Iterative SINR Precoding with Deep Learning Design for Network Massive MIMO Systems

In this paper, we investigate the sum-rate maximization problem for network massive MIMO systems in which all base stations (BSs) transmit data cooperatively. We introduce the iterative signal-to-interference-plus-noise ratio (ISINR) precoder by alternately iterating the SINR and precoding vectors. By exploring the ISINR precoder structure, we represent the precoder using some low-dimensional parameters, which can be generated with deep learning networks to improve real-time computation. Thus, the precoding vectors can be directly computed without iterations. Simulation results show that the ISINR precoding algorithm can reduce computational complexity while increasing sum-rate, and that deep learning design can achieve similar performance while further reducing complexity.