Radio Environment Map Prediction Method Based on Improved Environment Coding
Deep learning has the advantages of faster prediction speed and higher accuracy than the traditional method of predicting Radio Environment Map(REM).However,in order to ensure the accuracy of prediction,it's required to design complex model network for training,which leads to a lot of training time.In order to reduce the training time of the model and realize the rapid construction of REM,a method with an improved deep learning model combined with environment coding is proposed for REM construction.In the deep learning network structure,the Mobile Vision Transformer(MobileViT)module is used to replace the convolution module of the traditional model,which increases the global view of the model.In the input data pre-processing,the radio wave propagation mechanism is introduced to improve the interpretability of the model and the one-dimensional entropy of the image.The environmental coding is carried out by combining the path loss calculated by the empirical formula and the antenna position map,and then the resulting encoding map is used together with the urban environment map as the common input.The simulation results show that the improved model has a faster convergence rate in the training stage.The proposed data preprocessing method can accelerate the model training.
radio propagationREMpath loss predictiondeep learningimage generation
万方数据
维普
