Robust MU-MIMO Communication via Learning-Enhanced Receiver and Site-Specific Channel Modeling

Renwei Ou, Ruisi Li

Keywords: Deep leaning, MU-MIMO Communication

Abstract: As the evolution toward next-generation wireless networks gains momentum, there is a growing need for intelligent and environment-aware communication systems. This paper presents an environment-aware neural receiver for multi-user multiple-input multiple-output (MU-MIMO) uplink detection. By integrating site-specific channel modelling with a prior-enhanced deep learning architecture, our framework achieves robust signal detection under realistic propagation conditions while maintaining compatibility with existing wireless infrastructure. The proposed method incorporates linear minimum mean square error (LMMSE) derived log-likelihood ratio (LLR) priors to stabilize training convergence and employs a lightweight MobileNet-based 3D convolution (MBConv3D) network to refine channel estimates and optimize LLR outputs. Extensive simulations in urban scenarios demonstrate that our approach operates effectively in low-to-medium SNR ranges, closely approaching the performance bound of perfect channel state information (CSI). The results validate the framework's strong potential for practical 5G/6G deployments, offering a viable pathway toward intelligent receiver design for next-generation wireless systems.

Submission Number: 22