This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY 1

Causal RLGC(f ) Models for Transmission

Lines From Measured S -Parameters
Jianmin Zhang, Senior Member, IEEE, James L. Drewniak, Fellow, IEEE,
David J. Pommerenke, Senior Member, IEEE, Marina Y. Koledintseva, Senior Member, IEEE,
Richard E. DuBroff, Senior Member, IEEE, Wheling Cheng, Zhiping Yang, Senior Member, IEEE,
Qinghua B. Chen, Member, IEEE, and Antonio Orlandi, Fellow, IEEE

Abstract—Frequency-dependent causal RLGC(f ) models are model to represent a transmission line. Though many cross-
proposed for single-ended and coupled transmission lines. Dielec- sectional static and quasi-static numerical tools are able to eval-
tric loss, dielectric dispersion, and skin-effect loss are taken into ac- uate R, L, G, and C parameters for transmission lines, the condi-
count. The dielectric substrate is described by the two-term Debye
frequency dependence, and the transmission line conductors are of
tion is that the causal dielectric properties of the substrate materi-
finite conductivity. In this paper, three frequency-dependent RLGC als must be known prior to using these tools. For accuracy, the R,
models are studied. One is the known frequency-dependent ana- L, G, and C parameters must be evaluated at different frequen-
lytical RLGC model (RLGC-I), the second is the RLGC(f ) model cies with different values of complex permittivity, which is cum-
(RLGC-II) proposed in this paper, and the third (RLGC-III) is bersome for a wide frequency range of interest. In addition, these
same as the RLGC-II, but with causality enforced by the Hilbert cross-sectional tools are electrostatic field solvers in nature, and
transform in frequency domain. The causalities of the three RLGC
models are corroborated in the time domain by examining the
therefore, they may be suitable only for finding R, L, G, and C
propagation of a well-defined pulse through three different trans- parameters at low frequencies, where quasi-static approximation
mission lines: a single-ended stripline, a single-ended microstrip is still valid, while accuracy at higher frequencies is lost. Further-
line, and an edge-coupled differential stripline pair. A clear time- more, the frequency-dependent permittivity properties of a par-
domain start point is shown on each received pulse for the RLGC-II ticular PCB substrate (typically, of flame retardant (FR)-4 type)
model and the RLGC-III model, where their corresponding start are usually not known exactly. This is because dielectric proper-
points overlap. This indicates that the proposed RLGC(f ) model
(RLGC-II) is causal. Good agreement of simulated and measured ties for PCBs vary in a substantial range, depending on process
S-parameters has also been achieved in the frequency domain technology and constituents contained in the substrates even for
for the three transmission lines by using the proposed frequency- the same PCB manufacturer. To develop a frequency-dependent
dependent RLGC(f ) model. causal RLGC(f ) model, not only both the dielectric loss and the
Index Terms—Causality, dielectric materials, Hilbert trans- conductor loss must be taken into account, but also the dielectric
forms, scattering parameters, transmission line modeling. dispersion. However, it is very difficult or even impossible to find
appropriate data in literature or in manufacturer’s databases that
I. INTRODUCTION could be applicable for describing a particular PCB in terms
of dielectric frequency dispersion, e.g., using a Debye or a
IGNAL integrity analysis and channel modeling in high-
S speed digital systems are becoming more and more im-
portant as operating frequencies increase. When the on-board
Lorentzian model [1]. If the dielectric dispersion is not taken
into account and the dielectric is described by a constant real
permittivity with a constant loss tangent, the dielectric response
frequencies are above hundreds of megahertz, or especially in turns out to be noncausal. Indeed, real and imaginary parts of
the gigahertz range, traces on printed circuit boards (PCBs) no dielectric permittivity ε(ω) in any physically realizable linear
longer behave as simple conductors, but exhibit high-frequency dielectric medium are not independent of each other. They must
effects, and behave as transmission lines. Accurate models to satisfy the Kramers–Kronig relationships [2], which are analo-
simulate high-frequency effects, such as dielectric dispersion, gous to Hilbert transforms for any passive linear filter [3], [4]
skin-effect loss, and crosstalk, become critical, so it is desir-
able to obtain an accurate frequency-dependent causal RLGC(f )   ∞
 1 εim ag (x)

 ε (ω) = P dx + 1
 real π −∞ x − ω
Manuscript received April 11, 2009; revised September 10, 2009.
 ∞ (1)
J. Zhang, Z. Yang, and Q. B. Chen are with Cisco Systems, Inc., 
 1 εreal (x) − 1
San Jose, CA 95134 USA (e-mail: [email protected]; [email protected];  εim ag (ω) = − P
 dx
[email protected]). π −∞ x−ω
J. L. Drewniak, D. J. Pommerenke, M. Y. Koledintseva, and R. E. DuBroff
are with the Missouri University of Science and Technology (formerly Univer-
sity of Missouri-Rolla), Rolla, MO 65409 USA (e-mail: [email protected]; where P denotes the Cauchy principal value that expands the
[email protected]; [email protected]; [email protected]).
W. Cheng is with Juniper Networks, Sunnyvale, CA 94089 USA (e-mail:
class of functions for which the integral exists. As follows from
[email protected]). (1), εreal (ω) can be determined from εim ag (ω) at any particular
A. Orlandi is with the University of L’Aquila, L’Aquila, I-67010, Italy frequency if εim ag (ω) is known over the entire frequency range,
(e-mail: [email protected]). or vice versa. Hence, an RLGC model of a transmission line can
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. be causal only if the properties of the surrounding dielectrics
Digital Object Identifier 10.1109/TEMC.2009.2035055 have causal representations.

