Modeling of liquid crystal tunable microwave phase shifter exploiting artificial neural networks

In the last few years, numerous frequency agile devices that require materials with specific properties (anisotropic, ferroelectric, etc.) have been developed. The empirical models are faster and easier to use for the design engineer than any numerical technique. However, it is difficult to develop accurate closed-form relationships of devices for a large range of geometric and materials parameters. Thus, we suggest two fast and accurate methods, which combine analytical relationships with an artificial neural network trained with electromagnetic data, provided by finite element software. These methods are illustrated for a tunable microwave phase shifter. Closed-form relationships have been developed and are presented in this paper. A comparison between the response of the tunable microwave phase shifter, using the methods exploiting ANN, and electromagnetic simulation data are presented. The result of these methods shows good agreement with the predicted data.