This DOA-33 thing from bryston

[mikep]

The input stage of this simple discrete amp is particularly interesting. Why is this not more common? no current sources, fully differential/complementary. I mocked it up in the simulator and it aparently performs pretty well. Can't swing as near to the rails as some other designs and, I dunno why, but the sim tells me that the current in the output transistors is in the micro amps even driving +20 into 600ohms !? never seen it lie like this before.


having troubble posting an image. here is a link to a pdf with the schematic I am talking about:
http://www.bryston.ca/BrystonSite05/pdfs/PREAmplifiers/BP26+MPS2_SCHEMATICS_2005-01-17.pdf

mike

 

[Samuel Groner]

No current sources.

Sure it has two current sources--two lousy ones though (the 100k resistors).

Why is this not more common?

Because CM performance is unhappy--simulate CMRR and you'll see what I mean. There are applications where this is no problem, but a general-purpose opamp should have (very) good CMRR.

Can't swing as near to the rails as some other designs.

It doesn't look to me as if this were true, at least in inverting configuration. But I might be wrong...

The sim tells me that the current in the output transistors is in the micro amps even driving +20 dBu into 600 ohms!?

Quiescent current or transient/peak current? The former doesn't change with load.

Samuel

 

[cuelist]

Sentec, a well respected Swedish HiFi manufacturer in the 70's and 80's had a similar gainblock design using 6 transistors. They called it "double differential" and it was used in most of their designs.

I recall totally symmetrical circuits being very trendy in the 80's particularly with certain high end audio manufacturers (Mark Levinson etc).

Technical indulgence if you ask me :green:

 

[mikep]

yeah, I meant no ACTIVE current sources. but the overall simplicity isn't really what I was talking about. It is the dual long tailled pairs at the input that got my attention, one NPN, one PNP. are there any other examples of this out there? anyway, you could hop this up with current mirrors to improve the CMRR, right?

EDIT: and I don't know what I was thinking about the "swinging to the rails" bit.

mike

 

[tk@halmi]

Douglas Self's books are full with this type of design. He adds front and back of VAS buffering so they look a great deal more complex. However, it is the way to go if you want to banish distortion from you circuits. This one may not have the open loop gain to yield the good numbers Self reports, and it is the simplest version of a fully symmetrical opamp that may be practical.
Needs depend on the exact application. We have a lot of general purpose opamps on the market that people are trying to avoid. If it were to be used as a differential line input stage it should benefit from higher CMRR using at least constant current sources.
Adding current mirrors to the LTP changes gain and shifts poles to mess around with your bode plots. Stability may suffer so you will have to play with it a bit.

 

[mikep]

[quote author="tk@halmi"]Douglas Self's books are full with this type of design. [/quote]

ding ding ding ding! time to go to the library...

 

[Samuel Groner]

Before you run: I don't think he has written more than a few words about this (at least in the copies I've got). Better check Hood...

It is the dual long tailled pairs at the input that got my attention, one NPN, one PNP.

It looks nice on paper, but the question is: what's the benefit? There are some, mainly first-order cancellation of input bias current and a rather good slew-rate symmetry, but that's about it. Not much for a significantly increased complexity.

You could hop this up with current mirrors to improve the CMRR?

Not sure how much a current mirror does without a current source. Actually the CMRR of these complementary topology is rather good with a (actually two) current source and resistive load. Above 100 dB simulated IIRC.

Are there any other examples of this out there?

www.groupdiy.com/index.php?topic=11105
Studer has some power amps, IIRC. And Borbely likes it a lot, as does Per-Anders Sjöström.

Samuel

 

[clintrubber]

As long as you don't need a larger common mode range than the most suited choice from either a PNP or NPN kind of diff.input pair can give you, then I thought there's not much reason to use it (and that's even apart from the increased complexity).

Never used it (no need so far), so note this is not from experience, just from recollecting what I understood/have read.

Exact details escape me fro now, but IIRIC it was related to issues with the takeover-point of the two diffpairs (call it the crossover-point if you like :wink: )

 

[cuelist]

Designing "general purpose" building blocks will generally lead to more complex circuits than any one application needs.

I like the "old school" 4-6 transistor designs. OK so they won't fit every applications but when do you need lowest possible noise AND high CMRR AND high current drive AND xxx.... all the same time?

With perhaps 3-4 circuit variations you can pretty much cover any need.

Wasn't Einstein who said: "things should be a simple as possible but not more simple than that".

 

[tk@halmi]

[quote author="Samuel Groner"]Before you run: I don't think he has written more than a few words about this (at least in the copies I've got). Better check Hood...[/quote]

OK, it is not Self, not Hood, but Randy G. Slone: "High-Power Audio Amplifier Construction Manual". ISBN: 0071341196. Slone lives and breathes fully sym. topology.

 

[tk@halmi]

[quote author="cuelist"]Wasn't Einstein who said: "things should be a simple as possible but not more simple than that".[/quote]

Yes, but Einstein had some serious flaws of theory. One revealed by his act of giving up his child for adoption so he could get a cushy job at the patent office.

 

[bcarso]

haha Einstein.

The fully complementary stuff has indeed been around a lot. Borbely likes to do it with FETs, and in some cases since depletion mode devices allow it he simply puts a resistor between each diff pair's commoned source connections---no resistors to rails, let alone I sources.

Was it William of Occam that said "Transistors shold not be multiplied beyond necessity?"

No, his statement was a bit more general.

EDIT: reconsideration

However, the offset current due to the summed base currents does change, so for relatively high source/feedback impedances that could become an issue.

There's another approach that resembles the Borbely approach but allows much higher slew rate, but I don't have time to discuss it right now.

 

[adrianh]

If I am right and I get my common mode rejection
with a transformer then it is no big issue.

The transformer has a common mode limit of around 2000V

 

[bcarso]

CMR is sucky all right---about 54dB from quick sims.

If you add even more transistors and make each stage into a current-mirrored push-pull drive to each driver Q, as mikep speculates, it goes up to about 108dB (of course the simulator assumes perfectly matched transistors). The slew rate is not particularly good but it is indeed symmetrical---about 10V/us for a set of reasonable values. Sims were done with a 1k source impedance, which probably helps to limit the input signal rise/fall times and makes the pulse response look cleaner.

The distortion of the existing circuit for a good-sized output swing and gain of 10, driving a stiffish feedback network (1k, 111ohms) is about 10 ppm at 1kHz, again according to simulation. With the more complicated version it gets down to about 1.4ppm. Needless to say such numbers are, by themselves, likely to be sonically meaningless. For what it's worth the sim says most of the distortion is second still, with a fair amount of third and not a lot of much higher, which indicates that even with all the mirror symmetry you are still not symmetrical---the PNPs and NPNs are just not that well-matched. There's probably a place to tweak that would tune out the second.

If anyone wants to see the improved circuit example just PM me and I'll send a pdf. There are a couple of idealizations to make life easier, like some voltage sources inserted for bias, but they could be made real without too much trouble.