Manuscript

An Improved Passivity Enforcement Algorithm for

Transmission Line Models using Passive Filters
H.M. Jeewantha De Silva, Mohammad Shafieipour

Abstract-- This paper proposes a simple but effective method perturbation of the fitted parameters and passivity is enforced
based on shunt passive filters to enforce passivity on a frequency as a solution to a constrained optimization problem. However,
dependent transmission line model for multi-conductor cables and the derivation and implementation of such algorithms are
overhead lines. The passivity enforcement algorithm is applied to
a widely-used frequency dependent line model in EMT-type complicated, as they require many matrix linearization and
software, the Universal Line Model. The passivity violating regions eigenvalue sensitivity calculations. Furthermore, these methods
of the transmission line model are identified using the frequency are typically valid for eliminating small passivity violations,
sweep method. The passive shunt series RLC filters are added to which are commonly due to approximations in the linearization
the nodes of the transmission lines to eliminate passivity violations. process. In addition, with these methods there is no guarantee
Examples of multi-conductor underground cable systems are that the convergence is always achieved, as it depends on
presented to demonstrate the validity of the proposed approach.
several factors and for large transmission line systems with
Keywords: Electromagnetic transients, passivity enforcement, many conductors, these methods may require significant
phase domain model, passive filters. computation time (e.g. several minutes depending on the case).
Alternatively, passivity can be enforced analytically through
I. INTRODUCTION Hamiltonian matrix [6]. This approach is widely applied to
admittance-based realization of a frequency dependent
W IDEBAND transmission line models are widely used in
electromagnetic transient (EMT) studies such as
temporary over-voltages, switching over-voltages, network
component or network equivalent. The work in [7], [8] extend
this method to transmission lines. However, they are limited to
modal domain models based on constant transformation
resonance, lightning over-voltages, etc. These models
matrices. Note that for underground cables and vertically
accurately consider frequency dependency as well as
asymmetrical transmission lines, the transformation matrix is
distributed nature of the line parameters for frequencies ranging
frequency dependent. Again, the derivation and the
from 0 Hz to a few MHz. In this paper, transmission line refers
implementation of this enforcement algorithm is tedious and
to both overhead lines and cables.
requires significant effort. The computer memory and time
The time domain implementation of a transmission line
requirements for the algorithm can be significantly high for
involves several steps, which are summarized as follows. First,
large transmission line configurations with several
the line parameters such as propagation function and
conductors/cables.
characteristic admittance are formulated in frequency domain
Reference [9] discusses a filter-based method to enforce
for several frequency samples [1]. Next, by applying the
passivity for a two-layer network equivalent. A passivity
“Vector Fitting” technique, the frequency domain
enforcement method for multi-conductor transmission lines via
characteristics are approximated using continuous rational
filters is proposed in [10]. However, a drawback of this method
functions [2], [3]. Finally, the recursive convolution method is
is that the corrected model eliminates the natural decoupling of
applied to represent the transmission line equations as a
the transmission line. In EMT-type programs, the natural
standard EMT-type model. This includes a shunt conductance
decoupling of frequency dependent transmission line is a
and a parallel current source.
significant advantage as it divides the system into small
Transmission lines are passive as a matter of physical reality.
subsystems, which leads to faster simulations.
However, due to the errors in approximating frequency domain
This paper proposes an improved passivity enforcement
characteristics using rational functions as well as occasional
algorithm using passive filters for transmission line models in
frequency domain approximations, the resulting model may
EMT-type software. An improved quality factor estimation for
become non-passive [4]. It is observed that a non-passive model
passive filters is introduced. Compared to [10], an advantage of
may lead to unstable time domain simulations. One of the major
the proposed method is that the natural de-coupling of the
challenges of frequency dependent transmission line models is
transmission line is also maintained. Additionally, the proposed
to enforce the stability of the time domain simulations.
method does not require an iterative procedure to converge
Several passivity enforcement algorithms have been
numerically. Rather, it uses successive steps to enforce
proposed [4-6]. Some of these methods [4], [5] are based on
passivity at the local level until passivity is enforced globally.

