LIDAR and Position-Aided mmWave Beam Selection With Non-Local CNNs and Curriculum Training

Efficient millimeter wave (mmWave) beam selection in vehicle-to-infrastructure (V2I) communication is a crucial yet challenging task due to the narrow mmWave beamwidth and high user mobility. To reduce search overhead of iterative discovery procedures, contextual information from light detection ranging (LIDAR) sensors mounted on vehicles has been leveraged by data-driven methods produce useful side information. In this paper, we propose lightweight neural network (NN) architecture along with corresponding LIDAR preprocessing, which significantly outperforms previous works. Our solution comprises multiple novelties that improve both convergence speed final accuracy model. particular, define novel loss function inspired knowledge distillation idea, introduce curriculum training approach exploiting line-of-sight (LOS)/non-line-of-sight (NLOS) information, non-local attention module performance for more NLOS cases. Simulation results benchmark datasets show that, utilizing solely data receiver position, our NN-based scheme can achieve 79.9% throughput an exhaustive sweeping without any 95% searching among as few 6 beams. typical V2I scenario, proposed method considerably reduces time required desired throughput, comparison inverse fingerprinting hierarchical schemes.