Simplest discrete op-amps

I've admired the several discrete op-amp designs that have been posted here and elsewhere. But being a minimalist at heart, and someone who does not consider himself an expert in design with discrete transistors, I wonder if good performance could be obtained with simpler circuits.

In theory, you can implement an op-amp with as few as three transistors, but it'd only be good for driving high-Z loads. But add a two-transistor output stage and that's still only five transistors. Imagine a conventional diff amp--NPNs or even n-channel JFETs--with a resistive "tail", followed by a PNP voltage amplifier (with the usual string of biasing diodes and resistor as its collector load) coupled to a complementary output section.

It's a textbook circuit, really, but doesn't seem to find much use either in the designs posted here or in commercial products. Is it because of real, serious shortcomings or simply because it just doesn't look slick enough? Active current sources, sinks and mirrors are cool but are they absolute necessities? How much compliance and DC accuracy and how stable of a tempco of gain, input offset currents, etc. do we really need for our little audio projects? The quest for perfection is laudable, but... Well, it's just something to think about...

You don't need a differential pair for sound. A first, you are more flexible with a feedback impedance than with input impedance; second, you don't need to compensate even harmonics by using symmetry because of sonical reasons. And third, with discrete elements you are free to vary all topology for optimal results, not just surrounding like with OpAmps that are perfect building blocks when sit in a textbook, otherwise they are complex beasts with own characters.

[quote author="NewYorkDave"]I've admired the several discrete op-amp designs that have been posted here and elsewhere. But being a minimalist at heart, and someone who does not consider himself an expert in design with discrete transistors, I wonder if good performance could be obtained with simpler circuits.

In theory, you can implement an op-amp with as few as three transistors, but it'd only be good for driving high-Z loads. But add a two-transistor output stage and that's still only five transistors. Imagine a conventional diff amp--NPNs or even n-channel JFETs--with a resistive "tail", followed by a PNP voltage amplifier (with the usual string of biasing diodes and resistor as its collector load) coupled to a complementary output section.

It's a textbook circuit, really, but doesn't seem to find much use either in the designs posted here or in commercial products. Is it because of real, serious shortcomings or simply because it just doesn't look slick enough? Active current sources, sinks and mirrors are cool but are they absolute necessities? How much compliance and DC accuracy and how stable of a tempco of gain, input offset currents, etc. do we really need for our little audio projects? The quest for perfection is laudable, but... Well, it's just something to think about...[/quote]

You are correct the common opamp can be executed using 3 basic stages; the input differential long tail pair, the interstate level shifter (typically common emitter gain stage) and a unity gain output buffer (typically complementary common collectors).

In an absolute minimal configuration using maybe 4-5 transistors and a handful of resistors you can actually make a working opamp with enough open loop gain to kind of work with negative feedback. However this crude opamp will only work over a limited range of input conditions, and deliver horrible performance compared to what we are used to today.

I have a 300 page book on my shelf that is a collection of IEEE papers on the subject of integrated circuit opamp design and the sundry performance improvements accomplished up to the time the book was published (1978). These technical papers explain what all the extra parts do. I suspect the advances to opamps made since then would fill several more books.

Once upon a time it could make sense to roll your own opamp (typically for low input noise), but these days the benefit available is disappearing fast.

JR

PS: As an obscure footnote to this discussion, how many remember the Norton Amp?. It was an oddball variation on opamps that was made up of just 2 of the three basic opamp stages (output and intermediate level shifter).

In an absolute minimal configuration using maybe 4-5 transistors and a handful of resistors you can actually make a working opamp with enough open loop gain to kind of work with negative feedback. However this crude opamp will only work over a limited range of input conditions, and deliver horrible performance compared to what we are used to today.

Click to expand...

Couldn't disagree more. Check out the FETbloak from TK.

Here is what Quad Eight was up to with very simple opamps run on +/-28V rails. Pretty good sounding stuff. But if you look at the evolution, they upped the transistor count for better performance in later years. The AM-10 has 10 transistors.

Early QE DOA 5 transistors
http://www.avensonaudio.com/tech/Quadeight/mm100schematicsm1.jpg
Later QE DOA 10 transistors
http://www.avensonaudio.com/tech/Quadeight/quad8am10op.jpg

I think there is a certain level of complexity that you need for a general purpose opamp, but often you can design something simpler if you know the conditions that it will be used.

Also check out Nelson Pass' article:
http://www.passdiy.com/pdf/diyopamp.pdf

He starts with 3 transistors, and builds up form there, but stops before reaching a class-AB output stage.

I've been using his 5-transistor version a lot myself. No current booster on the back end, but current sources on both preceeding stages, mainly so it'll work with a larger range of supply voltages...you saw them in my photos of the line mixer prototype.

