Advances of Neural Network Modeling Methods for RF / Microwave Applications

─ This paper provides an overview of recent advances of neural network modeling techniques which are very useful for RF/microwave modeling and design. First, we review neural network inverse modeling method for fast microwave design. Conventionally, design parameters are obtained using optimization techniques by multiple evaluations of EM-based models, which take a long time. To avoid this problem, neural network inverse models are developed in a special way, such that they provide design parameters quickly for a given specification. The method is used to design complex waveguide dual mode filters and design parameters are obtained faster than the conventional EM-based technique while retaining comparable accuracy. We also review recurrent neural network (RNN) and dynamic neural network (DNN) methods. Both RNN and DNN structures have the dynamic modeling capabilities and can be trained to learn the analog nonlinear behaviors of the original microwave circuits from input-output dynamic signals. The trained neural networks become fast and accurate behavioral models that can be subsequently used in systemlevel simulation and design replacing the CPUintensive detailed representations. Examples of amplifier and mixer behavioral modeling using the neural-network-based approach are also presented. Index Terms─ Behavioral modeling, computer aided design, neural network.