Multi-Modal Neural Radio Radiance Field for Localized Statistical Channel Modelling

This paper presents MM-LSCM, a self-supervised multi-modal neural radio radiance field framework for localized statistical channel modeling (LSCM) for next-generation network optimization. Traditional LSCM methods rely solely on RSRP data, limiting their ability to model environmental structures that affect signal propagation. To address this, we propose a dual-branch neural architecture that integrates RSRP data and LiDAR point cloud information, enhancing spatial awareness and predictive accuracy. MM-LSCM leverages volume-rendering-based multi-modal synthesis to align radio propagation with environmental obstacles and employs a self-supervised training approach, eliminating the need for costly labeled data. Experimental results demonstrate that MM-LSCM significantly outperforms conventional methods in channel reconstruction accuracy and robustness to noise, making it a promising solution for real-world wireless network optimization.