Unified automated knowledge-based neural network modeling for microwave devices

This paper proposes an advance and unified algorithm for automated knowledge-based neural network (NN) model generation. Automated model generation (AMG) method has been applied to generate knowledge-based NN model with a nonlinear input mapping structure. On the basis of the existing literature, the proposed unified method takes both input and output mapping, both linear and nonlinear mapping into consideration during the automated modeling process. It can automatically detect the quality of the empirical model and choose the most suitable and compact mapping structures for the final knowledge-based NN model, which makes the modeling process more flexible than existing methods. The algorithm is demonstrated through microwave modeling examples.