查看文章

Millimeter wave (mmWave) could be a key technology to support high data rate demands for automated vehicles. MmWave needs array gain for the best performance, but this requires correctly pointing the beam, known as beam alignment. Dynamic blockages make beam alignment challenging in the vehicular setting. This paper proposes to leverage a vehicle's position along with past beam measurements to rank desirable pointing directions that can reduce the required beam training to a small set of pointing directions. The ranking is conducted using a learning-to-rank approach, which is a popular machine learning method used in recommender systems. The learning uses a kernel based model, and a new metric for evaluating ranked lists of pointing directions tailored to beam alignment is proposed. The proposed method provides a scalable framework for exploiting context information.