Vision-Aided Beam Allocation for Indoor mmWave Communications

Abstract

This paper presents a vision-aided beam allocation scheme to help conquer the non-trivial issue such as blockage or link failure scenarios of the millimeter wave (mmWave) indoor wireless communication systems. Particularly, a traditional beam allocation scheme degrades the beam training performance due to a non-convex optimization problem, which contain a combinatorial number of local optima and make them extremely challenging for conventional solvers. Hence, we propose a vision-aided beam allocation scheme to overcome the beam optimization issue and enhance the beam training performance in this paper. We employ a camera at the mmWave access point and leverage their scene information to spontaneously sort out the best allocated beam. We also exploit a machine learning tool to predict the allocated mmWave beam from the camera RGB scene. The simulation results show the performance of the proposed vision-aided solutions in terms of beam training and testing performance.