And they work OK. They take a little getting used to...modern monolithics allow you to be lazy, and these don't. Some of the workarounds are seen as undesirable in some circles. Some of the problems and workarounds:

• -unity gain instability...Miller compensation is not optional for low gains.
  -temperature variable DC offset...so you either match & trim, or AC couple. Or some of both. It strikes me as silly to build a caveman 3-transistor opamp, and then use a DC servo to compensate.
  -limited output drive...but this is an opportunity to bypass it when it's not required, and build it to suit the situation when it is.
  -no definitive specifications...if you need to know GBW, or bias current, or whatever, you're gonna have to figure out how to measure it yourself.

As for commercial products, I believe the gyrators in the Avedis EQ are 3-transistor based. The mic pre in the transistor version of the Soundcraft Series 2 is 4 transistors.

If you don't really need true-differential inputs, take a look at e.g. the functional modules in old Neumann stuff - http://www.gyraf.dk/schematics/Neu_v475.gif

- or the Calrec sub-modules in e.g. the 1061 http://www.gyraf.dk/schematics/Calrec_1061.gif

..this is probably as simple as it gets..

Jakob E.

an op-amp with as few as three transistors, but it'd only be good for driving high-Z loads

Click to expand...

Define opamp.

Look at the studer mic pre for some "three device" circuit topologi that can be made to have low output impedance.

Define opamp.

Click to expand...

The meaning is pretty well-known, but since you asked:

http://en.wikipedia.org/wiki/Operational_amplifier

As for non-opamp topologies: yes, I am quite aware of them--the "ring of three", for example, is a relatively simple noninverting amplifier that can give very good performance. I never studied Neve circuits closely but I know I've seen that circuit in some of their amps. It was even executed in tubes years before that, an AC-coupled example being the Pultec MB-1.

But in this particular thread, I wanted to talk about op-amps--circuits conforming to the basic definition of the first paragraph in the link above. The good intentions behind the "you don't need an op-amp" replies are appreciated but are beside the point.

If a five-transistor op-amp isn't good enough--and we need to know what we mean by "good enough", too--then let's talk about why.

By the way, Brad, that first QE op-amp you posted is almost exactly the topology I was describing earlier (but with collector bootstrapping added). Thanks for the link.

If a five-transistor op-amp isn't good enough--and we need to know what we mean by "good enough", too--then let's talk about why.

Click to expand...

CMRR is shabby and o/l gain depends heavily on the output load condition. If you address this by adding two more transistors, you'll end up with something similar I have designed (SGA-SOA-1_documentation.pdf).

Samuel

OK, good, now we're getting somewhere :grin:. So, the CMRR of the diff amp is poor with a resistive current source... Check. Performance of the voltage amp and output stage biasing can be improved with a CC load... Check. I see you also added HF degeneration in the emitters of the diff amp, as well as input protection... Nice touch.

So, Sam's op-amp is a better general-purpose amp than the five-transistor job. But wouldn't you say the five-transistor job could still be useful when input and output conditions are controlled to some extent?

As an aside, this is why tubes are more fun (to me) than BJTs. You can make a well-behaved, predictable circuit with only a couple of tubes. BJTs are unruly beasts and need to be bound and gagged before they behave themselves :razz:

[quote author="NewYorkDave"]

Define opamp.

Click to expand...

If a five-transistor op-amp isn't good enough--and we need to know what we mean by "good enough", too--then let's talk about why.

[/quote]

There are too many to list. That's why there are so many part numbers and they filled a 300 page book 25 years ago.

What is good enough will be application specific. The obvious example is compare a low gain inverting vs. non-inverting topology. The non-inverting version will reveal many limitation of the simplest implementation.

In modern manufacturing where we can pay as much to place a part as the part costs there is merit to accomplishing a task with less parts, but in most cases this means using an IC since it's only one pop, vs. 10 or more for a bare bones opamp.

The incremental cost to add 10 or 20 transistors to an IC is trivial which may lead to rather complicated designs but properly done there will be benefit from the extra parts. So if the benefit is real and cost minimal it's all good.

I have designed and used discrete opamps in the past when it made sense, these days the GP opamps cover so many bases so well it doesn't make much sense IMO.

JR

All good points... But since we're hobbyists here, for the most part, and not manufacturers--and this is a forum to discuss electronics theory--it's still a good mental exercise to ponder the whys and why nots, with cost-effectiveness being a secondary consideration.

[quote author="NewYorkDave"]All good points... But since we're hobbyists here, for the most part, and not manufacturers--and this is a forum to discuss electronics theory--it's still a good mental exercise to ponder the whys and why nots, with cost-effectiveness being a secondary consideration.[/quote]

Cost/benefit is at the heart of all engineering. Of course if the benefit desired is expanded knowledge or pride of design I can understand the mental math.

I learned a lot from IC opamp design technical papers. While they are often dealing with idiosyncrasies of small geometry integrated devices and IC specific limitations, the electronic circuit theory used often translates to discrete design, and these technical papers explain why they did stuff with scientific basis and clarity, often lacking from inspecting an XYZ product schematic, even with the designer's comments.

Opamp design is a simultaneous pursuit of optimal input characteristics "and" optimal output characteristic "and" extremely high gain at high frequency "with" minimal delay/phase shift, so that gain can be used for negative feedback.

