See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260135192

A New Algorithm for Wire Fault Location Using Time-Domain Reflectometry

Article  in  IEEE Sensors Journal · December 2013

DOI: 10.1109/JSEN.2013.2294193

CITATIONS READS
45 549

2 authors:

Qinghai Shi Olfa Kanoun

Technische Universität Chemnitz Technische Universität Chemnitz
30 PUBLICATIONS   236 CITATIONS    543 PUBLICATIONS   2,239 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dynamics of energy harvesting in small devices View project

ISINA (Interdisciplinary Symposium for Women in the MINT area including Mathematics, Informatics, Engineering, Natural Sciences and Technology) View project

All content following this page was uploaded by Qinghai Shi on 24 March 2014.

The user has requested enhancement of the downloaded file.

IEEE SENSORS JOURNAL, VOL. 14, NO. 4, APRIL 2014 1171

A New Algorithm for Wire Fault Location Using

Time-Domain Reflectometry
Qinghai Shi, Student Member, IEEE, and Olfa Kanoun, Senior Member, IEEE

Abstract— In time domain reflectometry (TDR) attenuation signal includes information about changes of wire impedance
and dispersion of the reflected signal limit the reachable accuracy and can be therefore used to detect wire faults.
for wire faults location. Because time of flight is evaluated, Over the last decade, many methods, such as Time
the wire faults with small impedance changing are difficult to
locate. In this paper, a novel method for the TDR-based wire Domain Reflectometry (TDR), Frequency Domain Reflectom-
fault detection is presented by transfer function analysis in the etry (FDR), Time-Frequency Domain Reflectometry (TFDR)
time domain. For the determination of the transfer function, a and Spectrum Time Domain Reflectometry (STDR) [1]–[8]
deconvolution should be carried out. Thereby, an inverse problem were developed. They use in general different incident signals
is to be solved by an adaptive filter approach. Adaptive filters are and signal processing methods. However, these techniques can
able to reduce spurious noise of the deconvolution and lead to
an acceptable deconvolution estimate. Therefore, a high signal-to- locate hard faults (open and short circuits) that produce big
noise-ratio can be reached. The filter’s stopband characteristics reflection but they are not always able to locate the small
are optimized by optimization technique to reduce the noise anomalies such as frays or chafes, whose small reflections are
components of the transfer function in the frequency domain. confused by the noise of measurement.
For that a nonlinear fitting procedure is proposed using the Some authors have demonstrated success locating soft faults
Riad–Parruck optimization criterion. The developed method can
locate both hard faults (open and short circuits) and soft faults in a controlled laboratory environment without the impedance
with small impedance changes, and identify the type of wire faults changes from mechanical vibration, movement and moisture
simultaneously in a controlled laboratory environment (without [9]–[12]. In these literatures the wire is normally fixed on a
the impedance changes from mechanical vibration, movement, table or other surface to prevent movement and vibration, and
and moisture). The algorithm using adaptive filters and optimiza- carefully measured with minimal measurement noise.
tion techniques is proposed in this paper for the traditional TDR
method, but it is general for most other reflectometry approaches. In [13], the authors have provided the condition of the
The estimated wirings are coaxial cables and twisted pair cables, location of the soft faults using TDR, FDR and SSTDR
which are used in electrical and power distribution systems. methods: the impedance variation in the environment of the
Index Terms— Wire fault location, inverse problem of wire system because of the vibration and movement must be
deconvolution, adaptive filters, optimization techniques, time smaller than the impedance changes due to the soft faults itself.
domain reflectometry (TDR), coaxial cable, twisted-pair cable Some improved methods [13]–[15] use the baseline method,
and transfer function. in which the output signal of the faulty wiring is compared
with the output of the healthy wiring, in order to detect and
I. I NTRODUCTION locate soft faults. This baseline approach is a natural fit to find
soft faults, but it is difficult to obtain a perfect baseline in a

