Robust Passivity Enforcement of Frequency Dependent Transmission Line Models

Bjørn Gustavsen
SINTEF Energy Research
N-7465 Trondheim, Norway
[email protected]

where i and r denote incident and reflected wave.

Abstract
Passivity must be enforced on rational macromodels in Matrices H and Yc are respectively the matrices of
order to ensure a stable time domain simulation. This paper propagation characteristic admittance. They are obtained
investigates procedures for enforcing passivity for directly from the p.u.l. series impedance matrix Z and shunt
transmission line models based on the Method of admittance matrix Y, and the line length l
Characteristics. Adding a conductive correction term
externally to the line terminals is shown to be a robust and Yc = Z −1 ZY (3)
simple procedure for ensuring a passive model. The − ZY l
perturbation of the line model behavior is reduced by shaping H=e (4)
the correction term using a low order rational function.
In the ULM [6], the rational fitting for Yc and H are on
Introduction the form
The modeling of linear devices and systems by rational N
R
macromodels has reached considerable interest, due to the Yc ≅ ∑ s − ma +D (5)
potential high accuracy of the resulting model. The modeling m =1 m
can be done from tabulated data in either the frequency
domain or the time domain, resulting in a model on state- G ⎛ Ng
R m, g ⎞ − sτ
space form or pole-residue form that can be easily included in H≅ ∑ ⎜⎜ ∑ s − a ⎟ e g
⎟
(6)
g =1 ⎝ m =1 m, g ⎠
most simulation programs. Although the modeling process is
fairly straightforward by usage of robust fitting techniques
Poles and delays for the fitting of H in (6) are precalculated
such as vector fitting [1], passivity should also be enforced in
by fitting the modes hi of H. (Nearly equal modes are lumped
order to ensure a stable simulation. This is usually achieved
by a postprocessing step where the model parameters are before doing the fitting, thereby reducing the order of the final
perturbed [2],[3],[4]. Techniques are also available for model).
passivity checking via algebraic tests [4]. N
r
This paper focuses on passivity enforcement for hi ≅ ∑ s − ma e− sτ i , i = 1,..., n (7)
transmission line models based on the Method of m =1 m ,i
Characteristics (MoC). The previous techniques for passivity
enforcement are not directly applicable to MoC based models Nodal Admittance Matrix And Passivity
since the latter makes use of delay terms. For that purpose, an The admittance matrix with respect to the line terminals can
algorithm is introduced that adds a rational correction term be calculated directly from the rational approximations for H
externally to the line terminals (ports). The approach is and Yc as follows [8]
demonstrated for use with the so-called Universal Line Model
(ULM) [6] as implemented in the PSCAD simulation Yn = ⎡ A B ⎤ (8)
environment [7]. ⎣⎢ B A ⎥⎦
where
The Universal Line Model
In the simulation of electromagnetic transients in power A = Yc (I + H 2 )(I − H 2 ) −1 (9)
systems, the Universal Line Model (ULM) is frequently
B = −2Yc H (I − H 2 ) −1 (10)
applied. This is a direct formulation of the MoC in phase
coordinates, without assumption of a real transformation The model is passive if the nodal admittance stamp (8) is
matrix. The main strength of this approach is the ability to positive for all frequencies, i.e. [2],[9]
accurately simulate situations where the propagation function
has small but important off-diagonal elements. This is
eig (Re{Yn ( s )}) > 0 ∀s, s = jω (11)
sometimes the situation in EMC problems such as undesired
coupling between a power line and telecommunication line.
Consider a line with ends k and m. In the frequency An approach for passivity checking based on algebraic
domain, the voltage-current relation at end k can be written tests is shown in [10].

i k = Yc v k − 2i ki (1) Passivity Enforcement by External Correction Term

Large passivity violations can occur outside the fitting
i ki = HT i mr (2) band. Most EMTP-type programs add a small shunt

169 SPI2007
conductance to Y in order to ensure a physical (passive)
behavior at low frequencies [11], see [12] for a recent Correction
discussion. With ULM, passivity violations can with short
lines also occur at high frequencies. This problem is overcome MoC
by multiplying H with a low-pass-filter [13].
One way of removing any remaining passivity violations is Fig. 2 Rational correction term (two-conductor line)
to add losses by connecting an external model to the line Example: Coaxial cable
terminals. Using the Simplistic Approach in [2, Section VII], As an example is used a homogenous two-conductor
a correction matrix Dcorr is established incrementally in a coaxial cable of 10 km length, buried 1 m below the surface
frequency sweep over the interval(s) containing passivity of a lossy ground, see Fig. 3. The series impedance Z is
violations. At the first frequency, Dcorr is initialized to zero. At calculated including skin effects in conductors and earth [13].
each frequency, Re{Yn}+Dcorr is separated into two modal
decompositions that respectively contain the positive and
negative eigenvalues.
1.0 m
−1 −1 −1
Re{Yn } + Dcorr = TΛT = TΛ pos T + TΛ neg T (12)

