Presentation Information
[B-1C-22]Performance Evaluation of Deep Unfolding-Based MIMO Precoding in Ray-Tracing Indoor Propagation Environments
◎Shumpei Tabuchi1, Kazuma Tomimoto1, Toshiki Hozen1, Shujiro Hatada1, Ryo Yamaguchi1 (1. SoftBank Corp.)
Keywords:
MU-MIMO,Deep Unfolding
AI/ML-based optimization of wireless systems is expected to play an important role in sixth-generation (6G) mobile communications. In multi-user MIMO downlink systems, spectral efficiency decreases as the number of users increases due to inter-user interference. The weighted minimum mean-square error (WMMSE) method achieves high precoding performance, but its iterative matrix-inversion operations cause high computational complexity. This paper evaluates a Deep Unfolding-based MU-MIMO precoding method based on IAIDNN, which unfolds WMMSE iterations into a neural network. To assess its effectiveness under more realistic propagation conditions, indoor ray-tracing channels are generated using NVIDIA Sionna. A 50 m × 50 m × 8 m indoor environment with PEC and wood wall materials is considered, and the spectral efficiency is compared with WMMSE, zero-forcing (ZF), and block diagonalization (BD) methods. The results show that, in the wood environment, ZF and BD suffer from performance degradation as the number of users increases, whereas the Deep Unfolding-based method outperforms them. These results indicate that Deep Unfolding-based precoding can potentially suppress performance degradation under realistic indoor propagation conditions.