You don't necessary need ultrasound, but a modulation scheme to put the sound energy outside of hearing range. Ultrasound can be unreliable since phones have di

You don't necessary need ultrasound, but a modulation scheme to put the sound energy outside of hearing range. Ultrasound can be unreliable since phones have different frequency response that affects sensitivity. Also compression tends to apply low pass filter, so overall the range is quite variable. To combat this, you can add forward error correction code, and have the audio source transmit data at low bit rate. Also, it's possible to modulate data using spread spectrum sequence (similar to GPS and ultrawideband) to reduce the energy below hearing threshold. The state of the art is not to embed data at all, but use audio fingerprinting, aka Shazam.