Design Of Sub-6GHz Microstrip Antenna Based On Machine Learning

With the continuous development of Internet of Things technology and 5G communication technology,people’s demand for portable smart devices continues to increase,which has led to a large demand for high-performance antennas.In the antenna design process,although the traditional electromagnetic software full-wave simulation can obtain relatively accurate simulation results,it will consume a lot of computing resources and time resources,which is not conducive to a large number of repeated calculation optimization and sensitivity analysis.Machine learning-assisted optimization（MLAO）is now a hot research topic.The basic principle is that the antenna optimization process can be simplified to input various geometric parameters of the antenna and output various electromagnetic responses of the antenna.The analogy is a black box whose input is a geometric parameter and the output is an electromagnetic response,which is consistent with the regression application of machine learning,both of which need to find the mapping relationship between input and output,and the two can be combined with each other.Starting from MLAO,this study takes the sub-6GHz antenna that shines in 5G communication system as the research object,and uses machine learning method to design two kinds of antennas.First,based on the basic machine learning theory,the deep learning theory is explained,and the formula of the multi-layer perceptron algorithm in deep learning is deduced.On this basis,this study uses the Python-based Keras framework to complete the programming,and completes the verification of the program and training model on the double-T antenna unit and the double-T periodic structure antenna（metasurface antenna）,and summarizes the machine The general steps of learning and the generality of machine learning-assisted antenna optimization methods are studied to improve the reusability of the program.Second,two microstrip antennas operating in the Sub-6GHz frequency band are designed based on the validated machine learning model.In the design process,the CST modeling and simulation method is used to obtain the training and testing data sets.In this study,some geometric parameters of the antenna are used as the input of the machine learning model,and the electromagnetic response(S₁₁)of the antenna is used as the output of the model.During the training process,the K-fold cross-validation method and the dropout regularization method were used to optimize the model training results.Finally,the training model successfully realized the mapping from the antenna geometric parameters to the antenna electromagnetic response.On this basis,the geometric parameters of the above two antennas are optimized and compared with the results of the electromagnetic full-wave simulation.The machine learning optimization results are slightly better than the electromagnetic simulation optimization results.Two antennas are finally designed,one of which is an 8-port folded monopole MIMO antenna.Its operating frequency bands are 3.59-3.82GHz and 4.89-5.09GHz,and the in-band gains are 3.73d Bi and 1.52d Bi.The isolation of different ports can reach-15d B or less;the second is a 4-port square slotted MIMO antenna,its working frequency is 3.52-3.73GHz,the in-band gain is 3.5d Bi,and the isolation of different ports can reach-20d B or less.The research results of this study provide a general research method for machine learning-assisted antenna design,which is of great help to improve the efficiency of antenna design.The combination of machine learning and antenna design can also provide experience and open the way for future intelligent optimization of antennas.