0018-9375/$26.00 © 2009 IEEE

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY

A lossy transmission line RLGC model was first proposed dipoles under an alternating field was modeled by Gd [13]. The
for coaxial cables, where there is no radiation loss. In the early p.u.l. capacitance of the RLGC model defined in [12] is constant
studies of loss for transmission lines, only skin-effect loss was over the entire frequency range of interest. Thus, the model
considered, while dielectric loss was neglected [5]. Wigington given in [12] is noncausal. However, using the above-mentioned
and Nahman √ [5] had shown that the simple skin-effect loss model, good agreement between measured and simulated S-
followed a f law, and later, Nahman [6] presented a transient parameters has been reported in [14], while causality has still
model of a lossy cable by taking into account both the skin- remained a problem to be fixed.
effect loss and the dielectric loss. Nahman’s representation of In this paper, a frequency-dependent causal RLGC(f ) model
the dielectric loss was (RLGC-II) is proposed and detailed in Section II-A for single-
ended transmission lines. Section II-B describes how to ob-
G (f ) = Af ε (f ) (2)
tain the unknowns for this causal model from measured S-
with a geometry-related constant A and frequency-dependent parameters using an optimization technique, genetic algorithm
permittivity ε(f ). Nahman also theoretically discussed causal (GA), and how to enforce causality in a single-ended line RLGC
responses from the point of view of cable loss. However, no model (RLGC-III). For a coupled transmission line, frequency-
causal RLGC model was developed at that time. Instead, his dependent causal RLGC(f ) models including RLGC-II model
interest was focused on a graphical transient analysis technique. and RLGC-III model (causality enforced) are presented Sec-
In later studies [7], [8], Nahman and coworkers extended the tion II-C. Based on the above-mentioned models, three cases:
skin-effect loss model to higher frequencies. Arabi et al. [9] im- 1) a single-ended stripline; 2) a single-ended microstrip line;
proved the skin-effect
√ loss model by adding a nonlinear phase and 3) an edge-coupled differential stripline, are studied in
term (Ro /(2L f ) versus 1) to the total phase in the transmis- Section III. Time-domain causality examination and frequency-
sion line propagation term e−γ l to take into account the phase domain S-parameter comparison between the RLGC circuit sim-
effects due to the inductance variation of the transmission line at ulations and the corresponding measurements are also presented
higher frequencies. In the aforementioned nonlinear phase term, in Section III. Conclusions are summarized in Section IV.
Ro is the per-unit-length (p.u.l.) resistance determined by the
transmission line structure and L is the transmission line p.u.l.
inductance. No dielectric loss model was formulated in [9]. II. MODEL DEVELOPMENT AND PARAMETER EXTRACTION
Svensson and Dermer developed a lossy RLC model, where
both the dielectric loss and the skin-effect loss were taken into The causal RLGC(f ) model (RLGC-II) proposed in this paper
account [10]. A physical relaxation model used in their study to is derived from the analytical model given in [12] by using a
describe the dielectric loss is given by causal dielectric representation to solve the noncausal phenom-
 τ2 ena. Parameters (unknowns) in the proposed model are found
a/τ from measured S-parameters using a GA. In reality, the prac-
ε = ε1 + dτ (3)
τ 1 1 + jωτ tical dimensions of a transmission line differ from its nominal
where τ is the relaxation time, a is the “relaxation strength,” and dimensions due to manufacturing tolerances in PCB fabrication
ε1 is the dielectric permittivity without the contribution from and due to the surface roughness. The latter impacts the surface
relaxation term. The integrand in (3) is a Debye term [11], but resistance of the transmission line and can be taken into account
after integration, permittivity is different from the Debye model. as causing an equivalent decrease of nominal conductivity of
It is important to note that the parameters in the relaxation model transmission line conductors in the frequency range of interest.
are unknown, and it is not easy to find them. Since nominal To account for these nonideal effects and improve the model
geometrical dimensions in [10] were directly used for the skin- accuracy, the known nominal conditions are used to estimate
effect loss calculation, the accuracy of the developed lossy RLC initial parameter ranges in the GA parameter search, instead of
model was limited by manufacturing tolerance, such as trace using them to directly calculate some of the unknowns. This is
width, trace thickness, substrate dielectric thickness, as well as especially beneficial, when a model is developed from measure-
conductor surface roughness. ments (e. g., from the measured S-parameters), as those nonideal
An analytical frequency-dependent RLGC model was intro- effects have already been incorporated in the measurements.
duced in the high-optimized simulation program with integrated
circuit emphasis (HSPICE), where dielectric loss, conductor
loss, and high-frequency phase shift due to inductance variation A. Frequency-Dependent Causal RLGC(f ) Model and
were taken into account [12]. This is the RLGC-I model men- S-Parameter Representation for Single-Ended
tioned in the Abstract. The dielectric loss and the conductor loss Transmission Lines
were modeled as The lossy transmission line RLGC model based on [12] was

G (f ) = G0 + f Gd developed in [14], and good agreement was obtained between
√ (4) simulated and measured S-parameters in the frequency do-
R (f ) = R0 + f (1 + j) Rs
main. However, the noncausal phenomena have remained. As
where Ro and Rs are the p.u.l. dc resistance and skin-effect mentioned in Section I, the constant p.u.l. capacitance and the
resistance, respectively. In (4), Go was used to model the shunt constant dielectric loss assumed in the model imply that the
current due to free electrons in imperfect dielectrics, and the dielectric representation is noncausal. To fix the problem asso-
power loss due to the dielectric polarization and rotation of ciated with the noncausal dielectric representation, the two-term

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: CAUSAL RLGC(f ) MODELS FOR TRANSMISSION LINES 3

Debye model is used in the proposed causal RLGC(f ) model are summarized as follows:
   √
εs1 − ε∞ εs2 − ε∞ 

R (f ) = R0 + f (1 + j) Rs
ε (ω) = ε∞ + + ε0 (5) 

1 + jωτ1 1 + jωτ2  L (f ) = L∞
(12)

 G (f ) = G0 + 2πf Kg εrim ag ε0
where ε∞ is the high-frequency relative permittivity, ε0 is the 

permittivity in free space, εs 1 and τ 1 are the static dielectric  C (f ) = Kg εr ε0 .
real
constant and the relaxation time constant of the first-term De-
bye components, while εs 2 and τ 2 correspond to the second The imaginary part included in the frequency-dependent re-
Debye term. As reported in [15], the two-term Debye model sistance term can be considered as an effective inductance using
can successfully describe dielectric properties for FR-4 dielec- the following transform:
tric substrates up to 20 GHz. By separating the real and the Rs
imaginary parts, (5) can be rewritten as L (f ) = √ . (13)
2π f
 
εs1 − ε∞ εs2 − ε∞ By adding (13) to the p.u.l. inductance L(f ) and subtract-
ε (ω) = ε∞ + + ε0
1 + (ωτ1 )2 1 + (ωτ2 )2 ing it from the p.u.l. resistance R(f ), the proposed frequency-
  dependent causal RLGC(f ) model is
(εs1 − ε∞ ) τ1 (εs2 − ε∞ ) τ2
− jω + ε0  √
1 + (ωτ1 )2 1 + (ωτ2 )2  R (f ) = R0 + f Rs

 √
= εrreal ε0 − jεrim ag ε0 (6) 
 L (f ) = L∞ + Rs / 2π f
(14)

 G (f ) = G0 + 2πf Kg εrim ag ε0
where 

  C (f ) = Kg εr ε0 .
ε s 1 −ε ∞ ε s 2 −ε ∞ real
 r
 εreal = ε∞ + 1+(ω τ 1 ) 2 + 1+(ω τ 2 ) 2
(7) Causality of this model will be further examined.

