Modern communication systems may benefit from the availability of sensor data leveraged by sophisticated machine learning algorithms. We recently described how LIDAR (light detection and ranging) on a vehicle can be used for line-of-sight detection and to reduce the overhead associated with link configuration in millimeter wave communication systems. LIDAR is used in autonomous driving for high resolution mapping and positioning. In this paper, we present new LIDAR-based features for machine learning and compare the previously proposed distributed architecture with two centralized schemes: using a single LIDAR located at the base station (BS) and fusing LIDAR data from neighboring vehicles at the BS. We also quantify the advantages of LIDAR-based solutions over solutions based on connected vehicles informing their positions. We use deep convolutional neural networks to process images composed of LIDAR data and/or positions. Using co-simulation of communications and LIDAR in a vehicle-to-infrastructure (V2I) scenario, we find that the distributed LIDAR-based architecture provides robust performance irrespective of car penetration rate, outperforming the single LIDAR at BS and position-based solutions. Under the simulated conditions, the benefits of a centralized data fusion over distributed processing are not significant, meaning that machine learning for line-of-sight detection and beam selection can be conveniently executed at vehicles equipped with LIDAR.
Paper
Full text
Position and LIDAR-Aided mmWave Beam Selection using Deep Learning
Semantic Scholar · Engineering · 2019
Abstract
Modern communication systems may benefit from the availability of sensor data leveraged by sophisticated machine learning algorithms. We recently described how LIDAR (light detection and ranging) on a vehicle can be used for line-of-sight detection and to reduce the overhead associated with link configuration in millimeter wave communication systems. LIDAR is used in autonomous driving for high resolution mapping and positioning. In this paper, we present new LIDAR-based features for machine learning and compare the previously proposed distributed architecture with two centralized schemes: using a single LIDAR located at the base station (BS) and fusing LIDAR data from neighboring vehicles at the BS. We also quantify the advantages of LIDAR-based solutions over solutions based on connected vehicles informing their positions. We use deep convolutional neural networks to process images composed of LIDAR data and/or positions. Using co-simulation of communications and LIDAR in a vehicle-to-infrastructure (V2I) scenario, we find that the distributed LIDAR-based architecture provides robust performance irrespective of car penetration rate, outperforming the single LIDAR at BS and position-based solutions. Under the simulated conditions, the benefits of a centralized data fusion over distributed processing are not significant, meaning that machine learning for line-of-sight detection and beam selection can be conveniently executed at vehicles equipped with LIDAR.