Deep Learning-Based Rate-Splitting Multiple Access for Massive MIMO-OFDM Systems With Imperfect CSIT

Abstract

Due to the high dimensionality of the channel state information (CSI) in massive multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, acquiring accurate CSI at the transmitter (CSIT) with limited feedback overhead is difficult, severely degrading the performance of conventional SDMA beamforming techniques. To this end, this paper proposes a deep learning (DL)-based end-to-end (E2E) rate-splitting multiple access (RSMA) beam-forming scheme for massive MIMO-OFDM systems, including an analog beamforming network (ABN) and a model-driven RSMA digital beamforming network (RDBN). We adopt an E2E training approach to jointly train the proposed ABN and MRBN to obtain better beamforming performance. Numerical results show that the proposed DL-based E2E RSMA beam-forming scheme significantly improves the system capacity and outperforms the state-of-the-art schemes.