Model-Driven Deep Learning for Massive Multiuser MIMO Constant Envelope Precoding

Abstract

Constant envelope (CE) precoding design is of great interest for massive multiuser multi-input multi-output systems because it can significantly reduce hardware cost and power consumption. However, existing CE precoding algorithms are hindered by excessive computational overhead. In this letter, a novel model-driven deep learning (DL)-based network that combines DL with conjugate gradient algorithm is proposed for CE precoding. Specifically, the original iterative algorithm is unfolded and parameterized by trainable variables. With the proposed architecture, the variables can be learned efficiently from training data through unsupervised learning approach. Thus, the proposed network learns to obtain the search step size and adjust the search direction. Simulation results demonstrate the superiority of the proposed network in terms of multiuser interference suppression capability and computational overhead.