Full Newton method for electromagnetic inverse scattering, utilizing
explicit second order derivatives
The full Newton's method is considered as an optimisation approach to the
inverse electromagnetic scattering problem in the frequency-domain. For the
sake of accuracy and computational efficiency, the gradient and the Hessian
of the cost-functional, with respect to parameter functions in the
L2-space, are derived explicitly by means of the adjoint scattering problem
and the first and second order Frechet differentials of the cost-functional.
The theory is presented for the general case involving a three-dimensional
scatterer consisting of general linear media, so called bianisotropic
media. Numerical results are presented for a non-uniform transmission line
problem. The numerical implementation, when reducing to a finite
dimensional parameter space, and a regularization technique, for the
resulting ill-conditioned Hessian matrix, are presented.
For the reconstruction of one or two parameters, the algorithm is tested on
synthetic reflection data contaminated with Gaussian noise. The algorithm
is also tested on measured reflection data to reconstruct a piecewise
constant shunt-capacitance.