 εrim ag = ω (ε s 1 −ε ∞ )τ 1
+ (ε s 2 −ε ∞ )τ 2
. The propagation constant for a piece of transmission line with
1+(ω τ 1 ) 2 1+(ω τ 2 ) 2
the p.u.l. parameters given in (14) is [16]
The frequency-dependent loss tangent is then calculated as a
ratio of imaginary and real parts of the permittivity γ= (R (f ) + j2πf L (f )) (G (f ) + j2πf C (f )) (15)

εrim ag and the characteristic impedance of the line is [16]

tan (δ)f = . (8) 
εrreal (R (f ) + j2πf L (f ))
Zc = . (16)
The frequency-dependent p.u.l. capacitance can then be eval- (G (f ) + j2πf C (f ))
uated using
The ABCD matrix for a uniform transmission line piece with
C (f ) = Kg εrreal εo (9) length l can be calculated as [17]
   cosh (γl) Zo · sinh (γl)

where Kg is a geometry-related constant. The p.u.l. shunt con- A B
   
ductance due to the dielectric loss is calculated via ABCD =  = 
.

G (f ) = G0 + 2πf C tan (δ)f (10) C D sinh (γl) /Zo cosh (γl)
(17)
where Go is the shunt conductance at dc due to free electrons Finally, the S-parameters for a single-ended transmission line
in imperfect dielectrics. Substituting (8) and (9) into (10), the are obtained from the ABCD matrix as [16]
frequency-dependent p.u. l shunt conductance is rewritten as
   A +B /Z 0 −C Z 0 −D 2(A D −B C ) 
S 11 S12 A +B /Z 0 +C Z 0 +D A +B /Z 0 +C Z 0 +D
G (f ) = G0 + 2πf Kg εrim ag ε0 . (11) =  (18)
S21 S22 2 −A +B /Z 0 −C Z 0 +D
A +B /Z 0 +C Z 0 +D A +B /Z 0 +C Z 0 +D
The p.u.l. inductance L∞ of a transmission line at high fre-
quencies depends on its cross-sectional geometry and perme- where Zo is the reference impedance.
ability of the surrounding material. For a given transmission
line, an assumption that the p.u.l. inductance L∞ is constant
B. GA Implementation and Causality Enforcement for a Single-
at high frequencies is reasonable, since dielectric materials are
nonmagnetic. Ended Transmission Line
The frequency-dependent resistance of the conductor, approx- The frequency-dependent RLGC(f ) model (14) is potentially
imated as in (4), seems reasonable, since good agreement be- causal. However, its further examination is necessary, because
tween simulations using (4) and measured S-parameters was the R(f ) term in the model (12) is known to be approximate.
demonstrated in [14]. This approximation is inherited in the In order to check the causal characteristics of the model, two
causal RLGC(f ) model proposed here. Based on the earlier dis- different methods are used in the GA optimization for find-
cussions, the p.u.l. parameters of the proposed RLGC(f ) model ing the model parameters. In the first approach, only measured

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY

S-parameters are used as the objective data for the model param-
eter searching. In the second approach, in addition to the mea-
sured S-parameters used as objective data, the Hilbert transform
is implemented on the minimum-phase part of the wave propa-
gation term e−γ l to enforce the causality requirements during the
parameter search. The causal feature of the proposed model can
then be explored by comparing the wave propagation using the
parameters from the two different approaches aforementioned.
Since the first approach (model RLGC-II) is similar to the second
one (model RLGC-III), except neglecting the causality require-
ment enforcement in RLGC-II, the GA implementation is then Fig. 1. GA flowchart of the parameter extraction for the RLGC(f ) models.
focused on the model RLGC-III in the further discussion.
As follows from (14), to build up a frequency-dependent the evaluated propagation term H(fi ) are defined as an objective
causal RLGC(f ) model for a single-ended transmission line, a function, where the subscript letter “c” stands for causal
set of unknowns must be obtained: Ro , Rs , L∞ , Go , Kg , εrreal ,
and εrim ag . Equation (7) indicates that the εrreal and the εrim ag 1
∆=
can be represented over the entire frequency span by the five N
 
parameters ε∞ , εs1 , τ 1 , εs2 , and τ 2 . The causal RLGC(f ) model   2  2
  ||P 2m1 (f i ) |−|P 2e1 (f i ) || ||S 2m1 (f i ) |−|S 2e 1 (f i ) || 

for a single-ended transmission line is then formulated as  
 

 
 m ax |P 2m1 |
+ m ax |S 2m1 | 

  
 N   2


RLGC (f ) = Ψ (ε∞ , εs1 , τ1 , εs2 , τ2 , R0 , Rs , L∞ , G0 , Kg ) .  ||Im (H (f i ))|−|Im (H c (f i ))||
 +
(19)   
 i=1  
m ax|Im (H c (f i ))|

  2 

 
 

This is straightforward for striplines. However, for a mi-  
+
||Re(H (f i ))|−|Re(H c (f i ))|| 

