Improved discrete op-amp

[Mr Evil]

Following on from the work described in a previous thread, I have managed to accomplish my goal of approaching the performance of decent IC op-amps with a simple discrete circuit that can be impemented small enough to replace an IC.

The following results were achieved with through-hole components on a breadboard, and so are probably a little optmistic given that the available SMT parts tend to be inferior, but it's promising nonetheless.

I determined that the major source of distortion was not the output stage as I had first thought, but the Vas, presumably due to the nonlinear input impedance. Cascoding this stage fixed it up nicely, reducing measured distortion from worse than an OPA604 to immeasurable, maybe even better than any op-amp I currently possess.

The attached schematic shows the circuit used. Compared to my first attempt, this one has degeneration resistors on the current mirror (they don't help distortion much, but they do help DC offset a lot - hopefully enough that I can do away with the adjustment pot.) and a cascoded Vas. I did try a MOSFET input stage too, since it's easier to buy dual ones compared to JFETs, but they are far too noisy so I'll have to put up with hand-matching single JFETs instead.

[Mr Evil]

Attached to this post is a comparison of distortion at 1kHz (ignore the spiky garbage below 1kHz). For these measurements gain was set to 100, non-inverting. The red trace is my circuit: Noise is an acceptable level, no distortion components are visible above the noise. Yellow is an NE5534: Noise a bit lower than mine, but there is some 3rd harmonic present (barely visible at -112dB). Blue is an OPA604, which is thoroughly beaten.

[darkfenriz]

Very interesting.
My only question is why do you cascode this way?
It depends on VAS current mode of course, but still VAS mosfet works with low Vds, probably in "'triode range'' (??? word-by-word translation from Polish). Adding a diode or two to change the voltage on the gate of J7 may change things and I suspect you don't need rail to rail voltage swing.
Don't know if I'm right, just a suspicion...
regards

[mzzj]

Damn! you have been faster
I have been planning similar project for a while now, altough I am not sure if i try to fit it at DIP8 size or if i do bigger units with bigger component count . Bipolars were on my mind, cheap as dirt and availlability is not an issue no matter where you live.
I Think i had to upgrade my soundcard before that, any recommendations on 24bit cards for measurements?

[ghg]

Best soundcard for audio measurements, so far:

E-MU 1212m:
http://www.emu.com/products/product....1&product=9872

RMAA data:
http://www.fixup.net/products/benchmarks/1212m2448.htm

Double the cost and you get the E-MU 1820m,
same data, but a convenient breakout box:
http://www.emu.com/products/product....1&product=9871

[boraomega]

I've tried something similar long, long time ago (12-15 years ago) with more than satisfactory results. Unfortunately, at that time, no simulation software was available to me, but practical results were more than good.

[Mr Evil]

Originally posted by darkfenriz
Very interesting.
My only question is why do you cascode this way?
It depends on VAS current mode of course, but still VAS mosfet works with low Vds, probably in "'triode range'' (??? word-by-word translation from Polish). Adding a diode or two to change the voltage on the gate of J7 may change things and I suspect you don't need rail to rail voltage swing.
Don't know if I'm right, just a suspicion...
regards

I cascoded like that because it only takes one component to do it. Extra components make it increasingly difficult to squeeze the circuit into the required volume, which is the main consideration. Yes, "triode range" is the right term, or you might say it's operating in the "ohmic region". It is a bit weird, but it seems to work. I did try altering the cascode bias voltage, but it's hard to measure the difference. Sims say nothing much changes.

Originally posted by boraomega
I've tried something similar long, long time ago (12-15 years ago) with more than satisfactory results. Unfortunately, at that time, no simulation software was available to me, but practical results were more than good.

Nice, but how small is it

[dimitri]

I like it!

[Ultima Thule]

Nice, though the Miller cap from the VAS input is connected directly to the output, interesting...

Michael

[mlloyd1]

mr evil:

very nice!

i am curious if you ever played around with this one during your investigations:
http://www.forsselltech.com/JFET%20Opamp.PDF

also, did you try other OPA604 parts? i saw your previous graphs in the other posts and was suprised how the OPA604 faired. just wondered if you might have a defective one?

mlloyd1