DocumentCode
3597034
Title
Efficient Jacobian Computation for High-Frequency Inverse Problem Solutions
Author
Yunpeng Song ; Nikolova, Natalia K.
Author_Institution
Dept. of Electr. & Comput. Eng., McMaster Univ., Hamilton, ON
fYear
2007
Firstpage
1
Lastpage
5
Abstract
Response Jacobians (gradients) can significantly improve the convergence of the reconstruction algorithms used in inverse problem solutions. However, the lack of efficient methods for computing response Jacobians has limited the applications of gradient-based algorithms to inverse problems when 3D numerical electromagnetic (EM) forward solvers are used. In this paper, we present the most recent developments in the self-adjoint sensitivity analysis (SASA) method for the computation of the response Jacobians with time-domain EM solvers. To our knowledge, our approach is the most computationally efficient method for EM sensitivity analysis with time-varying field solutions. It can deal with all types of optimizable (model) parameters, material parameters and shape parameters, of both dielectric and conducting objects in the structure of interest. Verification is carried out through the analysis of lossy dielectric structures.
Keywords
Jacobian matrices; computational electromagnetics; convergence of numerical methods; gradient methods; inverse problems; sensitivity analysis; time-domain analysis; 3D numerical electromagnetic forward solver; conducting object; convergence; dielectric object; gradient-based algorithm; high-frequency inverse problem solution; lossy dielectric structure; material parameter; optimizable parameter; reconstruction algorithm; response Jacobian computation; self-adjoint sensitivity analysis; shape parameter; time-domain analysis; time-varying field solution; Jacobian computation; Time domain analysis; adjoint-variable method; sensitivity analysis;
fLanguage
English
Publisher
iet
Conference_Titel
Antennas and Propagation, 2007. EuCAP 2007. The Second European Conference on
Print_ISBN
978-0-86341-842-6
Type
conf
Filename
4458782
Link To Document