Diffusion Particle Filter For MIMO Channel Estimation

Channel estimation is essential for unlocking the full potential of multiple-input multiple-output (MIMO) systems. Traditional pilot-based techniques often exhibit suboptimal performance when the channel prior deviates from the Gaussian assumption. Recently, diffusion models have emerged as a powerful generative framework, which are capable of learning unknown statistical patterns in wireless networks. However, the existing diffusion-based channel estimation methods either rely on imperfect approximations or suffer from high computational cost, limiting their real-time applicability. In this paper, we integrate the diffusion model into the particle filter, and propose a novel MIMO channel estimation method. Specifically, our method leverages the learned probabilistic structure of the diffusion model to guide particle updates, ensuring both the estimation accuracy and computing efficiency. The simulation results demonstrate that the proposed method achieves superior performance and has favourable adaptability across a range of channel conditions.