TH2E-6

A Conducting Sheet Model for Efficient Wide Band FDTD Analysis of Planar Waveguides and Circuits
A. Lauer and I. Wolff Institute of Mobile and Satellite Communication Techniques, D-47475 Kamp-Lintfort, Germany, e-mail lauer@imst.de

Abstract-A simple wide band equivalent circuit for the surface as impedance of conducting sheets is introduced into the three ditanh((1 + j ) & ) H s mensional Yee FDTD scheme. The model is based on the plane Y o = 2Ys = 2- = G skin effect, thus the frequency dependence of losses and of the ES (l+j)a (1) inner inductivity is included. Stability considerations are pre0 8 1 Cl=--, WO=--, Goo=-=Kt. sented as well as numerical results for the attenuation coefficient WO pot2 R0,o of microstrip and coplanar waveguides in comparison to referHerein t is the sheets thickness and CJ is its conductivity. ence data.
7

I. INTRODUCTION

A second order rational approximation2 for real value arguments x yields

x f(x) = -

tanhfi

mi:
W . 1
~

Metallic losses in microwave waveguides and circuits cannot fi (2) be treated efficiently using straight forward FDTD simula0.00037835 x2 + 0.089376 x 0.999241 tions, since the skin effect is much smaller and faster than qx) = 0.00936184~ +0.419520x+ ~ 1 the wave propagation itself. Thus the simulation time would be increased by magnitudes. the approximation interval was [0,80], the maximum relative Therefore thin conducting sheet approximations have been in- error was estimated to be 0.076%. troduced into the Yee [l] FDTD scheme, which work well in For the needed complex3 = mono-frequent and narrow band applications [2]-[4]. (1 + j)2i2, the relative error This contribution presents a conducting sheet approximation - h(x)l/lf(x)l is less than suitable for wide band FDTD simulations, which properly 3 % for Cl < 10, less than 6 % models losses as well as the inner inductivity under assump- for R < 30 and less than 10 % tion of the plane skin effect. for R < 50. In section I1 a simple approximative wide band equivalent cir- Figure 1 shows an equivalent cuit consisting of two inductors and three resistors is proposed circuit proposed to realize yo *I Yl = Go,~h(Cl). As seen in for modeling the sheets surface impedance. Equivalent Circuit for the Section I11 shows how this circuit is introduced into Yees section 111, this circuit can be Fig. second order rational surface admitFDTD scheme, yielding the need of only two additional easily introduced into the Yee tmCeapproximation 2ys = % y,, nodes per tangential electric node on the conducting sheet. scheme. Stability considerations follow in section IV. A formulation, that does not Numerical results for the attenuation coefficient of microstrip depend on the sheets parameters K and t , is and coplanar waveguides are given in section V and compared 1 1 h((l+j)a) =g+-+--, ri +JRIi r2+J012 (3) to those of measurements, static two dimensional FD calculations [5] and an approximate formula for MSL ohmic losses [61. with dimensionless

).(fI

CIRCUIT FOR THE SURFACE 11. A SIMPLE EQUIVALENT IMPEDANCE OF THINCONDUCTING SHEETS

li = Go,o W L i ,

Ti

= Go,oRi7

g=-

G0,o

i E {1721.
(4)

Assuming a plane skin effect, the surface admittance Y, (rsp. of the ,.numapprox.. package 20btained using the ,,minimax,, the admittance of a square Part of the sheet yo) can be written the symbolic algebra program MAPLEV(TM). Optimization criterion w i l ~
minimization of the maximum relative error. For a 2 pm thick Au sheet wi)/(2K) is approx. 6 GHZ. Yielding Corresponding to the currents in the two inductors of the equivalent cirup to a frequency of 60 GHz. cuit.

< 10

1589
0-7803-5135-5/99/$10.00 0 1999 IEEE 1999 IEEE MTT-S Digest

Equation (3) can be solved analytically, resulting in

rl = 7.482301652,
11

with
Ev=

r2
12

= 1.211859662, = 0.959481195,
(5)

1 (l+E++E3-+E-E+)&, 1 4 1

= 0.353893318,

g = 0.040414818.

