Adaptive frequency sampling using linear Bayesian vector fitting

Abstract

The authors present a novel Bayesian approach to adaptively select frequency samples to obtain a rational macromodel of device responses over a broad frequency range while performing as few electromagnetic simulations as possible. The method leverages a Bayesian approach to vector fitting to construct a data‐driven uncertainty measure. The presented technique is demonstrated by application to a double semi‐circular patch antenna and is shown to accurately and efficiently construct a rational macromodel over the frequency range of interest.

Statistical modeling of frequency responses using linear Bayesian vector fitting

“…Using LB‐VF models of different orders helps mitigate the conditioning on the number of poles, because of the weighted average being taken. In Reference 10, the 10 highest order models were used, to maximally utilize this effect. Naturally, taking more model orders into account increases the computational cost.…”

“…With the aid of these examples, it is shown that such uncertainty information can be used for various purposes, such as characterizing device responses in the presence of noisy or missing data, to verify functionality and compliance, or for adaptive sampling. Note that in Reference 10, the idea of using Bayesian VF for adaptive sampling was introduced first, but only for a simple one‐port example of an antenna. In this work, the linear Bayesian vector fitting (LB‐VF) framework is fully developed and tested on two multiport systems (a microwave filter and a RAM memory channel) for various applications.…”

Bayesian Optimization for Microwave Devices Using Deep GP Spectral Surrogate Models

“…Then, s values can also be acquired at different frequencies for each parameter combination p, without modifying the objective definition. This enables the usage of adaptive frequency sampling schemes [22] in the simulation software, thus further accelerating data acquisition during the model training.…”

A Bayesian Approach to Adaptive Frequency Sampling

“…The application of this framework to a representative example will confirm its aptness and efficiency. A prototypical version of this framework was presented in [9], on a one-port example.…”