Machine-Learning-Assisted Scenario Classification Using Large-Scale Fading Characteristics and Geographic Information

With the continuous growth of mobile communications, great progress has been made in wireless channel measurements and modeling. The channel characteristics vary significantly in different scenarios, and the scenario factor can be used as an important reference for the channel model. To get an accurate variable to represent the wireless environment, the classification of radio propagation scenarios is investigated. The collection and processing of channel data and geographic information about landforms and buildings are described, and the principle of the K-Nearest Neighbors (KNN) algorithm and its improved method weighted KNN (WKNN) algorithm are discussed in detail. The method proposed in this paper provides a feasible scheme for accurately defining and classifying wireless communication scenarios, only by taking advantage of the large-scale fading characteristics and geographic information system (GIS), which greatly reduces the workload and difficulty of the experiment. The result shows that when classifying scenarios in one province, the classifier works well. When faced with several provinces, the accuracy of KNN is above 90%, while that of WKNN is above 95%. The scenario classification method with high accuracy and low complexity brings convenience to the development of mobile communications.