IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 52, NO.

4, NOVEMBER 2010 1

Estimation of the Mathematical Parameters III. N UMERICAL S OLUTION FOR THE CORRELATION BETWEEN
of Double Exponential Pulses (α, β) AND (tr , tfwhm )
Using the Nelder Mead Algorithm Nevertheless, the physical and mathematical parameters have to be
transformed by solving a nonlinear system of equations, which cannot
Mathias Magdowski, Graduate Student Member, IEEE, and be solved analytically. As stated in [2] for an effective numerical
Ralf Vick, Senior Member, IEEE solution, it is convenient to use the double exponential function not
as a function of β and α, but as the ratio of both:
Abstract—Transient pulses for EMC problems, such as the high- β
altitude electromagnetic pulse and ultra-wideband pulses, are often λ= . (3)
α
described by a double exponential pulse. Such a pulse shape is specified
physically by the three characteristic parameters rise time tr , pulse width For a pulse with positive polarity, α < β and λ > 1. In the same
tfwhm , and maximum amplitude Emax . The mathematical description manner, the physical parameters should not be the pulse width tfwhm
is a double exponential function with the parameters α, β, and E0 . In and rise time tr , but also the ratio of both:
practice, it is often necessary to transform the two groups of parameters
into each other. This paper shows a novel relationship between the tfwhm
µ= . (4)
physical parameters tr and tfwhm on the one hand and the mathematical tr
parameters α and β on the other. It is shown that the least squares
method in combination with the Nelder Mead simplex algorithm is
It has been shown in [2] that µ > 4.291 for all double exponential
appropriate to determine an approximate closed-form formula between pulses. Using only the ratios has the following advantage: If two
these parameters. Therefore, the extensive analysis of double exponential pulses have the same ratio λ, they will also have the same µ. If the
pulses is possible in a considerably shorter computation time. The overall parameters α and β of one pulse are larger than α and β of another
approximation error is less than 3.8 %. pulse with the same λ by a determined factor, then tr and tfwhm will
Index Terms—double exponential pulse, high-altitude electromagnetic be lower by that factor and vice-versa. One can also say that (α, β)
pulse, Nelder Mead, parameter estimation, pulse width, rise time, ultra- and (tr , tfwhm ) of two pulses are reciprocally proportional to each
wideband pulse.
other, if these two pulses share the same λ and µ. The advantage
of this simplification is that the nonlinear optimization has to be
I. I NTRODUCTION performed only in one dimension and not in two, as in [3].
A brief introduction about double exponential pulses and their im-
portance for the electromagnetic compatibility is summarized in [1]. A. Determination of (tr , tfwhm ) from (α, β)
For practical application it is often necessary to transform between
The rise time tr can be calculated by tr = t90% −t10% . The instants
the physical parameters rise time tr , pulse width tfwhm (full width at
of time t10% and t90% can be found by an iterative approach
half max), and maximum field strength Emax and the mathematical
ln e−αt10%,i − 0.1


parameters α, β, and E0 . k
A very effective relationship that uses only four assistant variables t10%,i+1 = − , (5)
β
is established in [2]. Since these four variables have different values ln e−αt90%,i − 0.9


for different ratios of β/α to give a reasonable overall fitting error, t90%,i+1 = − k
, (6)
this is not very straightforward. With the help of the least squares β
method and the Nelder Mead algorithm, an estimation of the physical with the starting parameters:
parameters from the mathematical ones is applied in [3]. Unfortu- ln 1 − 0.1


ln 1 − 0.9


k k
nately, these equations cannot be used for the inverse transformation. t10%,1 = − ; t90%,1 = − . (7)
β β
The main idea of this paper is to give closed-form formulas for the
mathematical parameters of the double exponential pulse from the The iteration ends if the relative error between tr,i and tr,i+1 is
physical ones with a method that is based on the algorithm presented smaller than a preset accuracy. This procedure is more precise but
in [3]. Finally, the estimation errors are analyzed and shown. also more time-consuming than using the closed-form formulae given
in [3]. The normalized rise time for different values of 1 < λ ≤ 200
II. D OUBLE E XPONENTIAL P ULSE S HAPES is shown in Fig. 1. The pulse width tfwhm can be calculated with
The double exponential shape is given in [1] as a similar approach. The normalized pulse width is also plotted in
Fig. 1.
E(t) = E0 k e−αt − e−βt h(t), (1)

where E0 is the amplitude, α and β are the characteristic mathe- B. Determination of (α, β) from (tr , tfwhm )
matical parameters and h(t) is the unit-step function. The amplitude The determination of α and β from tr and tfwhm is much more
factor k is necessary to create different double exponential pulse difficult. For a first start, the very rough approximations
shapes with variable parameters but a constant amplitude. Therefore
1 1
k is given in [3] by α= and β= (8)
"

