Multipath-Aware 3D Dynamic Radio Map Construction via Gaussian Splatting

Abstract

Accurate radio map (RM) construction is essential for wireless network optimization and coverage analysis. However, existing approaches typically rely on quasi-static assumptions, failing to handle the inherent uncertainty of dynamic environments, which leads to performance degradation. In this paper, we propose a novel framework termed the multipath-aware dynamic radio map (MAD-RM). Specifically, MAD-RM formulates dynamic RM construction as a dynamic wireless radiance field learning problem, mitigating the uncertainty of environment dynamics by jointly exploiting spatial coordinates and instantaneous multipath information. To this end, a spatiotemporal anchor encoder (STAE) is developed to efficiently capture such spatio-temporal context through attention mechanisms. Subsequently, we render the dynamic wireless radiance field by leveraging Gaussian splatting techniques, featuring a novel dynamic modulation mechanism. In addition, a coverage-aware (CA) loss is proposed to improve Received Signal Strength (RSS) prediction accuracy in poor-coverage regions, ensuring overall reliability. Simulation results demonstrate that MAD-RM consistently outperforms representative baselines in both point-wise RSS prediction accuracy and RM construction quality.