Microwave Filter Modeling and Intelligent Tuning

Abstract

Traditional filter tuning methods mainly entail tuning by electromagnetic simulation technology, which treats the tuned filter as an ideal model. However, the structure of the actual filter is relatively complex, and filter tuning becomes affected by the loss of resonant cavity, phase loading and high-order mode. In this study, to solve these problems, the tuning process was divided into four stages. First, the passband and suppression of the filter could be tuned to a reasonable range by using the phase attribute of the reflection characteristics. Secondly, the tuning model parameters (coupling matrix) were extracted by curve fitting and the improved Cauchy method. Thirdly, the tuning model of the actual filter was established by a complex neural network. Finally, the mapping relationship between the surrogate model and the actual tuning model was established by the improved space mapping algorithm. By optimizing the parameters of surrogate model, we quickly obtained the optimal position of the screws. The results of the tuning experiment with the eighth coaxial cavity filter revealed that the method had high accuracy and fast convergence speed.