0 : = - (1

+<++E- +&E+)
+ + +

0E,

111. FDTD IMPLEMENTATION OF THE EQUIVALENT CIRCUIT

TOobtain an equivalent circuit for the FDTD method itself, the Yee scheme for inhomogeneous material with electric and magnetic losses can be Written in a short Operator formulation 171, 181. Figure 2 shows the unit cell of the well known Yee FDTD scheme. Field components as well as the cell sizes and the (Cv& Gv)OF = (5+ - I)@! - (5- - I)@:, material properties are functions of the discrete position X = (8) ( i ,j , l ~ )Field ~ . components and other properties belonging to (Lilt RV)Oy= - l)O! - (5- - I)@:, the cell X are shown in the figure. For simpler writing, component and direction indices are numbered in a modulo three with sense, e. g. Eo E E,, El E y , E2 3 Ez and again E3 z E, A~A! A ~ A ~ and E-1 Ez. Upper indices E rs . H are used for the elecc v =.-+Ev, Gv = * 0 : , tric rsp. ma netic conductivity 0 rsp. gH and for the sub 4 (9) AEA! cell sizes AI?and AH. A lower index "+" m e 9 s v 1, a "Lv =+Pv, Rv = * 0 : . AV " stands for v - 1. The shift operators tvand are defined by E,,Ef,(X) = H,(X f T") and &Jl,(X) = H,(X - G),with Figure 3 shows an electrical field node of the equivthe unit vector in v-direchon !&. dent circuit belonging to (8): To implement the conducting sheet approximation, all electrical nodes belonging to tangential field components in the conductor are modified as shown in figure 4. @t B o additional nodes @,,I and Ov,2, which are treated like magZfOH netic nodes in the time stepping scheme, are introduced. The 'parasitic' capacitor Cv is needed to C" leave the time stepping algorithm unchanged, its value (discussed in Fig. 3. Equivalent circuit for section IV) can be chosen very ~~~~~~. node in the small, thus it does not disturb the surface admittance approximation. Since 0: = EVA: and Ofl, = HAfl, for the tangential magnetic field, the proper parameters to realize the conducting sheet's surface admittance Y1 are

(7) PV = ; i(1 + 5+ + 5- + 5-5+)P, 1 0 ; = - (1 5+ 55-5+) OH. 4 An equivalent circuit formulation for (6) is proposed in [71. A more general formulation, suitable for graded mesh FDTD schemes, requires magnetic and electric voltages to be introduced to maintain symmetry of the equivalent circuit. With 0: = EVA: and 0: = HvAf equation (6) yields

E+

+ziz

r'

AF:

Fig. 2. Yee cell at the position X with the field components and parameters accounted to this cell.

Gv = z G o , o g, " ;
1 =11, Go,o 00 1 Lv,2 = 12, Go,o 00
&,I

G I

&,I

1 = -rl,
G0,o

(10)

The time continuous Yee scheme for lossy material and graded mesh discretization is
(&vat

Rv,~ = -r2.
G0,O

1 + 0,")Ev = H (b' - 'IHA+

- -(b - l)H+>
4The summation of the OH can be done using four gyrators in parallel

(pvat +O,H)H

1 - -((5+ - 1)E- - -(C- AY A !

1 A ! 1 -

- 1)E+

(6) [7], in this case, the OH-nodes have the same structure as the OE-nodes. An
alternative is to use four 1:l transformers, in this case, OH is implemented as a current.

1590

V. NUMERICAL VALIDATION

For the empirical validation of the approximation, a microstrip and a coplanar waveguide, for which reference data is available, have been simulated to obtain attenuation coefficients. Both are on a 250 pm A1203 substrate, the conducting Au (0 = 41 MANm) films are 5 pm thick. The microstrip lines width is 225 pm, yielding a characteristic impedance of approx. 53.2 SZ. The coplanar waveguide has an inner conducFig. 4.Modified equivalent circuit for a tangential electrical field node in the tors width of w = 125pm, the distance between the inner conducting sheet. sides of the outer conductors is d = 225 pm, the characteristic At the modified nodes, @,,I and OV,2 are subtracted from the impedance is approx. 48.6 R. right side of (8). The newly introduced nodes are treated as A 1 = 10 mm long part of the waveguides was used for the simulation. Since a very high accuracy is needed for the simulation of losses < 0.1 dB, For the discretization in time, one of the well known stepping schemes for lossy FDTD cells can be used. At the edges of the conducting sheets, G, has to be halved as well as L,J, Lv,2 and & , I , Rv,2 are to be doubled for proper operation.
0

A Gaussian pulse modulated with a sinus function has been used for excitation in an electric wall. It has been observed, that an unmodulated Gaussian pulse excites small static fields, which disturb the discrete Fourier transform needed for postprocessing. Approx. 5 mm and 15 mm away from the exciting wall, the voltages on the transmission lines have been recorded for attenuation calculation.
No absorbing boundary conditions have been used. The waveguides total lengths have been chosen big enough, so that the reflected pulse does not disturb the simulation.

CONSIDERATIONS IV. STABILITY As shown in [8] a stability criterion for the unmodified Yee scheme can be written as

If the left hand side of (12) is interpreted as an estimate for the maximum angular resonance frequency of the FDTD scheme, an equation for C, of the modified cells can be obtained,

The use of (13) to calculate C, has two advantages: 1. The time step At remains unchanged.

2. In practical applications, oC, is in the range from AN, whereas IYmJ > 10 AN, e.g. the AN to
parasitic capacitor does not influence the accuracy of the simulations.

