Abstract
In order to reduce the computation time of hybrid precoding processing while improving the spectrum efficiency (SE) of massive multiple-input multiple-output (MIMO) systems, in this letter, we investigate the sub-array hybrid precoding based on the convolutional neural network (CNN). A constraint-relaxation alternating minimization (CR Alt-Min) algorithm is proposed to create the training set of the CNN. To reduce the computation time caused by iterations in the Alt-Min algorithm, a CNN-based algorithm is proposed. Simulation results show that the CNN-based algorithm reduces the computation time in hybrid precoding processing by an order of magnitude. Moreover, the maximum SE is improved by 26.64% by the CNN-based algorithm, compared with the Alt-Min algorithm.
Funding Information
  • National Key Research and Development Program of China (2017YFE0121600)

This publication has 15 references indexed in Scilit: