A Novel Combined Modelling And Optimization Technique For Microwave Components

This paper presents a novel combined parametric modelling and design optimization technique for microwave components utilizing the neural networks. The proposed technique provides an iterative mechanism between ANN model training and design optimization update. This iterative mechanism is fully automated and requires no manual intervention. Furthermore, the proposed technique overcomes the limitations of the common ANN optimization strategy where the fixed training region of the ANN model limits the freedom of design optimization. The proposed technique automatically enlarges the ANN training region until an optimization solution satisfying the user's design specification is met. Once the whole iterative process is finished, an accurate parametric model and an optimal solution satisfying the design specification are simultaneously generated. A parametric modelling and design optimization example of a wideband QuasiElliptic filter design is presented to demonstrate the validity of this technique.