first patent picture shows it's basically just 2 output transistors and a level shift system of two more transistors, with bias sources.
Do you want me to try to explain it?
If anyone is actually interested I could do that, it would probably help me to clarify my own ideas, There's certainly aspects the patent skips over, and AoE too.
Best wishes
David
I don't think there is a distortion optimum, you just get less distortion with more quiescent current. The same holds for a complimentary emitter follower without emitter resistors.
For a discrete implementation, you will have to add emitter resistors for thermal stability. Maybe you get a distortion compensation effect then?
For a discrete implementation, you will have to add emitter resistors for thermal stability. Maybe you get a distortion compensation effect then?
Hi Dave,Hi Jan, This is not some obscure circuit, known only to me, that I refuse to explain.
...
I hoped that someone would understand it better than I do, and explain the subtleties to me.
...
Best wishes, David
Attached is a Thesis with analysis of the Monticelli OPS by LAKSHMINARASIMHAN KRISHNAN p46-51.
Cheers, IanH
Hi Marcel
The thermal non uniformity in a discrete implementation is a concern, mentioned in the Art of Electronics, so thank you for the excellent observation.
My initial impression is that thermal stability should be much less of a problem than a typical amp because thermal instability is proportional to rail potential.
The amplifer is operated at rails just a tenth of a typical power amp (~ 5 to 6 V versus ~70 V for powerful woofer amps)
Or I could use emitter resistors a tenth of the usual value, only 0.015 ohms or thereabouts.
Maybe a thermally coupled bias transistor would be adequate to ensure thermal stability with only trace resistance?
If small resistors are used then my first idea is that it would similar to an EF. Transconductance would transition smoothly when the emitter resistor has quiescent bias of ~26 mV.
That implies amps of bias, so maybe in practice an optimum is not practical, just less distortion for more current, as you wrote.
Best wishes
David
Last edited:
Hi Dave,
Attached is a Thesis with analysis of the Monticelli OPS by LAKSHMINARASIMHAN KRISHNAN p46-51.
Hi Ian
I expected this would interest you, I remembered your interest in push pull outputs that keep both sides active, and the Monticelli does that trick.
Thanks for the attachment, just what I wanted.
It's for FET outputs and that has made me think more about the implications of FET versus BJT.
AoE doesn't discuss the difference, except to observe that Monticelli invented it for "MOSFETs, but it works especially well with BJTs".
Apparently because with BJTs it has a quite linear transfer function for current in/out.
But for audio amps we want volts out, maybe the quadratic transfer functions of FETs would work better here?
(The sum of squares linearization trick, also an interest of yours IIRC!)
Not sure how well power FETs would work at such low Vds. lot of capacitance modulation, any comments?
Best wishes
David
Attached is a Thesis with analysis of the Monticelli OPS by LAKSHMINARASIMHAN KRISHNAN p46-51.
Hi Ian
I expected this would interest you, I remembered your interest in push pull outputs that keep both sides active, and the Monticelli does that trick.
Thanks for the attachment, just what I wanted.
It's for FET outputs and that has made me think more about the implications of FET versus BJT.
AoE doesn't discuss the difference, except to observe that Monticelli invented it for "MOSFETs, but it works especially well with BJTs".
Apparently because with BJTs it has a quite linear transfer function for current in/out.
But for audio amps we want volts out, maybe the quadratic transfer functions of FETs would work better here?
(The sum of squares linearization trick, also an interest of yours IIRC!)
Not sure how well power FETs would work at such low Vds. lot of capacitance modulation, any comments?
Best wishes
David
Hi Dave,
What I have found to work better is cubelaws in the class-A region (and even beyond) because you get a smooth transition when leaving the Class-A region and greatly reduced high order harmonics at the higher power levels. I have more recently found that autobias-nonswitching circuits generate this sort of cubelaws naturally -- so they seem to me to be the most logical choice for a good sounding power amp with high efficiency and stable thermally and minimal components.
Re: MOSFET capacitance variations in LV amps -- I think that can be managed without serious distortion or stability problems for audio amps. The key is enough current to drive the gates at the desired slew rates; with MOSFETs it's not a huge amount of current.
All the best with your endeavour.
Cheers, IanH
Yes, I have used the sum of squarelaws in Class-A to get linearity with good efficiency in Class-A. But it does not work well for Class-AB, which I assume you are considering. With squarelaws, leaving the region where both conduct causes nonlinearity to appear - similar to gm-doubling in conventional Class-AB.
What I have found to work better is cubelaws in the class-A region (and even beyond) because you get a smooth transition when leaving the Class-A region and greatly reduced high order harmonics at the higher power levels. I have more recently found that autobias-nonswitching circuits generate this sort of cubelaws naturally -- so they seem to me to be the most logical choice for a good sounding power amp with high efficiency and stable thermally and minimal components.
Re: MOSFET capacitance variations in LV amps -- I think that can be managed without serious distortion or stability problems for audio amps. The key is enough current to drive the gates at the desired slew rates; with MOSFETs it's not a huge amount of current.
All the best with your endeavour.
Cheers, IanH
3 shows current mirrors. The currents are set by I9 and the ratios of device sizes (X, 2X, 4X, 20X for PNPs and Y, 2Y, 8Y, and 20Y for NPNs). That won't work with discrete transistors.
