This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

INA122 DC offset at output, caused by noise?

Other Parts Discussed in Thread: INA122, INA126

Hi, thanks in advance for your time and expertise.

I am using an INA122 to measure input and output voltage in an GaN inverter.

When I measure the sensor response with the scope when the circuit is not switching (no noise), voltage measured in the RC output filter is clean and perfectly linear.

When the circuit is in operation, but no connector is attached to the INA122 except output RC filter, measure is also OK.

When the sensor is connected to the FPGA ADC, DC offset value appears at the output. This DC offset error is reduced when the noise generated by the switching is also reduced. The response approaches then the previously measured one (almost ideal, orange in the picture).

The load of the INA122 is an RC filter, and two twisted cables going to the ADC of the FPGA.

Why is there a DC offset in the output? How is the noise affecting the INA ? How can I fix it? I've tried using a common-mode choke, but this did not fix the problem.

Thanks,

Iñigo

Sensor Input voltage is 0-400V which at INA122 input its 0-200mV and with G=5 means 0-1V output for the ADC.

In orange, measured sensor response when inverter not switching. 

In blue, measured response when inverter switching. The "3 staged trend" of the data points I think is just a comsecuence of the operating points, rather than INA Gain since noise makes measure approach ideal response.

In red: Measure moves towards good value when operating point is varied so that less switching noise is generated.

  • Hello user4700544,

    Any 'noise' or 'switching' that appears common to both inputs will be attenuated by the CMRR specification of the device. The amount of attenuation depends on the frequency of the 'switching' and the gain of the device. Please refer to the "Common-mode Rejection vs. Frequency" graph on page 4 of the INA122 datasheet.

    As an example, let's say that the frequency of the switching is 100Hz, the amplitude is 10mV, and the gain is 100V/V. At 100Hz the typical CMRR specification of the INA122 is only 60dB, which is equivalent to 1mV of offset per 1V change in input common-mode voltage. Since in this example the input common mode change is 10mV, the offset (referred to the input) is 10uV. This offset will be amplified by the gain (100V/V) so it will appear as a dc offset of up to +/-1mV. The same holds true for the power supplies. Any noise on the power supplies will be attenuated according to the device's PSRR specification as found in the data sheet. Please note that these errors are in addition to all other dc errors (e.g. initial input offset voltage). The following articles may be of use. Even though we wrote them with current sensing application in mind, the accuracy section and PCB layout tips are applicable to all applications.

    www.eetimes.com/document.asp
    www.eetimes.com/document.asp

    The best way to mitigate this is to filter the inputs, provide proper power supply decoupling, and perhaps to look at using a device with better ac CMRR performance. In general, 3 op amp instrumentation amplifiers have better CMRR performance than their 2 op amp counterparts. One issue, however, that you may have to concern yourself with is the input common-mode voltage vs. output swing of 3 op amp instrumentation amplifiers, especially when operating with a single supply. For more information, please read the following FAQ and the associated article links:

    e2e.ti.com/.../3373.understanding-instrumentation-amplifier-vcm-vs-vout-plots

    Note that 2 op amp instrumentation amplifiers have a similar relationship but it is generally not as restrictive as the 3 op amp topology. I wrote an article on the 2 op amp topology that is also linked in the FAQ above.

    I notice you're using the INA122UA, which is the low-grade version of the device. You may want to evaluate the high-grade version (non-A). I also noticed that there is a 1Mohm resistor across the inputs, which actually sets the gain to 5.2V/V, but in your text you talk about a gain of 5V/V. If you really want a gain of 5V/V I recommend simply removing the gain-setting resistor. The resistor also generates noise that can couple into the circuit, so if it's not necessary I highly suggest removing it.

    Finally, you may want to try the INA126 high-grade version. It is also a 2 op amp instrumentation amplifier but appears to have better CMRR performance.

    Hope this helps!
  • Hi Pete, firstly thank you for your quick answer.

    The 1MOhm resistor, as you pointed out, is not connected since high value resistors always add noise.

    The input common-mode noise should be filtered out with a CM choke along with a DM filter inside the EMI filter box. This would be conducted or radiated noise picked in the cables that run to the sensor input.
    However, I thought that the GND connection for the input bias current return might be inserting noise from the ground return of the ADC, appearing as common mode signal at the input. 

    Since the INA122 accepts only 0.1V of common-mode voltage below Vcc- (GND in single-supply as in this case), I tried connecting that centre node to 2,7V instead of GND to increase the roof of VCM. This did not solve the problem. This also increased medium-high freq noise superimposed. I guess thats because of the Zener parasitic impedances in high frecuency.

    So, following your advice I am going to try also the INA126U.

    I'll soon give you feedback with my findings!
    Meanwhile, thanks and let me know if you come up with any other alternative to improve the solution.

    Iñigo