A Comprehensive Study in 5G MU-MIMO Beamforming: From 3GPP Codebook to AI-Based Precoding Techniques

This work utilizes channel reciprocity to study and evaluate several beamforming techniques for a Time-Division Duplex (TDD) Multi-User Multiple-Input Multiple-Output (MU-MIMO) system [2] for a 5G network. Our system consists of a gNodeB with 32 transmit antennas offering 12 randomly spaced User Equipments (UEs) inside a practical 3GPP-compliant Cluster Delay Line (CDL) [3] channel environment. By applying uplink Sounding Reference Signals (SRS), the gNodeB estimates downlink channel state information (CSI); the Proportional Fair (PF) [4] algorithm manages user scheduling. We present a comprehensive study and a performance comparison across three different categories of beamforming techniques: traditional linear precoding methods including Block Diagonalization (BD) [5] and Regularized Zero-Forcing (RZF) [6], the 3GPP CSI Codebook Type II [10] that wisely referenced for its complex orthogonal beam combinations for multi-users, and an AI-based algorithm [7] leveraging an unsupervised deep neural network to optimize beamforming matrices. Our analysis focuses on system throughput under single and double-transmitting layer conditions. The main contribution of this study is a detailed mathematical implementation, along with a performance investigation of these several beamforming techniques within a complex MU-MIMO CDL channel. To the best of our knowledge, this scenario has not been extensively examined in previous works. This study offers important direction for the implementation of next 5G and beyond wireless communication technologies.