Millimeter wave systems rely on narrow beams (beamforming) and dense cell deployments for reliable communication. Tracking these beams from multiple cells can increase power consumption and signaling overhead. Therefore, a mobile needs to selectively and smartly track beams under power/overhead constraints. In this paper, we propose a fully data-driven, long short-term memory (LSTM)-based, selective link tracking approach. These approaches are developed for both fixed and adaptive power/overhead constraints, which also predict the magnitude of the best performing beam. The algorithms are validated in simulations of a 28GHz 5G New Radio (NR)-like system in an urban area with realistic navigation routes utilizing detailed ray-tracing. The simulations demonstrate that the proposed methods outperform classic and deep reinforcement learning (RL) approaches in terms of tracking accuracy, power saving and overhead for both analog and digital beamforming architectures. We also argue that the prediction of the proposed method can be easily performed on a digital signal processor of a modern chipset with minimal resource consumption.
Read full abstract