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

Diffusion model-based channel estimators have shown impressive performance but suffer from high computational complexity because they rely on iterative reverse sampling. This letter proposes a sampling-free diffusion transformer (DiT)-based channel estimator, termed SF-DiT-CE, for low-complexity MIMO channel estimation. Exploiting angular-domain sparsity of MIMO channels, we train a lightweight DiT to directly predict the true channels from their perturbed observations and noise levels. At inference, we first obtain an initial channel estimate using the least-squares (LS) method, which can be viewed as the true channel corrupted by Gaussian noise. The DiT then takes this estimate and its corresponding noise scale as inputs 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. The code is available at: https://github.com/c-res/SF-DiT-CE