In the traditional finite difference time domain (FDTD) method: the studied
structure is modeled as elementary cells, which sizes have to be small eno
ugh to get close to the reality. Then, this discretization is applied to th
e whole calculation volume. even though some zones do not need such fine di
scretization. Consequently, this numerical tool requires an important calcu
lation capacity to increase the precision of the grid mesh. For very large
dimension structures, computer limitations no longer allow a sufficiently a
ccurate discretization. The aim of this article is to discuss some methods
which allow the EM study of large structures containing elements of very un
equal size as compared to the wavelengths used, while preserving an accepta
ble calculation cost. This approach combines finite differences of variable
precision and different-sized cells in the same calculation domain. It all
ows a fine discretization of small structures and a rough discretization of
the other elements of the Ehl problem. This spatial subgridding gives a ve
ry original possibility of undertaking EM zooms on some specific parts of t
he calculation domain thus allowing a better modelling of the studied struc
ture while preserving good accuracy and an acceptable cost.