TIME-DOMAIN ANALYSIS OF

FIELD-EXCITEDTRANSMISSION LINE SYSTEMS

BY USING MODEL-BASED PARAMETER ESTIMATION

M. D'Amore, IEEE Fellow M. S.Sarto, IEEE Member

Department of Electrical Engineering - University of Rome "La Sapienza"

Via Eudossiana, 18 - 00184 Rome, Italy

Abstract : The time-domain analysis of field-excited lossy frequency window by using only the coefficients of their rational
transmission tine (TL) networks with nonlinear loads is based on the approximating function. The samples of the spectral quantities for
nodal approach in the time-domain. The computation of the nodal the inverse DFT computation are then obtained by evaluating
admittance or impedance matrix coefficients of the linear part of the rational functions, which are known analytically. As a result, the
network is required in the frequency-domain, in order to obtain the computation-time is nearly-independent of the frequency- sampling
corresponding transient functions by the inverse Discrete Fourier step.
Transform (DF"). The model-based parameter estimation (MBPE) The proposed procedure will be applied to the analysis of the
technique is applied to improve the efficiency of the method in case induced effects on large-dimension field-excited TL networks. The
of large-dimension configurations. The developed procedure allows efficiency of the method will be assessed with reference to the
to represent transcendent functions over very wide bandwidths by required computation-time.
using only the Coefficients of their rational approximating functions.
Good accuracy and considerable reduction of the computation-time e-Do& A&&
are achieved. The EMP-induced effects on three signal mesh-type
networks, having increasing geometrical complexity, with nonlinear Let consider a field-excited network with multiconductor
loads, are calculated in order to assess the efficiency of the proposed dissipative lines and nonlinear loads, above a lossy ground plane.
method. The transient response of the system is obtained by applying the
nodal approach in the time-domain, having suitably represented each
Introduction multiconductor line-section in the frequency-domain by an
equivalent PI-type circuit containing impressed shunt current
The analysis of multiconductor networks excited by transient sources, induced by the extemal EM field [2].
electromagnetic (EM)fields, containing nonlinear protective devices, The proposed approach leads to the following nonlinear matrix
has been performed by using a combined frequency- and time- equation system of Volterra second type:
domain procedure based on the nodal approach 11-31. The developed t
method allows an accurate simulation of the configuration in the i(t) = - i,(t) + y(t-z) v(z) dz (la)
0
frequency-domain and the computation of the transient response of t
the system by solving nonlinear integral equation systems of v(t) = vo(t) + 1 z(t-z) i(z) dz (lb)
0
Volterra second type. The transient nodal admittance matrix and
with the nonlinear boundary conditions:
source vector coefficients have been computed by means of the
inverse D m . Therefore, a considerable amount of data has been i(t) =A t v(t) 1 I v(t) =f, [ i(t) 1 (2)
handled in the frequency-domain, because of the fast transient in which all the excitations are supposed to begin after t=O.
sources exciting the system, in order to avoid the Gibb's In eqs.( 1,2) i(t) and v(t) represent the vectors of the unknown
phenomenon in the inverse DFT calculation. In case of networks currents and voltages involving the nonlinear loads, whose
having complex configurations, the described numerical procedure is characteristics are supposed to be monotone; i,(t) and vo(t) are the
quite time-consuming. The efficiency of the method should be impressed current source and open-ended voltage vectors; y(t) and
improved by reducing the computation-time and the computer z(t) the transient nodal admittance and impedance matrices. The
memory occupation. time-domain model expressed by eqs.( 1,2) is solved by using a
The purpose of this paper is to use the MBPE approach to numerical technique based on the trapezoidal rule, which reduces the
increase the numerical efficiency of the developed procedure for the integral matrix equation of Volterra type to a nonlinear algebraic
time-domain analysis of large-dimension field-excited TL systems, equation system. Suitable algorithms are developed in order to
containing nonlinear elements. The MBPE technique is a numerical achieve the solution in case of strong nonlinearities.
procedure widely applied to the treatment of electromagnetic The coefficients of vectors i,(t), vo(t) and matrices y(t), z(t) are
problems and based on Prony's method [4-61. This technique allows computed by the inverse DFT of the corresponding quantities in the
to represent with very good accuracy a generic function in a wide frequency-domain:

