Shadow fading-aided path loss modeling in street canyons at 18 GHz

Accurate propagation modeling is essential for network planning and performance evaluation. This paper presents a measurement-based analysis of large-scale path loss (PL) and shadow fading (SF) at 18 GHz in a line-of-sight (LoS) street-canyon scenario, using three receiver heights. Several PL models are evaluated, and a height-dependent close-in model (CI-H) is identified as a suitable baseline, achieving a global root-mean-square error (RMSE) of 2.48 dB while preserving physical interpretability. Based on CI-H, SF is extracted and statistically characterized, exhibiting near-zero mean and a global standard deviation of 2.47 dB, with increasing variability at larger distances. Using the spatial structure of SF, spatial interpolation and machine-learning (ML) models are tested under both dense-route interpolation and spatially blocked validation. Under random grouped interpolation, Ordinary Kriging (OK) and Gaussian Process Regression (GPR) reduce the PL RMSE to 1.35 dB and 1.21 dB, respectively. When continuous 5–20 m route segments are excluded, GPR yields 2.21–2.38 dB, improving upon the corresponding 2.51–2.53 dB CI-H values by 0.15–0.32 dB. Under nested spatial cross-validation, GPR yields 2.32 dB, 0.21 dB below the 2.53 dB CI-H baseline. Thus, the large gains obtained with nearby measurements decrease substantially as spatial separation increases, although GPR retains a modest and consistent advantage on unseen route segments.