Extraction of Coupling Matrix for Bandpass Filters Based on Magnitude of S-parameters

A practically comprehensive magnitude-based model extraction framework for microwave band-pass filters is proposed. Unlike existing methods that require both magnitude and phase information of S-parameters, the new framework can extract the coupling matrix (CM) only using the magnitude information alone even in a highly ambiguous circumstance. The proposed framework consists of two major parts: 1) construction and training of convolutional neural network (CNN), with which a unique nonlinear mapping from the magnitude information to the CM is established; and 2) ambiguity identification and removal schemes, by which the non-physical solutions due to the absence of phase information are eliminated. To demonstrate the effectiveness of the proposed framework, examples of both ideal and practical filters are provided. The proposed framework provides the wireless industry with a useful tool for the robotic automatic-tuning of microwave bandpass filters.