U SING the wiring in electrical systems, vehicles and

power distribution systems, has caused an increased
requirement for electrical characterization of wiring networks
realistic environment. If the wirings vibrate and move enough
that the changes in the baseline may exceed the changes due
to the small anomalies, so that these soft faults cannot be
and location of wire faults. There are several emerging
detected.
approaches for wire faults location and characterization. The
Correlation algorithms have been applied to detect and
most widely used technique for wire fault location is Time
locate the small discontinuities [5]. An intrinsic limitation of
Domain Reflectometry (TDR). For that, a high-frequency
this technique is the attenuation and dispersion of reflected
signal or a pulse signal is sent down the cable. The reflected
signal that can limit the maximum distance of wire fault
Manuscript received September 19, 2013; revised November 29, 2013 detection and can affect the accuracy of fault location.
and December 2, 2013; accepted December 3, 2013. Date of publication Cepstrum [16] and Pecstrum [17] algorithms have been
December 5, 2013; date of current version February 21, 2014. This work was also applied using transfer function to detect and locate wire
supported in part by German Federal Ministry of Economics and Technology
and in part by the Central Innovation Program of German Federation of faults. Because the noise of the deconvolution for the transfer
Industrial Research Associations. This is an expanded paper from the IEEE function can’t be proposed, these techniques were only applied
SENSORS 2012 Conference. The associate editor coordinating the review of to locate hard faults that cause a large reflection, which has a
this paper and approving it for publication was Prof. Lorenzo Lo Monte.
The authors are with the Chair for Measurement and Sensor Technology, high SNR in the TDR measurement.
Chemnitz University of Technology, Chemnitz 09126, Germany (e-mail: In this paper a novel approach is presented for the wire fault
[email protected]; [email protected]). location. By signal processing the noise is reduced, so that
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. even soft wire faults can be detected and characterized with a
Digital Object Identifier 10.1109/JSEN.2013.2294193 high accuracy in a controlled laboratory environment (without

1530-437X © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1172 IEEE SENSORS JOURNAL, VOL. 14, NO. 4, APRIL 2014

impedance variation in the environment of the cable due to The curve fitting methods are expensive to compute and
vibration, movement and moisture). noisy due to the numerical sensitivity. For the minimum
Section II describes the principle of deconvolution for variance deconvolution (MVD) algorithms, extra computations
determination of the transfer function. For the realization are required to compute the autoregressive moving average
different filter principles are investigated. Section III shows (ARMA) coefficients [24]. For Wiener filtering based methods
two adaptive filters together with optimization techniques for (i.e., Wiener filters and spectral extrapolation), the amount
solving the inverse problem of deconvolution and to determine of computations required is relatively moderate. Numerical
the transfer function in the time domain. ill conditions can be taken care of simply by introducing
Section IV and V show the experimental investigation and a constant noise desensitizing parameter. However, wiener
analysis of results, respectively. Finally, Section VI gives the filters and spectral extrapolation suffers seriously from the
conclusions. band-limit of the input signals. It is difficult to find the optimal
noise desensitizing parameter in Wiener filters and reliable
II. D ECONVOLUTION frequency range in spectral extrapolation.
Iterative deconvolution techniques are applied to design an
For a linear, time invariant and causal system with a zero adaptive filter to reduce the deconvolution noise and yield
initial state, the relation between incident signal x(t), transfer the acceptable estimate of impulse response h(t). With the
function h(t) and reflected signal y(t) can be written as follows optimization criteria the optimal parameter of the adaptive
∗ filters can be calculated in order to get a “best” acceptable
y(t) = x(t) h(t) (1)
deconvolution estimate. In this study iterative deconvolu-
where “*” denotes the convolution operation. tion techniques are implemented to solve the deconvolution
The convolution can be represented in the frequency domain problem because the optimization techniques enables a higher
by the following equation degree of freedom to the user for the compromise the signal
distortion and noise reduction.
Y(ω) = X(ω) · H(ω) (2)
where X(ω), H(ω), and Y(ω) denote the frequency domain
III. A DAPTIVE F ILTERS AND O PTIMIZATION T ECHNIQUES
forms of x(t), h(t) and y(t).
Since deconvolution is the inverse process of the A. State of the Art
convolution operation, this inverse operation involves the
determination of either x(t) or h(t) assuming that there is an Matched filters have been implemented in [25] and cited in
adequate knowledge of the two remaining time functions. The [26] to locate cable faults. The matched filter improves the
process of determining h(t) knowing both x(t) and y(t) is a SNR by reducing the noise’s spectral bandwidth to that of the
system identification procedure. The corresponding operation input signal. It is obtained by correlating a known signal or
in the frequency domain is derived as template with an unknown signal to detect the presence of
the template in the unknown signal [27]. This is equivalent to
H(ω) = Y(ω)/ X (ω) (3) convolve the unknown signal with a time-reversed version of
the template. In this study the unknown signal is the reflected
Next, Eq. (4) is applied to compute H(ω) to which the inverse signal. The known signal is the incident signal. This means
fast Fourier transform can be used to yield h(t). that the system must know in advance what kind of incident
h(t) = IFFT{H(ω)} (4) signals it wishes to detect. Matched filter is easy to implement
in reflectometry method. However, the main drawback of
The zeros of X(ω) are also zeros of Y(ω). At these zeros, the matched filter approach is that the measurement system
in the Eq. (3) H(ω) becomes indeterminate. Hence applying requires a prior knowledge of the incident signal. If this
(3) will result in spurious noise errors. In this region, the information is inaccurate, the matched filter performs poorly.
deconvolution errors in the H(ω) will be large and the results In [25], the simulated fault condition is proposed with a
will contain some high values that can be viewed as a sequence zero-mean white noise which is critical applied in the realistic
of spike or δ(ω) functions. Then in time domain form of measurement.
impulse response h(t) with inverse Fourier transform these In this study the adaptive filters and optimization techniques
Dirac delta functions δ(ω) produce error contributions spread are applied to improve SNR. The prior knowledge of the
over the entire time domain transform epoch or most of it. incident signal in this algorithm is unnecessary. The optimiza-
If the error components are high, they can hide all or most tion techniques are implemented to find the best acceptable
the information of h(t). parameter of the adaptive filters, so that the impulse response
The deconvolution operation needs to be modified beyond has the same bandwidth as the incident signal in order to
the division to yield a low noise acceptable estimate of reduce the spectra of the noise and improve the SNR. In this
impulse response h(t). Numerous approaches for the decon- study the RP techniques is applied as the optimization criteria
volution, such as Wiener filters, spectral extrapolation, mini- because it is relative simple and the iterative routine can
mum variance deconvolution, curve fitting methods and iter- be easy implemented. This optimization technique enables a
ative deconvolution techniques, have been developed in the choice of optimum solutions through an investigation of the
literature [18]–[23]. accuracy and the noise indicators, which give a higher degree
SHI AND KANOUN: NEW ALGORITHM FOR WIRE FAULT LOCATION 1173