crostrip transmission line, the dielectric represented by a two- m ax|Re(H c (f i ))| 
term Debye model is not exactly the one that corresponds to the
substrate material. Instead, it is an effective dielectric, where (20)
both the substrate material and free space have to be taken where |S21
m
(fi )| and |S21
e
(fi )| are the magnitudes of the mea-
into account. The effective Debye model simplifies the problem sured (index m) and evaluated (index e) S-parameters at fre-
and improves the accuracy of the p.u.l. parameters in solving quency fi , while |P21
m
(fi )| and |P21e
(fi )| are the corresponding
microstrip transmission line problems, since the filling factor as- phases. The parameters max|S21 | and max|P21
m m
| are the max-
sociated with the microstrip geometry configuration and derived imum absolute values over the entire frequency range of in-
from electrostatic fields is eliminated in the related formulation. terest. Both propagation terms Hc (fi ) and H(fi ) are detailed
The search of ten unknowns in (19) becomes an optimization later in this section. Re(H(fi )) and Re(Hc (fi )) are the real
procedure. This is because the number of the equations of the parts of H(fi ) and Hc (fi ), respectively, while Im(H(fi )) and
S-parameters at different frequency points is much greater than Im(Hc (fi )) are the corresponding imaginary parts. The pur-
the number of the unknowns. Since GAs are powerful, robust, pose of introducing max|S21 m
|, max|P21m
|, max|Im (Hc (fi ))| ,
and efficient in global searching and optimization due to their and max|Re (Hc (fi ))| in (20) is to normalize the difference
mechanics of natural selection and natural genetics [18], a GA in each term so that ∆ is unitless, and each term is equally
code is developed to search the global optimum values for the ten weighted.
unknowns [1]. To implement a GA for solving an optimization A fitness function, which is used to quantify the possible
problem, the problem itself must be represented and formulated solution as “good” or “bad,” is defined as [1]
mathematically. For the ten unknowns, it is necessary to define
ten initial parameter ranges, which a possible solution for each p = ∆−1/3 . (21)
unknown correspondingly belongs to. The initial ranges are es-
The higher the fitness index p, the more chance that the re-
timated based on the transmission line configurations including
lated value stays in the GA search pool as a parent to generate
the cross-sectional geometry, the conductivity of the transmis-
offspring for the next generation. Therefore, the fitness index
sion line conductor, and the constituent parameters of the sur-
is used for each possible solution to compete against the others
rounding dielectric materials. These known conditions are not
in their solution space. The expected unknowns of the causal
directly involved in some of calculations of the unknowns. They
RLGC(f ) model are obtained as ∆ is minimized and p con-
are, instead, only used to evaluate the initial parameter ranges
verges. A schematic program flowchart about the GA imple-
for the ten unknowns, which differ from the skin-effect loss,
mentation is shown in Fig. 1. More general topics about GA
directly calculated from the nominal dimensions. Therefore, the
operators, such as selection, recombination, and mutation, are
unknowns extracted from GA in the proposed model are ac-
described in [1] and [18]–[21], and they are beyond the interest
curate. This is because the measured S-parameters include all
of this paper.
the nonideal effects, such as the surface roughness and the di-
The wave propagation term for a transmission line section
mension deviations of the transmission line. The S-parameter
with length l is [16]
differences between the evaluation and the measurement, and
the differences between the causal propagation term Hc (fi ) and H (ω) = e−γ l (22)

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: CAUSAL RLGC(f ) MODELS FOR TRANSMISSION LINES 5

where the propagation constant γ is defined based on the atten- The real part and the imaginary part shown in (30) now
uation constant α and phase constant β as meet the Kramers–Kronig relations, and the phase Pm can be
uniquely determined from θ (ω) by applying Hilbert transform,
γ = α + jβ. (23) i.e.,
Then the propagation term can be represented as Pm = −Im (Hilbert (θ (ω))) (31)
−α l−j β l −θ (ω )−j ϕ(ω )
H (ω) = e =e (24) where Im stands for the operation of obtaining the imaginary
part after applying the Hilbert transform on θ (ω). The phase
where θ (ω) = αl and ϕ (ω) = βl. In general, ϕ (ω) cannot be
θ (ω) is
uniquely determined from θ (ω) even if H (ω) is the transform
of a causal function of h(t) [22]. For the unique determination θ (ω) = − ln |H (ω)| . (32)
of ϕ (ω) from θ (ω), it is necessary to impose certain conditions
Substituting (32) into (31), the phase Pm of the minimum
on the transfer function H (ω). According to linear system the-
phase function can then be restored from the natural logarithm
ory [23], any stable system function can be represented by the
of the magnitude of the transfer function (24) as
product of a minimum phase function Hm (ω) and an all-pass
function Ha (ω) that has a unitary magnitude over the entire Pm = −Im (Hilbert (− ln (|H (ω)|))) . (33)
frequency span of interest as The causal propagation term Hc (ω)corresponding to the
H (ω) = Hm (ω) Ha (ω) . (25) transmission line propagation term H (ω) = e−γ l is built as
√
There is the Bode condition for separating the transfer func- Hc (ω) = |H (ω)| e−j P m e−j l L ∞C ∞
. (34)
tion H (ω) into a minimum phase function and an all-pass func- The GA objective function (20) and the fitness function (21)
tion, according to which the minimum phase function is defined can be estimated based on both the causality requirements and
as [6], [24] the S-parameter differences between the measurements and the
 
γ (ω) GA evaluations. The best solution (model parameters) is ob-
lim ⇒ 0. (26) tained for the RLGC(f ) model as the causality requirements are
ω →∞ jω
met, and the S-parameter discrepancies are minimized.
Substituting (7), (14), and (15) into (26), we get
   C. Frequency-Dependent Causal RLGC(f ) Model for Coupled

   R (f ) + j L + R√ Transmission Lines
 ∞
s
γ (ω) 1 ω 2π f
lim = lim  The method to enforce the causality requirement in a coupled
ω →∞ jω ω →∞ j  G o +K g ω ε i m a g ε o
r

ω + jKg εrreal εo transmission line is more complicated than in a single-ended

case. When compared with the single-ended model given in
"
1 (19), two more unknowns must be added to the RLGC(f ) model
= lim jL∞ Kg εrim ag εo + jKg εrreal εo (RLGC-II) to take into account the coupling effects. One un-
ω →∞ j
known is the p.u.l. mutual inductance Lm and the other is a
= L∞ C∞ (27) geometry-related factor Kg m . The first unknown is associated
with inductive coupling and the second one is related to ca-
where C∞ and L∞ are the p.u.l. capacitance and inductance pacitive coupling. The frequency-dependent causal RLGC(f )
evaluated at infinite frequency. For a transmission line with the model for coupled transmission lines is then represented by 12
two-term Debye dielectric representation, the C∞ corresponds unknowns as
to the “optical limit” dielectric constant ε∞ . # $
ε∞ , εs1 , τ1 , εs2 , τ2 , R0 ,
The relation shown in (27) indicates that a transmission line RLGC (f ) = Ψ . (35)
described by the RLGC(f ) model (RLGC-II) is not a minimum- Rs , L∞ , G0 , Kg , Lm , Kg m
phase system in the general case. The causality
√
can be enforced
The initial parameter range of the p.u.l. mutual inductance
by removing the linear phase term e−j l L ∞ C ∞ from the total
Lm can be evaluated from the initial parameter range of L∞ as
phase. The transfer function (24) can then be rewritten as
  L∞ and Lm are related by the coupling factor m as
√
H (ω) = e−θ (ω )−j P m e−j l L ∞ C ∞ (28) Lm
m= (36)
√ L∞
where Pm + l L∞ C∞ = ϕ (ω) and Pm is the minimum phase.
where the two lines in the coupled differential stripline are as-
By comparing (25) with (28), the all-pass function and the
sumed exactly same.
minimum-phase function is split as
The initial parameter range for the geometry factor Kg m can

Hm (ω) = e−θ (ω )−j P m be evaluated from the dimensions of the cross section of the
√ (29) coupled line. The p.u.l. mutual capacitance and the p.u.l. mutual
Ha (ω) = e−j l L ∞ C ∞ .
shunt conductance of the transmission line can be evaluated as
The phase from the minimum phase function in (29) is then Cm (f ) = Kg m εrreal εo (37)
θ (ω) + jPm = − ln (Hm (ω)) . (30) Gdm (f ) = 2πf Kg m εrim ag ε0 . (38)

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY

To simplify the enforcement of the causality requirement in and the insertion phase in the range of 30◦ –150◦ [25]. For the
the coupled differential pair in the model RLGC-III, it is better coupled stripline, a new deembedding method was developed
to separate even and odd wave propagation modes along the and used to eliminate the port effects [26].
transmission line. The propagation constants for the even mode The causality of a single-ended transmission line was tested
γe and the odd mode γo dd are calculated from using a well-defined time-domain pulse [27]
γe = (R (f ) + j2πf Le (f )) (Ge (f ) + j2πf Ce (f )) (39)
10 (t/τ0 )n −t /τ 0
y (t) = e (46)
γo dd = (R (f )+j2πf Lo dd (f )) (Go dd (f )+j2πf Co dd (f )) 1 + (t/τ0 )n
(40)
that was launched at one end (driving end) of the transmission
where Le , Ge , Ce , Lo dd , Go dd , and Co dd are evaluated from [17] line and observed at the other end (receiving end) with τ0 =
 0.1 ns and n = 4. Then y(t) was normalized to 1 by using
 L = L (f ) + Lm
 e y(t)/max(y(t)). A MATLAB code was written to realize the
Ge = G (f ) − Gdm (41) observation. The time-domain pulse y(t) launched at the driv-

 C = C (f ) − C
e m ing end was converted into the frequency domain using a fast
 Fourier transform (FFT) to obtain its frequency-domain spectra.
 L = L (f ) − Lm The propagation term e−γ l obtained from the RLGC(f ) model
 o dd
Go dd = G (f ) + Gdm (42) was multiplied by the frequency spectra of y(t), and the re-

C sult was converted back to time domain using the inverse FFT
= C (f ) + C .
o dd m
(IFFT), i.e., FFT→IFFT procedure was done. For the coupled
The characteristic impedance for the even mode Ze and the transmission line, the procedure was similar to the single-ended
odd mode Zo dd can be calculated as cases. But the coupled wave propagation was decomposed into
 even and odd modes, and the causality was examined for each
(R (f ) + j2πf Le (f )) mode separately.
Ze = (43) To clearly show the causal/noncausal phenomena, each of
(Ge (f ) + j2πf Ce (f ))
 the three studied transmission line cases was represented by
(R (f ) + j2πf Lo dd (f )) three different RLGC models. The first model was the RLGC-I
Zo dd = . (44) introduced in [12], and the model parameters were extracted
(Go dd (f ) + j2πf Co dd (f ))
using the method reported in [14]. The second method RLGC-II
The differential impedance and the common impedance is was the RLGC(f ) model, proposed in this paper, where the ten
determined from (or 12) unknowns were obtained from the method described in
 Section II, with measured S-parameters as the only objective
Zcom = 0.5Ze
(45) data. The third model RLGC-III was same as the RLGC-II, but
Zdif = 2Zo dd . with the enforced causality requirement, and hence, the parame-
ters in these models could differ. The ten (or 12) unknowns of the
For a coupled transmission line, the S-parameters can be eas-
third model (RLGC-III) were extracted using (20) as an objec-
ily evaluated for the common mode and the differential mode if
tive function with causality requirements enforced in the model
the p.u.l. R, L, G, and C parameters of the RLGC(f ) model and
parameter extraction. Therefore, the causality in the RLGC-
the transmission line length are known. While the S-parameters
III model was guaranteed. Along with the causal/noncausal
are under evaluation, the causality requirements have to be en-
phenomenon observation, the S-parameters of each studied
forced for both even and odd modes. As long as the coupled
case were modeled by the three RLGC(f ) representations.
problem is decomposed into even and odd modes, the earlier
Comparisons of the obtained S-parameters and comparison of
described procedure for single-ended transmission line can be
the time-domain waveforms at receiving end for the three stud-
directly used for solving coupled problems.
ied cases were conducted.
The first studied case is a single-ended stripline. It is built
III. CASE STUDIES
on layer 7 within an eight-layer board having FR-4 as sub-
Three cases, a single-ended microstrip, a single-ended strates. The cross-sectional dimensions of the stripline, the test
stripline, and an edge-coupled differential stripline, were stud- board, and the measurement reference plane after TRL calibra-
ied based on the method described in Section II. All the S- tion are shown in Fig. 2. The thickness of the copper is t =
parameters were measured using 8720ES VNA with ATN-4112 1.35 mil, the width of the trace is w = 12.5 mil, and the to-
A S-parameters test set. The measurement frequency range was tal height between the reference planes is b = 27.7 mil. The
from 200 MHz to 20 GHz. “Thru-Reflect-Line” (TRL) cali- stripline length after TRL calibration is 7976 mil. Fig. 3 demon-
bration was used to remove the coaxial-connector-related port strates the comparison between the measured S-parameters, the
effects in the single-ended cases [16]. Two TRL calibration kits corresponding results of HSPICE simulation using the RLGC-I
were designed and built on their corresponding test boards. The model with the extracted model parameters, and the results of the
frequency span of each TRL calibration pattern was broken into RLGC(f ) model with the parameters extracted using the above-
three frequency ranges to meet the usable bandwidth for a sin- mentioned RLGC-II and RLGC-III methods. Comparison of the
gle LINE/THRU (TRL calibration standards) pair less than 8:1 time-domain waveforms at the receiving end is shown in Fig. 4.

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: CAUSAL RLGC(f ) MODELS FOR TRANSMISSION LINES 7

Fig. 2. Test board, TRL measurement reference plane, and cross-sectional Fig. 5. Test board and the cross-sectional dimensions of the single-ended
dimensions of the single-ended stripline. microstrip line.

