Simulation-driven Design Optimization And Modeling For Microwave Engineering

Slawomir Koziel, Xin-She Yang, Qi-Jun Zhang

• Publisher: ICP
• Year: 2013
• Print ISBN: 9781848169166
• eBook ISBN: 9781848169227
• Topic: Biological Sciences
• Subtopic: Science General
• Index: Biological Sciences

Computer-aided full-wave electromagnetic (EM) analysis has been used in microwave engineering for the past decade. Initially, its main application area was design verification. Today, EM-simulation-driven optimization and design closure become increasingly important due to the complexity of microwave structures and increasing demands for accuracy. In many situations, theoretical models of microwave structures can only be used to yield the initial designs that need to be further fine-tuned to meet given performance requirements. In addition, EM-based design is a must for a growing number of microwave devices such as ultra-wideband (UWB) antennas, dielectric resonator antennas and substrate-integrated circuits. For circuits like these, no design-ready theoretical models are available, so design improvement can only be obtained through geometry adjustments based on repetitive, time-consuming simulations. On the other hand, various interactions between microwave devices and their environment, such as feeding structures and housing, must be taken into account, and this is only possible through full-wave EM analysis.

Electromagnetic simulations can be highly accurate, but they tend to be computationally expensive. Therefore, practical design optimization methods have to be computationally efficient, so that the number of CPU-intensive high-fidelity EM simulations is reduced as much as possible during the design process. For the same reasons, techniques for creating fast yet accurate models of microwave structures become crucially important.

In this edited book, the authors strive to review the state-of-the-art simulation-driven microwave design optimization and modeling. A group of international experts specialized in various aspects of microwave computer-aided design summarize and review a wide range of the latest developments and real-world applications. Topics include conventional and surrogate-based design optimization techniques, methods exploiting adjoint sensitivity, simulation-based tuning, space mapping, and several modeling methodologies, such as artificial neural networks and kriging. Applications and case studies include microwave filters, antennas, substrate integrated structures and various active components and circuits. The book also contains a few introductory chapters highlighting the fundamentals of optimization and modeling, gradient-based and derivative-free algorithms, metaheuristics, and surrogate-based optimization techniques, as well as finite difference and finite element methods.

Contents:

• Introduction to Optimization and Gradient-Based Methods (Xin-She Yang and Slawomir Koziel)
• Derivative-Free Methods and Metaheuristics (Xin-She Yang and Slawomir Koziel)
• Surrogate-Based Optimization (Slawomir Koziel, Leifur Leifsson, and Xin-She Yang)
• Space Mapping (Slawomir Koziel, Stanislav Ogurtsov, Qingsha S Cheng, and John W Bandler)
• Tuning Space Mapping (Qingsha S Cheng, John W Bandler, and Slawomir Koziel)
• Robust Design Using Knowledge-Based Response Correction and Adaptive Design Specifications (Slawomir Koziel, Stanislav Ogurtsov, and Leifur Leifsson)
• Simulation-Driven Design of Broadband Antennas Using Surrogate-Based Optimization (Slawomir Koziel and Stanislav Ogurtsov)
• Neural Networks for Radio Frequency/Microwave Modeling (Chuan Zhang, Lei Zhang, and Qi-Jun Zhang)
• Parametric Modeling of Microwave Passive Components Using Combined Neural Network and Transfer Function (Yazi Cao, Venu-Madhav-Reddy Gongal-Reddy, and Qi-Jun Zhang)
• Parametric Sensitivity Macromodels for Gradient-Based Optimization (Krishnan Chemmangat, Francesco Ferranti, Tom Dhaene, and Luc Knockaert)
• Neural Space Mapping Methods for Electromagnetics-Based Yield Estimation (José E Rayas-Sánchez)
• Neural Network Inverse Modeling for Microwave Filter Design (Humayun Kabir, Ying Wang, Ming Yu, and Qi-Jun Zhang)
• Simulation-Driven Design of Microwave Filters for Space Applications (Elena Díaz Caballero, José Vicente Morro Ros, Héctor Esteban González, Vicente Enrique Bôria Esbert, Carmen Bachiller Martín, and Ángel Belenguer Martinez)
• Time Domain Adjoint Sensitivities: The Transmission Line Modeling (TLM) Case (Mohamed H Bakr and Osman S Ahmed)
• Boundary Conditions for Two-Dimensional Finite-Element Modeling of Microwave Devices (Tian-Hong Loh and Christos Mias)
• Boundary Conditions for Three-Dimensional Finite-Element Modeling of Microwave Devices (Tian-Hong Loh and Christos Mias)

Readership: Graduates, lecturers, and researchers in electrical engineering, as well as engineers who use numerical optimization in their design work. This book will be of great interest to researchers in the fields of microwave engineering, antenna design, and computational electromagnetics.

Chapter 1

Introduction to Optimization and Gradient-Based Methods

Xin-She Yang and Slawomir Koziel

Optimization is everywhere, and this is especially true for simulation-driven optimization and modeling in microwave engineering. Optimization is an important paradigm itself with a wide range of applications. In almost all applications in engineering and industry, we are always trying to optimize something — whether to minimize the cost and energy consumption, or to maximize the profit, output, performance, and efficiency. In reality, resources, time, and money are always limited; consequently, optimization is far more important in practice (Gill et al., 1981; Arora, 1989; Yang, 2010; Yang and Koziel, 2010). The optimal use of available resources of any sort requires a paradigm shift in scientific thinking; this is because most real-world applications have far more complicated factors and parameters which affect how the system behaves.

