MIMO Channel Estimation Using Transformer-Based Generative Diffusion Models

doi:10.1109/tvt.2026.3680259

MIMO Channel Estimation Using Transformer-Based Generative Diffusion Models

2026-04-02

0

PRE

AI

DOI：10.1109/tvt.2026.3680259

Original

Original request for help

Save

Abstract

En 中文

Recent researches indicate that artificial intelligence (AI) holds significant potential for channel estimation. However, most existing AI-based channel estimators rely on training channel samples, leading to reduced estimation accuracy on out-of-distribution data. To address this problem, we propose a robust channel estimator for multiple-input multiple-output (MIMO) system called Transformer-based diffusion model (TDM), which is trained under an unsupervised process and independent of channel distribution. We perform the Transformer structure to accurately denoise from noisy channel matrix. A posterior sampling is developed at inference time to channel reconstruction. Experimental results demonstrate the superiority and out-of-distribution robustness of the TDM scheme.

Keywords:

MIMO channel estimation

generative AI

diffusion models

attention mechanism

MIMO Channel Estimation Using Transformer-Based Generative Diffusion Models

Abstract

Journal

Organization