C"3-1-7 $4.0001QB4IEEE 258

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.
 i,(t) = F-' [ I~(O)1 , vo(t) = F-' [ Vo(o) 1 d Ri
(3) fi(w)=X (9)
y(t) = F-I [ Y(0)I . z(t) = F" [ Z(0) I i=l
~

(jw- pi)
in which Z(w)= Y -'(U). The MBPE allows the complete representation of a generic
In case of fast transient sources such as the EMP, the most function by estimating the coefficients of the rational approximation,
time-consuming process in the numerical procedure consists in the obtained by using only few samples of the function itself. The
frequency-domain evaluation of the sampled values for the technique is applied following the algorithm described in Appendix,
calculation of the inverse DFT, because accurate results are obtained which involves a point-matching procedure and leads to a 2d-order
only if a wide frequency window is considered. The proposed linear equation system, whose solutions are the terms No ... Nn, Do ...
technique, based on the MBPE method, allows a consistent reduction Dd-, in eq.(4), having assumed D,=l and d samples of F(o) as
of the time and computer-memory in the computation of the transient forcing quantities.
quantities. In particular, the procedure is very convenient to be Difficulties can arise in the treatment of the 2d-order equation
applied to the assessment of the v(v+1)/2 coefficients of matrix Y(o) system because the 2d x 2d matrix of the coefficients involves terms
or Z(O), in case of large-dimension networks having a great number like wkwith k=l ...d. Therefore, computer overflow problems occur in
v of nodes. case of high values of w.Moreover, if F(w) contains a great number
of resonance peaks, it should be approximated by a rational function
MBPE-Procedure Freauencv-Domain with high value of d, over a wide frequency-window; as a
consequence, the full 2d x 2d matrix, including elements with very
Let consider the generic coefficient of matrix Y(o) or Z(W) different values, is ill-conditioned and the computation of the
represented by function Fij(o);in the following the subscripts i and j unknown coefficients is not accurate at all.
will be omitted for semplicity. F(w) is a transcendent function In order to avoid computer overflow problems, the calculation is
because refers to a distributed parameter system. However, since the carried out by using the normalized frequency a*= w/wo, coo being
physical structure has natural resonant frequencies, F(w) can be minimum or maximum value of the sampled points. In this case, the
approximated as a rational function in the following form: poles and residues are:
I\ N(jo) No + N I (io)+ ..... + Nn (jw)" pi=p:w, , Ri=R;w, , i=l ...d (10)
F(o) = -= (4)
D(jw) D, + D, (iw) + ..... + Dd ( j ~ ) ~ where p; and R; are the normalized ones.
in which N(jo) and D(jo) are polynomials of jo,having order n and The ill-conditioning problem is overcome by breaking up the
d, respectively. Let notice that the inverse FT of 6(0) is expected to whole frequency window into several sub-domains, containing only
be a real function; therefore, all its poles pi, i=l ...d, corresponding to few resonance frequencies, in which F(o) can be approximated by a
the roots of DCjo), are supposed to be real or complex conjugated low-order rational function. The frequency-range of the generic kIh
pairs and d is an even number. That means NCjw) and DOw) are sub-domain is Amk = ok- wk-', wk and o ~being
. ~the maximum and
polynomials with real coefficients. minimum value, respectively. The obtained approximation of F(w) in
As F(w) represents the transfering function of a physical system, the kthsub-domain is the rational function Gk(w) given in eq.(9).
d is assumed either equal to n or n+l. In particular, if the asymptotic Therefore, @a) is expressed in the following form:
value of F(o), for o approaching infinite, is equal to zero, it is set
d=n+l. Otherwise, if this value differs from zero:
lim F(w)= F,,, where q is the number of the frequency-windows and U is the step-
O+- function.
the following modified function is considered: The choice of d=10 to represent function $,(U) having only two
peaks over the window allows to overcome both the problems of
FM(o)= F(w) - F,, (6) overflow and ill-conditioning.
FM(w) has a null asymptotic value and can be represented in
rational form $"(w) by using d=n+l. The inverse FT of 6(w) will be Critical Points of the Procedure
then expressed as:
The main problem in the use of the described procedure consists
F-' [ P(w) ] = F [ fi'(w) J + FhfS(t) (7)
in the choice of the sub-domains for the rational approximation of
6(t) being the Dirac's delta function. In the following, d will be F(o) and of the d sampling points inside each frequency-window.
always assumed equal to n+l. Concerning the first point, let begin with the case of a
The residues Ri, i=l ...d, are given by the following well-known single-conductor lossless line, which can be considered as a very
expression: simple network constituted by only one 1 long line-section and two
NP,)
R, = ~ (8) ports.
WPl) Let enforce a current pulse 6,(t) at one end of the line. The
in which D'(pi) is the (jo)-derivative of DCjw) calculated for jo=pi; observed voltage at the other side will consist of a pulse SY(t-7)
let notice that eq.(8) is valid if p, is a single pole of @(w). Function having a time-delay T = l/c = 2 d 0 , with c = 3.108 m s-'. Therefore,
$(a)can be espressed in the following form: the fundamental resonance frequency of the line is orand is related

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.
 to the path the current and voltage waves have to cover between the The previous considerations can also be extended to case of lossy
two sides. The Fourier transform of the 2-delayed pulse GV(t-.r),is the networks, still representing valid criteria for the choice of Am.
complex function exp(-jurs) = cos(o2) - j sin(w) with period equal
to r=2d.r=or, and is just the transfering function of the line between In conclusion, the developed procedure consists in the following
the two ports. steps:
As a result, the frequency-spectra of the real and imaginary parts determination of the period r of the real and imaginary part of the
of the transfering function F(w), representing the generic mutual coefficients of Z(O) or Y(o);
coefficient of matrix Z(o),will be even and odd functions, determination of the window width Ao;
respectively, showing two peaks in each period r. Therefore, the computation of the coefficients of the rational approximating
rational approximation of F(w) can be computed by using ten-order function fik(o) over each sub-domain by using only ten equidistant
expressions over each r-wide interval. samples;
Let now observe the voltage waveform at the same side where reconstruction of the frequency-spectrum of $(a)from the
the current pulse is enforced. In case of mismatched line, the computed parameters.
transient response will be constituted of a peak at time t d , followed
by an other one after a 2.r-delay. In fact, the current and voltage
waves leaving from one side of the Line, have to cover a 21-long path
to reach again the starting point. The proposed procedure is applied t o the analysis of three
Consequently, if F ( o ) represents the generic proper term of field-excited TL networks having different configurations, in order to
matrix Z(w),the frequency-spectra of the real and imaginary parts verify the efficiency of the method, assessed in terms of the required
will be even and odd functions, respectively, having two peaks in time for the computation of the sampled values of the nodal
each period r/2. In this case, the width of the windows for the impedance matrix coefficients in the frequency-domain.
rational approximation of F(w) have to be chosen equal to r/2, if it is At first, the four-port network sketched in Fig.1 is considered.
set d=10. Each line-section is constituted by three parallel 2 m long wires with
Finally, let consider a more complex network constituted by L 1 mm radius, 1 cm apart, 70 cm distant from the aluminium
line-sections with different lenghts I,, 1, ... I,. A current pulse reference plane. T h e network i s stressed by an impinging
enforced at node j will get the node k by following different paths. electromagnetic plane wave having incident angle 6=30° and
The observed voltage at node k will then consists of several pulses azimuth angle w=lOo, represented in the time-domain as a double
having a different delay .r,=l,lc. The corresponding spectra of the real exponential function with time-constants 2, = 250 ns, T~ = 2 ns, and
and imaginary parts of F(o) are obtained by the superposition of peak value E, = 50 kV/m. The ports of the network are closed on 400
different functions exp(-jmi). The period r of the resulting function, R resistances parallel connected t o varistors, whose static
which is the same for all the coefficients of matrix Z(w), will be characteristic is described by the following function:
equal to the minimum common multiple among the periods r,,T2 ... v = V, [ i / % 1 l'a (12)
rL,with ri=2x/.ri=ori.The fundamental resonance frequency with V, = 5 kV, I, = 0.1 A, a = 30. The analysis of the configuration
or=2xc/A of the network corresponds to the maximum factor A
is performed by using the voltage formulation (lb). The MBPE
common to I,, I, ... IL, and is equal to r. In this case, function F(o) is
method is applied to the computation of the elements of the modified
characterized by several peaks inside each q-wide interval. Eq.(lI)
twelve-order impedance matrix ZM(o),which is obtained by
is then used by considering sub-intervals of r with the same width
subtracting the high-frequency nodal impedance matrix Z,, from
Am, equal to the breadth of the smallest windows containing only
Z(o)= Y(o)-'. As all the line-sections have the same length, the
two peaks of F(w). In case of tree-type networks, it correspond to the
maximum common factor A is equal to 2m and the period is
longest path connecting pairs of nodes. If the system has a
r/2x=clA=150 MHz. The width A d 2 n of the generic sub-interval is
loop-structure this path, including the same node only once, can be
assumed to be of 30 MHz, corresponding to the maximum extension
calculated by considering the maximum number of line sections
of the network. Over each 30 MHz-wide interval, tenth-order rational
which connect the farest ports.
functions are computed by using ten equidistant samples.
The last remark concerns the computation of the poles and
residues of the rational approximating function $(a).It is important
to point out that the computed roots of the denominator of the
rational function are just curve-fitting poles and do not represent the 3

actual poles of the system. In fact, in each Aw-wide window a .Y

two-peak complex function is approximated as a ten-order rational
function; the corresponding ten poles are situated whatever along the
frequency-axis and their contribution in any point external to the
considered frequency-window is not null. In other words, each
ten-pole set referring to a Am-wide window produces a good
* &
approximation of the transcendent function only inside that interval,
whereas outside other ten-pole sets are used, Fig.1- Sketch of the four-port field-excited multiconductor network.

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.
 100 -

~ I""

150 300 0 150 300

frequency [MHz] frequency [MHz]
Fig.4 - Frequency-spectra of the real (-) and imaginary (---) parts of
40 I I ky,l(l,l)for the network in Fig.1 with 3 m long line-sections.

1 17

-40 '
0
I

150
I
300
(a) (b)
Fig.5 - Geometrical configuration of the multiconductor networks
frequency [MHz] with seven- (a) and nineteen- (b) ports.
(b)
Fig.2 - Frequency-spectra of the real (-) and imaginary (---) parts of
ky,l(l.l)(a) and ky,ql,3)(b) for the network in Fig.1.
Figs.2a and 2b show the frequency spectra of the real and
imaginary parts of the generic proper and mutual coefficients zy,l(l,l)
and ky,4(l,3K respectively, computed by using the proposed
procedure. Zy,l(l,l)jsthe coefficient (1,l) of the input matrix
impedance at port 1; Zy,4(13) is the coefficient (1,3) of the transfering
matrix impedance between port 2 and 4. Let notice that all the
curves in Figs.2a,b have the period of 150 MHz; moreover, it is 0 150 300
interesting to point out that the real parts (continuous line) are even frequency [MHz]
functions, whereas the imaginary parts (hidden line) odd functions. Fig.6 - Frequency-spectra of the real parts ky,,,,,3,(-) and 2y.7(1.3)
The waveforms of the corresponding frequency-spectra of the exact (---) for the seven-port network in FigSa.
impedance matrix ZMare nearly coincident. The maximum error is
not greater than few units per cent as shown in Fig.3, with reference The 150 MHz periodicity of the impedance matrix coefficients
to the real and imaginary parts of ?y.4(l,3). disappears in Fig.4, where the real and imaginary frequency-spectra
of k ~ , l ( lare
+ l represented
) when all the line-sections in the network
are 3 m long; the waveforms have period r/2n = c/3 = 1 0 0 MHz.
Successively. the seven-port network in FigSa is analyzed. The
structure is obtained by adding other four elementary triangle-
networks to the previous configuration; therefore, it keeps the same
geometrical characteristics of the network in Fig. 1 but an increased
complexity. The periodicity of the frequency-spectra of the
impedance matrix coefficients is always l-/2~=150MHz, but the
width of each sub-interval is Awl211 = 25 MHz, because of the
increased dimension of the network. Fig.6 shows the
1
frequency-spectra of the real part of both the approximated and exact
0 150 300
frequency [MHz] coefficients ky,7(l,3) and 2y.,(,.3); the two curves cannot be
Fig.3 - Percentage error in the computation of the real (-) and distinguished and are very similar to the waveform in Fig.2b
imaginary (---) parts of ?:,%,>) for the network in Fig.1. ?y,4(l,3),
referring to the real part of except for two additional peaks
261

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.
 around 75 MHz, in each period. These peaks give evidence of the 10
increased dimension of the network.
..
.........__
._.-
I
The computation-times required by the MBPE method and by the 0
exact procedure are reported in the graph of Fig.7 as functions of the
assessed samples of a generic coefficient of the impedance matrix, E -10
with reference to both the first and second configuration. The shown s
m
data are obtained by running the programs on a RISC-based -20
>
processor with 33 MHz-clock. The hidden lines, corresponding to the
obtained curves for the exact procedure, increase linearly with the -30
number of samples. At the contrary, the continuous lines, referring
to the MBPE method, are nearly horizontal; infact, whatever is the 40
0 50 100
wished number of samples, the algorithm always requires the same time [ns]
computation-time to evaluate the coefficients of the rational (a)
approximation. The sampled values are then obtained by using the
achieved analytical representation. Let notice that good accuracy in
lo
the inverse DFT computation is reached by using at least 2000
samples; that means 910 seconds with the exact method and only 52
seconds via MBPE, for the network in FigSa.
At last, the large dimension network in FigSb, with 19 ports and
42 three-conductor line-sections is considered. The structure is
obtained by joining 24 elementary triangle-networks. The period
r / 2 n is always 150 MHz, but Aw/2n is set equal to 10 MHz. The
frequency spectra of the real part of coefficients ky,19(l,3)
Zy,19(1,3)
and
are represented in Fig.8. The maximum error is of only few -10 '
0 50
I I
100
unit per cent. time Ins]
(b)
I I I I IIII I I I 1 1 1111 I I I I 1 1 1 1 1 I Fig9 - Induced voltages on conductor 1 at ports 1 , 4 , 6 , 15 of the
network in Fig.Sb, terminated on 400 Q resistances parallel
connected (b) or not (a) to 5 kV varistors.

It it very interesting to note that in this case the computation-time

of 3000 samples required by the exact procedure is equal to 4346
seconds, that is more than one hour, whereas it is only 689 seconds
by using the MBPE method. These times are obtained by running
both the programs on a RISC-based processor with 66 MHz-clock.
I 1 I I 1 1 1 1 1 I 1 I I 1 1 l 1 1 Finally the transient voltages at each port of the three described
100 lo00 loo00 networks were computed. Figs.9a,b show the induced voltages on
number of samples conductor 1 at ports 1,4, 6, 15 of the largest network of FigSb, with
Fig.7 - Computation-time using the MBPE-method (-) and the or without varistors.
exact procedure (---),for the networks in Figs.1 (a) and 5a (b).
Conclusions
-" I 1
I The developed procedure based on the MBPE method allows a
consistent reduction of the computation-time in the transient analysis
of transmission line networks, containing nonlinear loads, excited by
fast transient electromagnetic sources. The method allows to
compute the rational approximation of the coefficients of the nodal
admittance or impedance matrix by using only few samples in the
frequency-domain. In fact, especially in case of large-dimension
configurations the evaluation of the sampled values of these terms
represents the most time-consuming process in the whole procedure,
0 150 300 because a considerable number of samples is necessary in order to
frequency [MHz] achieve good accuracy in the inverse DFT calculation.
Fig.8 - Frequency-spectra of the real parts of 2:,19(1,3)
(-) and Practical criteria, based on the geometrical configuration of the
Zy,,9(l,3)(---)for the nineteen-port network. network, are given for the choice of the sampling-frequencies.

262

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.
 Accurate results and very little error are obtained by breaking up the means the degree n of the numerator is not greater than the degree d
whole frequency bandwidth in several sub-domains. The of the denominator.
transcendent functions are then approximated over each sub-domain In (13) the unknown coefficients of N(s) and D(s) can be
by using low-order rational expressions. The efficiency of the calculated by using a point- matching algorithm, that is forcing the
procedure has been evaluated with reference to the required approximated function @(s)to be equal to the exact function F(s)
computation-time, expressed as function of the number of the only for few values of the complex frequency S=(T + jw. By setting
calculated samples. In case of a nineteen-port network with 42 D, = 1, m=n+d+l sampled values of F(s) are necessary to compute
three-conductor line-sections, the time saving for the computation of the n+d+l unkown coefficients in eq.(13). Therefore, the following
3000 samples by using the MBPE technique was about the 84%. linear equation system is obtained:
Further investigations will be devoted to the possibility of
N(s.) No + N, si + ..... + Nn sin
estimating the actual poles and residues of the system. In this case, it F(si) = F(sJ = = , i=l ...m (14)
will possible to express analitycally the inverse DFT of the D(sJ Do + D, si + ..... + S:
transcendent functions as a sum of exponential terms. Eq.( 14) can be rewritten in the following form:
F(s,) D(si) = N(sJ , i=l ...m (15)
Acknowledmnent
or by using a matrix expression:
This work was performed by the economic support of the AX=B
Commission of the European Communities to the Science Project where:
"Electromagnetic Compatibility: Fast Transients in
F, F, s1 ... F, sp' -1 -sI ,.. -sq -
Telecommunication and Power Apparatus and Systems".
F, F,s, ... F, si-' -1 -sa ... -si
A = ... ... ... (17a)
References ... ... ...
- F, Fmsm ... Fm's: -1 -sm ... -si -
M. D' Amore, M. S . Sarto, "EMP-coupling to multiconductor
is an mxm matrix and
dissipative lines with nonlinear loads above a lossy ground",
---
Proc. 10* Int. Zurich on EMC, Zurich, March 9-1 1, 1993. X = [Do D, ... D,, No N, ... Nn It (17b)
M. D'Amore, M. S . Sarto, "Time response of a network
containing field-excited multiconductor lossy lines with B = [ -F, S; -F, si ... -Fds: ... ... ..
nonlinear loads", Proc. 1993 IEEE EMC Int. Symp,, Dallas, are the m-dimensional vectors of the unknown coefficients and the
August 9-13, 1993. sampled values, respectively. In eqs.(l7a-c) Fi is the ith sampled
M. D' Amore, M. S . Sarto, "Transient analysis of field-excited value of F(s), calculated for s=si. The algebraic equation system (16)
multiconductor lossy networks with nonlinear dynamic loads: an is solved by using well known numerical techniques, such as the
integral equation approach", Proc. ICEEA, Turin, September low-up (LU) decomposition.
1417,1993. Let consider the case in which F(w) is a function defined in the
S . Chakrabarti, K. R. Demarest, E. K. Miller, "An extended frequency-domain. Let suppose that its poles and residues are real or
frequency-domain Prony's method for transfer function complex pairs and the degree of D u o ) is an even number. The
parameter estimation", Int. Journal of Numerical Modelling, Vol. equation system (15) can be rewritten by using a real formulation. In
6, pp. 269-281, Nov., 1993. fact, by setting si=jwi:
E. K. Miller, G. J. Burke, "Using model-based parameter
F(oi) = F, = FR(oi)+ j F'(wi) = F: +j F/ (18)
estimation to increase the physical interpretability and numerical
efficiency of computational electromagnetics", Computer with F: and F: real numbers and
Phvsics Communication, Vo1.68, pp.43-75, 1991. (-1)h o?h , k=2h
J. N. Brittingham, E. K. Miller, J. L. Willows, "Pole extraction
from real-frequency information", ELQE. m, Vo1.68, No.2,
(io,)k =
{ j (-11,
, k=2h+ 1 (19)

the real and the imaginary part in eq.( 15) can be separated:
Feb., 1980.
d/2
C
h=I
(-Uh (QZh [ F r D,, - N,,, I =-
- wi F: Dd+l-2h (14~; (2Oa)
ApDendix Rational Aporoximation of a Function by .MBPE
d/2
A
h=l
(-l)h [ - Fit D,, + W, Nd+,-,,- aiF r D,+,-,, ] = (14 F; (20b)
Let F(s) the pole-residue approximation of the generic function
F(s) defined in Laplace domain: with i=l ...d. Eqs.(20a,b) are obtained by supposing that n=d-1.
No + NI s + ..... + N,, s" However, similar expressions can be achieved for whatever n. Let
N(s)
F(s) = -= (13) notice that in this case only d sampled values of F(o) are necessary
D(s) Do + D, s + ..... + D, sd to calculate the real unknown coefficients Do, D, ... D,.,, No,N , ...
in which No ... Nn, Do ... D are unknown complex coefficients. For N,.,, having assumed D,=l. In fact, by using expression (15), d
R
physical causal systems, F(s) is a proper rational function, that it sampled values lead to 2d forcing quantities.
263

Authorized licensed use limited to: Siemens AG GBS B&S SOL GMS. Downloaded on December 30,2024 at 05:31:10 UTC from IEEE Xplore. Restrictions apply.