Op-amp low frequency/DC linearity errors

[splin]

Is there any easy way to determine the linearity of an op-amp from the datasheet? I'd like to build a test circuit to measure the linearity of some other components, so I'd like to know that an op-amp, configured as a unity gain buffer is not contributing more than .1ppm error over a limited output range. Measuring such errors is not trivial so can they be estimated? Datasheets rarely seem to mention linearity other than where THD is specified.

The large signal gain is obviously one of the most important factors as is CMRR. The latter can be avoided by operating with a 0V common mode voltage, but what other important factors apply? Presumably the more inherently linear the various stages of the amplifier are, the less the need for high gain?

Would an inverting or non-inverting arrangement be better from a linearity point of view?

I'm guessing that the output stage is the major contributor to non-linearity so how big an output swing could I use before linearity deteriorates significantly - I don't need more than 5V swing, but even +/- 1V would be useable? Would 80 or 90% of the minimum specified swing be reasonable? Would using amps with higher supply voltages help much?

Could anyone recommend suitable op-amps? Offset and drift are not really problems but low noise is important. It would be driven by a low impedance source so input resistance and bias currents would not be an issue.

What about audio amps such as the LME49990 which is advertised as 'ultra low distortion, ultra low noise'?

For AC signals (< 100Hz), the open loop gain roll-off will be a factor but presumably other frequency dependant non-linearity errors appear that aren't applicable at DC?

Sorry for the number of questions but any insights you can provide would be appreciated,

Splin

[moffy]

Look up an article by Jim Williams of Linear Technology called: "Max Wein, Hewlett Packard and a rainy Sunday afternoon". He basically reduces the distortion of a wein bridge to immeasurable levels. Covers similar ground to you. It is the feedback that gives the low distortion, so if you have lots of gain then distortion will be low. It is better to invert than not that way you avoid common mode input stage distortion.

[splin]

Look up an article by Jim Williams of Linear Technology called: "Max Wein, Hewlett Packard and a rainy Sunday afternoon". He basically reduces the distortion of a wein bridge to immeasurable levels. Covers similar ground to you. It is the feedback that gives the low distortion, so if you have lots of gain then distortion will be low. It is better to invert than not that way you avoid common mode input stage distortion.

Thanks, an interesting read. It doesn't really answer any of my questions though; he even ended up keeping the non-inverting arrangement but used another op-amp to remove the common mode voltage.

I did find Nat-Semis's AN-1671 'Gain and Linearity Testing for Precision Operational Amplifiers' which is quite relevant although I am having trouble understanding it - especially Fig. 2 which labels the upper and lower traces but appears to have 5! (or 3 with 2 of them rapidly transitioning between 2 levels).

It still doesn't help in how to identify an op-amp which might meet the << 1ppm DC non-linearity requirement. It does show the results of testing the LM4562 (aka. LME49720):

    "The gain is therefore up near 10 million to 40 million, and the nonlinearity is down near 1/20 part per million, also."

So perhaps the ultra-low distortion audio amps are the way to go. I see that the LTC1150 has a remarkably high  'Large-Signal Voltage Gain' (typical) of 180dB so perhaps even better, but it is a bit noisy? Trouble is it may have inherently poor non-linearity and thus rely on the high gain to compensate?

Any suggestions as to which op-amps might provide the best non-linearity performance?

[c4757p]

I see that the LTC1150 has a remarkably high  'Large-Signal Voltage Gain' (typical) of 180dB so perhaps even better, but it is a bit noisy? Trouble is it may have inherently poor non-linearity and thus rely on the high gain to compensate?

Transistors are inherently nonlinear. The voltage->current transfer function of a BJT is exponential, and of a MOSFET is... well, complicated, but also not linear. Any transistor-based linear circuit must rely on gain and feedback to compensate for nonlinearity.

[f5r5e5d]

http://www.sg-acoustics.ch/analogue_audio/ic_opamps/ Groner tests more characteristics than most op amp manu datasheets in a common format

http://electronicdesign.com/analog/supply-bootstrapping-reduces-distortion-op-amp-circuits shows a way around input impedance nonlinearity

http://www.ti.com/lit/an/sboa015/sboa015.pdf shows some composite op amp compensation options

[splin]

I see that the LTC1150 has a remarkably high  'Large-Signal Voltage Gain' (typical) of 180dB so perhaps even better, but it is a bit noisy? Trouble is it may have inherently poor non-linearity and thus rely on the high gain to compensate?

Transistors are inherently nonlinear. The voltage->current transfer function of a BJT is exponential, and of a MOSFET is... well, complicated, but also not linear. Any transistor-based linear circuit must rely on gain and feedback to compensate for nonlinearity.

Of course. Apologies if I didn't make it clear, but what I was getting at is how good a figure of merit is Avol alone for non-linearity (when CMMR is not an issue because there is not common mode voltage)? The LCT1150 has very high gain, but if its linearity without feedback is particularly poor the linearity performance with feedback may be worse, or no better, than a lower gain amp with better inherent linearity. I don't know, hence the question.

Linearity below 1ppm won't matter to most, but for a 6.5 digit or better DMM it does. Or for example, in another old thread about Kelvin Varley dividers there was discussion about the need for very high input resistance buffers, > 10^15 ohms, to maintain .1ppm accuracy. But what would you use to implement that buffer? It would need its linearity to be rather better than .1ppm. Both of these applications have the additional problem of a 10V or common mode range.

Perhaps its not a difficult problem but I don't have the expertise to know how to tell which opamps might be suitable based on the datasheet specs.

Thanks, Splin


[EDIT: added DMM after '6.5 digit or better']

[splin]

http://www.sg-acoustics.ch/analogue_audio/ic_opamps/ Groner tests more characteristics than most op amp manu datasheets in a common format

http://electronicdesign.com/analog/supply-bootstrapping-reduces-distortion-op-amp-circuits shows a way around input impedance nonlinearity

http://www.ti.com/lit/an/sboa015/sboa015.pdf shows some composite op amp compensation options

Thank you - some excellent references which will take a while to read. At a first glance many of the tests are measuring THD so is there a rule of thumb of how to relate (low frequency) distortion figures to DC non-linearity?