#−1 tfwhm tr
α α
ln
−α α−β
β
ln
−β α−β
β are used. Then a search algorithm is started. The parameters α and β
k = k(α, β) = e −e . (2) are slightly changed in a specific interval upwards and downwards in
each step. The pulse widths and rise times are calculated for each step
With the additional k-factor the maximum amplitude can be as described in Section III-A. These calculated physical parameters
converted by Emax = E0 . are compared to the target parameters for each changed variant of α
Manuscript received February 2, 2010; revised June 4, 2010.
and β. At the end of each step the best fitting variant is chosen as
M. Magdowski and R. Vick are with the Otto von Guericke University the start for the next step. If the best fitting variant is the same as
Magdeburg, Germany. the previous step, the specific interval of change is decreased by half.

0000–0000/00$00.00 © 2009 IEEE

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 52, NO. 4, NOVEMBER 2010 2

2.5 TABLE I
tfwhm · α PARAMETERS OF THE APPROXIMATION OF α
tr · α
2 Parameter Value
X0 4.661 034 × 10−5
tr · α, tfwhm · α

1.5 X1 0.842 411

X2 1.660 852
X3 8.887 335
1 X4 0.624 129

0.5 TABLE II
PARAMETERS OF THE APPROXIMATION OF β

0 Parameter Value
20 40 60 80 100 120 140 160 180 200
β/α Y1 2.167 370
Y2 0.360 349
Y3 0.013 289
Fig. 1. Rise time tr and pulse width tfwhm normalized to α for different Y4 1.358 701
ratios of β to α. Y5 0.137 057
Y6 122.802 861
2.5 Y7 1.314 850

2
exact numerical solution
approximate solution
α · tr , β · tr

1.5 10−1
β · tr
α · tr
1
α · tr

10−2
0.5

0
20 40 60 80 100 120 140 160 180 200 10−3
tfwhm/tr
101 102 103
tfwhm/tr
Fig. 2. Parameters α and β normalized to tr for different ratios of tfwhm
to tr .
Fig. 3. Comparison of exact and approximated values of the parameter α
normalized to tr for different ratios of tfwhm to tr .
Therefore, the search is quite fast at the beginning and becomes more
and more precise near the end. This is done until the desired pulse
parameters tr and tfwhm and the calculated parameters have a relative This approximation model has seven dimensionless parameters
error that is smaller than the preset accuracy. This procedure is quite Y1 . . . Y7 . Using a sum of more exponential functions would be more
simple but very time consuming. Even so, it offers the possibility precise, but would also require more parameters. The closed-form
to transform the physical parameters of the double exponential pulse approximation formulas are used with the criterion to minimize the
to the mathematical ones. The normalized parameters α and β are following sums in a specific region µ ∈ [µmin , µmax ]:
plotted for different ratios µ in Fig. 2. The calculation error is only
a matter of time and can be as low as 10−15 . "µ
max
#
X0 ...X4
X 2
(α − αapprox ) −→ min , (11)
IV. A PPROXIMATION WITH C LOSED F ORM F ORMULAS µmin
Because of the hyperbolic character of α · tr in Fig. 2, a modified
"µ #
max
Y1 ...Y7
X 2
hyperbolic function was chosen to approximate the exact value of α: (β − βapprox ) −→ min . (12)
X0 1 X1 µmin
αapprox =
tr
+ ·
tr
X2 X4 . (9)
tfwhm
− X3
tr This is known as the least squares method [4]. By applying of
The variables X0 . . . X4 are the dimensionless parameters of the the Nelder Mead algorithm [5] in the region 4.291 < µ ≤ 1 000,
approximation model. Due to the exponential character of β · tr , the parameters shown in Tables I and II for the approximation of
which is shown in Fig. 2, and the desire for a simple mathematical the mathematical parameters α and β from the physical ones tr and
expression, a sum of three exponential functions was selected to tfwhm can be derived.
approximate the precise magnitude of β: A comparison between the exact numerical and approximated ana-
1
h t lytical solution for α and β can be found in Fig. 3 and Fig. 4. Despite
− fwhm ·Y3
βapprox = · Y1 − Y2 · e tr
these plots being logarithmically scaled in both axis directions, almost
tr (10)
t
− fwhm ·Y5
t
− fwhm ·Y7
i no differences between the exact and the approximated curve can be
− Y4 · e tr −Y6 · e t r . found.
IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 52, NO. 4, NOVEMBER 2010 3

