Machine-Learning-Assisted Multi-Dimensional Multipath Parameter Estimation for Ultra-Wideband Large-Scale Arrays

Artikel i vetenskaplig tidskrift, 2026

Multi-dimensional multipath parameter estimation is essential for characterizing wireless channels. As the system bandwidth and array size increase for 6G communication and sensing systems, the assumption of independence among different channel parameter dimensions no longer holds, which inevitably necessitates multi-dimensional joint parameter search. Moreover, the improved resolution in delay and spatial domains due to ultra-wideband system bandwidth and large array aperture will further increase the computational burden. In this letter, we propose a machine-learning-assisted maximum likelihood estimation (MLE) approach to tackle this critical problem. The proposed method avoids exhaustive dense-grid evaluation in the state-of-the-art MLE algorithms, significantly reducing the complexity and computational burden. Simulation results demonstrate the effectiveness of the proposed method, and an indoor ultra-wideband large-scale channel measurement campaign is further conducted to validate its practical performance.