Reinforcement Learning Algorithm Implementation for Wideband Beam Tracking

Our study develops an innovative beam tracking method that merges subspace tracking with reinforcement learning. The study solves wideband beam alignment problems at different frequencies by decomposing subspaces and training a dynamic Reinforcement Learning model. Our system improves signal quality and performance through beamforming and feed-back control with dimension reduction technology. Through testing our system shows detection accuracy at 95% with 30dB SNR and decreased computing requirements versus traditional processing methods. Our system development will lead to major performance improvements across dynamic wideband communication networks.