Discrete Op Amp: Why is This Noisy?

[Joe Kramer]

Essentially, you are tacking on a follower stage to the output of your discrete opamp. I'll use a JFET as an example, but it would work for any type of transistor. Connect the gate of the JFET to the junction of the collector and resistor on the output transistor. Drain goes to V+ and the source goes to ground through a resistor 1k-10k (the value matters less when you are using feedback but in your case, because you are running open loop, it directly affects the output impedence). The junction of that resistor and the source now becomes the output of your opamp.

Thanks Jay--very clear.  Actually I'm thinking along the lines of . . .

Actually I was thinking more of micpre's and such. This is also where my post about Prodigy-Pro forum comes in.  icon_mrgreen

We're on the same page Coriolis!  Probably shouldn't annoy the folks here with such meanderings, but here's an idea I'd like to pursue: Discrete op amp with Ge xstrs, phantom power, transformer in/out.  Hence the inquiry about followers, because the need to drive a fairly heavy load.  But my goals for linearity wouldn't be anywhere near what a Jensen 990 could do (thanks Jay!).  I'd be more interested in "character" over precision.  Still the 990 is supposed to sound quite good, and doesn't look like a totally impossible build.  Would that output stage be considered push pull?

Joe

 

[Gilles C]

I would suggest to check this first, related to the JE-990

http://1176neve.tripod.com/id9.html

More designs here http://www.prodigy-pro.com/forum/viewtopic.php?t=281
Gilles

[Joe Kramer]

Hey Friends,

Haven't had hands-on time to do this mod yet, but I'd like to hit the bull's-eye when I do, hence some questions.  I'm not going to have room on my board for a trimmer so I'm going to do as RG says and knock down my bias resistors by a 10th, then run a 470K ohm from the divider junction to the base of Q1.  But when I do this, will the ratio I've established still hold?  That is, if I go from 5.6M/560K to 560K/56K, will the circuit still see essentially the same bias point?  And how about if I drop it another decimal point, and use 56K/5.6K, same thing, right?  Thanks for the help!

Joe

[Jay Doyle]

Hey Friends,

Haven't had hands-on time to do this mod yet, but I'd like to hit the bull's-eye when I do, hence some questions.  I'm not going to have room on my board for a trimmer so I'm going to do as RG says and knock down my bias resistors by a 10th, then run a 470K ohm from the divider junction to the base of Q1.  But when I do this, will the ratio I've established still hold?  That is, if I go from 5.6M/560K to 560K/56K, will the circuit still see essentially the same bias point?  And how about if I drop it another decimal point, and use 56K/5.6K, same thing, right?  Thanks for the help!

Joe

Yup, for a voltage divider, it is the RATIO that matters not the absolute values.

I think there is a post missing of yours that I never saw.

[Joe Kramer]

Yup, for a voltage divider, it is the RATIO that matters not the absolute values.

I think there is a post missing of yours that I never saw.

Hey Jay,

Many thanks!  Bench time is minimal for me lately, and that's going to help me immensely to be able to jump in and have that change be right on target.  I'm not sure what you mean about the missing post though. . . .

Joe

[WGTP]

Just found this, another Discrete Op Amp again by Jay Doyle.  I think it is in Schematics II.   Note Mu-Amp input and Jfet clippers.  Cool stuff. 

http://www.diystompboxes.com/pedals/schems/Shaka%20Discrete.pdf

[gaussmarkov]

Yup, for a voltage divider, it is the RATIO that matters not the absolute values.

is this correct?  RG sez

The intent of providing a bias voltage is to keep the reference at some middle voltage. If you do this with resistors, any current you suck out of the middle point (or push into it!) will move the reference voltage around. How much depends on the resistor values.

the whole article is here:  Designing Bias Supply (Vbias or Vb) Networks for Effects

[Mark Hammer]

Somebody needs to post either the link to the schem or the schem itself for the ROD-10.  That little table-top half-rack is a compendium of discrete op-amp circuits and circuit tweaks.  Takes the sort of stuff you've seen in the BD-2 and other Boss pedals and then goes farther with it.

[mac]

Not exactly the same but check this out:

http://www.diystompboxes.com/analogalchemy/misc/diodeopamp.html

I built a MXR+ and  Red Fuzz with it and it sounds very good.


mac

[WGTP]

http://aronnelson.com/gallery/KHE/Boss_ROD10_Schematic

Interesting variations.   

[Mark Hammer]

Thank you sir.

For the uninitiated, I do believe that every one of those 3-transistor combinations where you have a pair of FET "facing" each other, followed by a bipolar with a path between collector and the gate of one of those FETs is a discrete op-amp.

[R.G.]

