LiDAR-Guided Predictive Beam Optimization in Real-World Mobile V2I mmWave Networks

Abstract

The proposed method utilizes light detection and ranging (LiDAR) sensory data to predict the optimal beams in highly mobile vehicle-to-infrastructure (V2I) scenarios, and it is evaluated using the DeepSense 6G dataset. Simulation results show that the Temporal Neural Network (TCN) based method outperforms the state-of-the-art model, achieving 98.3% top-1 accuracy in future beam prediction while reducing beam training overhead by 80% and achieving an average power saving of 79.53%. Remarkably, the LiDAR-assisted method operates independently of previous beam information or calibration, providing a reliable alternative to conventional techniques. These results indicate that beam alignment challenges in mmWave and terahertz communication systems can be effectively mitigated by leveraging LiDAR-guided sensing and TCN-based predictive learning.