Essential requirements for complem. output devices (NPN/PNP)

[clintrubber]

Hello,

When stuffing a DOA-PCB some thinking about the relevant requirements for complementary output devices came up.

What makes a NPN & PNP a nice pair ? There's obviously the Vbe-requirement and it'd be nice if speed was alike as well. And hFe.
Other things needing to be alike ?


Special case:

How about (differences in) current gain ? I figure that as they're both large enough all will be fine.
But with both being 'high enough', what when one of them was even higher, like a hFE of 300 vs 200 ?
Would this lower THD even a bit further during the relevant signal-half or are other things going to happen ?


Regards,

Peter

 

[NewYorkDave]

DOA-PCB

How can the PCB be DOA if you haven't even finished stuffing it yet? Talk about pessimism!

 

[clintrubber]

:green: :wink:

Uh, maybe because of the complimentary output devices ? Free stuff just can't work :wink:

 

[Samuel Groner]

Good question!

I guess you want a matching criterion to get as low distortion as possible?

I don't think Vbe and "speed" (I think we'd need to define that a bit more precisely--I assume you mean fT) are per se an important matching characteristics. The general voltage/current transfer curve should very likely match for best even-order distortion cancellation, however (you've got a curve tracer of course). Parasitic capacities might be important as well.

Beta matching is only important if you leave class A.

I think high beta alone does not reduce distortion of the output stage. It does reduce loading of the previous stage which in turn might reduce overall distortion, but this is architecture dependent.

But my knowledge is somewhat fussy here as well so I'm looking forward to more competent answers!

Samuel

 

[tk@halmi]

In practice all that matching of the output pair makes little difference if any. What design are you using?

 

[clintrubber]

[quote author="tk@halmi"]In practice all that matching of the output pair makes little difference if any. What design are you using?[/quote]

Was finishing some 1731/25.. etc and then the thought came up when I measured hFE of the output-devices I had around. One of the two compl. was much higher, like 300 vs 200.
As usual the input-devices are matched w.r.t. hFE & Vbe and the subsequent transistors not - it'd be a bit strange a 'matching'-requirement would pop up again near the output.

But the question about those complementary requirements is of course somewhat different than for a diff-pair (which is usually of the same s*x like NPN+NPN etc - - although exceptions of a diff-pair consisting of a thyristor+beam power tetrode will likely be mentioned here soon :grin: )

And the question was mainly about how differences in hFE would influence things in principle. Indeed the numbers are already such that it won't matter much or at all in practice, but the idea was a little brain-triggering to understand the principle.

Thanks,

Peter

 

[clintrubber]

[quote author="Samuel Groner"]Good question!

I guess you want a matching criterion to get as low distortion as possible?
[/quote]Not especially: the DOA's at hand are about 're-creating the classics', so any dirt in the original may enter now again :wink:

I don't think Vbe and "speed" (I think we'd need to define that a bit more precisely--I assume you mean fT) are per se an important matching characteristics.

Speed as in fT, correct. About the same order of magnitude would be OK I figure, to have comparable stability-situations for pos & neg signal halves. But that's in principle - stability will already have been managed elsewhere so then vast fT-differences at the output devices won't matter (I now state without much further thinking).

Beta matching is only important if you leave class A.

Right, and once having enough gain around anyway the results of differences for output-NPN & -PNP will be little I expect.
Must admit I'm too lazy to solder & measure a DOA with a pair with around 200 [-] & replace with 200 & 300 & measure again. I should unlazy myself in this respect...

I think high beta alone does not reduce distortion of the output stage. It does reduce loading of the previous stage which in turn might reduce overall distortion, but this is architecture dependent.

Agreed - if the previous stage has enough current going anyway it might not care.

Bye,

Peter

 

[tk@halmi]

Sorry if I missed the original question. Perhaps it was already mentioned that degeneration of the output pair through emitter resistors is working for you to reduce gain mismatch as well.
Intrinsically PNPs will be slower than NPNs even when they are intended to be complementary devices. So in the end your ft is limited by the PNP device. Good old global feedback makes its effects small.

PS.: Just a word on curve tracers. They are fine for characterization during solid state device development and it is an integral part of yield improvement for semiconductor fabs. Understanding underlying principles is imperative to design. Accounting for parameter spread is part of design as well. After spending countless hours of tracing and matching I sold my curve tracer to a local audiophile recently.

 

[clintrubber]

[quote author="tk@halmi"]Sorry if I missed the original question. Perhaps it was already mentioned that degeneration of the output pair through emitter resistors is working for you to reduce gain mismatch as well. [/quote]
True ! & wasn't mentioned yet.

After spending countless hours of tracing and matching I sold my curve tracer to a local audiophile recently.

Enter the gainbloake, right ? No more matching of diff-pair devices ! :wink:

I see the potential problem of having the means (curve tracers, THD-meters, LCR-meters, ...) available... it's easy to overdo & analyze it and a lot of components should just be attached to PCBs without much testing !

I really wonder how much testing/matching was & is done for the commercial DOA's !

Bye,

Peter

 

[tk@halmi]

Enter the gainbloake, right ? No more matching of diff-pair devices

Well, the single ended input created its own set of problems like DC offset to mention one. It seems like there is always another trade off we don't want. The final penalties being expense and time.

I really wonder how much testing/matching was & is done for the commercial DOA's

Most likely the input pair was carefully matched and the rest was just packed in.
On a large scale one would buy parts that are already graded to some extent (by the thousands). Have the assembly line make a few thousand opamps. Shove them through test jigs and throw out the ones that don't meet spec. Adjust the price to make up for the cull. The per-unit cost will be less than half of the little guy who is making them individually by careful matching. Standard manufacturing principles apply.
There are actually far more the dual (and pretty closely matched) surface mount transistors available than through hole ones.
If you picked surface mount technology you could have a thousand 252O style opamps made for less than $5 a piece and get a few bad apples only.

 

[CJ]

There are a lot of pairs already matched in the catalogs.

I think you need a base of equal thickness, with equal amounts of doped carriers to form a matched pair.

What was the question?

 

[clintrubber]

[quote author="CJ"]There are a lot of pairs already matched in the catalogs.
[/quote]
Sure, start from compl. pairs that mention the other s&x (NPN/PNP) already in the datasheet. But then still so, you can buy a pair 'together' or get the NPN of brand 1 and the PNP from another corner of the world - both still being the types that are meant for each other. Same idea, but likely somewhat different results.

OK, if it'd really matter the circuit should go back to the drawing board (not this thread-category but the piece of furniture).

 

[clintrubber]

[quote author="tk@halmi"]

I really wonder how much testing/matching was & is done for the commercial DOA's

Most likely the input pair was carefully matched and the rest was just packed in.
...
If you picked surface mount technology you could have a thousand 252O style opamps made for less than $5 a piece and get a few bad apples only.[/quote]

Nice summary :thumb:
I'm sure the most of us are not as half as pragmatic when making our own DIY-boxes as one needs to be when really producing stuff. That'll each be the essence of producing (& staying in business) and the essence of DIY (make stuff that could quite easily be as good as - or at least in our minds it simply should since we overengineered so much :wink: )
here