Antenna design exploiting adjoint sensitivity‐based geometry evolution

Abstract

The authors present a new approach for evolutionary antenna design. Through exploiting efficient adjoint sensitivity analysis techniques, the antenna structure evolves to better satisfy the design constraints. The coordinates of a selected number of control vertices are chosen as optimisation parameters thus enabling evolution to arbitrary shapes. The authors approach is illustrated through the design of a number of microstrip structures.

Rapid EM-Driven Antenna Dimension Scaling Through Inverse Modeling

“…Experience-driven parameter sweeps, which are nowadays the dominant EM-simulation-based design approaches, both in academia and industry, allow for obtaining acceptable results in practical timeframes, however, only a limited number of geometry parameters can be handled this way and the designs obtained are hardly optimal. A reduction in the design cost can be achieved by using, among others, gradientbased search with adjoint sensitivities [5], or surrogate-based optimization (SBO) techniques [6], [7]. In the latter, direct optimization of the high-fidelity EM antenna model is replaced by iterative construction and re-optimization of a cheaper representation of the structure (a so-called surrogate model).…”

Optimization-driven design of compact UWB MIMO antenna

“…The state of the art analysis indicates that considerable reduction of the simulation-driven design effort can be obtained either through gradient search with adjoint sensitivities [11], [12], or by employing fast surrogate models [13], [14]. The first option is conceptually simpler and with strong theoretical foundations, although commercial availability of adjoint technology in EM solvers is still very limited.…”

On Rapid Re-Design of UWB Antennas with Respect to Substrate Permittivity

“…Methods such as the gradient-based search with numerical derivatives [1] or population-based metaheuristics (often used for global optimization [2,3]) may be prohibitively expensive. A possible way of alleviating these difficulties are surrogate-assisted techniques [4,5] as well as the gradient search with adjoint sensitivities [6,7]. Surrogate-based methods exploit faster representations of the antenna structure under design, typically constructed from coarsediscretization EM simulation models (faster but less accurate), which are appropriately corrected to be used as prediction tools for finding better designs [4,5,8].…”