of freedom to the user for the compromise the signal distortion

and noise reduction.

B. Two Adaptive Filters

The inverse problem of deconvolution for the transfer
function in the time domain includes the optimization of
the filter and definition of optimal parameter. The optimum
filter interpolates the impulse response H(ω) at the zeros of
measured input signal Xm (ω) to yield the estimated impulse
response Hest (ω). A class of these filters F(ω) is given by:

1
F(ω) = G(ω)
(5)
1+ |Xm (ω)|2

where G(ω) is a positive function of the frequency depends

on the deconvolution techniques and Xm (ω) is the frequency
domain form of the measured input signal. The filter is
designed, so that the quantity G(ω)/|Xm (ω)|2 is small and
F(ω) is almost unity in the regions where Xm (ω) is not
very small. But if Xm (ω) has a small value, the quantity
G(ω)/|Xm (ω)|2 becomes significant and causes adequate filter
attenuation to minimize the noise components.
Fig. 1. Algorithm of the optimization technieques.
Two adaptive filters are applied and compared in this study
to reduce to spurious noise components and produce an accept-
able deconvolution estimate. For the function G(ω) different C. Optimization Techniques
formulations can be principally chosen. In this paper we
Optimization techniques for the adaptive filter are used in
investigate two methods: The automated optimums compensa-
order to find the optimal filter parameter λ.
tion deconvolution and the automated-regularization technique
The error function e(t) is defined as
[28]–[31].
In [28] for the automated optimums compensation decon- e(t) = y(t) − hest (t)∗ xm (t) (10)
volution G(ω) in Eq. 5 is defined as:
where y(t) is the reflected signal without noise, hest (t) is the
G(ω) = λ (6) estimated transfer function and xm (t) is the measured input
signal.
The filter is adaptive, therefore, λ is defined after a number The convergence of iterative techniques can be monitored
of iterations and a specified criterion is realized between by forming the root-mean square error of the estimate,

