- The voltage produced by the pH electrode is read by a low bias-current buffer such as the LT1793 or ADA4627. The outer shield of the BNC will be attached to PCB ground with as low-impedance path possible. Should I compensate for average bias current, with a low leakage thin film type resistor?
Bias compensation works only in opamps without internal compensation! (only in simple bipolar opamps). So don't do that.
- Because guarding the shield is impossible (connected to reference electrode) guarding can only be used on the PCB near the opamp input. Is it worth guarding this short of a distance?
Yes, its even more important since there is short distance to low impedance voltage sources. See https://www.analog.com/en/design-center/reference-designs/circuits-from-the-lab/cn0407.html
- The boards will be meticulously cleaned in IPA in ultrasonic cleaner and then baked at a low temperature to increase insulation resistance.
Ok
- Since voltage can easily be induced on the shield of the BNC, I think a low impedance connection is necessary. Is any advantage gained by producing a virtual ground to drive the reference electrode?
You can measure then in pseudodifferential manner, so might help to cancel ground noise effects.
- Since the response of the pH electrode is slow (10s of seconds) analog filtering can be used. Would a small capacitor directly across the pH electrode suffice (film/low leakage type) or is it better to leave the electrode directly connected to buffer input and have a second stage filter for AC >~1Hz (active butterworth?)
Direct connection is better. Cap adds area where current can leak.
- Is it advised to shield the amplifier circuit on-board? I will already try and employ best EMC/noise practices: solid ground plane, mindful to separate digital and analog sections, rise time limiting. Should I use ferrite beads where signals enter/exit the shield?
That could help. If you will see excess unknown offset, that could be a EMI pickup (demodulation) by opamp.
- I am planning to use a boost converter to generate +/-18V supply (https://www.ti.com/lit/an/slua288/slua288.pdf) with an LC filter to reduce switching noise (I am using ADP1613 with output current of ~100mA per supply). That supply will be placed physically far from the analog section. Near the analog section, +/-15V LDO with good noise performance will be used to supply the opamps. Are ferrite beads, or additional filters required in this application?
Probably not necessary. However if your environment is water tank with salt at some industrial facility I will strongly recommend to fully isolate power and communication signals. I saw many times full phase voltage on that tanks.
- Output of amplifier is read with a 16 bit (hope for ~12 bits ENOB) A/D integrated into a Renesas MCU. Should I set amplifier gain/offset to maximize A/D range assuming there is no noise or DC offset on my signal? Or should I leave some breathing room at the sacrifice of resolution?
Leave 10%, set maximum oversampling to increase ENOB, add 50/60Hz postprocessing filter, delta-sigma ADC are superior for that applications.
Yes, thank you for your feedback. I will try and select an opamp with internal guard drive, or add guarding externally. I do wonder - is it worthwhile to bend up the leads of the Opamp input and make the connection to BNC directly in-air. Certainly would reduce leakage, although I am not sure this is necessary for pA range of leakage. I do have access to insulation resistance tester where I can measure intrinsic leakage of my board. Dielectric polarization of the FR4 must also contribute, no?
My MCU can measure in so called pseudo-differential but the acquisitions are a few ms apart. This would remove DC common mode voltages, no?
ADC is 16 bit, sadly not S/D (and the S/D resources on this MCU are already sampling output from another amplifier circuit) -
https://www.renesas.com/us/en/document/apn/16-bit-ad-converter-performance-ra2a1?language=en.
The tank environment is very salty - up to 2000mS! There are pumps submerged into the tank - operating at line voltage but switched with a relay. These pumps will be off during acquisition, although when they are on they must contribute significant noise. There is also another sensor in the tank - an EC sensor which comprises two stainless steel electrodes: one "grounded" through a voltage to current amp and another excited with a 100Hz +/-10V square wave to prevent electrode polarization. This too will not be powered during pH acquisition, but the voltage to current amp is always connected. I based the circuit loosely off of this app note:
https://www.analog.com/media/en/analog-dialogue/volume-50/number-4/articles/fully-automatic-self-calibrated-conductivity-measurement-system.pdf. I do wonder what interference and/or ground loops will be formed from this strategy... the amplifier directly after the pH electrode is powered +/-15V so should have plenty of headroom.
Unfortunately isolation would be difficult, since the MCU which has the A/D resources is not isolated. I hope the "grounded" EC electrode will wick away the large potentials (AC, DC, or electrostatic as the container is plastic) but there still may be a considerable 60Hz noise component. Also, second stage filtering will hopefully have sufficient rejection to those higher frequency disturbances. I don't have a good intuition... Perhaps this is one advantage to an instrumentation amplifier - wouldn't that kind of common mode noise be removed? I do still worry about any DC offsets that might change between calibrations.