An Efficient Wireless Propagation Loss Prediction Model Based on 3-D Terrain Features Extracted by Deep Learning

This letter proposes a path loss prediction method based on a convolutional neural network (CNN) by extracting features, such as terrain obstacles and building distribution. Twenty prediction tasks are carried out based on different configurations of the wave propagation mode, the number and height distribution of buildings, and whether there is a blocking effect, and a good accuracy is achieved. Meanwhile, the prediction time cost of the CNN and that of ray tracing are comprehensively compared, and in general, the CNN has stable and higher computing efficiency. It can be seen that the CNN has greater advantages when dealing with complex environments and multiple propagation mechanisms by processing topographic maps attributed to its better feature extraction and generalization capabilities.