A DNN-based Multi-Objective Precoding for Gaussian MIMO Networks

Abstract

This paper investigates a precoding design for a two-user multiple-input multiple-output (MIMO) network with various objectives, including simultaneous wireless information and power transfer, energy harvesting, and security. Conventionally, precoding and power allocation matrices for these objectives are obtained via different solutions. While in some cases analytic solutions are known, in other cases only time-consuming iterative methods are available. To overcome this issue and unify the solutions for multi-objective networks, a deep learning-enabled framework is proposed in this paper. The proposed deep neural network (DNN)-based precoding learns how to optimize multiple objective functions and find their corresponding input covariance matrices concurrently, efficiently, and reliably. Compared to conventional iterative precoding methods, the proposed approach reduces on-the-fly computational complexity 91.19% while reaching near-optimal performance (99.64% of the optimal solution). The proposed DNN-based precoding can flexibly adapt itself to the different needs of the network and is faster and more robust than transitional approaches, making it an attractive solution for current and future communication networks.