Research On Intelligent Propagation Model Of Wireless Channel Based On Deep Learning

In wireless communication system,wireless propagation model is a fundamental task of network planning and system optimization.Empirical propagation model is simple,but the formulas are set up for specific systems,need a lot of computation to adjust parameters.ray-tracing method is often very time-consuming.Intelligent communication model based on the measured data driven is considered as a solution in meeting the commonality and accuracy demand of propagation model modeling.Based on the above requirements,the paper main research work has been done as follows:1.In this paper,an intelligent propagation model framework based on feedforward neural network is proposed,which takes engineering parameters,map data as inputs features to classify and predict the coverage strength of RSRP in corresponding areas.In this framework,the input features are pre-mapped to obtain high-order vectors and then sent to the feedforward neural network for prediction.in which the high-order premapping algorithm including two parts:clutter index embedded network and feature generation network based on LightGBM.2.The clutter index embedded network maps the clutter index in the map parameters to get the dense coding vector.Revealing the continuity implied by clutter index as the category feature to help the subsequent model to fit the data feature.The solution improves the accuracy of intelligent model to 0.875 and provides model efficiency at the same time.3.Based on lightgbm environment feature generation network:in this paper,based on the previous research,nine artificial design features with good performance are selected.In addition,we proposes the automatic feature generator based on LightGBM.After training lightgbm with measured environmental data,high-order features are obtained.The experiment shows that the high-order features generated by LightgBM are well correlated with the RSRP coverage targets to be classified,and the accuracy of the model is improved.the framework of intelligent propagation model based on feedforward neural network feeds the input features(clutter index is encoded by embedded vector),artificial feature combination and generated high-order features into the feedforward neural network to fit the propagation model.Compared with several machine learning and deep learning algorithms,the experimental results show that the intelligent propagation model based on high-order feature pre-mapping proposed in this paper performs relatively well in the classification task of RSRP coverage strength and weak,with an accuracy of 0.90.The work done in this paper provides concrete new technology for the study of intelligent propagation model,and the analysis and modeling of wireless channel characteristic data.