H.M.J. De Silva is with the Manitoba Hydro international, Canada (e-mail:

[email protected]). Paper submitted to the International Conference on Power Systems Transients
(IPST2021) in Belo Horizonte, Brazil June 6-10, 2021.
M. Shafieipour is with Safe Engineering Services & technologies ltd., Laval,
QC H7L6E8, Canada (e-mail: [email protected])
This is different from approaches based on linearization, which combination of log and linear scales. The frequency range
may lead to numerical divergence. It should be noted that in this should cover the bandwidth of frequencies in time domain
paper, we use a widely used wideband transmission line model simulations [4].
in an EMT-type program [13], namely the Universal Line
Model [2] to demonstrate the effectiveness of the propose III. PASSIVITY ENFORCEMENT FOR TRANSMISSION LINES VIA
technique. However, it is expected that the introduced method PASSIVE SHUNT FILTERS
can be used with other transmission line models. A. Preliminaries
One possible drawback of the proposed approach is that the
accuracy of the fitted function may be decreased by adding a The fundamental concept behind the passivity enforcement by
filter. However, the passivity corrections are limited only to the filters is that the eigenvalues of a matrix can be changed by
passivity violating regions. Therefore, the inaccuracy of the rest modifying the diagonal elements of the matrix. If every
of the spectrum is insignificant. Furthermore, if the passivity diagonal element (Dii) of the matrix is increased by a small
violations are small, the error is negligible as shown in the value (dDii), all eigenvalues are increased by the same value
examples studied in Section IV. (dDii).
Addition of shunt conductance [4] and the accurate earth
II. PASSIVITY RELATED TO TRANSMISSION LINE AND CABLES return impedance and admittance formulas [13] can improve
passivity conditions in frequency domain parameter
A. Definition of passivity calculations. However, this may not eliminate all violations due
The passivity of the transmission line model is guaranteed; to unavoidable error accumulation in the curve-fitting
if and only if its transfer admittance function, Y(s) is positive procedure.
real for any given frequency. The transfer admittance is related The series RLC filter can be used to enforce passivity [9] and
to the sending and receiving-end voltages and currents as they are added as shunt elements to the terminals of the
transmission line model at both ends as shown in Fig. 1. A shunt
I  V  (1)
 I   Y V 
k k
RLC branch is relatively straightforward to implement in EMT-
 m  m Type software using existing routines.
First, passivity-violating regions are identified using a
The transfer admittance matrix can also be written in terms passivity identification method discussed in Section II.
of the propagation function A(s) and characteristic admittance
Yc(s) matrices (s = jω) [4, 10]
( I  A2 )1 ( I  A2 )Yc ( I  A2 ) 1 (2 A)Yc  (2)
Y  
 ( I  A2 )1 (2 A)Yc ( I  A2 ) 1 ( I  A2 )Yc 

The necessary and sufficient conditions for the transmission Fig. 1. Transmission line with passive filters
line model to be passive are [4, 10]
 a  Y  s  is analytic s :Re(s)0 (3) For each frequency-violating band, the transfer function of the
filter is defined as
 b  Y ( s)  Y ( s )
 c  H  s   Y T  s   Y  s   Hermitian Matrix  𝐹(𝑤) =
𝐾𝜆0 (4)
𝜔 𝜔
is positive definite 1 + 𝑗𝑄 ( − 0 )
𝜔0 𝜔

