A 3D High-Resolution Joint Location and Beamforming Prediction Model for IRS-Aided Wireless Networks

Abstract

Fifth generation and beyond (5GB) technology requires low latency, high capacity, and constant connectivity for safety and reliable service. Multiple-input multiple-output (MIMO) and millimeter wave (mmWave) technologies can help meet these needs. However, MIMO can cause extra overhead due to massive channel feedback, and mmWave signals weaken over short distances, leading to limited coverage. Intelligent reflecting surfaces (IRSs) and highly directive active beamforming are recommended to address coverage and overhead issues. Most IRS research focuses on optimizing phase shifts in two dimensions. This paper introduces a three-dimensional model to jointly evaluate user location and IRS phase shift optimization. Additionally, phase constants are derived from optimal phase shifts to limit training overhead. A random forest learning algorithm is proposed, using optimal phase constants and codebook indices to train for each estimated location. Data transmission utilizes the Doppler effect to predict the possible locations of a user. In this way, the trained model can perform high-resolution joint beamforming for the current and future locations of the user. Simulation results show that the model accurately predicts phase shifts without needing channel state information while keeping complexity and training overhead low.