A Fast Neural Network Solution for Optimal SCSI Aided Two-User MISO Broadcast Beamforming

This letter re-investigates the difficult statistical sum-rate maximization for two-user multiple-input single-output (MISO) broadcast beamforming (BF). The identified linear structure of the Pareto-optimal (PO) BF vectors interprets the optimal statistical channel state information (SCSI) aided two-user MISO broadcast BF vectors as the linear combinations of the statistical maximum-ratio transmission (MRT) and the statistical zero-forcing (ZF) BF. Inspired by this knowledge, a SCSI enabled BF neural network "MRT-ZF-BNN" is constructed for fast determining optimal two-user MISO statistical BF. The simulation results confirm the improved sum-rate and computational efficiency of the devised MRT-ZF-BNN over the existing baselines.