The Application Of Machine Learning Algorithm In Antenna Design

5G comunacation brings new opportunities and challenges to antenna design.The urban or indoor communication antenna needs to have functions like large radiation coverage range,compact size,low profile and high gain,etc.In addition,because the complex signal transmission scene of 5G communication,the antenna also needs to have diversity,multi-polarization and circular polarization functions.However,traditional antenna design methods need a lot of simulation time and will cost a lot of computational storage.Meanwhile,it is difficult for traditional methods to deal with multi-objective global optimization problems.To solve this problem,this thesis uses machine learning algorithms to help optimize antennas.This thesis studies 5G indoor antenna and machine learning-assisted antenna optimization algorithm,and the specific results are as follows:Firstly,two kinds of 5G indoor antennas are proposed,including the wideband pattern reconfigurable loop antenna and the wideband circularly polarized antenna.The pattern reconfigurable antenna can provide stable conical and broadside beams.The circularly polarized antenna can generate the circularly polarized beams in a broad band.In the design of the pattern reconfigurable antenna,the characteristic mode theory is used to slot the oatch and introduce reactance loading into the patch structure of the antenna.As the reslut,the radiation characteristics of the antenna are successfully improved.Then,an air gap and a X-shaped isolation structure are introduced into the antenna structure to improve the impedance matching characteristics of the antenna in the wide band.Finally,a feeding network which can change the current distribution on the patch surface is designed.After processing and testing,the antenna proposed in this thesis has wide-band,small-size and beam reconfigurable functions,which is very suitable for 5G indoor communication scenarios.In the design of broadband circular polarization antenna,the double ring patch loaded with cross finger capacitor is used as the main radiator,which reduces the antenna size and improves the impedance matching characteristic of the antenna.The circular polarization of the antenna is realized in a wide frequency range and the symmetry of the radiation pattern is improved.Moreover,impedance matching characteristics can be improved over a wider bandwidth.The final machining test results show that the antenna has a compact structure,can achieve a wide working bandwidth and high gain,and is a 5G indoor antenna with good performance.Then,in order to solve the problem that the optimization process takes too much time and energy in the design of the above two antennas,and it is difficult to select the global optimal result when there are too many parameters and optimization indexes.In this thesis,a multi-objective optimization algorithm for multi-input parameters is implemented based on k-nearest neighbor algorithm.In this thesis,two methods are proposed to realize multi-objective optimization algorithm,one is to construct objective function based on penalty function,the other is to construct weighted synthesis objective function.After verification by two examples,this thesis proposes two methods for their respective application scenarios,and the two antennas proposed in this thesis are optimized.In this thesis,the optimization algorithm only need small data sets can produce available agent model,and effectively improve the accuracy of the model prediction through active learning,compared with the traditional method of optimization results.Using the proposed optimization algorithm can save much time,and have better antenna matching feature,gain and axial ratio bandwidth indicators,etc.