<p> The thesis presents an efficient self-adjoint approach to the S-parameter sensitivity analysis based on full-wave electromagnetic (EM) time-domain simulations with two commonly used numerical techniques: the finite-difference time-domain (FDTD) method and the transmission-line matrix (TLM) method. Without any additional simulations, we extract the response gradient with respect to all the design variables making use of the full-wave solution already generated by the system analysis. It allows the computation of the S-parameter derivatives as an independent post-process with negligible overhead. The sole requirement is the ability of the solver to export the field solution at user-defined points. Most in-house and commercial solvers have this ability, which makes our approach readily applicable to practical design problems.</p> <p> In the TLM-based self-adjoint techniques, we propose an algorithm to convert the electrical and magnetic field solutions into TLM voltages. The TLM-based discrete adjoint variable method (AVM) is originally developed to use incident and reflected voltages as the state variables. Our conversion algorithm makes the TLM-AVM method applicable to all time-domain commercial solvers, FDTD simulators included, with
comparable accuracy and less memory overhead. Our approach is illustrated through waveguide examples using a TLM-based commercial simulator.</p> <p> Currently, our TLM-based self-adjoint approach is limited to loss-free homogeneous problems. However, our FDTD-based self-adjoint approach is valid for lossy inhomogeneous cases as well. The FDTD-based self-adjoint technique needs only the E-field values as the state variables. In order to make it also applicable to a TLM-based solver, whose mesh grid is displaced from the FDTD grid, we interpolate the E-field solution from the TLM mesh to that on the FDTD mesh. Our FDTD-based approach is validated through the response derivatives computation with respect to both shape and constitutive parameters in waveguide and antenna structures. The response derivatives can be used not only to guide a gradient-based optimizer, but also to provide a sufficient good initial guess for the solution of nonlinear inverse problems.</p> <p> Suggestions for further research are provided.</p> / Thesis / Master of Applied Science (MASc)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/21887 |
| Date | 06 1900 |
| Creators | Li, Ying |
| Contributors | Nikolova, Natalia, Bakr, Mohamed, Electrical and Computer Engineering |
| Source Sets | McMaster University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0029 seconds