Design And Optimization Of Antenna Array And Metasurface Reflectarray Based On Neural Network

In recent years,the words such as artificial intelligence and machine learning appear in our life more and more frequently,which are related to our life closely.This technology can be seen in various fields in our life including smart speaker,intelligent voice assistant and face recognition system etc.After the slow development of traditional machine learning,with the proposed deep learning and the arrival of the era of big data and the constant update of hardware equipment,it has seen rapid process.The arrival of 5G technology also makes the automatic driving technology further develop towards an ideal application direction.Neural network(NN)is one of the important machine learning algorithms,whose model imitates the way of biological neural network to process information.It is widely used in medical information detection,image recognition and other fields because of its high accuracy and not easy to overfit.The design and optimization of antenna array and metasurface is a time-consuming process.More importantly,it's difficult for not only beginners who are lack of specialized knowledge,but also those who have a certain theoretical basis or design experience,because the design and optimization of antenna array and metasurface is tightly related to the design environment.In addition,the optimization process depends on not only the designer's specialized knowledge and experiences,but also the performance of computers and the complexity of array itself.To overcome the above problems,in this thesis the NN-based methods have been proposed to design and optimize antenna array and metasurface reflectarray.The main contributions are as follow: Firstly,the antenna array design and optimization based on neural network has been developed.The linear antenna array and the planar antenna array are designed and optimized respectively.Artificial neural network(ANN)and convex optimization(CVX)are applied in the design,in which the training accuracy and testing accuracy of the proposed ANN model are more than 90%.By comparing the design results with the full wave simulation results,the good agreement between two results shows the accuracy of the proposed method.With the proposed method,a 1-by-11 linear array and a 7-by-7 planar array with microstrip antenna as elementary antenna are designed.In addition,a 1-by-41 linear array is also designed to show the high efficiency of the proposed method.All the designed arrays have achieved the design goal,including the desirable main lobe direction and the side lobe level lower than 15 d B compared with the main lobe level.The comparisons of the time cost consumed by the proposed method and the full wave simulation in a single run in all examples demonstrate high efficiency of the ANN-based method in the antenna array design and optimization.Compared with the traditional method,the proposed method is not only efficient,but also takes the coupling effect of the array elements into account,so that the design results are more agree with the actual design.Secondly,the design and optimization of metasurface reflectarray based on NN has been developed.In this method,the NN is used to optimally design the topology unit structure.In the NN model,the key parameters of the unit structure are taken as the input variables,and the amplitude and phase characteristics of the unit structure are used as the output variables.With the proposed NN model,the time consuming of full wave simulation process is neglected and the performance of the unit cell can be rapidly obtained.The training accuracy and testing accuracy of the proposed NN model are more than 90%.With the proposed unit cell,the metasurface-based reflectarray for the generation of orbital angular momentum(OAM)vortex waves with different mode numbers has been designed according to the OAM theory.Simulation results show that the metasurface reflectarrays designed by the proposed method can radiate the wave carrying the OAM in a wide band.The proposed method can quickly calculate the geometric dimensions of the metasurface according to the phase requirements,it's convenient,the coupling effect between the units is small,and no repeated optimization design process.