accuracy and noise content indicators (Eq. 14 and 15).  ⎡ ⎤2
 N−1
1  n
The resulting estimate of the impulse response Hest (ω) of ⎣y (n) −
ei =  hi (j) · x (n − j)⎦ (11)
the optimums compensation technique is: N
n=0 j=0
∗
Hest (ω) = Hm (ω) · F(ω) = Ym (ω) · Xm (ω)/[|Xm (ω)| + λ]
2
where hi (j) is the ith iterative estimate and N is the number
(7) of data points.
The algorithm of the optimization technique is given in
where Hm (ω), Xm (ω), Ym (ω) are the measured impulse Fig. 1. The measured input and output signals are acquired in
response, input signal and output signal in the frequency the time domain and transformed into the frequency domain
domain, respectively. The automated-regularization technique by the Fourier transform. They are processed by the adaptive
[30], [31] proposes the formulation: filter and transformed to the time domain functions by using
of the inverse Fourier transform. The data is stored for each
G(ω) = λ · ω4 (8) of the different values of filter parameters [32]. After each of
the filter parameters are processed, the optimization criteria
The resulting estimate of the impulse response Hest (ω) of are performed for the filter parameter. Then the optimal
the automated-regularization technique is filter parameter is selected and the optimal estimate of the
deconvolution solution Hopt (ω) can be calculated.
Hest (ω) = Hm (ω) · F(ω) Guillaume-Nahman (GN) and Riad-Parruck (RP) criteria are
= Ym (ω) · Xm ∗ (ω)/[|Xm (ω)|2 + λ · ω4 ] (9) typical optimization techniques for deconvolution.
The GN technique is a heuristic approach using frequency
which is similar to (7) except replacing λ by λ · ω4 [30], [31]. domain regularization. The optimal deconvolution solution is
1174 IEEE SENSORS JOURNAL, VOL. 14, NO. 4, APRIL 2014

obtained by monitoring when the standard deviation of the

imaginary portion of the estimated solution is minimal after
using of the inverse Fourier transform [31].
The RP technique is an approach to find the optimal decon-
volution result for impulse response data [28]. The optimal
deconvolution result is obtained if the empirical mean value
and the standard deviation of the tail portion of the impulse
response satisfy certain conditions.
In this study the RP techniques is applied as the optimization
criteria because it is relative simple and the iterative routine
can be easy implemented. The RP technique enables a choice
of optimum solutions through an investigation of the accuracy
and the noise indicators. This gives a higher degree of freedom
to the user for the compromise the signal distortion and noise
reduction.
The RP technique states that for the time duration limited Fig. 2. Experimental TDR system.
impulse response:
h (t) = 0, T1 < t < T2 (12)
used to calculate optimum impulse response between incident
A step response is obtained by integrating the impulse and reflected signals. Using inverse Fourier transform the
response, must be flat for in Eq. (12) defined time zone [28] impulse response in the time domain can be calculated and
then the location of wire faults can be determined by the time

T2 of flight and the velocity of signal propagation through the
w (t) = h (t) dt, T1 < t < T2 . (13) cable.
T1 A pulse signal was generated as the incident signal:
Using frequency domain regularization, it is known that as the x(t) = A · t · e-B·t , t > 0; x(t) = 0, t ≤ 0 (18)
filter parameter λ increased, the step response leads to very
small error content. The accuracy indicator A (λ) and noise The spectrum of the incident signal can be set easily as:
indicator N(λ) are defined for the iterative estimate as [28]: A
H (ω) = 2 (19)
1  B+j·ω
N
A (λ) = w (n) , T1 < t < T2 (14)
N The parameters A and B are adjusted to choose the ampli-
n=1

 tude and transition duration of the applied pulse x(t) and
1  N
N (λ) = 
consequently the bandwidth of signal could be adjusted. In
(w (n) − A (λ))2 , T1 < t < T2 (15)
N this study A = 109 and B = 108 . The rise and fall times of
n=1
the incident signal are 10 ns and 32 ns, respectively.
It is expected that, for small values of the filter parameter λ, RG-58 C/U coaxial cable with 5.2, 10.2 and 20.2 m length
A(λ) decreases at a much slower rate compared to that of the and twisted pair cable from LAPP KABEL with 84 m length
noise indicator N(λ). Consequently, it is expected that a range are used for these experiments. These two different cables with
of λ exists for which A(λ) is very close to A(λ = 0) while hard faults (open and short circuits) and the soft cable faults
N(λ) is much reduced compared to N(λ = 0). The optimum are investigated. The wire faults are described in this study
value λopt for the iterative filter parameter can be defined as: by using different load impedances at the end point of the
  estimated cable. Therefore, the cable length in every estimate
A (λ = 0) − A λopt
1 (16) is the position of the cable fault. As an example a coaxial
A (λ = 0) cable with a length of 5.2 m by different load impedances are
and proposed to identify the types of wire faults.
 
