Deep-SOR detection for massive MIMO systems

https://doi.org/10.1016/j.aeue.2025.155815

In massive MIMO systems, particularly in highly loaded scenarios where the number of transmit antennas approaches that of receive antennas, symbol detection faces significant challenges, including increased computational complexity and degraded performance. To address these issues, in the paper we propose a deep learning (DL)-assisted successive over-relaxation (SOR) detector. This detector utilizes two relaxation vectors to enhance performance, which are determined through DL training. Additionally, we introduce a convergence theorem and conduct simulations to validate their determination. Finally, simulation and complexity analysis results demonstrate that the proposed detector achieves superior performance with a moderate computational cost, especially in highly loaded scenarios.

@article{LU2025155815, title = {Deep-SOR detection for massive MIMO systems}, journal = {AEU - International Journal of Electronics and Communications}, pages = {155815}, year = {2025}, issn = {1434-8411}, doi = {https://doi.org/10.1016/j.aeue.2025.155815}, url = {https://www.sciencedirect.com/science/article/pii/S1434841125001566}, author = {Hoang-Yang Lu and S. Pourmohammad Azizi and Shyi-Chyi Cheng}, keywords = {Massive multiple-input multiple-output, Deep learning, Successive over-relaxation}, abstract = {In massive MIMO systems, particularly in highly loaded scenarios where the number of transmit antennas approaches that of receive antennas, symbol detection faces significant challenges, including increased computational complexity and degraded performance. To address these issues, in the paper we propose a deep learning (DL)-assisted successive over-relaxation (SOR) detector. This detector utilizes two relaxation vectors to enhance performance, which are determined through DL training. Additionally, we introduce a convergence theorem and conduct simulations to validate their determination. Finally, simulation and complexity analysis results demonstrate that the proposed detector achieves superior performance with a moderate computational cost, especially in highly loaded scenarios.} }