Enhancing computation performance of rational approximation for frequency-dependent network equivalents with parallelism and complex vector fitting.

This work examines two strategies for enhancing the rational approximation of Frequency-Dependent Network Equivalents (FDNE) using an 8-port FDNE featuring a frequency response marked by numerous peaks and valleys. Firstly, we employ Complex Vector Fitting (CVF), an alternative to the Vector Fitting (VF). CVF eliminates the constraint of complex conjugate pairs and was originally conceived for modeling baseband equivalents through scattering parameters. The implications of CVF for admittance (or impedance) matrix synthesis have not yet been previously reported in specialized literature. To enhance code performance and remove dependence on commercial software such as MATLAB®, VF and CVF were implemented in the C-language, utilizing a low-level linear algebra package and exploiting parallelism. We evaluated performance by varying the model order, number of ports, and frequency samples. The results confirm the feasibility of our approach, prompting a more in-depth exploration of the potential benefits regarding FDNE realization. • The constraint relaxation of the Complex Vector Fitting enhanced model's accuracy. • The Complex Vector Fitting achieved a model order reduction of 48%. • The parallel implementation improved computational efficiency using free libraries. [ABSTRACT FROM AUTHOR]