Fig. 6. S-parameter comparison between the RLGC(f ) models (RLGC-I, II,

and III) and the measurements for the microstrip line.

Fig. 3. S-parameter comparison between the RLGC(f ) models (RLGC-I, II,

and III) and the measurements for the single-ended stripline.

Fig. 7. Comparison of the time-domain waveforms at the receiving end for

the microstrip line.

Fig. 4. Comparison of the time-domain waveforms at the receiving end for

the single-ended stripline.
and the RLGC(f ) model with two sets of model parameters
extracted from the RLGC-II and RLGC-III models are com-
The second studied case is a single-ended microstrip trans- pared in Fig. 6. The time-domain waveform comparison at the
mission line. It is built on the top layer on a 26-layer receiving end is shown in Fig. 7.
board with FR-4 substrates. The cross-sectional dimensions The third studied case is an edge-coupled differential stripline.
of the microstrip and the test board are shown in Fig. 5, It is built on layer 7 within an eight-layer board with FR-4
where t = 2.4 mil, t1 = 1.35 mil, w = 5.75 mil, and h = as substrates. The cross-sectional dimensions of the coupled
3.65 mil. The microstrip is 6976 mil long after TRL calibration. stripline and the test board shown in Fig. 8 are t = 1.35 mil,
S-parameters obtained from measurement, HSPICE simulation w = 10 mil, and s = 12.5 mil. The coupled pair is 6905.5 mil
using the RLGC-I model with the extracted model parameters, long after TRL calibration [26]. Common-mode S-parameter

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY

Fig. 10. Differential-mode S-parameter comparison between the RLGC(f )

models (RLGC-I, II, and III) and the measurements for the edge-coupled
Fig. 8. Test board and the cross-sectional dimensions of the edge-coupled stripline.
stripline.

Fig. 11. Comparison of the time-domain even mode waveforms at the receiv-
ing end for the edge-coupled stripline.

Fig. 9. Common-mode S-parameter comparison between the RLGC(f ) mod-

els (RLGC-I, II, and III) and the measurements for the edge-coupled stripline.

comparison between the measurement and simulations using

the RLGC-I model with extracted model parameters, and the
RLGC(f ) model with two sets of model parameters from the
RLGC-II and the RLGC-III model extractions is shown in Fig. 9,
while the differential-mode S-parameter comparison is shown in
Fig. 10. The time-domain waveforms for the even mode and odd
mode propagations through the stripline are shown in Figs. 11
and 12, respectively.
S-parameter comparisons, including the magnitude and the
phase shown in Figs. 3, 6, 9, and 10, demonstrate that the pro-
posed RLGC(f ) model (RLGC-II), the analytical RLGC model
(RLGC-I), and the causality enforced model (RLGC-III) agree
well with the measurements. The maximum magnitude differ- Fig. 12. Comparison of the time-domain odd mode waveforms at the receiving
ence between simulation and measurement is less than 1 dB up end for the edge-coupled stripline.
to 20 GHz for the three studied cases, and the phase differences
are hard to distinguish. However, the time-domain waveforms
shown in Figs. 4, 7, 11, and 12 clearly demonstrate that the IV. DISCUSSION AND CONCLUSION
RLGC(f ) models (RLCG-II and RLCG-III) are causal, but the Frequency-dependent causal RLGC(f ) models are proposed
RLGC-I model is not. for single-ended and coupled transmission lines, and the

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: CAUSAL RLGC(f ) MODELS FOR TRANSMISSION LINES 9

methodology for building the models directly from measured [15] J. Zhang, M. Y. Koledintseva, R. E. DuBroff, D. J. Pommerenke,
S-parameters is developed. Time-domain waveforms in the J. L. Drewniak, K. N. Rozanov, G. Antonini, and A. Orlandi, “Character-
ization of dispersive dielectrics using planar transmission line structures
three studied cases clearly show causal phenomena for the pro- and genetic algorithm,” submitted for publication.
posed RLGC(f ) models while showing noncausal phenomena [16] D. M. Pozar, Microwave Engineering, 2nd ed. New York: Wiley, 1998.
for the RLGC-I model. For the proposed RLGC(f ) model, the [17] R. Mongia, I. Bahl, and P. Bhartia, RF and Microwave Coupled-Line
Circuits. Norwood, MA: Artech House, 1999, pp. 23–64.
start points of the received waveforms at the receiving end with [18] Y. Rahmat-Samii and E. Michielssen, Electromagnetic Optimization by
model parameters extracted from S-parameters only (RLGC-II) Genetic Algorithms. New York: Wiley, 1999, pp. 1–93.
and from the S-parameters plus enforced causality requirements [19] L. Davis, Handbook of Genetic Algorithms. New York: Van Nostrand
(RLGC-III) are exactly same. This observation indicates that Reinhold, 1991, pp. 1–99.
[20] D. Quagliarella, J. Periaux, C. Poloni, and G. Winter, Genetic Algorithms
the proposed RLGC(f ) model (RLGC-II) is causal, as expected. and Evolution Strategy in Engineering and Computer Science. New
Therefore, enforcing the causality requirements in the model York: Wiley, 1998, pp. 289–309.
parameter extraction is not necessary. [21] B. L. Miller and D. E. Goldberg, “Genetic algorithms, tournament selec-
tion, and the effects of noise,” IlliGAL Rep. 95006, Jul. 1995.
Although the studied cases are tested on PCBs with FR-4- [22] A. Papoulis, The Fourier Integral and Its Applications. New York:
type epoxy resin fiber-glass-filled substrates in the frequency McGraw-Hill, 1962.
range from 200 MHz to 20 GHz, the approach presented herein [23] S. K. Mitra, Digital Signal Processing—A Computer-Based Approach,
2nd ed. New York: McGraw-Hill, 2001, pp. 203–277.
is more general than that typically used to analyze PCBs. It [24] H. W. Bode, Network Analysis and Feedback Amplifier Design. New
can also be applied to on-silicon interconnects, provided that York: Van Nostrand, 1945.
the TEM/quasi-TEM conditions are fulfilled for transmission [25] Agilent Technologies, “Agilent network analysis applying the 8510 TRL
lines, and causal relations for permittivity are valid for substrate calibration for non-coaxial measurements,” Agilent Technol., Santa Clara,
CA, May 2001, Product Note 8510-8A.
dielectric materials. For higher frequency applications, or for [26] J. Zhang, Q. B. Chen, Z. Qiu, J. L. Drewniak, and A. Orlandi, “Using a
modeling other types of substrate dielectric materials with more single-ended TRL calibration pattern to de-embed coupled transmission
complex-shaped frequency dispersion, more than two Debye lines,” presented at the Int. Symp. Electromagn. Compat., Austin, TX,
Aug. 17–21, 2009.
terms may be needed. [27] IEC, “’Part 4-2: Testing and measurement techniques–Electrostatic dis-
charge immunity test,” unpublished.

REFERENCES
[1] J. Zhang, M. Y. Koledintseva, J. L. Drewniak, D. J. Pommerenke,
R. E. DuBroff, Z. Yang, W. Cheng, K. N. Rozanov, G. Antonini, and Jianmin Zhang (S’02–M’07–SM’09) received the
A. Orlandi, “Reconstruction of dielectric material parameters for disper- B.S. degree from Southeast University, Nanjing,
sive substrate using a genetic algorithm,” IEEE Trans. Electromagn. China, in 1985, and the M.S. and Ph.D. degrees
Compat., vol. 50, no. 3, pp. 704–714, Aug. 2008. in electrical engineering from the University of
[2] L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media. Missouri–Rolla, Rolla (renamed as Missouri Univer-
New York: Pergamon, 1960, pp. 239–312. sity of Science and Technology in 2008), in 2003 and
[3] S. L. Hahn, Hilbert Transforms in Signal Processing. Norwood, MA: 2007, respectively.
Artech House, 1996. During 1985, he was with the Nanjing Electronic
[4] A. D. Poularikas, The Transforms and Applications Handbook. Boca Equipment Research Institute, China, as a Hardware
Raton, FL: CRC Press, 1996, pp. 463–627. Engineer. In 2007, he joined Cisco Systems, Inc., San
[5] R. L. Wigington and N. S. Nahman, “Transient analysis of coaxial cables Jose, CA, as a Senior Hardware Engineer, where he
considering skin effect,” in Proc. IRE, Feb., 1957, vol. 45, pp. 166–174. is involved in signal integrity and power integrity R&D for high-speed inter-
[6] N. S. Nahman, “A discussion on the transient analysis of coaxial cables connects and design and analysis for high-performance networking products at
considering high-frequency losses,” IRE Trans. Circuit Theory, vol. 9, the printed circuit board level, package level, and system level. His research
pp. 144–152, Jun. 1962. interests include signal integrity, power integrity, SerDes modeling, and elec-
[7] D. R. Holt and N. S. Nahman, “Coaxial-line pulse-response error due tromagnetic interference/electromagnetic compatibility for high-speed digital
to a planar skin-effect approximation,” IEEE Trans. Instrum. Meas., systems. He is the author or coauthor of more than 30 technical papers and is
vol. IM-21, no. 4, pp. 515–519, Nov. 1972. the inventor of three issued patents.
[8] N. S. Nahman and D. R. Holt, “Transient Dr. Zhang was the recipient of the Best Symposium Paper Award and the
√ analysis of coaxial cables us-
ing skin effect approximation A + B f,” IEEE Trans. Circuit Theory, Best Student Symposium Paper Award from the IEEE Electromagnetic Com-
vol. CT-19, no. 5, pp. 443–451, Sep. 1972. patibility Society in 2006, and the Conference Best Session Paper Award in
[9] T. R. Arabi, A. T. Murphy, T. K. Sarkar, and R. F. Harrington, “On the signal integrity from the International Microelectronics and Packaging Society
modeling of conductor and substrate losses in multiconductor, multidi- in 2007.
electric transmission line systems,” IEEE Trans. Microw. Theory Tech.,
vol. 39, no. 7, pp. 1090–1097, Jul. 1991.
[10] C. Svensson and G. E. Dermer, “Time domain modeling of lossy inter- James L. Drewniak (S’85–M’90–SM’01–F’06) re-
connects,” IEEE Trans. Adv. Packag., vol. 24, no. 2, pp. 191–196, May ceived the B.S., M.S., and Ph.D. degrees in electrical
2001. engineering from the University of Illinois, Urbana-
[11] P. S. Neelakanta, Handbook of Electromagnetic Materials—Monolithic Champaign, in 1985, 1987, and 1991, respectively.
and Composite Versions and Their Applications. Boca Raton, FL: CRC In 1991, he joined the Electrical and Com-
Press, 1995, pp. 31–56. puter Engineering (ECE) Department, University of
[12] Star-HSPICE Manual, HSPICE, Fremont, CA, release 2001.2, Jun. 2001, Missouri–Rolla, Rolla (renamed as Missouri Univer-
ch.18. sity of Science and Technology in 2008), where he
[13] D. D. Pollock, Physical Properties of Materials for Engineers, 2nd ed. was the Director of the Materials Research Center
Boca Raton, FL: CRC Press, 1993, pp. 499–575. during 2002–2007, and has been one of the Principal
[14] J. Zhang, M. Koledintseva, J. Drewniak, G. Antonini, A. Orlandi, and Investigators in the Electromagnetic Compatibility
K. Rozanov, “Extracting R, L, G, C parameters of dispersive planar trans- Laboratory and a Full Professor in the ECE Department. His research interests
mission lines from measured S-parameters using a genetic algorithm,” include electromagnetic compatibility (EMC) in high-speed digital and mixed-
in Proc. 2004 IEEE Int. Symp. Electromagn. Compat., Santa Clara, CA, signal designs, electronic packaging, microelectromechanical systems, EMC in
Aug., vol. 2, pp. 572–576. power-electronic-based systems, and numerical modeling for EMC applications.

