Deep Learning for Beam Management: Opportunities, State-of-the-Arts and Challenges

Benefiting from huge bandwidth resources, millimeter-wave (mmWave) communications provide one of the most promising technologies for the fifth-generation wireless networks. To compensate for high pathloss of mmWave signals, large antenna arrays are equipped at base stations and user equipment to establish directional beamforming, where beam management is adopted to acquire and track the optimal beam pair with the maximum received power. Naturally, narrow beams are expected to achieve larger beamforming gain, whereas it could impose enormous training overhead and high sensitivity to blockages. Fortunately, the amazing success of deep learning (DL) has stimulated increasing interest in applying it to address those issues. In this article, we first elaborate the motivations of applying DL in beam management. Then, the current state-of-the-arts are reviewed, where their research routes and key features are discussed. Finally, challenges and future opportunities are summarized, highlighting DL design insights and novel beam management mechanisms. We hope this article can stimulate more striking ideas and exciting contributions for DL assisted beam management.