R39 can be quite a bit larger, as the ADC is rather high impedance and the signal is not fast anyway. So some 2.2 K or so should be OK. One could add additional clamping at the base of T10 (e.g. a 3.6-5.1 V zener to ground).
Additional diodes directly at the ADC input may not help very much, as the chip internal "diodes" (usually actually parasitic path to the substrate) often have a relatively low forward voltage.
It could make sense to have a diode directly at the input of the OP to avoid a large negative voltage, just in case.
Depending on the supply current to the ADC, one could consider using a 50 Ohms or similar resistor instead of the ferrite bead - this would limit the current in case of a latch-up and reduce possible damage.
Thanks!
I was thinking about increasing R39 as well. Supply current of the ADC when operating is max. 300µA, according to the datasheet.
Regarding the diodes at the ADC input the datasheet says this:
Keep the absolute voltage of any input within the range shown in Equation 3 to prevent the ESD diodes from turning on.
GND – 0.3 V < V(AINX) < VDD + 0.3 V (3)
If the voltages on the input pins can potentially violate these conditions, use external Schottky diodes and series resistors to limit the input current to safe values (see the Absolute Maximum Ratings table).
So I thought, additional diodes would be a good idea. But then those diodes would not protect the opamp inputs and voltage could go there and make damage...
I have a reverse diode on the output of the PSU but that diode gets disconnected if the relais is turned off. So in this case the sense input of the ADC is connected with the load (which might be a fan or a coil) without any reverse biased diode. I guess that is one huge mistake I made. Actually, the relais should disconnect the sense wires from the sense input as well or there should at least be some protection of the sense input like a reverse diode..
So what I'll try (currently building up the sense stage on a breadboard):
- increase R39
- zener at T10 base (actually: what would that do? according to my understanding the npn is current driven so if it conducts or not - and how much - can't be nailed down to a specific voltage if I'm correct?)
- maybe still try some diodes at the ADC inputs, as the datasheet suggests that? or would that be pointless in this context?
- replace L6 with a resistor (or put a resistor in series?)