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY

David J. Pommerenke (M’98–SM’03) received the Zhiping Yang (S’97–M’00–SM’05) received the
Ph.D. degree from the Technical University of Berlin, B.S. and M.S. degrees from Tsinghua University,
Berlin, Germany, in 1996. Beijing, China, in 1994 and 1997, respectively, and
He was with Hewlett Packard for five years. In the Ph.D. degree from the University of Missouri-
2001, he joined the Electromagnetic Compatibility Rolla, Rolla (renamed as the Missouri University of
Laboratory, University Missouri–Rolla, Rolla (re- Science and Technology in 2008), in 2000.
named as the Missouri University of Science and He was a Principal Signal Integrity Engineer with
Technology in 2008), where he is currently a Tenured Apple Computer. He was a Technical Leader with
Professor in the Electrical and Computer Engineer- Cisco System, Inc., San Jose, CA, where he is cur-
ing Department. Besides other professional activities, rently a Senior Manager of Hardware Engineering.
he is the U.S. representative of the Electrostatic Dis- He was a member of the Technical Staff with Nuova
charge (ESD) Standard Setting Group within the IEC TC77b. He was a Distin- Systems, Inc. His research interests include signal integrity and power integrity
guished Lecturer for the IEEE Electromagnetic Compatibility (EMC) Society methodology development for die/package/board codesign, application of em-
during 2006–2007. He is the author or coauthor of more than 100 papers and bedded passives, extraction of material properties at high frequency, and high-
the Inventor of ten patents. His research interests include system-level ESD, speed differential signaling technology. He is actively involved with I/O buffer
numerical simulations, EMC measurement methods, and instrumentations. information spec (IBIS) standard activities. His research in power integrity
greatly increases IBIS model’s simulation accuracy under nonideal power sup-
ply conditions. In 2005, the IBIS Standard Committee ratified BIRD95.6 pro-
Marina Y. Koledintseva (M’96–SM’03) received posal, which was coauthored by him and his colleagues in Cisco Systems and
the M.S. (highest honors) and Ph.D. degrees from Teraspeed Consulting Group. He is the author or coauthor of more than 20
the Moscow Power Engineering Institute (Technical research papers. He holds six issued and two pending U.S. patents.
University)—MPEI (TU), Moscow, Russia, in 1984
and 1996, respectively.
During 1983–1999, she was a Researcher with the
Ferrite Laboratory, MPEI (TU), where she was an
Associate Professor from 1997 to 1999. In January
2000, she joined the Electromagnetic Compatibility
(EMC) Laboratory, Missouri University of Science
Qinghua B. Chen (S’95–M’96) received the
and Technology (MS&T), Rolla (known as the Uni-
versity of Missouri–Rolla before 2008), as a Visiting Professor, where she has B.S.E.E. and M.S.E.E. degrees from Tsinghua Uni-
versity, Beijing, China, and the Ph.D. degree from
been a Research Professor in the Electrical and Computer Engineering Depart-
Texas A&M University, College Station.
ment since 2005. Her current research interests include microwave engineering,
He worked with Andiamo Systems, Inc., as a Man-
interaction of electromagnetic fields with ferrites and composite media, their
modeling, and application for electromagnetic compatibility. She is the author ager and a Technical Leader, where he was in charge
of high-speed signal integrity activities. He was also
or coauthor of more than 150 papers. She is the holder of seven patents.
with Raza Foundries, Inc., Nplab, Inc., and Texas
Dr. Koledintseva is a member of the Education, TC-9 Computational Elec-
Instruments, Inc., as a Technical Leader/ Senior De-
tromagnetics, and TC-11 (Nanotechnology) Committees of the IEEE Electro-
magnetic Compatibility Society. sign Engineer, where he was engaged in high-speed
IC/system designs. He is currently a Senior Engineer-
ing Manager with Cisco Systems, Inc., San Jose, CA, where he is engaged with
high-speed high-performance networking product R&D.
Richard E. DuBroff (S’74–M’77–SM’84) received
the B.S.E.E. degree from Rensselaer Polytechnic In-
stitute, Troy, NY, in 1970, and the M.S. and Ph.D.
degrees in electrical engineering from the University
of Illinois, Urbana-Champaign, in 1972 and 1976,
respectively.
From 1976 to 1978, he was a Postdoctoral Re- Antonio Orlandi (M’90–SM’97–F’07) was born in
searcher with the Ionosphere Radio Laboratory, Uni- Milan, Italy, in 1963. He received the Laurea degree
versity of Illinois, Urbana-Champaign, where he was in electrical engineering from the University of Rome
involved with backscatter inversion of ionospheric “La Sapienza,” Rome, Italy, in 1988.
electron density profiles. From 1978 to 1984, he was a From 1988 to 1990, he was with the Depart-
Research Engineer in the Geophysics Branch of Phillips Petroleum, Bartlesville, ment of Electrical Engineering, University of Rome
OK. Since 1984, he has been with the University of Missouri, Rolla (renamed “La Sapienza.” Since 1990, he has been with the
as the Missouri University of Science and Technology in 2008), where he was Department of Electrical Engineering, University of
an Associate Chairman for graduate studies from 1991 to 1996 and from 2002 L’Aquila, L’Aquila, Italy, where he is currently a Full
to 2009, and is currently a Professor in the Electrical and Computer Engineering Professor and the Chair of the UAq Electromagnetic
Department. Compatibility (EMC) Laboratory. He is the author or
coauthor of more than 200 technical papers published in the field of EMC in
lightning protection systems and power drive systems. His research interests
include numerical methods and modeling techniques to approach signal/power
Wheling Cheng received the Bachelor’s degree from integrity, and EMC/electromagnetic interference issues in high-speed digital
Chiao-Tung University, Hsinchu, Taiwan, in 1986, systems.
and the Ph.D. degree from Stanford University, Dr. Orlandi is a member of the Education, TC-9 Computational Electro-
Stanford, CA, in 1995. magnetics, and the Chairman of the TC-10 “Signal Integrity” Committees of
During 1995–1997, she was with nChip, where the IEEE EMC Society. He was the recipient of the IEEE TRANSACTIONS ON
she was involved in multichip module development ELECTROMAGNETIC COMPATIBILITY Best Paper Award in 1997, the IEEE EMC
and fabrication. During 1997–1999, she was with LSI Society Technical Achievement Award in 2003, the IBM Shared University Re-
Logic, where she was involved with signal integrity search Award in 2004–2006, the Computer Simulation Technology University
analysis and package design. From 1999 to July 2009, Award in 2004, and the IEEE International Symposium on EMC Best Paper
she was with Cisco Systems, Inc., where she was en- Award in 2009. He was an Associate Editor of the IEEE TRANSACTIONS ON
gaged in signal integrity and power integrity analysis. ELECTROMAGNETIC COMPATIBILITY from 1996 to 2000 and the IEEE TRANS-
In August 2009, she joined Juniper Networks, Sunnyvale, CA, as a member of ACTIONS ON MOBILE COMPUTING from 2001 to 2006, and has been the Sym-
the Technical Staff. Her current research interests include 3-D modeling of posium Chairman of the TC-5 “Signal Integrity” Technical Committee of the
high-speed interconnects and SerDes channel analysis. International Zurich Symposium and Technical Exhibition on EMC since 1999.

Authorized licensed use limited to: Missouri University of Science and Technology. Downloaded on February 14,2010 at 13:00:16 EST from IEEE Xplore. Restrictions apply.