Recent Advances in Modeling and Optimization of Multi-State Filters

The multi-state tunble filter in microwave area is widely used in many aspects. Actually we hope there can be a geometrically determined filter which possesses different frequency characteristics. And we can change its filter states by conveniently altering some specific parameters. To precisely and efficiently achieve the design of this kind of tunable filter, this paper review a unique technique with the help of ANN optimization and multiple novel algorithms. Considering the particularity of the filter, we use two sets of parameters as input during modeling. Parameters related to geometric dimensions are non-tunable parameters, and parameters related to filter state turning are tunable parameters. After optimization, we can obtain the value of the parameters under the desired response. The whole model consists of multiple sub-models and corresponds to the tuning states. All sub-models share non-tunable parameters and each sub-model has its corresponding tunable parameters. In order to ensure the accuracy of the optimization, both tunable and non-tunable parameters will be obtained simultaneously and constrained by all tuning states. Moreover, a trust-region algorithm is used to improve the convergence in the process of multiple tuning states optimization.