Research On Machine Learning For Antenna Array Optimization Design

As a component of the wireless communication system,the antenna bears the important responsibility of sending and receiving electromagnetic waves.Nowadays,with the continuous improvement of communication requirements,the design of antenna arrays is becoming more and more complicated.For complex antenna arrays,computing resources such as computing time and memory consumption for optimal design also increase.This thesis aims to take advantage of machine learning's ability to fit non-linear functions while being less time-consuming,and to combine antenna array optimization design with machine learning to reduce the cost of traditional antenna optimization design.The main research of this thesis is to apply machine learning to the optimization design of antenna array,which is mainly divided into two aspects: the first is an algorithm that combines antenna array analysis and machine learning.This thesis implements antenna array analysis by designing a Deep Neural Network(DNN)structure.The trained DNN can predict the radiation field of the antenna array in real time.The predicted results are in good agreement with the theoretical results.Compared with traditional algorithms,the method used in this thesis has a significant increase in computing speed.The second is an algorithm combining Dolph-Chebyshev synthesis and machine learning.This thesis implements the Dolph-Chebyshev synthesis method by designing a DNN.The trained DNN can predict the current distribution that meets the design requirements.The results are in good agreement with the theoretical results.At the end of this thesis,the problem of initialization of hyperparameters in the training process is studied,and the optimization ideas for the initialization of hyperparameters of the DNN designed in this thesis are given.The specific work of this thesis is as follows:Firstly,this thesis introduces the basic theory of antenna array analysis and synthesis,introduces the development of machine learning and its advantages,and explains in detail the current domestic and foreign research on the combination of antenna array optimization and machine learning,provides a theory for the research work of this thesis.Next,for the antenna array analysis,this thesis designs the DNN combined with linear array analysis and planar array analysis,and analyzes the training process,training results and applicability.The simulation shows that the trained DNN can realize the antenna array analysis,and the accuracy of the prediction result is above 90%.Compared with the traditional algorithm,the calculation time of the algorithm in this thesis is significantly reduced,which effectively demonstrates the feasibility of combining DNN and antenna array analysis.Then,for the Dolph-Chebyshev synthesis,this thesis designs the DNN combined with Chebyshev linear array synthesis and Chebyshev planar array synthesis.The training process,training results and applicability are analyzed,and a good performance DNN is obtained.The simulation shows that the trained DNN can realize the Dolph-Chebyshev synthesis method,and the accuracy of the prediction results is above 90%,which effectively demonstrates the feasibility of combining the DNN designed in this thesis with the DolphChebyshev synthesis.The learning rate,the number of hidden layer neurons,the number of hidden layer layers and the number of iterations is the four main hyperparameters in the training process.Their initialization has a vital impact on the performance of the DNN.In order to further improve the performance of DNN,this thesisairtiu then trains a more complex Chebyshev linear array synthesis model,gives the optimization ideas for the initialization of four hyperparameters,and conducts simulation calculations.A good training effect is obtained,and the performance of the original DNN is effectively improved.Finally,this thesis summarizes the research content,illustrates the feasibility of combining antenna array optimization design with machine learning,and provides a reference for antenna array optimization design.