N λopt  N (λ = 0) (17)
V. R ESULTS AND A NALYSIS
The optimization criterion is used to find the optimum
The estimated wirings are coaxial cables and twisted pair
solution of the deconvolution.
cables, which are used in electrical and power distribution
systems. All results in this study are measured in a controlled
IV. E XPERIMENTAL S ETUP laboratory environment.
Fig. 2 shows the experimental TDR system. It consists Fig. 3 and Fig. 4 show the TDR trace of an open circuit
of a directional bridge, an arbitrary waveform generator and at the end point of a 10 m coaxial cable with 50  charac-
an oscilloscope (ADC) which are connected to a PC with teristic impedance and a short circuit at the end point of an
GPIB cable for automatic control of the instruments. Iterative 84 m twisted pair cable with 120  characteristic impedance,
deconvolution, adaptive filter and optimization techniques are respectively. The TDR method works by sending a short pulse
SHI AND KANOUN: NEW ALGORITHM FOR WIRE FAULT LOCATION 1175

Fig. 3. TDR trace of an open circuit at the end point of a 10.2 m coaxial Fig. 5. TDR trace of a 56  load impedance at the end point of a 20.2 m
cable. coaxial cable.

Fig. 4. TDR trace of a short circuit at the end point of a 84 m twisted pair Fig. 6. Spectrum of incident and reflected signals of Fig. 3.
cable.

signal down to the cable. The wave can be reflected whenever

a signal traveling in a cable line encounters an impedance
discontinuity. The amount of signal reflected back is calculated
by the reflection coefficient:
ρ = (ZL − Z0 )/(ZL + Z0 ) (20)
where Z0 is the characteristic impedance of the cable,
ZL is the impedance at the fault and ρ is the reflection
coefficient.
The type of the faults can be estimated from the amplitude
of reflected signals. If the cable has an open circuit then the Fig. 7. H(ω) of an open circuit at the end point of a 10.2 m coaxial cable.
reflected signal is positive and has approximately the same
amplitude as the amplitude of incident wave by short cable
(Fig. 3). If the cable is very long then the reflected signal has a this reason the accuracy of traditional TDR method is limited
clearly attenuation and dispersion (Fig. 4) because of dielectric to approximately from 1 to 5 m for power cables with length
losses of the insulator of cable and the limited conductivity of above 2 km.
conductor in unshielded twisted pair cables. A new algorithm must be used to avoid the human inter-
The distance d between a reflection and injection points can pretation in the measurement process and automated locate
be calculated by d = υ·t/2, where υ is the velocity of the signal and detect hard and soft cable faults using signal processing.
propagation into the cable and t is the time interval between Transfer function between incident and reflected signals is
the incident and reflected signals (time of flight). applied in this study to detect and locate the cable fault.
The reflected signals are attenuated and distorted because Fig. 6 shows the magnitude of the incident and reflected
of the frequency dependent signal propagation, conductive signals in the frequency domain for an open circuit at the
and dielectric loss in the cable system. Small impedance end point of a 10.2 m coaxial cable (Fig. 3). The division
changes in the cable are more difficult to detect because of between both signals is used to calculate the spectrum of the
the smaller reflections (in Fig. 5 there is 6  impedance deconvoled transfer function H(ω).
changes). The dispersion and attenuation of signals in cables Fig. 7 shows the result of the transfer function H(ω). These
would make it hard to clarify the edge of the reflected signal. Dirac delta function in high frequency domain produce error
Consequently, the transitions at input point of the incident contributions spread over the entire time domain transform
signal are smoothed. Therefore, the measurement result is epoch or most of it. Consequently, all the information of h(t)
strongly dependent on subjective interpretation. Because of are hidden by them (Fig. 8).
1176 IEEE SENSORS JOURNAL, VOL. 14, NO. 4, APRIL 2014

Fig. 8. Fault location of an open circuit at the end point of a 10.2 m coaxial Fig. 11. Fault location of a 56  at the end point of a 20.2 m cable with
cable without adaptive filters. adaptive filters.

