David - The current mirrors are necessary in practice to generate voltage references that track Vbe.
This is an interesting circuit, and one that I was not aware of (it post-dates the first edition of Art of Electronics). Thanks for starting the thread!
The intended use is monolithic op-amps. The problem being addressed (output voltage swing) can also be solved by adding supply rails (my discrete transistor amplifier does that).
The crossover region may look better, but the open-loop distortion and output impedance are both high due to the common emitter. In contrast, a class AB emitter follower has 0.1-0.3% THD and a few tenths of an ohm output impedance in open-loop.
The Monticelli output stage has to be used with high negative feedback. That too is a downside for a discrete transistor amplifier.
Ed
This is an interesting circuit, and one that I was not aware of (it post-dates the first edition of Art of Electronics). Thanks for starting the thread!
The intended use is monolithic op-amps. The problem being addressed (output voltage swing) can also be solved by adding supply rails (my discrete transistor amplifier does that).
The crossover region may look better, but the open-loop distortion and output impedance are both high due to the common emitter. In contrast, a class AB emitter follower has 0.1-0.3% THD and a few tenths of an ohm output impedance in open-loop.
The Monticelli output stage has to be used with high negative feedback. That too is a downside for a discrete transistor amplifier.
Ed
My impression is that the drivers operate class-A, which would make it a poor choice for a power amp, no?
For me personally, it's just yet another idea around an analog output stage.
I say that retrospectively.
Something that was novel 40 years ago, but if I just speak for myself, I am kinda loosing interest in those things.
About efficiency and power consumption; we know all theoretical efficiency limits for power amplifiers.
So to put it in the whole picture, it will just shave off a few percent at most, seen from the total system efficiency.
The efficiency of any Class-A or a good performing Class-AB is already bad to begin with.
So in the end it's just a tiny little less bad I guess.
You don't use these systems to be efficient to begin with, but more because someone just likes them for a specific reason.
So I don't really see the point of trying.
I can totally see this was a thing back than when Class-D power amplifiers weren't performing as well as they do today.
Ya, I found it interesting because it explained the OPA1611 etc. But initially I made no comment because who cares? DIYA is full of both those who love crude "warm sounding" (aka ~2nd order distortion) amps, class-A amps, tube amps, and those using 20+transistors in pursuit of ppm THD. A classic blameless circuit is what I consider reasonable, with perhaps a few more transistors for protection. I will probably play with the idea in spice, but I don't expect any revelations. It seems to be overly complicated/ redundant so perhaps there is something I still don't understand. We'll see.
No (except for the intention of minimizing distortion).
The difference is that Renardson has an adjustment to minimize distortion and Monticelli's approach is adjustment-free.
Are there actually commercial power amplifiers, where Monticelli's approach was realized ?
Last edited:
You could use MOSFETs or Darlington transistors as output devices, although Darlingtons spoil the rail-to-rail behaviour. You don't need as much driving current then.
I doubt if that will work well in a Monticelli-style output stage, the output transistor is then no longer part of the translinear loop with the bias spreader and common-base stage.
I don't know. It is surely used a lot in op-amps since the patent expired, but power amplifiers?
I don't think this is quite true, let me explain.
I already mentioned that the basic circuit in the patent #1 has no mirrors.
The next step in the chain from conceptual to physical implementation is actually shown in the OPA1611 pic. in post #2.
The idealized bias V sources for the level shifters have been implemented as current sources and pairs of diode connected transistors, in 2 translinear loops (as Marcel pointed out).
This tracks the Vbe with temperature, necessary in practice (as you say) but it doesn't require a mirror.
The next step is to physically implement the level shifter current sources, still shown as idealized in the OPA1611.
Since there is a diode connected transistor available in exactly the convenient spot, it just cries out to be connected like a mirror - to set up a current source transistor Vbe.
But it doesn't need to be, we could use any current source, so the "mirror" is not strictly necessary, even if the bias source temperature compensation is.
So my idea is that any "mirror" imperfections won't matter much, as I discussed in my previous post.
The signal doesn't flow thru the "mirror", as it does in some audio amplifiers.
See what I mean?
My pleasure, as I try to explain a circuit at least it becomes clearer to me, hopefully to others🙂
Well, an emitter follower is just a circuit with inherent local 100% -ve feedback, so naturally has lower output impedance.
