Environment-Aware Satellite-Terrestrial Channel Knowledge Map

Abstract

Within the channel knowledge map (CKM) paradigm for environment-aware channel prediction, this paper constructs a satellite-terrestrial channel gain map (ST-CGM) for low-Earth orbit (LEO) satellites operating over mountainous regions, explicitly accounting for terrain-induced signal shadowing. However, building an ST-CGM is challenging because channel data are often limited to sparse samples of satellite transmitter positions and terrestrial user locations. Such sparsity can impair model generalization and reduce prediction accuracy due to insufficient spatial characterization of the propagation environment. To address this, we propose a deep neural network (DNN)-based method that integrates ray tracing data to construct the ST-CGM. The proposed approach supports offline model training and efficient online channel-gain prediction. Specifically, ray-tracing data are used to capture the impact of complex terrain on channel gain across different transmitter-receiver configurations, while the DNN learns the resulting nonlinear spatial propagation patterns. By leveraging synthetic ray-tracing samples, the proposed method improves the spatial coverage of training data and enables the DNN to generalize from sparse observations to unmeasured locations. Simulation results show that the proposed method significantly improves channel prediction accuracy for specific user locations compared to the 3GPP TR 38.811 non-terrestrial network (NTN) model and other measurement-based benchmarks, especially in non-line-of-sight scenarios.