ML-based Multiple BS Beam Tracking and Positioning with Digital Twin

초록

This paper proposes an Aerial Omniverse Digital Twin (AODT)-driven machine learning (ML) beam tracking and positioning method for mmWave systems with multiple base stations (BSs). Conventional beam sweeping methods suffer from either excessive training overhead under full beam sweeping or coarse angle estimation under partial beam sweeping, while many ML approaches rely on simplified channel models or limited datasets. Thus, the proposed method is designed to improve positioning accuracy in digital twin (DT) mmWave environments with multiple BSs by refining the angle of arrival (AoA) through residual learning using AODT-generated channel data and accurate ground-truth positions. By utilizing codebook-based partial beam sweeping and residual learning, the proposed method estimates the residual between the selected beam-center angle and the true AoA, enables continuous angle tracking through AoA refinement, and performs least-squares positioning utilizing refined AoA estimates from multiple BSs. Case studies show that, by learning the residual between the coarse beam-center angle and the true angle of arrival (AoA), the proposed method obtains refined AoA estimates that lead to more accurate positioning, thereby confirming its effectiveness in DT mmWave systems with multiple BSs.

키워드