But it's well known that global feedback is a more effective use of gain than separate local loops (examples in D. Self's Audio Power Amplifier book)
Once we run the whole amp in closed loop, as it is actually used, then the output impedance will be at least as low for the CE outputs.
Analog Devices address this in one of their application notes.
Or that's the theory anyway.
Best wishes
David
Last edited:
No.
Did you read the OP, where it is explained this is for a switch mode class H, and the impact is vastly more than a few percent?
The simulations I have done of the switch mode class H have it as comparable to the better Class D.
With the Monticelli circuit the quiescent rails would be ~5V +- and if I scale the numbers from the patent then around 100 to 200 mA quiescent.
So only 1 or 2 W quiescent, that is actually better than a Class D that sits there with a million transitions from rail to rail every second, just to keep the output silent!
Not to mention less EMI.
Class D will be more efficient for maximum sustained power, where their outputs are driven to saturation, and the switch losses are a smaller fraction.
But that's not close to my use case, a home theater with reference level peak capacity, because sustained 105 dB is intolerable.
Best wishes
David
Last edited:
This is my first tentative; I adopted a symmetrical drive, as the circuit lends itself to such a configuration.
The driven nodes are the bases of the OP transistors, and I had to correct the area of Q3 to get consistent results:
The results are disappointing: despite the low closed-loop gain (4), the THD is 0.67%, and the frequency behaviour is dismal:
I then tried a different injection point:
As the circuit was not stable, I had to add a summary compensation: C1, C2. Their location and value is certainly not optimal, but the THD has fallen to 0.031%, and the FR has improved too:
The driven nodes are the bases of the OP transistors, and I had to correct the area of Q3 to get consistent results:
The results are disappointing: despite the low closed-loop gain (4), the THD is 0.67%, and the frequency behaviour is dismal:
I then tried a different injection point:
As the circuit was not stable, I had to add a summary compensation: C1, C2. Their location and value is certainly not optimal, but the THD has fallen to 0.031%, and the FR has improved too:
Thanks, I can't run sims at the moment so I am very interested to see your results.
That is worse than I hoped, my first suspicion is that it just needs more gain!
Have you looked at the open loop tranfer function of just the output section, comparable to the Art of Electronics plot?
Best wishes
David
That is worse than I hoped, my first suspicion is that it just needs more gain!
Have you looked at the open loop tranfer function of just the output section, comparable to the Art of Electronics plot?
Best wishes
David
Last edited:
The circuit relies on the currents being proportional to the device sizes. While the PNP half could be built with a second current source of equal value to I9, the best approach is to mirror I9's current.
Getting the right voltages at the bases of Q16 and Q17 relies on the currents through Q23, Q24, Q26, and Q27 all being equal to I9.
I don't know how much error can be tolerated. I do see that this thing is all mirrors. 🙂
I don't have Doug's book, but I believe that claim applies only in the narrow context of local versus global feedback being the only trade-off, while everything else remains constant. We would have no disagreements if only design were so simple.
In practice, there are many degrees of freedom. I like an amplifier that performs reasonably well in open-loop, and then becomes great in closed loop. This is a design style, not an optimization problem.
Ed
I was thinking of something like this as a possible discrete implementation. Connection to the earlier amplifier stages not shown, as I don't know what those stages will look like.
R3N and R3P are there to improve the thermal stability, but they mess up the harmonic mean control law. I hope that can be fixed to some extent with RoptN and RoptP.
R1N, R2N, R1P and R2P are basically cheap current source implementations. As the voltages across them don't vary much in the crossover region, I think you can get away with using resistors, otherwise they have to be replaced with better current source implementations.
Unfortunately, the Darlington output stage costs some voltage headroom.
To adjust the P side's minimum current, disconnect the earlier stages, close S1N and S2N, short the output to ground via a current meter and trim RVP.
To adjust the N side's minimum current, disconnect the earlier stages, close S1P and S2P, short the output to ground via a current meter and trim RVN.
One could also connect R1N and R2N to the other side of R3N, and connect R1P and R2P to the other side of R3P. The signal voltages across R1N, R2N, R1P and R2P are then somewhat smaller, so the current variations are also somewhat smaller.
Here it is, green trace.The magenta one is an ideal straight line, for reference and comparison. Red and blue are the currents through the output transistors. The current transfer is 45/0.2=225: