Research On Analysis And Prediction Of Array Near Field Spatial Power Synthesis

In electronic information systems,the electromagnetic protection capabilities of some devices exposed to specific high-power environments need to be tested.In order to satisfy the system electromagnetic compatibility requirements of special near-field electromagnetic radiation during the inspection,power synthesis can be achieved by designing the amplitude/phase of transmitting antenna array.After that,the expected field strength and power density can be obtained at a specified spatial position.The synthesis field strength is affected by the distance between the antenna array and target position.In a certain near-field range of the array,the reduction of distance can improve the power synthesis intensity of the target position,but it will lead to the coupling intensification between array and the device under test.Therefore,it is necessary to study the spatial power synthesis of array near field in depth.The topic of this thesis originated from the project of *** project.When the device under test exists in the near-field area of the array,the change of the field strength distribution at the power synthesis target with the material and structural characteristics of the device under test is studied.Meanwhile,in order to improve the computational efficiency of solving electromagnetic distribution,a neural network algorithm combined with full-wave simulation is proposed to establish the field intensity and power analysis model.The factors that affect the prediction accuracy of neural network algorithms and the generalization ability of the final model are analyzed in detail.The main work of this thesis includes:1.Designed a high-power horn antenna that works in the typical X-band,and studied its port matching and far-field radiation characteristics,and the comparison of the theoretical and simulation results of the power synthesis results after the array.Firstly,the structure of the horn antenna feed waveguide is optimized to meet the requirements of low profile and the gain of the main radiation direction in the whole frequency band is greater than 11.8d Bi.Secondly,the power synthesis method of array near field based on phase compensation is used to compare the electric field distribution results of theoretical analysis and full-wave simulation in the near field region between 10-element linear array and 10×10 plane array by taking a half-wave dipole antenna as an example.Finally,the placement mode of the 10-element high-power horn antenna array is determined by analyzing the synthetic field strength under the given target coordinates.2.The field intensity distribution of high-power horn linear array at the target is studied when there is a typical test device in the near field area.Among them,the equipment under test is a typical total reflection metal plate,a dielectric plate or a receiving antenna.Some factors that affect the peak value of the electric field at the target point are analyzed,including: the length of the total reflection metal plate;the thickness,length and dielectric constant of the dielectric plate;the distance,elevation angle and operating frequency of the horn receiving antenna.3.Based on the machine learning method,a prediction model of the field strength distribution and the received power of the receiving antenna port is established.When establishing the prediction model of oral field strength,multivariate sampling was designed to obtain input samples for training and testing.After that,batch simulation was performed in MATLAB with FEKO to obtain the output electric field and power corresponding to each input sample.And 10-fold cross-validation is used as a model selection and model verification method to discuss the influence of sample size and network architecture on the accuracy of training results.The performance of Levenberg-Marquard algorithm,Bayesian regularization algorithm and genetic algorithm for optimizing hyperparameters is compared.In the received power prediction model,the accuracy of the prediction model established by support vector regression,Gaussian process regression and artificial neural network on the test set is discussed.A new test set was designed according to the value range of the variables,which verified the generalization ability of the best neural network model in analyzing electric field distribution and received power.