Residual Diffusion Models for Variable-Rate Joint Source Channel Coding of MIMO CSI

TL;DR

RD-JSCC tackles the CSI feedback bottleneck in FDD MIMO by fusing a lightweight autoencoder with a residual diffusion denoiser, enabling adaptive, high-fidelity reconstruction under challenging channel conditions. The two-stage decoding and chi-prediction allow efficient low-latency inference, while preprocessing to the angular-delay domain exploits channel sparsity. Key contributions include residual-diffusion formulation, two-stage training, and Matryoshka representation learning for variable-rate compression, achieving state-of-the-art NMSE and BLER performance on COST2100 with flexible complexity controls. The approach offers practical benefits for real-world CSI feedback, providing robust performance across diverse channel distributions and estimation imperfections, with explicit mechanisms for low-latency operation and dynamic rate adaptation.

Abstract

Despite significant advancements in deep learning-based CSI compression, some key limitations remain unaddressed. Current approaches predominantly treat CSI compression as a source coding problem, neglecting transmission errors. In finite block length regimes, separate source and channel coding proves suboptimal, with reconstruction performance deteriorating significantly under challenging channel conditions. While existing autoencoder-based compression schemes can be readily extended to support joint source-channel coding, they struggle to capture complex channel distributions and exhibit poor scalability with increasing parameter count. To overcome these inherent limitations of autoencoder-based approaches, we propose Residual-Diffusion Joint Source-Channel Coding (RD-JSCC), a novel framework that integrates a lightweight autoencoder with a residual diffusion module to iteratively refine CSI reconstruction. Our flexible decoding strategy balances computational efficiency and performance by dynamically switching between low-complexity autoencoder decoding and sophisticated diffusion-based refinement based on channel conditions. Comprehensive simulations demonstrate that RD-JSCC significantly outperforms existing autoencoder-based approaches in challenging wireless environments. Furthermore, RD-JSCC offers several practical features, including a low-latency 2-step diffusion during inference, support for multiple compression rates with a single model, robustness to fixed-bit quantization, and adaptability to imperfect channel estimation.

Figures (11)

• Figure 1: Low-complexity CSI encoder.
• Figure 2: Low-complexity CSI decoder.
• Figure 3: A lightweight autoencoder compresses the MIMO CSI matrix into a low-dimensional latent representation, which is transmitted over a noisy channel. Decoding occurs in two stages: (1) a low-complexity decoder produces a coarse reconstruction, and (2) a diffusion-based denoising model refines the output for enhanced quality.
• Figure 4: RD-JSCC achieves an order-of-magnitude improvement in performance over state-of-the-art deep JSCC baselines on the COST2100 outdoor dataset for feedback bandwidth of $k=16$.
• Figure 5: Uncoded BLER performance comparison between U-Net based JSCC and RD-JSCC approaches. The diffusion-based method shows continued improvement at high SNR, achieving BLER $\approx 10^{-6}$ at 5 dB while the U-Net approach saturates around $10^{-4}$.

Theorems & Definitions (3)

• Remark 1
• Remark 2
• Remark 3