1.1. Introduction

Contemporary engineering design is heavily based on computer simulations. This introduces additional difficulties to optimization. Growing demand for accuracy and ever-increasing complexity of structures and systems results in the simulation process being more and more time-consuming. In many engineering fields, the evaluation of a single design can take as long as several days or even weeks. On the other hand, simulation-based objective functions are inherently noisy, which makes the optimization process even more difficult. Still, simulation-driven design becomes a must for a growing number of areas, which creates a need for robust and efficient optimization methodologies that can yield satisfactory designs even given the presence of analytically intractable objectives and limited computational resources.

For any optimization problem, the integrated components of the optimization process are optimization algorithm, an efficient numerical simulator, and a realistic representation of the physical processes we wish to model and optimize. This is often a time-consuming process, and in many cases, the computational costs are usually very high. Once we have a good model, the overall computation costs are determined by the optimization algorithms used for search and the numerical solver used for simulation.

Search algorithms are the tools and techniques of achieving optimality of the problem of interest. This search for optimality is complicated further by the fact that uncertainty is almost always presents in real-world systems. Therefore, we seek not only the optimal design but also robust design in engineering and industry. Optimal solutions, which are not robust enough, are not practical in reality. Suboptimal solutions or good robust solutions are often the choice in such cases.

Simulations are often the most time-consuming part. In many applications, an optimization process often involves the evaluation of objective function many times, often thousands and even millions of configurations. Such evaluations often involve the use of extensive computational tools such as a computational fluid dynamics simulator or a finite-element solver. This is the step that is most time-consuming, often taking 50% to 90% of the overall computing time. Therefore, we have to balance the accuracy (high-fidelity) and allowable computational time. In many cases, some fast approximations (often with reduced fidelity) can be used for most parts of the search iterations and the high-fidelity model used for double-checking the good designs. This combination of approximations and variable fidelity helps to reduce the overall computational time significantly.

Optimization problems can be formulated in many ways. For example, the commonly used method of least-squares is a special case of maximum-likelihood formulations. By far the most widely used formulation is to write a nonlinear optimization problem as

subject to the constraints

where f_i, h_j and g_k are in general nonlinear functions. Here the design vector x = (x₁, x₂, …, x_n) can be continuous, discrete, or mixed in n-dimensional space. The functions f_i are called objective or cost functions, and when M > 1, the optimization is multiobjective or multicriteria (Sawaragi et al., 1985). It is possible to combine different objectives into a single objective, and we will focus on the single-objective optimization problems in most parts of this book. It is worth pointing out here that we write the problem as a minimization problem, but it can also be written as a maximization by simply replacing f_i(x) by – f_i(x).

When all functions are nonlinear, we are dealing with nonlinear constrained problems. In some special cases when f_i, h_j, g_k are linear, the problem becomes linear, and we can use the widely linear programming techniques such as the simplex method. When some design variables can only take discrete values (often integers), while other variables are real continuous, the problem is of mixed type, which is often difficult to solve, especially for large-scale optimization problems.

A very special class of optimization is convex optimization (Boyd and Vandenberghe, 2004), which has guaranteed global optimality. Any optimal solution is also the global optimum, and most importantly, there are efficient algorithms of polynomial time to solve such problems (Conn et al., 2000). These efficient algorithms such as the interior-point methods (Karmarkar, 1984) are widely used and have been implemented in many software packages.

1.2. Main Challenges in Optimization

There are three main issues in simulation-driven optimization and modeling, and they are: the efficiency of an algorithm, and the efficiency and accuracy of a numerical simulator, and assigning the right algorithms to the right problem. Despite their importance, there is no satisfactory rule or guidelines to handle such issues. Obviously, we try to use the most efficient algorithms available, but the actual efficiency of an algorithm may depend on many factors such as the inner working of an algorithm, the information need (such as objective functions and their derivatives), and implementation details. The efficiency of a solver is even more complicated, depending on the actual ...

Table of contents

• Cover
• Title Page
• Title Page1
• Copyright Page
• List of Contributors
• PREFACE
• Acknowledgments
• Table of Contents
• 1.  Introduction to Optimization and Gradient-Based Methods
• 2.  Derivative-Free Methods and Metaheuristics
• 3.  Surrogate-Based Optimization
• 4.  Space Mapping
• 5.  Tuning Space Mapping
• 6.  Robust Design Using Knowledge-Based Response Correction and Adaptive Design Specifications
• 7.  Simulation-Driven Design of Broadband Antennas Using Surrogate-Based Optimization
• 8.  Neural Networks for Radio Frequency/Microwave Modeling
• 9.  Parametric Modeling of Microwave Passive Components Using Combined Neural Network and Transfer Function
• 10.  Parametric Sensitivity Macromodels for Gradient-Based Optimization
• 11.  Neural Space Mapping Methods for Electromagnetics-Based Yield Estimation
• 12.  Neural Network Inverse Modeling for Microwave Filter Design
• 13.  Simulation-Driven Design of Microwave Filters for Space Applications
• 14.  Time Domain Adjoint Sensitivities: The Transmission Line Modeling (TLM) Case
• 15.  Boundary Conditions for Two-Dimensional Finite-Element Modeling of Microwave Devices
• 16.  Boundary Conditions for Three-Dimensional Finite-Element Modeling of Microwave Devices
• Index