Efficient Modeling of Ku-Band High Power Dielectric Resonator Filter With Applications of Neural Networks

This paper presents a modeling method for high power dielectric resonator (DR) filter. An efficient and accurate model for DR filter is developed by preserving only the necessary information of the generalized scattering matrix (GSM) for each junction of the filter structure. Doing so, the sizes of the GSMs are dramatically reduced for the training and testing of the neural network (NN) models of each critical junction of the DR filter. The neural models are subsequently connected using a transmission line model for each mode in a cylindrical dielectric rod with conductor housing. Calculations of the propagation constant for each mode are given. The final model contains NN and transmission line models and is thus very fast. The method is described in details using a Ku-band high power DR filter. Efficiency and accuracy of the model are demonstrated through comparison with full-wave simulations and measurements. Excellent agreements are observed for both near-band and wide-band results.