Help needed: low voltage high output impedance sensor measurement circuit

[arg004]

Hi everyone,

I hope you're all doing well.

I'm a graduate student working on building a voltage measurement circuit for one of our sensors. The sensor outputs a voltage ranging from -500 mV to +500 mV, depending on the concentration of certain ions present in the soil. It has a very high output impedance, in the range of several megaohms.

I've built a basic circuit using an INA333 instrumentation amplifier and an STM32L431 microcontroller. I'm using a REF2030 as the voltage reference, which provides dual outputs of 1.5V and 3V. For setting the gain, I used a 1% 100k ohm resistor. The circuit is powered by a 9V battery, and the ADC sample rate is set to 1Hz.

I tested the circuit with various voltages generated by a Keithley 2450 source meter, ranging from -500 mV to +500 mV. The measured output is within 1 mV of the source, which is acceptable for my application.

However, when I connect the sensor, the readings are inconsistent. At times, the circuit provides accurate results within a few millivolts, as verified with an HP 34401A multimeter. But frequently, it produces erratic and inaccurate readings. I can't determine the cause of this issue—whether it's related to the INAMP bias current, the ADC, or grounding.

One observation is that when I connect the circuit's ground to a small wire buried in the same soil as the sensor, the readings stabilize, but the voltage drops by 20-30 mV.

Any insights into what might be causing these problems would be greatly appreciated. Thanks in advance.

[ap]

The INA333 essentially measures two voltages referenced to GND and multiplies the difference between the inputs by a factor. Your sensor is not refereced to GND but floating. You need to tie one of its outputs to GND.

[voltsandjolts]

https://e2e.ti.com/support/amplifiers-group/amplifiers/f/amplifiers-forum/570147/ina333-ina333-input-bias-current-return-path

[Kleinstein]

The main problem is likely the input bias path. In the simplest case one of the inputs / sensro terminals should be tied to ground or an mid point voltage.

For the ADC it can help to rund the ADC much faster (e.g. 10-100 kHz) and than average over multiple main periods (e.g. 100 ms). This suppresses main hum related interference. One may still want to look for clipping on the faster data, e.g. also get the min and max values.

[arg004]

Thanks a lot, guys. Yes, it was likely the bias current return path issue. I did a quick test adding a 10 Meg ohm resistor connected to each input and the other sides tied to the 1.5V reference. It seems to be stable now, although not as stable as the multimeter but no more incorrect readings. I'll do more testing to be sure.
I tried averaging the ADC values earlier, but it was not very effective as many of the values were incorrect in the first place.

[kleiner Rainer]

Another possible source of error is EMI. We are surrounded by inconspicuous RF sources, commonly called mobile phone. Today, any measuring device working with low-level signals needs RF input filtering and shielding. The power supply also needs good input filtering on the mains side, since there is so much crap floating around on the power grid.

RF filtering is described in this application manual starting at page 5-12:

https://www.analog.com/media/en/training-seminars/design-handbooks/designers-guide-instrument-amps-complete.pdf

Even if the INA333 already has built in EMC filtering, more filtering is always better. In a two-stage approach I would put the added filter directly at the input connector. A metal case for the circuit is part of the EMI mitigation. Another improvement can be had by using twisted pair wiring since that rejects RF very well. Ethernet cables are cheap and available everywhere, and you get four pairs with different colours.

Hope that helps.

Greetings,

Rainer DG1SMD

[voltsandjolts]

Place the amplifier very close to the sensor if possible, avoiding EMI pickup from long wires with the high impedance signal.