DNN-based CSI-RS Port Virtualization Matrix Design in Massive MIMO System

Abstract

In massive multiple-input multiple-output (MIMO) systems, the channel state information (CSI) is needed to achieve high data rate. However, getting the CSI of the full antennas is hard since channel state information reference signal (CSI-RS) overhead is large. Therefore the CSI-RS port virtualization matrix which groups the antennas in a single CSI-RS port at the base station is used. In this paper, we propose a deep neural network (DNN) based CSI-RS port virtualization matrix design scheme for a time division duplex (TDD) massive MIMO system. The proposed scheme utilizes both the CSI-RS and sounding reference signal to get the CSI and estimates the CSI-RS port virtualization matrix in terms of codebook index. Simulation result shows that the proposed DNN based scheme can achieve up to 25.8% performance gain, particularly in high uplink signal-to-noise ratio regions where more accurate uplink channel information can be obtained.