Generative Diffusion Model-based Compression of MIMO CSI

TL;DR

The paper tackles the challenge of compressing CSI for MIMO systems in scenarios requiring channel prediction and side information. It proposes a fixed-rate encoding with a trainable codebook and a conditional diffusion-based decoder that leverages side information (UL CSI) to reconstruct future DL CSI. The approach is trained with a joint objective that couples codebook alignment and reconstruction likelihood, and is implemented with a CNN-style encoder/decoder and a U-Net diffusion backbone. Across CDL-C Sionna simulations and COST2100 data, the method achieves substantial rate-distortion gains, including up to more than a factor of two improvements at about half the rate, underscoring diffusion-based CSI compression as a viable path for next-generation wireless systems.

Abstract

While neural lossy compression techniques have markedly advanced the efficiency of Channel State Information (CSI) compression and reconstruction for feedback in MIMO communications, efficient algorithms for more challenging and practical tasks-such as CSI compression for future channel prediction and reconstruction with relevant side information-remain underexplored, often resulting in suboptimal performance when existing methods are extended to these scenarios. To that end, we propose a novel framework for compression with side information, featuring an encoding process with fixed-rate compression using a trainable codebook for codeword quantization, and a decoding procedure modeled as a backward diffusion process conditioned on both the codeword and the side information. Experimental results show that our method significantly outperforms existing CSI compression algorithms, often yielding over twofold performance improvement by achieving comparable distortion at less than half the data rate of competing methods in certain scenarios. These findings underscore the potential of diffusion-based compression for practical deployment in communication systems.

Figures (5)

• Figure 1: System Model (Coding for computing with side information)
• Figure 2: Proposed compression framework.
• Figure 3: Magnitude visualization of CSI samples: The task is to predict ${\mathbf{Z}}$ (future DL CSI) from compressed ${\mathbf{X}}$ (current DL CSI), leveraging ${\mathbf{Y}}$ (UL CSI) as side information.
• Figure 4: Rate-distortion curves for experiments in Sec. \ref{['subsection_csi_prediction']}.
• Figure 5: Rate-distortion curves for experiments in Sec. \ref{['subsection_csi_compression']}.