Positive eigenvalues are ensured for Re{Yn}+Dcorr at a ρsoil=100 Ω⋅m

given frequency by adding the negated value of the second Fig. 3 132 kV cable
term in (12) to Dcorr. This procedure ensures that all
eigenvalues of the combined model are non-negative at all CABLE DATA
frequencies contained in the sweep. No iterations are needed Item Property
since each update of Dcorr leads to higher losses at all Core OD=39 mm , ρ= 3.365E–8 Ω⋅m
frequencies. The procedure is illustrated in Fig. 1. Insulation t=18.25 mm, εr=2.85
Sheath t=0.22 mm , ρ= 1.718E–8 Ω⋅m
Dcorr = 0 Jacket t=4.53 mm, εr=2.51
for k=1:Ns
In the fitting process, eight poles are used for the fitting of
Re{Y( sk )} + Dcorr → TΛ pos T −1 + TΛ neg T −1
Yc (5) and 12 poles are used for fitting each of the modes of H
Dcorr : = Dcorr − TΛ neg T−1 (7). Out-of-band passivity violations at low frequencies are
end avoided by introducing an artificial shunt conductance at low
frequencies such that a trapped voltage will discharge with a
Fig. 1 Calculating conductive correction term
one second time constant. Out-of-band passivity violations at
high frequencies are avoided by multiplying H with a low-
The resulting correction term is a nodal conductance matrix
pass filter (10 MHz cut-off frequency) and asymptotic
that gives an instantaneous coupling between all line
passivity is enforced for Yc [13].
terminals, see Fig. 2. The correction can be limited in
Fig. 4 shows the rational fitting of Yc using an 8th order
bandwidth by multiplying with a suitable band-pass filter. If
common pole set with weighting equal to the inverse
the passivity violations are located between frequencies
magnitude of the matrix trace. The weighting is seen to ensure
{ω 1,ω 2}, it is proposed to use a correction high relative accuracy at low frequencies where the matrix
elements are small.
−1.01 ⋅ (−10ω 2 ) ⋅ s
P( s ) = Dcorr (13) Fig. 5 shows the rational fitting of H using a 12th order
( s − (−0.1ω 1))( s − (−10ω 2 )) approximation for each of the two modes. It is seen that all
elements are well fitted, including the small element that
The two poles in (13) are respectively placed at one decade represents the induced sheath voltage.
below and above ω 1 and ω 2 to ensure that the imaginary part Figs 6 and 7 (expanded view) show the resulting
of the band-pass filter is negligible between ω 1 and ω 2. The eigenvalues of Re{Yn}, calculated from the fitted model by
factor 1.01 ensures that the real part of the band-pass filter is (8)-(10). Some of the eigenvalues are negative in the
greater than unity at the boundaries ω 1 and ω 2. For practical frequency range 15.3 Hz–575 Hz, thereby making the line
implementation in a time domain simulation program it may model non-passive.
be advantageous to expand (13) into a sum of pole-residue
terms

R1 R2
P( s) = + (14)
s − a1 s − a2

170 SPI2007
Fig. 4 Rational fitting of Yc (8th order) Fig. 7 Eigenvalues of Re{Yn}. Expanded view

Fig. 5 Rational fitting of H (12th order per mode)

Fig. 8 Effect of external correction term on eig(Re{Yn})
Time Domain Simulation
In the following example is used the obtained model in a
transient simulation, see Fig. 9. The simulation is with the
PSCAD software [7] with the correction term (14) included
via a user-defined subroutine [14].

