Sampling-Free Diffusion Transformers for Low-Complexity MIMO Channel Estimation

TL;DR

This work tackles the high computational burden of diffusion-based MIMO channel estimators caused by iterative reverse sampling. It proposes SF-DiT-CE, a sampling-free diffusion transformer trained under a VE forward process to directly predict clean channels from a perturbed, LS-initialized observation, operating on angular-domain representations. Key contributions include a lightweight DiT architecture, VE-consistent training with X-prediction, and a single forward-pass inference that markedly reduces latency while achieving superior NMSE and robustness to distribution shifts. The approach enables real-time, diffusion-prior-based channel estimation in high-dimensional MIMO systems with competitive accuracy and substantially lower computational cost.

Abstract

Diffusion model-based channel estimators have shown impressive performance but suffer from high computational complexity because they rely on iterative reverse sampling. This paper proposes a sampling-free diffusion transformer (DiT) for low-complexity MIMO channel estimation, termed SF-DiT-CE. Exploiting angular-domain sparsity of MIMO channels, we train a lightweight DiT to directly predict the clean channels from their perturbed observations and noise levels. At inference, the least square (LS) estimate and estimation noise condition the DiT to recover the channel in a single forward pass, eliminating iterative sampling. Numerical results demonstrate that our method achieves superior estimation accuracy and robustness with significantly lower complexity than state-of-the-art baselines.

Figures (4)

• Figure 1: The architecture of the adopted DiT model.
• Figure 2: Comparison of different channel estimators on CDL-C dataset.
• Figure 3: Comparison of diffusion-based channel estimators on CDL-D dataset.
• Figure 4: Impact of (a) prediction objective and (b) loss function on NMSE of the proposed approach.

Theorems & Definitions (1)

• Remark 1