Parallel EM optimization approach to microwave filter design using feature assisted neuro-transfer functions

This paper proposes a new parallel EM optimization approach to microwave filter design using feature assisted neuro-transfer functions. We propose to modify the conventional optimization objective function by adding a new feature-parameter based optimization objective. The purpose is to use feature parameters to assist optimization to overcome the problem of bad starting points. We use feature zeros of the response transfer function as feature parameters, and develop surrogate models of feature zeros as functions of geometrical variables using parallel EM simulations. This approach use surrogate to guide EM optimization to simultaneously identify and adjust feature zeros into suitable frequency bands while forcing the responses to meet specifications. The proposed optimization has a better chance of avoiding local minima and reaches optimal EM solution faster.