BEMF zero crossing in BLDC motors?

[ifonlyeverything]

From my understanding (which may be completely wrong), BEMF of a brushless DC motor is AC in nature i.e. it swings negative to positive. One common way of sensing zero crossing is forming a virtual ground between all 3 phases and then measuring BEMF of the undriven (floating) phase relative to this virtual ground. I've seen several schematics showing this, discrete implementations for hobbyists like to use the LM339. One such example is here https://simple-circuit.com/wp-content/uploads/2018/03/pic18f4550-sensorless-bldc-motor-controller-bemf-circuit.png I've also seen motor controller implementations that measure phase voltage with an ADC.

How is this possible when negative voltage inputs typically damages comparator and ADC inputs? LM339, for example, does not like negative inputs per https://www.ti.com/lit/an/snoaa35b/snoaa35b.pd and the ADCs on my STM32 have an absolute minimum of -0.3V relative to ground.

[f36]

The BEMF on each phase goes positive and negative with respect to the virtual ground (that you're trying to make with the resistor network), not your circuit ground, so your comparator or ADCs will never see a negative voltage (ignoring any effects of parasitics in the circuit construction).

The voltage on each phase will be pinned to ground or your bus voltage when one of the MOSFETs for the phase is on, and the BEMF will be between those two extremes when both MOSFETs are off. If the voltage on one phase tried for whatever reason to go above or below those limits, the body diodes on the FETs would start conducting and clamp the voltage.

If you probe the phase outputs to your motor while it's running with the oscilloscope ground connected to your circuit ground, you'll see something like this. Note the position of the waveform relative to the baseline.