Deep Learning-Based Hybrid Precoding for FDD Massive MIMO-OFDM Systems with a Limited Pilot and Feedback Overhead

Due to the large dimension of the channel state information (CSI) in frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) orthogonal frequency di-vision multiplexing (OFDM) systems, achieving spectral-efficient hybrid precoding with a limited pilot and feedback overhead is difficult. To this end, this paper proposes a deep learning (DL)-based hybrid precoding scheme for FDD massive MIMO-OFDM systems to jointly model the downlink pilot training, uplink CSI feedback, and downlink multi-user broadband hybrid precoding modules as an end-to-end (E2E) neural network. We adopt an E2E training method to jointly train all neural network modules with the sum throughput as the optimization goal so that the explicit channel estimation at the users and the explicit channel reconstruction at the base station (BS) can be avoided with reduced pilot and feedback overhead. Numerical results show that the proposed DL-based E2E scheme outperforms state-of-the-art schemes.