Research On Metasurface Optimization Design Method Based On Neural Network

Metamaterials are artificial composite materials composed of subwavelength units,which have electromagnetic properties that natural materials do not have.Metamaterials are two-dimensional metamaterials,which have attracted great attention in recent years due to their simple structure,easy processing,easy integration and other characteristics.However,for a long time,the design of metasurface units often depends on the experience of designers,and for the continuous application of metasurface layers,designers often need to master a large number of different physical knowledge to calculate the S-parameter distribution of metasurface.For some more complex applications,the S-parameter distribution is difficult to be directly calculated.The introduction of neural network is an important way to solve the difficulties in metasurface design.This paper proposes to combine the neural network with the metasurface near field imaging model and the metasurface far field RCS forming model to realize near-field reconstruction based on metasurface and far-field reconstruction based on metasurface.(1)Near-field reconstruction based on metasurface.Aiming at the problems such as slow convergence speed and not flexible use of GS algorithm,this paper combines the neural network with the metasurface near-field imaging model,regards the metasurface unit as the neuron of the neural network,regards the entire metasurface as the hidden layer,regards the incident field and the electric field distribution on the imaging surface as the input layer and output layer of the neural network respectively,and introduces the back propagation algorithm.Near field reconstruction is realized based on metasurface.Then,based on this algorithm,the detection of multichannel electromagnetic beam carrying orbital angular momentum is realized,which solves the problem that the traditional method needs to replace multiple spiral phase plates.The incident field of multichannel orbital angular momentum is quantized as input matrix,and the target image is taken as a label and put into the network for training,so as to obtain the metasurface phase distribution.The obtained phase distribution is modeled,simulated and tested.The results show that the simulation results are consistent with the experimental results,and both are compound expectations.(2)Far-field reconstruction based on metasurface.In view of the problem that the phase distribution corresponding to the metasurface cannot be calculated directly according to the metasurface target RCS in traditional electromagnetic illusions,this paper proposes to combine the neural network with the metasurface far-field RCS forming model,and also regard the metasurface as the neuron of the neural network,the entire metasurface as the hidden layer,and the incident field as the input layer.The RCS in the specified solid Angle space is regarded as the output layer,and the back propagation algorithm is introduced to realize the far field reconstruction based on the metasurface.Based on this algorithm,the automobile RCS are reconstructed,and the automobile RCS are obtained by simulation.The incident wave electric field distribution is quantized into a matrix as the input of the neural network,and the obtained automobile RCS are put into the network as a label for training,and the metasurface phase distribution is obtained.The obtained phase distribution is modeled and simulated based on the PB phase principle.The simulation results are in agreement with the expectation.