AI/ML-Driven Channel Estimation and Beamforming for 6G Massive MIMO Systems: A Performance Analysis

Sixth-generation (6G) massive MIMO systems operating over sparse, fast-varying mmWave and terahertz channels present new challenges to pilot-assisted channel estimation and tracking, with standard Least Squares (LS), Minimum Mean Square Error (MMSE), and Orthogonal Matching Pursuit (OMP) approaches yielding either prohibitively high pilot overhead or inaccurate channel state information (CSI). This article proposes an innovative hybrid approach that leverages deep neural networks and sparse Bayesian learning to achieve robust estimation, tracking, and precoding of massive MIMO channels, with the performance evaluated under eight widely used criteria and compared to LS, MMSE, OMP, and conventional Sparse Bayesian Learning (SBL) methods. Simulation results demonstrate that the proposed approach provides up to 29x better bit-error-rate (BER) than LS at 10 dB SNR, 18.2 dB NMSE gain, 80% higher spectral efficiency than zero-forcing precoding while maintaining channel tracking accuracy at 300 km/h and reducing pilot overhead to 5-8% of the frame. Additionally, the approach enables the use of 256-QAM and 10+ Tbps data rates at terahertz bands for short-range 6G applications.