Hi Ed
The first picture shows the basic concept without any current mirrors, just current sources and bias potentials.
Current mirrors are a convenient implementation, especially in an IC but not essential, I think.
In any case, I don't think a currrent mirror implementation necessarily won't work with discrete transistors, there are plenty of audio amps with discrete current mirrors.
Not as well matched as an IC but perfectly usable.
Or it could be done with monolithic dual and quad transistors.
The outputs can't be done this way in a practical implementation, but it's similar to the bias spreader in a conventional amp, dissimilar transistors can be thermally coupled and work pretty well.
Do you see any specific problems?
Best wishes
David
Hi Ed
The first picture shows the basic concept without any current mirrors, just current sources and bias potentials.
Current mirrors are a convenient implementation, especially in an IC but not essential, I think.
In any case, I don't think a currrent mirror implementation necessarily won't work with discrete transistors, there are plenty of audio amps with discrete current mirrors.
Not as well matched as an IC but perfectly usable.
Or it could be done with monolithic dual and quad transistors.
The outputs can't be done this way in a practical implementation, but it's similar to the bias spreader in a conventional amp, dissimilar transistors can be thermally coupled and work pretty well.
Do you see any specific problems?
Best wishes
David
Discrete transistor current mirrors are stabilized with emitter degeneration resistors and are often inside the global feedback loop. The circuit has multiple transistors that need to track. You can try to make it work, but I think that other approaches are more suitable for discrete transistors.
Ed
Ed
What about this approach concerning distortion remove in Class AB output buffers of power amplifiers with only 20-50mA idle resp. quiescent current ?
http://www.renardson-audio.com/B-amp.html
https://www.renardson-audio.com/mjr9.html
found under
https://www.renardson-audio.com
An other approach was realized by Anders Thule (Thule Audio, virtual Class-A) - go to post #19+78 under
https://www.diyaudio.com/community/threads/what-happen-to-thule-audio.126352/page-4
and post #5 under
https://www.diyaudio.com/community/threads/thule-ia100-dc-on-output-when-hot.363092/
There is in use a controlled Vbe multiplier by an additional current source
Last edited:
It's a nuisance that the quiescent current depends on more than one mirror ratio. You could make some ratios adjustable, as is done in conventional complementary emitter follower circuits, but then you need to adjust several trimming potmeters to adjust the quiescent current.
Maybe something for transistor arrays and ThermalTrak output transistors?
I don't think this will be a problem, the current mirrors are not an essential part of the circuit, the first patent pic. shows the basic idea and there are no current mirrors.
The apparent current mirrors are just a convient way to implement current sources to set up the quiescent.
This realization came to me after I looked closely at the AoE X Chapters, the authors themselves have published it on-line so here's a copy to refer to as I try to explain.
Notice that the complementary transistors are scaled to match, so the NPN and PNP outputs (Q14 & Q13) are X20, the level shifters (Q3 & Q17) are X2, and so on.
EXCEPT the current sources Q15 & Q18...!
What's that about?
If you look at the patent (3rd circuit in my first post) it shows a similar mismatch and AoE have faithfully copied it.
But the reason it exists is to adjust for the emitter current of the driver transistor shown in the patent.
And the AoE circuit doesn't model that (just has a current source), so it's mismatched by a factor of 2 (X4 versus X8).
But it still works fine! A mismatch is detectable on plot 4x.94 below but the output 4x.93 looks excellent.
Best wishes
David
Last edited:
It is notable the way the 3rd patent circuit is done, with the bias chains in the top and bottom halves both set from the same source, with a real current mirror (25').
Whereas in the 1st, basic concept explanation, there are two independent sources.
Probably just an implementation example, for the reasons in the above post I don't think this is a problem, the ratios don't have to track perfectly.
I do have a stash of ThermalTrak output transistors if I need them!
Best wishes
David
Last edited:
At low bias the square law behaviour of MOSFETs breaks down and I hoped to exploit this.
I don't yet fully understand this area, and it's often not well modelled in SPICE.
Have to do a bit more research.
Best wishes
David
At very low currents, MOSFETs behave exponentially (weak inversion or subthreshold region). There is a wide range between weak and strong inversion called moderate inversion where they are neither exponential nor sort of quadratic. Some old Spice models only model the strong inversion region.
Last edited:
I phrased post #15 quite inaccurately. What I meant is that it is inconvenient that the circuit depends on two translinear loops, the two loops consisting of the output transistors, common base transistors and bias spreaders, and that they influence each other. You probably have to come up with some scheme where you can adjust them independently.
Maybe something like this? Disconnect the circuit that drives the Monticelli stage and the NPN part, connect the collector of the PNP common base transistor to the negative rail, short the output and adjust the PNP side, then do the opposite to adjust the NPN side? You have to keep in mind that you actually adjust the minimum rather than the quiescent current then, as with the NPNs disconnected, the PNP common base transistor runs at as much current as it normally does when the NPN output transistor is turned on hard.
Maybe something like this? Disconnect the circuit that drives the Monticelli stage and the NPN part, connect the collector of the PNP common base transistor to the negative rail, short the output and adjust the PNP side, then do the opposite to adjust the NPN side? You have to keep in mind that you actually adjust the minimum rather than the quiescent current then, as with the NPNs disconnected, the PNP common base transistor runs at as much current as it normally does when the NPN output transistor is turned on hard.