Adaptive Beam Tracking With the Unscented Kalman Filter for Millimeter Wave Communication

Millimeter wave (mmWave) communication links for 5G cellular technology require high beamforming gain to overcome channel impairments and achieve high throughput. While much work has focused on estimating mmWave channels and designing beamforming schemes, the time dynamic nature of mmWave channels quickly renders estimates stale and increases sounding overhead. We model the underlying time dynamic state space of mmWave channels with a linear Gauss-Markov process and design sounding beamformers suitable for tracking in an unscented Kalman filtering framework. Given an initial channel estimate, simulation results show that unscented filtering efficiently leads to better estimates than the extended Kalman filter with limited sounding and allows forward prediction for higher sustained beamforming gain during data transmission. Moreover, from tracked prior channel estimates, optimal and constrained suboptimal beams are adaptively chosen for low sounding overhead while minimizing estimation error.