Machine Learning Assisted Optimization And Antenna Design Technology

With the development of wireless communication technology,there are more requirements for antenna designs,for example miniaturization,broadband and multi-frequency.Antenna and arrays are becoming more and more complex,thus leading to a wider set of degrees of design freedom.Traditional full-wave electromagnetic(EM)simulation can be very accurate but also very time-consuming,so it is not suitable for efficient antenna optimization and sensitivity analysis,which need a large number of repeated simulations.Recently,machine-learning-assisted optimization(MLAO)has attracted more and more attention and has been applied to antenna designs.Firstly,the thesis compares the prediction performance of the artificial neural network(ANN),support vector machine(SVM)and gaussian process regression(GPR),which are common surrogate models ued in EM field.It is concluded GPR is the best to approximate antenna models with high dimension,small sample set and many locally optimal solutions.Then,some optimization techniques based on GPR and the framework of MLAO applied to antenna designs are introduced,followed by the generality study,stress test and robustness test of this optimization method.Secondly,based on the research of MLAO,a broadband reflectarray antenna and a dual-band antenna for wireless local area network(WLAN)are designed and optimized.The reflectarray antenna works at Q-band.The optimal unit cell is found after 36 iterations(full-wave EM simulations)with 4 design parametrs and 15 samples.The simulation and test results show that the reflectarray antenna has a wide gain bandwidth and good radiation performance.In the optimization process of the miniaturized dual-band antenna,there are13 design parameters and 30 samples.The optimal design is found after 79 iterations.The size of the dual-band antenna is only 20×15 mm~2,and it can effectively cover the 2.4/5 GHz band.The two cases show that MLAO can help to improve the antenna design efficiency.Finally,an automatic optimization software is designed to assist the antenna design process using machine learning method.The software is developed based on the function of MATLAB graphical user interface(GUI),and supports the call of full-wave EM simulation software HFSS and CST for joint simulation optimization,parametric modeling is carried out according to the existing model.In addition to optimization,the software offers the real-time monitoring of the design parameters and targets convergence trend.The users can suspend and adjust the parameter range during optimization.The automatic optimization software further improves the antenna optimization efficiency and code reuse rate.