Artificial Neural Networks and Space Mapping for EM-Based Modeling and Design of Microwave Circuits

Abstract

This chapter reviews the intersection of two major CAD technologies for modeling and design of RF and microwave circuits: artificial neural networks (ANNs) and space mapping (SM). A brief introduction to ANNs is first presented, starting from elementary concepts associated to biological neurons. Electromagnetic (EM)-based modeling and design optimization of microwave circuits using ANNs is addressed. The conventional and most widely used neural network approach for RF and microwave design optimization is explained, followed by brief descriptions of typical enhancing techniques, such as decomposition, design of experiments, clusterization, and adaptive data sampling. More advanced approaches for ANN-based design exploiting microwave knowledge are briefly reviewed, including the hybrid EM-ANN approach, the prior knowledge input method, and knowledge-based neural networks. Computationally efficient neural SM methods for highly accurate EM-based design optimization are surveyed, contrasting different strategies for developing suitable (input and output) neural mappings. A high-level perspective is kept throughout the chapter, emphasizing the main ideas associated with these innovative techniques. A tutorial example using commercially available CAD tools is finally presented to illustrate the efficiency of the neural SM methods.