OPA2192: Linearity problems

Hi Patrick,

I haven't understood why you are changing R1 and R2. What is the meaning of this:

"For a gain of 1:
- R1: from 180 to 1k5
- R2: from 180 to 1k5

For a gain of 0.06818
- R1: from 180 to 22k
- R2: from 180 to 22k"

Hhm, when changing R1 and R2 you will change the gain, as well? No idea, what you mean :-)

What shunt resistor do you use? And what is your measuring current range?

Kai

Hi Kai,

I try to isolate the problem and therefore there is no shunt connected. Instead I use a high precision adjustable voltage source to simulate the shunt voltage.
As the adjustable voltage source has a range of -10V...10V  it is difficult to step the input voltage in a 0...80mV range.

That is why I first reduced the gain from 8.3333 to 1 to get an input range of 0...0.6827V. As I have seen a linearity problem with the gain of 1 as well I have also tried a gain of 0.06818 (input range 0....10.012V). Also with this gain there is some non-linearity visible.

So there is no shunt, just this first OPAMP stage. I try to figure out what affects the linearity of this first stage or how can I approve the linearity.

Regards,

Patrick

Hello Pruf,

The linearity errors you are observing in your OPA192 circuit might be due to the P-channel, N-channel transition region of the input differential pairs in response to a wide common-mode voltage change. The common-mode voltage change at the non-inverting input is inversely proportional to the gain the circuit uses. The input common-voltage would have to cross within the transition region for this to be a contributor to the linearity error.

One way to test if that is the cause is to increase the supply voltage from the current +/-5 V, to something considerably higher such as +/-12 V or +/-15 V and the run the linearity test again. If the linearity is much improved, then that pretty well points to input cross-over distortion as being the source of the non-linearity error. Using higher voltage supplies moves the transition region further away from the mid-scale range of the op amp as when low voltage supplies are used. 

Regards, Thomas

Precision Amplifiers Applications Engineering

Hi Thomas,

Thanks for the answer. That was actually my first thought. But as of my understanding the voltage of the inputs should stay lower than the transition region when I have a gain of 1 (input voltage < 0.6826V and common voltage range is equal or close to GND which is the case).
According to the datasheet the transition region is in the range of (V+)-3V to (V+)-1.5V. In my case 2V to 3.5V

I now switched platform having the same first stage on a DUAL-DIYAMP-EVM board from TI. This allows me to be more flexible to change parameters (e.g. supply voltage).
I also switched to better measurement equipment (same precision multimeter on input and output).

Question: Is the linearity dependend in any way from the value of the resistors (keeping the same gain)? E.g. changing R10/R11 and R1/R2 from 1.5k to 15k (gain=1).

Best regards,

Patrick

Hey Patrick, 

While Tom helps you with the first stage linearity I ran your filter design through some tools I have developed. Not sure where you got your RC values, and I am sure they are nominally fine for what your are trying to do. A lot of the legacy stuff based on 1970's text books leaves some dynamic range on the table - this is easy to improve as shown this attached file. Not a huge deal, but kind of free integrated noise reduction - two added things, (I have quite a range of published articles on these topics, but I won't bore you with those now). 

1. You don't need that series resistor into the 2nd stage V+ input -the OPA192 does not have matched input bias currents

2. You might consider a final RC filter to set an integration bound on the broadband noise, I was just guessing 100kHz for the noise markers I used. 

Here is the design update discussion, 

4th order MFB redesing to reduce noise.docx

here is the updated TINA file, 

OPA192 improved 4th order MFB.TSC

Hi Michael,

That is very interesting. Thanks a lot for your explanations and the calculations. Although we have some other frame conditions (like the offset voltage of the last stage, that we have to change when altering the gain) we have to consider those inputs in future designs. Noise is also one of the topics we have to tightly monitor in this stages. Tutorials like the TIPL 1311 and following as well as inputs like yours are a great help. Especially if one is not doing precision amplifier designs every day.

One question to your point 1:
So I assume that the OPA2192 is internally bias current compensated. Is in this case the series ressitor into V+ input  counterproductive? Means not having this resistor improves behaviour?

General question:
I have seen that TI has the OPA2388 that, compared to OPA2192, has zero crossover. But supply range is only +/- 2.75V. Do you expect that TI has a new OPAMP like the OPA2388 with a higher supply range?
Or is this technologically not possible?

Best regards,

Patrick

Thomas,

I did some measurements on the DUAL-DIYAMP-EVM board. First with a gain of 1. Result: almost perfect linearity.

Next I changed to a gain of 8.3333. I did two measurements. With the difficulty of stepping a 0...80mV input voltage I get also very acceptable results.
Probably most of the error results from not beeing able to get the values from the multimeters (FLUKE 8845A) at exact the same time.
Attached are the two results with the gain of 8.3333.
I will now try to check the results against our real target board...

Best regards,

Patrick

Hi Patrick,

hhm, how are you simulating the shunt voltage without introducing common mode noise? Or by other words, how the both signal grounds (of the voltage source and your circuit) are connected to each other? I hope you connect both signal grounds directly together and generate two different ultra precise voltages, one for R1 and the other for R2? But both voltages referenced to signal ground of voltage source which is connected to signal ground of your circuit?

The OPA2192 is a modern state-of-the-art OPAmp offering 0.00008% THD. This corresponds to an unlinearity of 8 x 10^-7. Even if you take into account a gain of R10 / R1 = 15 and an increase of unlinearity by a factor of 15 you get an unlinearity of only 15 x 8 x 10^-7 = 1.2 x 10^-5. I think it can be excluded that your measured unlinearity of up to 5 x 10^-4 is caused by the OPA2192.

Thomas has discussed a very interesting issue with the resistors. I aggree with him that you should increase the resistors R1, R2, R10 and R11 by a factor of 10. This will decrease the heating and the temperature drift. But also the linearity behaviour is the best for resistors in the 1k...30k range. It depends on the resistor type, of course. For your experiment I would choose state-of-the-art resistors, through hole, with +/-15ppm temperature drift and +/-0.1% manufacturing tolerances.

Kai

Patrick,

In your original problematic setup, are you using the same Fluke meter to measure the input and output sequentially?  Or are you using two different meters?  There's no right answer here...  Two meters leads towards seeing the mismatch of the linearity of the two meters but cancellation of the 1/f noise.  A single meter takes out the non-linearity mismatch of the meters, but adds susceptibility to 1/f noise since measurements taken at two different points in time.  On the noise topic, are you measuring at the output of the first stage or at the output of your entire circuit post-filters?  If the latter, I do wonder if the difference in noise content being not perfectly Gaussian is causing you grief in your meter(s).  You might even consider adding a low frequency RC LPF to the front of your meter to limit the bandwidth to near DC to see if this is noise contribution, or truly DC non-linearity.

Thanks,
Scott

No Patrick, the x192 I think are higher V CMOS devices, they do not have matched input bias currents - so adding a resistor on the V+ input does not cancel that error, just adds noise, The offset current is the same magnitude as the bias current - a clear indication of not matching in this term. 

Patrick,

We have high voltage zero-drift parts like the OPA2189.  It is not a true "zero crossover" device like the OPA2388, but for an autozero/chopper amplifier this amounts to a difference in AC performance only.  At low frequencies, the autozero/chopper removes the offset almost completely, regardless of the source (package stress, 1/f noise, PMOS to NMOS transitions, etc).

Thanks,
Scott 

Hi Scott,

a cap can be put in parallel to the 9k1 resistor to decrease the noise and improve the stability. A 100n cap would give a corner frequency of 175Hz.

Kai

Hi,

It turns out that the Non-Linearity on our target board was due to a second current measurement circuit with 10 times higher gain that is connected parallel to the measurement circuit I was talking about (for 10 times higher sensitivity in the lower current range).
At the point of approx. 60mV input voltage the output of the first stage of the second current measurement circuit went into its limit (~5V). This unlinearity disturbed the original measurement circuit due to cabling resistance in the range of 200mOhm.

OPA2192 is not the problem!

Thanks to all the contributors for their inputs in this thread. By the way I did learn a lot about input bias current and noise.

Best regards,

Patrick

Hi Patrick,

nice to hear that you could solve the issue!

Good luck :-)

Kai