for detecting the Zero-cross of a sensorless BLDC I use a starpoint reproduction and give this Neutral-Point-signal and the evaluated Phase to a comparator. At higher speed there is no problem with zero- cross detection. But at lower speed, the zero crossing detection is asymmetrical: The falling edge comes late, the rising edge comes very soon. So the following phases of the runngin motor are long - short - long - short ....
Who knows something about this effect?