Deep learning and diffusion models for uplink channel estimation in MU-MIMO systems

Periyakarupan Gurusamy Sivabalan Velmurugan, Soundarapandian Suriya Ramkumar, Sundarrajan Jayaraman Thiruvengadam, Vinoth Babu Kumaravelu,
Deep learning and diffusion models for uplink channel estimation in MU-MIMO systems,
Results in Engineering,
Volume 28,
2025,
107930,
ISSN 2590-1230,
https://doi.org/10.1016/j.rineng.2025.107930.
(https://www.sciencedirect.com/science/article/pii/S2590123025039817)
Abstract: Accurate channel estimation is essential for reliable and high-capacity wireless communications. This paper proposes two learning-based models for uplink channel estimation in multiuser multiple-input multiple-output (MU-MIMO) systems: a deep neural network (DNN) regression framework and a diffusion-based (DB) estimator with transformer denoiser. The DNN, integrated with pilot-symbol-informed pre-processing, achieves up to a 3 dB signal to noise ratio (SNR) gain at a mean square error of 10−2 over least-squares (LS) and minimum mean-square error (MMSE) estimators, while reducing input dimensionality and computational cost. The DB estimator, employing a transformer encoder as a denoiser, effectively captures inter-user and antenna dependencies and accounts for noise uncertainty. With pilot-assisted pre-processing, the DB estimator provides an additional 0.5 dB gain in high-SNR regimes. Simulation results under different MU-MIMO settings, with the number of users, base station antennas, and pilot sequence length changed, demonstrate that both approaches surpass conventional channel estimators in terms of accuracy, scalability, and robustness uniformly.
Keywords: Channel estimation; Deep learning; Deep neural network; Diffusion model; Massive MIMO; MU-MIMO