Beam alignment for millimeter wave vehicular communications

Abstract

Millimeter wave (mmWave) has the potential to provide vehicles with high data rate communications that will enable a whole new range of applications. Its use, however, is not straightforward due to its challenging propagation characteristics. One approach to overcome the propagation challenge is the use of directional beams, but it requires a proper alignment and presents a challenging engineering problem, especially under the high vehicular mobility. In this dissertation, fast and efficient beam alignment solutions suitable for vehicular applications are developed. To better quantify the problem, first the impact of directional beams on the temporal variation of the channels is investigated theoretically. The proposed model includes both the Doppler effect and the pointing error due to mobility. The channel coherence time is derived, and a new concept called the beam coherence time is proposed for capturing the overhead of mmWave beam alignment. Next, an efficient learning-based beam alignment framework is proposed. The core of this framework is the beam pair selection methods that use side information (position in this case) and past beam measurements to identify promising beam directions and eliminate unnecessary beam training. Three offline learning methods for beam pair selection are proposed: two statistics-based and one machine learning-based methods. The two statistical learning methods consist of a heuristic and an optimal selection that minimizes the misalignment probability. The third one uses a learning-to-rank approach from the recommender system literature. The proposed approach shows an order of magnitude lower overhead than existing standard (IEEE 802.11ad) enabling it to support large arrays at high speed. Finally, an online version of the optimal statistical learning method is developed. The solution is based on the upper confidence bound algorithm with a newly introduced risk-aware feature that helps avoid severe misalignment during the learning. Along with the online beam pair selection, an online beam pair refinement is also proposed for learning to adapt the codebook to the environment to further maximize the beamforming gain. The combined solution shows a fast learning behavior that can quickly achieve positive gain over the exhaustive search on the original (and unrefined) codebook. The results show that side information can help reduce mmWave link configuration overhead.