Research On Multipath Interference Suppression Of Microwave Signal Based On Machine Learning And Neural Network

With the development of global informationization,information transmission efficiency is becoming increasingly important.This paper focuses on multipath interference in the microwave anechoic chamber,and proposes machine learning and neural network to carry out corresponding interference suppression research,so as to improve the measurement accuracy of the anechoic chamber.The main idea of this paper is to extract radiation data through simulation,apply machine learning and neural network to multipath interference suppression,and train interference suppression model based on simulation data.Theoretically,the mechanism of interference suppression is analyzed.It is also proved in simulation experiments that the proposed method has good suppression accuracy.In this paper,the microwave anechoic chamber is modeled,and the relevant suppression methods are researched for the multipath interference.The specific research work is summarized as follows:1.Model the microwave anechoic chamber communication environment.First,the anechoic chamber is modeled to simulate communication in an interference environment and a non-interference environment.In order to simulate the influence of different interference sources on interference,multi-group simulation is carried out for different materials,shapes and amounts of interference sources.The materials are PEC,aluminum and copper,and the shapes are plane,concave,convex and wave surface.The difference in materials and shapes makes the interference more diverse,and the amount of interference sources increases can significantly enhance the overall interference.10 sampling points are taken under each interference source,and the signal with multipath and its corresponding signal without multipath are extracted for subsequent model training.2.Proposed suppression method based on machine learning,which uses polynomial regression and random forest.Firstly,according to the theory of polynomial regression,the signal propagation expression is modeled.The signal with multipath is input into the expression,and the signal without multipath is used as the output supervised model to fit the interference factor in the anechoic chamber.Simulation experiments show that polynomial regression can effectively suppress interference,and the suppression accuracy is about 93.88%,but the sensitivity to subtle changes of interference signals is poor.Then using random forest,the suppression model is generated by the same training method.The suppression accuracy is about 95.01%,there is no limitation of modeling expression,and it has good sensitivity to subtle or strongly changes of interference signals.3.Proposed suppression method based on neural network.Using the same simulation data,the learning rate is consistent with the polynomial regression,both 0.01,and the gradient descent method updates the parameters to train the model.The neural network can converge in about 15 iterations,and its suppression accuracy is about 97.11%,while the polynomial regression converges about 40 iterations.It converges faster than polynomial regression,and the suppression accuracy is better than polynomial regression and random forest.It has excellent sensitivity to subtle or strongly changes of interference signals,and also has better stability and generalization.