Localized channel modeling is paramount for 5G cellular network optimization. However, the existing channel models either apply to general scenarios and can not well match the localized geographical structures of physical environment, or have high computational complexity. In this paper, we propose a localized statistical channel modeling (LSCM) based on the radio map, which is aware of the targeted propagation environment. Instead of using channel impulse response, LSCM solely relies on the reference signal receiving power (RSRP), which can be expressed as a linear model of the channel’s angular power spectrum (APS). Based on this, we formulate the task of channel modeling as a sparse recovery problem where the non-zero entries of the APS indicate the channel paths’ power and angles of departure. In order to achieve good channel modeling performance, we propose a multi-grid-based APS estimation scheme which can make our LSCM more accurate. To exploit the similarities of the channel paths from adjacent grids, construct the radio map by interpolating the RSRP, and cluster the grids in beamspace. A novel regularization term that constrains both sparsity and similarity is also proposed to address the multigrid-based LSCM problem. At last, extensive simulations based on both synthetic and real RSRP measurements are presented to verify the performance of the proposed channel model.

Channel modeling;Network optimization;Radio map;RSRP;Sparse recovery