The complex poles and residues always appear in conjugate where, Q is the quality factor, ω is the angular frequency, λ0 is
pairs in vector fitting algorithms. Hence, the first two conditions the negative eigenvalue with the largest magnitude in that
are always satisfied. To fulfill the third condition, the violating band, and ω0 is the angular frequency at which the
eigenvalues of the Hermitian matrix should be positive for most negative eigenvalue occurs. In (4), the factor K > 1.0
every frequency ω. ensures that the corrected eigenvalues are positive by at least a
pre-determined (small) amount (e.g. K = 1.0001). The series
B. Passivity Identification
resistance, inductance and capacitance values are computed as
The frequency sweep method is used to identify violating
regions. In this method, the eigenvalues of the Hermitian matrix 1 (5a)
is computed for the frequency range of interest (e.g. 0.001 Hz 𝑅=
𝐾𝜆0
to a few MHz) and negative eigenvalues are monitored to 𝑄𝑅
identify passivity violating regions. A potential drawback of 𝐿= (5b)
𝜔0
this method is that there can be missing negative eigenvalues
1
between two frequency samples. This can be avoided by 𝐶= (5c)
analyzing the eigenvalue characteristics as a function of 𝑅𝑄𝜔0
frequency and by taking sufficient number of samples in
 B. Improved estimation of the quality factor (Q) criteria is used for positive R and L values.
Selecting a proper quality factor for the filter is critical for
successful enforcement of passivity. A small quality factor may 𝜆𝐿 𝜔𝐿 2 (8)
< 𝜆𝑈 < 𝜆𝐿
lead to an over-passivity compensation and a large quality 𝜔𝑈 2
factor can lead to an under-passivity compensation. Fig. 2
shows the effect of quality factor on the magnitude of series where λL and λU are the magnitudes of the negative eigenvalue
RLC filter transfer function. at ωL and ωU, respectively, with ωL being the first frequency
sample. If the above criteria is not met, λU is replaced with (9)

𝜔𝐿 2 (9)
𝜆𝑈 = 0.5𝜆𝐿 ( + 1)
𝜔𝑈 2

Similarly, for a low-pass filter, the requirement for R and C to

be positive is

𝜆𝑈 𝜔𝐿 2 (10)
< 𝜆𝐿 < 𝜆𝑈
𝜔𝑈 2

where, ωU is the last frequency sample. If the above criteria is

Fig. 2. Filter characteristics with different quality factors Q
not met, λL is replaced with (11)
Reference [2] suggests computing the quality factor based
𝜔𝐿 2 (11)
on inequalities. The quality factor was determined based on the 𝜆𝐿 = 0.5𝜆𝑈 ( + 1)
magnitude of the filter admittance function [2]. However, it 𝜔𝑈 2
should be noted that only the real part of the filter admittance D. Algorithm
contributes to the eigenvalues of the transfer admittance matrix.
The addition of sufficient shunt conductances to the
Therefore, the quality factor can be evaluated as
admittance matrix improves the condition of passivity at very
low frequencies. For a given transmission line model, the
√√2 − 1 passivity violations are first determined by evaluating the
𝑄1 = 𝜔 𝜔 (6a)
( 𝑈 − 0) eigenvalues of the Hermitian matrix (based on curve-fitted Yc
𝜔0 𝜔𝑈 and A) for a set of frequency samples as discussed in Section II.
√√2 − 1 For each frequency, the most negative eigenvalue of the
𝑄2 = 𝜔 (6b)
𝜔 Hemitian matrix is selected. The passivity violating regions are
( 0 − 𝐿)
𝜔𝐿 𝜔0 accordingly identified based on the most negative eigenvalues
𝑄 = min⁡(𝑄1 , 𝑄2 ) (6c)
in each frequency. For each violating band, the frequency (ω0)
at which the most negative eigenvalue occurs, is determined.
Starting with the negative eigenvalue with largest magnitude
where, ωU and ωL are the upper and lower angular frequencies
for all violating bands, filters are added one by one until all
between which the eigenvalue curve is negative. The diagonal
eigenvalues are positive. If the violating band is within the
elements of the transfer admittance matrix [Y(ω)] are updated
lower and upper bounds of the frequency samples, the series
to include the admittance contribution of the filter F(ω)
RLC filters are added to correct the violation as discussed in
Section III-C. If there are violations at the upper or lower
𝑌(𝑖, 𝑖) = 𝐹(𝜔) + 𝑌(𝑖, 𝑖) (7)
frequency samples, a low-pass filter (series RC) or a high-pass
filter (series RL) is added, respectively. Fig. 3 shows the
C. Elimination of violations at upper and lower passivity enforcement algorithm in a flow chart.
bounds
A low-pass filter (series RC) or high-pass filter (series RL) can
be used to eliminate violations at the upper or lower frequency
samples in the frequency spectrum, respectively. The filter
parameters are derived by substituting the magnitude of the
negative eigenvalues and frequencies at the beginning and at
the end of the violating region (i.e. λL, λU, ωL, ωU). Depending
on the frequencies and the eigenvalues at the boundaries of the
violating region, the filter parameters (such as R, L, or C) can
be negative. For a high-pass filter, the following inequality
 The orders of the characteristic admittance (Yc) and the
propagation function (H) are 14 and 58, respectively. There are
four modes in the propagation function. The fitting rms errors
corresponding to the Yc and H are 0.0954 % and 0.07314 %,
respectively. The curve-fitting frequency range is 0.1 Hz to 1
MHz with 100 frequency samples. The residue/pole ratio of the
propagation function is 3.64.
The passivity of the cable system is examined using the
frequency sweep method for a frequency range from 0.1 mHz
to 1 MHz. The eigenvalues of the Hermitian matrix are shown
in Fig. 5 before and after the addition of sufficient shunt
conductances to the admittance formulation. The presence of
negative values in the original cable eigenvalues of H indicate
that the original model is non-passive and therefore may lead to
numerical instability in time domain simulations. However,
with the addition of the shunt conductances (e.g. 1.0e-10 Ω.m)
the large passivity violations at low frequencies are eliminated.

Fig. 3 Flow chart of the passivity enforcement algorithm

IV. NUMERICAL RESULTS

A. Application Example 1
The proposed passivity enforcement algorithm is
demonstrated using a 2 km long three-phase underground cable
system as shown in Fig. 4. The cable layout is flat with 0.1 m
distance between the cables. The data for the cable system is
Fig. 5 Eigenvalues of Hemittian matrix H (dotted lines: original cable; solid
shown in Table I. The frequency dependent phase domain lines: cable with a shunt conductance)
model (Universal Line Model) [2] in PSCAD/EMTDC
commercial software [13] is used for demonstration. The addition of shunt conductances does not always guarantee
a passive model. There can be negative eigenvalues at other
frequencies. These values are then removed by passive filters
as discussed in Section III. It can be seen that with the addition
of the filters, the line model becomes passive (see Fig. 6)..

Fig. 4 Three-phase underground cable configuration

TABLE I
TRANSMISSION LINE DATA
Cable data
Conductor Outer Radius 0.022 m
Inner Ins. Outer Radius 0.0395 m
Sheath Outer Radius 0.044 m
Outer Ins. Outer Radius 0.0475 m
Inner Ins. Capacitance 0.3 uF/km
Outer Ins. Relative Permittivity 2.3
Conductor Dc Resistance 0.046 ohms/km
Sheath Resistivity 2.826e-8 Ωm
Outer Ins. Relative Permittivity 2.3
Other Fig. 6 Eigenvalues of Hemittian matrix H (dotted lines: without filters; solid
Ground resistivity 100 Ωm lines: with filters)
Length of the line 2 km
Five filters with characteristics listed in Table II are added to
the nodes of the line model. In Table II, F0, λ0, R, L, and C are
the frequency at the negative eigenvalue λ0, negative eigenvalue
(magnitude of the largest negative eigenvalue in the violating
band), resistance, inductance and capacitance of the RLC series
filter, respectively.

TABLE II
PASSIVE FILTER DATA
F0 (Hz) λ0 R (Ω) L (H) C (F)
59.3 7.3185e-07 1.366257e6 4.739e3 1.5191e-09
172.6 1.5024e-07 6.655176e6 7.009e3 1.2119e-10
6.6 9.4675e-08 1.0561364e7 2.34257e5 2.4661e-09
12.2 4.8082e-08 2.0795796e7 1.40435e6 1.1971e-10
569.313 2.7594e-08 3.6236519e7 3.4450e4 2.2685e-12
Fig. 8 The receiving-end voltage of phase A (Solid line: with passivity
enforcement; dotted line: without passivity enforcement)
Fig. 7 compares the transfer admittance function (Y) of the line
model before and after the addition of the filters. The maximum A short circuit test is conducted to verify the accuracy of the
error is around 6e-7. This demonstrates that the error due to the simulation in time domain (see Fig 9). The phase A at the
addition of passive filters is very small in the frequency domain. sending-end of the cable is energized with step voltage (1 V)
and all other conductors are grounded through a 0.01 Ω
resistance.
The receiving-end voltage is compared with a solution obtained
via Numerical Laplace Transform (NLT) technique [14] (see
Fig 10). The time domain results from the simulation show a
close agreement with the NLT. This demonstrates the accuracy
of the proposed passivity enforcement algorithm.

Fig. 7 Top graph: Magnitude of diagonals of the original (solid curve) and
modified (dotted curve) transfer admittance between 1 Hz to 1 MHz;
Bottom graph: Difference between actual and modified transfer admittance

The sending-end of the cable is energized with 225 kV (L-L)

RMS three-phase voltage source and all other conductors are
kept open. A breaker is connected between the cable and the
source. The breaker is initially closed but opened at t = 1.0 s.
The receiving-end voltage of phase A is shown in Fig. 8. It is
clear that without the proposed passivity enforcement
technique, the simulation is unstable. When the breaker is open, Fig. 9 Short circuit configuration for the cable system
the receiving-end voltage should approach zero, as there is no
source acting on cable. This can be seen from the waveform
corresponding to the passive model. However, the waveform
corresponding to the non-passive model deviates from the
solution after the breaker is opened and exhibits numerical
instability towards the end of the simulation.
A non-passive model can give stable or unstable simulations
depending on many parameters including external circuit
parameters, time step and circuit breaker operation, etc.
 Fig. 10 The receiving-end voltage of phase A (Solid line: time domain Fig. 11 Eigenvalues of Hemittian matrix H (dotted lines: without filters;
simulation with passivity enforcement; dotted line: NLT solution without solid lines: with filters)
filters)
Two filters are added to compensate for passivity violations and
B. Application Example II the filter parameters are as shown in Table IV.
In this section, the passivity enforcement method is TABLE IV
demonstrated with an example having two cable circuits in PASSIVE FILTER DATA
parallel. The horizontal distance between the circuits is 1.0 m F0 (Hz) λ0 R (Ω) L (H) C (F)
and between the cables is 0.2 m (flat configuration). The depth 26.7353 5.6527e-06 176890.0166 2580.2026 1.3735e-08
12.2749 2.1085e-06 474221.4298 21457.765 7.8347e-09
of the cables is 2.0 m. The cable system data is shown in Table
III.
TABLE III In this example, all conductors of the cable system are kept
TRANSMISSION LINE DATA open. The first conductor of the sending-end is energized with
Cable data step voltage with initial ramp. The time step is 1.0 µs and the
Conductor Outer Radius 0.030 m
Inner Ins. Outer Radius 0.056 m
length of the simulation is 0.1 s. The voltage at the first
Sheath Outer Radius 0.060 m conductor of the receiving-end is observed. Figs. 12 and 13
Outer Ins. Outer Radius 0.065 m show voltage waveforms with and without the proposed
Inner Ins. Capacitance 0.205 uF/km passivity enforcement algorithm, respectively. With passive
Outer Ins. Relative Permittivity 2.3
Conductor Resistivity 1.7241e-08 Ωm
filters, the voltage waveform is stable.
Sheath Resistivity 2.8264e-08 Ωm
Outer Ins. Relative Permittivity 2.3
Other
Ground resistivity 100 Ωm
Length of the line 10 km
Shunt conductance 1.0e-9 Ohms.m

The plot of eigenvalues of the Hermittian matrix is shown in

Fig. 11. It can be seen that the passivity is successfully enforced
and there are no negative eigenvalues after the passivity
enforcement procedure.

Fig. 12 Receiving-end voltage of core-conductor of first cable with passivity

enforcement
 [8] A. Chinea, S. Grivet-Talocia, “A Passivity Enforcement Scheme for
Delay-Based Transmission Line Macromodels”, Microwave and Wireless
Components Letters, IEEE Volume 17, Issue 8, pp. 562 – 564, August.
2007.
[9] D. Shu, X. Xie, Z. Yan and V. Dinavahi, "A Two-Layer Network
Equivalent With Local Passivity Compensation With Applications to
Hybrid Simulations of MMC-Based AC–DC Grids," in IEEE
Transactions on Power Systems, vol. 34, no. 6, pp. 4514-4524, Nov. 2019.
[10] B. Gustavsen, "Passivity Enforcement for Transmission Line Models
Based on the Method of Characteristics", IEEE Transactions on Power
Delivery, vol. 23, no. 4, pp. 2286-2293, Oct. 2008.
[11] C. Chen , E. Gad, R. Achar, “Passivity Verification in Delay-Based
Macromodels of Multiconductor Electrical Interconnects”, IEEE Trans.
Circuit and systems, vol. 52, October 2005.
[12] H. Xue, A. Ametani, J. Mahseredjian, Y. Baba, F. Rachidi and I. Kocar,
“Transient Responses of Overhead Cables due to Mode Transition in High
Frequencies”, IEEE Transactions on Electromag. Compatibility, vol. 60,
no. 3, pp.785-794, June 2018.
[13] "PSCAD/EMTDC manual," Manitoba Hydro International ltd.,
Available: https://hvdc.ca/pscad/. [Accessed 16 Oct. 2020].
[14] P. Moreno and A. Ramirez, "Implementation of the Numerical Laplace
Fig. 13 Receiving-end voltage of core-conductor of first cable without passivity Transform: A Review Task Force on Frequency Domain Methods for
enforcement EMT Studies, Working Group on Modeling and Analysis of System
Transients Using Digital Simulation, General Systems Subcommittee,
V. CONCLUSION IEEE Power Engineering Society," in IEEE Transactions on Power
Delivery, vol. 23, no. 4, pp. 2599-2609, Oct. 2008
This paper proposes an improved filter-based passivity
enforcement algorithm to ensure the stability of transmission
line models in EMT-type software. The inclusion of adequate
shunt conductance and the use of accurate earth return formula
can enhance the condition of passivity at very low and very high
frequencies respectively. The remaining passivity violations
(generated by curve-fitting procedure) are eliminated by adding
RLC, RC or RL series filters. The band-pass filter eliminates
passivity violations with improved quality factor estimation.
The physical characteristics of the low- and high-pass filters are
enforced.
Numerical results show that the stability of the time domain
simulations involving a three-phase underground cable system
can be ensured using the proposed method. This method can be
easily implemented in EMT-type software to improve stability
of the transmission line models.

REFERENCES
[1] L.M Wedepohl, “Application of Matrix Methods to the Solution of
Traveling Wave Phenomena in Polyphase Systems”, Proc IEE, 110, pp.
2200-2212, 1963
[2] A. Morched, B. Gustavsen, M. Tartibi, “A Universal Model for Accurate
Calculation of Electromagnetic Transients on Overhead Lines and
Underground cables” , IEEE Transactions on Power Delivery, Vol. 14,
No 3, July 1999.
[3] B. Gustavsen, A. Semlyen, “Simulation of transmission line transients
using Vector fitting and modal decomposition”, IEEE Transactions on
Power Delivery, Vol. 13, No 2, April 1998
[4] H. M. J. De Silva, A. M. Gole, J. E. Nordstrom and L. M. Wedepohl,
"Robust Passivity Enforcement Scheme for Time-Domain Simulation of
Multi-Conductor Transmission Lines and Cables," in IEEE Transactions
on Power Delivery, vol. 25, no. 2, pp. 930-938, April 2010
[5] B. Gustavsen, “Enforcing Passivity for Admittance Matrices
Approximated by Rational Functions”, IEEE Transactions on Power
Systems, Vol. 16, No 1, February 2001.
[6] C. Chen, D. Saraswat, E. Gad, M. Nakhla, R. Achar and M. C. E. Yagoub,
"Passivity Enforcement for Method of Characteristics-based
Macromodels", Proc. International Symposium on Signals, Systems and
Electronics (ISSSE), Montreal, Canada, July 2007.
[7] C. Chen, D. Saraswat, E. Gad, M. Nakhla, R. Achar, M.C.E. Yagoub,
“Passivity Enforcement for Method of Characteristics-based
Multiconductor Transmission Line Macromodels”, Signals, Systems and
Electronics, 2007. ISSSE '07. International Symposium, pp. 25 – 28 July
30 2007-August 2 2007.