RECENT TRENDS IN ADJOINT SENSITIVITY ANALYSIS FOR TRANSMISSION-LINE MODELLING METHOD

Abstract

<p> This thesis addresses recent trends and developments of the adjoint-variable method (AVM) for microwave structures with the time-domain transmission-line modeling (TD-TLM) method. </p> <p> Design sensitivity analysis of high-frequency (HF) structures is concerned with estimating the sensitivity of the response with respect to the design parameters. This information is essential at different stages of the design cycle such as the optimization, tolerance analysis, and yield analysis. </p> <p> Traditional approaches of sensitivity calculations involve estimating the sensitivities thought fmite-difference approximations. They suffer from formidable simulation time, as the full-wave analysis of practical HF structure requires extensive computational time. For a structure with N design parameters, at least N+l system analyses are required to extract the design response and its sensitivities. The adjoint variable method, on the other hand, supplies the sensitivity information in a very efficient way. Using at most two system analysis, the algorithm provides the design responses and its sensitivities, regardless of the number of the design parameters. </p> <p> In this thesis two contributions have been achieved which aims at enhancing the efficiency of the TLM-A VM framework. The first contribution is a reformulation of the AVM. This reformulation results in casting both the original and the adjoint systems in mathematically identical forms. It is shown that both systems can thus be modeled using a single TLM simulator with the only difference in the excitation. The second contribution focuses on generalizing the A VM algorithm by employing it for more advanced TLM nodes. The compatibility of the symmetrical condensed node (SCN) with the AVM algorithm has been verified in previous work for a general 3-D problem. Here, this is extended to include the hybrid symmetrical condensed node (HSCN), which is more efficient in terms of memory saving and simulation time. The new approaches are all illustrated through sensitivity estimation of different waveguide structures. </p>