Beruflich Dokumente
Kultur Dokumente
&
y
r
(6)
Substituting these variables into the Maxwell's equations,
we can obtain the 'a' parts of the feld components fIst and
then use them to calculate te remaining components.
During the simulation, we save the transfoned fled
variables at all time steps, and by perfoning inverse Fourier
transfon on (5) and (6), we can obtain the original feld
variables fom transfoned ones in te time domain.
B. Spectral FDTD Formulation
The Spectral FDTD approach is one of direct feld
periodic FDTD methods. Instead of splitting fled
components, it applies PBC by using a constant transverse
wavenumber (CTW) wave instead of conventional single
angle incident wave
[
5
][
61
.
The fequency domain of CTW is defmed as:
Et = exp(Jxx )exp(Jyy )exp(Jz (z -zo))
exp( -
k
g / )exp( mt
g
)
(7)
HI
= exp(Jxx )exp(yy )exp(z (z - Zo ))YTE
(8)
exp( -
k
g / )exp( mt g)
(9)
Where ko = 27/e , and kx ' k
y
represent constat
transverse wavenumbers, exp(- kg / (2) corresponds to a
Gaussian pulse used to limit the badwidth of wave
whileexp(Jlfo )is used to shif the Gauss pulse in te time
domain. Te CTW is calculated by applying inverse Fourier
transfor on (7) and (8) and is introduced into the
computational area using total-feldlscatter-feld formulation.
The feld components update follows standad FDTD
leapfog scheme ad standard UPML layers are implemented
as ABC.
C Frequency-Angle Mapping Formulation
Both Split-Field FDTD and Spectral FDTD are wide-band
approaches, their primary distinction lies in the fequency
angle mapping system. Since Split-Field performs additional
feld-splitting for phase compensating throughout the grid
and uses standard Gaussian plane wave as incident wave, a
fll range of fequency response coresponding to one
incident wave angle can be obtained by a single r of FDTD
program.
Unlike the conventional FDTD approach, the incident
wave in Spectal FDTD is CTW wave defmed as a time
domain wave with trasverse wavenumbers(independent of
fequency).As is shown in Fig. 1,
y
x
Fig.1 Multiple incident angles carried by a CTW wave
The CTW has constant transverse wavenumbers (kx ,k
y
)
and it caries a continuous range of incident agles, each
angle corresponds to a fequency component with
relationship by fowing equations:
B=sin-1(kj/ko) ( 10)
r= tan-l (k
y
/
kx
),
k
}
=k
; +
k
As is shown in Fig2, by changing the value of kj' both
angel and fequency can scan within the desired range, which
dbb
makes it possible to obtain multiple agle and fequency
responses in a mush fewer simulation counts compared to
single fequency approach. For example, in order to simulate
o 15G fequency response at incident angle B = 30
of a
periodic structure, we repeat simulation Z3 times with
kx = 2; sin(30
0
) / ee,
( 1 1)
f = 15 ^ 10
9
(n -1)/25,n = 1,2 .. .25,ee = 3 ? 10
Ib6lS
Z
O C
. b. T. Tb.
cc
Fig.2 Angel-Frequency Curves obtained by 25 times of simulation
III. NUMRCAL RESULTS AD COMASON
A. numerical results
The frst periodic structure being aalysed here consists of
simple straight diploes arrayed in a fashion of rectangular
grid. The inter-element spacing is 17.8mm, Dipole Length
is12.7mm, Dipole Width is1.27mm. We apply both Split
Field FDTD and Spectral FDTD method in the simulation
and compare the results with result obtained by CST M
Studio. Te size of mesh cells is 56x 56x 100 in both Split
Field FDTD ad Spectral FDTD. The time and space costs of
both methods ae listed in Tablel ad results ae plotted in
Fig.3. As can be seen in Fig.3, results of Split-Field FDTD
and Spectral FDTD show good agreement with CST result,
which verif te accuracy of both methods.
TALE 1 TNPbPL LLb LL NMLb
Time Time Time Memory Disk Space
Steps Cost Cost/Step Cost Cost (MB)
(Min.) (Min.) (MB)
Split- Z 1+ .3J ++J 3JV
Field
Spectral +V + .Z1 J3 +
gg ..........................
gg _ . _ . _---,---------. --------
gy ........ ........., ....... .
U
>
........ , ......... , ........
g
.
