MmWave Vehicular Beam Alignment Leveraging Online Learning

Beam alignment accuracy is critical for millimeter wave communications employing beamforming techniques. However, due to the high cost and complexity of the traditional solutions, especially beam sweeping, learning-based methods that use contextual data to select optimum beam orientations are emerging as a desirable technology for mobility applications such as vehicle to everything. However, the majority of available systems involve supervised learning, in which the training data is acquired ahead of time. In this study, we create an online learning algorithm for beam pair selection using a multi-armed bandit framework. The learning machine recognizes beam directions in a specified codebook separated by 3 dB beamwidths and makes recommendations for a set of beam combinations to complete beam training. In vehicular communication applications, the learning algorithm's robustness to location inaccuracy provides a trade-off between overhead and alignment efficiency.