I do believe that every one of those 3-transistor combinations where you have a pair of FET "facing" each other, followed by a bipolar with a path between collector and the gate of one of those FETs is a discrete op-amp.

I took a look. You're correct, all of the two-FET diffamps and a bipolar following is a discrete opamp. Or rather,  a discrete feedback amp. They're all identical.

Why use those instead of integrated opamps? Good question.

I think it is because there is no frequency compensation needed, so they have full output through the whole audio range. A diffamp followed by a bipolar gain stage is a two stage amplifier. If it's done properly, a two stage amplifier is inherently stable in feedback. There are only two time constants to add phase shift (at the frequencies that matter) before the gain gets below unity, so no compensation is needed. The whole genre of two-bipolar feedback pairs used this dodge as well.

That little table-top half-rack is a compendium of discrete op-amp circuits and circuit tweaks.

They did throw in everything but the kitchen sink.

OD I is a tube screamer-ish distortion with asymmetrical clipping. OD II is the same, but with symmetrical diode clipping. OD III seems to be banging two series discrete opamps against  a reduced (5.6V) power supply, and DIST is very similar to OD III but with the bigger 8V power supply and then further limited with diodes to ground. FUZZ is actually our old friend the Univox Super Fuzz octave section, but modified for not-symmetrical rectification. I think this would FWR very imperfectly, which seems to be the point. I'm not surey why the didn't also go ahead and toss in a more perfect FWR section for a more prominent octave.

The tone control section is a modified feedback Baxendall, but with a gyrator-inductor midrange control following it (opamps 1a, 1b, 2a, and 2b). Lots of miscellaneous low pass filters and JFET switching sprinkled around, too.

[Mark Hammer]

This is the sort of circuit/unit that could be very instructive if accompanied by illustrative sound clips.  Being able to see the circuit differences, and hear them as well has always been tremendously instructive for me.

[Eb7+9]

Thank you sir.

For the uninitiated, I do believe that every one of those 3-transistor combinations where you have a pair of FET "facing" each other, followed by a bipolar with a path between collector and the gate of one of those FETs is a discrete op-amp.

it depends what definition you go with - if you go back before the 90's you still see people confusing op-amps for OTA's in papers (and at work !) ... what you're missing to really call this a bone-fide op-amp is : (i) greater open-loop gain, and (ii) a voltage buffer at the output ... a traditional Miller-integrating op-amp can be seen simply as a voltage buffered OTA - that's the part people missed in the past ... so in this case you've got a low-drive op-amp/OTA if you will - where the output drive is provided by a 2k collector load - very marginal drive compared to if you followed the 2k load with, say, an emitter follower and took the output at the emitter ...

OD makers realized at some point that using a high-drive op-amp to clip diodes made for sharp "clip" corners on the waveform - and therefore producing more "fizz" in the sound ... using these circuits in conjunction with diodes in the FBK path gives "soft" clipping ... that's the deal here ...

ps. one obvious problem with the Jensen op-amp clone is in the way the output devices are biased ... in my extimation using 1n4148's for turn-on diodes will likely produce an idling current that's too high in the output pair ...

[WGTP]

http://aronnelson.com/gallery/KHE/Boss_ROD10_Notes_Corrections?full=1

Here is the waveform visual part.  i can't help with the soundclips.   

Thanks for the explainations of the circuit. 

[R.G.]

what you're missing to really call this a bone-fide op-amp is : (i) greater open-loop gain, and (ii) a voltage buffer at the output ...

I guess that we all pay homage to the immortal William Jefferson Clinton's "it all depends on what your definition of "is" is."

The best working definition of an operational amplifier is one where the operation is set by the elements around the opamp, not the opamp itself. Clearly that is what is happening here. The dual JFET stage could be considered an OTA if you modulated the gain with source current - but then all diffamps are like that. The bipolar transistor is running as a transimpedance opamp - you feed it a current, it puts out a voltage. It's analogous to the voltage gain stage in the stock Linn-style power amp and the voltage amplifier stage in the monolithic opamp.

While the definition of the ideal opamp requires nearly infinite gain and near voltage-source output drive, these qualities are relative. The circuit as shown has an open loop gain of 44.5db - that's a gain of 167. So it's "ideal" within engineering tolerances up to gains of 16 or so when the non-infinite gain starts peeking through. The -3db point of the circuit is heavily dependent on the actual devices, but it's somewhere over 190kHz for reasonable guesstimates of parts. The output loading plays heavily into this, but for all loads over about 22K, the loading can be ignored, because there is enough current available to let it look "ideal-ish".

It's worth noting that the same circuit with a current source load instead of the 2.2K collector resistor has an open loop gain of 58db, clearly within the opamp range.

So it's low-ish gain for an opamp, but works without an output voltage buffer for many loads.

