Research On Design Method Of Substrate Integrated Waveguide Device Based On Neural Network

Substrate Integrated Waveguide(SIW),as a new type of guided wave structure,has been widely used in the design of microwave devices due to the advantages of low loss,small volume,high quality factor,and easy integration with other structures.But,the current design methods of SIW devices,which still need to do lots of iterative processes to optimize the geometric parameters,are very time-consuming.Neural networks have been regarded as an effective alternative technology for traditional modeling methods due to their powerful modeling capabilities of nonlinear problems and fast response speed,and have been successfully applied to the modeling of various microwave devices.To model the SIW devices fastly and accurately,the design method of SIW devices has been deeply studied based on neural networks in this paper.By incorporating the knowledge of filter decomposition with the inverse neural network,we build a coarse model that can synthesize the dimensions of a SIW filter.However,due to the characteristics of the SIW structure itself and the errors generated in the merge of the sub-models,the results of the coarse model are very different from the ideal response.We propose a novel calibrated coarse model from the perspective of the coupling matrix to correct the errors generated in the coarse model.Besides,this paper also proposes an equivalent de-embedding technique which is simpler than the thrureflect-line(TRL)calibration technique to accurately extract the generalized scattering matrix(GSM)of the SIW iris.An H-plane 5th order SIW filter is synthesized by the proposed model and the design efficiency is greatly improved by using Python drive HFSS to achieve automated simulation.The result shows that the SIW filter whose frequency response is very close to the ideal one can be synthesized with only a few hundred training data,which prove the feasibility and accuracy of the proposed model.Two improved coupling matrix models,pre-tuned and collaborative tuning,are proposed based on the neural network to solve the problem of the passband shift of the coupled-resonator SIW filter synthesized by the traditional coupling matrix model.At the beginning of the design of the pre-tuned model,the basic parameters of the SIW resonator are pre-determined by considering a variety of parameters affecting the resonance frequency.The center frequency will be used for the input of the neural network of collaborative tuning mode,which means the parameters affecting the resonance frequency will be determined by the neural network.The same C-band third-order bandpass filter based on SIW regular triangle cavity is synthesized by the two improved models.The results show that both methods can not only solve the problem of frequency shift but also can acquire a good frequency response which fully meets the design goals and is no necessity to do the time-consuming optimization process.Finally,the two improved coupling matrix models are combined with the calibrated coarse model to make the frequency response of the SIW filter very close to the ideal response.It further verifies that the calibrated coarse model has wide adaptability and accuracy,not only can be combined with different types of circuit models but also show high accuracy.