Opamp + outputstage?

palstanturhin

Member
2011-06-17 11:26 am
Before I go do something really really stupid (design a PCD for it);
what do you think?
 

Attachments

  • circuit.png
    circuit.png
    111.4 KB · Views: 378

mchambin

Member
2011-03-13 8:21 pm
I like the op amp + output stage way of making a PA.

Are you sure of enough phase margin for stability ? There is a lot between op amp output and load.

I like the bias and pre-drivers using the same BJTs; Consider, one step beyond: Using a dual pnp + a dual npn.
 
Before I go do something really really stupid (design a PCD for it);
what do you think?

Build it on proto board first.

For the design itself, you may want to use better than zener regulators for the op-amp supply. They will work but they're horridly inefficient and should one of the zeners fail you will end up with 40 volts on the opamp.
 

palstanturhin

Member
2011-06-17 11:26 am
I like the op amp + output stage way of making a PA.

Are you sure of enough phase margin for stability ? There is a lot between op amp output and load.
It simulates quite well and even allowes somewhat capacitive load. But I honestly do not know how accurate the simulator Spice-models are here?

Using a dual pnp + a dual npn.
That is exactly what I have in mind... ;-)
I just did not have model for the BC846UPN.
 

palstanturhin

Member
2011-06-17 11:26 am
Build it on proto board first.

For the design itself, you may want to use better than zener regulators for the op-amp supply. They will work but they're horridly inefficient and should one of the zeners fail you will end up with 40 volts on the opamp.

Will consider, but I also want to keep it super simple.
The input stage draws about 20mA of current and driving the diamond requires microamperes...
I have, however, 1W zeners in mind.
 
Before I go do something really really stupid (design a PCD for it);
what do you think?

There are two unnecessary transistors, the two for bias aren't needed as the emitter followers driven by the opamp also connect 2Vbe across those lines. You will need some emitter resistors to stabilize the bias I think.

An output stage with gain is often more distorting that a simple follower, but its one way to bridge the gap between +/-15V opamp-land and the main rails.
 

wg_ski

Member
2007-10-10 5:21 pm
An output stage with gain is often more distorting that a simple follower, but its one way to bridge the gap between +/-15V opamp-land and the main rails.

More often than not, class AB “pro” amplifiers are built this way. An op amp either feeding an output stage with gain, or feeding a gain stage and then followers. QSC and all of its clones, Peavey, Crest, all the older good pre-Harmon-buyout Crown stuff, vintage stuff like BGW and Altec, and a lot of no-name low end. I’ve seen many different topologies. Designed plenty of them myself. You can achieve high loop gain, low offset and well matched input devices without the need for doing any component matching whatsoever in production. They are not THAT hard to stabilize, especially because the intermediate gain stages can have a lot of local degeneration which you normally don’t want in a “blameless”. They also seem to be easier to make bulletproof with the proper choice of op amp and internal current limiting/clamping.
 

palstanturhin

Member
2011-06-17 11:26 am
There are two unnecessary transistors, the two for bias aren't needed as the emitter followers driven by the opamp also connect 2Vbe across those lines.

This is quite interesting.
In simulation they seem to make a significant difference in outputstage bias current (25mA vs 1.35A)?

The thermal stability is one of my major concerns though. In theory it is CFP, but in practice it is Darlington devices at the output.

I am not too worried about the distortions the outputstage makes. Opamp seems to do a good job error correcting.
I also tried MOSFETs at the first simulation. Got THD down to 0.00005% 1kHz/1W/8ohm, but already a 100pF capacitance at the output made an high power oscillator out of it.
 
This is quite interesting.
In simulation they seem to make a significant difference in outputstage bias current (25mA vs 1.35A)?
Yes because they are in parallel they change the bias voltage and you have no other means to adjust it. Rather than use a pair of transistors like this what about a bias pot?

