A Diffusion-Model Based Approach for Low-Overhead and Accurate Channel Estimation in LEO-MIMO Systems

This letter investigates the synergy of diffusion models (DMs) and correlation-sparsity features for channel estimation in low earth orbit (LEO) satellite multiple-input multiple-output (MIMO) systems. We exploit the received signal as the guidance of DM to learn the structured features of LEO satellite channels. By extracting the high spatial correlation in the space-frequency domain and sparsity in the angle-delay domain, the DM approach exhibits potential in enabling accurate channel estimation and pilot overhead reduction. Simulation results show that the proposed approach achieves approximately 36.93% higher estimation accuracy than three advanced deep learning-based estimators and 79.39% greater pilot overhead reduction than the compressed sensing-based approach. In addition, model robustness to scenario dynamics, orbital dynamics, and frequency offset is analyzed.