Probably they are using the opamp structure for easy setting of bias conditions - which it does very well indeed - and then letting the

What is probably going on is that they're using the oapmp type structure to

a traditional Miller-integrating op-amp can be seen simply as a voltage buffered OTA - that's the part people missed in the past ... so in this case you've got a low-drive op-amp/OTA if you will - where the output drive is provided by a 2k collector load - very marginal drive compared to if you followed the 2k load with, say, an emitter follower and took the output at the emitter ...

Actuall, the input diff stage is a voltage to current stage, a transconductance amp. It's not an operational transconductance amp unless you arrange to modulate it by varying the current, the idea being to let the external parts set the conditions, and part of what RCA defined OTA as meaning. The voltage gain stage takes in a current, puts out a voltage, so it's a transimpedance stage. The combination of the two is a voltage-> current ->voltage, or a pure voltage gain. The output buffer in the normal opamp is just that - a current buffer. I grant you that it's tricky as most people have never heard of transimpedance stages, but it's not an OTA. It is a low drive, low open loop gain operational amplifier.

If you followed the output with an emitter follower, you would have to put in dominant pole compensation to keep it from oscillating in all likelihood. The phase shifts from the diffamp/voltage stage/ and follower stage would add up, the follower would let it run to much higher frequencies, and you'd get classic gain-phase oscillation unless you were very careful. Three time constants, gain and feedback give you an oscillator unless you try hard not to.

OD makers realized at some point that using a high-drive op-amp to clip diodes made for sharp "clip" corners on the waveform

Almost. It's the low open loop gain making the corners softer, not the current drive.

[mac]

...I wonder if we need all those extras components between the diff pair and the output transistor/s in a real opamp, ie, how they affect the tone?


mac

[R.G.]

...I wonder if we need all those extras components between the diff pair and the output transistor/s in a real opamp, ie, how they affect the tone?

To a very large extent, the whole opamp game is to make what's in the opamp circuit NOT matter. That's the whole point of operational amplifiers - you can (mostly) ignore the innards and just have the external components set how it operates. To that end, the makers of the ICs put in all the parts that they need to to make the opamp perform as a good opamp for their typical markets. They don't put in stuff they don't need to make it a good opamp, as they see opamps.

But that's not what you asked.

There are issues with mainstream opamps. The high gain and more than two stages forces the use of frequency compensation, and that starts rolling that high gain off at quite low frequencies. The high gain also makes the entry into the nonlinear regions of operation very abrupt. Abrupt changes in audio signals are always associated with high frequency content.

As long as the opamps in effects never run into clipping or nonlinearities, they are very, very neutral and clear sounding. Let the opamp bang against the power supplies and you hear it as buzzy distortion. But we mostly don't do that - we make the opamp drive diodes and transistors into nonlinearities, and listen to those distortions.

In the case of these discrete opamps, it may be good that the open loop gain is low. Low open loop gain lets the underlying amplifier show through, and softens up the edges of nonlinearity by there not being enough gain to cover up the nonlinearities.

Is that good?

Who knows???

I think that sometimes it is, sometimes it isn't. And most likely, like every other situation in musical electronics, it depends on the circuit conditions you have around it and what you're trying to do. Some musical circuits need good DC accuracy. None of that in low-gain discretes. Some need soft, reliable transition into overload. That point goes to low gain discretes. Some need good high frequency gain, as in filters. Point to IC opamps. Some need... well, you see where this is going. You can't really say "Ahah! This opamp has lots of junk between the diffamp and the output buffer, therefore it sounds worse than (whatever else)." with any degree of accuracy if the circuits are not also stated.

Even among IC opamps there is a big variation in the innards, and therefore in the performance: open loop gain, input bias current, input offset voltage, open loop output impedance, dominant pole location, large signal high frequency response, output current, current limiting (the earliest ones went up in a puff of smoke if you touched the wrong pin), input pin common mode voltage range, input differential voltage range, input and output voltage compliance, power supply rejection, and all the other things on datasheets.

The issue is not simple at all. And there isn't a simple answer to your question, as far as I know. It depends - on everything else.

Sorry.

[gaussmarkov]

great info, R.G.  i have been wondering about the transistor vs. op amp choice.  i have read people's preferences for discretes, observing that ICs sound harsh to them, but not seen any explanation of why that might be.

and it's nice to hear that there are lots of variables left to fool around with.     sometimes it seems like there isn't much left to discover--like it's all been done.  "it depends" suggests to me lots of opportunity for musicality:  tweakable tones and dynamics produced with different circuits.

[mac]

The issue is not simple at all. And there isn't a simple answer to your question, as far as I know. It depends - on everything else.
Sorry.

Thanks, RG. You answered my question completely.


mac