Fig. 9. Gain-magnitude frequency response of two filters.

Fig. 12. Fault location of different load impedances of a 5.2 m coaxial cable.

pass bandwidth in the frequency domain. In the optimum

compensation technique, λ is constant. Therefore, the stop
band of the adaptive filter exits wherever |X(ω)| is small.
Consequently, it utilizes an adaptive filter of the same type
as the incident signal.
The automated-regularization technique enhances the
low-frequency components due to the ω4 term. This is because
λ·ω4 is very small at low frequencies resulting in |F(ω)| of
approximately 1. It always utilizes a low pass filter with pass
band extending up to highest frequency component in |X(ω)|.
Fig. 10. Fault location of an open circuit at the end point of a 10.2 m coaxial
cable with adaptive filters.
Soft faults are estimated also with optimum compensation
techniques and automated-regularization techniques. Fig. 11
shows a soft fault location at the end point of a 20.2 m
Adaptive filters and optimization techniques are used in coaxial cable. This soft fault is described by using a 56 
this study to reduce the noise of deconvolution and find the load impedance at the end point of the 20.2 m coaxial cable.
optimum deconvolution solution. In TDR trace (Fig. 5) the reflected signal is very difficult to
Fig. 9 shows the gain-magnitude frequency response of detect and the soft faults with small impedance changes can’t
these two adaptive filters in the frequency domain. The best be located without signal processing. With the deconvolution
acceptable parameter of the adaptive filters can be found by and adaptive filters these soft faults can be easily detected and
the optimization technique, so that the transfer function has located and the SNR can be enhanced.
the same bandwidth as the input signal in order to reduce the Fig. 12 shows faults location of different load impedances
spectra of the noise and improve the SNR. for a coaxial cable with a length of 5.2 m. Cable length is
Fig. 10 shows the transfer function in the time domain as a considered as the position of the cable fault to the input point.
result of using of two filter techniques. The accuracy for wire With the amplitude of the transfer function in the time domain
faults location can be improve from this algorithm, because the the type of cable faults can be identified.
time interval between incident and reflected signal is unessen-
tial for the wire fault location. The position of the wire fault VI. C ONCLUSION
can be easily found from the result of this algorithm in Fig. 10. A new algorithm for the wire fault detection and location is
In this estimate the pulse signal in Eq. (18) is applied proposed based on transfer function analysis, adaptive filters
as the incident signal for TDR measurement that has a low and optimization techniques using TDR measurement.
SHI AND KANOUN: NEW ALGORITHM FOR WIRE FAULT LOCATION 1177

