A Deep-Learning Method for Path Loss Prediction Using Geospatial Information and Path Profiles

Beyond 5G/6G should provide services everywhere, and it is necessary to expand area coverage and develop high-frequency bands from millimeter waves to terahertz waves. Based on these issues, clarifying radio propagation characteristics and modeling techniques is important for the system and area design of beyond 5G/6G, which will utilize various frequencies in any environment. We have developed a site-specific path loss model by extracting features of the propagation environment by machine learning using images of three regions as input data: the transmitting point, the receiving point, and the region between both points. However, image scaling is required in the region between the points to keep the image size constant in accordance with the distance. Therefore, even if the propagation path is the same, the effect on the propagation characteristics caused by shadowing is different. In this article, we propose a method to parameterize the environment on the propagation path in the region between the transmitting and receiving points with a constant size regardless of distance and combine it with images around the points. Since the dominant path that contributes to path loss characteristics depends on the urban structure between transmitting and receiving points, parameterizing the environment on the propagation path should improve the estimation accuracy. We demonstrate the effectiveness of the proposed method through an evaluation using 800-MHz and 2-GHz measured data in urban, suburban, and rural areas.