[2] C. Mao and H. Zhou, “Novel parameter estimation of double exponential

pulse (EMP, UWB) by statistical means,” IEEE Trans. Electromagn.
2
Compat., vol. 50, no. 1, pp. 97 – 100, February 2008.
[3] M. Camp and H. Garbe, “Parameter estimation of double exponential
pulses (EMP, UWB) with least squares and Nelder Mead algorithm,” IEEE
1.5 exact numerical solution Trans. Electromagn. Compat., vol. 46, no. 4, pp. 675 – 678, November
β · tr

2004.
approximate solution
[4] C. L. Lawson and R. J. Hanson, Solving least squares problems, ser.
Prentice-Hall series in automatic computation. Englewood Cliffs, NJ:
1 Prentice Hall, June 1974.
[5] J. A. Nelder and R. Mead, “A simplex method for function minimization,”
The Computer Journal, vol. 7, no. 4, pp. 308 – 313, January 1965.

101 102 103

tfwhm/tr

Fig. 4. Comparison of exact and approximated values of the parameter β

normalized to tr for different ratios of tfwhm to tr .

4
relative error of αapprox
relative error of βapprox
2
relative error in %

0
Mathias Magdowski (GSM’09) was born in Wol-
mirstedt, Germany in 1984. He received his Dipl.-
−2 Ing. degree in electrical engineering from the Otto-
von-Guericke University Magdeburg, Germany in
2008, where he is currently pursuing his Ph.D.
degree at the Institute for Fundamental Electrical
−4 Engineering and Electromagnetic Compatibility.
101 102 103
His current research interests include numerical
tfwhm/tr methods for modeling EMC problems.

Fig. 5. Relative estimation errors of α and β for different ratios of tfwhm

to tr .

V. A NALYSIS OF E STIMATION E RRORS

The relative errors are used to examine the precision of the method.
The estimation errors in (9) and (10) are less than 3.8 % and 3.2 %,
respectively, in the interval 4.291 ≤ µ ≤ 1 000. These estimation
errors are plotted in Fig. 5. As compared to the approximation that
was derived in [2], the relative error in this work increased marginally,
but the presented approach does not require a lookup table. The
relative error is still lower than the error that results from the
approximation formulas that were developed by the two-dimensional
optimization in [3]. All discussed approximations are computationally Ralf Vick (M’99 – SM’05) studied electrical engi-
much more efficient than solving the nonlinear system of equations neering at Dresden University of Technology, Ger-
numerically. many. He received his Ph.D. degree in 1995.
He had been an EMC consultant in private prac-
tice for several years, he was involved in several
VI. C ONCLUSION projects, e.g. the EMC of the German Frigate F124
This paper shows that the Nelder Mead simplex algorithm in and investigation of the radiation of power line car-
rier systems. Since 2004 he has been full professor in
combination with the least squares method is appropriate to determine
electrical engineering. He is now head of the EMC
the relationship between the commonly used physical parameters rise chair at Otto-von-Guericke University Magdeburg,
time and pulse width and the mathematical parameters α and β of Germany.
double exponential pulse shapes. With the help of these approximate
correlations it is possible to perform an extensive analysis of double
exponential pulse shapes with a very wide range of parameters in a
considerably smaller computing time. The estimation errors of α and
β are all less than 3.8 %.

R EFERENCES
[1] Int. Electrotech. Commiss., Testing and measurement techniques - HEMP
immunity test methods for equipment and systems, November 2001, IEC
61000-4-25.

The author has requested enhancement of the downloaded file. All in-text references underlined in blue are linked to publications on ResearchGate.