For many problems in radio frequency (RF) and electromagnetics (EM) engineering, it is impractical to perform numerous experiments on the physical world directly. Even computer aided design (CAD)-based simulations often require a substantial investment in terms of computational cost, due to the complexity of the systems under study and the high signal bandwidth needed in modern engineering applications.
In this framework, data-efficient optimization techniques are of paramount importance, especially when a high number of design parameters is considered (curse of dimensionality). In particular, optimization of modern RF and EM systems is a non-trivial task given that such systems typically have a non-linear response to its control parameters, such as the width and length of the metallic traces of an interconnect, and multiple local minima can exist in the chosen design space, which is the typical case for power amplifier design, for example.
In this tutorial, a novel approach based on Bayesian interference is presented: the main idea is to perform
optimization on surrogate models, which mimic the real optimization problem, but contrary to the latter, are very
cheap to evaluate. However, differently from other surrogate-based optimization strategies, the models are defined
in a probabilistic framework, which consists of a prior distribution that captures our beliefs about the behavior of
the unknown objective function and an observation model that describes the data generation mechanism.
The main features of Bayesian optimization (BO) are thoroughly explained in this tutorial and relevant application
examples are used to describe the characteristics of state-of-the-art BO techniques.
Tom Dhaene received the Ph.D. degree in electrotechnical engineering from Ghent University, Ghent, Belgium, in 1993. From 1989 to 1993, he was a Research Assistant with the Department of Information Technology, Ghent University. In 1993, he was with EDA Company Alphabit (currently part of Keysight Technologies), Ghent. He was one of the key developers of the planar EM simulator ADS momentum. Since 2007, he has been a Full Professor with the Department of Information Technology, Ghent University/imec, Ghent. He has authored or co-authored over 300 peer-reviewed papers and abstracts in international conference proceedings, journals, and books. He is the holder of five U.S. patents. His current research interests include the different aspects of full-wave electromagnetic circuit modeling, transient simulation, and time-domain characterization of high-frequency and high speed interconnections
Domenico Spina received the M.S. degree (summa cum laude) in electronics engineering from the University of L'Aquila in 2010. He completed a joint Ph.D. program in 2014, obtaining the Ph.D. degree "Doctor of Electrical Engineering" from the Ghent University and "Dottore di Ricerca in Ingegneria Elettrica e dell'Informazione" from the University of L'Aquila. Since 2015, he has been a postdoctoral researcher at the Internet Technology and Data Science Lab (IDLab) of the Department of Information Technology (INTEC) at Ghent University, and he is currently affiliated with imec. His current research interests include machine learning for electrical and microwave engineering, modeling and simulation, system identification, sensitivity and uncertainty analysis.