We introduce an approach to carry out layout optimization of metallic antenna parts. An optimization technique first developed for the optimization of load-bearing elastic structures is adapted for the purpose of metallic antenna design. The local conductivity values in a given region are used as design variables and are iteratively updated by a gradient-based optimization algorithm. Given a set of time-domain signals from exterior sources, the design objective is here to maximize the energy received by the antenna and transmitted to a coaxial cable. The optimization proceeds through a sequence of coarsely-defined lossy designs with successively increasing details and less losses as the iterations proceed. The objective function gradient is derived based on the FDTD discretization of Maxwell's equations and is expressed in terms of field solutions of the original antenna problem and an adjoint field problem. The same FDTD code, but with different wave sources, is used for both the original antenna problem and the adjoint problem. For any number of design variables, the gradient is evaluated on the basis of only two FDTD simulations, one for the original antenna problem and another for the adjoint field problem. We demonstrate the capability of the method by optimizing the radiating patch of both UWB monopole and microstrip antennas. The UWB monopole is designed to radiate over a wide frequency band 1-10 GHz, while the microstrip patch is designed for single and dual frequency band operation. In these examples, there are more than 20,000 design variables, and the algorithm typically converges in less than 150 iterations. The optimization results show a promising use of the proposed approach as a general method for conceptual design of near-resonance metallic antennas.

We show how the computational power and programmability of modern graphics processing units (GPUs) can be used to efficiently solve large-scale pixel-based material distribution problems using a gradient-based optimality criterion method. To illustrate the principle, a so-called topology optimization problem that results in a constrained nonlinear programming problem with over 4 million decision variables is solved on a commodity GPU.

We present an approach to design from scratch planar microwave antennas for the purpose of ultra-wideband (UWB) near-field sensing. Up to about 120 000 design variables associated with square grids on planar substrates are subject to design, and a numerical optimization algorithm decides, after around 200 iterations, for each edge in the grid whether it should consist of metal or a dielectric. The antenna layouts produced with this approach show UWB impedance matching properties and near-field coupling coefficients that are Hat over a much wider frequency range than a standard UWB antenna. The properties of the optimized antennas are successfully cross-verified with a commercial software and, for one of the designs, also validated experimentally. We demonstrate that an antenna optimized in this way shows a high sensitivity when used for near-field detection of a phantom with dielectric properties representative of muscle tissue.