..........................
g| ..........................
g] ........ ........., ...... ..
g_ ....................... ..
g
_
_._._---.---------.--------
C5TM5tudo
5pectrzll1
5plt-ledI1
-----------------_ .' ._----- _. ! . _------
, , ,
'. -- -- '. - -- -----... --. -. ----. -_. -.'--
"g
_
>
g
_
_
_
|
_
>
_
g
_g
Frcqucucc/(GHz)
Fig.3 Power refection coefcient of a dipole array
To illustrate the fequency-agle mapping system of
SFDTD, we then simulate a periodic structue consists of
square patches. The inter-element spacing is 1 Om and square
size is Scm ? Scm. To calculate fequency response at J
degree, we var k _ fom 0 to 66.72 , as can be seen in FigA
and Fig.5, trough proper interolation, fequency response
other than 0 degree and 30 degrees are obtained as well.
,
l --Incident Angle
--
r
----------l-----------
r'
--
Incident Angle=lO
g
g .......................... .
gg .--.... ---.... ---- ... -->-...
----------:-.--.-------: --- - ---
j g.
E
U.>
_
g
.|
t
.
U
]
4
---------- -----------.--- ----
-----_._--.-_ .. _--_ ..-. -- . ---
.......... ' .............. .. .. .
. . . _ _ .
.
. _ -: - .. _ .. _ _ . . .. : .... -
g_ _.........
... _.... ..
, . , ,
, , , ,
, . , .
, . , .
_____ .' ____ ..- ----- - --------_.' - -'" ------- '. -.', ---------
, . , .
, . , .
, . , ,
, . , ,
, . , ,
, , , ,
, . , .
-----------r-----------}-----------: ------------: ----------
, , ,
, , ,
, , ,
, . ,
, , ,
, , ,
, , , ,
--------- --,--- ------ --r---- ------- ,----- ------,----------
, , , ,
, , , ,
, , , ,
, , , ,
, , ,
, , , ,
, , , ,
----------
:
-----------,........ ....----------
:
----------
, , , ,
, , , ,
, , , ,
, , , ,
, , , ,
, , , ,
- - - _ < .............. ......... '
, , ,
, , , ,
, , , ,
, . , .
, . , .
, . , .
, , , ,
--- r -- --; --: - : ----. .
, , ,
, , ,
, , ,
, , ,
, ,
, , ,
, , , ,
------,-----------r----------,-----------.----------
, . , .
, , , .
, , , ,
, , ,
, , ,
, , , ,
, . , .
g_ .......... , . ..... . . .. . . .. - ... .. -r--- . ---. --
, , , ,
"
g
--
--
--
>
-
g
Frcqucucy/(GHz)
Fig.4 Power refection coefcient of a square patch aray with incident angle
0degree ad 10 degrees
dbb
.9
.b
' 0.7
L
G
.I
.
_ .4
.J
.
-Incident Angle=20o
-Incident . AgleJ"
, ,
----------- _ ---------- ------------,------------->-----------
, ,
, ,
. . _. .. _. .. .......... _.: .. _ ..._... _ . .
:
...... .... .
, ,
, ,
, ,
, ,
, ,
, ,
, .
--------- ------------,------------.-----------
, ,
-------- ------------.------------. -----------
, ,
, ,
, ,
, ,
, ,
, ,
, .
- - ------- -----------_._----------_ . _----------
, ,
, ,
, ,
------- --_._-_._-_.'--_._--------,_._-_._-_.-
, ,
, ,
, ,
- . __ .. - ._' .. .. ., ............ ., .......... .
, ,
, ,
, ,
, ,
, ,
-- _ ._- . . . _ _ . . _ _ .. -
.-
............
Z +
cc Hz
D
Fig.S Power refection coefcient of a square patch array with incident angle
20 degrees ad 30 degrees
B. Comparison of to methods
Furtermore, we compare the simulation speed and
resouce consuption of the two methods based on Table 1.
1) Stabilit and Simulation Speed
Since SFDTD follows standad FDTD scheme, it inherit
the spatial and time sample density requirements of
conventional FDTD.
While Split-Field FDTD shares the same stability
requirements in spatial sampling, due to feld variables
splitting, the time sampling requirement of Split-Field FDTD
is stricter tha Spectral FDTD and is relevant to incident
angle of the plane wave.
Due to feld variables splitting, split-fled FDTD takes up
more time and memory in a single r and is less stable
compared with sided when dealing with same problem,
especially in the case of grazing agle since simulation time
of split-FDTD increases in proportion of incident angle while
SFDTD's consumption remains constant.
2) Resource Consumption
Meanwhile, in the Split-Field FDTD, time domain data of
all time steps need to be stored for time shifing operations to
obtain the original feld data. Although we can pre-calculate
the time shif value for individual grids ad store the feld
variables of adjacent time steps during the simulation, it will
greatly increase the memory usage and slow down te
overall speed. On the other hand, since SFDTD is a direct
feld scheme, data of much fewer time steps are needed for
storage compared to Split-Field FDTD, which saves a
considerable amount of storage space ad processing time.
IV. CONCLUSION: