Transformer-enabled hybrid precoding for TDD large-scale antenna arrays systems with channel sensing

Abstract

Hybrid precoding is a crucial technique for massive multiple-input multiple-output (MIMO) systems owing to its capability to offer an adequate beamforming gain while reducing the hardware cost. However, the nonconvex objective functions and constraints pose great challenges to hybrid precoders design. The conventional precoding method that contains a two-step process including channel estimation and precoding design based on such estimate is not necessarily optimal to tackle this problem. In this article, a transformer-empowered approach waiving high-dimensional channel estimation is proposed to design precoders with the goal of simplifying the complicated hybrid precoding problem into the optimization of neural network structure. Specifically, the proposed approach learns channel sensing from uplink pilots and then operates downlink hybrid precoding depended on interleaved-polymerization-transformer-based analog precoding network (IPTAP-Net) which decomposes on a peruser basis and conventional linear digital precoding algorithm to reduce computational complexity in multi-user systems. Simulations show that the proposed methodology acquires remarkable performance improvement and strong robustness, as compared to state-of-the-art hybrid precoding schemes. Furthermore, proposed approach develops a generalizable talent for manifold multi-user cells.