Deep Learning Aided Beam Prediction in Dual-Band Multi-Cell Massive MIMO System

Sub-6GHz and millimeter-wave (mmWave) are two primary frequency resources in the fifth-generation mobile communication technology (5G) and beyond. Their coordination permits higher reliability in a dual-band multi-cell massive multiple-input multiple-output (MIMO) system. However mmWave is extremely sensitive to multipath fading and interference under different communication scenarios for the poor penetration, thus making the beam prediction critical. In this paper, we propose a strategy to predict the mmWave beams through sub-6GHz channels without estimating channel state information (CSI). We first establish a dual-band multi-cell massive MIMO system model and formulate the problem by selecting beam vectors to maximize the transmission rate of the system. Since the essence of beam prediction is similar to supervised classification, we then construct a deep neural network (DNN) to explore the mapping relationship between sub-6GHz and mmWave channels based on their spatial correlation properties. The numerical results show that the proposed DNN can predict the mmWave beams with an accuracy of over 90%.