Semantic Segmentation of Multipath Fading Channel-Based Regional Map

Wireless communication technology is evolving rapidly, where multiple-input–multiple-output (MIMO) technology plays a crucial role by effectively leveraging the diversity of spatial multipath channels. Most MIMO algorithms are designed with the simple but effective spatial correlation assumption, which assumes homological multipath characteristics for all elements of the antenna array. However, this ideal assumption may not always hold, which can be broken by the heterogeneity in multipath effects across regions. Thus, identifying and categorizing these heterogeneous regions is essential for both optimization and deployment in next-generation wireless communication systems. In this letter, we treat the heterogeneous as semantic in multipath fading domain, and propose to segment the regional map into different partitions. In detail, this letter introduces a multistacked U-shaped network (U-net) model, designed for effective channel segmentation. The model is trained on datasets generated through ray tracing (RT) methods across diverse scenarios. Extensive experiments demonstrate that the proposed data-driven model achieves a segmentation accuracy of 78.931%, effectively identifying complex multipath regions, while operating several thousand times faster than RT methods.