Low-Cost and Power-Efficient Massive MIMO Precoding: Architecture and Algorithm Designs

This study proposes two low-cost precoding architectures for massive multiuser multiple-input multiple-output systems to alleviate multiuser interference, where one has the objective of reducing the power consumption of the conventional 1-bit precoding architecture and the another has the objective of improving the performance of the conventional 1-bit precoding schemes. However, jointly identifying the precoding factor and the precoding vector of the proposed precoding architectures is intractable because they are coupled together. We address this issue by utilizing an alternating minimization (AltMin) framework to optimize alternatively the precoding factor and the precoding vector. Nevertheless, this framework experiences difficulty in optimizing the precoding vector, which we address by proposing a machine learning-inspired algorithm developed from the cross-entropy optimization (CEO) framework to obtain a near-optimal precoding vector. Although the CEO-based AltMin algorithm provides excellent performance, the complexity of the CEO part is relatively high. We thus develop an AltMin algorithm using a sequential greedy descent algorithm as a low-complexity counterpart of the CEO-based AltMin algorithm. Simulation results reveal that the proposed precoding algorithms can successfully determine the weights of the proposed precoding architectures, thereby providing considerable performance improvements over previous 1-bit precoding algorithms. Moreover, the hardware complexity and power consumption of the proposed precoding architectures are considerably lower than those of previous 1-bit precoding architectures.