The novelty of this new technique is that the SNR of [8] C. Furse, M. Safavi, P. Smith, and C. Lo, “Feasibility of spread
deconvolution, which is necessary for determining the trans- spectrum sensors for location of arcs on live wires,” IEEE Sensors J.,
vol. 5, no. 6, pp. 1445–1449, Dec. 2005.
fer function in the time domain, is significantly improved. [9] J. P. Steiner and W. L. Weeks, “Time-domain reflectometry for
For deconvolution an inverse problem is solved by an adaptive monitoring cable changes: Feasibility study,” Dept. Electr. Comput.
filter approach. Adaptive filters are able to reduce spurious Eng., Univ. Utah, Palo Alto, CA, USA: Tech. Rep. EPRI GS-6642,
Feb. 1990.
noise of the deconvolution and lead to an acceptable deconvo-
[10] B. Waddoups, C. Furse, and M. Schmidt, “Analysis of reflectometry
lution estimate. The optimal filter parameters are determined for detection of chafed aircraft wiring insulation,” in Proc.
by the optimization technique using the Riad-Parruck opti- 5th Joint FAA/DoD/NASA Aging Aircraft Conf., Sep. 2001,
mization criterion. It enables the choice of optimal solutions pp. 2–20.
[11] B. Waddoups, “Analysis of reflectometry for detection of chafed aircraft
under consideration of accuracy and noise indicators. This wiring insulation,” M.S. thesis, Dept. Electr. Comput. Eng., Utah State
gives a higher degree of freedom for the compromise between Univ., Logan, UT, USA, 2001.
signal distortion and noise reduction, so that the optimal [12] S. Wu, “An iterative inversion method for transmission line fault
location,” Ph.D. dissertation, Dept. Electr. Comput. Eng., Univ. Utah,
deconvolution solution can be found and the spurious noise can Salt Lake City, UT, USA, 2011.
be reduced. Soft wire faults with small impedance differences [13] L. A. Griffiths, R. Parakh, C. Furse, and B. Baker, “The invisible fray:
(≤ 10 ) can be also detected and located by this algorithm A critical analysis of the use of reflectometry for fray location,” IEEE
Sensors J., vol. 6, no. 3, pp. 697–706, Jun. 2006.
using the TDR in a controlled laboratory environment. The
[14] C. Furse, Y. C. Chung, C. Lo, and P. Pendayala, “A critical comparison
examples presented in this paper show that a wire fault location of reflectometry methods for location of wiring faults,” Smart Struct.
deviation smaller than 0.4 m can be realized by this algorithm Syst., vol. 2, no. 1, pp. 25–46, 2006.
using the TDR method in a controlled laboratory environment. [15] C. Furse, P. Smith, and M. Diamond, “Feasibility of reflectometry
for nondestructive evaluation of prestressed concrete anchors,” IEEE
These two properties can be improved by the choice of suitable Sensors J., vol. 9, no. 11, pp. 1322–1329, Nov. 2009.
incident signals. [16] C. Chen and L. Roemer, “The application of cepstrum technique in
The type of cable fault can be identified by the amplitude of power cable fault detection,” in Proc. IEEE Int. Conf. Acoust., Speech,
the transfer function in the time domain. If the load impedance Signal Process., vol. 1. Apr. 1976, pp. 764–767.
[17] L. Van Biesen, J. Renneboog, and A. Barel, “High accuracy
of the cable fault is higher than the characteristic impedance location of faults on electrical lines using digital signal processing,”
of the cable, the amplitude of transfer function is positive. IEEE Trans. Instrum. Meas., vol. 39, no. 1, pp. 175–179,
Otherwise it is negative. The value of the load impedance is Feb. 1990.
[18] J. B. Thomas, An Introduction to Statistical Communication Theory.
also proportional to the amplitude of transfer function, if they New York, NY, USA: Wiley, 1967.
are at the same position on the cable. [19] A. T. Walden, “Robust deconvolution by modified Wiener filtering,”
The method is generic, but the investigation in this paper Geophysics, vol. 53, no. 2, pp. 186–191, Feb. 1988.
was carried out exemplarily with coaxial cables and twisted [20] S. L. Marple, Digital Spectral Analysis with Applications. Upper Saddle
River, NJ, USA: Prentice-Hall, 1987.
pair cables, because these are often used in electrical and [21] J. M. Mendel and J. Kormylo, “New fast optimal white noise estimator
power distribution systems. for deconvolution,” IEEE Trans. Geosci. Electron., vol. 15, no. 1,
pp. 32–41, Jan. 1977.
[22] N. Levinson, “The Wiener RMS (root mean square) error criterion
ACKNOWLEDGMENT in filter design and prediction,” J. Math. Phys., vol. 25, pp. 261–278,
The authors wish to thank Uwe Troeltzsch for support on Jan. 1947.
[23] S. Zohar, “Fortran subroutine for the solution of Toeplitz sets of linear
simulation with Matlab and also of GEMAC mbH for support equations,” IEEE Trans. Acoust., Speech, Signal Process., vol. 27, no. 6,
on the cable samples. pp. 656–658, Dec. 1979.
[24] S. K. Sin and C. H. Chen, “A comparison of deconvolution techniques
for the ultrasonic nondestructive evalution of materials,” IEEE Trans.
R EFERENCES Image Process., vol. 1, no. 1, pp. 3–10, Jan. 1992.
[1] C. R. Sharma, C. Furse, and R. R. Harrison, “Low power [25] M. N. Barkesseh, M. Das, and N. A. Kheir, “An identification based
STDR CMOS-sensor for locating faults in aging aircraft wiring,” IEEE matched filtering method for fault detection,” in Proc. Amer. Control
Sensors J., vol. 7, no. 1, pp. 43–50, Jan. 2007. Conf., May 1990, pp. 1907–1908.
[2] Q. Shi and O. Kanoun, “Analysis of the parameters of a lossy coaxial [26] S. R. Schuet, D. A. Timucin, and K. R. Wheeler, “A model-based
cable for cable fault detection,” Trans. Syst., Signals Devices, vol. 7, probabilistic inversion framework for wire fault detection using
no. 4, pp. 311–325, 2012. TDR,” in Proc. IEEE Instrum. Meas. Technol. Conf., May 2010,
[3] C. Buccella, M. Feliziani, and G. Manzi, “Detection and localiza- pp. 422–425.
tion of defects in shielded cables by time-domain measurements with [27] L. Doyle, Essentials of Cognitive Radio. Cambridge, U.K.: Cambridge
UWB pulse injection and clean algorithm postprocessing,” IEEE Trans. Univ. Press, 2009, pp. 100–101.
Electromagn. Compat., vol. 46, no. 4, pp. 597–605, Nov. 2004. [28] B. Parruck and S. M. Riad, “An optimization criterion for iterative
[4] Q. Shi, U. Troeltzsch, and O. Kaonun, “Detection and localization of deconvolution,” IEEE Trans. Instrum. Meas., vol. 32, no. 1, pp. 137–140,
cable faults by time and frequency domain measurements systems,” in Mar. 1983.
Proc. 7th Int. Mult-Conf. Syst., SSD, Jun. 2010, pp. 1–6. [29] N. S. Nahman and M. E. Guillaume, “Deconvolution of time domain
[5] E. Song, Y. Shin, P. Stone, J. Wang, T.-S. Choe, J.-G. Yook, et al., waveforms in the presence of noise,” Dept. Commerce, Nat. Bureau
“Detection and location of multiple wiring faults via time-frequency- Stand., Washington, DC, USA, Tech. Note 1047, 1981.
domain reflectometry,” IEEE Trans. Electromagn. Compat., vol. 51, [30] N. S. Nahman and M. E. Guillaume, Some Results Using the Guillaume-
no. 1, pp. 131–138, Feb. 2009. Nahman Automated Deconvolution Method. Memphis, TN, USA: URSi,
[6] Q. Shi and O. Kanoun, “Application of deconvolution for wire fault Jan. 1983.
location using time domain reflectometry,” in Proc. IEEE Sensors, [31] N. S. Nahman, “Software correction of measured pulse data,” Fast
Nov. 2012, pp. 1380–1383. Electr. Opt. Meas., vols. 108–109, pp. 351–417, Jan. 1986.
[7] P. S. Smith, “Spread spectrum time domain reflectometry,” Ph.D. [32] O. Kanoun, U. Troeltzsch, and H.-R. Traenkler, “Benefits of evolutionary
dissertation, Dept. Elect. Comput. Eng., Utah State Univ., Logan, UT, strategy in modeling of impedance spectra,” Electrochim. Acta, vol. 51,
USA, 2003. nos. 8–9, pp. 1453–1461, Jan. 2006.
1178 IEEE SENSORS JOURNAL, VOL. 14, NO. 4, APRIL 2014

