[SOLVED] Choosing transistors for current mirrors?

[iimagine]

If you're referring to the audio amp, I'm not sure it would be particularly useful to anyone but me. It's designed from scratch, using no other completed design...

You and I have something in common I'm only interested in the current mirror part. It can be useful to many other things other than audio stuffs.

 

[Darkcobra]

You and I have something in common I'm only interested in the current mirror part. It can be useful to many other things other than audio stuffs.

Well you might like the schematic after all then, as I've been playing with the current mirror in it recently, and it demonstrates a few tricks that perhaps you haven't seen.

Note for any experts who look this over: This amp is intentionally DC coupled, as it's not strictly for audio. I'm aware it has crossover distortion. The feedback is also set up for better stability into possibly capacitance loads, not for minimal distortion. If you see any other issues, please don't tell me. This circuit is a personal exercise, and I'm trying to figure it out for myself! However, if I've given [iimagine] any incorrect info regarding the operation of current mirrors in particular, do correct me.



Q16 and R4 form a current limiting diode (CLD), set for 5mA. This serves as the reference current for the current mirror. Rather than throwing away the 5mA, it's perfect for powering the long-tailed pair; which operates at a lower voltage, so that the CLD never goes out of compliance (meaning the voltage differential across it is never so small that the CLD stops operating properly).

Q9 is the reference side of the current mirror. Normally you'd have only one transistor on the output side, and get one output current the same as the reference. But you can also have more than one output transistor, to generate multiple equal current sources from a single reference. Here I'm taking advantage of that, with two 5mA outputs (Q11 and Q12), and by placing them in parallel I get a higher current of 10mA to drive the biasing network.

By doing it this way, I save a few parts and mA; at least compared to the ugly way I was doing it before. There is one potential downside. If you overdrive the amp so hard the output sufficiently nears the upper rail, so will the biasing network. Q11 and Q12 go out of compliance. And a bit of this feeds back through Q9 and Q16 to the long-tailed pair. Unintentional feedback paths are generally a bad thing, but it seems not to have any noticeable ill effects in this particular circuit, under what are already extreme conditions.

I understand you can also generate one or more currents lower than the reference, by adding an emitter degeneration resistor to the output transistor(s), Widlar-style. I haven't experimented much with this yet.

 

[iimagine]

Normally you'd have only one transistor on the output side, and get one output current the same as the reference. But you can also have more than one output transistor, to generate multiple equal current sources from a single reference. Here I'm taking advantage of that, with two 5mA outputs (Q11 and Q12), and by placing them in parallel I get a higher current of 10mA to drive the biasing network.

That's clever! Something that I have never thought of. Thanks for sharing.

 

[godfreyl]

please don't tell me ... I'm trying to figure it out for myself!

I like that attitude! It's a refreshing change from all the "Please give me the circuit" type posts from people who want to get a passing grade without doing any work.

Anyway, I'd like to point out a couple of things you may or may not be aware of, and the simulator won't reveal:
The first is thermal effects. As a rule of thumb, a transistor's vBE decreases about 2mV per degree C, for the same collector current. For a given temperature, increasing the vBE by about 26mV doubles the collector current.

The result is that if we keep vBE constant and increase the junction temerature by about 15 degrees, we can expect the collecter current to double. Since increased current will generally lead to higher temperatures, there's a nasty positive feedback reaction. This can result in thermal runaway, where temperature and current increase until the device burns out.

Normally we only worry about this with class AB output stages, however it could also be a concern with a 2 transistor mirror. For example if the output transistor of the mirror has vCE=20V and it starts with 5mA collector current, it will be dissipating 100mW. That's enough to cause significant heating in a small device like a TO92. Without doing the math, I strongly suspect you'd end up with a burnt transistor or two.

If you're using dual (or quad) matched transistors, this is less of a problem. With both transistors on the same piece of silicon, the temperature difference between them will be much smaller.

The other thing to watch out for is that in simulators, all the transistors are perfectly matched i.e. every 2n2222 is exactly the same as every other 2n2222. In real life they can differ quite a lot. On the other hand, I suspect that datasheets are often overly pessimistic regarding the spread of vBE.

All of which makes emitter resistors seem like a good idea as they can:
a) Increase the output impedance (your original problem)
b) Provide thermal stability
c) Make the circuit much less sensitive to component mismatching

Interesting amp design, btw, especially the output stage. ;-)

 

[FvM]

I see that degeneration resistors reduce the achievable output swing, but 100 mV voltage drop (the 10 ohm variant in the previous discussed test circuit) already helps a lot.

Although self-heating of semiconductors isn't included in the standard SPICE models and simulations, you can make your own behavioral models that take account for it.

 

[Darkcobra]

Anyway, I'd like to point out a couple of things you may or may not be aware of, and the simulator won't reveal:
The first is thermal effects.

I'm glad you mentioned that. I do know about thermal effects, and on smaller circuits, I usually estimate dissipation for certain components and make sure it doesn't result in a significant temperature rise. Sometimes I'll do an ambient temperature sweep analysis too.

I have done neither on this circuit yet. And now that you mention it, I think you're right, those transistors will probably burn.

So I have posted a flawed current mirror example, and that should be known to anyone seeing it.

The other thing to watch out for is that in simulators, all the transistors are perfectly matched i.e. every 2n2222 is exactly the same as every other 2n2222. In real life they can differ quite a lot.

I know about that too. Which I now realize, in combination with the thermal issue, will lead to horrible current sharing in that parallel current output arrangement.

Will definitely have to consider emitter resistors, and probably a different package like SOT-223 as well.

Interesting amp design, btw, especially the output stage.

I know what "interesting" sometimes means, but if so, that's ok. :wink:

I keep a copy of every previous version, plus a log of what I changed and why. Should I reach the point where I'm confident in the design and get it working properly in the real world, I'd love to post them all up on my website so people can see how it evolved. It would be educational for some, humorous for others.

I see that degeneration resistors reduce the achievable output swing, but 100 mV voltage drop (the 10 ohm variant in the previous discussed test circuit) already helps a lot.

100mV isn't bad at all. I'm still playing catch-up with new info I've absorbed from this thread and a few others I'm monitoring, so I haven't tried using the resistors yet. But I will soon.

Although self-heating of semiconductors isn't included in the standard SPICE models and simulations, you can make your own behavioral models that take account for it.

That would be very useful. I'll do some Googling and see how it's done.

 

[godfreyl]

I know what "interesting" sometimes means....

Was meant in a good way. For instance, the inclusion of Q4 and Q10 is unusual, but a nice touch. They'll do a good job of making sure the output devices switch off when they should, and as fast as reasonably possible. That could be especially important as you mentioned "unusual loads" and "not just for audio".

Also, while the actual mirror has problems, I think the idea of mirroring the input stage current source to get the 2'nd stage current source is quite neat. I don't remember seeing that before.

 

[Darkcobra]

Was meant in a good way.

I honestly didn't expect that, so it's a pleasant surprise and very encouraging. Thanks!