Space Mapping for Electromagnetic-Simulation-Driven Design Optimization

Abstract

Space mapping (SM) has been one of the most popular surrogate-based optimization techniques in microwave engineering to date. By exploiting the knowledge embedded in the underlying coarse model (e.g., an equivalent circuit), SM allows dramatic reduction of the computational cost while optimizing electromagnetic (EM)-simulated structures such as filters or antennas. While potentially very efficient, SM is not always straightforward to implement and set up, and may suffer from convergence problems. In this chapter, we discuss several variations of an SM optimization algorithm aimed at improving SM performance for design problems involving EM simulations. These include SM with constrained parameter extraction and surrogate model optimization designed to overcome the problem of selecting preassigned parameters for implicit SM, SM with response surface approximation coarse models that maintain SM efficiency when a fast coarse model is not available, and SM with sensitivity which takes advantage of adjoint sensitivity (which has recently become commercially available in EM simulators) to improve the convergence properties and further reduce the computational cost of SM algorithms. Each variation of the SM algorithm presented here is illustrated using a real-world microwave design example.