After adjusting the mirror for reflecting beam back to the AOM, the QPD signal shows the better alignment. Before the voltage different readout is about 1.2 V, now it's reduced to ~500 mV.

We'll try to add a translational stage for the mirror for better alignment.