Rate-Distortion-Function-Inspired CSI Feedback for FDD Massive MIMO-OFDM Communications

Abstract

In this paper, we investigate the downlink channel state information (CSI) feedback for frequency-division duplexing (FDD) massive multiple-input multiple-output (MIMO) communications with orthogonal frequency division multiplexing (OFDM) modulation. By formulating CSI feedback as a lossy compression problem, we study its fundamental limit governed by the rate–distortion (R–D) function. To obtain a tractable surrogate for high-dimensional CSI, we leverage a variational upper bound that can be computed via variational autoencoder (VAE) training as a conservative benchmark. Building on this upper bound, we establish a set of sufficient optimality conditions that provide theoretical guidance for practical design. Guided by these conditions, we propose a VAE–based neural network, termed the Rate–Distortion-Function-Inspired Network (RDFI-Net), for CSI feedback. Specifically, RDFI-Net employs the hierarchical autoregressive probabilistic model to strengthen latent entropy modeling for efficient entropy coding, and incorporates enhanced neural modules to approximate the derived optimality conditions. Furthermore, we develop a low-complexity rate adaptation scheme that enables variable-rate CSI feedback without training and managing multiple models. Numerical results show that our proposed approach achieves superior performance.