Data-Driven Radio Map Caching for Beam Management Under Positioning Uncertainty

Abstract

Millimeter-wave (mmWave) beam management in roadside unit (RSU) networks requires exhaustive beam sweeping over large codebooks, incurring high overhead. This letter proposes a data-driven radio map caching framework where the RSU trains decoupled transmit (TX) and receive (RX) beam maps from sparse channel measurements and uploads them once to the user equipment (UE). The UE queries both maps locally using its GPS position and blockage state to retrieve top- $K$ candidate transmit and receive beams and selects the final pair from a short on-air pilot sweep. This removes the recurring per-beam measurement-and-report loop of conventional beam management: no real-time channel state information (CSI) feedback, no per-beam received-power reporting, and only two lightweight beam-index messages per alignment. Evaluations show that caching a small number of beam candidates jointly resolves both impairments, recovering near-optimal beamforming gain under severe positioning error while remaining robust to training sparsity down to a few hundred measurements, on standard hardware with no positioning infrastructure beyond GPS.