Simple(ish) discrete op-amp

[Mr Evil]

I've been pondering making a design for a discrete op-amp. Something that is not too complex and yet offers performance as good as IC op-amps.

Attached is one design I am considering (only simulated at the moment - it's too damned hot to do any real work). It has a single-ended input instead of differential for simplicity (lower component count plus no need for Miller compensation). This makes it 'current feedback' (as in low-impedance feedback node - not to be confused with 'current feedback' meaning feedback proportional to output current).

Biasing for the Vas is accomplished with current mirrors. This helps with symmetry, in particular compensating for the potentially unequal Vgs of the two Vas JFETs.

Both the input stage and Vas are cascoded, which in conjunction with the current mirrors provides both constant current and constant voltage operation for those transistors, minimizing thermal memory distortion.

Output stage uses JFETs, for simplicity.

Simulated performance compares well with IC op-amps, with 134dB of open-loop gain, 30MHz GBP, very high input impedance, low output impedance, low distortion and around 180V/us of slew-rate. It's not quite as simple as I had set out to create, but it's not too bad.

Comments and criticism welcome!

[vAD]

At first sight like all it is good, but something guards. Did not look more deeply.
For the information look.

[darkfenriz]

Hi
what are J5 & J6 for?
How about compensation and stability issues?
What's the minimal (closed loop) gain?

I like its simplicity, but I don't understand the issues above so it's hard to comment now.

best regards

[Nelson Pass]

Originally posted by Mr Evil
Comments and criticism welcome!

Interesting and imaginative. Let us know how the simulations
sound.

[Mr Evil]

Originally posted by vAD
At first sight like all it is good, but something guards. Did not look more deeply.
For the information look.

Argh! Every time I think up something that could be a little novel, someone posts something prior that's very similar! Hmph.

Originally posted by darkfenriz
Hi
what are J5 & J6 for?
How about compensation and stability issues?
What's the minimal (closed loop) gain?

I like its simplicity, but I don't understand the issues above so it's hard to comment now.

best regards

J5/6 are cascodes. Normally you see JFET cascodes with their gates connected to the source of the other transistor, but having thought a little about it I think it's better to connect the cascode's gate to the gate of the other transistor (I've seen this done somewhere else, which is what made me think about it). My reasoning is this: Normally, the cascode FET holds Vds of the other transistor at its Vgs. Ideally this is a constant, but because of finite gain of the cascode it will actually change with current. However, if it's referenced to the gate of the other transistor instead, assuming both transistors are identical, then the change in the cascode's Vgs will be exactly matched by the Vgs of the other transistor, meaning Vds is independent of current.

Compensation is not needed. It should be unity-gain stable, with the caveat that comes with current-feedback designs that the resistor from output to feedback node must be present and of sufficiently high value (I haven't worked out the minimum yet). Of course that might be different in reality.

Originally posted by Nelson Pass


Interesting and imaginative. Let us know how the simulations
sound.

Don't worry, I'll be exercising real transistors as soon as the temperature drops below sweltering.

[darkfenriz]

well
I thought J7/J8 were cascodes for bjt VAS and adding another common gate stage (J5/6) doesn't change things much. Do I miss something here?
regards

[Mr Evil]

Well you're right that the BJTs do provide gain, and if the JFETs weren't there then they would work entirely as the Vas, but the JFETs provide gain too, and so are also the Vas.

[ilimzn]

In practise, your current mirrors will need emitter resistors. And, when you add them, you can easily multiply current without needing multiple transistors - just divide the corresponding leg emitter resistor with the current multiplier.

Otherwise, it looks very interesting! Let us know how it works for real when you ave the chance to try it out...

[Workhorse]

EVIL'S Intersting CIRCUIT...very imaginative....waiting to get the real life response.....go and build up the discrete opamp fast, nice and easy.......

[vAD]

Excuse, it was very fast the thrown sight, but all the same, there will be a feeling that it to work well.

Originally posted by Mr Evil

J5/6 are cascodes. Normally you see JFET cascodes with their gates connected to the source of the other transistor, but having thought a little about it I think it's better to connect the cascode's gate to the gate of the other transistor (I've seen this done somewhere else, which is what made me think about it). My reasoning is this: Normally, the cascode FET holds Vds of the other transistor at its Vgs. Ideally this is a constant, but because of finite gain of the cascode it will actually change with current. However, if it's referenced to the gate of the other transistor instead, assuming both transistors are identical, then the change in the cascode's Vgs will be exactly matched by the Vgs of the other transistor, meaning Vds is independent of current.

Compensation is not needed. It should be unity-gain stable, with the caveat that comes with current-feedback designs that the resistor from output to feedback node must be present and of sufficiently high value (I haven't worked out the minimum yet). Of course that might be different in reality.

Very interestingly, did not know, whether there is really working circuit.
ilimzn (In practise, your current mirrors will need emitter resistors.) completely I support.