Qinghai Shi received the master’s degree in elec- Olfa Kanoun (SM’06) has been a University Pro-
trical engineering from the Dresden University of fessor of measurement and sensor technology with
Technology, Dresden, Germany, in 2008. Since Chemnitz University of Technology, Germany, since
2009, he has been a Research Assistant and Ph.D. 2007. She studied electrical engineering and infor-
Student with the Chair for Measurement and Sensor mation technology from the Technical University,
Technology, Chemnitz University of Technology, Munich, in 1996, where she specialized in the field
Germany. His research concerns cable fault detec- of electronics. During the Ph.D. degree with the Uni-
tion and location with time domain reflectometry, versity of the Bundeswehr, Munich, she developed
impedance spectroscopy and network analyzer, mod- a novel calibration free temperature measurement
eling of the transmission line, network topology, bio- method and was awarded in 2001 by the Com-
impedance measurement, design of sensor systems, mission of Professors in Measurement Technology,
and methods for signal processing. AHMT e. V., Germany.
As a Senior Scientist from 2001 to 2006, she founded a working group on
impedance spectroscopy and carried out since then simultaneously research
on energy autonomous sensor systems. In 2004, she founded the IEEE
Instrumentation and Measurement Chapter within the section Germany.
Her research interests focuses on sensors, measurement systems and mea-
surement methods, battery diagnosis methods, material testing by impedance
spectroscopy, availability and conversion of ambient energy, design of energy
autonomous systems, energy conversion from electrostatic field, smart energy
management, strain, and pressure measurement by sensors based on carbon
nanotubes.

View publication stats