Short list of input characteristics ; DC offset, bias, stability, operating point range, CMRR, PSRR (commonly referred to input), noise current, noise voltage, noise...

output characteristics; drive capability, signal swing, slew rate/rise time, SC protection, crossover distortion, source impedance...

overall. GBW and stability considerations.

Optimizing with different degrees of attention to these sundry characteristics gives individual models their niche. Just adding more internal voltage gain is useless unless that gain comes with minimal phase shift/delay so tricks like cascoding input stages can help there. Degenerating input differential pairs can improve slew rate but will compromise noise and dc offset. Darlington inputs reduce bias current but increase offset, etc.

Opamp design is the balancing of multiple characteristics. It is probably wise to be a little cautious of any new design that drives one specification into the dirt as it may be at the expense of another. Extremely high slew rate opamps designed for S/H applications may not be optimal for audio linearity. Chopper stabilized inputs will give great DC characteristics but with limited bandwidth.

Have fun..

JR

[quote author="JohnRoberts"]In an absolute minimal configuration using maybe 4-5 transistors and a handful of resistors you can actually make a working opamp with enough open loop gain to kind of work with negative feedback. However this crude opamp will only work over a limited range of input conditions, and deliver horrible performance compared to what we are used to today.
[/quote]
What? Have you ever built and listened to a discrete opamp? Sounds like another bookworm-know-it-all found this forum.

[quote author="JohnRoberts"]I have a 300 page book on my shelf that is a collection of IEEE papers on the subject of integrated circuit opamp design and the sundry performance improvements accomplished up to the time the book was published (1978). These technical papers explain what all the extra parts do. I suspect the advances to opamps made since then would fill several more books.
[/quote]
:roll: Here we go...

[quote author="NewYorkDave"]I've admired the several discrete op-amp designs that have been posted here and elsewhere. But being a minimalist at heart, and someone who does not consider himself an expert in design with discrete transistors, I wonder if good performance could be obtained with simpler circuits.[/quote]
If you allow cap output coupling try one of the Boaks (pride permitting :green: ). I even have some PCBs left to get started.

[quote author="tk@halmi"][quote author="JohnRoberts"]In an absolute minimal configuration using maybe 4-5 transistors and a handful of resistors you can actually make a working opamp with enough open loop gain to kind of work with negative feedback. However this crude opamp will only work over a limited range of input conditions, and deliver horrible performance compared to what we are used to today.
[/quote]
What? Have you ever built and listened to a discrete opamp? Sounds like another bookworm-know-it-all found this forum.

[quote author="JohnRoberts"]I have a 300 page book on my shelf that is a collection of IEEE papers on the subject of integrated circuit opamp design and the sundry performance improvements accomplished up to the time the book was published (1978). These technical papers explain what all the extra parts do. I suspect the advances to opamps made since then would fill several more books.
[/quote]
:roll: Here we go...[/quote]

Yes, I designed a discrete opamp into a console (Loft) summing amp back in the late '70s (used LM394 pair for input devices). These days you can buy off the shelf parts that work.

Sorry to waste your time. I should be doing other things now anyhow.

JR

C'mon, guys, don't fight. This was shaping up to be an illuminating discussion.

I actually found this forum a while back when I was researching just this topic, trying to make the simplest possible discrete op amp. I have some opinions, you may disagree with them.

I think diff-pairs can do more (sonic) harm than good unless they are relatively "fancy" (current source in the tail, current mirror load, etc.), or if there is lots of OL gain. but a resistive loaded common source JFET does a pretty good job as a minimal input stage. for better or worse, you end up with a current feedback amp.

I think a current source for the voltage gain stage is necessary.

a simple class-A output follower works good enough with a resistive load.

so I think you can build a functional amp with enough OL gain to be useful with only 4 transistors. not the most hi-fi thing in the world but pretty good sounding.

something else to consider in the name of simplicity is using all transistors of the same type. Ive seen amps posted to this forum that are all BC550/560. Ive made my own from only 2n4401/4403 with good results.

Id like to hear some opinions on other small bipolars that would be a good comprimise for this kind of general purpose use.

[quote author="RogerFoote"]It was John Roberts... These were really nice audio kits. I checked my 1981 Popular Electronics Builders Guide, and the compander I built was an article by John Roberts. And If memory serves, he was doing stuff for Loft at the time as well.

I don't get it sometimes... This forum sucks up to people who can't even solder and then chase away veteran audio guys. Does that make sense?

Even if this isn't the same John Roberts, we should extend the benefit of doubt.

Sheesh![/quote]

Hi, yes I'm the same John Roberts. I put the kit company to sleep after automation gutted the economic incentive for all but hard core hobby types. FWIW Heathkit went from $100M to zero in the same time frame.

An no it takes more than one critic to chase me off, but I see little point in wasting time defending my honor or trading ad hominum epithets. I really do have other stuff I should be doing.

Later,

JR