The MSL was discretized using 10 cells for both, the conductors width and the substrate height. The backside metalization was simulated as a conducting sheet, too. The CPW waveguides inner conductor was discretized using 11 cells, the gap with 5 cells. The substrate was discretized with an eight cell graded mesh. Figure 5 shows a comparison of the approximative surface admittance Yl obtained from voltage and current FDTD results in a microstrip waveguides backside metalization and the theoretical results from (1). Obviously the conductance as well as the inner inductivity are very accurately modeled in the whole frequency range. The Figures 6 and 7 compare the simulated attenuation with measurements, static FD results and an approximation for MSL losses. The FDTD results do agree very well with the reference data over the whole frequency range. Radiation losses have not been taken into account in the FDTD simulations.

VI. CONCLUSIONS It is also possible to eliminate the capacitors from the modified nodes, but in this case the time step possibly has to be A wide band FDTD model based on a simple equivalent circuit has been presented for conducting sheet modeling. The adjusted to ensure stability.
1591

sheets conductance as well as its the inner inductivity, depending on the frequency due to the skin effect, are properly modeled. Excellent results have been obtained simulating lossy microstrip and coplanar waveguides. Because the approximation proposed here has almost no impact on computation time and computer memory consumption, it is suitable for the simulation for most passive planar microwave circuits.

9
0.2

T 3
v

REFERENCES
[I] K. S . Yee, Numerical solution of initial boundary value problems in[2] [3] [4]

8
I

0.1

[5]

[6] 171

[8]

volving Maxwells equations in isotropic media, IEEE T r m . Antennus Prupugut., vol. AP-14, pp. 302-307, 1966. J. G.Maloney and G. S . Smith, The efficient modeling of thin material sheets in the finite-differencetime-domain (FDTD) method: IEEE Truns. Antennus Propugut., vol. AP-40, pp. 323-330, 1992. 0 . 0 L.-K. Wu and L.-T. Han, Implementation and application of resistive 0 10 20 sheet boundary condition in the finite-difference time-domain method: f(GHz) -+ IEEE Truns. Antennus Propugur., vol. AP-40, pp. 628-633, 1992. J. G.Maloney and G. S . Smith, A comparison of methods for modeling electrically thon dielectric and conducting sheets in the finite-difference fig. 6. Attenuation -IS21 I of a 1 = 10 mm long microstrip waveguide (E, = time-domain (FDTD) method, IEEE Trans. Microwme Theory Tech., 9.8, h = 250 pm, w = 225 pm, t = 5 pm, (J = 41 MANm) in dependence of the frequency f . The thick solid line represents the FDTD results, the thin vol. MTT-41, pp. 690494, 1993. G. Kibuka, R. BertenbuS, M. Naghed, and I. Wolff, C o p l a n ~ lumped solid line is measured using open circuits of two different lengths, the other elements and their application in filters on ceramic and gallium w- line is obtained using an approximativeformula [6]. The FDTD results do not senide substrates: in 19th Europeun Microwave Cunference Pmceed- start at a frequency of zero, because a modulated Gaussian pulse was used for excitation. ings, pp. 656-661, 1989. E. Hammerstad and F. Bekkadal, Micrustrip Hundbook, vol. STF 44 A 74169 of E U B report. I. J. Craddock, C. J. Railton, and J. P. McGeehan, Derivation and npplication of a passive equivalent circuit for the Finite Difference Time Domain algorithm, IEEE Micruwuve und Guided Wave Lett.. vol. 6, pp. 4 0 4 2 , 1996. A. Lauer and I. Wolff, Stable and efficient ABCs for graded mesh fdtd simulations, in IEEE MTT-S Int. Micruwuve Symp. Digest, pp. 461464.1998.

200
160

T
I

T 1
W

120

s
-c

80
40

o
-40 -80
v.v

~~

10

20 f(GHz)

30

40

10

20

f(GHz)

Fig. 5. Real and imaginary P a r t of the admittance Yo in dependence of the frequency f . The solid lines represents the approximation obtained from voltage and current FDTD results in a microstrip waveguides backside metalimtion (t = 5 pm, o = 41 MNVm), the thin line shows the exact results from the plane skin effect analysis. The FDTD results do not start at a frequency of zero, because a modukuted Gaussian pulse was used for excitation.

Fig. 7. Attenuation -IS211 of a 1 = 10 mm long coplanar waveguide (E, = 9.8, h = 250 pm, w = 125 pm, d = 225 pm,t = 5 pm, o = 41 MNVm) in dependence of the frequency .f. The thick solid line represents the FDTD results, the thin solid line is measured using open circuits of two different lengths, the other line is obtained using a static Finite Difference method in conjunction with an ohmic loss approximation. The FDTD results do not start at a frequency of zero, because a modulated Gaussian pulse was used for excitation.

1592