An Efficient Beamforming Architecture to Handle the Trade-Off Between Performance and Hardware Complexity in Multiuser Massive MISO Systems

Beamforming is the fundamental concept of wireless communications to serve several users through multiple-antenna transceivers. The advent of massive multiple-input multiple-output (MIMO) systems leads to an investigation into beamforming to reach optimal performance. However, fully-digital beamforming implementation has to face important challenges in terms of consumed power and cost of the RF chains and converters. To solve the problem, some studies focus on reducing hardware complexity by dividing the beamforming into digital and analog parts, known as hybrid beamforming (HBF). In HBF systems, the dominant part of the hardware complexity depends on the architecture of the analog network, which determines the required RF components. In this paper, we categorize the analog beamforming parts based on the connectivity between RF-chains and antennas. Moreover, we show the impacts of the connection on the analog beamforming matrix. To this aim, we propose an efficient connectivity architecture in which the antennas are divided into fully-connected and singly-connected groups so that we can deal with the hardware complexity-performance trade-off. In addition, to achieve further performance improvements, we propose a dynamic architecture that adjusts the connection of the RF-chain antenna to the channel state information (CSI) through a switch network. Analytically, we propose an efficient approach to calculate the zero-forcing precoder, which is proper for both fixed and dynamic architectures. A two-part algorithm based on the greedy search method has also been developed to obtain switch states in dynamic architecture and then implemented by a deep neural network (DNN). The simulation results confirm the theoretical analysis and the suitability of the proposed architecture.