Abstract
An overview of neural network-based modeling techniques and their applications in microwave modeling and design is presented. The neural network represent RF/microwave components with the help of training data that are pairs of model input-output (IO) data generated from detailed microwave simulation or measurement. Neural networks have significant advantages over other techniques for multidimensional function approximation as they permit a compact representation of a multidimensional function, requiring minimal storage of coefficients and being very efficient to evaluate. The neural network can produce a parametric model by exploiting existing microwave knowledge in the form of empirical/analytical/equivalent model during neural network development. Neural network maps an existing model to match a new device with a technique called Neuro-Space Mapping (Neuro-SM).