Geometry Scaling of Microwave Filters Using an Adaptive Homotopy Continuation Method

In microwave engineering, efficient geometry scaling of microwave filters can be a challenging task. In this paper, a new procedure for geometry scaling deploying an adaptive homotopy continuation method is proposed to redesign filters. The proposed approach aims at forming a database with appropriate data distribution and then an Artificial Neural Network (ANN) is trained upon it. Starting from one design, the database is constructed using homotopy continuation with adaptive steps and a local polynomial regression model. Designs in the database consider both the geometry parameters and inherently shifting characteristics during the scaling process, such as the nominal position of transmission zeros. Then given the required bandwidth and centre frequency, the well-trained ANN serves as an inverse model to output the filter geometry, which is later adjusted by a tuning procedure. The proposed method is demonstrated using a fourth-order coaxial bandpass filter, where three groups of design specifications are required.