Low Complexity Beam Selection Scheme for Terahertz Systems: A Machine Learning Approach

Terahertz (THz) communication has been envisioned as a promising candidate to support ultra-broadband for future beyond fifth generation (5G). In the THz communication system, hybrid beamforming is essential for overcoming the serious attenuation due to the extremely high frequency in THz band. However, the current beam selection schemes in hybrid beamforming architectures still suffer the enormous computational complexity. To tackle this issue, we consider the beam selection problem as a multi-class classification model, and then propose a low complexity beam selection method based on random forest classification (RFC), which belongs to machine learning algorithms and sufficiently depends on training samples. Simulation results show that our proposed RFC based beam selection scheme is capable of providing a better tradeoff between sum-rate and complexity by choosing the appropriate parameter settings compared with some existing beam selection schemes.