FDTD
by
electromagnetics.info
The FDTD method can be classed as a special case of the
finite difference method , a standard method for solving PDEs. It is a time-stepping method, wherein starting
at time t=0, the electromagnetic properties are updated at each time level to give the evolving solution.
The FDTD method can be used in 1, 2 and 3 dimensions, but it is the more practical 3D case
that is the most important and which the Yee algorithm addresses. It is the 3 dimensional FDTD
solution of electromagnetic problems that is considered here.
1 Modelling
For a linear, isotropic nondispersive materials (i.e. materials
having field-independent, direction-independent, and frequency-independent
electric and magnetic properties), then Maxwell's curl equations take the
following form:
where H=[Hx,Hy,Hz] is the magnetic field intensity,
E=[Ex,Ey,Ez] is the electric field intensity,
m is the permeability, e is the permittivity,
r is the magnetic resistivity
and s is the conductivity. (Note that the magnetic resistivity (r) is
generally taken to be zero. Its inclusion above is simply
to demonstrate the similar form of the two equations.)
In general m, r, e and s can vary throughout the medium. For the background to these equations, click
here .
The individual components of the equations (1)
and (2) are as follows
|
|
¶Hx ¶t
|
= - |
1 m
|
( |
¶Ey ¶z
|
- |
¶Ez ¶y
|
- rHx ) , |
| (3) |
|
|
¶Hy ¶t
|
= - |
1 m
|
( |
¶Ez ¶x
|
- |
¶Ex ¶z
|
- rHy) , |
| (4) |
|
|
¶Hz ¶t
|
= - |
1 m
|
( |
¶Ex ¶y
|
- |
¶Ey ¶x
|
- rHz ) , |
| (5) |
from equation (1) and
|
|
¶Ex ¶t
|
= - |
1 e
|
( |
¶Hz ¶y
|
- |
¶Hy ¶z
|
- sEx ) , |
| (6) |
|
|
¶Ey ¶t
|
= - |
1 e
|
( |
¶Hx ¶z
|
- |
¶Hz ¶x
|
- sEy) , |
| (7) |
|
|
¶Ez ¶t
|
= - |
1 e
|
( |
¶Hy ¶x
|
- |
¶Hx ¶y
|
- sEz ) . |
| (8) |
The system of six coupled PDEs forms the basis of the FD-TD numerical algorithm.
2 The Yee Algorithm
In 1966, Kane Yee [2] introduced a set of finite-difference equations for the time-dependent
Maxwell's curl equations system for a lossless material (r = 0 and s = 0). In order
to derive the method it is
assumed that the three-dimensional region of interest is divided into a mesh
of cubic cells. Equation (2) is used to update E at each time level (n), equation (1) is used
to update H at each mid time level (n+[1/2]). The mesh used for updating E and
H is different; the edge of the cubes in one mesh lie at the centres of the cubes in the
other mesh; the cubes interlace each other.
The components of the electromagnetic field Ex, Ey, Ez, Hx, Hy, Hz are computed at different
points in the domain. The above figure shows the mesh for computing H from E with
a typical origin. The coordinates show the positioning of the electromagnetic components with respect
to that origin. The finite difference equivalent of equation (1) for computing H at
the [1/2] time level from the original time level is as follows:
|
Hx|[1/2],0,0[1/2] = A[1/2],0,0Hx|[1/2],0,0-[1/2]+ C[1/2],0,0 |
é ê
ë
|
- |
Ez|[1/2],[1/2],00 - Ez|[1/2],-[1/2],00 Dy
|
+ |
Ey|[1/2],0,[1/2][1/2] - Ey|[1/2],0,-[1/2]0 Dz
|
ù ú
û
|
|
|
|
Hy|0,[1/2],0[1/2] = A0,[1/2],0 Hy|0,[1/2],0-[1/2]+ C0,[1/2],0 |
é ê
ë
|
- |
Ex|0,[1/2],[1/2]0 - Ex|0,[1/2],-[1/2]0 Dz
|
+ |
Ez|[1/2],[1/2],0[1/2] - Ez|-[1/2],[1/2],00 Dx
|
ù ú
û
|
|
|
|
Hz|0,0,[1/2][1/2] = A0,0,[1/2] Hz|0,0,[1/2]-[1/2]+C0,0,[1/2] |
é ê
ë
|
- |
Ey|[1/2],0,[1/2]0 - Ey|-[1/2],0,[1/2]0 Dx
|
+ |
Ex|0,[1/2],[1/2][1/2] - Ey|0,-[1/2],[1/2]0 Dy
|
ù ú
û
|
|
|
where
|
A*,*,* = 1.0 ,C*,*,* = |
Dt m*,*,*
|
, |
|
and parameter r is presumed zero.
For computing E from H the following figure shows a cube from the mesh
in the region of the same origin. The following finite difference
equivalents of equation (2) may be derived
|
Ex|1,[1/2],[1/2]1 = B1,[1/2],[1/2] Ex|1,[1/2],[1/2]0+ D1,[1/2],[1/2] |
é ê
ë
|
|
Hz|1,1,[1/2][1/2] - Hz|1,0,[1/2][1/2] Dy
|
- |
Hy|1,[1/2],1[1/2] - Hy|1,[1/2],0[1/2] Dz
|
ù ú
û
|
|
|
|
Ey|[1/2],1,[1/2]1 = B[1/2],1,[1/2] Ey|[1/2],1,[1/2]0+ D[1/2],1,[1/2] |
é ê
ë
|
|
Hx|[1/2],1,1[1/2] - Hx|[1/2],1,0[1/2] Dz
|
- |
Hz|1,1,[1/2][1/2] - Hz|0,1,[1/2][1/2] Dx
|
ù ú
û
|
|
|
|
Ez|[1/2],[1/2],11 = B[1/2],[1/2],1 Ez|[1/2],[1/2],10+ D[1/2],[1/2],1 |
é ê
ë
|
|
Hy|1,[1/2],1[1/2] - Hy|0,[1/2],1[1/2] Dx
|
- |
Hx|[1/2],1,1[1/2] - Hx|[1/2],0,1[1/2] Dy
|
ù ú
û
|
|
|
where
Most simply, the complete electromagnetic domain of interest is taken to be a cuboid with Nx, Ny,
and Nz cuboids in the x,y and z directions.
