Objective Guide to Op-Amp Rolling - Part 1

[peufeu]

Here's an experiment that I would find interesting, so I'm suggesting it.

- Put opamp in unity gain buffer configuration and ground the input, so it outputs 0V.

- Connect output of opamp to output of audio source via suitable resistor, for example 1k.

- Play signal and measure distortion at the output of the opamp.

This will show the distortion added by the opamp's output stage due to load current.

You can also measure intermodulation by using two outputs from your analyzer at different frequencies, and feeding them into the opamp's output via two resistors. Or you can send one signal in the input, and one in the output via a resistor.

Since the output is a virtual ground, there is very little actual signal there, so the distortion can't hide behind it, and it is much easier to measure. Likewise, the ADC will not contribute its own distortion, since it is only acquiring the very small voltage at the output of the opamp, not the full signal amplitude. In the case of intermodulation, the ADC will see the full amplitude of one frequency but not the other, so it does not contribute its own IMD.

This will not show distortion due to other mechanisms, like input stage nonlinearity due to common mode. Only output.

Another interesting experiment is to measure distortion in the same way, with a small DC current pulled from the output of the opamp. You can vary the current, and the distortion profile will change depending on where the offset current shifts the signal on the output stage crossover of the opamp.

 

[valerianf]

I remember a long time ago placing OPAxxx inside my Pioneer Laserdisc.
The sound was really improved.
But nowadays, are the OPA still on a socket on the PCB?
I do not see myself moding a 11channels AVR (I did it for 5 channels but what a pain).

 

[tomchr]

Your current measurements still only show the noise floor of your circuit and the THD+N floor of the analyzer. I.e., in the cases of the LME49720 (= LM4562) and some of the other opamps, you're not seeing the performance of the opamp. You're seeing the performance of your instrumentation.

I dug up the circuit National/TI used for the LM4562 from its data sheet. See attached. With the values shown the distortion is magnified by a factor of 101 (not 100 as it says in the figure).

Tom

 

[syn08]

I dug up the circuit National/TI used for the LM4562 from its data sheet. See attached. With the values shown the distortion is magnified by a factor of 101 (not 100 as it says in the figure).

Classic noise gain manipulation, I think Walter Jung baptized it.

 

[tomchr]

Probably. I know the guys at National didn't invent it. But if you want to show the THD of the opamp you gotta pull a few tricks.

Tom

 

[orchardaudio]

Here are results with the following added:

• Texas Instruments NE5532
• New Japan Radio NJM2732
• Analog Device LT1169

Since many people requested the NE5532, I have that particular result highlighted.

This is still with a 100k load.

 

[solderdude]

The opamps themselves are loaded with 9.0k by the way.

 

[orchardaudio]

Here is the non-inverting unity gain configuration

 

[tomchr]

You're still just seeing the noise of the feedback resistors and the THD+N floor of the AP. For most of the opamps, the results say nothing about the opamp under test.

Tom

 

[AnalogSteph]

I have literally never seen an NJM2742 in any audio product - at 40 nV/sqrt(Hz) of noise density and 2 MHz of GBW this hardly is the most promising part either, so no wonder. Thus the choice of this part seems puzzling. (Likewise, having both LM4562 and LME49720, which as mentioned are literally the same die.) Also, how did you operate NJM2732, a low-voltage part specified up to +6 V (presumably CMOS), from a +/-12 V supply without the magic smoke escaping?

By contrast, NJM4580s and NJM2068s are common as dirt and as such I would recommend acquiring some, even 4565s are still found occasionally (the current Focusrite Saffires are using its "reboot" version NJM8065 to drive the headphone out, it's let's say not exactly ideal for that). The NJM4556A headphone driver is another good one to try. In low-voltage circuits you also see NJM2115, as well as NJM2122 for microphone inputs (very low Vn but not remotely unity gain stable)... I think the Behringer UMC202HD actually uses both.

Note that there are actually two different LM833s out there, LM833N is the original NatSemi part while LM833P actually is TI's take on the MC33078 circuit - another type that used to be quite common at one point, btw.

Likewise, not every NE5532 is created equal... TI's seem to be among the most powerful but afflicted by rather high levels of common-mode distortion. (Study Samuel Groner's setup in this regard.)

Some of TI's OPA16xx lineup (SMD only) has become quite common - OPA1678s are increasingly seeping into audio interfaces, OPA1612s are found in high-end DACs, and I bet if you look around you'll find OPA1622s, the FET input OPA1642s and OPA1656s used as well.

Not sure what the AP software has in terms of graphing options, but I'm not even colorblind yet still find identifying the curves extremely tricky.

Low-gain distortion measurement does have merit, as there are some chips like the LM833 that seem to be using some sort of distortion compensation taking effect then, which wouldn't be active the same at high noise gain. That said, having some sort of distortion magnification setup would definitely be useful. Samuel Groner did a lot of things right, so study his methodology closely.

 

[tomchr]

Yeah. Especially the maroon and the sorta-brownish colours can be hard to tell apart. You can change the colours, though.

Tom

 

[orchardaudio]

Yeah. Especially the maroon and the sorta-brownish colours can be hard to tell apart. You can change the colours, though.

Tom

I have tried to me no matter what I do it just blends together too much. I can export the data to excel if somebody here want to plot it that way. I really don't have too much time to mess with this right now. I will see if I can do something better for future tests.

 

[JohnYang1997]

I have tried to me no matter what I do it just blends together too much. I can export the data to excel if somebody here want to plot it that way. I really don't have too much time to mess with this right now. I will see if I can do something better for future tests.

You can do like this. Insanely bright color can actually look ok in the final graph. And you can alternate high contrast colors. Also thicken some or all traces can help in some graphs too.

 

[pma]

Years ago (2013) I measured some not small differences between the opamps in the CCIF IMD 19+20kHz test in a circuit with gain +101x, as attached.