LiDAR data-based base station selection using deep learning

Abstract

BS selection plays an important role in establishing communication links in the mm-Wave-based communication systems. For selecting the best BS, each BS performs a handshake with the UE. This selection is further challenging due to the increased communication overhead imposed by the handshake between multiple BSs and UEs. In this research, we investigate BS selection for AVs using DL. Due to the rigid foundation of signal processing algorithms in statistics and information theory, they do not account for non-linearities and imperfections in the system, which can be mitigated by DL-based communication systems. Further, FL is applied where BS broadcasts its position to all nodes to reduce communication overhead. The dataset generation is described using the RT technique and LiDAR sensor. Finally, the simulation results verify that the proposed algorithm performs considerably better, with an accuracy of 1.7 times better than GPS-based selection and a reduction of 96.38% in overall data size using FL-based selection, thereby reducing communication overhead significantly. The generality of the proposed model has been further tested by using techniques like TL for variation in the number of samples for training, the city used for simulation, and the number of BSs. Furthermore, hyperparameter tuning is performed using PSO on CNN utilizing LiDAR data.