• DocumentCode
    2322192
  • Title

    Sensitivity analysis of electrically small objects in lossy inhomogeneous structures

  • Author

    Song, Yunpeng ; Nikolova, Natalia N.

  • Author_Institution
    McMaster Univ., Hamilton
  • fYear
    2007
  • fDate
    9-15 June 2007
  • Firstpage
    4453
  • Lastpage
    4456
  • Abstract
    Accurate response Jacobians can significantly improve the convergence and the resolution of the reconstruction algorithms used in inverse microwave imaging. As the electrical size of an object decreases, however, the accuracy of the computed Jacobians with respect to its shape and constitutive parameters deteriorates. This is especially valid if finite-difference (FD) approximations are used. In this paper, we investigate the accuracy of the Jacobians provided by our self-adjoint sensitivity analysis (SASA) of lossy dielectric structures. We focus on time-domain solutions. The accuracy of the response gradients is investigated for electrically small objects. Recommendations are given for a proper choice of shape parameters when the object´s size becomes smaller than certain limits. We show that our approach yields accurate Jacobians even for objects as small as one grid cell while the FD approximations are prone to numerical noise.
  • Keywords
    Jacobian matrices; finite difference methods; image reconstruction; microwave imaging; sensitivity analysis; Jacobians; finite-difference approximations; lossy inhomogeneous structures; reconstruction algorithms; sensitivity analysis; Convergence; Dielectric losses; Finite difference methods; Image resolution; Jacobian matrices; Microwave imaging; Reconstruction algorithms; Sensitivity analysis; Shape; Time domain analysis;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Antennas and Propagation Society International Symposium, 2007 IEEE
  • Conference_Location
    Honolulu, HI
  • Print_ISBN
    978-1-4244-0877-1
  • Electronic_ISBN
    978-1-4244-0878-8
  • Type

    conf

  • DOI
    10.1109/APS.2007.4396531
  • Filename
    4396531