DNN-based Sum-Rate Maximization of Multicell MISO Networks

Abstract

A beamforming vector design based on a deep neural network (DNN) is proposed for multicell multi-input single-output channels with scalar information exchange and local channel state information (CSI). The beamforming vectors are designed making zero generating-interference to the selected interference-free users (IFUs). The set of IFUs is chosen by the DNN based on supervised learning where the inputs can be obtained with only local CSI and limited scalar information exchange. Simulation results show that the DNN is well-trained in estimating the unknown CSI from the inputs with only local CSI in multicell networks.