The thermal stability is one of my major concerns though. In theory it is CFP, but in practice it is Darlington devices at the output.
Indeed I reckon its not thermally stable

I am not too worried about the distortions the outputstage makes. Opamp seems to do a good job error correcting.
I also tried MOSFETs at the first simulation. Got THD down to 0.00005% 1kHz/1W/8ohm, but already a 100pF capacitance at the output made an high power oscillator out of it.
I'd concentrate on reducing distortion at high slew rates like 30kHz signal, as this is the hardest load for an output stage. You can also try injecting a dynamic load into the other end of the load resistor which will see how well the feedback lowers the output impedance. Ditto for power rails - inject deliberate ripple in the simulations and see how much gets to the output.
 

palstanturhin

Member
2011-06-17 11:26 am
A small update...

I made a proto of the output stage only, just to test its thermal behaviour.

Diamond temperature is critical. A slight increase in its temperature, be it for what ever reason (touching it with your finger, for example), makes a massive difference at the output bias, from milliamperes up to hundreds of milliamps! But as the diamond cools down, the output bias decreases back to normal.
So diamond must definitely be temperature compensated.

And as you may already guessed, surprisingly (?), the output does behave like a CFP usually behaves. Which is nice...

I think I have to proto the rest of the circuit next...
 
A small update...

I made a proto of the output stage only, just to test its thermal behaviour.

Diamond temperature is critical. A slight increase in its temperature, be it for what ever reason (touching it with your finger, for example), makes a massive difference at the output bias, from milliamperes up to hundreds of milliamps! But as the diamond cools down, the output bias decreases back to normal.
So diamond must definitely be temperature compensated.

And as you may already guessed, surprisingly (?), the output does behave like a CFP usually behaves. Which is nice...

I think I have to proto the rest of the circuit next...

You should probably look into adding emitter resistors to your output transistors at the supply rails. It will cost you a bit of gain but you will get better stability in return. Try about an ohm first and adjust from there.
 

palstanturhin

Member
2011-06-17 11:26 am
Very true.
However, I am planning to use a Darlington at the output, even though in the schematic it is drawn as two separate transistors. D2390/B1560 to be excact, which were ripped from a blown HarmanKardon AV-receiver...

So; in that Darlington I have a driver built in. When the Darlington gets hotter, also the driver inside it gets hotter, which makes the driver conduct more, which makes the output transistor conduct less???

That is why I just must build the whole thing to see what actually happens in real life?
Is it enough to keep it stable, or do I need something else?
 
CFP thermal stability is mostly set by the driver and bias circuitry.
The output transistor is not that important.
That said, is for a true CFP.
Here we have an extrapolation, an output stage with gain which is prone to thermal runaway. See Ron Elliot ESP site.

Usually that's true. But in this case he has darlingtons on his outputs connected to have voltage gain ... and they will have tons of it without emitter resistors or other current constraining parts. Even the smallest changes in temperature will upset the operating points in large ways, as he's already seen.

Adding emitter resistors allows him to take control of the gain in that output stage and potentially stabilize it's operation simply by limiting the gain.
 

palstanturhin

Member
2011-06-17 11:26 am
Firstly, thanks to everybody for the valuable input on this matter! Most appreciated!

Unfortunately I have not progressed much with the proto.
As I have resurrected the original idea with the MOSFETs at the output.
In simulation, a more muscular diamond solves the oscillation problem with capacitive loads, which I first had. Diamond with 2x 139/140.
Low value drain resistors (0.1R) does not hurt the THD much. In simulation still 0.00015% 1W/1kHz/8R, 0.005% @20kHz full power. SnR around 110dB. Tolerates almost 1uF parallel capacitance to the load.

I could monitor with an other opamp the bias from the drain resistors, between the drains, and send via optocoupler back to the diamond?
I could do the same also with the Darlingtons?

The simple and easy would be gone, and hard and difficult introduced?
But I have also done some preliminary sketching on a PCB, and there would be a nice little land left just for an active bias servo on it...

Active bias servo also eliminates an other problem I have here, which is the varying supply voltages...

I need to research some more...