Beam Management Solution Using Q-Learning Framework

The beam management is a procedure that properly selects the beams to overcome strong path loss attenuation. This procedure is specially important in 5G-NR (new radio) deployments that operate in millimeter waves frequencies. We propose a novel beam tracking solution that is based on a reinforcement learning framework. More specifically, using the Q-learning algorithm, the user equipments learn over time the best set of beams to maximize their own signal-to-noise to interference ratio. Our framework takes into account reference signals in the 5G technical specification and propose a measurement protocol to implement the Q-learning. The proposed method shows better spectral efficiency than the beam sweeping technique for the multi-user MIMO case.