V1
V3
V4

Fig. 9 Step voltage excitation

Fig. 6 Eigenvalues of Re{Yn}. Figs 10 and 11 respectively show the simulated voltage at
the far end core conductor and sheath conductor when the
A correction term (13) is calculated and added to the four cable core conductor is subjected to a step voltage excitation.
line terminals. The result in Fig. 8 shows that the eigenvalues The sheath conductor is grounded at the sending end but is
are modified to become positive at all frequencies, thereby open at the far end. The simulated voltage is shown before
giving a passive model. It is also seen that the eigenvalues are and after adding the external correction term (13). In addition
unaffected at very low frequencies due to the band pass is shown the deviation between the two responses. It is seen
function in (13). that the effect of the correction term is insignificant. The
relative impact is higher for the response in Fig. 11 since it is
nearly two magnitudes smaller than the one in Fig. 10.

171 SPI2007
 model perturbation is reduced by appropriate shaping of the
correction term by a band pass function. Application to one
case taken from power systems modeling showed that the
passivity was achieved with a only a insignificant change to a
transient response.
References
[1] B. Gustavsen, and A. Semlyen, “Rational approximation
of frequency domain responses by vector fitting”, IEEE
Trans. Power Delivery, vol. 14, no. 3, pp. 1052-1061, July
1999.
[2] B. Gustavsen, and A. Semlyen, ”Enforcing passivity for
admittance matrices approximated by rational functions”,
IEEE Trans. Power Systems, vol. 16, no. 1, pp. 97-104,
Feb. 2001.
[3] S. Grivet-Talocia, “Passivity enforcement via perturbation
Fig. 10 Far end core voltage (V3) of Hamiltonian matrices”, IEEE Trans. Circuits and
Systems–I, vol. 51, no. 9, pp. 1755-1769, Sept. 2004.
[4] D. Saraswat, R. Achar, and M.S. Nakhla, “A fast
algorithm and practical considerations for passive
macromodeling of measured/simulated data”, IEEE Trans.
Advanced Packaging, vol. 27, no. 1, pp. 57-70, Feb. 2004.
[6] A. Morched, B. Gustavsen, and M. Tartibi, “A universal
model for accurate calculation of electromagnetic
transients on overhead lines and underground cables”,
IEEE Trans. Power Delivery, vol. 14, no. 3, pp. 1032-
1038, July 1999.
[7] http://pscad.com/
[8] B. Gustavsen, “Validation of frequency dependent
transmission line models”, IEEE Trans. Power Delivery,
vol. 20, no. 2, pp. 925-933, April 2005.
[9] S. Boyd, and L.O. Chua, “On the passivity criterion for
Fig. 11 Far end sheath voltage (V4) LTI n-ports”, Circuit Theory and Applications, vol. 10, pp.
323-333, 1982.
Discussion [10]E. Ghad, C. Chen, M. Nakhla, and R. Achar, “Passivity
The proposed approach for passivity enforcement is highly verification in delay-based macromodels of elcectrical
robust and has been successfully applied to a number of line interconnects”, IEEE Trans. Circuits and Systems – I, vol.
cases, including a six-conductor cable system [13]. This 52, no. 10, pp. 2173-2187, October 2005.
method is fail-proof in the sense that it cannot diverge; a [11]J.R. Marti, “The problem of frequency dependence in
single sweep will guarantee passivity at the frequency samples transmission line modelling”, PhD Thesis, University of
contained in the sweep. A potential drawback is that the size British Columbia, Canada, 1981.
of the perturbation is not minimal, but this should not be a [12]S. Grivet-Talocia, and F. Canavero, “DC-compliant
problem as long as the needed correction is small, which is macromodels based on the method of characteristics for
usually the case of in-band passivity violations. (In-band frequency-dependent transmission lines”, in Proc.
passivity violations generally decrease in size when increasing Electronics Systemintegration Technology Conference,
the fitting order.) An undesired side effect is that causality is Dresden, Germany, September 5-7, 2006, pp. 56-61.
violated due to the instantaneous coupling between the line
[13]B. Gustavsen, “Passivity enforcement for transmission
ends. This coupling is however insignificant when the needed
line models based on the method of characteristics”, IEEE
correction is small, see for instance Figs 6 and 7.
Trans. Power Delivery, submitted.
The passivity enforcement approach was in this work used
[14]B. Gustavsen, and O. Mo, “Interfacing convolution based
together with the Universal Line Model, but it is equally well
linear models to an electromagnetic transients program”,
applicable to other MoC based models.
International Conference on Power System Transients
Conclusions (IPST), 2007, submitted.
Passivity enforcement for MOC based transmission line
models can be achieved by adding a rational correction term
to the line terminals. The correction term is derived directly
from the model nodal admittance matrix via a frequency
sweep, giving a conductance matrix. The bandwidth of the

172 SPI2007