From the information given by the components in this cell, Hx,Hy,Hz at the n-[1/2]th time level
and Ex,Ey,Ez at the nth time level, Hx,Hy,Hz at the n+[1/2]th time level can be computed
using the following formulae:
For i=1..Nx, j=1..Ny, k=1..Nz
|
Hx|i+[1/2],j,kn+[1/2] = Ai+[1/2],j,kHx|i+[1/2],j,kn-[1/2]+ Ci+[1/2],j,k |
é ê
ë
|
- |
Ez|i+[1/2],j+[1/2],kn - Ez|i+[1/2],j-[1/2],kn Dy
|
+ |
Ey|i+[1/2],j,k+[1/2]n+[1/2] - Ey|i+[1/2],j,k-[1/2]n Dz
|
ù ú
û
|
|
|
|
Hy|i,j+[1/2],kn+[1/2] = Ai,j+[1/2],k Hy|i,j+[1/2],kn-[1/2]+ Ci,j+[1/2],k |
é ê
ë
|
- |
Ex|i,j+[1/2],k+[1/2]n - Ex|i,j+[1/2],k-[1/2]n Dz
|
+ |
Ez|i+[1/2],j+[1/2],kn+[1/2] - Ez|i-[1/2],j+[1/2],kn Dx
|
ù ú
û
|
|
|
|
Hz|i,j,k+[1/2]n+[1/2] = Ai,j,k+[1/2] Hz|i,j,k+[1/2]n-[1/2]+Ci,j,k+[1/2] |
é ê
ë
|
- |
Ey|i+[1/2],j,k+[1/2]n - Ey|i-[1/2],j,k+[1/2]n Dx
|
+ |
Ex|i,j+[1/2],k+[1/2]n+[1/2] - Ey|i,j-[1/2],k+[1/2]n Dy
|
ù ú
û
|
|
|
where
|
A*,*,* = 1, C*,*,* = |
Dt m*,*,*
|
|
|
|
Ex|i+1,j+[1/2],k+[1/2]n+1 = Bi+1,j+[1/2],k+[1/2] Ex|i+1,j+[1/2],k+[1/2]n + |
|
|
Di+1,j+[1/2],k+[1/2] |
é ê
ë
|
|
Hz|i+1,j+1,k+[1/2]n+[1/2] - Hz|i+1,j,k+[1/2]n+[1/2] Dy
|
- |
Hy|i+1,j+[1/2],k+1n+[1/2] - Hy|i+1,j+[1/2],kn+[1/2] Dz
|
ù ú
û
|
|
|
|
Ey|i+[1/2],j+1,k+[1/2]n+1 = Bi+[1/2],j+1,k+[1/2] Ey|i+[1/2],j+1,k+[1/2]n+ |
|
|
Di+[1/2],j+1,k+[1/2] |
é ê
ë
|
|
Hx|i+[1/2],j+1,k+1n+[1/2] - Hx|i+[1/2],j+1,kn+[1/2] Dz
|
- |
Hz|i+1,j+1,k+[1/2]n+[1/2] - Hz|i,j+1,k+[1/2]n+[1/2] Dx
|
ù ú
û
|
|
|
|
Ez|i+[1/2],j+[1/2],k+1n+1 = Bi+[1/2],j+[1/2],k+1 Ez|i+[1/2],j+[1/2],k+1n+ |
|
|
Di+[1/2],j+[1/2],k+1 |
é ê
ë
|
|
Hy|i+1,j+[1/2],k+1n+[1/2] - Hy|i,j+[1/2],k+1n+[1/2] Dx
|
- |
Hx|i+[1/2],j+1,k+1n+[1/2] - Hx|i+[1/2],j,k+1n+[1/2] Dy
|
ù ú
û
|
|
|
where
Practicalities
Cell size and time step
The cells must be small enough to accurately represent the material
distribution and to give results to the detail required. However,
the 3D FDTD is a computationally-intensive method and if the cells
are small then the simulation could easily become impractical.
Generally it is advisable to keep the number of cells to the
order of 1 million on modern computers.
3 Boundary Condition
For reasons of efficiency, the FD-TD method is generally applied in a closed region
where most of the activity is taking place. The remaining region is modelled
by applying a suitable condition on the boundary of the FD-TD region. The
application of a suitable boundary condition is critically important in FD-TD.
For simplicity, let the FD-TD region be a cuboid, the boundary condition must be placed
on the six faces of the cuboid. The most common boundary condition that is applied
is the absorbing boundary condition (ABC) in which it is desired that all
outward-propagating numerical wave analogues to exit the FD-TD domain
and spurious reflections are supressed. A typical example is the one introduced
by Mur [1].
References
- [1]
- G.Mur (1981). Absorbing Boundary Conditions for the
Numerical Simulation of Waves. IEEE Transactions on Magnetic Compatability,
23, 1981, 377-382.
- [2]
- K.S.Yee (1966). "Numerical Solution of Initial Boundary Value
Problems involving Maxwell's Equations", IEEE Trans.
Antennas and Propagation, 14, 302-307.